xref: /freebsd/sys/kern/kern_malloc.c (revision 5b56413d04e608379c9a306373554a8e4d321bc0)
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 
35 /*
36  * Kernel malloc(9) implementation -- general purpose kernel memory allocator
37  * based on memory types.  Back end is implemented using the UMA(9) zone
38  * allocator.  A set of fixed-size buckets are used for smaller allocations,
39  * and a special UMA allocation interface is used for larger allocations.
40  * Callers declare memory types, and statistics are maintained independently
41  * for each memory type.  Statistics are maintained per-CPU for performance
42  * reasons.  See malloc(9) and comments in malloc.h for a detailed
43  * description.
44  */
45 
46 #include <sys/cdefs.h>
47 #include "opt_ddb.h"
48 #include "opt_vm.h"
49 
50 #include <sys/param.h>
51 #include <sys/systm.h>
52 #include <sys/asan.h>
53 #include <sys/kdb.h>
54 #include <sys/kernel.h>
55 #include <sys/lock.h>
56 #include <sys/malloc.h>
57 #include <sys/msan.h>
58 #include <sys/mutex.h>
59 #include <sys/vmmeter.h>
60 #include <sys/proc.h>
61 #include <sys/queue.h>
62 #include <sys/sbuf.h>
63 #include <sys/smp.h>
64 #include <sys/sysctl.h>
65 #include <sys/time.h>
66 #include <sys/vmem.h>
67 #ifdef EPOCH_TRACE
68 #include <sys/epoch.h>
69 #endif
70 
71 #include <vm/vm.h>
72 #include <vm/pmap.h>
73 #include <vm/vm_domainset.h>
74 #include <vm/vm_pageout.h>
75 #include <vm/vm_param.h>
76 #include <vm/vm_kern.h>
77 #include <vm/vm_extern.h>
78 #include <vm/vm_map.h>
79 #include <vm/vm_page.h>
80 #include <vm/vm_phys.h>
81 #include <vm/vm_pagequeue.h>
82 #include <vm/uma.h>
83 #include <vm/uma_int.h>
84 #include <vm/uma_dbg.h>
85 
86 #ifdef DEBUG_MEMGUARD
87 #include <vm/memguard.h>
88 #endif
89 #ifdef DEBUG_REDZONE
90 #include <vm/redzone.h>
91 #endif
92 
93 #if defined(INVARIANTS) && defined(__i386__)
94 #include <machine/cpu.h>
95 #endif
96 
97 #include <ddb/ddb.h>
98 
99 #ifdef KDTRACE_HOOKS
100 #include <sys/dtrace_bsd.h>
101 
102 bool	__read_frequently			dtrace_malloc_enabled;
103 dtrace_malloc_probe_func_t __read_mostly	dtrace_malloc_probe;
104 #endif
105 
106 #if defined(INVARIANTS) || defined(MALLOC_MAKE_FAILURES) ||		\
107     defined(DEBUG_MEMGUARD) || defined(DEBUG_REDZONE)
108 #define	MALLOC_DEBUG	1
109 #endif
110 
111 #if defined(KASAN) || defined(DEBUG_REDZONE)
112 #define	DEBUG_REDZONE_ARG_DEF	, unsigned long osize
113 #define	DEBUG_REDZONE_ARG	, osize
114 #else
115 #define	DEBUG_REDZONE_ARG_DEF
116 #define	DEBUG_REDZONE_ARG
117 #endif
118 
119 typedef	enum {
120 	SLAB_COOKIE_SLAB_PTR		= 0x0,
121 	SLAB_COOKIE_MALLOC_LARGE	= 0x1,
122 	SLAB_COOKIE_CONTIG_MALLOC	= 0x2,
123 } slab_cookie_t;
124 #define	SLAB_COOKIE_MASK		0x3
125 #define	SLAB_COOKIE_SHIFT		2
126 #define	GET_SLAB_COOKIE(_slab)						\
127     ((slab_cookie_t)(uintptr_t)(_slab) & SLAB_COOKIE_MASK)
128 
129 /*
130  * When realloc() is called, if the new size is sufficiently smaller than
131  * the old size, realloc() will allocate a new, smaller block to avoid
132  * wasting memory. 'Sufficiently smaller' is defined as: newsize <=
133  * oldsize / 2^n, where REALLOC_FRACTION defines the value of 'n'.
134  */
135 #ifndef REALLOC_FRACTION
136 #define	REALLOC_FRACTION	1	/* new block if <= half the size */
137 #endif
138 
139 /*
140  * Centrally define some common malloc types.
141  */
142 MALLOC_DEFINE(M_CACHE, "cache", "Various Dynamically allocated caches");
143 MALLOC_DEFINE(M_DEVBUF, "devbuf", "device driver memory");
144 MALLOC_DEFINE(M_TEMP, "temp", "misc temporary data buffers");
145 
146 static struct malloc_type *kmemstatistics;
147 static int kmemcount;
148 
149 #define KMEM_ZSHIFT	4
150 #define KMEM_ZBASE	16
151 #define KMEM_ZMASK	(KMEM_ZBASE - 1)
152 
153 #define KMEM_ZMAX	65536
154 #define KMEM_ZSIZE	(KMEM_ZMAX >> KMEM_ZSHIFT)
155 static uint8_t kmemsize[KMEM_ZSIZE + 1];
156 
157 #ifndef MALLOC_DEBUG_MAXZONES
158 #define	MALLOC_DEBUG_MAXZONES	1
159 #endif
160 static int numzones = MALLOC_DEBUG_MAXZONES;
161 
162 /*
163  * Small malloc(9) memory allocations are allocated from a set of UMA buckets
164  * of various sizes.
165  *
166  * Warning: the layout of the struct is duplicated in libmemstat for KVM support.
167  *
168  * XXX: The comment here used to read "These won't be powers of two for
169  * long."  It's possible that a significant amount of wasted memory could be
170  * recovered by tuning the sizes of these buckets.
171  */
172 struct {
173 	int kz_size;
174 	const char *kz_name;
175 	uma_zone_t kz_zone[MALLOC_DEBUG_MAXZONES];
176 } kmemzones[] = {
177 	{16, "malloc-16", },
178 	{32, "malloc-32", },
179 	{64, "malloc-64", },
180 	{128, "malloc-128", },
181 	{256, "malloc-256", },
182 	{384, "malloc-384", },
183 	{512, "malloc-512", },
184 	{1024, "malloc-1024", },
185 	{2048, "malloc-2048", },
186 	{4096, "malloc-4096", },
187 	{8192, "malloc-8192", },
188 	{16384, "malloc-16384", },
189 	{32768, "malloc-32768", },
190 	{65536, "malloc-65536", },
191 	{0, NULL},
192 };
193 
194 u_long vm_kmem_size;
195 SYSCTL_ULONG(_vm, OID_AUTO, kmem_size, CTLFLAG_RDTUN, &vm_kmem_size, 0,
196     "Size of kernel memory");
197 
198 static u_long kmem_zmax = KMEM_ZMAX;
199 SYSCTL_ULONG(_vm, OID_AUTO, kmem_zmax, CTLFLAG_RDTUN, &kmem_zmax, 0,
200     "Maximum allocation size that malloc(9) would use UMA as backend");
201 
202 static u_long vm_kmem_size_min;
203 SYSCTL_ULONG(_vm, OID_AUTO, kmem_size_min, CTLFLAG_RDTUN, &vm_kmem_size_min, 0,
204     "Minimum size of kernel memory");
205 
206 static u_long vm_kmem_size_max;
207 SYSCTL_ULONG(_vm, OID_AUTO, kmem_size_max, CTLFLAG_RDTUN, &vm_kmem_size_max, 0,
208     "Maximum size of kernel memory");
209 
210 static u_int vm_kmem_size_scale;
211 SYSCTL_UINT(_vm, OID_AUTO, kmem_size_scale, CTLFLAG_RDTUN, &vm_kmem_size_scale, 0,
212     "Scale factor for kernel memory size");
213 
214 static int sysctl_kmem_map_size(SYSCTL_HANDLER_ARGS);
215 SYSCTL_PROC(_vm, OID_AUTO, kmem_map_size,
216     CTLFLAG_RD | CTLTYPE_ULONG | CTLFLAG_MPSAFE, NULL, 0,
217     sysctl_kmem_map_size, "LU", "Current kmem allocation size");
218 
219 static int sysctl_kmem_map_free(SYSCTL_HANDLER_ARGS);
220 SYSCTL_PROC(_vm, OID_AUTO, kmem_map_free,
221     CTLFLAG_RD | CTLTYPE_ULONG | CTLFLAG_MPSAFE, NULL, 0,
222     sysctl_kmem_map_free, "LU", "Free space in kmem");
223 
224 static SYSCTL_NODE(_vm, OID_AUTO, malloc, CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
225     "Malloc information");
226 
227 static u_int vm_malloc_zone_count = nitems(kmemzones);
228 SYSCTL_UINT(_vm_malloc, OID_AUTO, zone_count,
229     CTLFLAG_RD, &vm_malloc_zone_count, 0,
230     "Number of malloc zones");
231 
232 static int sysctl_vm_malloc_zone_sizes(SYSCTL_HANDLER_ARGS);
233 SYSCTL_PROC(_vm_malloc, OID_AUTO, zone_sizes,
234     CTLFLAG_RD | CTLTYPE_OPAQUE | CTLFLAG_MPSAFE, NULL, 0,
235     sysctl_vm_malloc_zone_sizes, "S", "Zone sizes used by malloc");
236 
237 /*
238  * The malloc_mtx protects the kmemstatistics linked list.
