2 * Copyright (c) 1987, 1991, 1993
3 * The Regents of the University of California.
4 * Copyright (c) 2005-2009 Robert N. M. Watson
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
10 * 1. Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in the
14 * documentation and/or other materials provided with the distribution.
15 * 4. Neither the name of the University nor the names of its contributors
16 * may be used to endorse or promote products derived from this software
17 * without specific prior written permission.
19 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31 * @(#)kern_malloc.c 8.3 (Berkeley) 1/4/94
35 * Kernel malloc(9) implementation -- general purpose kernel memory allocator
36 * based on memory types. Back end is implemented using the UMA(9) zone
37 * allocator. A set of fixed-size buckets are used for smaller allocations,
38 * and a special UMA allocation interface is used for larger allocations.
39 * Callers declare memory types, and statistics are maintained independently
40 * for each memory type. Statistics are maintained per-CPU for performance
41 * reasons. See malloc(9) and comments in malloc.h for a detailed
45 #include <sys/cdefs.h>
46 __FBSDID("$FreeBSD$");
51 #include <sys/param.h>
52 #include <sys/systm.h>
54 #include <sys/kernel.h>
56 #include <sys/malloc.h>
57 #include <sys/mutex.h>
58 #include <sys/vmmeter.h>
61 #include <sys/sysctl.h>
67 #include <vm/vm_pageout.h>
68 #include <vm/vm_param.h>
69 #include <vm/vm_kern.h>
70 #include <vm/vm_extern.h>
71 #include <vm/vm_map.h>
72 #include <vm/vm_page.h>
74 #include <vm/uma_int.h>
75 #include <vm/uma_dbg.h>
78 #include <vm/memguard.h>
81 #include <vm/redzone.h>
84 #if defined(INVARIANTS) && defined(__i386__)
85 #include <machine/cpu.h>
91 #include <sys/dtrace_bsd.h>
93 dtrace_malloc_probe_func_t dtrace_malloc_probe;
97 * When realloc() is called, if the new size is sufficiently smaller than
98 * the old size, realloc() will allocate a new, smaller block to avoid
99 * wasting memory. 'Sufficiently smaller' is defined as: newsize <=
100 * oldsize / 2^n, where REALLOC_FRACTION defines the value of 'n'.
102 #ifndef REALLOC_FRACTION
103 #define REALLOC_FRACTION 1 /* new block if <= half the size */
107 * Centrally define some common malloc types.
109 MALLOC_DEFINE(M_CACHE, "cache", "Various Dynamically allocated caches");
110 MALLOC_DEFINE(M_DEVBUF, "devbuf", "device driver memory");
111 MALLOC_DEFINE(M_TEMP, "temp", "misc temporary data buffers");
113 MALLOC_DEFINE(M_IP6OPT, "ip6opt", "IPv6 options");
114 MALLOC_DEFINE(M_IP6NDP, "ip6ndp", "IPv6 Neighbor Discovery");
116 static struct malloc_type *kmemstatistics;
117 static int kmemcount;
119 #define KMEM_ZSHIFT 4
120 #define KMEM_ZBASE 16
121 #define KMEM_ZMASK (KMEM_ZBASE - 1)
123 #define KMEM_ZMAX 65536
124 #define KMEM_ZSIZE (KMEM_ZMAX >> KMEM_ZSHIFT)
125 static uint8_t kmemsize[KMEM_ZSIZE + 1];
127 #ifndef MALLOC_DEBUG_MAXZONES
128 #define MALLOC_DEBUG_MAXZONES 1
130 static int numzones = MALLOC_DEBUG_MAXZONES;
133 * Small malloc(9) memory allocations are allocated from a set of UMA buckets
136 * XXX: The comment here used to read "These won't be powers of two for
137 * long." It's possible that a significant amount of wasted memory could be
138 * recovered by tuning the sizes of these buckets.
143 uma_zone_t kz_zone[MALLOC_DEBUG_MAXZONES];
162 * Zone to allocate malloc type descriptions from. For ABI reasons, memory
163 * types are described by a data structure passed by the declaring code, but
164 * the malloc(9) implementation has its own data structure describing the
165 * type and statistics. This permits the malloc(9)-internal data structures
166 * to be modified without breaking binary-compiled kernel modules that
167 * declare malloc types.
