xref: /freebsd/sys/kern/kern_malloc.c (revision cd8537910406e68d4719136a5b0cf6d23bb1b23b)
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 #ifdef DEBUG_REDZONE
114 #define	DEBUG_REDZONE_ARG_DEF	, unsigned long osize
115 #define	DEBUG_REDZONE_ARG	, osize
116 #else
117 #define	DEBUG_REDZONE_ARG_DEF
118 #define	DEBUG_REDZONE_ARG
119 #endif
120 
121 /*
122  * When realloc() is called, if the new size is sufficiently smaller than
123  * the old size, realloc() will allocate a new, smaller block to avoid
124  * wasting memory. 'Sufficiently smaller' is defined as: newsize <=
125  * oldsize / 2^n, where REALLOC_FRACTION defines the value of 'n'.
126  */
127 #ifndef REALLOC_FRACTION
128 #define	REALLOC_FRACTION	1	/* new block if <= half the size */
129 #endif
130 
131 /*
132  * Centrally define some common malloc types.
133  */
134 MALLOC_DEFINE(M_CACHE, "cache", "Various Dynamically allocated caches");
135 MALLOC_DEFINE(M_DEVBUF, "devbuf", "device driver memory");
136 MALLOC_DEFINE(M_TEMP, "temp", "misc temporary data buffers");
137 
138 static struct malloc_type *kmemstatistics;
139 static int kmemcount;
140 
141 #define KMEM_ZSHIFT	4
142 #define KMEM_ZBASE	16
143 #define KMEM_ZMASK	(KMEM_ZBASE - 1)
144 
145 #define KMEM_ZMAX	65536
146 #define KMEM_ZSIZE	(KMEM_ZMAX >> KMEM_ZSHIFT)
147 static uint8_t kmemsize[KMEM_ZSIZE + 1];
148 
149 #ifndef MALLOC_DEBUG_MAXZONES
150 #define	MALLOC_DEBUG_MAXZONES	1
151 #endif
152 static int numzones = MALLOC_DEBUG_MAXZONES;
153 
154 /*
155  * Small malloc(9) memory allocations are allocated from a set of UMA buckets
156  * of various sizes.
157  *
158  * Warning: the layout of the struct is duplicated in libmemstat for KVM support.
159  *
160  * XXX: The comment here used to read "These won't be powers of two for
161  * long."  It's possible that a significant amount of wasted memory could be
162  * recovered by tuning the sizes of these buckets.
163  */
164 struct {
165 	int kz_size;
166 	const char *kz_name;
167 	uma_zone_t kz_zone[MALLOC_DEBUG_MAXZONES];
168 } kmemzones[] = {
169 	{16, "malloc-16", },
170 	{32, "malloc-32", },
171 	{64, "malloc-64", },
172 	{128, "malloc-128", },
173 	{256, "malloc-256", },
174 	{384, "malloc-384", },
175 	{512, "malloc-512", },
176 	{1024, "malloc-1024", },
177 	{2048, "malloc-2048", },
178 	{4096, "malloc-4096", },
179 	{8192, "malloc-8192", },
180 	{16384, "malloc-16384", },
181 	{32768, "malloc-32768", },
182 	{65536, "malloc-65536", },
183 	{0, NULL},
184 };
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 static SYSCTL_NODE(_vm, OID_AUTO, malloc, CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
217     "Malloc information");
218 
219 static u_int vm_malloc_zone_count = nitems(kmemzones);
220 SYSCTL_UINT(_vm_malloc, OID_AUTO, zone_count,
221     CTLFLAG_RD, &vm_malloc_zone_count, 0,
222     "Number of malloc zones");
223 
224 static int sysctl_vm_malloc_zone_sizes(SYSCTL_HANDLER_ARGS);
225 SYSCTL_PROC(_vm_malloc, OID_AUTO, zone_sizes,
226     CTLFLAG_RD | CTLTYPE_OPAQUE | CTLFLAG_MPSAFE, NULL, 0,
227     sysctl_vm_malloc_zone_sizes, "S", "Zone sizes used by malloc");
228 
229 /*
230  * The malloc_mtx protects the kmemstatistics linked list.
231  */
232 struct mtx malloc_mtx;
233 
234 static int sysctl_kern_malloc_stats(SYSCTL_HANDLER_ARGS);
235 
236 #if defined(MALLOC_MAKE_FAILURES) || (MALLOC_DEBUG_MAXZONES > 1)
237 static SYSCTL_NODE(_debug, OID_AUTO, malloc, CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
238     "Kernel malloc debugging options");
239 #endif
240 
241 /*
242  * malloc(9) fault injection -- cause malloc failures every (n) mallocs when
243  * the caller specifies M_NOWAIT.  If set to 0, no failures are caused.
244  */
245 #ifdef MALLOC_MAKE_FAILURES
246 static int malloc_failure_rate;
247 static int malloc_nowait_count;
248 static int malloc_failure_count;
249 SYSCTL_INT(_debug_malloc, OID_AUTO, failure_rate, CTLFLAG_RWTUN,
250     &malloc_failure_rate, 0, "Every (n) mallocs with M_NOWAIT will fail");
251 SYSCTL_INT(_debug_malloc, OID_AUTO, failure_count, CTLFLAG_RD,
252     &malloc_failure_count, 0, "Number of imposed M_NOWAIT malloc failures");
253 #endif
254 
255 static int
256 sysctl_kmem_map_size(SYSCTL_HANDLER_ARGS)
257 {
258 	u_long size;
259 
260 	size = uma_size();
261 	return (sysctl_handle_long(oidp, &size, 0, req));
262 }
263 
264 static int
265 sysctl_kmem_map_free(SYSCTL_HANDLER_ARGS)
266 {
267 	u_long size, limit;
268 
269 	/* The sysctl is unsigned, implement as a saturation value. */
270 	size = uma_size();
271 	limit = uma_limit();
272 	if (size > limit)
273 		size = 0;
274 	else
275 		size = limit - size;
276 	return (sysctl_handle_long(oidp, &size, 0, req));
277 }
278 
279 static int
280 sysctl_vm_malloc_zone_sizes(SYSCTL_HANDLER_ARGS)
281 {
282 	int sizes[nitems(kmemzones)];
283 	int i;
284 
285 	for (i = 0; i < nitems(kmemzones); i++) {
286 		sizes[i] = kmemzones[i].kz_size;
287 	}
288 
289 	return (SYSCTL_OUT(req, &sizes, sizeof(sizes)));
290 }
291 
292 /*
293  * malloc(9) uma zone separation -- sub-page buffer overruns in one
294  * malloc type will affect only a subset of other malloc types.