239  */
240 struct mtx malloc_mtx;
241 
242 static int sysctl_kern_malloc_stats(SYSCTL_HANDLER_ARGS);
243 
244 #if defined(MALLOC_MAKE_FAILURES) || (MALLOC_DEBUG_MAXZONES > 1)
245 static SYSCTL_NODE(_debug, OID_AUTO, malloc, CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
246     "Kernel malloc debugging options");
247 #endif
248 
249 /*
250  * malloc(9) fault injection -- cause malloc failures every (n) mallocs when
251  * the caller specifies M_NOWAIT.  If set to 0, no failures are caused.
252  */
253 #ifdef MALLOC_MAKE_FAILURES
254 static int malloc_failure_rate;
255 static int malloc_nowait_count;
256 static int malloc_failure_count;
257 SYSCTL_INT(_debug_malloc, OID_AUTO, failure_rate, CTLFLAG_RWTUN,
258     &malloc_failure_rate, 0, "Every (n) mallocs with M_NOWAIT will fail");
259 SYSCTL_INT(_debug_malloc, OID_AUTO, failure_count, CTLFLAG_RD,
260     &malloc_failure_count, 0, "Number of imposed M_NOWAIT malloc failures");
261 #endif
262 
263 static int
264 sysctl_kmem_map_size(SYSCTL_HANDLER_ARGS)
265 {
266 	u_long size;
267 
268 	size = uma_size();
269 	return (sysctl_handle_long(oidp, &size, 0, req));
270 }
271 
272 static int
273 sysctl_kmem_map_free(SYSCTL_HANDLER_ARGS)
274 {
275 	u_long size, limit;
276 
277 	/* The sysctl is unsigned, implement as a saturation value. */
278 	size = uma_size();
279 	limit = uma_limit();
280 	if (size > limit)
281 		size = 0;
282 	else
283 		size = limit - size;
284 	return (sysctl_handle_long(oidp, &size, 0, req));
285 }
286 
287 static int
288 sysctl_vm_malloc_zone_sizes(SYSCTL_HANDLER_ARGS)
289 {
290 	int sizes[nitems(kmemzones)];
291 	int i;
292 
293 	for (i = 0; i < nitems(kmemzones); i++) {
294 		sizes[i] = kmemzones[i].kz_size;
295 	}
296 
297 	return (SYSCTL_OUT(req, &sizes, sizeof(sizes)));
298 }
299 
300 /*
301  * malloc(9) uma zone separation -- sub-page buffer overruns in one
302  * malloc type will affect only a subset of other malloc types.
303  */
304 #if MALLOC_DEBUG_MAXZONES > 1
305 static void
306 tunable_set_numzones(void)
307 {
308 
309 	TUNABLE_INT_FETCH("debug.malloc.numzones",
310 	    &numzones);
311 
312 	/* Sanity check the number of malloc uma zones. */
313 	if (numzones <= 0)
314 		numzones = 1;
315 	if (numzones > MALLOC_DEBUG_MAXZONES)
316 		numzones = MALLOC_DEBUG_MAXZONES;
317 }
318 SYSINIT(numzones, SI_SUB_TUNABLES, SI_ORDER_ANY, tunable_set_numzones, NULL);
319 SYSCTL_INT(_debug_malloc, OID_AUTO, numzones, CTLFLAG_RDTUN | CTLFLAG_NOFETCH,
320     &numzones, 0, "Number of malloc uma subzones");
321 
322 /*
323  * Any number that changes regularly is an okay choice for the
324  * offset.  Build numbers are pretty good of you have them.
325  */
326 static u_int zone_offset = __FreeBSD_version;
327 TUNABLE_INT("debug.malloc.zone_offset", &zone_offset);
328 SYSCTL_UINT(_debug_malloc, OID_AUTO, zone_offset, CTLFLAG_RDTUN,
329     &zone_offset, 0, "Separate malloc types by examining the "
330     "Nth character in the malloc type short description.");
331 
332 static void
333 mtp_set_subzone(struct malloc_type *mtp)
334 {
335 	struct malloc_type_internal *mtip;
336 	const char *desc;
337 	size_t len;
338 	u_int val;
339 
340 	mtip = &mtp->ks_mti;
341 	desc = mtp->ks_shortdesc;
342 	if (desc == NULL || (len = strlen(desc)) == 0)
343 		val = 0;
344 	else
345 		val = desc[zone_offset % len];
346 	mtip->mti_zone = (val % numzones);
347 }
348 
349 static inline u_int
350 mtp_get_subzone(struct malloc_type *mtp)
351 {
352 	struct malloc_type_internal *mtip;
353 
354 	mtip = &mtp->ks_mti;
355 
356 	KASSERT(mtip->mti_zone < numzones,
357 	    ("mti_zone %u out of range %d",
358 	    mtip->mti_zone, numzones));
359 	return (mtip->mti_zone);
360 }
361 #elif MALLOC_DEBUG_MAXZONES == 0
362 #error "MALLOC_DEBUG_MAXZONES must be positive."
363 #else
364 static void
365 mtp_set_subzone(struct malloc_type *mtp)
366 {
367 	struct malloc_type_internal *mtip;
368 
369 	mtip = &mtp->ks_mti;
370 	mtip->mti_zone = 0;
371 }
372 
373 static inline u_int
374 mtp_get_subzone(struct malloc_type *mtp)
375 {
376 
377 	return (0);
378 }
379 #endif /* MALLOC_DEBUG_MAXZONES > 1 */
380 
381 /*
382  * An allocation has succeeded -- update malloc type statistics for the
383  * amount of bucket size.  Occurs within a critical section so that the
384  * thread isn't preempted and doesn't migrate while updating per-PCU
385  * statistics.
386  */
387 static void
388 malloc_type_zone_allocated(struct malloc_type *mtp, unsigned long size,
389     int zindx)
390 {
391 	struct malloc_type_internal *mtip;
392 	struct malloc_type_stats *mtsp;
393 
394 	critical_enter();
395 	mtip = &mtp->ks_mti;
396 	mtsp = zpcpu_get(mtip->mti_stats);
397 	if (size > 0) {
398 		mtsp->mts_memalloced += size;
399 		mtsp->mts_numallocs++;
400 	}
401 	if (zindx != -1)
402 		mtsp->mts_size |= 1 << zindx;
403 
404 #ifdef KDTRACE_HOOKS
405 	if (__predict_false(dtrace_malloc_enabled)) {
406 		uint32_t probe_id = mtip->mti_probes[DTMALLOC_PROBE_MALLOC];
407 		if (probe_id != 0)
408 			(dtrace_malloc_probe)(probe_id,
409 			    (uintptr_t) mtp, (uintptr_t) mtip,
410 			    (uintptr_t) mtsp, size, zindx);
411 	}
412 #endif
413 
414 	critical_exit();
415 }
416 
417 void
418 malloc_type_allocated(struct malloc_type *mtp, unsigned long size)
419 {
420 
421 	if (size > 0)
422 		malloc_type_zone_allocated(mtp, size, -1);
423 }
424 
425 /*
426  * A free operation has occurred -- update malloc type statistics for the
427  * amount of the bucket size.  Occurs within a critical section so that the
428  * thread isn't preempted and doesn't migrate while updating per-CPU
429  * statistics.