169 static uma_zone_t mt_zone;
172 SYSCTL_ULONG(_vm, OID_AUTO, kmem_size, CTLFLAG_RDTUN, &vm_kmem_size, 0,
173 "Size of kernel memory");
175 static u_long kmem_zmax = KMEM_ZMAX;
176 SYSCTL_ULONG(_vm, OID_AUTO, kmem_zmax, CTLFLAG_RDTUN, &kmem_zmax, 0,
177 "Maximum allocation size that malloc(9) would use UMA as backend");
179 static u_long vm_kmem_size_min;
180 SYSCTL_ULONG(_vm, OID_AUTO, kmem_size_min, CTLFLAG_RDTUN, &vm_kmem_size_min, 0,
181 "Minimum size of kernel memory");
183 static u_long vm_kmem_size_max;
184 SYSCTL_ULONG(_vm, OID_AUTO, kmem_size_max, CTLFLAG_RDTUN, &vm_kmem_size_max, 0,
185 "Maximum size of kernel memory");
187 static u_int vm_kmem_size_scale;
188 SYSCTL_UINT(_vm, OID_AUTO, kmem_size_scale, CTLFLAG_RDTUN, &vm_kmem_size_scale, 0,
189 "Scale factor for kernel memory size");
191 static int sysctl_kmem_map_size(SYSCTL_HANDLER_ARGS);
192 SYSCTL_PROC(_vm, OID_AUTO, kmem_map_size,
193 CTLFLAG_RD | CTLTYPE_ULONG | CTLFLAG_MPSAFE, NULL, 0,
194 sysctl_kmem_map_size, "LU", "Current kmem allocation size");
196 static int sysctl_kmem_map_free(SYSCTL_HANDLER_ARGS);
197 SYSCTL_PROC(_vm, OID_AUTO, kmem_map_free,
198 CTLFLAG_RD | CTLTYPE_ULONG | CTLFLAG_MPSAFE, NULL, 0,
199 sysctl_kmem_map_free, "LU", "Free space in kmem");
202 * The malloc_mtx protects the kmemstatistics linked list.
204 struct mtx malloc_mtx;
206 #ifdef MALLOC_PROFILE
207 uint64_t krequests[KMEM_ZSIZE + 1];
209 static int sysctl_kern_mprof(SYSCTL_HANDLER_ARGS);
212 static int sysctl_kern_malloc_stats(SYSCTL_HANDLER_ARGS);
215 * time_uptime of the last malloc(9) failure (induced or real).
217 static time_t t_malloc_fail;
219 #if defined(MALLOC_MAKE_FAILURES) || (MALLOC_DEBUG_MAXZONES > 1)
220 static SYSCTL_NODE(_debug, OID_AUTO, malloc, CTLFLAG_RD, 0,
221 "Kernel malloc debugging options");
225 * malloc(9) fault injection -- cause malloc failures every (n) mallocs when
226 * the caller specifies M_NOWAIT. If set to 0, no failures are caused.
228 #ifdef MALLOC_MAKE_FAILURES
229 static int malloc_failure_rate;
230 static int malloc_nowait_count;
231 static int malloc_failure_count;
232 SYSCTL_INT(_debug_malloc, OID_AUTO, failure_rate, CTLFLAG_RWTUN,
233 &malloc_failure_rate, 0, "Every (n) mallocs with M_NOWAIT will fail");
234 SYSCTL_INT(_debug_malloc, OID_AUTO, failure_count, CTLFLAG_RD,
235 &malloc_failure_count, 0, "Number of imposed M_NOWAIT malloc failures");
239 sysctl_kmem_map_size(SYSCTL_HANDLER_ARGS)
243 size = vmem_size(kmem_arena, VMEM_ALLOC);
244 return (sysctl_handle_long(oidp, &size, 0, req));
248 sysctl_kmem_map_free(SYSCTL_HANDLER_ARGS)
252 size = vmem_size(kmem_arena, VMEM_FREE);
253 return (sysctl_handle_long(oidp, &size, 0, req));
257 * malloc(9) uma zone separation -- sub-page buffer overruns in one
258 * malloc type will affect only a subset of other malloc types.