295  */
296 #if MALLOC_DEBUG_MAXZONES > 1
297 static void
298 tunable_set_numzones(void)
299 {
300 
301 	TUNABLE_INT_FETCH("debug.malloc.numzones",
302 	    &numzones);
303 
304 	/* Sanity check the number of malloc uma zones. */
305 	if (numzones <= 0)
306 		numzones = 1;
307 	if (numzones > MALLOC_DEBUG_MAXZONES)
308 		numzones = MALLOC_DEBUG_MAXZONES;
309 }
310 SYSINIT(numzones, SI_SUB_TUNABLES, SI_ORDER_ANY, tunable_set_numzones, NULL);
311 SYSCTL_INT(_debug_malloc, OID_AUTO, numzones, CTLFLAG_RDTUN | CTLFLAG_NOFETCH,
312     &numzones, 0, "Number of malloc uma subzones");
313 
314 /*
315  * Any number that changes regularly is an okay choice for the
316  * offset.  Build numbers are pretty good of you have them.
317  */
318 static u_int zone_offset = __FreeBSD_version;
319 TUNABLE_INT("debug.malloc.zone_offset", &zone_offset);
320 SYSCTL_UINT(_debug_malloc, OID_AUTO, zone_offset, CTLFLAG_RDTUN,
321     &zone_offset, 0, "Separate malloc types by examining the "
322     "Nth character in the malloc type short description.");
323 
324 static void
325 mtp_set_subzone(struct malloc_type *mtp)
326 {
327 	struct malloc_type_internal *mtip;
328 	const char *desc;
329 	size_t len;
330 	u_int val;
331 
332 	mtip = &mtp->ks_mti;
333 	desc = mtp->ks_shortdesc;
334 	if (desc == NULL || (len = strlen(desc)) == 0)
335 		val = 0;
336 	else
337 		val = desc[zone_offset % len];
338 	mtip->mti_zone = (val % numzones);
339 }
340 
341 static inline u_int
342 mtp_get_subzone(struct malloc_type *mtp)
343 {
344 	struct malloc_type_internal *mtip;
345 
346 	mtip = &mtp->ks_mti;
347 
348 	KASSERT(mtip->mti_zone < numzones,
349 	    ("mti_zone %u out of range %d",
350 	    mtip->mti_zone, numzones));
351 	return (mtip->mti_zone);
352 }
353 #elif MALLOC_DEBUG_MAXZONES == 0
354 #error "MALLOC_DEBUG_MAXZONES must be positive."
355 #else
356 static void
357 mtp_set_subzone(struct malloc_type *mtp)
358 {
359 	struct malloc_type_internal *mtip;
360 
361 	mtip = &mtp->ks_mti;
362 	mtip->mti_zone = 0;
363 }
364 
365 static inline u_int
366 mtp_get_subzone(struct malloc_type *mtp)
367 {
368 
369 	return (0);
370 }
371 #endif /* MALLOC_DEBUG_MAXZONES > 1 */
372 
373 /*
374  * An allocation has succeeded -- update malloc type statistics for the
375  * amount of bucket size.  Occurs within a critical section so that the
376  * thread isn't preempted and doesn't migrate while updating per-PCU
377  * statistics.
378  */
379 static void
380 malloc_type_zone_allocated(struct malloc_type *mtp, unsigned long size,
381     int zindx)
382 {
383 	struct malloc_type_internal *mtip;
384 	struct malloc_type_stats *mtsp;
385 
386 	critical_enter();
387 	mtip = &mtp->ks_mti;
388 	mtsp = zpcpu_get(mtip->mti_stats);
389 	if (size > 0) {
390 		mtsp->mts_memalloced += size;
391 		mtsp->mts_numallocs++;
392 	}
393 	if (zindx != -1)
394 		mtsp->mts_size |= 1 << zindx;
395 
396 #ifdef KDTRACE_HOOKS
397 	if (__predict_false(dtrace_malloc_enabled)) {
398 		uint32_t probe_id = mtip->mti_probes[DTMALLOC_PROBE_MALLOC];
399 		if (probe_id != 0)
400 			(dtrace_malloc_probe)(probe_id,
401 			    (uintptr_t) mtp, (uintptr_t) mtip,
402 			    (uintptr_t) mtsp, size, zindx);
403 	}
404 #endif
405 
406 	critical_exit();
407 }
408 
409 void
410 malloc_type_allocated(struct malloc_type *mtp, unsigned long size)
411 {
412 
413 	if (size > 0)
414 		malloc_type_zone_allocated(mtp, size, -1);
415 }
416 
417 /*
418  * A free operation has occurred -- update malloc type statistics for the
419  * amount of the bucket size.  Occurs within a critical section so that the
420  * thread isn't preempted and doesn't migrate while updating per-CPU
421  * statistics.
422  */
423 void
424 malloc_type_freed(struct malloc_type *mtp, unsigned long size)
425 {
426 	struct malloc_type_internal *mtip;
427 	struct malloc_type_stats *mtsp;
428 
429 	critical_enter();
430 	mtip = &mtp->ks_mti;
431 	mtsp = zpcpu_get(mtip->mti_stats);
432 	mtsp->mts_memfreed += size;
433 	mtsp->mts_numfrees++;
434 
435 #ifdef KDTRACE_HOOKS
436 	if (__predict_false(dtrace_malloc_enabled)) {
437 		uint32_t probe_id = mtip->mti_probes[DTMALLOC_PROBE_FREE];
438 		if (probe_id != 0)
439 			(dtrace_malloc_probe)(probe_id,
440 			    (uintptr_t) mtp, (uintptr_t) mtip,
441 			    (uintptr_t) mtsp, size, 0);
442 	}
443 #endif
444 
445 	critical_exit();
446 }
447 
448 /*
449  *	contigmalloc:
450  *
451  *	Allocate a block of physically contiguous memory.
452  *
453  *	If M_NOWAIT is set, this routine will not block and return NULL if
454  *	the allocation fails.
