2 * SPDX-License-Identifier: BSD-3-Clause
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)
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
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.
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
34 * @(#)kern_malloc.c 8.3 (Berkeley) 1/4/94
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
48 #include <sys/cdefs.h>
49 __FBSDID("$FreeBSD$");
54 #include <sys/param.h>
55 #include <sys/systm.h>
57 #include <sys/kernel.h>
59 #include <sys/malloc.h>
60 #include <sys/mutex.h>
61 #include <sys/vmmeter.h>
63 #include <sys/queue.h>
66 #include <sys/sysctl.h>
70 #include <sys/epoch.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>
85 #include <vm/uma_int.h>
86 #include <vm/uma_dbg.h>
89 #include <vm/memguard.h>
92 #include <vm/redzone.h>
95 #if defined(INVARIANTS) && defined(__i386__)
96 #include <machine/cpu.h>
102 #include <sys/dtrace_bsd.h>
104 bool __read_frequently dtrace_malloc_enabled;
105 dtrace_malloc_probe_func_t __read_mostly dtrace_malloc_probe;
108 #if defined(INVARIANTS) || defined(MALLOC_MAKE_FAILURES) || \
109 defined(DEBUG_MEMGUARD) || defined(DEBUG_REDZONE)
110 #define MALLOC_DEBUG 1
114 * When realloc() is called, if the new size is sufficiently smaller than
115 * the old size, realloc() will allocate a new, smaller block to avoid
116 * wasting memory. 'Sufficiently smaller' is defined as: newsize <=
117 * oldsize / 2^n, where REALLOC_FRACTION defines the value of 'n'.
119 #ifndef REALLOC_FRACTION
120 #define REALLOC_FRACTION 1 /* new block if <= half the size */
124 * Centrally define some common malloc types.
126 MALLOC_DEFINE(M_CACHE, "cache", "Various Dynamically allocated caches");
127 MALLOC_DEFINE(M_DEVBUF, "devbuf", "device driver memory");
128 MALLOC_DEFINE(M_TEMP, "temp", "misc temporary data buffers");
130 static struct malloc_type *kmemstatistics;
131 static int kmemcount;
133 #define KMEM_ZSHIFT 4
134 #define KMEM_ZBASE 16
135 #define KMEM_ZMASK (KMEM_ZBASE - 1)
137 #define KMEM_ZMAX 65536
138 #define KMEM_ZSIZE (KMEM_ZMAX >> KMEM_ZSHIFT)
139 static uint8_t kmemsize[KMEM_ZSIZE + 1];
141 #ifndef MALLOC_DEBUG_MAXZONES
142 #define MALLOC_DEBUG_MAXZONES 1
144 static int numzones = MALLOC_DEBUG_MAXZONES;
147 * Small malloc(9) memory allocations are allocated from a set of UMA buckets
150 * Warning: the layout of the struct is duplicated in libmemstat for KVM support.
152 * XXX: The comment here used to read "These won't be powers of two for
153 * long." It's possible that a significant amount of wasted memory could be
154 * recovered by tuning the sizes of these buckets.
159 uma_zone_t kz_zone[MALLOC_DEBUG_MAXZONES];
164 {128, "malloc-128", },
165 {256, "malloc-256", },
166 {384, "malloc-384", },
167 {512, "malloc-512", },
168 {1024, "malloc-1024", },
169 {2048, "malloc-2048", },
170 {4096, "malloc-4096", },
171 {8192, "malloc-8192", },
172 {16384, "malloc-16384", },
173 {32768, "malloc-32768", },
174 {65536, "malloc-65536", },
179 SYSCTL_ULONG(_vm, OID_AUTO, kmem_size, CTLFLAG_RDTUN, &vm_kmem_size, 0,
180 "Size of kernel memory");
182 static u_long kmem_zmax = KMEM_ZMAX;
183 SYSCTL_ULONG(_vm, OID_AUTO, kmem_zmax, CTLFLAG_RDTUN, &kmem_zmax, 0,
184 "Maximum allocation size that malloc(9) would use UMA as backend");
186 static u_long vm_kmem_size_min;
187 SYSCTL_ULONG(_vm, OID_AUTO, kmem_size_min, CTLFLAG_RDTUN, &vm_kmem_size_min, 0,
188 "Minimum size of kernel memory");
190 static u_long vm_kmem_size_max;
191 SYSCTL_ULONG(_vm, OID_AUTO, kmem_size_max, CTLFLAG_RDTUN, &vm_kmem_size_max, 0,
192 "Maximum size of kernel memory");
194 static u_int vm_kmem_size_scale;
195 SYSCTL_UINT(_vm, OID_AUTO, kmem_size_scale, CTLFLAG_RDTUN, &vm_kmem_size_scale, 0,
196 "Scale factor for kernel memory size");
198 static int sysctl_kmem_map_size(SYSCTL_HANDLER_ARGS);
199 SYSCTL_PROC(_vm, OID_AUTO, kmem_map_size,
200 CTLFLAG_RD | CTLTYPE_ULONG | CTLFLAG_MPSAFE, NULL, 0,
201 sysctl_kmem_map_size, "LU", "Current kmem allocation size");
203 static int sysctl_kmem_map_free(SYSCTL_HANDLER_ARGS);
204 SYSCTL_PROC(_vm, OID_AUTO, kmem_map_free,
205 CTLFLAG_RD | CTLTYPE_ULONG | CTLFLAG_MPSAFE, NULL, 0,
206 sysctl_kmem_map_free, "LU", "Free space in kmem");
208 static SYSCTL_NODE(_vm, OID_AUTO, malloc, CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
209 "Malloc information");
211 static u_int vm_malloc_zone_count = nitems(kmemzones);
212 SYSCTL_UINT(_vm_malloc, OID_AUTO, zone_count,
213 CTLFLAG_RD, &vm_malloc_zone_count, 0,
214 "Number of malloc zones");
216 static int sysctl_vm_malloc_zone_sizes(SYSCTL_HANDLER_ARGS);
217 SYSCTL_PROC(_vm_malloc, OID_AUTO, zone_sizes,
218 CTLFLAG_RD | CTLTYPE_OPAQUE | CTLFLAG_MPSAFE, NULL, 0,
219 sysctl_vm_malloc_zone_sizes, "S", "Zone sizes used by malloc");
222 * The malloc_mtx protects the kmemstatistics linked list.
