xref: /freebsd/sys/kern/kern_malloc.c (revision bdafb02fcb88389fd1ab684cfe734cb429d35618)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1987, 1991, 1993
5  *	The Regents of the University of California.
6  * Copyright (c) 2005-2009 Robert N. M. Watson
7  * Copyright (c) 2008 Otto Moerbeek <otto@drijf.net> (mallocarray)
8  * All rights reserved.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  *	@(#)kern_malloc.c	8.3 (Berkeley) 1/4/94
35  */
36 
37 /*
38  * Kernel malloc(9) implementation -- general purpose kernel memory allocator
39  * based on memory types.  Back end is implemented using the UMA(9) zone
40  * allocator.  A set of fixed-size buckets are used for smaller allocations,
41  * and a special UMA allocation interface is used for larger allocations.
42  * Callers declare memory types, and statistics are maintained independently
43  * for each memory type.  Statistics are maintained per-CPU for performance
44  * reasons.  See malloc(9) and comments in malloc.h for a detailed
45  * description.
46  */
47 
48 #include <sys/cdefs.h>
49 __FBSDID("$FreeBSD$");
50 
51 #include "opt_ddb.h"
52 #include "opt_vm.h"
53 
54 #include <sys/param.h>
55 #include <sys/systm.h>
56 #include <sys/kdb.h>
57 #include <sys/kernel.h>
58 #include <sys/lock.h>
59 #include <sys/malloc.h>
60 #include <sys/mutex.h>
61 #include <sys/vmmeter.h>
62 #include <sys/proc.h>
63 #include <sys/sbuf.h>
64 #include <sys/smp.h>
65 #include <sys/sysctl.h>
66 #include <sys/time.h>
67 #include <sys/vmem.h>
68 
69 #include <vm/vm.h>
70 #include <vm/pmap.h>
71 #include <vm/vm_pageout.h>
72 #include <vm/vm_param.h>
73 #include <vm/vm_kern.h>
74 #include <vm/vm_extern.h>
75 #include <vm/vm_map.h>
76 #include <vm/vm_page.h>
77 #include <vm/uma.h>
78 #include <vm/uma_int.h>
79 #include <vm/uma_dbg.h>
80 
81 #ifdef DEBUG_MEMGUARD
82 #include <vm/memguard.h>
83 #endif
84 #ifdef DEBUG_REDZONE
85 #include <vm/redzone.h>
86 #endif
87 
88 #if defined(INVARIANTS) && defined(__i386__)
89 #include <machine/cpu.h>
90 #endif
91 
92 #include <ddb/ddb.h>
93 
94 #ifdef KDTRACE_HOOKS
95 #include <sys/dtrace_bsd.h>
96 
97 bool	__read_frequently			dtrace_malloc_enabled;
98 dtrace_malloc_probe_func_t __read_mostly	dtrace_malloc_probe;
99 #endif
100 
101 #if defined(INVARIANTS) || defined(MALLOC_MAKE_FAILURES) ||		\
102     defined(DEBUG_MEMGUARD) || defined(DEBUG_REDZONE)
103 #define	MALLOC_DEBUG	1
104 #endif
105 
106 /*
107  * When realloc() is called, if the new size is sufficiently smaller than
108  * the old size, realloc() will allocate a new, smaller block to avoid
109  * wasting memory. 'Sufficiently smaller' is defined as: newsize <=
110  * oldsize / 2^n, where REALLOC_FRACTION defines the value of 'n'.
111  */
112 #ifndef REALLOC_FRACTION
113 #define	REALLOC_FRACTION	1	/* new block if <= half the size */
114 #endif
115 
116 /*
117  * Centrally define some common malloc types.
118  */
119 MALLOC_DEFINE(M_CACHE, "cache", "Various Dynamically allocated caches");
120 MALLOC_DEFINE(M_DEVBUF, "devbuf", "device driver memory");
121 MALLOC_DEFINE(M_TEMP, "temp", "misc temporary data buffers");
122 
123 static struct malloc_type *kmemstatistics;
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	65536
131 #define KMEM_ZSIZE	(KMEM_ZMAX >> KMEM_ZSHIFT)
132 static uint8_t kmemsize[KMEM_ZSIZE + 1];
133 
134 #ifndef MALLOC_DEBUG_MAXZONES
135 #define	MALLOC_DEBUG_MAXZONES	1
136 #endif
137 static int numzones = MALLOC_DEBUG_MAXZONES;
138 
139 /*
140  * Small malloc(9) memory allocations are allocated from a set of UMA buckets
141  * of various sizes.
142  *
143  * XXX: The comment here used to read "These won't be powers of two for
144  * long."  It's possible that a significant amount of wasted memory could be
145  * recovered by tuning the sizes of these buckets.
146  */
147 struct {
148 	int kz_size;
149 	char *kz_name;
150 	uma_zone_t kz_zone[MALLOC_DEBUG_MAXZONES];
151 } kmemzones[] = {
152 	{16, "16", },
153 	{32, "32", },
154 	{64, "64", },
155 	{128, "128", },
156 	{256, "256", },
157 	{512, "512", },
158 	{1024, "1024", },
159 	{2048, "2048", },
160 	{4096, "4096", },
161 	{8192, "8192", },
162 	{16384, "16384", },
163 	{32768, "32768", },
164 	{65536, "65536", },
165 	{0, NULL},
166 };
167 
168 /*
169  * Zone to allocate malloc type descriptions from.  For ABI reasons, memory
170  * types are described by a data structure passed by the declaring code, but
171  * the malloc(9) implementation has its own data structure describing the
172  * type and statistics.  This permits the malloc(9)-internal data structures
173  * to be modified without breaking binary-compiled kernel modules that
174  * declare malloc types.