430  */
431 void
432 malloc_type_freed(struct malloc_type *mtp, unsigned long size)
433 {
434 	struct malloc_type_internal *mtip;
435 	struct malloc_type_stats *mtsp;
436 
437 	critical_enter();
438 	mtip = &mtp->ks_mti;
439 	mtsp = zpcpu_get(mtip->mti_stats);
440 	mtsp->mts_memfreed += size;
441 	mtsp->mts_numfrees++;
442 
443 #ifdef KDTRACE_HOOKS
444 	if (__predict_false(dtrace_malloc_enabled)) {
445 		uint32_t probe_id = mtip->mti_probes[DTMALLOC_PROBE_FREE];
446 		if (probe_id != 0)
447 			(dtrace_malloc_probe)(probe_id,
448 			    (uintptr_t) mtp, (uintptr_t) mtip,
449 			    (uintptr_t) mtsp, size, 0);
450 	}
451 #endif
452 
453 	critical_exit();
454 }
455 
456 /*
457  *	contigmalloc:
458  *
459  *	Allocate a block of physically contiguous memory.
460  *
461  *	If M_NOWAIT is set, this routine will not block and return NULL if
462  *	the allocation fails.
463  */
464 #define	IS_CONTIG_MALLOC(_slab)						\
465     (GET_SLAB_COOKIE(_slab) == SLAB_COOKIE_CONTIG_MALLOC)
466 #define	CONTIG_MALLOC_SLAB(_size)					\
467     ((void *)(((_size) << SLAB_COOKIE_SHIFT) | SLAB_COOKIE_CONTIG_MALLOC))
468 static inline size_t
469 contigmalloc_size(uma_slab_t slab)
470 {
471 	uintptr_t va;
472 
473 	KASSERT(IS_CONTIG_MALLOC(slab),
474 	    ("%s: called on non-contigmalloc allocation: %p", __func__, slab));
475 	va = (uintptr_t)slab;
476 	return (va >> SLAB_COOKIE_SHIFT);
477 }
478 
479 void *
480 contigmalloc(unsigned long size, struct malloc_type *type, int flags,
481     vm_paddr_t low, vm_paddr_t high, unsigned long alignment,
482     vm_paddr_t boundary)
483 {
484 	void *ret;
485 
486 	ret = (void *)kmem_alloc_contig(size, flags, low, high, alignment,
487 	    boundary, VM_MEMATTR_DEFAULT);
488 	if (ret != NULL) {
489 		/* Use low bits unused for slab pointers. */
490 		vsetzoneslab((uintptr_t)ret, NULL, CONTIG_MALLOC_SLAB(size));
491 		malloc_type_allocated(type, round_page(size));
492 	}
493 	return (ret);
494 }
495 
496 void *
497 contigmalloc_domainset(unsigned long size, struct malloc_type *type,
498     struct domainset *ds, int flags, vm_paddr_t low, vm_paddr_t high,
499     unsigned long alignment, vm_paddr_t boundary)
500 {
501 	void *ret;
502 
503 	ret = (void *)kmem_alloc_contig_domainset(ds, size, flags, low, high,
504 	    alignment, boundary, VM_MEMATTR_DEFAULT);
505 	if (ret != NULL) {
506 		/* Use low bits unused for slab pointers. */
507 		vsetzoneslab((uintptr_t)ret, NULL, CONTIG_MALLOC_SLAB(size));
508 		malloc_type_allocated(type, round_page(size));
509 	}
510 	return (ret);
511 }
512 
513 /*
514  *	contigfree (deprecated).
515  *
516  *	Free a block of memory allocated by contigmalloc.
517  *
518  *	This routine may not block.
519  */
520 void
521 contigfree(void *addr, unsigned long size __unused, struct malloc_type *type)
522 {
523 	free(addr, type);
524 }
525 #undef	IS_CONTIG_MALLOC
526 #undef	CONTIG_MALLOC_SLAB
527 
528 #ifdef MALLOC_DEBUG
529 static int
530 malloc_dbg(caddr_t *vap, size_t *sizep, struct malloc_type *mtp,
531     int flags)
532 {
533 #ifdef INVARIANTS
534 	int indx;
535 
536 	KASSERT(mtp->ks_version == M_VERSION, ("malloc: bad malloc type version"));
537 	/*
538 	 * Check that exactly one of M_WAITOK or M_NOWAIT is specified.
539 	 */
540 	indx = flags & (M_WAITOK | M_NOWAIT);
541 	if (indx != M_NOWAIT && indx != M_WAITOK) {
542 		static	struct timeval lasterr;
543 		static	int curerr, once;
544 		if (once == 0 && ppsratecheck(&lasterr, &curerr, 1)) {
545 			printf("Bad malloc flags: %x\n", indx);
546 			kdb_backtrace();
547 			flags |= M_WAITOK;
548 			once++;
549 		}
550 	}
551 #endif
552 #ifdef MALLOC_MAKE_FAILURES
553 	if ((flags & M_NOWAIT) && (malloc_failure_rate != 0)) {
554 		atomic_add_int(&malloc_nowait_count, 1);
555 		if ((malloc_nowait_count % malloc_failure_rate) == 0) {
556 			atomic_add_int(&malloc_failure_count, 1);
557 			*vap = NULL;
558 			return (EJUSTRETURN);
559 		}
560 	}
561 #endif
562 	if (flags & M_WAITOK) {
563 		KASSERT(curthread->td_intr_nesting_level == 0,
564 		   ("malloc(M_WAITOK) in interrupt context"));
565 		if (__predict_false(!THREAD_CAN_SLEEP())) {
566 #ifdef EPOCH_TRACE
567 			epoch_trace_list(curthread);
568 #endif
569 			KASSERT(0,
570 			    ("malloc(M_WAITOK) with sleeping prohibited"));
571 		}
572 	}
573 	KASSERT(curthread->td_critnest == 0 || SCHEDULER_STOPPED(),
574 	    ("malloc: called with spinlock or critical section held"));
575 
576 #ifdef DEBUG_MEMGUARD
577 	if (memguard_cmp_mtp(mtp, *sizep)) {
578 		*vap = memguard_alloc(*sizep, flags);
579 		if (*vap != NULL)
580 			return (EJUSTRETURN);
581 		/* This is unfortunate but should not be fatal. */
582 	}
583 #endif
584 
585 #ifdef DEBUG_REDZONE
586 	*sizep = redzone_size_ntor(*sizep);
587 #endif
588 
589 	return (0);
590 }
591 #endif
592 
593 /*
594  * Handle large allocations and frees by using kmem_malloc directly.
