xref: /freebsd/sys/kern/kern_malloc.c (revision 68d75eff68281c1b445e3010bb975eae07aac225)
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/queue.h>
64 #include <sys/sbuf.h>
65 #include <sys/smp.h>
66 #include <sys/sysctl.h>
67 #include <sys/time.h>
68 #include <sys/vmem.h>
69 #ifdef EPOCH_TRACE
70 #include <sys/epoch.h>
71 #endif
72 
73 #include <vm/vm.h>
74 #include <vm/pmap.h>
75 #include <vm/vm_domainset.h>
76 #include <vm/vm_pageout.h>
77 #include <vm/vm_param.h>
78 #include <vm/vm_kern.h>
79 #include <vm/vm_extern.h>
80 #include <vm/vm_map.h>
81 #include <vm/vm_page.h>
82 #include <vm/vm_phys.h>
83 #include <vm/vm_pagequeue.h>
84 #include <vm/uma.h>
85 #include <vm/uma_int.h>
86 #include <vm/uma_dbg.h>
87 
88 #ifdef DEBUG_MEMGUARD
89 #include <vm/memguard.h>
90 #endif
91 #ifdef DEBUG_REDZONE
92 #include <vm/redzone.h>
93 #endif
94 
95 #if defined(INVARIANTS) && defined(__i386__)
96 #include <machine/cpu.h>
97 #endif
98 
99 #include <ddb/ddb.h>
100 
101 #ifdef KDTRACE_HOOKS
102 #include <sys/dtrace_bsd.h>
103 
104 bool	__read_frequently			dtrace_malloc_enabled;
105 dtrace_malloc_probe_func_t __read_mostly	dtrace_malloc_probe;
106 #endif
107 
108 #if defined(INVARIANTS) || defined(MALLOC_MAKE_FAILURES) ||		\
109     defined(DEBUG_MEMGUARD) || defined(DEBUG_REDZONE)
110 #define	MALLOC_DEBUG	1
111 #endif
112 
113 /*
114  * When realloc() is called, if the new size is sufficiently smaller than
115  * the old size, realloc() will allocate a new, smaller block to avoid
116  * wasting memory. 'Sufficiently smaller' is defined as: newsize <=
117  * oldsize / 2^n, where REALLOC_FRACTION defines the value of 'n'.
118  */
119 #ifndef REALLOC_FRACTION
120 #define	REALLOC_FRACTION	1	/* new block if <= half the size */
121 #endif
122 
123 /*
124  * Centrally define some common malloc types.
125  */
126 MALLOC_DEFINE(M_CACHE, "cache", "Various Dynamically allocated caches");
127 MALLOC_DEFINE(M_DEVBUF, "devbuf", "device driver memory");
128 MALLOC_DEFINE(M_TEMP, "temp", "misc temporary data buffers");
129 
130 static struct malloc_type *kmemstatistics;
131 static int kmemcount;
132 
133 #define KMEM_ZSHIFT	4
134 #define KMEM_ZBASE	16
135 #define KMEM_ZMASK	(KMEM_ZBASE - 1)
136 
137 #define KMEM_ZMAX	65536
138 #define KMEM_ZSIZE	(KMEM_ZMAX >> KMEM_ZSHIFT)
139 static uint8_t kmemsize[KMEM_ZSIZE + 1];
140 
141 #ifndef MALLOC_DEBUG_MAXZONES
142 #define	MALLOC_DEBUG_MAXZONES	1
143 #endif
144 static int numzones = MALLOC_DEBUG_MAXZONES;
145 
146 /*
147  * Small malloc(9) memory allocations are allocated from a set of UMA buckets
148  * of various sizes.
149  *
150  * XXX: The comment here used to read "These won't be powers of two for
151  * long."  It's possible that a significant amount of wasted memory could be
152  * recovered by tuning the sizes of these buckets.
153  */
154 struct {
155 	int kz_size;
156 	char *kz_name;
157 	uma_zone_t kz_zone[MALLOC_DEBUG_MAXZONES];
158 } kmemzones[] = {
159 	{16, "16", },
160 	{32, "32", },
161 	{64, "64", },
162 	{128, "128", },
163 	{256, "256", },
164 	{512, "512", },
165 	{1024, "1024", },
166 	{2048, "2048", },
167 	{4096, "4096", },
168 	{8192, "8192", },
169 	{16384, "16384", },
170 	{32768, "32768", },
171 	{65536, "65536", },
172 	{0, NULL},
173 };
174 
175 /*
176  * Zone to allocate malloc type descriptions from.  For ABI reasons, memory
177  * types are described by a data structure passed by the declaring code, but
178  * the malloc(9) implementation has its own data structure describing the
179  * type and statistics.  This permits the malloc(9)-internal data structures
180  * to be modified without breaking binary-compiled kernel modules that
181  * declare malloc types.
182  */
183 static uma_zone_t mt_zone;
184 static uma_zone_t mt_stats_zone;
185 
186 u_long vm_kmem_size;
187 SYSCTL_ULONG(_vm, OID_AUTO, kmem_size, CTLFLAG_RDTUN, &vm_kmem_size, 0,
188     "Size of kernel memory");
189 
190 static u_long kmem_zmax = KMEM_ZMAX;
191 SYSCTL_ULONG(_vm, OID_AUTO, kmem_zmax, CTLFLAG_RDTUN, &kmem_zmax, 0,
192     "Maximum allocation size that malloc(9) would use UMA as backend");
193 
194 static u_long vm_kmem_size_min;
195 SYSCTL_ULONG(_vm, OID_AUTO, kmem_size_min, CTLFLAG_RDTUN, &vm_kmem_size_min, 0,
196     "Minimum size of kernel memory");
197 
198 static u_long vm_kmem_size_max;
199 SYSCTL_ULONG(_vm, OID_AUTO, kmem_size_max, CTLFLAG_RDTUN, &vm_kmem_size_max, 0,
200     "Maximum size of kernel memory");
201 
202 static u_int vm_kmem_size_scale;
203 SYSCTL_UINT(_vm, OID_AUTO, kmem_size_scale, CTLFLAG_RDTUN, &vm_kmem_size_scale, 0,
204     "Scale factor for kernel memory size");
205 
206 static int sysctl_kmem_map_size(SYSCTL_HANDLER_ARGS);
207 SYSCTL_PROC(_vm, OID_AUTO, kmem_map_size,
208     CTLFLAG_RD | CTLTYPE_ULONG | CTLFLAG_MPSAFE, NULL, 0,
209     sysctl_kmem_map_size, "LU", "Current kmem allocation size");
210 
211 static int sysctl_kmem_map_free(SYSCTL_HANDLER_ARGS);
212 SYSCTL_PROC(_vm, OID_AUTO, kmem_map_free,
213     CTLFLAG_RD | CTLTYPE_ULONG | CTLFLAG_MPSAFE, NULL, 0,
214     sysctl_kmem_map_free, "LU", "Free space in kmem");
215 
216 /*
217  * The malloc_mtx protects the kmemstatistics linked list.