175  */
176 static uma_zone_t mt_zone;
177 static uma_zone_t mt_stats_zone;
178 
179 u_long vm_kmem_size;
180 SYSCTL_ULONG(_vm, OID_AUTO, kmem_size, CTLFLAG_RDTUN, &vm_kmem_size, 0,
181     "Size of kernel memory");
182 
183 static u_long kmem_zmax = KMEM_ZMAX;
184 SYSCTL_ULONG(_vm, OID_AUTO, kmem_zmax, CTLFLAG_RDTUN, &kmem_zmax, 0,
185     "Maximum allocation size that malloc(9) would use UMA as backend");
186 
187 static u_long vm_kmem_size_min;
188 SYSCTL_ULONG(_vm, OID_AUTO, kmem_size_min, CTLFLAG_RDTUN, &vm_kmem_size_min, 0,
189     "Minimum size of kernel memory");
190 
191 static u_long vm_kmem_size_max;
192 SYSCTL_ULONG(_vm, OID_AUTO, kmem_size_max, CTLFLAG_RDTUN, &vm_kmem_size_max, 0,
193     "Maximum size of kernel memory");
194 
195 static u_int vm_kmem_size_scale;
196 SYSCTL_UINT(_vm, OID_AUTO, kmem_size_scale, CTLFLAG_RDTUN, &vm_kmem_size_scale, 0,
197     "Scale factor for kernel memory size");
198 
199 static int sysctl_kmem_map_size(SYSCTL_HANDLER_ARGS);
200 SYSCTL_PROC(_vm, OID_AUTO, kmem_map_size,
201     CTLFLAG_RD | CTLTYPE_ULONG | CTLFLAG_MPSAFE, NULL, 0,
202     sysctl_kmem_map_size, "LU", "Current kmem allocation size");
203 
204 static int sysctl_kmem_map_free(SYSCTL_HANDLER_ARGS);
205 SYSCTL_PROC(_vm, OID_AUTO, kmem_map_free,
206     CTLFLAG_RD | CTLTYPE_ULONG | CTLFLAG_MPSAFE, NULL, 0,
207     sysctl_kmem_map_free, "LU", "Free space in kmem");
208 
209 /*
210  * The malloc_mtx protects the kmemstatistics linked list.
211  */
212 struct mtx malloc_mtx;
213 
214 #ifdef MALLOC_PROFILE
215 uint64_t krequests[KMEM_ZSIZE + 1];
216 
217 static int sysctl_kern_mprof(SYSCTL_HANDLER_ARGS);
218 #endif
219 
220 static int sysctl_kern_malloc_stats(SYSCTL_HANDLER_ARGS);
221 
222 /*
223  * time_uptime of the last malloc(9) failure (induced or real).
224  */
225 static time_t t_malloc_fail;
226 
227 #if defined(MALLOC_MAKE_FAILURES) || (MALLOC_DEBUG_MAXZONES > 1)
228 static SYSCTL_NODE(_debug, OID_AUTO, malloc, CTLFLAG_RD, 0,
229     "Kernel malloc debugging options");
230 #endif
231 
232 /*
233  * malloc(9) fault injection -- cause malloc failures every (n) mallocs when
234  * the caller specifies M_NOWAIT.  If set to 0, no failures are caused.
235  */
236 #ifdef MALLOC_MAKE_FAILURES
237 static int malloc_failure_rate;
238 static int malloc_nowait_count;
239 static int malloc_failure_count;
240 SYSCTL_INT(_debug_malloc, OID_AUTO, failure_rate, CTLFLAG_RWTUN,
241     &malloc_failure_rate, 0, "Every (n) mallocs with M_NOWAIT will fail");
242 SYSCTL_INT(_debug_malloc, OID_AUTO, failure_count, CTLFLAG_RD,
243     &malloc_failure_count, 0, "Number of imposed M_NOWAIT malloc failures");
244 #endif
245 
246 static int
247 sysctl_kmem_map_size(SYSCTL_HANDLER_ARGS)
248 {
249 	u_long size;
250 
251 	size = uma_size();
252 	return (sysctl_handle_long(oidp, &size, 0, req));
253 }
254 
255 static int
256 sysctl_kmem_map_free(SYSCTL_HANDLER_ARGS)
257 {
258 	u_long size, limit;
259 
260 	/* The sysctl is unsigned, implement as a saturation value. */
261 	size = uma_size();
262 	limit = uma_limit();
263 	if (size > limit)
264 		size = 0;
265 	else
266 		size = limit - size;
267 	return (sysctl_handle_long(oidp, &size, 0, req));
268 }
269 
270 /*
271  * malloc(9) uma zone separation -- sub-page buffer overruns in one
272  * malloc type will affect only a subset of other malloc types.
273  */
274 #if MALLOC_DEBUG_MAXZONES > 1
275 static void
276 tunable_set_numzones(void)
277 {
278 
279 	TUNABLE_INT_FETCH("debug.malloc.numzones",
280 	    &numzones);
281 
282 	/* Sanity check the number of malloc uma zones. */
283 	if (numzones <= 0)
284 		numzones = 1;
285 	if (numzones > MALLOC_DEBUG_MAXZONES)
286 		numzones = MALLOC_DEBUG_MAXZONES;
287 }
288 SYSINIT(numzones, SI_SUB_TUNABLES, SI_ORDER_ANY, tunable_set_numzones, NULL);
289 SYSCTL_INT(_debug_malloc, OID_AUTO, numzones, CTLFLAG_RDTUN | CTLFLAG_NOFETCH,
290     &numzones, 0, "Number of malloc uma subzones");
291 
292 /*
293  * Any number that changes regularly is an okay choice for the
294  * offset.  Build numbers are pretty good of you have them.
295  */
296 static u_int zone_offset = __FreeBSD_version;
297 TUNABLE_INT("debug.malloc.zone_offset", &zone_offset);
298 SYSCTL_UINT(_debug_malloc, OID_AUTO, zone_offset, CTLFLAG_RDTUN,
299     &zone_offset, 0, "Separate malloc types by examining the "
300     "Nth character in the malloc type short description.");
301 
302 static void
303 mtp_set_subzone(struct malloc_type *mtp)
304 {
305 	struct malloc_type_internal *mtip;
306 	const char *desc;
307 	size_t len;
308 	u_int val;
309 
310 	mtip = mtp->ks_handle;
311 	desc = mtp->ks_shortdesc;
312 	if (desc == NULL || (len = strlen(desc)) == 0)
313 		val = 0;
314 	else
315 		val = desc[zone_offset % len];
316 	mtip->mti_zone = (val % numzones);
317 }
318 
319 static inline u_int
320 mtp_get_subzone(struct malloc_type *mtp)
321 {
322 	struct malloc_type_internal *mtip;
323 
324 	mtip = mtp->ks_handle;
325 
326 	KASSERT(mtip->mti_zone < numzones,
327 	    ("mti_zone %u out of range %d",
328 	    mtip->mti_zone, numzones));
329 	return (mtip->mti_zone);
330 }
331 #elif MALLOC_DEBUG_MAXZONES == 0
332 #error "MALLOC_DEBUG_MAXZONES must be positive."
