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 #define DEBUG_REDZONE_ARG_DEF , unsigned long osize
115 #define DEBUG_REDZONE_ARG , osize
117 #define DEBUG_REDZONE_ARG_DEF
118 #define DEBUG_REDZONE_ARG
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'.
127 #ifndef REALLOC_FRACTION
128 #define REALLOC_FRACTION 1 /* new block if <= half the size */
132 * Centrally define some common malloc types.
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");
138 static struct malloc_type *kmemstatistics;
139 static int kmemcount;
141 #define KMEM_ZSHIFT 4
142 #define KMEM_ZBASE 16
143 #define KMEM_ZMASK (KMEM_ZBASE - 1)
145 #define KMEM_ZMAX 65536
146 #define KMEM_ZSIZE (KMEM_ZMAX >> KMEM_ZSHIFT)
147 static uint8_t kmemsize[KMEM_ZSIZE + 1];
149 #ifndef MALLOC_DEBUG_MAXZONES
150 #define MALLOC_DEBUG_MAXZONES 1
152 static int numzones = MALLOC_DEBUG_MAXZONES;
155 * Small malloc(9) memory allocations are allocated from a set of UMA buckets
158 * Warning: the layout of the struct is duplicated in libmemstat for KVM support.
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.
167 uma_zone_t kz_zone[MALLOC_DEBUG_MAXZONES];
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", },
187 SYSCTL_ULONG(_vm, OID_AUTO, kmem_size, CTLFLAG_RDTUN, &vm_kmem_size, 0,
188 "Size of kernel memory");
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");
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");
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");
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");
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");
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");
216 static SYSCTL_NODE(_vm, OID_AUTO, malloc, CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
217 "Malloc information");
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");
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");
230 * The malloc_mtx protects the kmemstatistics linked list.
232 struct mtx malloc_mtx;
234 static int sysctl_kern_malloc_stats(SYSCTL_HANDLER_ARGS);
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");
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.
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");
256 sysctl_kmem_map_size(SYSCTL_HANDLER_ARGS)
261 return (sysctl_handle_long(oidp, &size, 0, req));
265 sysctl_kmem_map_free(SYSCTL_HANDLER_ARGS)
269 /* The sysctl is unsigned, implement as a saturation value. */
276 return (sysctl_handle_long(oidp, &size, 0, req));
280 sysctl_vm_malloc_zone_sizes(SYSCTL_HANDLER_ARGS)
282 int sizes[nitems(kmemzones)];
285 for (i = 0; i < nitems(kmemzones); i++) {
286 sizes[i] = kmemzones[i].kz_size;
289 return (SYSCTL_OUT(req, &sizes, sizeof(sizes)));
293 * malloc(9) uma zone separation -- sub-page buffer overruns in one
294 * malloc type will affect only a subset of other malloc types.
296 #if MALLOC_DEBUG_MAXZONES > 1
298 tunable_set_numzones(void)
301 TUNABLE_INT_FETCH("debug.malloc.numzones",
304 /* Sanity check the number of malloc uma zones. */
307 if (numzones > MALLOC_DEBUG_MAXZONES)
308 numzones = MALLOC_DEBUG_MAXZONES;
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");
315 * Any number that changes regularly is an okay choice for the
316 * offset. Build numbers are pretty good of you have them.
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.");
325 mtp_set_subzone(struct malloc_type *mtp)
327 struct malloc_type_internal *mtip;
333 desc = mtp->ks_shortdesc;
334 if (desc == NULL || (len = strlen(desc)) == 0)
337 val = desc[zone_offset % len];
338 mtip->mti_zone = (val % numzones);
342 mtp_get_subzone(struct malloc_type *mtp)
344 struct malloc_type_internal *mtip;
348 KASSERT(mtip->mti_zone < numzones,
349 ("mti_zone %u out of range %d",
350 mtip->mti_zone, numzones));
351 return (mtip->mti_zone);
353 #elif MALLOC_DEBUG_MAXZONES == 0
354 #error "MALLOC_DEBUG_MAXZONES must be positive."