218  */
219 struct mtx malloc_mtx;
220 
221 #ifdef MALLOC_PROFILE
222 uint64_t krequests[KMEM_ZSIZE + 1];
223 
224 static int sysctl_kern_mprof(SYSCTL_HANDLER_ARGS);
225 #endif
226 
227 static int sysctl_kern_malloc_stats(SYSCTL_HANDLER_ARGS);
228 
229 /*
230  * time_uptime of the last malloc(9) failure (induced or real).
231  */
232 static time_t t_malloc_fail;
233 
234 #if defined(MALLOC_MAKE_FAILURES) || (MALLOC_DEBUG_MAXZONES > 1)
235 static SYSCTL_NODE(_debug, OID_AUTO, malloc, CTLFLAG_RD, 0,
236     "Kernel malloc debugging options");
237 #endif
238 
239 /*
240  * malloc(9) fault injection -- cause malloc failures every (n) mallocs when
241  * the caller specifies M_NOWAIT.  If set to 0, no failures are caused.
242  */
243 #ifdef MALLOC_MAKE_FAILURES
244 static int malloc_failure_rate;
245 static int malloc_nowait_count;
246 static int malloc_failure_count;
247 SYSCTL_INT(_debug_malloc, OID_AUTO, failure_rate, CTLFLAG_RWTUN,
248     &malloc_failure_rate, 0, "Every (n) mallocs with M_NOWAIT will fail");
249 SYSCTL_INT(_debug_malloc, OID_AUTO, failure_count, CTLFLAG_RD,
250     &malloc_failure_count, 0, "Number of imposed M_NOWAIT malloc failures");
251 #endif
252 
253 static int
254 sysctl_kmem_map_size(SYSCTL_HANDLER_ARGS)
255 {
256 	u_long size;
257 
258 	size = uma_size();
259 	return (sysctl_handle_long(oidp, &size, 0, req));
260 }
261 
262 static int
263 sysctl_kmem_map_free(SYSCTL_HANDLER_ARGS)
264 {
265 	u_long size, limit;
266 
267 	/* The sysctl is unsigned, implement as a saturation value. */
268 	size = uma_size();
269 	limit = uma_limit();
270 	if (size > limit)
271 		size = 0;
272 	else
273 		size = limit - size;
274 	return (sysctl_handle_long(oidp, &size, 0, req));
275 }
276 
277 /*
278  * malloc(9) uma zone separation -- sub-page buffer overruns in one
279  * malloc type will affect only a subset of other malloc types.
280  */
281 #if MALLOC_DEBUG_MAXZONES > 1
282 static void
283 tunable_set_numzones(void)
284 {
285 
286 	TUNABLE_INT_FETCH("debug.malloc.numzones",
287 	    &numzones);
288 
289 	/* Sanity check the number of malloc uma zones. */
290 	if (numzones <= 0)
291 		numzones = 1;
292 	if (numzones > MALLOC_DEBUG_MAXZONES)
293 		numzones = MALLOC_DEBUG_MAXZONES;
294 }
295 SYSINIT(numzones, SI_SUB_TUNABLES, SI_ORDER_ANY, tunable_set_numzones, NULL);
296 SYSCTL_INT(_debug_malloc, OID_AUTO, numzones, CTLFLAG_RDTUN | CTLFLAG_NOFETCH,
297     &numzones, 0, "Number of malloc uma subzones");
298 
299 /*
300  * Any number that changes regularly is an okay choice for the
301  * offset.  Build numbers are pretty good of you have them.
302  */
303 static u_int zone_offset = __FreeBSD_version;
304 TUNABLE_INT("debug.malloc.zone_offset", &zone_offset);
305 SYSCTL_UINT(_debug_malloc, OID_AUTO, zone_offset, CTLFLAG_RDTUN,
306     &zone_offset, 0, "Separate malloc types by examining the "
307     "Nth character in the malloc type short description.");
308 
309 static void
310 mtp_set_subzone(struct malloc_type *mtp)
311 {
312 	struct malloc_type_internal *mtip;
313 	const char *desc;
314 	size_t len;
315 	u_int val;
316 
317 	mtip = mtp->ks_handle;
318 	desc = mtp->ks_shortdesc;
319 	if (desc == NULL || (len = strlen(desc)) == 0)
320 		val = 0;
321 	else
322 		val = desc[zone_offset % len];
323 	mtip->mti_zone = (val % numzones);
324 }
325 
326 static inline u_int
327 mtp_get_subzone(struct malloc_type *mtp)
328 {
329 	struct malloc_type_internal *mtip;
330 
331 	mtip = mtp->ks_handle;
332 
333 	KASSERT(mtip->mti_zone < numzones,
334 	    ("mti_zone %u out of range %d",
335 	    mtip->mti_zone, numzones));
336 	return (mtip->mti_zone);
337 }
338 #elif MALLOC_DEBUG_MAXZONES == 0
339 #error "MALLOC_DEBUG_MAXZONES must be positive."
340 #else
341 static void
342 mtp_set_subzone(struct malloc_type *mtp)
343 {
344 	struct malloc_type_internal *mtip;
345 
346 	mtip = mtp->ks_handle;
347 	mtip->mti_zone = 0;
348 }
349 
350 static inline u_int
351 mtp_get_subzone(struct malloc_type *mtp)
352 {
353 
354 	return (0);
355 }
356 #endif /* MALLOC_DEBUG_MAXZONES > 1 */
357 
358 int
359 malloc_last_fail(void)
360 {
361 
362 	return (time_uptime - t_malloc_fail);
363 }
364 
365 /*
366  * An allocation has succeeded -- update malloc type statistics for the
367  * amount of bucket size.  Occurs within a critical section so that the
368  * thread isn't preempted and doesn't migrate while updating per-PCU
369  * statistics.
370  */
371 static void
372 malloc_type_zone_allocated(struct malloc_type *mtp, unsigned long size,
373     int zindx)
374 {
375 	struct malloc_type_internal *mtip;
376 	struct malloc_type_stats *mtsp;
377 
378 	critical_enter();
379 	mtip = mtp->ks_handle;
380 	mtsp = zpcpu_get(mtip->mti_stats);
381 	if (size > 0) {
382 		mtsp->mts_memalloced += size;
383 		mtsp->mts_numallocs++;
384 	}
385 	if (zindx != -1)
386 		mtsp->mts_size |= 1 << zindx;
387 
388 #ifdef KDTRACE_HOOKS
389 	if (__predict_false(dtrace_malloc_enabled)) {
390 		uint32_t probe_id = mtip->mti_probes[DTMALLOC_PROBE_MALLOC];
391 		if (probe_id != 0)
392 			(dtrace_malloc_probe)(probe_id,
393 			    (uintptr_t) mtp, (uintptr_t) mtip,
394 			    (uintptr_t) mtsp, size, zindx);
395 	}
396 #endif
397 
398 	critical_exit();
399 }
400 
401 void
402 malloc_type_allocated(struct malloc_type *mtp, unsigned long size)
403 {
404 
405 	if (size > 0)
406 		malloc_type_zone_allocated(mtp, size, -1);
407 }
408 
409 /*
410  * A free operation has occurred -- update malloc type statistics for the
411  * amount of the bucket size.  Occurs within a critical section so that the
412  * thread isn't preempted and doesn't migrate while updating per-CPU
413  * statistics.