333 #else
334 static void
335 mtp_set_subzone(struct malloc_type *mtp)
336 {
337 	struct malloc_type_internal *mtip;
338 
339 	mtip = mtp->ks_handle;
340 	mtip->mti_zone = 0;
341 }
342 
343 static inline u_int
344 mtp_get_subzone(struct malloc_type *mtp)
345 {
346 
347 	return (0);
348 }
349 #endif /* MALLOC_DEBUG_MAXZONES > 1 */
350 
351 int
352 malloc_last_fail(void)
353 {
354 
355 	return (time_uptime - t_malloc_fail);
356 }
357 
358 /*
359  * An allocation has succeeded -- update malloc type statistics for the
360  * amount of bucket size.  Occurs within a critical section so that the
361  * thread isn't preempted and doesn't migrate while updating per-PCU
362  * statistics.
363  */
364 static void
365 malloc_type_zone_allocated(struct malloc_type *mtp, unsigned long size,
366     int zindx)
367 {
368 	struct malloc_type_internal *mtip;
369 	struct malloc_type_stats *mtsp;
370 
371 	critical_enter();
372 	mtip = mtp->ks_handle;
373 	mtsp = zpcpu_get(mtip->mti_stats);
374 	if (size > 0) {
375 		mtsp->mts_memalloced += size;
376 		mtsp->mts_numallocs++;
377 	}
378 	if (zindx != -1)
379 		mtsp->mts_size |= 1 << zindx;
380 
381 #ifdef KDTRACE_HOOKS
382 	if (__predict_false(dtrace_malloc_enabled)) {
383 		uint32_t probe_id = mtip->mti_probes[DTMALLOC_PROBE_MALLOC];
384 		if (probe_id != 0)
385 			(dtrace_malloc_probe)(probe_id,
386 			    (uintptr_t) mtp, (uintptr_t) mtip,
387 			    (uintptr_t) mtsp, size, zindx);
388 	}
389 #endif
390 
391 	critical_exit();
392 }
393 
394 void
395 malloc_type_allocated(struct malloc_type *mtp, unsigned long size)
396 {
397 
398 	if (size > 0)
399 		malloc_type_zone_allocated(mtp, size, -1);
400 }
401 
402 /*
403  * A free operation has occurred -- update malloc type statistics for the
404  * amount of the bucket size.  Occurs within a critical section so that the
405  * thread isn't preempted and doesn't migrate while updating per-CPU
406  * statistics.
407  */
408 void
409 malloc_type_freed(struct malloc_type *mtp, unsigned long size)
410 {
411 	struct malloc_type_internal *mtip;
412 	struct malloc_type_stats *mtsp;
413 
414 	critical_enter();
415 	mtip = mtp->ks_handle;
416 	mtsp = zpcpu_get(mtip->mti_stats);
417 	mtsp->mts_memfreed += size;
418 	mtsp->mts_numfrees++;
419 
420 #ifdef KDTRACE_HOOKS
421 	if (__predict_false(dtrace_malloc_enabled)) {
422 		uint32_t probe_id = mtip->mti_probes[DTMALLOC_PROBE_FREE];
423 		if (probe_id != 0)
424 			(dtrace_malloc_probe)(probe_id,
425 			    (uintptr_t) mtp, (uintptr_t) mtip,
426 			    (uintptr_t) mtsp, size, 0);
427 	}
428 #endif
429 
430 	critical_exit();
431 }
432 
433 /*
434  *	contigmalloc:
435  *
436  *	Allocate a block of physically contiguous memory.
437  *
438  *	If M_NOWAIT is set, this routine will not block and return NULL if
439  *	the allocation fails.
440  */
441 void *
442 contigmalloc(unsigned long size, struct malloc_type *type, int flags,
443     vm_paddr_t low, vm_paddr_t high, unsigned long alignment,
444     vm_paddr_t boundary)
445 {
446 	void *ret;
447 
448 	ret = (void *)kmem_alloc_contig(size, flags, low, high, alignment,
449 	    boundary, VM_MEMATTR_DEFAULT);
450 	if (ret != NULL)
451 		malloc_type_allocated(type, round_page(size));
452 	return (ret);
453 }
454 
455 void *
456 contigmalloc_domain(unsigned long size, struct malloc_type *type,
457     int domain, int flags, vm_paddr_t low, vm_paddr_t high,
458     unsigned long alignment, vm_paddr_t boundary)
459 {
460 	void *ret;
461 
462 	ret = (void *)kmem_alloc_contig_domain(domain, size, flags, low, high,
463 	    alignment, boundary, VM_MEMATTR_DEFAULT);
464 	if (ret != NULL)
465 		malloc_type_allocated(type, round_page(size));
466 	return (ret);
467 }
468 
469 /*
470  *	contigfree:
471  *
472  *	Free a block of memory allocated by contigmalloc.
473  *
474  *	This routine may not block.
475  */
476 void
477 contigfree(void *addr, unsigned long size, struct malloc_type *type)
478 {
479 
480 	kmem_free((vm_offset_t)addr, size);
481 	malloc_type_freed(type, round_page(size));
482 }
483 
484 #ifdef MALLOC_DEBUG
485 static int
486 malloc_dbg(caddr_t *vap, size_t *sizep, struct malloc_type *mtp,
487     int flags)
488 {
489 #ifdef INVARIANTS
490 	int indx;
491 
492 	KASSERT(mtp->ks_magic == M_MAGIC, ("malloc: bad malloc type magic"));
493 	/*
494 	 * Check that exactly one of M_WAITOK or M_NOWAIT is specified.