455  */
456 void *
457 contigmalloc(unsigned long size, struct malloc_type *type, int flags,
458     vm_paddr_t low, vm_paddr_t high, unsigned long alignment,
459     vm_paddr_t boundary)
460 {
461 	void *ret;
462 
463 	ret = (void *)kmem_alloc_contig(size, flags, low, high, alignment,
464 	    boundary, VM_MEMATTR_DEFAULT);
465 	if (ret != NULL)
466 		malloc_type_allocated(type, round_page(size));
467 	return (ret);
468 }
469 
470 void *
471 contigmalloc_domainset(unsigned long size, struct malloc_type *type,
472     struct domainset *ds, int flags, vm_paddr_t low, vm_paddr_t high,
473     unsigned long alignment, vm_paddr_t boundary)
474 {
475 	void *ret;
476 
477 	ret = (void *)kmem_alloc_contig_domainset(ds, size, flags, low, high,
478 	    alignment, boundary, VM_MEMATTR_DEFAULT);
479 	if (ret != NULL)
480 		malloc_type_allocated(type, round_page(size));
481 	return (ret);
482 }
483 
484 /*
485  *	contigfree:
486  *
487  *	Free a block of memory allocated by contigmalloc.
488  *
489  *	This routine may not block.
490  */
491 void
492 contigfree(void *addr, unsigned long size, struct malloc_type *type)
493 {
494 
495 	kmem_free((vm_offset_t)addr, size);
496 	malloc_type_freed(type, round_page(size));
497 }
498 
499 #ifdef MALLOC_DEBUG
500 static int
501 malloc_dbg(caddr_t *vap, size_t *sizep, struct malloc_type *mtp,
502     int flags)
503 {
504 #ifdef INVARIANTS
505 	int indx;
506 
507 	KASSERT(mtp->ks_version == M_VERSION, ("malloc: bad malloc type version"));
508 	/*
509 	 * Check that exactly one of M_WAITOK or M_NOWAIT is specified.
510 	 */
511 	indx = flags & (M_WAITOK | M_NOWAIT);
512 	if (indx != M_NOWAIT && indx != M_WAITOK) {
513 		static	struct timeval lasterr;
514 		static	int curerr, once;
515 		if (once == 0 && ppsratecheck(&lasterr, &curerr, 1)) {
516 			printf("Bad malloc flags: %x\n", indx);
517 			kdb_backtrace();
518 			flags |= M_WAITOK;
519 			once++;
520 		}
521 	}
522 #endif
523 #ifdef MALLOC_MAKE_FAILURES
524 	if ((flags & M_NOWAIT) && (malloc_failure_rate != 0)) {
525 		atomic_add_int(&malloc_nowait_count, 1);
526 		if ((malloc_nowait_count % malloc_failure_rate) == 0) {
527 			atomic_add_int(&malloc_failure_count, 1);
528 			*vap = NULL;
529 			return (EJUSTRETURN);
530 		}
531 	}
532 #endif
533 	if (flags & M_WAITOK) {
534 		KASSERT(curthread->td_intr_nesting_level == 0,
535 		   ("malloc(M_WAITOK) in interrupt context"));
536 		if (__predict_false(!THREAD_CAN_SLEEP())) {
537 #ifdef EPOCH_TRACE
538 			epoch_trace_list(curthread);
539 #endif
540 			KASSERT(1,
541 			    ("malloc(M_WAITOK) with sleeping prohibited"));
542 		}
543 	}
544 	KASSERT(curthread->td_critnest == 0 || SCHEDULER_STOPPED(),
545 	    ("malloc: called with spinlock or critical section held"));
546 
547 #ifdef DEBUG_MEMGUARD
548 	if (memguard_cmp_mtp(mtp, *sizep)) {
549 		*vap = memguard_alloc(*sizep, flags);
550 		if (*vap != NULL)
551 			return (EJUSTRETURN);
552 		/* This is unfortunate but should not be fatal. */
553 	}
554 #endif
555 
556 #ifdef DEBUG_REDZONE
557 	*sizep = redzone_size_ntor(*sizep);
558 #endif
559 
560 	return (0);
561 }
562 #endif
563 
564 /*
565  * Handle large allocations and frees by using kmem_malloc directly.
566  */
567 static inline bool
568 malloc_large_slab(uma_slab_t slab)
569 {
570 	uintptr_t va;
571 
572 	va = (uintptr_t)slab;
573 	return ((va & 1) != 0);
574 }
575 
576 static inline size_t
577 malloc_large_size(uma_slab_t slab)
578 {
579 	uintptr_t va;
580 
581 	va = (uintptr_t)slab;
582 	return (va >> 1);
583 }
584 
585 static caddr_t __noinline
586 malloc_large(size_t *size, struct malloc_type *mtp, struct domainset *policy,
587     int flags DEBUG_REDZONE_ARG_DEF)
588 {
589 	vm_offset_t kva;
590 	caddr_t va;
591 	size_t sz;
592 
593 	sz = roundup(*size, PAGE_SIZE);
594 	kva = kmem_malloc_domainset(policy, sz, flags);
595 	if (kva != 0) {
596 		/* The low bit is unused for slab pointers. */
597 		vsetzoneslab(kva, NULL, (void *)((sz << 1) | 1));
598 		uma_total_inc(sz);
599 		*size = sz;
600 	}
601 	va = (caddr_t)kva;
602 	malloc_type_allocated(mtp, va == NULL ? 0 : sz);
603 	if (__predict_false(va == NULL)) {
604 		KASSERT((flags & M_WAITOK) == 0,
605 		    ("malloc(M_WAITOK) returned NULL"));
606 	}
607 #ifdef DEBUG_REDZONE
608 	if (va != NULL)
609 		va = redzone_setup(va, osize);
610 #endif
611 	return (va);
612 }
613 
614 static void
615 free_large(void *addr, size_t size)
616 {
617 
618 	kmem_free((vm_offset_t)addr, size);
619 	uma_total_dec(size);
620 }
621 
622 /*
623  *	malloc:
624  *
625  *	Allocate a block of memory.
626  *
627  *	If M_NOWAIT is set, this routine will not block and return NULL if
628  *	the allocation fails.