224 struct mtx malloc_mtx;
226 static int sysctl_kern_malloc_stats(SYSCTL_HANDLER_ARGS);
228 #if defined(MALLOC_MAKE_FAILURES) || (MALLOC_DEBUG_MAXZONES > 1)
229 static SYSCTL_NODE(_debug, OID_AUTO, malloc, CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
230 "Kernel malloc debugging options");
234 * malloc(9) fault injection -- cause malloc failures every (n) mallocs when
235 * the caller specifies M_NOWAIT. If set to 0, no failures are caused.
237 #ifdef MALLOC_MAKE_FAILURES
238 static int malloc_failure_rate;
239 static int malloc_nowait_count;
240 static int malloc_failure_count;
241 SYSCTL_INT(_debug_malloc, OID_AUTO, failure_rate, CTLFLAG_RWTUN,
242 &malloc_failure_rate, 0, "Every (n) mallocs with M_NOWAIT will fail");
243 SYSCTL_INT(_debug_malloc, OID_AUTO, failure_count, CTLFLAG_RD,
244 &malloc_failure_count, 0, "Number of imposed M_NOWAIT malloc failures");
248 sysctl_kmem_map_size(SYSCTL_HANDLER_ARGS)
253 return (sysctl_handle_long(oidp, &size, 0, req));
257 sysctl_kmem_map_free(SYSCTL_HANDLER_ARGS)
261 /* The sysctl is unsigned, implement as a saturation value. */
268 return (sysctl_handle_long(oidp, &size, 0, req));
272 sysctl_vm_malloc_zone_sizes(SYSCTL_HANDLER_ARGS)
274 int sizes[nitems(kmemzones)];
277 for (i = 0; i < nitems(kmemzones); i++) {
278 sizes[i] = kmemzones[i].kz_size;
281 return (SYSCTL_OUT(req, &sizes, sizeof(sizes)));
285 * malloc(9) uma zone separation -- sub-page buffer overruns in one
286 * malloc type will affect only a subset of other malloc types.
288 #if MALLOC_DEBUG_MAXZONES > 1
290 tunable_set_numzones(void)
293 TUNABLE_INT_FETCH("debug.malloc.numzones",
296 /* Sanity check the number of malloc uma zones. */
299 if (numzones > MALLOC_DEBUG_MAXZONES)
300 numzones = MALLOC_DEBUG_MAXZONES;
302 SYSINIT(numzones, SI_SUB_TUNABLES, SI_ORDER_ANY, tunable_set_numzones, NULL);
303 SYSCTL_INT(_debug_malloc, OID_AUTO, numzones, CTLFLAG_RDTUN | CTLFLAG_NOFETCH,
304 &numzones, 0, "Number of malloc uma subzones");
307 * Any number that changes regularly is an okay choice for the
308 * offset. Build numbers are pretty good of you have them.
310 static u_int zone_offset = __FreeBSD_version;
311 TUNABLE_INT("debug.malloc.zone_offset", &zone_offset);
312 SYSCTL_UINT(_debug_malloc, OID_AUTO, zone_offset, CTLFLAG_RDTUN,
313 &zone_offset, 0, "Separate malloc types by examining the "
314 "Nth character in the malloc type short description.");
317 mtp_set_subzone(struct malloc_type *mtp)
319 struct malloc_type_internal *mtip;
325 desc = mtp->ks_shortdesc;
326 if (desc == NULL || (len = strlen(desc)) == 0)
329 val = desc[zone_offset % len];
330 mtip->mti_zone = (val % numzones);
334 mtp_get_subzone(struct malloc_type *mtp)
336 struct malloc_type_internal *mtip;
340 KASSERT(mtip->mti_zone < numzones,
341 ("mti_zone %u out of range %d",
342 mtip->mti_zone, numzones));
343 return (mtip->mti_zone);
345 #elif MALLOC_DEBUG_MAXZONES == 0
346 #error "MALLOC_DEBUG_MAXZONES must be positive."