414  */
415 void
416 malloc_type_freed(struct malloc_type *mtp, unsigned long size)
417 {
418 	struct malloc_type_internal *mtip;
419 	struct malloc_type_stats *mtsp;
420 
421 	critical_enter();
422 	mtip = mtp->ks_handle;
423 	mtsp = zpcpu_get(mtip->mti_stats);
424 	mtsp->mts_memfreed += size;
425 	mtsp->mts_numfrees++;
426 
427 #ifdef KDTRACE_HOOKS
428 	if (__predict_false(dtrace_malloc_enabled)) {
429 		uint32_t probe_id = mtip->mti_probes[DTMALLOC_PROBE_FREE];
430 		if (probe_id != 0)
431 			(dtrace_malloc_probe)(probe_id,
432 			    (uintptr_t) mtp, (uintptr_t) mtip,
433 			    (uintptr_t) mtsp, size, 0);
434 	}
435 #endif
436 
437 	critical_exit();
438 }
439 
440 /*
441  *	contigmalloc:
442  *
443  *	Allocate a block of physically contiguous memory.
444  *
445  *	If M_NOWAIT is set, this routine will not block and return NULL if
446  *	the allocation fails.
447  */
448 void *
449 contigmalloc(unsigned long size, struct malloc_type *type, int flags,
450     vm_paddr_t low, vm_paddr_t high, unsigned long alignment,
451     vm_paddr_t boundary)
452 {
453 	void *ret;
454 
455 	ret = (void *)kmem_alloc_contig(size, flags, low, high, alignment,
456 	    boundary, VM_MEMATTR_DEFAULT);
457 	if (ret != NULL)
458 		malloc_type_allocated(type, round_page(size));
459 	return (ret);
460 }
461 
462 void *
463 contigmalloc_domainset(unsigned long size, struct malloc_type *type,
464     struct domainset *ds, int flags, vm_paddr_t low, vm_paddr_t high,
465     unsigned long alignment, vm_paddr_t boundary)
466 {
467 	void *ret;
468 
469 	ret = (void *)kmem_alloc_contig_domainset(ds, size, flags, low, high,
470 	    alignment, boundary, VM_MEMATTR_DEFAULT);
471 	if (ret != NULL)
472 		malloc_type_allocated(type, round_page(size));
473 	return (ret);
474 }
475 
476 /*
477  *	contigfree:
478  *
479  *	Free a block of memory allocated by contigmalloc.
480  *
481  *	This routine may not block.
482  */
483 void
484 contigfree(void *addr, unsigned long size, struct malloc_type *type)
485 {
486 
487 	kmem_free((vm_offset_t)addr, size);
488 	malloc_type_freed(type, round_page(size));
489 }
490 
491 #ifdef MALLOC_DEBUG
492 static int
493 malloc_dbg(caddr_t *vap, size_t *sizep, struct malloc_type *mtp,
494     int flags)
495 {
496 #ifdef INVARIANTS
497 	int indx;
498 
499 	KASSERT(mtp->ks_magic == M_MAGIC, ("malloc: bad malloc type magic"));
500 	/*
501 	 * Check that exactly one of M_WAITOK or M_NOWAIT is specified.
502 	 */
503 	indx = flags & (M_WAITOK | M_NOWAIT);
504 	if (indx != M_NOWAIT && indx != M_WAITOK) {
505 		static	struct timeval lasterr;
506 		static	int curerr, once;
507 		if (once == 0 && ppsratecheck(&lasterr, &curerr, 1)) {
508 			printf("Bad malloc flags: %x\n", indx);
509 			kdb_backtrace();
510 			flags |= M_WAITOK;
511 			once++;
512 		}
513 	}
514 #endif
515 #ifdef MALLOC_MAKE_FAILURES
516 	if ((flags & M_NOWAIT) && (malloc_failure_rate != 0)) {
517 		atomic_add_int(&malloc_nowait_count, 1);
518 		if ((malloc_nowait_count % malloc_failure_rate) == 0) {
519 			atomic_add_int(&malloc_failure_count, 1);
520 			t_malloc_fail = time_uptime;
521 			*vap = NULL;
522 			return (EJUSTRETURN);
523 		}
524 	}
525 #endif
526 	if (flags & M_WAITOK) {
527 		KASSERT(curthread->td_intr_nesting_level == 0,
528 		   ("malloc(M_WAITOK) in interrupt context"));
529 		if (__predict_false(!THREAD_CAN_SLEEP())) {
530 #ifdef EPOCH_TRACE
531 			epoch_trace_list(curthread);
532 #endif
533 			KASSERT(1,
534 			    ("malloc(M_WAITOK) with sleeping prohibited"));
535 		}
536 	}
537 	KASSERT(curthread->td_critnest == 0 || SCHEDULER_STOPPED(),
538 	    ("malloc: called with spinlock or critical section held"));
539 
540 #ifdef DEBUG_MEMGUARD
541 	if (memguard_cmp_mtp(mtp, *sizep)) {
542 		*vap = memguard_alloc(*sizep, flags);
543 		if (*vap != NULL)
544 			return (EJUSTRETURN);
545 		/* This is unfortunate but should not be fatal. */
546 	}
547 #endif
548 
549 #ifdef DEBUG_REDZONE
550 	*sizep = redzone_size_ntor(*sizep);
551 #endif
552 
553 	return (0);
554 }
555 #endif
556 
557 /*
558  * Handle large allocations and frees by using kmem_malloc directly.