595  */
596 #define	IS_MALLOC_LARGE(_slab)						\
597     (GET_SLAB_COOKIE(_slab) == SLAB_COOKIE_MALLOC_LARGE)
598 #define	MALLOC_LARGE_SLAB(_size)					\
599     ((void *)(((_size) << SLAB_COOKIE_SHIFT) | SLAB_COOKIE_MALLOC_LARGE))
600 static inline size_t
601 malloc_large_size(uma_slab_t slab)
602 {
603 	uintptr_t va;
604 
605 	va = (uintptr_t)slab;
606 	KASSERT(IS_MALLOC_LARGE(slab),
607 	    ("%s: called on non-malloc_large allocation: %p", __func__, slab));
608 	return (va >> SLAB_COOKIE_SHIFT);
609 }
610 
611 static caddr_t __noinline
612 malloc_large(size_t size, struct malloc_type *mtp, struct domainset *policy,
613     int flags DEBUG_REDZONE_ARG_DEF)
614 {
615 	void *va;
616 
617 	size = roundup(size, PAGE_SIZE);
618 	va = kmem_malloc_domainset(policy, size, flags);
619 	if (va != NULL) {
620 		/* Use low bits unused for slab pointers. */
621 		vsetzoneslab((uintptr_t)va, NULL, MALLOC_LARGE_SLAB(size));
622 		uma_total_inc(size);
623 	}
624 	malloc_type_allocated(mtp, va == NULL ? 0 : size);
625 	if (__predict_false(va == NULL)) {
626 		KASSERT((flags & M_WAITOK) == 0,
627 		    ("malloc(M_WAITOK) returned NULL"));
628 	} else {
629 #ifdef DEBUG_REDZONE
630 		va = redzone_setup(va, osize);
631 #endif
632 		kasan_mark(va, osize, size, KASAN_MALLOC_REDZONE);
633 	}
634 	return (va);
635 }
636 
637 static void
638 free_large(void *addr, size_t size)
639 {
640 
641 	kmem_free(addr, size);
642 	uma_total_dec(size);
643 }
644 #undef	IS_MALLOC_LARGE
645 #undef	MALLOC_LARGE_SLAB
646 
647 /*
648  *	malloc:
649  *
650  *	Allocate a block of memory.
651  *
652  *	If M_NOWAIT is set, this routine will not block and return NULL if
653  *	the allocation fails.
654  */
655 void *
656 (malloc)(size_t size, struct malloc_type *mtp, int flags)
657 {
658 	int indx;
659 	caddr_t va;
660 	uma_zone_t zone;
661 #if defined(DEBUG_REDZONE) || defined(KASAN)
662 	unsigned long osize = size;
663 #endif
664 
665 	MPASS((flags & M_EXEC) == 0);
666 
667 #ifdef MALLOC_DEBUG
668 	va = NULL;
669 	if (malloc_dbg(&va, &size, mtp, flags) != 0)
670 		return (va);
671 #endif
672 
673 	if (__predict_false(size > kmem_zmax))
674 		return (malloc_large(size, mtp, DOMAINSET_RR(), flags
675 		    DEBUG_REDZONE_ARG));
676 
677 	if (size & KMEM_ZMASK)
678 		size = (size & ~KMEM_ZMASK) + KMEM_ZBASE;
679 	indx = kmemsize[size >> KMEM_ZSHIFT];
680 	zone = kmemzones[indx].kz_zone[mtp_get_subzone(mtp)];
681 	va = uma_zalloc_arg(zone, zone, flags);
682 	if (va != NULL) {
683 		size = zone->uz_size;
684 		if ((flags & M_ZERO) == 0) {
685 			kmsan_mark(va, size, KMSAN_STATE_UNINIT);
686 			kmsan_orig(va, size, KMSAN_TYPE_MALLOC, KMSAN_RET_ADDR);
687 		}
688 	}
689 	malloc_type_zone_allocated(mtp, va == NULL ? 0 : size, indx);
690 	if (__predict_false(va == NULL)) {
691 		KASSERT((flags & M_WAITOK) == 0,
692 		    ("malloc(M_WAITOK) returned NULL"));
693 	}
694 #ifdef DEBUG_REDZONE
695 	if (va != NULL)
696 		va = redzone_setup(va, osize);
697 #endif
698 #ifdef KASAN
699 	if (va != NULL)
700 		kasan_mark((void *)va, osize, size, KASAN_MALLOC_REDZONE);
701 #endif
702 	return ((void *) va);
703 }
704 
705 static void *
706 malloc_domain(size_t *sizep, int *indxp, struct malloc_type *mtp, int domain,
707     int flags)
708 {
709 	uma_zone_t zone;
710 	caddr_t va;
711 	size_t size;
712 	int indx;
713 
714 	size = *sizep;
715 	KASSERT(size <= kmem_zmax && (flags & M_EXEC) == 0,
716 	    ("malloc_domain: Called with bad flag / size combination"));
717 	if (size & KMEM_ZMASK)
718 		size = (size & ~KMEM_ZMASK) + KMEM_ZBASE;
719 	indx = kmemsize[size >> KMEM_ZSHIFT];
720 	zone = kmemzones[indx].kz_zone[mtp_get_subzone(mtp)];
721 	va = uma_zalloc_domain(zone, zone, domain, flags);
722 	if (va != NULL)
723 		*sizep = zone->uz_size;
724 	*indxp = indx;
725 	return ((void *)va);
726 }
727 
728 void *
729 malloc_domainset(size_t size, struct malloc_type *mtp, struct domainset *ds,
730     int flags)
731 {
732 	struct vm_domainset_iter di;
733 	caddr_t va;
734 	int domain;
735 	int indx;
736 #if defined(KASAN) || defined(DEBUG_REDZONE)
737 	unsigned long osize = size;
738 #endif
739 
740 	MPASS((flags & M_EXEC) == 0);
741 
742 #ifdef MALLOC_DEBUG
743 	va = NULL;
744 	if (malloc_dbg(&va, &size, mtp, flags) != 0)
745 		return (va);
746 #endif
747 
748 	if (__predict_false(size > kmem_zmax))
749 		return (malloc_large(size, mtp, DOMAINSET_RR(), flags
750 		    DEBUG_REDZONE_ARG));
751 
752 	vm_domainset_iter_policy_init(&di, ds, &domain, &flags);
753 	do {
754 		va = malloc_domain(&size, &indx, mtp, domain, flags);
755 	} while (va == NULL && vm_domainset_iter_policy(&di, &domain) == 0);
756 	malloc_type_zone_allocated(mtp, va == NULL ? 0 : size, indx);
757 	if (__predict_false(va == NULL)) {
758 		KASSERT((flags & M_WAITOK) == 0,
759 		    ("malloc(M_WAITOK) returned NULL"));
760 	}
761 #ifdef DEBUG_REDZONE
762 	if (va != NULL)
763 		va = redzone_setup(va, osize);
764 #endif
765 #ifdef KASAN
766 	if (va != NULL)
767 		kasan_mark((void *)va, osize, size, KASAN_MALLOC_REDZONE);
768 #endif
769 #ifdef KMSAN
770 	if ((flags & M_ZERO) == 0) {
771 		kmsan_mark(va, size, KMSAN_STATE_UNINIT);
772 		kmsan_orig(va, size, KMSAN_TYPE_MALLOC, KMSAN_RET_ADDR);
773 	}
774 #endif
775 	return (va);
776 }
777 
778 /*
779  * Allocate an executable area.
780  */
781 void *
782 malloc_exec(size_t size, struct malloc_type *mtp, int flags)
783 {
784 
785 	return (malloc_domainset_exec(size, mtp, DOMAINSET_RR(), flags));
786 }
787 
788 void *
789 malloc_domainset_exec(size_t size, struct malloc_type *mtp, struct domainset *ds,
790     int flags)
791 {
792 #if defined(DEBUG_REDZONE) || defined(KASAN)
793 	unsigned long osize = size;
794 #endif
795 #ifdef MALLOC_DEBUG
796 	caddr_t va;
797 #endif
798 
799 	flags |= M_EXEC;
800 
801 #ifdef MALLOC_DEBUG
802 	va = NULL;
803 	if (malloc_dbg(&va, &size, mtp, flags) != 0)
804 		return (va);
805 #endif
806 
807 	return (malloc_large(size, mtp, ds, flags DEBUG_REDZONE_ARG));
808 }
809 
810 void *
811 malloc_aligned(size_t size, size_t align, struct malloc_type *type, int flags)
812 {
813 	return (malloc_domainset_aligned(size, align, type, DOMAINSET_RR(),
814 	    flags));
815 }
816 
817 void *
818 malloc_domainset_aligned(size_t size, size_t align,
819     struct malloc_type *mtp, struct domainset *ds, int flags)
820 {
821 	void *res;
822 	size_t asize;
823 
824 	KASSERT(powerof2(align),
825 	    ("malloc_domainset_aligned: wrong align %#zx size %#zx",
826 	    align, size));
827 	KASSERT(align <= PAGE_SIZE,
828 	    ("malloc_domainset_aligned: align %#zx (size %#zx) too large",
829 	    align, size));
830 
831 	/*
832 	 * Round the allocation size up to the next power of 2,
833 	 * because we can only guarantee alignment for
834 	 * power-of-2-sized allocations.  Further increase the
835 	 * allocation size to align if the rounded size is less than
836 	 * align, since malloc zones provide alignment equal to their
837 	 * size.