349 mtp_set_subzone(struct malloc_type *mtp)
351 struct malloc_type_internal *mtip;
358 mtp_get_subzone(struct malloc_type *mtp)
363 #endif /* MALLOC_DEBUG_MAXZONES > 1 */
366 * An allocation has succeeded -- update malloc type statistics for the
367 * amount of bucket size. Occurs within a critical section so that the
368 * thread isn't preempted and doesn't migrate while updating per-PCU
372 malloc_type_zone_allocated(struct malloc_type *mtp, unsigned long size,
375 struct malloc_type_internal *mtip;
376 struct malloc_type_stats *mtsp;
380 mtsp = zpcpu_get(mtip->mti_stats);
382 mtsp->mts_memalloced += size;
383 mtsp->mts_numallocs++;
386 mtsp->mts_size |= 1 << zindx;
389 if (__predict_false(dtrace_malloc_enabled)) {
390 uint32_t probe_id = mtip->mti_probes[DTMALLOC_PROBE_MALLOC];
392 (dtrace_malloc_probe)(probe_id,
393 (uintptr_t) mtp, (uintptr_t) mtip,
394 (uintptr_t) mtsp, size, zindx);
402 malloc_type_allocated(struct malloc_type *mtp, unsigned long size)
406 malloc_type_zone_allocated(mtp, size, -1);
410 * A free operation has occurred -- update malloc type statistics for the
411 * amount of the bucket size. Occurs within a critical section so that the
412 * thread isn't preempted and doesn't migrate while updating per-CPU
416 malloc_type_freed(struct malloc_type *mtp, unsigned long size)
418 struct malloc_type_internal *mtip;
419 struct malloc_type_stats *mtsp;
423 mtsp = zpcpu_get(mtip->mti_stats);
424 mtsp->mts_memfreed += size;
425 mtsp->mts_numfrees++;
428 if (__predict_false(dtrace_malloc_enabled)) {
429 uint32_t probe_id = mtip->mti_probes[DTMALLOC_PROBE_FREE];
431 (dtrace_malloc_probe)(probe_id,
432 (uintptr_t) mtp, (uintptr_t) mtip,
433 (uintptr_t) mtsp, size, 0);
443 * Allocate a block of physically contiguous memory.
445 * If M_NOWAIT is set, this routine will not block and return NULL if
446 * the allocation fails.
449 contigmalloc(unsigned long size, struct malloc_type *type, int flags,
450 vm_paddr_t low, vm_paddr_t high, unsigned long alignment,
455 ret = (void *)kmem_alloc_contig(size, flags, low, high, alignment,
456 boundary, VM_MEMATTR_DEFAULT);
458 malloc_type_allocated(type, round_page(size));
463 contigmalloc_domainset(unsigned long size, struct malloc_type *type,
464 struct domainset *ds, int flags, vm_paddr_t low, vm_paddr_t high,
465 unsigned long alignment, vm_paddr_t boundary)
469 ret = (void *)kmem_alloc_contig_domainset(ds, size, flags, low, high,
470 alignment, boundary, VM_MEMATTR_DEFAULT);
472 malloc_type_allocated(type, round_page(size));
479 * Free a block of memory allocated by contigmalloc.
481 * This routine may not block.
484 contigfree(void *addr, unsigned long size, struct malloc_type *type)
487 kmem_free((vm_offset_t)addr, size);
488 malloc_type_freed(type, round_page(size));
493 malloc_dbg(caddr_t *vap, size_t *sizep, struct malloc_type *mtp,
499 KASSERT(mtp->ks_version == M_VERSION, ("malloc: bad malloc type version"));
501 * Check that exactly one of M_WAITOK or M_NOWAIT is specified.
503 indx = flags & (M_WAITOK | M_NOWAIT);
504 if (indx != M_NOWAIT && indx != M_WAITOK) {
505 static struct timeval lasterr;
506 static int curerr, once;
507 if (once == 0 && ppsratecheck(&lasterr, &curerr, 1)) {
508 printf("Bad malloc flags: %x\n", indx);
515 #ifdef MALLOC_MAKE_FAILURES
516 if ((flags & M_NOWAIT) && (malloc_failure_rate != 0)) {
517 atomic_add_int(&malloc_nowait_count, 1);
518 if ((malloc_nowait_count % malloc_failure_rate) == 0) {
519 atomic_add_int(&malloc_failure_count, 1);
521 return (EJUSTRETURN);
525 if (flags & M_WAITOK) {
526 KASSERT(curthread->td_intr_nesting_level == 0,
527 ("malloc(M_WAITOK) in interrupt context"));
528 if (__predict_false(!THREAD_CAN_SLEEP())) {
530 epoch_trace_list(curthread);
533 ("malloc(M_WAITOK) with sleeping prohibited"));
536 KASSERT(curthread->td_critnest == 0 || SCHEDULER_STOPPED(),
537 ("malloc: called with spinlock or critical section held"));
539 #ifdef DEBUG_MEMGUARD
540 if (memguard_cmp_mtp(mtp, *sizep)) {
541 *vap = memguard_alloc(*sizep, flags);
543 return (EJUSTRETURN);
544 /* This is unfortunate but should not be fatal. */
549 *sizep = redzone_size_ntor(*sizep);
557 * Handle large allocations and frees by using kmem_malloc directly.