559  */
560 static inline bool
561 malloc_large_slab(uma_slab_t slab)
562 {
563 	uintptr_t va;
564 
565 	va = (uintptr_t)slab;
566 	return ((va & 1) != 0);
567 }
568 
569 static inline size_t
570 malloc_large_size(uma_slab_t slab)
571 {
572 	uintptr_t va;
573 
574 	va = (uintptr_t)slab;
575 	return (va >> 1);
576 }
577 
578 static caddr_t
579 malloc_large(size_t *size, struct domainset *policy, int flags)
580 {
581 	vm_offset_t va;
582 	size_t sz;
583 
584 	sz = roundup(*size, PAGE_SIZE);
585 	va = kmem_malloc_domainset(policy, sz, flags);
586 	if (va != 0) {
587 		/* The low bit is unused for slab pointers. */
588 		vsetzoneslab(va, NULL, (void *)((sz << 1) | 1));
589 		uma_total_inc(sz);
590 		*size = sz;
591 	}
592 	return ((caddr_t)va);
593 }
594 
595 static void
596 free_large(void *addr, size_t size)
597 {
598 
599 	kmem_free((vm_offset_t)addr, size);
600 	uma_total_dec(size);
601 }
602 
603 /*
604  *	malloc:
605  *
606  *	Allocate a block of memory.
607  *
608  *	If M_NOWAIT is set, this routine will not block and return NULL if
609  *	the allocation fails.
610  */
611 void *
612 (malloc)(size_t size, struct malloc_type *mtp, int flags)
613 {
614 	int indx;
615 	caddr_t va;
616 	uma_zone_t zone;
617 #if defined(DEBUG_REDZONE)
618 	unsigned long osize = size;
619 #endif
620 
621 #ifdef MALLOC_DEBUG
622 	va = NULL;
623 	if (malloc_dbg(&va, &size, mtp, flags) != 0)
624 		return (va);
625 #endif
626 
627 	if (size <= kmem_zmax && (flags & M_EXEC) == 0) {
628 		if (size & KMEM_ZMASK)
629 			size = (size & ~KMEM_ZMASK) + KMEM_ZBASE;
630 		indx = kmemsize[size >> KMEM_ZSHIFT];
631 		zone = kmemzones[indx].kz_zone[mtp_get_subzone(mtp)];
632 #ifdef MALLOC_PROFILE
633 		krequests[size >> KMEM_ZSHIFT]++;
634 #endif
635 		va = uma_zalloc(zone, flags);
636 		if (va != NULL)
637 			size = zone->uz_size;
638 		malloc_type_zone_allocated(mtp, va == NULL ? 0 : size, indx);
639 	} else {
640 		va = malloc_large(&size, DOMAINSET_RR(), flags);
641 		malloc_type_allocated(mtp, va == NULL ? 0 : size);
642 	}
643 	if (flags & M_WAITOK)
644 		KASSERT(va != NULL, ("malloc(M_WAITOK) returned NULL"));
645 	else if (va == NULL)
646 		t_malloc_fail = time_uptime;
647 #ifdef DEBUG_REDZONE
648 	if (va != NULL)
649 		va = redzone_setup(va, osize);
650 #endif
651 	return ((void *) va);
652 }
653 
654 static void *
655 malloc_domain(size_t *sizep, int *indxp, struct malloc_type *mtp, int domain,
656     int flags)
657 {
658 	uma_zone_t zone;
659 	caddr_t va;
660 	size_t size;
661 	int indx;
662 
663 	size = *sizep;
664 	KASSERT(size <= kmem_zmax && (flags & M_EXEC) == 0,
665 	    ("malloc_domain: Called with bad flag / size combination."));
666 	if (size & KMEM_ZMASK)
667 		size = (size & ~KMEM_ZMASK) + KMEM_ZBASE;
668 	indx = kmemsize[size >> KMEM_ZSHIFT];
669 	zone = kmemzones[indx].kz_zone[mtp_get_subzone(mtp)];
670 #ifdef MALLOC_PROFILE
671 	krequests[size >> KMEM_ZSHIFT]++;
672 #endif
673 	va = uma_zalloc_domain(zone, NULL, domain, flags);
674 	if (va != NULL)
675 		*sizep = zone->uz_size;
676 	*indxp = indx;
677 	return ((void *)va);
678 }
679 
680 void *
681 malloc_domainset(size_t size, struct malloc_type *mtp, struct domainset *ds,
682     int flags)
683 {
684 	struct vm_domainset_iter di;
685 	caddr_t ret;
686 	int domain;
687 	int indx;
688 
689 #if defined(DEBUG_REDZONE)
690 	unsigned long osize = size;
691 #endif
692 #ifdef MALLOC_DEBUG
693 	ret= NULL;
694 	if (malloc_dbg(&ret, &size, mtp, flags) != 0)
695 		return (ret);
696 #endif
697 	if (size <= kmem_zmax && (flags & M_EXEC) == 0) {
698 		vm_domainset_iter_policy_init(&di, ds, &domain, &flags);
699 		do {
700 			ret = malloc_domain(&size, &indx, mtp, domain, flags);
701 		} while (ret == NULL &&
702 		    vm_domainset_iter_policy(&di, &domain) == 0);
703 		malloc_type_zone_allocated(mtp, ret == NULL ? 0 : size, indx);
704 	} else {
705 		/* Policy is handled by kmem. */
706 		ret = malloc_large(&size, ds, flags);
707 		malloc_type_allocated(mtp, ret == NULL ? 0 : size);
708 	}
709 
710 	if (flags & M_WAITOK)
711 		KASSERT(ret != NULL, ("malloc(M_WAITOK) returned NULL"));
712 	else if (ret == NULL)
713 		t_malloc_fail = time_uptime;
714 #ifdef DEBUG_REDZONE
715 	if (ret != NULL)
716 		ret = redzone_setup(ret, osize);
717 #endif
718 	return (ret);
719 }
720 
721 void *
722 mallocarray(size_t nmemb, size_t size, struct malloc_type *type, int flags)
723 {
724 
725 	if (WOULD_OVERFLOW(nmemb, size))
726 		panic("mallocarray: %zu * %zu overflowed", nmemb, size);
727 
728 	return (malloc(size * nmemb, type, flags));
729 }
730 
731 #ifdef INVARIANTS
732 static void
733 free_save_type(void *addr, struct malloc_type *mtp, u_long size)
734 {
735 	struct malloc_type **mtpp = addr;
736 
737 	/*
738 	 * Cache a pointer to the malloc_type that most recently freed
739 	 * this memory here.  This way we know who is most likely to
740 	 * have stepped on it later.
741 	 *
742 	 * This code assumes that size is a multiple of 8 bytes for
743 	 * 64 bit machines
744 	 */
745 	mtpp = (struct malloc_type **) ((unsigned long)mtpp & ~UMA_ALIGN_PTR);
746 	mtpp += (size - sizeof(struct malloc_type *)) /
747 	    sizeof(struct malloc_type *);
748 	*mtpp = mtp;
749 }
750 #endif
751 
752 #ifdef MALLOC_DEBUG
753 static int
754 free_dbg(void **addrp, struct malloc_type *mtp)
755 {
756 	void *addr;
757 
758 	addr = *addrp;
759 	KASSERT(mtp->ks_magic == M_MAGIC, ("free: bad malloc type magic"));
760 	KASSERT(curthread->td_critnest == 0 || SCHEDULER_STOPPED(),
761 	    ("free: called with spinlock or critical section held"));
762 
763 	/* free(NULL, ...) does nothing */
764 	if (addr == NULL)
765 		return (EJUSTRETURN);
766 
767 #ifdef DEBUG_MEMGUARD
768 	if (is_memguard_addr(addr)) {
769 		memguard_free(addr);
770 		return (EJUSTRETURN);
771 	}
772 #endif
773 
774 #ifdef DEBUG_REDZONE
775 	redzone_check(addr);
776 	*addrp = redzone_addr_ntor(addr);
777 #endif
778 
779 	return (0);
780 }
781 #endif
782 
783 /*
784  *	free:
785  *
786  *	Free a block of memory allocated by malloc.