357 mtp_set_subzone(struct malloc_type *mtp)
359 struct malloc_type_internal *mtip;
366 mtp_get_subzone(struct malloc_type *mtp)
371 #endif /* MALLOC_DEBUG_MAXZONES > 1 */
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
380 malloc_type_zone_allocated(struct malloc_type *mtp, unsigned long size,
383 struct malloc_type_internal *mtip;
384 struct malloc_type_stats *mtsp;
388 mtsp = zpcpu_get(mtip->mti_stats);
390 mtsp->mts_memalloced += size;
391 mtsp->mts_numallocs++;
394 mtsp->mts_size |= 1 << zindx;
397 if (__predict_false(dtrace_malloc_enabled)) {
398 uint32_t probe_id = mtip->mti_probes[DTMALLOC_PROBE_MALLOC];
400 (dtrace_malloc_probe)(probe_id,
401 (uintptr_t) mtp, (uintptr_t) mtip,
402 (uintptr_t) mtsp, size, zindx);
410 malloc_type_allocated(struct malloc_type *mtp, unsigned long size)
414 malloc_type_zone_allocated(mtp, size, -1);
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
424 malloc_type_freed(struct malloc_type *mtp, unsigned long size)
426 struct malloc_type_internal *mtip;
427 struct malloc_type_stats *mtsp;
431 mtsp = zpcpu_get(mtip->mti_stats);
432 mtsp->mts_memfreed += size;
433 mtsp->mts_numfrees++;
436 if (__predict_false(dtrace_malloc_enabled)) {
437 uint32_t probe_id = mtip->mti_probes[DTMALLOC_PROBE_FREE];
439 (dtrace_malloc_probe)(probe_id,
440 (uintptr_t) mtp, (uintptr_t) mtip,
441 (uintptr_t) mtsp, size, 0);
451 * Allocate a block of physically contiguous memory.
453 * If M_NOWAIT is set, this routine will not block and return NULL if
454 * the allocation fails.
457 contigmalloc(unsigned long size, struct malloc_type *type, int flags,
458 vm_paddr_t low, vm_paddr_t high, unsigned long alignment,
463 ret = (void *)kmem_alloc_contig(size, flags, low, high, alignment,
464 boundary, VM_MEMATTR_DEFAULT);
466 malloc_type_allocated(type, round_page(size));
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)
477 ret = (void *)kmem_alloc_contig_domainset(ds, size, flags, low, high,
478 alignment, boundary, VM_MEMATTR_DEFAULT);
480 malloc_type_allocated(type, round_page(size));
487 * Free a block of memory allocated by contigmalloc.
489 * This routine may not block.
492 contigfree(void *addr, unsigned long size, struct malloc_type *type)
495 kmem_free((vm_offset_t)addr, size);
496 malloc_type_freed(type, round_page(size));
501 malloc_dbg(caddr_t *vap, size_t *sizep, struct malloc_type *mtp,
507 KASSERT(mtp->ks_version == M_VERSION, ("malloc: bad malloc type version"));
509 * Check that exactly one of M_WAITOK or M_NOWAIT is specified.
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);
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);
529 return (EJUSTRETURN);
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())) {
538 epoch_trace_list(curthread);
541 ("malloc(M_WAITOK) with sleeping prohibited"));
544 KASSERT(curthread->td_critnest == 0 || SCHEDULER_STOPPED(),
545 ("malloc: called with spinlock or critical section held"));
547 #ifdef DEBUG_MEMGUARD
548 if (memguard_cmp_mtp(mtp, *sizep)) {
549 *vap = memguard_alloc(*sizep, flags);
551 return (EJUSTRETURN);
552 /* This is unfortunate but should not be fatal. */
557 *sizep = redzone_size_ntor(*sizep);
565 * Handle large allocations and frees by using kmem_malloc directly.