560 malloc_large_slab(uma_slab_t slab)
564 va = (uintptr_t)slab;
565 return ((va & 1) != 0);
569 malloc_large_size(uma_slab_t slab)
573 va = (uintptr_t)slab;
578 malloc_large(size_t *size, struct domainset *policy, int flags)
583 sz = roundup(*size, PAGE_SIZE);
584 va = kmem_malloc_domainset(policy, sz, flags);
586 /* The low bit is unused for slab pointers. */
587 vsetzoneslab(va, NULL, (void *)((sz << 1) | 1));
591 return ((caddr_t)va);
595 free_large(void *addr, size_t size)
598 kmem_free((vm_offset_t)addr, size);
605 * Allocate a block of memory.
607 * If M_NOWAIT is set, this routine will not block and return NULL if
608 * the allocation fails.
611 (malloc)(size_t size, struct malloc_type *mtp, int flags)
616 #if defined(DEBUG_REDZONE)
617 unsigned long osize = size;
620 MPASS((flags & M_EXEC) == 0);
623 if (malloc_dbg(&va, &size, mtp, flags) != 0)
627 if (size <= kmem_zmax) {
628 if (size & KMEM_ZMASK)
629 size = (size & ~KMEM_ZMASK) + KMEM_ZBASE;
630 indx = kmemsize[size >> KMEM_ZSHIFT];
631 zone = kmemzones[indx].kz_zone[mtp_get_subzone(mtp)];
632 va = uma_zalloc(zone, flags);
634 size = zone->uz_size;
635 malloc_type_zone_allocated(mtp, va == NULL ? 0 : size, indx);
637 va = malloc_large(&size, DOMAINSET_RR(), flags);
638 malloc_type_allocated(mtp, va == NULL ? 0 : size);
640 if (__predict_false(va == NULL)) {
641 KASSERT((flags & M_WAITOK) == 0,
642 ("malloc(M_WAITOK) returned NULL"));
646 va = redzone_setup(va, osize);
648 return ((void *) va);
652 malloc_domain(size_t *sizep, int *indxp, struct malloc_type *mtp, int domain,
661 KASSERT(size <= kmem_zmax && (flags & M_EXEC) == 0,
662 ("malloc_domain: Called with bad flag / size combination."));
663 if (size & KMEM_ZMASK)
664 size = (size & ~KMEM_ZMASK) + KMEM_ZBASE;
665 indx = kmemsize[size >> KMEM_ZSHIFT];
666 zone = kmemzones[indx].kz_zone[mtp_get_subzone(mtp)];
667 va = uma_zalloc_domain(zone, NULL, domain, flags);
669 *sizep = zone->uz_size;
675 malloc_domainset(size_t size, struct malloc_type *mtp, struct domainset *ds,
678 struct vm_domainset_iter di;
683 #if defined(DEBUG_REDZONE)
684 unsigned long osize = size;
686 MPASS((flags & M_EXEC) == 0);
689 if (malloc_dbg(&va, &size, mtp, flags) != 0)
692 if (size <= kmem_zmax) {
693 vm_domainset_iter_policy_init(&di, ds, &domain, &flags);
695 va = malloc_domain(&size, &indx, mtp, domain, flags);
696 } while (va == NULL &&
697 vm_domainset_iter_policy(&di, &domain) == 0);
698 malloc_type_zone_allocated(mtp, va == NULL ? 0 : size, indx);
700 /* Policy is handled by kmem. */
701 va = malloc_large(&size, ds, flags);
702 malloc_type_allocated(mtp, va == NULL ? 0 : size);
704 if (__predict_false(va == NULL)) {
705 KASSERT((flags & M_WAITOK) == 0,
706 ("malloc(M_WAITOK) returned NULL"));
710 va = redzone_setup(va, osize);
716 * Allocate an executable area.
719 malloc_exec(size_t size, struct malloc_type *mtp, int flags)
722 #if defined(DEBUG_REDZONE)
723 unsigned long osize = size;
729 if (malloc_dbg(&va, &size, mtp, flags) != 0)
732 va = malloc_large(&size, DOMAINSET_RR(), flags);
733 malloc_type_allocated(mtp, va == NULL ? 0 : size);
734 if (__predict_false(va == NULL)) {
735 KASSERT((flags & M_WAITOK) == 0,
736 ("malloc(M_WAITOK) returned NULL"));
740 va = redzone_setup(va, osize);
742 return ((void *) va);
746 malloc_domainset_exec(size_t size, struct malloc_type *mtp, struct domainset *ds,
750 #if defined(DEBUG_REDZONE)
751 unsigned long osize = size;
757 if (malloc_dbg(&va, &size, mtp, flags) != 0)
760 /* Policy is handled by kmem. */
761 va = malloc_large(&size, ds, flags);
762 malloc_type_allocated(mtp, va == NULL ? 0 : size);
763 if (__predict_false(va == NULL)) {
764 KASSERT((flags & M_WAITOK) == 0,
765 ("malloc(M_WAITOK) returned NULL"));
769 va = redzone_setup(va, osize);
775 mallocarray(size_t nmemb, size_t size, struct malloc_type *type, int flags)
778 if (WOULD_OVERFLOW(nmemb, size))
779 panic("mallocarray: %zu * %zu overflowed", nmemb, size);
781 return (malloc(size * nmemb, type, flags));
786 free_save_type(void *addr, struct malloc_type *mtp, u_long size)
788 struct malloc_type **mtpp = addr;
791 * Cache a pointer to the malloc_type that most recently freed
792 * this memory here. This way we know who is most likely to
793 * have stepped on it later.