260 #if MALLOC_DEBUG_MAXZONES > 1
262 tunable_set_numzones(void)
265 TUNABLE_INT_FETCH("debug.malloc.numzones",
268 /* Sanity check the number of malloc uma zones. */
271 if (numzones > MALLOC_DEBUG_MAXZONES)
272 numzones = MALLOC_DEBUG_MAXZONES;
274 SYSINIT(numzones, SI_SUB_TUNABLES, SI_ORDER_ANY, tunable_set_numzones, NULL);
275 SYSCTL_INT(_debug_malloc, OID_AUTO, numzones, CTLFLAG_RDTUN | CTLFLAG_NOFETCH,
276 &numzones, 0, "Number of malloc uma subzones");
279 * Any number that changes regularly is an okay choice for the
280 * offset. Build numbers are pretty good of you have them.
282 static u_int zone_offset = __FreeBSD_version;
283 TUNABLE_INT("debug.malloc.zone_offset", &zone_offset);
284 SYSCTL_UINT(_debug_malloc, OID_AUTO, zone_offset, CTLFLAG_RDTUN,
285 &zone_offset, 0, "Separate malloc types by examining the "
286 "Nth character in the malloc type short description.");
289 mtp_get_subzone(const char *desc)
294 if (desc == NULL || (len = strlen(desc)) == 0)
296 val = desc[zone_offset % len];
297 return (val % numzones);
299 #elif MALLOC_DEBUG_MAXZONES == 0
300 #error "MALLOC_DEBUG_MAXZONES must be positive."
303 mtp_get_subzone(const char *desc)
308 #endif /* MALLOC_DEBUG_MAXZONES > 1 */
311 malloc_last_fail(void)
314 return (time_uptime - t_malloc_fail);
318 * An allocation has succeeded -- update malloc type statistics for the
319 * amount of bucket size. Occurs within a critical section so that the
320 * thread isn't preempted and doesn't migrate while updating per-PCU
324 malloc_type_zone_allocated(struct malloc_type *mtp, unsigned long size,
327 struct malloc_type_internal *mtip;
328 struct malloc_type_stats *mtsp;
331 mtip = mtp->ks_handle;
332 mtsp = &mtip->mti_stats[curcpu];
334 mtsp->mts_memalloced += size;
335 mtsp->mts_numallocs++;
338 mtsp->mts_size |= 1 << zindx;
341 if (dtrace_malloc_probe != NULL) {
342 uint32_t probe_id = mtip->mti_probes[DTMALLOC_PROBE_MALLOC];
344 (dtrace_malloc_probe)(probe_id,
345 (uintptr_t) mtp, (uintptr_t) mtip,
346 (uintptr_t) mtsp, size, zindx);
354 malloc_type_allocated(struct malloc_type *mtp, unsigned long size)
358 malloc_type_zone_allocated(mtp, size, -1);
362 * A free operation has occurred -- update malloc type statistics for the
363 * amount of the bucket size. Occurs within a critical section so that the
364 * thread isn't preempted and doesn't migrate while updating per-CPU
368 malloc_type_freed(struct malloc_type *mtp, unsigned long size)
370 struct malloc_type_internal *mtip;
371 struct malloc_type_stats *mtsp;
374 mtip = mtp->ks_handle;
375 mtsp = &mtip->mti_stats[curcpu];
376 mtsp->mts_memfreed += size;
377 mtsp->mts_numfrees++;
380 if (dtrace_malloc_probe != NULL) {
381 uint32_t probe_id = mtip->mti_probes[DTMALLOC_PROBE_FREE];
383 (dtrace_malloc_probe)(probe_id,
384 (uintptr_t) mtp, (uintptr_t) mtip,
385 (uintptr_t) mtsp, size, 0);
395 * Allocate a block of physically contiguous memory.
397 * If M_NOWAIT is set, this routine will not block and return NULL if
398 * the allocation fails.
401 contigmalloc(unsigned long size, struct malloc_type *type, int flags,
402 vm_paddr_t low, vm_paddr_t high, unsigned long alignment,
407 ret = (void *)kmem_alloc_contig(kernel_arena, size, flags, low, high,
408 alignment, boundary, VM_MEMATTR_DEFAULT);
410 malloc_type_allocated(type, round_page(size));
417 * Free a block of memory allocated by contigmalloc.
419 * This routine may not block.
422 contigfree(void *addr, unsigned long size, struct malloc_type *type)
425 kmem_free(kernel_arena, (vm_offset_t)addr, size);
426 malloc_type_freed(type, round_page(size));
432 * Allocate a block of memory.
434 * If M_NOWAIT is set, this routine will not block and return NULL if
435 * the allocation fails.