568 malloc_large_slab(uma_slab_t slab)
572 va = (uintptr_t)slab;
573 return ((va & 1) != 0);
577 malloc_large_size(uma_slab_t slab)
581 va = (uintptr_t)slab;
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)
593 sz = roundup(*size, PAGE_SIZE);
594 kva = kmem_malloc_domainset(policy, sz, flags);
596 /* The low bit is unused for slab pointers. */
597 vsetzoneslab(kva, NULL, (void *)((sz << 1) | 1));
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"));
609 va = redzone_setup(va, osize);
615 free_large(void *addr, size_t size)
618 kmem_free((vm_offset_t)addr, size);
625 * Allocate a block of memory.
627 * If M_NOWAIT is set, this routine will not block and return NULL if
628 * the allocation fails.
631 (malloc)(size_t size, struct malloc_type *mtp, int flags)
637 unsigned long osize = size;
640 MPASS((flags & M_EXEC) == 0);
644 if (malloc_dbg(&va, &size, mtp, flags) != 0)
648 if (__predict_false(size > kmem_zmax))
649 return (malloc_large(&size, mtp, DOMAINSET_RR(), flags
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);
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"));
666 va = redzone_setup(va, osize);
668 return ((void *) va);
672 malloc_domain(size_t *sizep, int *indxp, struct malloc_type *mtp, int domain,
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);
689 *sizep = zone->uz_size;
695 malloc_domainset(size_t size, struct malloc_type *mtp, struct domainset *ds,
698 struct vm_domainset_iter di;
703 unsigned long osize = size;
706 MPASS((flags & M_EXEC) == 0);
710 if (malloc_dbg(&va, &size, mtp, flags) != 0)
714 if (__predict_false(size > kmem_zmax))
715 return (malloc_large(&size, mtp, DOMAINSET_RR(), flags
718 vm_domainset_iter_policy_init(&di, ds, &domain, &flags);
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"));
729 va = redzone_setup(va, osize);
735 * Allocate an executable area.
738 malloc_exec(size_t size, struct malloc_type *mtp, int flags)
741 return (malloc_domainset_exec(size, mtp, DOMAINSET_RR(), flags));
745 malloc_domainset_exec(size_t size, struct malloc_type *mtp, struct domainset *ds,
749 unsigned long osize = size;
759 if (malloc_dbg(&va, &size, mtp, flags) != 0)
763 return (malloc_large(&size, mtp, ds, flags DEBUG_REDZONE_ARG));
767 malloc_aligned(size_t size, size_t align, struct malloc_type *type, int flags)
769 return (malloc_domainset_aligned(size, align, type, DOMAINSET_RR(),
774 malloc_domainset_aligned(size_t size, size_t align,
775 struct malloc_type *mtp, struct domainset *ds, int flags)
780 KASSERT(align != 0 && powerof2(align),
781 ("malloc_domainset_aligned: wrong align %#zx size %#zx",
783 KASSERT(align <= PAGE_SIZE,
784 ("malloc_domainset_aligned: align %#zx (size %#zx) too large",
788 * Round the allocation size up to the next power of 2,
789 * because we can only guarantee alignment for
790 * power-of-2-sized allocations. Further increase the
791 * allocation size to align if the rounded size is less than
792 * align, since malloc zones provide alignment equal to their
795 asize = size <= align ? align : 1UL << flsl(size - 1);
797 res = malloc_domainset(asize, mtp, ds, flags);
798 KASSERT(res == NULL || ((uintptr_t)res & (align - 1)) == 0,
799 ("malloc_domainset_aligned: result not aligned %p size %#zx "
800 "allocsize %#zx align %#zx", res, size, asize, align));
805 mallocarray(size_t nmemb, size_t size, struct malloc_type *type, int flags)
808 if (WOULD_OVERFLOW(nmemb, size))
809 panic("mallocarray: %zu * %zu overflowed", nmemb, size);
811 return (malloc(size * nmemb, type, flags));
815 mallocarray_domainset(size_t nmemb, size_t size, struct malloc_type *type,
816 struct domainset *ds, int flags)
819 if (WOULD_OVERFLOW(nmemb, size))
820 panic("mallocarray_domainset: %zu * %zu overflowed", nmemb, size);
822 return (malloc_domainset(size * nmemb, type, ds, flags));
827 free_save_type(void *addr, struct malloc_type *mtp, u_long size)
829 struct malloc_type **mtpp = addr;
832 * Cache a pointer to the malloc_type that most recently freed
833 * this memory here. This way we know who is most likely to
834 * have stepped on it later.