795 * This code assumes that size is a multiple of 8 bytes for
798 mtpp = (struct malloc_type **) ((unsigned long)mtpp & ~UMA_ALIGN_PTR);
799 mtpp += (size - sizeof(struct malloc_type *)) /
800 sizeof(struct malloc_type *);
807 free_dbg(void **addrp, struct malloc_type *mtp)
812 KASSERT(mtp->ks_version == M_VERSION, ("free: bad malloc type version"));
813 KASSERT(curthread->td_critnest == 0 || SCHEDULER_STOPPED(),
814 ("free: called with spinlock or critical section held"));
816 /* free(NULL, ...) does nothing */
818 return (EJUSTRETURN);
820 #ifdef DEBUG_MEMGUARD
821 if (is_memguard_addr(addr)) {
823 return (EJUSTRETURN);
829 *addrp = redzone_addr_ntor(addr);
839 * Free a block of memory allocated by malloc.
841 * This routine may not block.
844 free(void *addr, struct malloc_type *mtp)
851 if (free_dbg(&addr, mtp) != 0)
854 /* free(NULL, ...) does nothing */
858 vtozoneslab((vm_offset_t)addr & (~UMA_SLAB_MASK), &zone, &slab);
860 panic("free: address %p(%p) has not been allocated.\n",
861 addr, (void *)((u_long)addr & (~UMA_SLAB_MASK)));
863 if (__predict_true(!malloc_large_slab(slab))) {
864 size = zone->uz_size;
866 free_save_type(addr, mtp, size);
868 uma_zfree_arg(zone, addr, slab);
870 size = malloc_large_size(slab);
871 free_large(addr, size);
873 malloc_type_freed(mtp, size);
879 * Zero then free a block of memory allocated by malloc.
881 * This routine may not block.
884 zfree(void *addr, struct malloc_type *mtp)
891 if (free_dbg(&addr, mtp) != 0)
894 /* free(NULL, ...) does nothing */
898 vtozoneslab((vm_offset_t)addr & (~UMA_SLAB_MASK), &zone, &slab);
900 panic("free: address %p(%p) has not been allocated.\n",
901 addr, (void *)((u_long)addr & (~UMA_SLAB_MASK)));
903 if (__predict_true(!malloc_large_slab(slab))) {
904 size = zone->uz_size;
906 free_save_type(addr, mtp, size);
908 explicit_bzero(addr, size);
909 uma_zfree_arg(zone, addr, slab);
911 size = malloc_large_size(slab);
912 explicit_bzero(addr, size);
913 free_large(addr, size);
915 malloc_type_freed(mtp, size);
919 * realloc: change the size of a memory block
922 realloc(void *addr, size_t size, struct malloc_type *mtp, int flags)
929 KASSERT(mtp->ks_version == M_VERSION,
930 ("realloc: bad malloc type version"));
931 KASSERT(curthread->td_critnest == 0 || SCHEDULER_STOPPED(),
932 ("realloc: called with spinlock or critical section held"));
934 /* realloc(NULL, ...) is equivalent to malloc(...) */
936 return (malloc(size, mtp, flags));
939 * XXX: Should report free of old memory and alloc of new memory to
943 #ifdef DEBUG_MEMGUARD
944 if (is_memguard_addr(addr))
945 return (memguard_realloc(addr, size, mtp, flags));
951 alloc = redzone_get_size(addr);
953 vtozoneslab((vm_offset_t)addr & (~UMA_SLAB_MASK), &zone, &slab);
956 KASSERT(slab != NULL,
957 ("realloc: address %p out of range", (void *)addr));
959 /* Get the size of the original block */
960 if (!malloc_large_slab(slab))
961 alloc = zone->uz_size;
963 alloc = malloc_large_size(slab);
965 /* Reuse the original block if appropriate */
967 && (size > (alloc >> REALLOC_FRACTION) || alloc == MINALLOCSIZE))
969 #endif /* !DEBUG_REDZONE */
971 /* Allocate a new, bigger (or smaller) block */
972 if ((newaddr = malloc(size, mtp, flags)) == NULL)
975 /* Copy over original contents */
976 bcopy(addr, newaddr, min(size, alloc));
982 * reallocf: same as realloc() but free memory on failure.