629  */
630 void *
631 (malloc)(size_t size, struct malloc_type *mtp, int flags)
632 {
633 	int indx;
634 	caddr_t va;
635 	uma_zone_t zone;
636 #ifdef DEBUG_REDZONE
637 	unsigned long osize = size;
638 #endif
639 
640 	MPASS((flags & M_EXEC) == 0);
641 
642 #ifdef MALLOC_DEBUG
643 	va = NULL;
644 	if (malloc_dbg(&va, &size, mtp, flags) != 0)
645 		return (va);
646 #endif
647 
648 	if (__predict_false(size > kmem_zmax))
649 		return (malloc_large(&size, mtp, DOMAINSET_RR(), flags
650 		    DEBUG_REDZONE_ARG));
651 
652 	if (size & KMEM_ZMASK)
653 		size = (size & ~KMEM_ZMASK) + KMEM_ZBASE;
654 	indx = kmemsize[size >> KMEM_ZSHIFT];
655 	zone = kmemzones[indx].kz_zone[mtp_get_subzone(mtp)];
656 	va = uma_zalloc(zone, flags);
657 	if (va != NULL)
658 		size = zone->uz_size;
659 	malloc_type_zone_allocated(mtp, va == NULL ? 0 : size, indx);
660 	if (__predict_false(va == NULL)) {
661 		KASSERT((flags & M_WAITOK) == 0,
662 		    ("malloc(M_WAITOK) returned NULL"));
663 	}
664 #ifdef DEBUG_REDZONE
665 	if (va != NULL)
666 		va = redzone_setup(va, osize);
667 #endif
668 	return ((void *) va);
669 }
670 
671 static void *
672 malloc_domain(size_t *sizep, int *indxp, struct malloc_type *mtp, int domain,
673     int flags)
674 {
675 	uma_zone_t zone;
676 	caddr_t va;
677 	size_t size;
678 	int indx;
679 
680 	size = *sizep;
681 	KASSERT(size <= kmem_zmax && (flags & M_EXEC) == 0,
682 	    ("malloc_domain: Called with bad flag / size combination."));
683 	if (size & KMEM_ZMASK)
684 		size = (size & ~KMEM_ZMASK) + KMEM_ZBASE;
685 	indx = kmemsize[size >> KMEM_ZSHIFT];
686 	zone = kmemzones[indx].kz_zone[mtp_get_subzone(mtp)];
687 	va = uma_zalloc_domain(zone, NULL, domain, flags);
688 	if (va != NULL)
689 		*sizep = zone->uz_size;
690 	*indxp = indx;
691 	return ((void *)va);
692 }
693 
694 void *
695 malloc_domainset(size_t size, struct malloc_type *mtp, struct domainset *ds,
696     int flags)
697 {
698 	struct vm_domainset_iter di;
699 	caddr_t va;
700 	int domain;
701 	int indx;
702 #ifdef DEBUG_REDZONE
703 	unsigned long osize = size;
704 #endif
705 
706 	MPASS((flags & M_EXEC) == 0);
707 
708 #ifdef MALLOC_DEBUG
709 	va = NULL;
710 	if (malloc_dbg(&va, &size, mtp, flags) != 0)
711 		return (va);
712 #endif
713 
714 	if (__predict_false(size > kmem_zmax))
715 		return (malloc_large(&size, mtp, DOMAINSET_RR(), flags
716 		    DEBUG_REDZONE_ARG));
717 
718 	vm_domainset_iter_policy_init(&di, ds, &domain, &flags);
719 	do {
720 		va = malloc_domain(&size, &indx, mtp, domain, flags);
721 	} while (va == NULL && vm_domainset_iter_policy(&di, &domain) == 0);
722 	malloc_type_zone_allocated(mtp, va == NULL ? 0 : size, indx);
723 	if (__predict_false(va == NULL)) {
724 		KASSERT((flags & M_WAITOK) == 0,
725 		    ("malloc(M_WAITOK) returned NULL"));
726 	}
727 #ifdef DEBUG_REDZONE
728 	if (va != NULL)
729 		va = redzone_setup(va, osize);
730 #endif
731 	return (va);
732 }
733 
734 /*
735  * Allocate an executable area.
736  */
737 void *
738 malloc_exec(size_t size, struct malloc_type *mtp, int flags)
739 {
740 
741 	return (malloc_domainset_exec(size, mtp, DOMAINSET_RR(), flags));
742 }
743 
744 void *
745 malloc_domainset_exec(size_t size, struct malloc_type *mtp, struct domainset *ds,
746     int flags)
747 {
748 #ifdef DEBUG_REDZONE
749 	unsigned long osize = size;
750 #endif
751 #ifdef MALLOC_DEBUG
752 	caddr_t va;
753 #endif
754 
755 	flags |= M_EXEC;
756 
757 #ifdef MALLOC_DEBUG
758 	va = NULL;
759 	if (malloc_dbg(&va, &size, mtp, flags) != 0)
760 		return (va);
761 #endif
762 
763 	return (malloc_large(&size, mtp, ds, flags DEBUG_REDZONE_ARG));
764 }
765 
766 void *
767 mallocarray(size_t nmemb, size_t size, struct malloc_type *type, int flags)
768 {
769 
770 	if (WOULD_OVERFLOW(nmemb, size))
771 		panic("mallocarray: %zu * %zu overflowed", nmemb, size);
772 
773 	return (malloc(size * nmemb, type, flags));
774 }
775 
776 #ifdef INVARIANTS
777 static void
778 free_save_type(void *addr, struct malloc_type *mtp, u_long size)
779 {
780 	struct malloc_type **mtpp = addr;
781 
782 	/*
783 	 * Cache a pointer to the malloc_type that most recently freed
784 	 * this memory here.  This way we know who is most likely to
785 	 * have stepped on it later.
786 	 *
787 	 * This code assumes that size is a multiple of 8 bytes for
788 	 * 64 bit machines
789 	 */
790 	mtpp = (struct malloc_type **) ((unsigned long)mtpp & ~UMA_ALIGN_PTR);
791 	mtpp += (size - sizeof(struct malloc_type *)) /
792 	    sizeof(struct malloc_type *);
793 	*mtpp = mtp;
794 }
795 #endif
796 
797 #ifdef MALLOC_DEBUG
798 static int
799 free_dbg(void **addrp, struct malloc_type *mtp)
800 {
801 	void *addr;
802 
803 	addr = *addrp;
804 	KASSERT(mtp->ks_version == M_VERSION, ("free: bad malloc type version"));
805 	KASSERT(curthread->td_critnest == 0 || SCHEDULER_STOPPED(),
806 	    ("free: called with spinlock or critical section held"));
807 
808 	/* free(NULL, ...) does nothing */
809 	if (addr == NULL)
810 		return (EJUSTRETURN);
811 
812 #ifdef DEBUG_MEMGUARD
813 	if (is_memguard_addr(addr)) {
814 		memguard_free(addr);
815 		return (EJUSTRETURN);
816 	}
817 #endif
818 
819 #ifdef DEBUG_REDZONE
820 	redzone_check(addr);
821 	*addrp = redzone_addr_ntor(addr);
822 #endif
823 
824 	return (0);
825 }
826 #endif
827 
828 /*
829  *	free:
830  *
831  *	Free a block of memory allocated by malloc.