787  *
788  *	This routine may not block.
789  */
790 void
791 free(void *addr, struct malloc_type *mtp)
792 {
793 	uma_zone_t zone;
794 	uma_slab_t slab;
795 	u_long size;
796 
797 #ifdef MALLOC_DEBUG
798 	if (free_dbg(&addr, mtp) != 0)
799 		return;
800 #endif
801 	/* free(NULL, ...) does nothing */
802 	if (addr == NULL)
803 		return;
804 
805 	vtozoneslab((vm_offset_t)addr & (~UMA_SLAB_MASK), &zone, &slab);
806 	if (slab == NULL)
807 		panic("free: address %p(%p) has not been allocated.\n",
808 		    addr, (void *)((u_long)addr & (~UMA_SLAB_MASK)));
809 
810 	if (__predict_true(!malloc_large_slab(slab))) {
811 		size = zone->uz_size;
812 #ifdef INVARIANTS
813 		free_save_type(addr, mtp, size);
814 #endif
815 		uma_zfree_arg(zone, addr, slab);
816 	} else {
817 		size = malloc_large_size(slab);
818 		free_large(addr, size);
819 	}
820 	malloc_type_freed(mtp, size);
821 }
822 
823 void
824 free_domain(void *addr, struct malloc_type *mtp)
825 {
826 	uma_zone_t zone;
827 	uma_slab_t slab;
828 	u_long size;
829 
830 #ifdef MALLOC_DEBUG
831 	if (free_dbg(&addr, mtp) != 0)
832 		return;
833 #endif
834 
835 	/* free(NULL, ...) does nothing */
836 	if (addr == NULL)
837 		return;
838 
839 	vtozoneslab((vm_offset_t)addr & (~UMA_SLAB_MASK), &zone, &slab);
840 	if (slab == NULL)
841 		panic("free_domain: address %p(%p) has not been allocated.\n",
842 		    addr, (void *)((u_long)addr & (~UMA_SLAB_MASK)));
843 
844 	if (__predict_true(!malloc_large_slab(slab))) {
845 		size = zone->uz_size;
846 #ifdef INVARIANTS
847 		free_save_type(addr, mtp, size);
848 #endif
849 		uma_zfree_domain(zone, addr, slab);
850 	} else {
851 		size = malloc_large_size(slab);
852 		free_large(addr, size);
853 	}
854 	malloc_type_freed(mtp, size);
855 }
856 
857 /*
858  *	realloc: change the size of a memory block
859  */
860 void *
861 realloc(void *addr, size_t size, struct malloc_type *mtp, int flags)
862 {
863 	uma_zone_t zone;
864 	uma_slab_t slab;
865 	unsigned long alloc;
866 	void *newaddr;
867 
868 	KASSERT(mtp->ks_magic == M_MAGIC,
869 	    ("realloc: bad malloc type magic"));
870 	KASSERT(curthread->td_critnest == 0 || SCHEDULER_STOPPED(),
871 	    ("realloc: called with spinlock or critical section held"));
872 
873 	/* realloc(NULL, ...) is equivalent to malloc(...) */
874 	if (addr == NULL)
875 		return (malloc(size, mtp, flags));
876 
877 	/*
878 	 * XXX: Should report free of old memory and alloc of new memory to
879 	 * per-CPU stats.
880 	 */
881 
882 #ifdef DEBUG_MEMGUARD
883 	if (is_memguard_addr(addr))
884 		return (memguard_realloc(addr, size, mtp, flags));
885 #endif
886 
887 #ifdef DEBUG_REDZONE
888 	slab = NULL;
889 	zone = NULL;
890 	alloc = redzone_get_size(addr);
891 #else
892 	vtozoneslab((vm_offset_t)addr & (~UMA_SLAB_MASK), &zone, &slab);
893 
894 	/* Sanity check */
895 	KASSERT(slab != NULL,
896 	    ("realloc: address %p out of range", (void *)addr));
897 
898 	/* Get the size of the original block */
899 	if (!malloc_large_slab(slab))
900 		alloc = zone->uz_size;
901 	else
902 		alloc = malloc_large_size(slab);
903 
904 	/* Reuse the original block if appropriate */
905 	if (size <= alloc
906 	    && (size > (alloc >> REALLOC_FRACTION) || alloc == MINALLOCSIZE))
907 		return (addr);
908 #endif /* !DEBUG_REDZONE */
909 
910 	/* Allocate a new, bigger (or smaller) block */
911 	if ((newaddr = malloc(size, mtp, flags)) == NULL)
912 		return (NULL);
913 
914 	/* Copy over original contents */
915 	bcopy(addr, newaddr, min(size, alloc));
916 	free(addr, mtp);
917 	return (newaddr);
918 }
919 
920 /*
921  *	reallocf: same as realloc() but free memory on failure.
922  */
923 void *
924 reallocf(void *addr, size_t size, struct malloc_type *mtp, int flags)
925 {
926 	void *mem;
927 
928 	if ((mem = realloc(addr, size, mtp, flags)) == NULL)
929 		free(addr, mtp);
930 	return (mem);
931 }
932 
933 #ifndef __sparc64__
934 CTASSERT(VM_KMEM_SIZE_SCALE >= 1);
935 #endif
936 
937 /*
938  * Initialize the kernel memory (kmem) arena.
939  */
940 void
941 kmeminit(void)
942 {
943 	u_long mem_size;
944 	u_long tmp;
945 
946 #ifdef VM_KMEM_SIZE
947 	if (vm_kmem_size == 0)
948 		vm_kmem_size = VM_KMEM_SIZE;
949 #endif
950 #ifdef VM_KMEM_SIZE_MIN
951 	if (vm_kmem_size_min == 0)
952 		vm_kmem_size_min = VM_KMEM_SIZE_MIN;
953 #endif
954 #ifdef VM_KMEM_SIZE_MAX
955 	if (vm_kmem_size_max == 0)
956 		vm_kmem_size_max = VM_KMEM_SIZE_MAX;
957 #endif
958 	/*
959 	 * Calculate the amount of kernel virtual address (KVA) space that is
960 	 * preallocated to the kmem arena.  In order to support a wide range
961 	 * of machines, it is a function of the physical memory size,
962 	 * specifically,
963 	 *
964 	 *	min(max(physical memory size / VM_KMEM_SIZE_SCALE,
965 	 *	    VM_KMEM_SIZE_MIN), VM_KMEM_SIZE_MAX)
966 	 *
967 	 * Every architecture must define an integral value for
968 	 * VM_KMEM_SIZE_SCALE.  However, the definitions of VM_KMEM_SIZE_MIN
969 	 * and VM_KMEM_SIZE_MAX, which represent respectively the floor and
970 	 * ceiling on this preallocation, are optional.  Typically,
971 	 * VM_KMEM_SIZE_MAX is itself a function of the available KVA space on
972 	 * a given architecture.