985 reallocf(void *addr, size_t size, struct malloc_type *mtp, int flags)
989 if ((mem = realloc(addr, size, mtp, flags)) == NULL)
995 * malloc_size: returns the number of bytes allocated for a request of the
999 malloc_size(size_t size)
1003 if (size > kmem_zmax)
1005 if (size & KMEM_ZMASK)
1006 size = (size & ~KMEM_ZMASK) + KMEM_ZBASE;
1007 indx = kmemsize[size >> KMEM_ZSHIFT];
1008 return (kmemzones[indx].kz_size);
1012 * malloc_usable_size: returns the usable size of the allocation.
1015 malloc_usable_size(const void *addr)
1017 #ifndef DEBUG_REDZONE
1026 #ifdef DEBUG_MEMGUARD
1027 if (is_memguard_addr(__DECONST(void *, addr)))
1028 return (memguard_get_req_size(addr));
1031 #ifdef DEBUG_REDZONE
1032 size = redzone_get_size(__DECONST(void *, addr));
1034 vtozoneslab((vm_offset_t)addr & (~UMA_SLAB_MASK), &zone, &slab);
1036 panic("malloc_usable_size: address %p(%p) is not allocated.\n",
1037 addr, (void *)((u_long)addr & (~UMA_SLAB_MASK)));
1039 if (!malloc_large_slab(slab))
1040 size = zone->uz_size;
1042 size = malloc_large_size(slab);
1047 CTASSERT(VM_KMEM_SIZE_SCALE >= 1);
1050 * Initialize the kernel memory (kmem) arena.
1059 if (vm_kmem_size == 0)
1060 vm_kmem_size = VM_KMEM_SIZE;
1062 #ifdef VM_KMEM_SIZE_MIN
1063 if (vm_kmem_size_min == 0)
1064 vm_kmem_size_min = VM_KMEM_SIZE_MIN;
1066 #ifdef VM_KMEM_SIZE_MAX
1067 if (vm_kmem_size_max == 0)
1068 vm_kmem_size_max = VM_KMEM_SIZE_MAX;
1071 * Calculate the amount of kernel virtual address (KVA) space that is
1072 * preallocated to the kmem arena. In order to support a wide range
1073 * of machines, it is a function of the physical memory size,
1076 * min(max(physical memory size / VM_KMEM_SIZE_SCALE,
1077 * VM_KMEM_SIZE_MIN), VM_KMEM_SIZE_MAX)
1079 * Every architecture must define an integral value for
1080 * VM_KMEM_SIZE_SCALE. However, the definitions of VM_KMEM_SIZE_MIN
1081 * and VM_KMEM_SIZE_MAX, which represent respectively the floor and
1082 * ceiling on this preallocation, are optional. Typically,
1083 * VM_KMEM_SIZE_MAX is itself a function of the available KVA space on
1084 * a given architecture.
1086 mem_size = vm_cnt.v_page_count;
1087 if (mem_size <= 32768) /* delphij XXX 128MB */
1088 kmem_zmax = PAGE_SIZE;
1090 if (vm_kmem_size_scale < 1)
1091 vm_kmem_size_scale = VM_KMEM_SIZE_SCALE;
1094 * Check if we should use defaults for the "vm_kmem_size"
1097 if (vm_kmem_size == 0) {
1098 vm_kmem_size = mem_size / vm_kmem_size_scale;
1099 vm_kmem_size = vm_kmem_size * PAGE_SIZE < vm_kmem_size ?
1100 vm_kmem_size_max : vm_kmem_size * PAGE_SIZE;
1101 if (vm_kmem_size_min > 0 && vm_kmem_size < vm_kmem_size_min)
1102 vm_kmem_size = vm_kmem_size_min;
1103 if (vm_kmem_size_max > 0 && vm_kmem_size >= vm_kmem_size_max)
1104 vm_kmem_size = vm_kmem_size_max;
1106 if (vm_kmem_size == 0)
1107 panic("Tune VM_KMEM_SIZE_* for the platform");
1110 * The amount of KVA space that is preallocated to the
1111 * kmem arena can be set statically at compile-time or manually
1112 * through the kernel environment. However, it is still limited to
1113 * twice the physical memory size, which has been sufficient to handle
1114 * the most severe cases of external fragmentation in the kmem arena.
1116 if (vm_kmem_size / 2 / PAGE_SIZE > mem_size)
1117 vm_kmem_size = 2 * mem_size * PAGE_SIZE;
1119 vm_kmem_size = round_page(vm_kmem_size);
1120 #ifdef DEBUG_MEMGUARD
1121 tmp = memguard_fudge(vm_kmem_size, kernel_map);
1127 #ifdef DEBUG_MEMGUARD
1129 * Initialize MemGuard if support compiled in. MemGuard is a
1130 * replacement allocator used for detecting tamper-after-free
1131 * scenarios as they occur. It is only used for debugging.