836 * This code assumes that size is a multiple of 8 bytes for
839 mtpp = (struct malloc_type **) ((unsigned long)mtpp & ~UMA_ALIGN_PTR);
840 mtpp += (size - sizeof(struct malloc_type *)) /
841 sizeof(struct malloc_type *);
848 free_dbg(void **addrp, struct malloc_type *mtp)
853 KASSERT(mtp->ks_version == M_VERSION, ("free: bad malloc type version"));
854 KASSERT(curthread->td_critnest == 0 || SCHEDULER_STOPPED(),
855 ("free: called with spinlock or critical section held"));
857 /* free(NULL, ...) does nothing */
859 return (EJUSTRETURN);
861 #ifdef DEBUG_MEMGUARD
862 if (is_memguard_addr(addr)) {
864 return (EJUSTRETURN);
870 *addrp = redzone_addr_ntor(addr);
880 * Free a block of memory allocated by malloc.
882 * This routine may not block.
885 free(void *addr, struct malloc_type *mtp)
892 if (free_dbg(&addr, mtp) != 0)
895 /* free(NULL, ...) does nothing */
899 vtozoneslab((vm_offset_t)addr & (~UMA_SLAB_MASK), &zone, &slab);
901 panic("free: address %p(%p) has not been allocated.\n",
902 addr, (void *)((u_long)addr & (~UMA_SLAB_MASK)));
904 if (__predict_true(!malloc_large_slab(slab))) {
905 size = zone->uz_size;
907 free_save_type(addr, mtp, size);
909 uma_zfree_arg(zone, addr, slab);
911 size = malloc_large_size(slab);
912 free_large(addr, size);
914 malloc_type_freed(mtp, size);
920 * Zero then free a block of memory allocated by malloc.
922 * This routine may not block.
925 zfree(void *addr, struct malloc_type *mtp)
932 if (free_dbg(&addr, mtp) != 0)
935 /* free(NULL, ...) does nothing */
939 vtozoneslab((vm_offset_t)addr & (~UMA_SLAB_MASK), &zone, &slab);
941 panic("free: address %p(%p) has not been allocated.\n",
942 addr, (void *)((u_long)addr & (~UMA_SLAB_MASK)));
944 if (__predict_true(!malloc_large_slab(slab))) {
945 size = zone->uz_size;
947 free_save_type(addr, mtp, size);
949 explicit_bzero(addr, size);
950 uma_zfree_arg(zone, addr, slab);
952 size = malloc_large_size(slab);
953 explicit_bzero(addr, size);
954 free_large(addr, size);
956 malloc_type_freed(mtp, size);
960 * realloc: change the size of a memory block
963 realloc(void *addr, size_t size, struct malloc_type *mtp, int flags)
970 KASSERT(mtp->ks_version == M_VERSION,
971 ("realloc: bad malloc type version"));
972 KASSERT(curthread->td_critnest == 0 || SCHEDULER_STOPPED(),
973 ("realloc: called with spinlock or critical section held"));
975 /* realloc(NULL, ...) is equivalent to malloc(...) */
977 return (malloc(size, mtp, flags));
980 * XXX: Should report free of old memory and alloc of new memory to
984 #ifdef DEBUG_MEMGUARD
985 if (is_memguard_addr(addr))
986 return (memguard_realloc(addr, size, mtp, flags));
992 alloc = redzone_get_size(addr);
994 vtozoneslab((vm_offset_t)addr & (~UMA_SLAB_MASK), &zone, &slab);
997 KASSERT(slab != NULL,
998 ("realloc: address %p out of range", (void *)addr));
1000 /* Get the size of the original block */
1001 if (!malloc_large_slab(slab))
1002 alloc = zone->uz_size;
1004 alloc = malloc_large_size(slab);
1006 /* Reuse the original block if appropriate */
1008 && (size > (alloc >> REALLOC_FRACTION) || alloc == MINALLOCSIZE))
1010 #endif /* !DEBUG_REDZONE */
1012 /* Allocate a new, bigger (or smaller) block */
1013 if ((newaddr = malloc(size, mtp, flags)) == NULL)
1016 /* Copy over original contents */
1017 bcopy(addr, newaddr, min(size, alloc));
1023 * reallocf: same as realloc() but free memory on failure.