832  *
833  *	This routine may not block.
834  */
835 void
836 free(void *addr, struct malloc_type *mtp)
837 {
838 	uma_zone_t zone;
839 	uma_slab_t slab;
840 	u_long size;
841 
842 #ifdef MALLOC_DEBUG
843 	if (free_dbg(&addr, mtp) != 0)
844 		return;
845 #endif
846 	/* free(NULL, ...) does nothing */
847 	if (addr == NULL)
848 		return;
849 
850 	vtozoneslab((vm_offset_t)addr & (~UMA_SLAB_MASK), &zone, &slab);
851 	if (slab == NULL)
852 		panic("free: address %p(%p) has not been allocated.\n",
853 		    addr, (void *)((u_long)addr & (~UMA_SLAB_MASK)));
854 
855 	if (__predict_true(!malloc_large_slab(slab))) {
856 		size = zone->uz_size;
857 #ifdef INVARIANTS
858 		free_save_type(addr, mtp, size);
859 #endif
860 		uma_zfree_arg(zone, addr, slab);
861 	} else {
862 		size = malloc_large_size(slab);
863 		free_large(addr, size);
864 	}
865 	malloc_type_freed(mtp, size);
866 }
867 
868 /*
869  *	zfree:
870  *
871  *	Zero then free a block of memory allocated by malloc.
872  *
873  *	This routine may not block.
874  */
875 void
876 zfree(void *addr, struct malloc_type *mtp)
877 {
878 	uma_zone_t zone;
879 	uma_slab_t slab;
880 	u_long size;
881 
882 #ifdef MALLOC_DEBUG
883 	if (free_dbg(&addr, mtp) != 0)
884 		return;
885 #endif
886 	/* free(NULL, ...) does nothing */
887 	if (addr == NULL)
888 		return;
889 
890 	vtozoneslab((vm_offset_t)addr & (~UMA_SLAB_MASK), &zone, &slab);
891 	if (slab == NULL)
892 		panic("free: address %p(%p) has not been allocated.\n",
893 		    addr, (void *)((u_long)addr & (~UMA_SLAB_MASK)));
894 
895 	if (__predict_true(!malloc_large_slab(slab))) {
896 		size = zone->uz_size;
897 #ifdef INVARIANTS
898 		free_save_type(addr, mtp, size);
899 #endif
900 		explicit_bzero(addr, size);
901 		uma_zfree_arg(zone, addr, slab);
902 	} else {
903 		size = malloc_large_size(slab);
904 		explicit_bzero(addr, size);
905 		free_large(addr, size);
906 	}
907 	malloc_type_freed(mtp, size);
908 }
909 
910 /*
911  *	realloc: change the size of a memory block
912  */
913 void *
914 realloc(void *addr, size_t size, struct malloc_type *mtp, int flags)
915 {
916 	uma_zone_t zone;
917 	uma_slab_t slab;
918 	unsigned long alloc;
919 	void *newaddr;
920 
921 	KASSERT(mtp->ks_version == M_VERSION,
922 	    ("realloc: bad malloc type version"));
923 	KASSERT(curthread->td_critnest == 0 || SCHEDULER_STOPPED(),
924 	    ("realloc: called with spinlock or critical section held"));
925 
926 	/* realloc(NULL, ...) is equivalent to malloc(...) */
927 	if (addr == NULL)
928 		return (malloc(size, mtp, flags));
929 
930 	/*
931 	 * XXX: Should report free of old memory and alloc of new memory to
932 	 * per-CPU stats.
933 	 */
934 
935 #ifdef DEBUG_MEMGUARD
936 	if (is_memguard_addr(addr))
937 		return (memguard_realloc(addr, size, mtp, flags));
938 #endif
939 
940 #ifdef DEBUG_REDZONE
941 	slab = NULL;
942 	zone = NULL;
943 	alloc = redzone_get_size(addr);
944 #else
945 	vtozoneslab((vm_offset_t)addr & (~UMA_SLAB_MASK), &zone, &slab);
946 
947 	/* Sanity check */
948 	KASSERT(slab != NULL,
949 	    ("realloc: address %p out of range", (void *)addr));
950 
951 	/* Get the size of the original block */
952 	if (!malloc_large_slab(slab))
953 		alloc = zone->uz_size;
954 	else
955 		alloc = malloc_large_size(slab);
956 
957 	/* Reuse the original block if appropriate */
958 	if (size <= alloc
959 	    && (size > (alloc >> REALLOC_FRACTION) || alloc == MINALLOCSIZE))
960 		return (addr);
961 #endif /* !DEBUG_REDZONE */
962 
963 	/* Allocate a new, bigger (or smaller) block */
964 	if ((newaddr = malloc(size, mtp, flags)) == NULL)
965 		return (NULL);
966 
967 	/* Copy over original contents */
968 	bcopy(addr, newaddr, min(size, alloc));
969 	free(addr, mtp);
970 	return (newaddr);
971 }
972 
973 /*
974  *	reallocf: same as realloc() but free memory on failure.
975  */
976 void *
977 reallocf(void *addr, size_t size, struct malloc_type *mtp, int flags)
978 {
979 	void *mem;
980 
981 	if ((mem = realloc(addr, size, mtp, flags)) == NULL)
982 		free(addr, mtp);
983 	return (mem);
984 }
985 
986 /*
987  * 	malloc_size: returns the number of bytes allocated for a request of the
988  * 		     specified size
989  */
990 size_t
991 malloc_size(size_t size)
992 {
993 	int indx;
994 
995 	if (size > kmem_zmax)
996 		return (0);
997 	if (size & KMEM_ZMASK)
998 		size = (size & ~KMEM_ZMASK) + KMEM_ZBASE;
999 	indx = kmemsize[size >> KMEM_ZSHIFT];
1000 	return (kmemzones[indx].kz_size);
1001 }
1002 
1003 /*
1004  *	malloc_usable_size: returns the usable size of the allocation.