1133 memguard_init(kernel_arena);
1138 * Initialize the kernel memory allocator
1142 mallocinit(void *dummy)
1147 mtx_init(&malloc_mtx, "malloc", NULL, MTX_DEF);
1151 if (kmem_zmax < PAGE_SIZE || kmem_zmax > KMEM_ZMAX)
1152 kmem_zmax = KMEM_ZMAX;
1154 for (i = 0, indx = 0; kmemzones[indx].kz_size != 0; indx++) {
1155 int size = kmemzones[indx].kz_size;
1156 const char *name = kmemzones[indx].kz_name;
1159 for (subzone = 0; subzone < numzones; subzone++) {
1160 kmemzones[indx].kz_zone[subzone] =
1161 uma_zcreate(name, size,
1163 mtrash_ctor, mtrash_dtor, mtrash_init, mtrash_fini,
1165 NULL, NULL, NULL, NULL,
1167 UMA_ALIGN_PTR, UMA_ZONE_MALLOC);
1169 for (;i <= size; i+= KMEM_ZBASE)
1170 kmemsize[i >> KMEM_ZSHIFT] = indx;
1173 SYSINIT(kmem, SI_SUB_KMEM, SI_ORDER_SECOND, mallocinit, NULL);
1176 malloc_init(void *data)
1178 struct malloc_type_internal *mtip;
1179 struct malloc_type *mtp;
1181 KASSERT(vm_cnt.v_page_count != 0, ("malloc_register before vm_init"));
1184 if (mtp->ks_version != M_VERSION)
1185 panic("malloc_init: type %s with unsupported version %lu",
1186 mtp->ks_shortdesc, mtp->ks_version);
1188 mtip = &mtp->ks_mti;
1189 mtip->mti_stats = uma_zalloc_pcpu(pcpu_zone_64, M_WAITOK | M_ZERO);
1190 mtp_set_subzone(mtp);
1192 mtx_lock(&malloc_mtx);
1193 mtp->ks_next = kmemstatistics;
1194 kmemstatistics = mtp;
1196 mtx_unlock(&malloc_mtx);
1200 malloc_uninit(void *data)
1202 struct malloc_type_internal *mtip;
1203 struct malloc_type_stats *mtsp;
1204 struct malloc_type *mtp, *temp;
1205 long temp_allocs, temp_bytes;
1209 KASSERT(mtp->ks_version == M_VERSION,
1210 ("malloc_uninit: bad malloc type version"));
1212 mtx_lock(&malloc_mtx);
1213 mtip = &mtp->ks_mti;
1214 if (mtp != kmemstatistics) {
1215 for (temp = kmemstatistics; temp != NULL;
1216 temp = temp->ks_next) {
1217 if (temp->ks_next == mtp) {
1218 temp->ks_next = mtp->ks_next;
1223 ("malloc_uninit: type '%s' not found", mtp->ks_shortdesc));
1225 kmemstatistics = mtp->ks_next;
1227 mtx_unlock(&malloc_mtx);
1230 * Look for memory leaks.
1232 temp_allocs = temp_bytes = 0;
1233 for (i = 0; i <= mp_maxid; i++) {
1234 mtsp = zpcpu_get_cpu(mtip->mti_stats, i);
1235 temp_allocs += mtsp->mts_numallocs;
1236 temp_allocs -= mtsp->mts_numfrees;
1237 temp_bytes += mtsp->mts_memalloced;
1238 temp_bytes -= mtsp->mts_memfreed;
1240 if (temp_allocs > 0 || temp_bytes > 0) {
1241 printf("Warning: memory type %s leaked memory on destroy "
1242 "(%ld allocations, %ld bytes leaked).\n", mtp->ks_shortdesc,
1243 temp_allocs, temp_bytes);
1246 uma_zfree_pcpu(pcpu_zone_64, mtip->mti_stats);
1249 struct malloc_type *
1250 malloc_desc2type(const char *desc)
1252 struct malloc_type *mtp;
1254 mtx_assert(&malloc_mtx, MA_OWNED);
1255 for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) {
1256 if (strcmp(mtp->ks_shortdesc, desc) == 0)
1263 sysctl_kern_malloc_stats(SYSCTL_HANDLER_ARGS)
1265 struct malloc_type_stream_header mtsh;
1266 struct malloc_type_internal *mtip;
1267 struct malloc_type_stats *mtsp, zeromts;
1268 struct malloc_type_header mth;
1269 struct malloc_type *mtp;
1273 error = sysctl_wire_old_buffer(req, 0);
1276 sbuf_new_for_sysctl(&sbuf, NULL, 128, req);
1277 sbuf_clear_flags(&sbuf, SBUF_INCLUDENUL);
1278 mtx_lock(&malloc_mtx);
1280 bzero(&zeromts, sizeof(zeromts));
1283 * Insert stream header.
1285 bzero(&mtsh, sizeof(mtsh));
1286 mtsh.mtsh_version = MALLOC_TYPE_STREAM_VERSION;
1287 mtsh.mtsh_maxcpus = MAXCPU;
1288 mtsh.mtsh_count = kmemcount;
1289 (void)sbuf_bcat(&sbuf, &mtsh, sizeof(mtsh));
1292 * Insert alternating sequence of type headers and type statistics.
1294 for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) {
1295 mtip = &mtp->ks_mti;
1298 * Insert type header.
1300 bzero(&mth, sizeof(mth));
1301 strlcpy(mth.mth_name, mtp->ks_shortdesc, MALLOC_MAX_NAME);
1302 (void)sbuf_bcat(&sbuf, &mth, sizeof(mth));
1305 * Insert type statistics for each CPU.