495 	 */
496 	indx = flags & (M_WAITOK | M_NOWAIT);
497 	if (indx != M_NOWAIT && indx != M_WAITOK) {
498 		static	struct timeval lasterr;
499 		static	int curerr, once;
500 		if (once == 0 && ppsratecheck(&lasterr, &curerr, 1)) {
501 			printf("Bad malloc flags: %x\n", indx);
502 			kdb_backtrace();
503 			flags |= M_WAITOK;
504 			once++;
505 		}
506 	}
507 #endif
508 #ifdef MALLOC_MAKE_FAILURES
509 	if ((flags & M_NOWAIT) && (malloc_failure_rate != 0)) {
510 		atomic_add_int(&malloc_nowait_count, 1);
511 		if ((malloc_nowait_count % malloc_failure_rate) == 0) {
512 			atomic_add_int(&malloc_failure_count, 1);
513 			t_malloc_fail = time_uptime;
514 			*vap = NULL;
515 			return (EJUSTRETURN);
516 		}
517 	}
518 #endif
519 	if (flags & M_WAITOK) {
520 		KASSERT(curthread->td_intr_nesting_level == 0,
521 		   ("malloc(M_WAITOK) in interrupt context"));
522 		KASSERT(curthread->td_epochnest == 0,
523 			("malloc(M_WAITOK) in epoch context"));
524 	}
525 	KASSERT(curthread->td_critnest == 0 || SCHEDULER_STOPPED(),
526 	    ("malloc: called with spinlock or critical section held"));
527 
528 #ifdef DEBUG_MEMGUARD
529 	if (memguard_cmp_mtp(mtp, *sizep)) {
530 		*vap = memguard_alloc(*sizep, flags);
531 		if (*vap != NULL)
532 			return (EJUSTRETURN);
533 		/* This is unfortunate but should not be fatal. */
534 	}
535 #endif
536 
537 #ifdef DEBUG_REDZONE
538 	*sizep = redzone_size_ntor(*sizep);
539 #endif
540 
541 	return (0);
542 }
543 #endif
544 
545 /*
546  *	malloc:
547  *
548  *	Allocate a block of memory.
549  *
550  *	If M_NOWAIT is set, this routine will not block and return NULL if
551  *	the allocation fails.
552  */
553 void *
554 (malloc)(size_t size, struct malloc_type *mtp, int flags)
555 {
556 	int indx;
557 	caddr_t va;
558 	uma_zone_t zone;
559 #if defined(DEBUG_REDZONE)
560 	unsigned long osize = size;
561 #endif
562 
563 #ifdef MALLOC_DEBUG
564 	va = NULL;
565 	if (malloc_dbg(&va, &size, mtp, flags) != 0)
566 		return (va);
567 #endif
568 
569 	if (size <= kmem_zmax && (flags & M_EXEC) == 0) {
570 		if (size & KMEM_ZMASK)
571 			size = (size & ~KMEM_ZMASK) + KMEM_ZBASE;
572 		indx = kmemsize[size >> KMEM_ZSHIFT];
573 		zone = kmemzones[indx].kz_zone[mtp_get_subzone(mtp)];
574 #ifdef MALLOC_PROFILE
575 		krequests[size >> KMEM_ZSHIFT]++;
576 #endif
577 		va = uma_zalloc(zone, flags);
578 		if (va != NULL)
579 			size = zone->uz_size;
580 		malloc_type_zone_allocated(mtp, va == NULL ? 0 : size, indx);
581 	} else {
582 		size = roundup(size, PAGE_SIZE);
583 		zone = NULL;
584 		va = uma_large_malloc(size, flags);
585 		malloc_type_allocated(mtp, va == NULL ? 0 : size);
586 	}
587 	if (flags & M_WAITOK)
588 		KASSERT(va != NULL, ("malloc(M_WAITOK) returned NULL"));
589 	else if (va == NULL)
590 		t_malloc_fail = time_uptime;
591 #ifdef DEBUG_REDZONE
592 	if (va != NULL)
593 		va = redzone_setup(va, osize);
594 #endif
595 	return ((void *) va);
596 }
597 
598 void *
599 malloc_domain(size_t size, struct malloc_type *mtp, int domain,
600     int flags)
601 {
602 	int indx;
603 	caddr_t va;
604 	uma_zone_t zone;
605 #if defined(DEBUG_REDZONE)
606 	unsigned long osize = size;
607 #endif
608 
609 #ifdef MALLOC_DEBUG
610 	va = NULL;
611 	if (malloc_dbg(&va, &size, mtp, flags) != 0)
612 		return (va);
613 #endif
614 	if (size <= kmem_zmax && (flags & M_EXEC) == 0) {
615 		if (size & KMEM_ZMASK)
616 			size = (size & ~KMEM_ZMASK) + KMEM_ZBASE;
617 		indx = kmemsize[size >> KMEM_ZSHIFT];
618 		zone = kmemzones[indx].kz_zone[mtp_get_subzone(mtp)];
619 #ifdef MALLOC_PROFILE
620 		krequests[size >> KMEM_ZSHIFT]++;
621 #endif
622 		va = uma_zalloc_domain(zone, NULL, domain, flags);
623 		if (va != NULL)
624 			size = zone->uz_size;
625 		malloc_type_zone_allocated(mtp, va == NULL ? 0 : size, indx);
626 	} else {
627 		size = roundup(size, PAGE_SIZE);
628 		zone = NULL;
629 		va = uma_large_malloc_domain(size, domain, flags);
630 		malloc_type_allocated(mtp, va == NULL ? 0 : size);
631 	}
632 	if (flags & M_WAITOK)
633 		KASSERT(va != NULL, ("malloc(M_WAITOK) returned NULL"));
634 	else if (va == NULL)
635 		t_malloc_fail = time_uptime;
636 #ifdef DEBUG_REDZONE
637 	if (va != NULL)
638 		va = redzone_setup(va, osize);
639 #endif
640 	return ((void *) va);
641 }
642 
643 void *
644 mallocarray(size_t nmemb, size_t size, struct malloc_type *type, int flags)
645 {
646 
647 	if (WOULD_OVERFLOW(nmemb, size))
648 		panic("mallocarray: %zu * %zu overflowed", nmemb, size);
649 
650 	return (malloc(size * nmemb, type, flags));
651 }
652 
653 #ifdef INVARIANTS
654 static void
655 free_save_type(void *addr, struct malloc_type *mtp, u_long size)
656 {
657 	struct malloc_type **mtpp = addr;
658 
659 	/*
660 	 * Cache a pointer to the malloc_type that most recently freed
661 	 * this memory here.  This way we know who is most likely to
662 	 * have stepped on it later.