1005  */
1006 size_t
1007 malloc_usable_size(const void *addr)
1008 {
1009 #ifndef DEBUG_REDZONE
1010 	uma_zone_t zone;
1011 	uma_slab_t slab;
1012 #endif
1013 	u_long size;
1014 
1015 	if (addr == NULL)
1016 		return (0);
1017 
1018 #ifdef DEBUG_MEMGUARD
1019 	if (is_memguard_addr(__DECONST(void *, addr)))
1020 		return (memguard_get_req_size(addr));
1021 #endif
1022 
1023 #ifdef DEBUG_REDZONE
1024 	size = redzone_get_size(__DECONST(void *, addr));
1025 #else
1026 	vtozoneslab((vm_offset_t)addr & (~UMA_SLAB_MASK), &zone, &slab);
1027 	if (slab == NULL)
1028 		panic("malloc_usable_size: address %p(%p) is not allocated.\n",
1029 		    addr, (void *)((u_long)addr & (~UMA_SLAB_MASK)));
1030 
1031 	if (!malloc_large_slab(slab))
1032 		size = zone->uz_size;
1033 	else
1034 		size = malloc_large_size(slab);
1035 #endif
1036 	return (size);
1037 }
1038 
1039 CTASSERT(VM_KMEM_SIZE_SCALE >= 1);
1040 
1041 /*
1042  * Initialize the kernel memory (kmem) arena.
1043  */
1044 void
1045 kmeminit(void)
1046 {
1047 	u_long mem_size;
1048 	u_long tmp;
1049 
1050 #ifdef VM_KMEM_SIZE
1051 	if (vm_kmem_size == 0)
1052 		vm_kmem_size = VM_KMEM_SIZE;
1053 #endif
1054 #ifdef VM_KMEM_SIZE_MIN
1055 	if (vm_kmem_size_min == 0)
1056 		vm_kmem_size_min = VM_KMEM_SIZE_MIN;
1057 #endif
1058 #ifdef VM_KMEM_SIZE_MAX
1059 	if (vm_kmem_size_max == 0)
1060 		vm_kmem_size_max = VM_KMEM_SIZE_MAX;
1061 #endif
1062 	/*
1063 	 * Calculate the amount of kernel virtual address (KVA) space that is
1064 	 * preallocated to the kmem arena.  In order to support a wide range
1065 	 * of machines, it is a function of the physical memory size,
1066 	 * specifically,
1067 	 *
1068 	 *	min(max(physical memory size / VM_KMEM_SIZE_SCALE,
1069 	 *	    VM_KMEM_SIZE_MIN), VM_KMEM_SIZE_MAX)
1070 	 *
1071 	 * Every architecture must define an integral value for
1072 	 * VM_KMEM_SIZE_SCALE.  However, the definitions of VM_KMEM_SIZE_MIN
1073 	 * and VM_KMEM_SIZE_MAX, which represent respectively the floor and
1074 	 * ceiling on this preallocation, are optional.  Typically,
1075 	 * VM_KMEM_SIZE_MAX is itself a function of the available KVA space on
1076 	 * a given architecture.
1077 	 */
1078 	mem_size = vm_cnt.v_page_count;
1079 	if (mem_size <= 32768) /* delphij XXX 128MB */
1080 		kmem_zmax = PAGE_SIZE;
1081 
1082 	if (vm_kmem_size_scale < 1)
1083 		vm_kmem_size_scale = VM_KMEM_SIZE_SCALE;
1084 
1085 	/*
1086 	 * Check if we should use defaults for the "vm_kmem_size"
1087 	 * variable:
1088 	 */
1089 	if (vm_kmem_size == 0) {
1090 		vm_kmem_size = mem_size / vm_kmem_size_scale;
1091 		vm_kmem_size = vm_kmem_size * PAGE_SIZE < vm_kmem_size ?
1092 		    vm_kmem_size_max : vm_kmem_size * PAGE_SIZE;
1093 		if (vm_kmem_size_min > 0 && vm_kmem_size < vm_kmem_size_min)
1094 			vm_kmem_size = vm_kmem_size_min;
1095 		if (vm_kmem_size_max > 0 && vm_kmem_size >= vm_kmem_size_max)
1096 			vm_kmem_size = vm_kmem_size_max;
1097 	}
1098 	if (vm_kmem_size == 0)
1099 		panic("Tune VM_KMEM_SIZE_* for the platform");
1100 
1101 	/*
1102 	 * The amount of KVA space that is preallocated to the
1103 	 * kmem arena can be set statically at compile-time or manually
1104 	 * through the kernel environment.  However, it is still limited to
1105 	 * twice the physical memory size, which has been sufficient to handle
1106 	 * the most severe cases of external fragmentation in the kmem arena.
1107 	 */
1108 	if (vm_kmem_size / 2 / PAGE_SIZE > mem_size)
1109 		vm_kmem_size = 2 * mem_size * PAGE_SIZE;
1110 
1111 	vm_kmem_size = round_page(vm_kmem_size);
1112 #ifdef DEBUG_MEMGUARD
1113 	tmp = memguard_fudge(vm_kmem_size, kernel_map);
1114 #else
1115 	tmp = vm_kmem_size;
1116 #endif
1117 	uma_set_limit(tmp);
1118 
1119 #ifdef DEBUG_MEMGUARD
1120 	/*
1121 	 * Initialize MemGuard if support compiled in.  MemGuard is a
1122 	 * replacement allocator used for detecting tamper-after-free
1123 	 * scenarios as they occur.  It is only used for debugging.