1026 reallocf(void *addr, size_t size, struct malloc_type *mtp, int flags)
1030 if ((mem = realloc(addr, size, mtp, flags)) == NULL)
1036 * malloc_size: returns the number of bytes allocated for a request of the
1040 malloc_size(size_t size)
1044 if (size > kmem_zmax)
1046 if (size & KMEM_ZMASK)
1047 size = (size & ~KMEM_ZMASK) + KMEM_ZBASE;
1048 indx = kmemsize[size >> KMEM_ZSHIFT];
1049 return (kmemzones[indx].kz_size);
1053 * malloc_usable_size: returns the usable size of the allocation.
1056 malloc_usable_size(const void *addr)
1058 #ifndef DEBUG_REDZONE
1067 #ifdef DEBUG_MEMGUARD
1068 if (is_memguard_addr(__DECONST(void *, addr)))
1069 return (memguard_get_req_size(addr));
1072 #ifdef DEBUG_REDZONE
1073 size = redzone_get_size(__DECONST(void *, addr));
1075 vtozoneslab((vm_offset_t)addr & (~UMA_SLAB_MASK), &zone, &slab);
1077 panic("malloc_usable_size: address %p(%p) is not allocated.\n",
1078 addr, (void *)((u_long)addr & (~UMA_SLAB_MASK)));
1080 if (!malloc_large_slab(slab))
1081 size = zone->uz_size;
1083 size = malloc_large_size(slab);
1088 CTASSERT(VM_KMEM_SIZE_SCALE >= 1);
1091 * Initialize the kernel memory (kmem) arena.
1100 if (vm_kmem_size == 0)
1101 vm_kmem_size = VM_KMEM_SIZE;
1103 #ifdef VM_KMEM_SIZE_MIN
1104 if (vm_kmem_size_min == 0)
1105 vm_kmem_size_min = VM_KMEM_SIZE_MIN;
1107 #ifdef VM_KMEM_SIZE_MAX
1108 if (vm_kmem_size_max == 0)
1109 vm_kmem_size_max = VM_KMEM_SIZE_MAX;
1112 * Calculate the amount of kernel virtual address (KVA) space that is
1113 * preallocated to the kmem arena. In order to support a wide range
1114 * of machines, it is a function of the physical memory size,
1117 * min(max(physical memory size / VM_KMEM_SIZE_SCALE,
1118 * VM_KMEM_SIZE_MIN), VM_KMEM_SIZE_MAX)
1120 * Every architecture must define an integral value for
1121 * VM_KMEM_SIZE_SCALE. However, the definitions of VM_KMEM_SIZE_MIN
1122 * and VM_KMEM_SIZE_MAX, which represent respectively the floor and
1123 * ceiling on this preallocation, are optional. Typically,
1124 * VM_KMEM_SIZE_MAX is itself a function of the available KVA space on
1125 * a given architecture.
1127 mem_size = vm_cnt.v_page_count;
1128 if (mem_size <= 32768) /* delphij XXX 128MB */
1129 kmem_zmax = PAGE_SIZE;
1131 if (vm_kmem_size_scale < 1)
1132 vm_kmem_size_scale = VM_KMEM_SIZE_SCALE;
1135 * Check if we should use defaults for the "vm_kmem_size"
1138 if (vm_kmem_size == 0) {
1139 vm_kmem_size = mem_size / vm_kmem_size_scale;
1140 vm_kmem_size = vm_kmem_size * PAGE_SIZE < vm_kmem_size ?