663 	 *
664 	 * This code assumes that size is a multiple of 8 bytes for
665 	 * 64 bit machines
666 	 */
667 	mtpp = (struct malloc_type **) ((unsigned long)mtpp & ~UMA_ALIGN_PTR);
668 	mtpp += (size - sizeof(struct malloc_type *)) /
669 	    sizeof(struct malloc_type *);
670 	*mtpp = mtp;
671 }
672 #endif
673 
674 #ifdef MALLOC_DEBUG
675 static int
676 free_dbg(void **addrp, struct malloc_type *mtp)
677 {
678 	void *addr;
679 
680 	addr = *addrp;
681 	KASSERT(mtp->ks_magic == M_MAGIC, ("free: bad malloc type magic"));
682 	KASSERT(curthread->td_critnest == 0 || SCHEDULER_STOPPED(),
683 	    ("free: called with spinlock or critical section held"));
684 
685 	/* free(NULL, ...) does nothing */
686 	if (addr == NULL)
687 		return (EJUSTRETURN);
688 
689 #ifdef DEBUG_MEMGUARD
690 	if (is_memguard_addr(addr)) {
691 		memguard_free(addr);
692 		return (EJUSTRETURN);
693 	}
694 #endif
695 
696 #ifdef DEBUG_REDZONE
697 	redzone_check(addr);
698 	*addrp = redzone_addr_ntor(addr);
699 #endif
700 
701 	return (0);
702 }
703 #endif
704 
705 /*
706  *	free:
707  *
708  *	Free a block of memory allocated by malloc.
709  *
710  *	This routine may not block.
711  */
712 void
713 free(void *addr, struct malloc_type *mtp)
714 {
715 	uma_slab_t slab;
716 	u_long size;
717 
718 #ifdef MALLOC_DEBUG
719 	if (free_dbg(&addr, mtp) != 0)
720 		return;
721 #endif
722 	/* free(NULL, ...) does nothing */
723 	if (addr == NULL)
724 		return;
725 
726 	slab = vtoslab((vm_offset_t)addr & (~UMA_SLAB_MASK));
727 	if (slab == NULL)
728 		panic("free: address %p(%p) has not been allocated.\n",
729 		    addr, (void *)((u_long)addr & (~UMA_SLAB_MASK)));
730 
731 	if (!(slab->us_flags & UMA_SLAB_MALLOC)) {
732 		size = slab->us_keg->uk_size;
733 #ifdef INVARIANTS
734 		free_save_type(addr, mtp, size);
735 #endif
736 		uma_zfree_arg(LIST_FIRST(&slab->us_keg->uk_zones), addr, slab);
737 	} else {
738 		size = slab->us_size;
739 		uma_large_free(slab);
740 	}
741 	malloc_type_freed(mtp, size);
742 }
743 
744 void
745 free_domain(void *addr, struct malloc_type *mtp)
746 {
747 	uma_slab_t slab;
748 	u_long size;
749 
750 #ifdef MALLOC_DEBUG
751 	if (free_dbg(&addr, mtp) != 0)
752 		return;
753 #endif
754 
755 	/* free(NULL, ...) does nothing */
756 	if (addr == NULL)
757 		return;
758 
759 	slab = vtoslab((vm_offset_t)addr & (~UMA_SLAB_MASK));
760 	if (slab == NULL)
761 		panic("free_domain: address %p(%p) has not been allocated.\n",
762 		    addr, (void *)((u_long)addr & (~UMA_SLAB_MASK)));
763 
764 	if (!(slab->us_flags & UMA_SLAB_MALLOC)) {
765 		size = slab->us_keg->uk_size;
766 #ifdef INVARIANTS
767 		free_save_type(addr, mtp, size);
768 #endif
769 		uma_zfree_domain(LIST_FIRST(&slab->us_keg->uk_zones),
770 		    addr, slab);
771 	} else {
772 		size = slab->us_size;
773 		uma_large_free(slab);
774 	}
775 	malloc_type_freed(mtp, size);
776 }
777 
778 /*
779  *	realloc: change the size of a memory block
780  */
781 void *
782 realloc(void *addr, size_t size, struct malloc_type *mtp, int flags)
783 {
784 	uma_slab_t slab;
785 	unsigned long alloc;
786 	void *newaddr;
787 
788 	KASSERT(mtp->ks_magic == M_MAGIC,
789 	    ("realloc: bad malloc type magic"));
790 	KASSERT(curthread->td_critnest == 0 || SCHEDULER_STOPPED(),
791 	    ("realloc: called with spinlock or critical section held"));
792 
793 	/* realloc(NULL, ...) is equivalent to malloc(...) */
794 	if (addr == NULL)
795 		return (malloc(size, mtp, flags));
796 
797 	/*
798 	 * XXX: Should report free of old memory and alloc of new memory to
799 	 * per-CPU stats.
800 	 */
801 
802 #ifdef DEBUG_MEMGUARD
803 	if (is_memguard_addr(addr))
804 		return (memguard_realloc(addr, size, mtp, flags));
805 #endif
806 
807 #ifdef DEBUG_REDZONE
808 	slab = NULL;
809 	alloc = redzone_get_size(addr);
810 #else
811 	slab = vtoslab((vm_offset_t)addr & ~(UMA_SLAB_MASK));
812 
813 	/* Sanity check */
814 	KASSERT(slab != NULL,
815 	    ("realloc: address %p out of range", (void *)addr));
816 
817 	/* Get the size of the original block */
818 	if (!(slab->us_flags & UMA_SLAB_MALLOC))
819 		alloc = slab->us_keg->uk_size;
820 	else
821 		alloc = slab->us_size;
822 
823 	/* Reuse the original block if appropriate */
824 	if (size <= alloc
825 	    && (size > (alloc >> REALLOC_FRACTION) || alloc == MINALLOCSIZE))
826 		return (addr);
827 #endif /* !DEBUG_REDZONE */
828 
829 	/* Allocate a new, bigger (or smaller) block */
830 	if ((newaddr = malloc(size, mtp, flags)) == NULL)
831 		return (NULL);
832 
833 	/* Copy over original contents */
834 	bcopy(addr, newaddr, min(size, alloc));
835 	free(addr, mtp);
836 	return (newaddr);
837 }
838 
839 /*
840  *	reallocf: same as realloc() but free memory on failure.
841  */
842 void *
843 reallocf(void *addr, size_t size, struct malloc_type *mtp, int flags)
844 {
845 	void *mem;
846 
847 	if ((mem = realloc(addr, size, mtp, flags)) == NULL)
848 		free(addr, mtp);
849 	return (mem);
850 }
851 
852 #ifndef __sparc64__
853 CTASSERT(VM_KMEM_SIZE_SCALE >= 1);
854 #endif
855 
856 /*
857  * Initialize the kernel memory (kmem) arena.