1124 	 */
1125 	memguard_init(kernel_arena);
1126 #endif
1127 }
1128 
1129 /*
1130  * Initialize the kernel memory allocator
1131  */
1132 /* ARGSUSED*/
1133 static void
1134 mallocinit(void *dummy)
1135 {
1136 	int i;
1137 	uint8_t indx;
1138 
1139 	mtx_init(&malloc_mtx, "malloc", NULL, MTX_DEF);
1140 
1141 	kmeminit();
1142 
1143 	if (kmem_zmax < PAGE_SIZE || kmem_zmax > KMEM_ZMAX)
1144 		kmem_zmax = KMEM_ZMAX;
1145 
1146 	for (i = 0, indx = 0; kmemzones[indx].kz_size != 0; indx++) {
1147 		int size = kmemzones[indx].kz_size;
1148 		const char *name = kmemzones[indx].kz_name;
1149 		int subzone;
1150 
1151 		for (subzone = 0; subzone < numzones; subzone++) {
1152 			kmemzones[indx].kz_zone[subzone] =
1153 			    uma_zcreate(name, size,
1154 #ifdef INVARIANTS
1155 			    mtrash_ctor, mtrash_dtor, mtrash_init, mtrash_fini,
1156 #else
1157 			    NULL, NULL, NULL, NULL,
1158 #endif
1159 			    UMA_ALIGN_PTR, UMA_ZONE_MALLOC);
1160 		}
1161 		for (;i <= size; i+= KMEM_ZBASE)
1162 			kmemsize[i >> KMEM_ZSHIFT] = indx;
1163 	}
1164 }
1165 SYSINIT(kmem, SI_SUB_KMEM, SI_ORDER_SECOND, mallocinit, NULL);
1166 
1167 void
1168 malloc_init(void *data)
1169 {
1170 	struct malloc_type_internal *mtip;
1171 	struct malloc_type *mtp;
1172 
1173 	KASSERT(vm_cnt.v_page_count != 0, ("malloc_register before vm_init"));
1174 
1175 	mtp = data;
1176 	if (mtp->ks_version != M_VERSION)
1177 		panic("malloc_init: type %s with unsupported version %lu",
1178 		    mtp->ks_shortdesc, mtp->ks_version);
1179 
1180 	mtip = &mtp->ks_mti;
1181 	mtip->mti_stats = uma_zalloc_pcpu(pcpu_zone_64, M_WAITOK | M_ZERO);
1182 	mtp_set_subzone(mtp);
1183 
1184 	mtx_lock(&malloc_mtx);
1185 	mtp->ks_next = kmemstatistics;
1186 	kmemstatistics = mtp;
1187 	kmemcount++;
1188 	mtx_unlock(&malloc_mtx);
1189 }
1190 
1191 void
1192 malloc_uninit(void *data)
1193 {
1194 	struct malloc_type_internal *mtip;
1195 	struct malloc_type_stats *mtsp;
1196 	struct malloc_type *mtp, *temp;
1197 	long temp_allocs, temp_bytes;
1198 	int i;
1199 
1200 	mtp = data;
1201 	KASSERT(mtp->ks_version == M_VERSION,
1202 	    ("malloc_uninit: bad malloc type version"));
1203 
1204 	mtx_lock(&malloc_mtx);
1205 	mtip = &mtp->ks_mti;
1206 	if (mtp != kmemstatistics) {
1207 		for (temp = kmemstatistics; temp != NULL;
1208 		    temp = temp->ks_next) {
1209 			if (temp->ks_next == mtp) {
1210 				temp->ks_next = mtp->ks_next;
1211 				break;
1212 			}
1213 		}
1214 		KASSERT(temp,
1215 		    ("malloc_uninit: type '%s' not found", mtp->ks_shortdesc));
1216 	} else
1217 		kmemstatistics = mtp->ks_next;
1218 	kmemcount--;
1219 	mtx_unlock(&malloc_mtx);
1220 
1221 	/*
1222 	 * Look for memory leaks.
1223 	 */
1224 	temp_allocs = temp_bytes = 0;
1225 	for (i = 0; i <= mp_maxid; i++) {
1226 		mtsp = zpcpu_get_cpu(mtip->mti_stats, i);
1227 		temp_allocs += mtsp->mts_numallocs;
1228 		temp_allocs -= mtsp->mts_numfrees;
1229 		temp_bytes += mtsp->mts_memalloced;
1230 		temp_bytes -= mtsp->mts_memfreed;
1231 	}
1232 	if (temp_allocs > 0 || temp_bytes > 0) {
1233 		printf("Warning: memory type %s leaked memory on destroy "
1234 		    "(%ld allocations, %ld bytes leaked).\n", mtp->ks_shortdesc,
1235 		    temp_allocs, temp_bytes);
1236 	}
1237 
1238 	uma_zfree_pcpu(pcpu_zone_64, mtip->mti_stats);
1239 }
1240 
1241 struct malloc_type *
1242 malloc_desc2type(const char *desc)
1243 {
1244 	struct malloc_type *mtp;
1245 
1246 	mtx_assert(&malloc_mtx, MA_OWNED);
1247 	for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) {
1248 		if (strcmp(mtp->ks_shortdesc, desc) == 0)
1249 			return (mtp);
1250 	}
1251 	return (NULL);
1252 }
1253 
1254 static int
1255 sysctl_kern_malloc_stats(SYSCTL_HANDLER_ARGS)
1256 {
1257 	struct malloc_type_stream_header mtsh;
1258 	struct malloc_type_internal *mtip;
1259 	struct malloc_type_stats *mtsp, zeromts;
1260 	struct malloc_type_header mth;
1261 	struct malloc_type *mtp;
1262 	int error, i;
1263 	struct sbuf sbuf;
1264 
1265 	error = sysctl_wire_old_buffer(req, 0);
1266 	if (error != 0)
1267 		return (error);
1268 	sbuf_new_for_sysctl(&sbuf, NULL, 128, req);
1269 	sbuf_clear_flags(&sbuf, SBUF_INCLUDENUL);
1270 	mtx_lock(&malloc_mtx);
1271 
1272 	bzero(&zeromts, sizeof(zeromts));
1273 
1274 	/*
1275 	 * Insert stream header.
1276 	 */
1277 	bzero(&mtsh, sizeof(mtsh));
1278 	mtsh.mtsh_version = MALLOC_TYPE_STREAM_VERSION;
1279 	mtsh.mtsh_maxcpus = MAXCPU;
1280 	mtsh.mtsh_count = kmemcount;
1281 	(void)sbuf_bcat(&sbuf, &mtsh, sizeof(mtsh));
1282 
1283 	/*
1284 	 * Insert alternating sequence of type headers and type statistics.
1285 	 */
1286 	for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) {
1287 		mtip = &mtp->ks_mti;
1288 
1289 		/*
1290 		 * Insert type header.
1291 		 */
1292 		bzero(&mth, sizeof(mth));
1293 		strlcpy(mth.mth_name, mtp->ks_shortdesc, MALLOC_MAX_NAME);
1294 		(void)sbuf_bcat(&sbuf, &mth, sizeof(mth));
1295 
1296 		/*
1297 		 * Insert type statistics for each CPU.
1298 		 */
1299 		for (i = 0; i <= mp_maxid; i++) {
1300 			mtsp = zpcpu_get_cpu(mtip->mti_stats, i);
1301 			(void)sbuf_bcat(&sbuf, mtsp, sizeof(*mtsp));
1302 		}
1303 		/*
1304 		 * Fill in the missing CPUs.