438 malloc(unsigned long size, struct malloc_type *mtp, int flags)
441 struct malloc_type_internal *mtip;
444 #if defined(DIAGNOSTIC) || defined(DEBUG_REDZONE)
445 unsigned long osize = size;
449 KASSERT(mtp->ks_magic == M_MAGIC, ("malloc: bad malloc type magic"));
451 * Check that exactly one of M_WAITOK or M_NOWAIT is specified.
453 indx = flags & (M_WAITOK | M_NOWAIT);
454 if (indx != M_NOWAIT && indx != M_WAITOK) {
455 static struct timeval lasterr;
456 static int curerr, once;
457 if (once == 0 && ppsratecheck(&lasterr, &curerr, 1)) {
458 printf("Bad malloc flags: %x\n", indx);
465 #ifdef MALLOC_MAKE_FAILURES
466 if ((flags & M_NOWAIT) && (malloc_failure_rate != 0)) {
467 atomic_add_int(&malloc_nowait_count, 1);
468 if ((malloc_nowait_count % malloc_failure_rate) == 0) {
469 atomic_add_int(&malloc_failure_count, 1);
470 t_malloc_fail = time_uptime;
475 if (flags & M_WAITOK)
476 KASSERT(curthread->td_intr_nesting_level == 0,
477 ("malloc(M_WAITOK) in interrupt context"));
479 #ifdef DEBUG_MEMGUARD
480 if (memguard_cmp_mtp(mtp, size)) {
481 va = memguard_alloc(size, flags);
484 /* This is unfortunate but should not be fatal. */
489 size = redzone_size_ntor(size);
492 if (size <= kmem_zmax) {
493 mtip = mtp->ks_handle;
494 if (size & KMEM_ZMASK)
495 size = (size & ~KMEM_ZMASK) + KMEM_ZBASE;
496 indx = kmemsize[size >> KMEM_ZSHIFT];
497 KASSERT(mtip->mti_zone < numzones,
498 ("mti_zone %u out of range %d",
499 mtip->mti_zone, numzones));
500 zone = kmemzones[indx].kz_zone[mtip->mti_zone];
501 #ifdef MALLOC_PROFILE
502 krequests[size >> KMEM_ZSHIFT]++;
504 va = uma_zalloc(zone, flags);
506 size = zone->uz_size;
507 malloc_type_zone_allocated(mtp, va == NULL ? 0 : size, indx);
509 size = roundup(size, PAGE_SIZE);
511 va = uma_large_malloc(size, flags);
512 malloc_type_allocated(mtp, va == NULL ? 0 : size);
514 if (flags & M_WAITOK)
515 KASSERT(va != NULL, ("malloc(M_WAITOK) returned NULL"));
517 t_malloc_fail = time_uptime;
519 if (va != NULL && !(flags & M_ZERO)) {
520 memset(va, 0x70, osize);
525 va = redzone_setup(va, osize);
527 return ((void *) va);
533 * Free a block of memory allocated by malloc.
535 * This routine may not block.
538 free(void *addr, struct malloc_type *mtp)
543 KASSERT(mtp->ks_magic == M_MAGIC, ("free: bad malloc type magic"));
545 /* free(NULL, ...) does nothing */
549 #ifdef DEBUG_MEMGUARD
550 if (is_memguard_addr(addr)) {
558 addr = redzone_addr_ntor(addr);
561 slab = vtoslab((vm_offset_t)addr & (~UMA_SLAB_MASK));
564 panic("free: address %p(%p) has not been allocated.\n",
565 addr, (void *)((u_long)addr & (~UMA_SLAB_MASK)));
567 if (!(slab->us_flags & UMA_SLAB_MALLOC)) {
569 struct malloc_type **mtpp = addr;
571 size = slab->us_keg->uk_size;
574 * Cache a pointer to the malloc_type that most recently freed
575 * this memory here. This way we know who is most likely to
576 * have stepped on it later.