858  */
859 void
860 kmeminit(void)
861 {
862 	u_long mem_size;
863 	u_long tmp;
864 
865 #ifdef VM_KMEM_SIZE
866 	if (vm_kmem_size == 0)
867 		vm_kmem_size = VM_KMEM_SIZE;
868 #endif
869 #ifdef VM_KMEM_SIZE_MIN
870 	if (vm_kmem_size_min == 0)
871 		vm_kmem_size_min = VM_KMEM_SIZE_MIN;
872 #endif
873 #ifdef VM_KMEM_SIZE_MAX
874 	if (vm_kmem_size_max == 0)
875 		vm_kmem_size_max = VM_KMEM_SIZE_MAX;
876 #endif
877 	/*
878 	 * Calculate the amount of kernel virtual address (KVA) space that is
879 	 * preallocated to the kmem arena.  In order to support a wide range
880 	 * of machines, it is a function of the physical memory size,
881 	 * specifically,
882 	 *
883 	 *	min(max(physical memory size / VM_KMEM_SIZE_SCALE,
884 	 *	    VM_KMEM_SIZE_MIN), VM_KMEM_SIZE_MAX)
885 	 *
886 	 * Every architecture must define an integral value for
887 	 * VM_KMEM_SIZE_SCALE.  However, the definitions of VM_KMEM_SIZE_MIN
888 	 * and VM_KMEM_SIZE_MAX, which represent respectively the floor and
889 	 * ceiling on this preallocation, are optional.  Typically,
890 	 * VM_KMEM_SIZE_MAX is itself a function of the available KVA space on
891 	 * a given architecture.
892 	 */
893 	mem_size = vm_cnt.v_page_count;
894 	if (mem_size <= 32768) /* delphij XXX 128MB */
895 		kmem_zmax = PAGE_SIZE;
896 
897 	if (vm_kmem_size_scale < 1)
898 		vm_kmem_size_scale = VM_KMEM_SIZE_SCALE;
899 
900 	/*
901 	 * Check if we should use defaults for the "vm_kmem_size"
902 	 * variable:
903 	 */
904 	if (vm_kmem_size == 0) {
905 		vm_kmem_size = (mem_size / vm_kmem_size_scale) * PAGE_SIZE;
906 
907 		if (vm_kmem_size_min > 0 && vm_kmem_size < vm_kmem_size_min)
908 			vm_kmem_size = vm_kmem_size_min;
909 		if (vm_kmem_size_max > 0 && vm_kmem_size >= vm_kmem_size_max)
910 			vm_kmem_size = vm_kmem_size_max;
911 	}
912 
913 	/*
914 	 * The amount of KVA space that is preallocated to the
915 	 * kmem arena can be set statically at compile-time or manually
916 	 * through the kernel environment.  However, it is still limited to
917 	 * twice the physical memory size, which has been sufficient to handle
918 	 * the most severe cases of external fragmentation in the kmem arena.
919 	 */
920 	if (vm_kmem_size / 2 / PAGE_SIZE > mem_size)
921 		vm_kmem_size = 2 * mem_size * PAGE_SIZE;
922 
923 	vm_kmem_size = round_page(vm_kmem_size);
924 #ifdef DEBUG_MEMGUARD
925 	tmp = memguard_fudge(vm_kmem_size, kernel_map);
926 #else
927 	tmp = vm_kmem_size;
928 #endif
929 	uma_set_limit(tmp);
930 
931 #ifdef DEBUG_MEMGUARD
932 	/*
933 	 * Initialize MemGuard if support compiled in.  MemGuard is a
934 	 * replacement allocator used for detecting tamper-after-free
935 	 * scenarios as they occur.  It is only used for debugging.
936 	 */
937 	memguard_init(kernel_arena);
938 #endif
939 }
940 
941 /*
942  * Initialize the kernel memory allocator
943  */
944 /* ARGSUSED*/
945 static void
946 mallocinit(void *dummy)
947 {
948 	int i;
949 	uint8_t indx;
950 
951 	mtx_init(&malloc_mtx, "malloc", NULL, MTX_DEF);
952 
953 	kmeminit();
954 
955 	if (kmem_zmax < PAGE_SIZE || kmem_zmax > KMEM_ZMAX)
956 		kmem_zmax = KMEM_ZMAX;
957 
958 	mt_stats_zone = uma_zcreate("mt_stats_zone",
959 	    sizeof(struct malloc_type_stats), NULL, NULL, NULL, NULL,
960 	    UMA_ALIGN_PTR, UMA_ZONE_PCPU);
961 	mt_zone = uma_zcreate("mt_zone", sizeof(struct malloc_type_internal),
962 #ifdef INVARIANTS
963 	    mtrash_ctor, mtrash_dtor, mtrash_init, mtrash_fini,
964 #else
965 	    NULL, NULL, NULL, NULL,
966 #endif
967 	    UMA_ALIGN_PTR, UMA_ZONE_MALLOC);
968 	for (i = 0, indx = 0; kmemzones[indx].kz_size != 0; indx++) {
969 		int size = kmemzones[indx].kz_size;
970 		char *name = kmemzones[indx].kz_name;
971 		int subzone;
972 
973 		for (subzone = 0; subzone < numzones; subzone++) {
974 			kmemzones[indx].kz_zone[subzone] =
975 			    uma_zcreate(name, size,
976 #ifdef INVARIANTS
977 			    mtrash_ctor, mtrash_dtor, mtrash_init, mtrash_fini,
978 #else
979 			    NULL, NULL, NULL, NULL,
980 #endif
981 			    UMA_ALIGN_PTR, UMA_ZONE_MALLOC);
982 		}
983 		for (;i <= size; i+= KMEM_ZBASE)
984 			kmemsize[i >> KMEM_ZSHIFT] = indx;
985 
986 	}
987 }
988 SYSINIT(kmem, SI_SUB_KMEM, SI_ORDER_SECOND, mallocinit, NULL);
989 
990 void
991 malloc_init(void *data)