838 	 */
839 	if (size == 0)
840 		size = 1;
841 	asize = size <= align ? align : 1UL << flsl(size - 1);
842 
843 	res = malloc_domainset(asize, mtp, ds, flags);
844 	KASSERT(res == NULL || ((uintptr_t)res & (align - 1)) == 0,
845 	    ("malloc_domainset_aligned: result not aligned %p size %#zx "
846 	    "allocsize %#zx align %#zx", res, size, asize, align));
847 	return (res);
848 }
849 
850 void *
851 mallocarray(size_t nmemb, size_t size, struct malloc_type *type, int flags)
852 {
853 
854 	if (WOULD_OVERFLOW(nmemb, size))
855 		panic("mallocarray: %zu * %zu overflowed", nmemb, size);
856 
857 	return (malloc(size * nmemb, type, flags));
858 }
859 
860 void *
861 mallocarray_domainset(size_t nmemb, size_t size, struct malloc_type *type,
862     struct domainset *ds, int flags)
863 {
864 
865 	if (WOULD_OVERFLOW(nmemb, size))
866 		panic("mallocarray_domainset: %zu * %zu overflowed", nmemb, size);
867 
868 	return (malloc_domainset(size * nmemb, type, ds, flags));
869 }
870 
871 #if defined(INVARIANTS) && !defined(KASAN)
872 static void
873 free_save_type(void *addr, struct malloc_type *mtp, u_long size)
874 {
875 	struct malloc_type **mtpp = addr;
876 
877 	/*
878 	 * Cache a pointer to the malloc_type that most recently freed
879 	 * this memory here.  This way we know who is most likely to
880 	 * have stepped on it later.
881 	 *
882 	 * This code assumes that size is a multiple of 8 bytes for
883 	 * 64 bit machines
884 	 */
885 	mtpp = (struct malloc_type **) ((unsigned long)mtpp & ~UMA_ALIGN_PTR);
886 	mtpp += (size - sizeof(struct malloc_type *)) /
887 	    sizeof(struct malloc_type *);
888 	*mtpp = mtp;
889 }
890 #endif
891 
892 #ifdef MALLOC_DEBUG
893 static int
894 free_dbg(void **addrp, struct malloc_type *mtp)
895 {
896 	void *addr;
897 
898 	addr = *addrp;
899 	KASSERT(mtp->ks_version == M_VERSION, ("free: bad malloc type version"));
900 	KASSERT(curthread->td_critnest == 0 || SCHEDULER_STOPPED(),
901 	    ("free: called with spinlock or critical section held"));
902 
903 	/* free(NULL, ...) does nothing */
904 	if (addr == NULL)
905 		return (EJUSTRETURN);
906 
907 #ifdef DEBUG_MEMGUARD
908 	if (is_memguard_addr(addr)) {
909 		memguard_free(addr);
910 		return (EJUSTRETURN);
911 	}
912 #endif
913 
914 #ifdef DEBUG_REDZONE
915 	redzone_check(addr);
916 	*addrp = redzone_addr_ntor(addr);
917 #endif
918 
919 	return (0);
920 }
921 #endif
922 
923 /*
924  *	free:
925  *
926  *	Free a block of memory allocated by malloc.
927  *
928  *	This routine may not block.
929  */
930 void
931 free(void *addr, struct malloc_type *mtp)
932 {
933 	uma_zone_t zone;
934 	uma_slab_t slab;
935 	u_long size;
936 
937 #ifdef MALLOC_DEBUG
938 	if (free_dbg(&addr, mtp) != 0)
939 		return;
940 #endif
941 	/* free(NULL, ...) does nothing */
942 	if (addr == NULL)
943 		return;
944 
945 	vtozoneslab((vm_offset_t)addr & (~UMA_SLAB_MASK), &zone, &slab);
946 	if (slab == NULL)
947 		panic("free: address %p(%p) has not been allocated",
948 		    addr, (void *)((u_long)addr & (~UMA_SLAB_MASK)));
949 
950 	switch (GET_SLAB_COOKIE(slab)) {
951 	case __predict_true(SLAB_COOKIE_SLAB_PTR):
952 		size = zone->uz_size;
953 #if defined(INVARIANTS) && !defined(KASAN)
954 		free_save_type(addr, mtp, size);
955 #endif
956 		uma_zfree_arg(zone, addr, slab);
957 		break;
958 	case SLAB_COOKIE_MALLOC_LARGE:
959 		size = malloc_large_size(slab);
960 		free_large(addr, size);
961 		break;
962 	case SLAB_COOKIE_CONTIG_MALLOC:
963 		size = contigmalloc_size(slab);
964 		kmem_free(addr, size);
965 		size = round_page(size);
966 		break;
967 	default:
968 		panic("%s: addr %p slab %p with unknown cookie %d", __func__,
969 		    addr, slab, GET_SLAB_COOKIE(slab));
970 		/* NOTREACHED */
971 	}
972 	malloc_type_freed(mtp, size);
973 }
974 
975 /*
976  *	zfree:
977  *
978  *	Zero then free a block of memory allocated by malloc.
979  *
980  *	This routine may not block.
981  */
982 void
983 zfree(void *addr, struct malloc_type *mtp)
984 {
985 	uma_zone_t zone;
986 	uma_slab_t slab;
987 	u_long size;
988 
989 #ifdef MALLOC_DEBUG
990 	if (free_dbg(&addr, mtp) != 0)
991 		return;
992 #endif
993 	/* free(NULL, ...) does nothing */
994 	if (addr == NULL)
995 		return;
996 
997 	vtozoneslab((vm_offset_t)addr & (~UMA_SLAB_MASK), &zone, &slab);
998 	if (slab == NULL)
999 		panic("free: address %p(%p) has not been allocated",
1000 		    addr, (void *)((u_long)addr & (~UMA_SLAB_MASK)));
1001 
1002 	switch (GET_SLAB_COOKIE(slab)) {
1003 	case __predict_true(SLAB_COOKIE_SLAB_PTR):
1004 		size = zone->uz_size;
1005 #if defined(INVARIANTS) && !defined(KASAN)
1006 		free_save_type(addr, mtp, size);
1007 #endif
1008 		kasan_mark(addr, size, size, 0);
1009 		explicit_bzero(addr, size);
1010 		uma_zfree_arg(zone, addr, slab);
1011 		break;
1012 	case SLAB_COOKIE_MALLOC_LARGE:
1013 		size = malloc_large_size(slab);
1014 		kasan_mark(addr, size, size, 0);
1015 		explicit_bzero(addr, size);
1016 		free_large(addr, size);
1017 		break;
1018 	case SLAB_COOKIE_CONTIG_MALLOC:
1019 		size = round_page(contigmalloc_size(slab));
1020 		explicit_bzero(addr, size);
1021 		kmem_free(addr, size);
1022 		break;
1023 	default:
1024 		panic("%s: addr %p slab %p with unknown cookie %d", __func__,
1025 		    addr, slab, GET_SLAB_COOKIE(slab));
1026 		/* NOTREACHED */
1027 	}
1028 	malloc_type_freed(mtp, size);
1029 }
1030 
1031 /*
1032  *	realloc: change the size of a memory block
1033  */
1034 void *
1035 realloc(void *addr, size_t size, struct malloc_type *mtp, int flags)
1036 {
1037 #ifndef DEBUG_REDZONE
1038 	uma_zone_t zone;
1039 	uma_slab_t slab;
1040 #endif
1041 	unsigned long alloc;
1042 	void *newaddr;
1043 
1044 	KASSERT(mtp->ks_version == M_VERSION,
1045 	    ("realloc: bad malloc type version"));
1046 	KASSERT(curthread->td_critnest == 0 || SCHEDULER_STOPPED(),
1047 	    ("realloc: called with spinlock or critical section held"));
1048 
1049 	/* realloc(NULL, ...) is equivalent to malloc(...) */
1050 	if (addr == NULL)
1051 		return (malloc(size, mtp, flags));
1052 
1053 	/*
1054 	 * XXX: Should report free of old memory and alloc of new memory to
1055 	 * per-CPU stats.