973 	 */
974 	mem_size = vm_cnt.v_page_count;
975 	if (mem_size <= 32768) /* delphij XXX 128MB */
976 		kmem_zmax = PAGE_SIZE;
977 
978 	if (vm_kmem_size_scale < 1)
979 		vm_kmem_size_scale = VM_KMEM_SIZE_SCALE;
980 
981 	/*
982 	 * Check if we should use defaults for the "vm_kmem_size"
983 	 * variable:
984 	 */
985 	if (vm_kmem_size == 0) {
986 		vm_kmem_size = mem_size / vm_kmem_size_scale;
987 		vm_kmem_size = vm_kmem_size * PAGE_SIZE < vm_kmem_size ?
988 		    vm_kmem_size_max : vm_kmem_size * PAGE_SIZE;
989 		if (vm_kmem_size_min > 0 && vm_kmem_size < vm_kmem_size_min)
990 			vm_kmem_size = vm_kmem_size_min;
991 		if (vm_kmem_size_max > 0 && vm_kmem_size >= vm_kmem_size_max)
992 			vm_kmem_size = vm_kmem_size_max;
993 	}
994 	if (vm_kmem_size == 0)
995 		panic("Tune VM_KMEM_SIZE_* for the platform");
996 
997 	/*
998 	 * The amount of KVA space that is preallocated to the
999 	 * kmem arena can be set statically at compile-time or manually
1000 	 * through the kernel environment.  However, it is still limited to
1001 	 * twice the physical memory size, which has been sufficient to handle
1002 	 * the most severe cases of external fragmentation in the kmem arena.
1003 	 */
1004 	if (vm_kmem_size / 2 / PAGE_SIZE > mem_size)
1005 		vm_kmem_size = 2 * mem_size * PAGE_SIZE;
1006 
1007 	vm_kmem_size = round_page(vm_kmem_size);
1008 #ifdef DEBUG_MEMGUARD
1009 	tmp = memguard_fudge(vm_kmem_size, kernel_map);
1010 #else
1011 	tmp = vm_kmem_size;
1012 #endif
1013 	uma_set_limit(tmp);
1014 
1015 #ifdef DEBUG_MEMGUARD
1016 	/*
1017 	 * Initialize MemGuard if support compiled in.  MemGuard is a
1018 	 * replacement allocator used for detecting tamper-after-free
1019 	 * scenarios as they occur.  It is only used for debugging.
1020 	 */
1021 	memguard_init(kernel_arena);
1022 #endif
1023 }
1024 
1025 /*
1026  * Initialize the kernel memory allocator
1027  */
1028 /* ARGSUSED*/
1029 static void
1030 mallocinit(void *dummy)
1031 {
1032 	int i;
1033 	uint8_t indx;
1034 
1035 	mtx_init(&malloc_mtx, "malloc", NULL, MTX_DEF);
1036 
1037 	kmeminit();
1038 
1039 	if (kmem_zmax < PAGE_SIZE || kmem_zmax > KMEM_ZMAX)
1040 		kmem_zmax = KMEM_ZMAX;
1041 
1042 	mt_stats_zone = uma_zcreate("mt_stats_zone",
1043 	    sizeof(struct malloc_type_stats), NULL, NULL, NULL, NULL,
1044 	    UMA_ALIGN_PTR, UMA_ZONE_PCPU);
1045 	mt_zone = uma_zcreate("mt_zone", sizeof(struct malloc_type_internal),
1046 #ifdef INVARIANTS
1047 	    mtrash_ctor, mtrash_dtor, mtrash_init, mtrash_fini,
1048 #else
1049 	    NULL, NULL, NULL, NULL,
1050 #endif
1051 	    UMA_ALIGN_PTR, UMA_ZONE_MALLOC);
1052 	for (i = 0, indx = 0; kmemzones[indx].kz_size != 0; indx++) {
1053 		int size = kmemzones[indx].kz_size;
1054 		char *name = kmemzones[indx].kz_name;
1055 		int subzone;
1056 
1057 		for (subzone = 0; subzone < numzones; subzone++) {
1058 			kmemzones[indx].kz_zone[subzone] =
1059 			    uma_zcreate(name, size,
1060 #ifdef INVARIANTS
1061 			    mtrash_ctor, mtrash_dtor, mtrash_init, mtrash_fini,
1062 #else
1063 			    NULL, NULL, NULL, NULL,
1064 #endif
1065 			    UMA_ALIGN_PTR, UMA_ZONE_MALLOC);
1066 		}
1067 		for (;i <= size; i+= KMEM_ZBASE)
1068 			kmemsize[i >> KMEM_ZSHIFT] = indx;
1069 
1070 	}
1071 }
1072 SYSINIT(kmem, SI_SUB_KMEM, SI_ORDER_SECOND, mallocinit, NULL);
1073 
1074 void
1075 malloc_init(void *data)
1076 {
1077 	struct malloc_type_internal *mtip;
1078 	struct malloc_type *mtp;
1079 
1080 	KASSERT(vm_cnt.v_page_count != 0, ("malloc_register before vm_init"));
1081 
1082 	mtp = data;
1083 	if (mtp->ks_magic != M_MAGIC)
1084 		panic("malloc_init: bad malloc type magic");
1085 
1086 	mtip = uma_zalloc(mt_zone, M_WAITOK | M_ZERO);
1087 	mtip->mti_stats = uma_zalloc_pcpu(mt_stats_zone, M_WAITOK | M_ZERO);
1088 	mtp->ks_handle = mtip;
1089 	mtp_set_subzone(mtp);
1090 
1091 	mtx_lock(&malloc_mtx);
1092 	mtp->ks_next = kmemstatistics;
1093 	kmemstatistics = mtp;
1094 	kmemcount++;
1095 	mtx_unlock(&malloc_mtx);
1096 }
1097 
1098 void
1099 malloc_uninit(void *data)
1100 {
1101 	struct malloc_type_internal *mtip;
1102 	struct malloc_type_stats *mtsp;
1103 	struct malloc_type *mtp, *temp;
1104 	uma_slab_t slab;
1105 	long temp_allocs, temp_bytes;
1106 	int i;
1107 
1108 	mtp = data;
1109 	KASSERT(mtp->ks_magic == M_MAGIC,
1110 	    ("malloc_uninit: bad malloc type magic"));
1111 	KASSERT(mtp->ks_handle != NULL, ("malloc_deregister: cookie NULL"));
1112 
1113 	mtx_lock(&malloc_mtx);
1114 	mtip = mtp->ks_handle;
1115 	mtp->ks_handle = NULL;
1116 	if (mtp != kmemstatistics) {
1117 		for (temp = kmemstatistics; temp != NULL;
1118 		    temp = temp->ks_next) {
1119 			if (temp->ks_next == mtp) {
1120 				temp->ks_next = mtp->ks_next;
1121 				break;
1122 			}
1123 		}
1124 		KASSERT(temp,
1125 		    ("malloc_uninit: type '%s' not found", mtp->ks_shortdesc));
1126 	} else
1127 		kmemstatistics = mtp->ks_next;
1128 	kmemcount--;
1129 	mtx_unlock(&malloc_mtx);
1130 
1131 	/*
1132 	 * Look for memory leaks.