1141 vm_kmem_size_max : vm_kmem_size * PAGE_SIZE;
1142 if (vm_kmem_size_min > 0 && vm_kmem_size < vm_kmem_size_min)
1143 vm_kmem_size = vm_kmem_size_min;
1144 if (vm_kmem_size_max > 0 && vm_kmem_size >= vm_kmem_size_max)
1145 vm_kmem_size = vm_kmem_size_max;
1147 if (vm_kmem_size == 0)
1148 panic("Tune VM_KMEM_SIZE_* for the platform");
1151 * The amount of KVA space that is preallocated to the
1152 * kmem arena can be set statically at compile-time or manually
1153 * through the kernel environment. However, it is still limited to
1154 * twice the physical memory size, which has been sufficient to handle
1155 * the most severe cases of external fragmentation in the kmem arena.
1157 if (vm_kmem_size / 2 / PAGE_SIZE > mem_size)
1158 vm_kmem_size = 2 * mem_size * PAGE_SIZE;
1160 vm_kmem_size = round_page(vm_kmem_size);
1161 #ifdef DEBUG_MEMGUARD
1162 tmp = memguard_fudge(vm_kmem_size, kernel_map);
1168 #ifdef DEBUG_MEMGUARD
1170 * Initialize MemGuard if support compiled in. MemGuard is a
1171 * replacement allocator used for detecting tamper-after-free
1172 * scenarios as they occur. It is only used for debugging.
1174 memguard_init(kernel_arena);
1179 * Initialize the kernel memory allocator
1183 mallocinit(void *dummy)
1188 mtx_init(&malloc_mtx, "malloc", NULL, MTX_DEF);
1192 if (kmem_zmax < PAGE_SIZE || kmem_zmax > KMEM_ZMAX)
1193 kmem_zmax = KMEM_ZMAX;
1195 for (i = 0, indx = 0; kmemzones[indx].kz_size != 0; indx++) {
1196 int size = kmemzones[indx].kz_size;
1197 const char *name = kmemzones[indx].kz_name;
1201 align = UMA_ALIGN_PTR;
1202 if (powerof2(size) && size > sizeof(void *))
1203 align = MIN(size, PAGE_SIZE) - 1;
1204 for (subzone = 0; subzone < numzones; subzone++) {
1205 kmemzones[indx].kz_zone[subzone] =
1206 uma_zcreate(name, size,
1208 mtrash_ctor, mtrash_dtor, mtrash_init, mtrash_fini,
1210 NULL, NULL, NULL, NULL,
1212 align, UMA_ZONE_MALLOC);
1214 for (;i <= size; i+= KMEM_ZBASE)
1215 kmemsize[i >> KMEM_ZSHIFT] = indx;
1218 SYSINIT(kmem, SI_SUB_KMEM, SI_ORDER_SECOND, mallocinit, NULL);
1221 malloc_init(void *data)
1223 struct malloc_type_internal *mtip;
1224 struct malloc_type *mtp;
1226 KASSERT(vm_cnt.v_page_count != 0, ("malloc_register before vm_init"));
1229 if (mtp->ks_version != M_VERSION)
1230 panic("malloc_init: type %s with unsupported version %lu",
1231 mtp->ks_shortdesc, mtp->ks_version);
1233 mtip = &mtp->ks_mti;
1234 mtip->mti_stats = uma_zalloc_pcpu(pcpu_zone_64, M_WAITOK | M_ZERO);
1235 mtp_set_subzone(mtp);
1237 mtx_lock(&malloc_mtx);
1238 mtp->ks_next = kmemstatistics;
1239 kmemstatistics = mtp;
1241 mtx_unlock(&malloc_mtx);
1245 malloc_uninit(void *data)
1247 struct malloc_type_internal *mtip;
1248 struct malloc_type_stats *mtsp;
1249 struct malloc_type *mtp, *temp;
1250 long temp_allocs, temp_bytes;
1254 KASSERT(mtp->ks_version == M_VERSION,
1255 ("malloc_uninit: bad malloc type version"));
1257 mtx_lock(&malloc_mtx);
1258 mtip = &mtp->ks_mti;
1259 if (mtp != kmemstatistics) {
1260 for (temp = kmemstatistics; temp != NULL;
1261 temp = temp->ks_next) {
1262 if (temp->ks_next == mtp) {
1263 temp->ks_next = mtp->ks_next;
1268 ("malloc_uninit: type '%s' not found", mtp->ks_shortdesc));
1270 kmemstatistics = mtp->ks_next;
1272 mtx_unlock(&malloc_mtx);
1275 * Look for memory leaks.