992 {
993 	struct malloc_type_internal *mtip;
994 	struct malloc_type *mtp;
995 
996 	KASSERT(vm_cnt.v_page_count != 0, ("malloc_register before vm_init"));
997 
998 	mtp = data;
999 	if (mtp->ks_magic != M_MAGIC)
1000 		panic("malloc_init: bad malloc type magic");
1001 
1002 	mtip = uma_zalloc(mt_zone, M_WAITOK | M_ZERO);
1003 	mtip->mti_stats = uma_zalloc_pcpu(mt_stats_zone, M_WAITOK | M_ZERO);
1004 	mtp->ks_handle = mtip;
1005 	mtp_set_subzone(mtp);
1006 
1007 	mtx_lock(&malloc_mtx);
1008 	mtp->ks_next = kmemstatistics;
1009 	kmemstatistics = mtp;
1010 	kmemcount++;
1011 	mtx_unlock(&malloc_mtx);
1012 }
1013 
1014 void
1015 malloc_uninit(void *data)
1016 {
1017 	struct malloc_type_internal *mtip;
1018 	struct malloc_type_stats *mtsp;
1019 	struct malloc_type *mtp, *temp;
1020 	uma_slab_t slab;
1021 	long temp_allocs, temp_bytes;
1022 	int i;
1023 
1024 	mtp = data;
1025 	KASSERT(mtp->ks_magic == M_MAGIC,
1026 	    ("malloc_uninit: bad malloc type magic"));
1027 	KASSERT(mtp->ks_handle != NULL, ("malloc_deregister: cookie NULL"));
1028 
1029 	mtx_lock(&malloc_mtx);
1030 	mtip = mtp->ks_handle;
1031 	mtp->ks_handle = NULL;
1032 	if (mtp != kmemstatistics) {
1033 		for (temp = kmemstatistics; temp != NULL;
1034 		    temp = temp->ks_next) {
1035 			if (temp->ks_next == mtp) {
1036 				temp->ks_next = mtp->ks_next;
1037 				break;
1038 			}
1039 		}
1040 		KASSERT(temp,
1041 		    ("malloc_uninit: type '%s' not found", mtp->ks_shortdesc));
1042 	} else
1043 		kmemstatistics = mtp->ks_next;
1044 	kmemcount--;
1045 	mtx_unlock(&malloc_mtx);
1046 
1047 	/*
1048 	 * Look for memory leaks.
1049 	 */
1050 	temp_allocs = temp_bytes = 0;
1051 	for (i = 0; i <= mp_maxid; i++) {
1052 		mtsp = zpcpu_get_cpu(mtip->mti_stats, i);
1053 		temp_allocs += mtsp->mts_numallocs;
1054 		temp_allocs -= mtsp->mts_numfrees;
1055 		temp_bytes += mtsp->mts_memalloced;
1056 		temp_bytes -= mtsp->mts_memfreed;
1057 	}
1058 	if (temp_allocs > 0 || temp_bytes > 0) {
1059 		printf("Warning: memory type %s leaked memory on destroy "
1060 		    "(%ld allocations, %ld bytes leaked).\n", mtp->ks_shortdesc,
1061 		    temp_allocs, temp_bytes);
1062 	}
1063 
1064 	slab = vtoslab((vm_offset_t) mtip & (~UMA_SLAB_MASK));
1065 	uma_zfree_pcpu(mt_stats_zone, mtip->mti_stats);
1066 	uma_zfree_arg(mt_zone, mtip, slab);
1067 }
1068 
1069 struct malloc_type *
1070 malloc_desc2type(const char *desc)
1071 {
1072 	struct malloc_type *mtp;
1073 
1074 	mtx_assert(&malloc_mtx, MA_OWNED);
1075 	for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) {
1076 		if (strcmp(mtp->ks_shortdesc, desc) == 0)
1077 			return (mtp);
1078 	}
1079 	return (NULL);
1080 }
1081 
1082 static int
1083 sysctl_kern_malloc_stats(SYSCTL_HANDLER_ARGS)
1084 {
1085 	struct malloc_type_stream_header mtsh;
1086 	struct malloc_type_internal *mtip;
1087 	struct malloc_type_stats *mtsp, zeromts;
1088 	struct malloc_type_header mth;
1089 	struct malloc_type *mtp;
1090 	int error, i;
1091 	struct sbuf sbuf;
1092 
1093 	error = sysctl_wire_old_buffer(req, 0);
1094 	if (error != 0)
1095 		return (error);
1096 	sbuf_new_for_sysctl(&sbuf, NULL, 128, req);
1097 	sbuf_clear_flags(&sbuf, SBUF_INCLUDENUL);
1098 	mtx_lock(&malloc_mtx);
1099 
1100 	bzero(&zeromts, sizeof(zeromts));
1101 
1102 	/*
1103 	 * Insert stream header.
1104 	 */
1105 	bzero(&mtsh, sizeof(mtsh));
1106 	mtsh.mtsh_version = MALLOC_TYPE_STREAM_VERSION;
1107 	mtsh.mtsh_maxcpus = MAXCPU;
1108 	mtsh.mtsh_count = kmemcount;
1109 	(void)sbuf_bcat(&sbuf, &mtsh, sizeof(mtsh));
1110 
1111 	/*
1112 	 * Insert alternating sequence of type headers and type statistics.
1113 	 */
1114 	for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) {
1115 		mtip = (struct malloc_type_internal *)mtp->ks_handle;
1116 
1117 		/*
1118 		 * Insert type header.
1119 		 */
1120 		bzero(&mth, sizeof(mth));
1121 		strlcpy(mth.mth_name, mtp->ks_shortdesc, MALLOC_MAX_NAME);
1122 		(void)sbuf_bcat(&sbuf, &mth, sizeof(mth));
1123 
1124 		/*
1125 		 * Insert type statistics for each CPU.
1126 		 */
1127 		for (i = 0; i <= mp_maxid; i++) {
1128 			mtsp = zpcpu_get_cpu(mtip->mti_stats, i);
1129 			(void)sbuf_bcat(&sbuf, mtsp, sizeof(*mtsp));
1130 		}
1131 		/*
1132 		 * Fill in the missing CPUs.