1133 	 */
1134 	temp_allocs = temp_bytes = 0;
1135 	for (i = 0; i <= mp_maxid; i++) {
1136 		mtsp = zpcpu_get_cpu(mtip->mti_stats, i);
1137 		temp_allocs += mtsp->mts_numallocs;
1138 		temp_allocs -= mtsp->mts_numfrees;
1139 		temp_bytes += mtsp->mts_memalloced;
1140 		temp_bytes -= mtsp->mts_memfreed;
1141 	}
1142 	if (temp_allocs > 0 || temp_bytes > 0) {
1143 		printf("Warning: memory type %s leaked memory on destroy "
1144 		    "(%ld allocations, %ld bytes leaked).\n", mtp->ks_shortdesc,
1145 		    temp_allocs, temp_bytes);
1146 	}
1147 
1148 	slab = vtoslab((vm_offset_t) mtip & (~UMA_SLAB_MASK));
1149 	uma_zfree_pcpu(mt_stats_zone, mtip->mti_stats);
1150 	uma_zfree_arg(mt_zone, mtip, slab);
1151 }
1152 
1153 struct malloc_type *
1154 malloc_desc2type(const char *desc)
1155 {
1156 	struct malloc_type *mtp;
1157 
1158 	mtx_assert(&malloc_mtx, MA_OWNED);
1159 	for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) {
1160 		if (strcmp(mtp->ks_shortdesc, desc) == 0)
1161 			return (mtp);
1162 	}
1163 	return (NULL);
1164 }
1165 
1166 static int
1167 sysctl_kern_malloc_stats(SYSCTL_HANDLER_ARGS)
1168 {
1169 	struct malloc_type_stream_header mtsh;
1170 	struct malloc_type_internal *mtip;
1171 	struct malloc_type_stats *mtsp, zeromts;
1172 	struct malloc_type_header mth;
1173 	struct malloc_type *mtp;
1174 	int error, i;
1175 	struct sbuf sbuf;
1176 
1177 	error = sysctl_wire_old_buffer(req, 0);
1178 	if (error != 0)
1179 		return (error);
1180 	sbuf_new_for_sysctl(&sbuf, NULL, 128, req);
1181 	sbuf_clear_flags(&sbuf, SBUF_INCLUDENUL);
1182 	mtx_lock(&malloc_mtx);
1183 
1184 	bzero(&zeromts, sizeof(zeromts));
1185 
1186 	/*
1187 	 * Insert stream header.
1188 	 */
1189 	bzero(&mtsh, sizeof(mtsh));
1190 	mtsh.mtsh_version = MALLOC_TYPE_STREAM_VERSION;
1191 	mtsh.mtsh_maxcpus = MAXCPU;
1192 	mtsh.mtsh_count = kmemcount;
1193 	(void)sbuf_bcat(&sbuf, &mtsh, sizeof(mtsh));
1194 
1195 	/*
1196 	 * Insert alternating sequence of type headers and type statistics.
1197 	 */
1198 	for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) {
1199 		mtip = (struct malloc_type_internal *)mtp->ks_handle;
1200 
1201 		/*
1202 		 * Insert type header.
1203 		 */
1204 		bzero(&mth, sizeof(mth));
1205 		strlcpy(mth.mth_name, mtp->ks_shortdesc, MALLOC_MAX_NAME);
1206 		(void)sbuf_bcat(&sbuf, &mth, sizeof(mth));
1207 
1208 		/*
1209 		 * Insert type statistics for each CPU.
1210 		 */
1211 		for (i = 0; i <= mp_maxid; i++) {
1212 			mtsp = zpcpu_get_cpu(mtip->mti_stats, i);
1213 			(void)sbuf_bcat(&sbuf, mtsp, sizeof(*mtsp));
1214 		}
1215 		/*
1216 		 * Fill in the missing CPUs.