578 * This code assumes that size is a multiple of 8 bytes for
581 mtpp = (struct malloc_type **)
582 ((unsigned long)mtpp & ~UMA_ALIGN_PTR);
583 mtpp += (size - sizeof(struct malloc_type *)) /
584 sizeof(struct malloc_type *);
587 uma_zfree_arg(LIST_FIRST(&slab->us_keg->uk_zones), addr, slab);
589 size = slab->us_size;
590 uma_large_free(slab);
592 malloc_type_freed(mtp, size);
596 * realloc: change the size of a memory block
599 realloc(void *addr, unsigned long size, struct malloc_type *mtp, int flags)
605 KASSERT(mtp->ks_magic == M_MAGIC,
606 ("realloc: bad malloc type magic"));
608 /* realloc(NULL, ...) is equivalent to malloc(...) */
610 return (malloc(size, mtp, flags));
613 * XXX: Should report free of old memory and alloc of new memory to
617 #ifdef DEBUG_MEMGUARD
618 if (is_memguard_addr(addr))
619 return (memguard_realloc(addr, size, mtp, flags));
624 alloc = redzone_get_size(addr);
626 slab = vtoslab((vm_offset_t)addr & ~(UMA_SLAB_MASK));
629 KASSERT(slab != NULL,
630 ("realloc: address %p out of range", (void *)addr));
632 /* Get the size of the original block */
633 if (!(slab->us_flags & UMA_SLAB_MALLOC))
634 alloc = slab->us_keg->uk_size;
636 alloc = slab->us_size;
638 /* Reuse the original block if appropriate */
640 && (size > (alloc >> REALLOC_FRACTION) || alloc == MINALLOCSIZE))
642 #endif /* !DEBUG_REDZONE */
644 /* Allocate a new, bigger (or smaller) block */
645 if ((newaddr = malloc(size, mtp, flags)) == NULL)
648 /* Copy over original contents */
649 bcopy(addr, newaddr, min(size, alloc));
655 * reallocf: same as realloc() but free memory on failure.
658 reallocf(void *addr, unsigned long size, struct malloc_type *mtp, int flags)
662 if ((mem = realloc(addr, size, mtp, flags)) == NULL)
668 * Wake the uma reclamation pagedaemon thread when we exhaust KVA. It
669 * will call the lowmem handler and uma_reclaim() callbacks in a
670 * context that is safe.
673 kmem_reclaim(vmem_t *vm, int flags)
676 uma_reclaim_wakeup();
681 CTASSERT(VM_KMEM_SIZE_SCALE >= 1);
685 * Initialize the kernel memory (kmem) arena.
694 if (vm_kmem_size == 0)
695 vm_kmem_size = VM_KMEM_SIZE;
697 #ifdef VM_KMEM_SIZE_MIN
698 if (vm_kmem_size_min == 0)
699 vm_kmem_size_min = VM_KMEM_SIZE_MIN;
701 #ifdef VM_KMEM_SIZE_MAX
702 if (vm_kmem_size_max == 0)
703 vm_kmem_size_max = VM_KMEM_SIZE_MAX;
706 * Calculate the amount of kernel virtual address (KVA) space that is
707 * preallocated to the kmem arena. In order to support a wide range
708 * of machines, it is a function of the physical memory size,
711 * min(max(physical memory size / VM_KMEM_SIZE_SCALE,
712 * VM_KMEM_SIZE_MIN), VM_KMEM_SIZE_MAX)
714 * Every architecture must define an integral value for
715 * VM_KMEM_SIZE_SCALE. However, the definitions of VM_KMEM_SIZE_MIN
716 * and VM_KMEM_SIZE_MAX, which represent respectively the floor and
717 * ceiling on this preallocation, are optional. Typically,
718 * VM_KMEM_SIZE_MAX is itself a function of the available KVA space on
719 * a given architecture.
721 mem_size = vm_cnt.v_page_count;
722 if (mem_size <= 32768) /* delphij XXX 128MB */
723 kmem_zmax = PAGE_SIZE;
725 if (vm_kmem_size_scale < 1)
726 vm_kmem_size_scale = VM_KMEM_SIZE_SCALE;
729 * Check if we should use defaults for the "vm_kmem_size"
732 if (vm_kmem_size == 0) {
733 vm_kmem_size = (mem_size / vm_kmem_size_scale) * PAGE_SIZE;
735 if (vm_kmem_size_min > 0 && vm_kmem_size < vm_kmem_size_min)
736 vm_kmem_size = vm_kmem_size_min;
737 if (vm_kmem_size_max > 0 && vm_kmem_size >= vm_kmem_size_max)
738 vm_kmem_size = vm_kmem_size_max;
742 * The amount of KVA space that is preallocated to the
743 * kmem arena can be set statically at compile-time or manually
744 * through the kernel environment. However, it is still limited to
745 * twice the physical memory size, which has been sufficient to handle
746 * the most severe cases of external fragmentation in the kmem arena.