1307 for (i = 0; i <= mp_maxid; i++) {
1308 mtsp = zpcpu_get_cpu(mtip->mti_stats, i);
1309 (void)sbuf_bcat(&sbuf, mtsp, sizeof(*mtsp));
1312 * Fill in the missing CPUs.
1314 for (; i < MAXCPU; i++) {
1315 (void)sbuf_bcat(&sbuf, &zeromts, sizeof(zeromts));
1318 mtx_unlock(&malloc_mtx);
1319 error = sbuf_finish(&sbuf);
1324 SYSCTL_PROC(_kern, OID_AUTO, malloc_stats,
1325 CTLFLAG_RD | CTLTYPE_STRUCT | CTLFLAG_MPSAFE, 0, 0,
1326 sysctl_kern_malloc_stats, "s,malloc_type_ustats",
1327 "Return malloc types");
1329 SYSCTL_INT(_kern, OID_AUTO, malloc_count, CTLFLAG_RD, &kmemcount, 0,
1330 "Count of kernel malloc types");
1333 malloc_type_list(malloc_type_list_func_t *func, void *arg)
1335 struct malloc_type *mtp, **bufmtp;
1339 mtx_lock(&malloc_mtx);
1341 mtx_assert(&malloc_mtx, MA_OWNED);
1343 mtx_unlock(&malloc_mtx);
1345 buflen = sizeof(struct malloc_type *) * count;
1346 bufmtp = malloc(buflen, M_TEMP, M_WAITOK);
1348 mtx_lock(&malloc_mtx);
1350 if (count < kmemcount) {
1351 free(bufmtp, M_TEMP);
1355 for (mtp = kmemstatistics, i = 0; mtp != NULL; mtp = mtp->ks_next, i++)
1358 mtx_unlock(&malloc_mtx);
1360 for (i = 0; i < count; i++)
1361 (func)(bufmtp[i], arg);
1363 free(bufmtp, M_TEMP);
1368 get_malloc_stats(const struct malloc_type_internal *mtip, uint64_t *allocs,
1371 const struct malloc_type_stats *mtsp;
1372 uint64_t frees, alloced, freed;
1379 for (i = 0; i <= mp_maxid; i++) {
1380 mtsp = zpcpu_get_cpu(mtip->mti_stats, i);
1382 *allocs += mtsp->mts_numallocs;
1383 frees += mtsp->mts_numfrees;
1384 alloced += mtsp->mts_memalloced;
1385 freed += mtsp->mts_memfreed;
1387 *inuse = *allocs - frees;
1388 return (alloced - freed);
1391 DB_SHOW_COMMAND(malloc, db_show_malloc)
1393 const char *fmt_hdr, *fmt_entry;
1394 struct malloc_type *mtp;
1395 uint64_t allocs, inuse;
1397 /* variables for sorting */
1398 struct malloc_type *last_mtype, *cur_mtype;
1399 int64_t cur_size, last_size;
1402 if (modif[0] == 'i') {
1403 fmt_hdr = "%s,%s,%s,%s\n";
1404 fmt_entry = "\"%s\",%ju,%jdK,%ju\n";
1406 fmt_hdr = "%18s %12s %12s %12s\n";
1407 fmt_entry = "%18s %12ju %12jdK %12ju\n";
1410 db_printf(fmt_hdr, "Type", "InUse", "MemUse", "Requests");
1412 /* Select sort, largest size first. */
1414 last_size = INT64_MAX;
1420 for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) {
1422 * In the case of size ties, print out mtypes
1423 * in the order they are encountered. That is,
1424 * when we encounter the most recently output
1425 * mtype, we have already printed all preceding
1426 * ties, and we must print all following ties.
1428 if (mtp == last_mtype) {
1432 size = get_malloc_stats(&mtp->ks_mti, &allocs,
1434 if (size > cur_size && size < last_size + ties) {
1439 if (cur_mtype == NULL)
1442 size = get_malloc_stats(&cur_mtype->ks_mti, &allocs, &inuse);
1443 db_printf(fmt_entry, cur_mtype->ks_shortdesc, inuse,
1444 howmany(size, 1024), allocs);
1449 last_mtype = cur_mtype;
1450 last_size = cur_size;
1454 #if MALLOC_DEBUG_MAXZONES > 1
1455 DB_SHOW_COMMAND(multizone_matches, db_show_multizone_matches)
1457 struct malloc_type_internal *mtip;
1458 struct malloc_type *mtp;
1462 db_printf("Usage: show multizone_matches <malloc type/addr>\n");
1466 if (mtp->ks_version != M_VERSION) {
1467 db_printf("Version %lx does not match expected %x\n",
1468 mtp->ks_version, M_VERSION);
1472 mtip = &mtp->ks_mti;
1473 subzone = mtip->mti_zone;
1475 for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) {
1476 mtip = &mtp->ks_mti;
1477 if (mtip->mti_zone != subzone)
1479 db_printf("%s\n", mtp->ks_shortdesc);
1484 #endif /* MALLOC_DEBUG_MAXZONES > 1 */