1217 		 */
1218 		for (; i < MAXCPU; i++) {
1219 			(void)sbuf_bcat(&sbuf, &zeromts, sizeof(zeromts));
1220 		}
1221 
1222 	}
1223 	mtx_unlock(&malloc_mtx);
1224 	error = sbuf_finish(&sbuf);
1225 	sbuf_delete(&sbuf);
1226 	return (error);
1227 }
1228 
1229 SYSCTL_PROC(_kern, OID_AUTO, malloc_stats, CTLFLAG_RD|CTLTYPE_STRUCT,
1230     0, 0, sysctl_kern_malloc_stats, "s,malloc_type_ustats",
1231     "Return malloc types");
1232 
1233 SYSCTL_INT(_kern, OID_AUTO, malloc_count, CTLFLAG_RD, &kmemcount, 0,
1234     "Count of kernel malloc types");
1235 
1236 void
1237 malloc_type_list(malloc_type_list_func_t *func, void *arg)
1238 {
1239 	struct malloc_type *mtp, **bufmtp;
1240 	int count, i;
1241 	size_t buflen;
1242 
1243 	mtx_lock(&malloc_mtx);
1244 restart:
1245 	mtx_assert(&malloc_mtx, MA_OWNED);
1246 	count = kmemcount;
1247 	mtx_unlock(&malloc_mtx);
1248 
1249 	buflen = sizeof(struct malloc_type *) * count;
1250 	bufmtp = malloc(buflen, M_TEMP, M_WAITOK);
1251 
1252 	mtx_lock(&malloc_mtx);
1253 
1254 	if (count < kmemcount) {
1255 		free(bufmtp, M_TEMP);
1256 		goto restart;
1257 	}
1258 
1259 	for (mtp = kmemstatistics, i = 0; mtp != NULL; mtp = mtp->ks_next, i++)
1260 		bufmtp[i] = mtp;
1261 
1262 	mtx_unlock(&malloc_mtx);
1263 
1264 	for (i = 0; i < count; i++)
1265 		(func)(bufmtp[i], arg);
1266 
1267 	free(bufmtp, M_TEMP);
1268 }
1269 
1270 #ifdef DDB
1271 static int64_t
1272 get_malloc_stats(const struct malloc_type_internal *mtip, uint64_t *allocs,
1273     uint64_t *inuse)
1274 {
1275 	const struct malloc_type_stats *mtsp;
1276 	uint64_t frees, alloced, freed;
1277 	int i;
1278 
1279 	*allocs = 0;
1280 	frees = 0;
1281 	alloced = 0;
1282 	freed = 0;
1283 	for (i = 0; i <= mp_maxid; i++) {
1284 		mtsp = zpcpu_get_cpu(mtip->mti_stats, i);
1285 
1286 		*allocs += mtsp->mts_numallocs;
1287 		frees += mtsp->mts_numfrees;
1288 		alloced += mtsp->mts_memalloced;
1289 		freed += mtsp->mts_memfreed;
1290 	}
1291 	*inuse = *allocs - frees;
1292 	return (alloced - freed);
1293 }
1294 
1295 DB_SHOW_COMMAND(malloc, db_show_malloc)
1296 {
1297 	const char *fmt_hdr, *fmt_entry;
1298 	struct malloc_type *mtp;
1299 	uint64_t allocs, inuse;
1300 	int64_t size;
1301 	/* variables for sorting */
1302 	struct malloc_type *last_mtype, *cur_mtype;
1303 	int64_t cur_size, last_size;
1304 	int ties;
1305 
1306 	if (modif[0] == 'i') {
1307 		fmt_hdr = "%s,%s,%s,%s\n";
1308 		fmt_entry = "\"%s\",%ju,%jdK,%ju\n";
1309 	} else {
1310 		fmt_hdr = "%18s %12s  %12s %12s\n";
1311 		fmt_entry = "%18s %12ju %12jdK %12ju\n";
1312 	}
1313 
1314 	db_printf(fmt_hdr, "Type", "InUse", "MemUse", "Requests");
1315 
1316 	/* Select sort, largest size first. */
1317 	last_mtype = NULL;
1318 	last_size = INT64_MAX;
1319 	for (;;) {
1320 		cur_mtype = NULL;
1321 		cur_size = -1;
1322 		ties = 0;
1323 
1324 		for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) {
1325 			/*
1326 			 * In the case of size ties, print out mtypes
1327 			 * in the order they are encountered.  That is,
1328 			 * when we encounter the most recently output
1329 			 * mtype, we have already printed all preceding
1330 			 * ties, and we must print all following ties.
1331 			 */
1332 			if (mtp == last_mtype) {
1333 				ties = 1;
1334 				continue;
1335 			}
1336 			size = get_malloc_stats(mtp->ks_handle, &allocs,
1337 			    &inuse);
1338 			if (size > cur_size && size < last_size + ties) {
1339 				cur_size = size;
1340 				cur_mtype = mtp;
1341 			}
1342 		}
1343 		if (cur_mtype == NULL)
1344 			break;
1345 
1346 		size = get_malloc_stats(cur_mtype->ks_handle, &allocs, &inuse);
1347 		db_printf(fmt_entry, cur_mtype->ks_shortdesc, inuse,
1348 		    howmany(size, 1024), allocs);
1349 
1350 		if (db_pager_quit)
1351 			break;
1352 
1353 		last_mtype = cur_mtype;
1354 		last_size = cur_size;
1355 	}
1356 }
1357 
1358 #if MALLOC_DEBUG_MAXZONES > 1
1359 DB_SHOW_COMMAND(multizone_matches, db_show_multizone_matches)
1360 {
1361 	struct malloc_type_internal *mtip;
1362 	struct malloc_type *mtp;
1363 	u_int subzone;
1364 
1365 	if (!have_addr) {
1366 		db_printf("Usage: show multizone_matches <malloc type/addr>\n");
1367 		return;
1368 	}
1369 	mtp = (void *)addr;
1370 	if (mtp->ks_magic != M_MAGIC) {
1371 		db_printf("Magic %lx does not match expected %x\n",
1372 		    mtp->ks_magic, M_MAGIC);
1373 		return;
1374 	}
1375 
1376 	mtip = mtp->ks_handle;
1377 	subzone = mtip->mti_zone;
1378 
1379 	for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) {
1380 		mtip = mtp->ks_handle;
1381 		if (mtip->mti_zone != subzone)
1382 			continue;
1383 		db_printf("%s\n", mtp->ks_shortdesc);
1384 		if (db_pager_quit)
1385 			break;
1386 	}
1387 }
1388 #endif /* MALLOC_DEBUG_MAXZONES > 1 */
1389 #endif /* DDB */
1390 
1391 #ifdef MALLOC_PROFILE
1392 
1393 static int
1394 sysctl_kern_mprof(SYSCTL_HANDLER_ARGS)
1395 {
1396 	struct sbuf sbuf;
1397 	uint64_t count;
1398 	uint64_t waste;
1399 	uint64_t mem;
1400 	int error;
1401 	int rsize;
1402 	int size;
1403 	int i;
1404 
1405 	waste = 0;
1406 	mem = 0;
1407 
1408 	error = sysctl_wire_old_buffer(req, 0);
1409 	if (error != 0)
1410 		return (error);
1411 	sbuf_new_for_sysctl(&sbuf, NULL, 128, req);
1412 	sbuf_printf(&sbuf,
1413 	    "\n  Size                    Requests  Real Size\n");
1414 	for (i = 0; i < KMEM_ZSIZE; i++) {
1415 		size = i << KMEM_ZSHIFT;
1416 		rsize = kmemzones[kmemsize[i]].kz_size;
1417 		count = (long long unsigned)krequests[i];
1418 
1419 		sbuf_printf(&sbuf, "%6d%28llu%11d\n", size,
1420 		    (unsigned long long)count, rsize);
1421 
1422 		if ((rsize * count) > (size * count))
1423 			waste += (rsize * count) - (size * count);
1424 		mem += (rsize * count);
1425 	}
1426 	sbuf_printf(&sbuf,
1427 	    "\nTotal memory used:\t%30llu\nTotal Memory wasted:\t%30llu\n",
1428 	    (unsigned long long)mem, (unsigned long long)waste);
1429 	error = sbuf_finish(&sbuf);
1430 	sbuf_delete(&sbuf);
1431 	return (error);
1432 }
1433 
1434 SYSCTL_OID(_kern, OID_AUTO, mprof, CTLTYPE_STRING|CTLFLAG_RD,
1435     NULL, 0, sysctl_kern_mprof, "A", "Malloc Profiling");
1436 #endif /* MALLOC_PROFILE */
1437