1277 temp_allocs = temp_bytes = 0;
1278 for (i = 0; i <= mp_maxid; i++) {
1279 mtsp = zpcpu_get_cpu(mtip->mti_stats, i);
1280 temp_allocs += mtsp->mts_numallocs;
1281 temp_allocs -= mtsp->mts_numfrees;
1282 temp_bytes += mtsp->mts_memalloced;
1283 temp_bytes -= mtsp->mts_memfreed;
1285 if (temp_allocs > 0 || temp_bytes > 0) {
1286 printf("Warning: memory type %s leaked memory on destroy "
1287 "(%ld allocations, %ld bytes leaked).\n", mtp->ks_shortdesc,
1288 temp_allocs, temp_bytes);
1291 uma_zfree_pcpu(pcpu_zone_64, mtip->mti_stats);
1294 struct malloc_type *
1295 malloc_desc2type(const char *desc)
1297 struct malloc_type *mtp;
1299 mtx_assert(&malloc_mtx, MA_OWNED);
1300 for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) {
1301 if (strcmp(mtp->ks_shortdesc, desc) == 0)
1308 sysctl_kern_malloc_stats(SYSCTL_HANDLER_ARGS)
1310 struct malloc_type_stream_header mtsh;
1311 struct malloc_type_internal *mtip;
1312 struct malloc_type_stats *mtsp, zeromts;
1313 struct malloc_type_header mth;
1314 struct malloc_type *mtp;
1318 error = sysctl_wire_old_buffer(req, 0);
1321 sbuf_new_for_sysctl(&sbuf, NULL, 128, req);
1322 sbuf_clear_flags(&sbuf, SBUF_INCLUDENUL);
1323 mtx_lock(&malloc_mtx);
1325 bzero(&zeromts, sizeof(zeromts));
1328 * Insert stream header.
1330 bzero(&mtsh, sizeof(mtsh));
1331 mtsh.mtsh_version = MALLOC_TYPE_STREAM_VERSION;
1332 mtsh.mtsh_maxcpus = MAXCPU;
1333 mtsh.mtsh_count = kmemcount;
1334 (void)sbuf_bcat(&sbuf, &mtsh, sizeof(mtsh));
1337 * Insert alternating sequence of type headers and type statistics.
1339 for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) {
1340 mtip = &mtp->ks_mti;
1343 * Insert type header.
1345 bzero(&mth, sizeof(mth));
1346 strlcpy(mth.mth_name, mtp->ks_shortdesc, MALLOC_MAX_NAME);
1347 (void)sbuf_bcat(&sbuf, &mth, sizeof(mth));
1350 * Insert type statistics for each CPU.
1352 for (i = 0; i <= mp_maxid; i++) {
1353 mtsp = zpcpu_get_cpu(mtip->mti_stats, i);
1354 (void)sbuf_bcat(&sbuf, mtsp, sizeof(*mtsp));
1357 * Fill in the missing CPUs.