1056 	 */
1057 
1058 #ifdef DEBUG_MEMGUARD
1059 	if (is_memguard_addr(addr))
1060 		return (memguard_realloc(addr, size, mtp, flags));
1061 #endif
1062 
1063 #ifdef DEBUG_REDZONE
1064 	alloc = redzone_get_size(addr);
1065 #else
1066 	vtozoneslab((vm_offset_t)addr & (~UMA_SLAB_MASK), &zone, &slab);
1067 
1068 	/* Sanity check */
1069 	KASSERT(slab != NULL,
1070 	    ("realloc: address %p out of range", (void *)addr));
1071 
1072 	/* Get the size of the original block */
1073 	switch (GET_SLAB_COOKIE(slab)) {
1074 	case __predict_true(SLAB_COOKIE_SLAB_PTR):
1075 		alloc = zone->uz_size;
1076 		break;
1077 	case SLAB_COOKIE_MALLOC_LARGE:
1078 		alloc = malloc_large_size(slab);
1079 		break;
1080 	default:
1081 #ifdef INVARIANTS
1082 		panic("%s: called for addr %p of unsupported allocation type; "
1083 		    "slab %p cookie %d", __func__, addr, slab, GET_SLAB_COOKIE(slab));
1084 #endif
1085 		return (NULL);
1086 	}
1087 
1088 	/* Reuse the original block if appropriate */
1089 	if (size <= alloc &&
1090 	    (size > (alloc >> REALLOC_FRACTION) || alloc == MINALLOCSIZE)) {
1091 		kasan_mark((void *)addr, size, alloc, KASAN_MALLOC_REDZONE);
1092 		return (addr);
1093 	}
1094 #endif /* !DEBUG_REDZONE */
1095 
1096 	/* Allocate a new, bigger (or smaller) block */
1097 	if ((newaddr = malloc(size, mtp, flags)) == NULL)
1098 		return (NULL);
1099 
1100 	/*
1101 	 * Copy over original contents.  For KASAN, the redzone must be marked
1102 	 * valid before performing the copy.
1103 	 */
1104 	kasan_mark(addr, alloc, alloc, 0);
1105 	bcopy(addr, newaddr, min(size, alloc));
1106 	free(addr, mtp);
1107 	return (newaddr);
1108 }
1109 
1110 /*
1111  *	reallocf: same as realloc() but free memory on failure.
1112  */
1113 void *
1114 reallocf(void *addr, size_t size, struct malloc_type *mtp, int flags)
1115 {
1116 	void *mem;
1117 
1118 	if ((mem = realloc(addr, size, mtp, flags)) == NULL)
1119 		free(addr, mtp);
1120 	return (mem);
1121 }
1122 
1123 /*
1124  * 	malloc_size: returns the number of bytes allocated for a request of the
1125  * 		     specified size
1126  */
1127 size_t
1128 malloc_size(size_t size)
1129 {
1130 	int indx;
1131 
1132 	if (size > kmem_zmax)
1133 		return (round_page(size));
1134 	if (size & KMEM_ZMASK)
1135 		size = (size & ~KMEM_ZMASK) + KMEM_ZBASE;
1136 	indx = kmemsize[size >> KMEM_ZSHIFT];
1137 	return (kmemzones[indx].kz_size);
1138 }
1139 
1140 /*
1141  *	malloc_usable_size: returns the usable size of the allocation.
1142  */
1143 size_t
1144 malloc_usable_size(const void *addr)
1145 {
1146 #ifndef DEBUG_REDZONE
1147 	uma_zone_t zone;
1148 	uma_slab_t slab;
1149 #endif
1150 	u_long size;
1151 
1152 	if (addr == NULL)
1153 		return (0);
1154 
1155 #ifdef DEBUG_MEMGUARD
1156 	if (is_memguard_addr(__DECONST(void *, addr)))
1157 		return (memguard_get_req_size(addr));
1158 #endif
1159 
1160 #ifdef DEBUG_REDZONE
1161 	size = redzone_get_size(__DECONST(void *, addr));
1162 #else
1163 	vtozoneslab((vm_offset_t)addr & (~UMA_SLAB_MASK), &zone, &slab);
1164 	if (slab == NULL)
1165 		panic("malloc_usable_size: address %p(%p) is not allocated",
1166 		    addr, (void *)((u_long)addr & (~UMA_SLAB_MASK)));
1167 
1168 	switch (GET_SLAB_COOKIE(slab)) {
1169 	case __predict_true(SLAB_COOKIE_SLAB_PTR):
1170 		size = zone->uz_size;
1171 		break;
1172 	case SLAB_COOKIE_MALLOC_LARGE:
1173 		size = malloc_large_size(slab);
1174 		break;
1175 	default:
1176 		__assert_unreachable();
1177 		size = 0;
1178 		break;
1179 	}
1180 #endif
1181 
1182 	/*
1183 	 * Unmark the redzone to avoid reports from consumers who are
1184 	 * (presumably) about to use the full allocation size.
1185 	 */
1186 	kasan_mark(addr, size, size, 0);
1187 
1188 	return (size);
1189 }
1190 
1191 CTASSERT(VM_KMEM_SIZE_SCALE >= 1);
1192 
1193 /*
1194  * Initialize the kernel memory (kmem) arena.
1195  */
1196 void
1197 kmeminit(void)
1198 {
1199 	u_long mem_size;
1200 	u_long tmp;
1201 
1202 #ifdef VM_KMEM_SIZE
1203 	if (vm_kmem_size == 0)
1204 		vm_kmem_size = VM_KMEM_SIZE;
1205 #endif
1206 #ifdef VM_KMEM_SIZE_MIN
1207 	if (vm_kmem_size_min == 0)
1208 		vm_kmem_size_min = VM_KMEM_SIZE_MIN;
1209 #endif
1210 #ifdef VM_KMEM_SIZE_MAX
1211 	if (vm_kmem_size_max == 0)
1212 		vm_kmem_size_max = VM_KMEM_SIZE_MAX;
1213 #endif
1214 	/*
1215 	 * Calculate the amount of kernel virtual address (KVA) space that is
1216 	 * preallocated to the kmem arena.  In order to support a wide range
1217 	 * of machines, it is a function of the physical memory size,
1218 	 * specifically,
1219 	 *
1220 	 *	min(max(physical memory size / VM_KMEM_SIZE_SCALE,
1221 	 *	    VM_KMEM_SIZE_MIN), VM_KMEM_SIZE_MAX)
1222 	 *
1223 	 * Every architecture must define an integral value for
1224 	 * VM_KMEM_SIZE_SCALE.  However, the definitions of VM_KMEM_SIZE_MIN
1225 	 * and VM_KMEM_SIZE_MAX, which represent respectively the floor and
1226 	 * ceiling on this preallocation, are optional.  Typically,
1227 	 * VM_KMEM_SIZE_MAX is itself a function of the available KVA space on
1228 	 * a given architecture.
1229 	 */
1230 	mem_size = vm_cnt.v_page_count;
1231 	if (mem_size <= 32768) /* delphij XXX 128MB */
1232 		kmem_zmax = PAGE_SIZE;
1233 
1234 	if (vm_kmem_size_scale < 1)
1235 		vm_kmem_size_scale = VM_KMEM_SIZE_SCALE;
1236 
1237 	/*
1238 	 * Check if we should use defaults for the "vm_kmem_size"
1239 	 * variable:
1240 	 */
1241 	if (vm_kmem_size == 0) {
1242 		vm_kmem_size = mem_size / vm_kmem_size_scale;
1243 		vm_kmem_size = vm_kmem_size * PAGE_SIZE < vm_kmem_size ?