1133 		 */
1134 		for (; i < MAXCPU; i++) {
1135 			(void)sbuf_bcat(&sbuf, &zeromts, sizeof(zeromts));
1136 		}
1137 
1138 	}
1139 	mtx_unlock(&malloc_mtx);
1140 	error = sbuf_finish(&sbuf);
1141 	sbuf_delete(&sbuf);
1142 	return (error);
1143 }
1144 
1145 SYSCTL_PROC(_kern, OID_AUTO, malloc_stats, CTLFLAG_RD|CTLTYPE_STRUCT,
1146     0, 0, sysctl_kern_malloc_stats, "s,malloc_type_ustats",
1147     "Return malloc types");
1148 
1149 SYSCTL_INT(_kern, OID_AUTO, malloc_count, CTLFLAG_RD, &kmemcount, 0,
1150     "Count of kernel malloc types");
1151 
1152 void
1153 malloc_type_list(malloc_type_list_func_t *func, void *arg)
1154 {
1155 	struct malloc_type *mtp, **bufmtp;
1156 	int count, i;
1157 	size_t buflen;
1158 
1159 	mtx_lock(&malloc_mtx);
1160 restart:
1161 	mtx_assert(&malloc_mtx, MA_OWNED);
1162 	count = kmemcount;
1163 	mtx_unlock(&malloc_mtx);
1164 
1165 	buflen = sizeof(struct malloc_type *) * count;
1166 	bufmtp = malloc(buflen, M_TEMP, M_WAITOK);
1167 
1168 	mtx_lock(&malloc_mtx);
1169 
1170 	if (count < kmemcount) {
1171 		free(bufmtp, M_TEMP);
1172 		goto restart;
1173 	}
1174 
1175 	for (mtp = kmemstatistics, i = 0; mtp != NULL; mtp = mtp->ks_next, i++)
1176 		bufmtp[i] = mtp;
1177 
1178 	mtx_unlock(&malloc_mtx);
1179 
1180 	for (i = 0; i < count; i++)
1181 		(func)(bufmtp[i], arg);
1182 
1183 	free(bufmtp, M_TEMP);
1184 }
1185 
1186 #ifdef DDB
1187 DB_SHOW_COMMAND(malloc, db_show_malloc)
1188 {
1189 	struct malloc_type_internal *mtip;
1190 	struct malloc_type_internal *mtsp;
1191 	struct malloc_type *mtp;
1192 	uint64_t allocs, frees;
1193 	uint64_t alloced, freed;
1194 	int i;
1195 
1196 	db_printf("%18s %12s  %12s %12s\n", "Type", "InUse", "MemUse",
1197 	    "Requests");
1198 	for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) {
1199 		mtip = (struct malloc_type_internal *)mtp->ks_handle;
1200 		allocs = 0;
1201 		frees = 0;
1202 		alloced = 0;
1203 		freed = 0;
1204 		for (i = 0; i <= mp_maxid; i++) {
1205 			mtsp = zpcpu_get_cpu(mtip->mti_stats, i);
1206 			allocs += mtip->mti_stats[i].mts_numallocs;
1207 			frees += mtip->mti_stats[i].mts_numfrees;
1208 			alloced += mtip->mti_stats[i].mts_memalloced;
1209 			freed += mtip->mti_stats[i].mts_memfreed;
1210 		}
1211 		db_printf("%18s %12ju %12juK %12ju\n",
1212 		    mtp->ks_shortdesc, allocs - frees,
1213 		    (alloced - freed + 1023) / 1024, allocs);
1214 		if (db_pager_quit)
1215 			break;
1216 	}
1217 }
1218 
1219 #if MALLOC_DEBUG_MAXZONES > 1
1220 DB_SHOW_COMMAND(multizone_matches, db_show_multizone_matches)
1221 {
1222 	struct malloc_type_internal *mtip;
1223 	struct malloc_type *mtp;
1224 	u_int subzone;
1225 
1226 	if (!have_addr) {
1227 		db_printf("Usage: show multizone_matches <malloc type/addr>\n");
1228 		return;
1229 	}
1230 	mtp = (void *)addr;
1231 	if (mtp->ks_magic != M_MAGIC) {
1232 		db_printf("Magic %lx does not match expected %x\n",
1233 		    mtp->ks_magic, M_MAGIC);
1234 		return;
1235 	}
1236 
1237 	mtip = mtp->ks_handle;
1238 	subzone = mtip->mti_zone;
1239 
1240 	for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) {
1241 		mtip = mtp->ks_handle;
1242 		if (mtip->mti_zone != subzone)
1243 			continue;
1244 		db_printf("%s\n", mtp->ks_shortdesc);
1245 		if (db_pager_quit)
1246 			break;
1247 	}
1248 }
1249 #endif /* MALLOC_DEBUG_MAXZONES > 1 */
1250 #endif /* DDB */
1251 
1252 #ifdef MALLOC_PROFILE
1253 
1254 static int
1255 sysctl_kern_mprof(SYSCTL_HANDLER_ARGS)
1256 {
1257 	struct sbuf sbuf;
1258 	uint64_t count;
1259 	uint64_t waste;
1260 	uint64_t mem;
1261 	int error;
1262 	int rsize;
1263 	int size;
1264 	int i;
1265 
1266 	waste = 0;
1267 	mem = 0;
1268 
1269 	error = sysctl_wire_old_buffer(req, 0);
1270 	if (error != 0)
1271 		return (error);
1272 	sbuf_new_for_sysctl(&sbuf, NULL, 128, req);
1273 	sbuf_printf(&sbuf,
1274 	    "\n  Size                    Requests  Real Size\n");
1275 	for (i = 0; i < KMEM_ZSIZE; i++) {
1276 		size = i << KMEM_ZSHIFT;
1277 		rsize = kmemzones[kmemsize[i]].kz_size;
1278 		count = (long long unsigned)krequests[i];
1279 
1280 		sbuf_printf(&sbuf, "%6d%28llu%11d\n", size,
1281 		    (unsigned long long)count, rsize);
1282 
1283 		if ((rsize * count) > (size * count))
1284 			waste += (rsize * count) - (size * count);
1285 		mem += (rsize * count);
1286 	}
1287 	sbuf_printf(&sbuf,
1288 	    "\nTotal memory used:\t%30llu\nTotal Memory wasted:\t%30llu\n",
1289 	    (unsigned long long)mem, (unsigned long long)waste);
1290 	error = sbuf_finish(&sbuf);
1291 	sbuf_delete(&sbuf);
1292 	return (error);
1293 }
1294 
1295 SYSCTL_OID(_kern, OID_AUTO, mprof, CTLTYPE_STRING|CTLFLAG_RD,
1296     NULL, 0, sysctl_kern_mprof, "A", "Malloc Profiling");
1297 #endif /* MALLOC_PROFILE */
1298