1359 for (; i < MAXCPU; i++) {
1360 (void)sbuf_bcat(&sbuf, &zeromts, sizeof(zeromts));
1363 mtx_unlock(&malloc_mtx);
1364 error = sbuf_finish(&sbuf);
1369 SYSCTL_PROC(_kern, OID_AUTO, malloc_stats,
1370 CTLFLAG_RD | CTLTYPE_STRUCT | CTLFLAG_MPSAFE, 0, 0,
1371 sysctl_kern_malloc_stats, "s,malloc_type_ustats",
1372 "Return malloc types");
1374 SYSCTL_INT(_kern, OID_AUTO, malloc_count, CTLFLAG_RD, &kmemcount, 0,
1375 "Count of kernel malloc types");
1378 malloc_type_list(malloc_type_list_func_t *func, void *arg)
1380 struct malloc_type *mtp, **bufmtp;
1384 mtx_lock(&malloc_mtx);
1386 mtx_assert(&malloc_mtx, MA_OWNED);
1388 mtx_unlock(&malloc_mtx);
1390 buflen = sizeof(struct malloc_type *) * count;
1391 bufmtp = malloc(buflen, M_TEMP, M_WAITOK);
1393 mtx_lock(&malloc_mtx);
1395 if (count < kmemcount) {
1396 free(bufmtp, M_TEMP);
1400 for (mtp = kmemstatistics, i = 0; mtp != NULL; mtp = mtp->ks_next, i++)
1403 mtx_unlock(&malloc_mtx);
1405 for (i = 0; i < count; i++)
1406 (func)(bufmtp[i], arg);
1408 free(bufmtp, M_TEMP);
1413 get_malloc_stats(const struct malloc_type_internal *mtip, uint64_t *allocs,
1416 const struct malloc_type_stats *mtsp;
1417 uint64_t frees, alloced, freed;
1424 for (i = 0; i <= mp_maxid; i++) {
1425 mtsp = zpcpu_get_cpu(mtip->mti_stats, i);
1427 *allocs += mtsp->mts_numallocs;
1428 frees += mtsp->mts_numfrees;
1429 alloced += mtsp->mts_memalloced;
1430 freed += mtsp->mts_memfreed;
1432 *inuse = *allocs - frees;
1433 return (alloced - freed);
1436 DB_SHOW_COMMAND(malloc, db_show_malloc)
1438 const char *fmt_hdr, *fmt_entry;
1439 struct malloc_type *mtp;
1440 uint64_t allocs, inuse;
1442 /* variables for sorting */
1443 struct malloc_type *last_mtype, *cur_mtype;
1444 int64_t cur_size, last_size;
1447 if (modif[0] == 'i') {
1448 fmt_hdr = "%s,%s,%s,%s\n";
1449 fmt_entry = "\"%s\",%ju,%jdK,%ju\n";
1451 fmt_hdr = "%18s %12s %12s %12s\n";
1452 fmt_entry = "%18s %12ju %12jdK %12ju\n";
1455 db_printf(fmt_hdr, "Type", "InUse", "MemUse", "Requests");
1457 /* Select sort, largest size first. */
1459 last_size = INT64_MAX;
1465 for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) {
1467 * In the case of size ties, print out mtypes
1468 * in the order they are encountered. That is,
1469 * when we encounter the most recently output
1470 * mtype, we have already printed all preceding
1471 * ties, and we must print all following ties.
1473 if (mtp == last_mtype) {
1477 size = get_malloc_stats(&mtp->ks_mti, &allocs,
1479 if (size > cur_size && size < last_size + ties) {
1484 if (cur_mtype == NULL)
1487 size = get_malloc_stats(&cur_mtype->ks_mti, &allocs, &inuse);
1488 db_printf(fmt_entry, cur_mtype->ks_shortdesc, inuse,
1489 howmany(size, 1024), allocs);
1494 last_mtype = cur_mtype;
1495 last_size = cur_size;
1499 #if MALLOC_DEBUG_MAXZONES > 1
1500 DB_SHOW_COMMAND(multizone_matches, db_show_multizone_matches)
1502 struct malloc_type_internal *mtip;
1503 struct malloc_type *mtp;
1507 db_printf("Usage: show multizone_matches <malloc type/addr>\n");
1511 if (mtp->ks_version != M_VERSION) {
1512 db_printf("Version %lx does not match expected %x\n",
1513 mtp->ks_version, M_VERSION);
1517 mtip = &mtp->ks_mti;
1518 subzone = mtip->mti_zone;
1520 for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) {
1521 mtip = &mtp->ks_mti;
1522 if (mtip->mti_zone != subzone)
1524 db_printf("%s\n", mtp->ks_shortdesc);
1529 #endif /* MALLOC_DEBUG_MAXZONES > 1 */