1244 		    vm_kmem_size_max : vm_kmem_size * PAGE_SIZE;
1245 		if (vm_kmem_size_min > 0 && vm_kmem_size < vm_kmem_size_min)
1246 			vm_kmem_size = vm_kmem_size_min;
1247 		if (vm_kmem_size_max > 0 && vm_kmem_size >= vm_kmem_size_max)
1248 			vm_kmem_size = vm_kmem_size_max;
1249 	}
1250 	if (vm_kmem_size == 0)
1251 		panic("Tune VM_KMEM_SIZE_* for the platform");
1252 
1253 	/*
1254 	 * The amount of KVA space that is preallocated to the
1255 	 * kmem arena can be set statically at compile-time or manually
1256 	 * through the kernel environment.  However, it is still limited to
1257 	 * twice the physical memory size, which has been sufficient to handle
1258 	 * the most severe cases of external fragmentation in the kmem arena.
1259 	 */
1260 	if (vm_kmem_size / 2 / PAGE_SIZE > mem_size)
1261 		vm_kmem_size = 2 * mem_size * PAGE_SIZE;
1262 
1263 	vm_kmem_size = round_page(vm_kmem_size);
1264 
1265 	/*
1266 	 * With KASAN or KMSAN enabled, dynamically allocated kernel memory is
1267 	 * shadowed.  Account for this when setting the UMA limit.
1268 	 */
1269 #if defined(KASAN)
1270 	vm_kmem_size = (vm_kmem_size * KASAN_SHADOW_SCALE) /
1271 	    (KASAN_SHADOW_SCALE + 1);
1272 #elif defined(KMSAN)
1273 	vm_kmem_size /= 3;
1274 #endif
1275 
1276 #ifdef DEBUG_MEMGUARD
1277 	tmp = memguard_fudge(vm_kmem_size, kernel_map);
1278 #else
1279 	tmp = vm_kmem_size;
1280 #endif
1281 	uma_set_limit(tmp);
1282 
1283 #ifdef DEBUG_MEMGUARD
1284 	/*
1285 	 * Initialize MemGuard if support compiled in.  MemGuard is a
1286 	 * replacement allocator used for detecting tamper-after-free
1287 	 * scenarios as they occur.  It is only used for debugging.
1288 	 */
1289 	memguard_init(kernel_arena);
1290 #endif
1291 }
1292 
1293 /*
1294  * Initialize the kernel memory allocator
1295  */
1296 /* ARGSUSED*/
1297 static void
1298 mallocinit(void *dummy)
1299 {
1300 	int i;
1301 	uint8_t indx;
1302 
1303 	mtx_init(&malloc_mtx, "malloc", NULL, MTX_DEF);
1304 
1305 	kmeminit();
1306 
1307 	if (kmem_zmax < PAGE_SIZE || kmem_zmax > KMEM_ZMAX)
1308 		kmem_zmax = KMEM_ZMAX;
1309 
1310 	for (i = 0, indx = 0; kmemzones[indx].kz_size != 0; indx++) {
1311 		int size = kmemzones[indx].kz_size;
1312 		const char *name = kmemzones[indx].kz_name;
1313 		size_t align;
1314 		int subzone;
1315 
1316 		align = UMA_ALIGN_PTR;
1317 		if (powerof2(size) && size > sizeof(void *))
1318 			align = MIN(size, PAGE_SIZE) - 1;
1319 		for (subzone = 0; subzone < numzones; subzone++) {
1320 			kmemzones[indx].kz_zone[subzone] =
1321 			    uma_zcreate(name, size,
1322 #if defined(INVARIANTS) && !defined(KASAN) && !defined(KMSAN)
1323 			    mtrash_ctor, mtrash_dtor, mtrash_init, mtrash_fini,
1324 #else
1325 			    NULL, NULL, NULL, NULL,
1326 #endif
1327 			    align, UMA_ZONE_MALLOC);
1328 		}
1329 		for (;i <= size; i+= KMEM_ZBASE)
1330 			kmemsize[i >> KMEM_ZSHIFT] = indx;
1331 	}
1332 }
1333 SYSINIT(kmem, SI_SUB_KMEM, SI_ORDER_SECOND, mallocinit, NULL);
1334 
1335 void
1336 malloc_init(void *data)
1337 {
1338 	struct malloc_type_internal *mtip;
1339 	struct malloc_type *mtp;
1340 
1341 	KASSERT(vm_cnt.v_page_count != 0,
1342 	    ("malloc_init() called before vm_mem_init()"));
1343 
1344 	mtp = data;
1345 	if (mtp->ks_version != M_VERSION)
1346 		panic("malloc_init: type %s with unsupported version %lu",
1347 		    mtp->ks_shortdesc, mtp->ks_version);
1348 
1349 	mtip = &mtp->ks_mti;
1350 	mtip->mti_stats = uma_zalloc_pcpu(pcpu_zone_64, M_WAITOK | M_ZERO);
1351 	mtp_set_subzone(mtp);
1352 
1353 	mtx_lock(&malloc_mtx);
1354 	mtp->ks_next = kmemstatistics;
1355 	kmemstatistics = mtp;
1356 	kmemcount++;
1357 	mtx_unlock(&malloc_mtx);
1358 }
1359 
1360 void
1361 malloc_uninit(void *data)
1362 {
1363 	struct malloc_type_internal *mtip;
1364 	struct malloc_type_stats *mtsp;
1365 	struct malloc_type *mtp, *temp;
1366 	long temp_allocs, temp_bytes;
1367 	int i;
1368 
1369 	mtp = data;
1370 	KASSERT(mtp->ks_version == M_VERSION,
1371 	    ("malloc_uninit: bad malloc type version"));
1372 
1373 	mtx_lock(&malloc_mtx);
1374 	mtip = &mtp->ks_mti;
1375 	if (mtp != kmemstatistics) {
1376 		for (temp = kmemstatistics; temp != NULL;
1377 		    temp = temp->ks_next) {
1378 			if (temp->ks_next == mtp) {
1379 				temp->ks_next = mtp->ks_next;
1380 				break;
1381 			}
1382 		}
1383 		KASSERT(temp,
1384 		    ("malloc_uninit: type '%s' not found", mtp->ks_shortdesc));
1385 	} else
1386 		kmemstatistics = mtp->ks_next;
1387 	kmemcount--;
1388 	mtx_unlock(&malloc_mtx);
1389 
1390 	/*
1391 	 * Look for memory leaks.
1392 	 */
1393 	temp_allocs = temp_bytes = 0;
1394 	for (i = 0; i <= mp_maxid; i++) {
1395 		mtsp = zpcpu_get_cpu(mtip->mti_stats, i);
1396 		temp_allocs += mtsp->mts_numallocs;
1397 		temp_allocs -= mtsp->mts_numfrees;
1398 		temp_bytes += mtsp->mts_memalloced;
1399 		temp_bytes -= mtsp->mts_memfreed;
1400 	}
1401 	if (temp_allocs > 0 || temp_bytes > 0) {
1402 		printf("Warning: memory type %s leaked memory on destroy "
1403 		    "(%ld allocations, %ld bytes leaked).\n", mtp->ks_shortdesc,
1404 		    temp_allocs, temp_bytes);
1405 	}
1406 
1407 	uma_zfree_pcpu(pcpu_zone_64, mtip->mti_stats);
1408 }
1409 
1410 struct malloc_type *
1411 malloc_desc2type(const char *desc)
1412 {
1413 	struct malloc_type *mtp;
1414 
1415 	mtx_assert(&malloc_mtx, MA_OWNED);
1416 	for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) {
1417 		if (strcmp(mtp->ks_shortdesc, desc) == 0)
1418 			return (mtp);
1419 	}
1420 	return (NULL);
1421 }
1422 
1423 static int
1424 sysctl_kern_malloc_stats(SYSCTL_HANDLER_ARGS)
1425 {
1426 	struct malloc_type_stream_header mtsh;
1427 	struct malloc_type_internal *mtip;
1428 	struct malloc_type_stats *mtsp, zeromts;
1429 	struct malloc_type_header mth;
1430 	struct malloc_type *mtp;
1431 	int error, i;
1432 	struct sbuf sbuf;
1433 
1434 	error = sysctl_wire_old_buffer(req, 0);
1435 	if (error != 0)
1436 		return (error);
1437 	sbuf_new_for_sysctl(&sbuf, NULL, 128, req);
1438 	sbuf_clear_flags(&sbuf, SBUF_INCLUDENUL);
1439 	mtx_lock(&malloc_mtx);
1440 
1441 	bzero(&zeromts, sizeof(zeromts));
1442 
1443 	/*
1444 	 * Insert stream header.