1305 		 */
1306 		for (; i < MAXCPU; i++) {
1307 			(void)sbuf_bcat(&sbuf, &zeromts, sizeof(zeromts));
1308 		}
1309 	}
1310 	mtx_unlock(&malloc_mtx);
1311 	error = sbuf_finish(&sbuf);
1312 	sbuf_delete(&sbuf);
1313 	return (error);
1314 }
1315 
1316 SYSCTL_PROC(_kern, OID_AUTO, malloc_stats,
1317     CTLFLAG_RD | CTLTYPE_STRUCT | CTLFLAG_MPSAFE, 0, 0,
1318     sysctl_kern_malloc_stats, "s,malloc_type_ustats",
1319     "Return malloc types");
1320 
1321 SYSCTL_INT(_kern, OID_AUTO, malloc_count, CTLFLAG_RD, &kmemcount, 0,
1322     "Count of kernel malloc types");
1323 
1324 void
1325 malloc_type_list(malloc_type_list_func_t *func, void *arg)
1326 {
1327 	struct malloc_type *mtp, **bufmtp;
1328 	int count, i;
1329 	size_t buflen;
1330 
1331 	mtx_lock(&malloc_mtx);
1332 restart:
1333 	mtx_assert(&malloc_mtx, MA_OWNED);
1334 	count = kmemcount;
1335 	mtx_unlock(&malloc_mtx);
1336 
1337 	buflen = sizeof(struct malloc_type *) * count;
1338 	bufmtp = malloc(buflen, M_TEMP, M_WAITOK);
1339 
1340 	mtx_lock(&malloc_mtx);
1341 
1342 	if (count < kmemcount) {
1343 		free(bufmtp, M_TEMP);
1344 		goto restart;
1345 	}
1346 
1347 	for (mtp = kmemstatistics, i = 0; mtp != NULL; mtp = mtp->ks_next, i++)
1348 		bufmtp[i] = mtp;
1349 
1350 	mtx_unlock(&malloc_mtx);
1351 
1352 	for (i = 0; i < count; i++)
1353 		(func)(bufmtp[i], arg);
1354 
1355 	free(bufmtp, M_TEMP);
1356 }
1357 
1358 #ifdef DDB
1359 static int64_t
1360 get_malloc_stats(const struct malloc_type_internal *mtip, uint64_t *allocs,
1361     uint64_t *inuse)
1362 {
1363 	const struct malloc_type_stats *mtsp;
1364 	uint64_t frees, alloced, freed;
1365 	int i;
1366 
1367 	*allocs = 0;
1368 	frees = 0;
1369 	alloced = 0;
1370 	freed = 0;
1371 	for (i = 0; i <= mp_maxid; i++) {
1372 		mtsp = zpcpu_get_cpu(mtip->mti_stats, i);
1373 
1374 		*allocs += mtsp->mts_numallocs;
1375 		frees += mtsp->mts_numfrees;
1376 		alloced += mtsp->mts_memalloced;
1377 		freed += mtsp->mts_memfreed;
1378 	}
1379 	*inuse = *allocs - frees;
1380 	return (alloced - freed);
1381 }
1382 
1383 DB_SHOW_COMMAND(malloc, db_show_malloc)
1384 {
1385 	const char *fmt_hdr, *fmt_entry;
1386 	struct malloc_type *mtp;
1387 	uint64_t allocs, inuse;
1388 	int64_t size;
1389 	/* variables for sorting */
1390 	struct malloc_type *last_mtype, *cur_mtype;
1391 	int64_t cur_size, last_size;
1392 	int ties;
1393 
1394 	if (modif[0] == 'i') {
1395 		fmt_hdr = "%s,%s,%s,%s\n";
1396 		fmt_entry = "\"%s\",%ju,%jdK,%ju\n";
1397 	} else {
1398 		fmt_hdr = "%18s %12s  %12s %12s\n";
1399 		fmt_entry = "%18s %12ju %12jdK %12ju\n";
1400 	}
1401 
1402 	db_printf(fmt_hdr, "Type", "InUse", "MemUse", "Requests");
1403 
1404 	/* Select sort, largest size first. */
1405 	last_mtype = NULL;
1406 	last_size = INT64_MAX;
1407 	for (;;) {
1408 		cur_mtype = NULL;
1409 		cur_size = -1;
1410 		ties = 0;
1411 
1412 		for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) {
1413 			/*
1414 			 * In the case of size ties, print out mtypes
1415 			 * in the order they are encountered.  That is,
1416 			 * when we encounter the most recently output
1417 			 * mtype, we have already printed all preceding
1418 			 * ties, and we must print all following ties.
1419 			 */
1420 			if (mtp == last_mtype) {
1421 				ties = 1;
1422 				continue;
1423 			}
1424 			size = get_malloc_stats(&mtp->ks_mti, &allocs,
1425 			    &inuse);
1426 			if (size > cur_size && size < last_size + ties) {
1427 				cur_size = size;
1428 				cur_mtype = mtp;
1429 			}
1430 		}
1431 		if (cur_mtype == NULL)
1432 			break;
1433 
1434 		size = get_malloc_stats(&cur_mtype->ks_mti, &allocs, &inuse);
1435 		db_printf(fmt_entry, cur_mtype->ks_shortdesc, inuse,
1436 		    howmany(size, 1024), allocs);
1437 
1438 		if (db_pager_quit)
1439 			break;
1440 
1441 		last_mtype = cur_mtype;
1442 		last_size = cur_size;
1443 	}
1444 }
1445 
1446 #if MALLOC_DEBUG_MAXZONES > 1
1447 DB_SHOW_COMMAND(multizone_matches, db_show_multizone_matches)
1448 {
1449 	struct malloc_type_internal *mtip;
1450 	struct malloc_type *mtp;
1451 	u_int subzone;
1452 
1453 	if (!have_addr) {
1454 		db_printf("Usage: show multizone_matches <malloc type/addr>\n");
1455 		return;
1456 	}
1457 	mtp = (void *)addr;
1458 	if (mtp->ks_version != M_VERSION) {
1459 		db_printf("Version %lx does not match expected %x\n",
1460 		    mtp->ks_version, M_VERSION);
1461 		return;
1462 	}
1463 
1464 	mtip = &mtp->ks_mti;
1465 	subzone = mtip->mti_zone;
1466 
1467 	for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) {
1468 		mtip = &mtp->ks_mti;
1469 		if (mtip->mti_zone != subzone)
1470 			continue;
1471 		db_printf("%s\n", mtp->ks_shortdesc);
1472 		if (db_pager_quit)
1473 			break;
1474 	}
1475 }
1476 #endif /* MALLOC_DEBUG_MAXZONES > 1 */
1477 #endif /* DDB */
1478