1445 	 */
1446 	bzero(&mtsh, sizeof(mtsh));
1447 	mtsh.mtsh_version = MALLOC_TYPE_STREAM_VERSION;
1448 	mtsh.mtsh_maxcpus = MAXCPU;
1449 	mtsh.mtsh_count = kmemcount;
1450 	(void)sbuf_bcat(&sbuf, &mtsh, sizeof(mtsh));
1451 
1452 	/*
1453 	 * Insert alternating sequence of type headers and type statistics.
1454 	 */
1455 	for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) {
1456 		mtip = &mtp->ks_mti;
1457 
1458 		/*
1459 		 * Insert type header.
1460 		 */
1461 		bzero(&mth, sizeof(mth));
1462 		strlcpy(mth.mth_name, mtp->ks_shortdesc, MALLOC_MAX_NAME);
1463 		(void)sbuf_bcat(&sbuf, &mth, sizeof(mth));
1464 
1465 		/*
1466 		 * Insert type statistics for each CPU.
1467 		 */
1468 		for (i = 0; i <= mp_maxid; i++) {
1469 			mtsp = zpcpu_get_cpu(mtip->mti_stats, i);
1470 			(void)sbuf_bcat(&sbuf, mtsp, sizeof(*mtsp));
1471 		}
1472 		/*
1473 		 * Fill in the missing CPUs.
1474 		 */
1475 		for (; i < MAXCPU; i++) {
1476 			(void)sbuf_bcat(&sbuf, &zeromts, sizeof(zeromts));
1477 		}
1478 	}
1479 	mtx_unlock(&malloc_mtx);
1480 	error = sbuf_finish(&sbuf);
1481 	sbuf_delete(&sbuf);
1482 	return (error);
1483 }
1484 
1485 SYSCTL_PROC(_kern, OID_AUTO, malloc_stats,
1486     CTLFLAG_RD | CTLTYPE_STRUCT | CTLFLAG_MPSAFE, 0, 0,
1487     sysctl_kern_malloc_stats, "s,malloc_type_ustats",
1488     "Return malloc types");
1489 
1490 SYSCTL_INT(_kern, OID_AUTO, malloc_count, CTLFLAG_RD, &kmemcount, 0,
1491     "Count of kernel malloc types");
1492 
1493 void
1494 malloc_type_list(malloc_type_list_func_t *func, void *arg)
1495 {
1496 	struct malloc_type *mtp, **bufmtp;
1497 	int count, i;
1498 	size_t buflen;
1499 
1500 	mtx_lock(&malloc_mtx);
1501 restart:
1502 	mtx_assert(&malloc_mtx, MA_OWNED);
1503 	count = kmemcount;
1504 	mtx_unlock(&malloc_mtx);
1505 
1506 	buflen = sizeof(struct malloc_type *) * count;
1507 	bufmtp = malloc(buflen, M_TEMP, M_WAITOK);
1508 
1509 	mtx_lock(&malloc_mtx);
1510 
1511 	if (count < kmemcount) {
1512 		free(bufmtp, M_TEMP);
1513 		goto restart;
1514 	}
1515 
1516 	for (mtp = kmemstatistics, i = 0; mtp != NULL; mtp = mtp->ks_next, i++)
1517 		bufmtp[i] = mtp;
1518 
1519 	mtx_unlock(&malloc_mtx);
1520 
1521 	for (i = 0; i < count; i++)
1522 		(func)(bufmtp[i], arg);
1523 
1524 	free(bufmtp, M_TEMP);
1525 }
1526 
1527 #ifdef DDB
1528 static int64_t
1529 get_malloc_stats(const struct malloc_type_internal *mtip, uint64_t *allocs,
1530     uint64_t *inuse)
1531 {
1532 	const struct malloc_type_stats *mtsp;
1533 	uint64_t frees, alloced, freed;
1534 	int i;
1535 
1536 	*allocs = 0;
1537 	frees = 0;
1538 	alloced = 0;
1539 	freed = 0;
1540 	for (i = 0; i <= mp_maxid; i++) {
1541 		mtsp = zpcpu_get_cpu(mtip->mti_stats, i);
1542 
1543 		*allocs += mtsp->mts_numallocs;
1544 		frees += mtsp->mts_numfrees;
1545 		alloced += mtsp->mts_memalloced;
1546 		freed += mtsp->mts_memfreed;
1547 	}
1548 	*inuse = *allocs - frees;
1549 	return (alloced - freed);
1550 }
1551 
1552 DB_SHOW_COMMAND_FLAGS(malloc, db_show_malloc, DB_CMD_MEMSAFE)
1553 {
1554 	const char *fmt_hdr, *fmt_entry;
1555 	struct malloc_type *mtp;
1556 	uint64_t allocs, inuse;
1557 	int64_t size;
1558 	/* variables for sorting */
1559 	struct malloc_type *last_mtype, *cur_mtype;
1560 	int64_t cur_size, last_size;
1561 	int ties;
1562 
1563 	if (modif[0] == 'i') {
1564 		fmt_hdr = "%s,%s,%s,%s\n";
1565 		fmt_entry = "\"%s\",%ju,%jdK,%ju\n";
1566 	} else {
1567 		fmt_hdr = "%18s %12s  %12s %12s\n";
1568 		fmt_entry = "%18s %12ju %12jdK %12ju\n";
1569 	}
1570 
1571 	db_printf(fmt_hdr, "Type", "InUse", "MemUse", "Requests");
1572 
1573 	/* Select sort, largest size first. */
1574 	last_mtype = NULL;
1575 	last_size = INT64_MAX;
1576 	for (;;) {
1577 		cur_mtype = NULL;
1578 		cur_size = -1;
1579 		ties = 0;
1580 
1581 		for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) {
1582 			/*
1583 			 * In the case of size ties, print out mtypes
1584 			 * in the order they are encountered.  That is,
1585 			 * when we encounter the most recently output
1586 			 * mtype, we have already printed all preceding
1587 			 * ties, and we must print all following ties.
1588 			 */
1589 			if (mtp == last_mtype) {
1590 				ties = 1;
1591 				continue;
1592 			}
1593 			size = get_malloc_stats(&mtp->ks_mti, &allocs,
1594 			    &inuse);
1595 			if (size > cur_size && size < last_size + ties) {
1596 				cur_size = size;
1597 				cur_mtype = mtp;
1598 			}
1599 		}
1600 		if (cur_mtype == NULL)
1601 			break;
1602 
1603 		size = get_malloc_stats(&cur_mtype->ks_mti, &allocs, &inuse);
1604 		db_printf(fmt_entry, cur_mtype->ks_shortdesc, inuse,
1605 		    howmany(size, 1024), allocs);
1606 
1607 		if (db_pager_quit)
1608 			break;
1609 
1610 		last_mtype = cur_mtype;
1611 		last_size = cur_size;
1612 	}
1613 }
1614 
1615 #if MALLOC_DEBUG_MAXZONES > 1
1616 DB_SHOW_COMMAND(multizone_matches, db_show_multizone_matches)
1617 {
1618 	struct malloc_type_internal *mtip;
1619 	struct malloc_type *mtp;
1620 	u_int subzone;
1621 
1622 	if (!have_addr) {
1623 		db_printf("Usage: show multizone_matches <malloc type/addr>\n");
1624 		return;
1625 	}
1626 	mtp = (void *)addr;
1627 	if (mtp->ks_version != M_VERSION) {
1628 		db_printf("Version %lx does not match expected %x\n",
1629 		    mtp->ks_version, M_VERSION);
1630 		return;
1631 	}
1632 
1633 	mtip = &mtp->ks_mti;
1634 	subzone = mtip->mti_zone;
1635 
1636 	for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) {
1637 		mtip = &mtp->ks_mti;
1638 		if (mtip->mti_zone != subzone)
1639 			continue;
1640 		db_printf("%s\n", mtp->ks_shortdesc);
1641 		if (db_pager_quit)
1642 			break;
1643 	}
1644 }
1645 #endif /* MALLOC_DEBUG_MAXZONES > 1 */
1646 #endif /* DDB */
1647