748 if (vm_kmem_size / 2 / PAGE_SIZE > mem_size)
749 vm_kmem_size = 2 * mem_size * PAGE_SIZE;
751 vm_kmem_size = round_page(vm_kmem_size);
752 #ifdef DEBUG_MEMGUARD
753 tmp = memguard_fudge(vm_kmem_size, kernel_map);
757 vmem_init(kmem_arena, "kmem arena", kva_alloc(tmp), tmp, PAGE_SIZE,
759 vmem_set_reclaim(kmem_arena, kmem_reclaim);
761 #ifdef DEBUG_MEMGUARD
763 * Initialize MemGuard if support compiled in. MemGuard is a
764 * replacement allocator used for detecting tamper-after-free
765 * scenarios as they occur. It is only used for debugging.
767 memguard_init(kmem_arena);
772 * Initialize the kernel memory allocator
776 mallocinit(void *dummy)
781 mtx_init(&malloc_mtx, "malloc", NULL, MTX_DEF);
787 if (kmem_zmax < PAGE_SIZE || kmem_zmax > KMEM_ZMAX)
788 kmem_zmax = KMEM_ZMAX;
790 mt_zone = uma_zcreate("mt_zone", sizeof(struct malloc_type_internal),
792 mtrash_ctor, mtrash_dtor, mtrash_init, mtrash_fini,
794 NULL, NULL, NULL, NULL,
796 UMA_ALIGN_PTR, UMA_ZONE_MALLOC);
797 for (i = 0, indx = 0; kmemzones[indx].kz_size != 0; indx++) {
798 int size = kmemzones[indx].kz_size;
799 char *name = kmemzones[indx].kz_name;
802 for (subzone = 0; subzone < numzones; subzone++) {
803 kmemzones[indx].kz_zone[subzone] =
804 uma_zcreate(name, size,
806 mtrash_ctor, mtrash_dtor, mtrash_init, mtrash_fini,
808 NULL, NULL, NULL, NULL,
810 UMA_ALIGN_PTR, UMA_ZONE_MALLOC);
812 for (;i <= size; i+= KMEM_ZBASE)
813 kmemsize[i >> KMEM_ZSHIFT] = indx;
817 SYSINIT(kmem, SI_SUB_KMEM, SI_ORDER_SECOND, mallocinit, NULL);
820 malloc_init(void *data)
822 struct malloc_type_internal *mtip;
823 struct malloc_type *mtp;
825 KASSERT(vm_cnt.v_page_count != 0, ("malloc_register before vm_init"));
828 if (mtp->ks_magic != M_MAGIC)
829 panic("malloc_init: bad malloc type magic");
831 mtip = uma_zalloc(mt_zone, M_WAITOK | M_ZERO);
832 mtp->ks_handle = mtip;
833 mtip->mti_zone = mtp_get_subzone(mtp->ks_shortdesc);
835 mtx_lock(&malloc_mtx);
836 mtp->ks_next = kmemstatistics;
837 kmemstatistics = mtp;
839 mtx_unlock(&malloc_mtx);
843 malloc_uninit(void *data)
845 struct malloc_type_internal *mtip;
846 struct malloc_type_stats *mtsp;
847 struct malloc_type *mtp, *temp;
849 long temp_allocs, temp_bytes;
853 KASSERT(mtp->ks_magic == M_MAGIC,
854 ("malloc_uninit: bad malloc type magic"));
855 KASSERT(mtp->ks_handle != NULL, ("malloc_deregister: cookie NULL"));
857 mtx_lock(&malloc_mtx);
858 mtip = mtp->ks_handle;
859 mtp->ks_handle = NULL;
860 if (mtp != kmemstatistics) {
861 for (temp = kmemstatistics; temp != NULL;
862 temp = temp->ks_next) {
863 if (temp->ks_next == mtp) {
864 temp->ks_next = mtp->ks_next;
869 ("malloc_uninit: type '%s' not found", mtp->ks_shortdesc));
871 kmemstatistics = mtp->ks_next;
873 mtx_unlock(&malloc_mtx);
876 * Look for memory leaks.
878 temp_allocs = temp_bytes = 0;
879 for (i = 0; i < MAXCPU; i++) {
880 mtsp = &mtip->mti_stats[i];
881 temp_allocs += mtsp->mts_numallocs;
882 temp_allocs -= mtsp->mts_numfrees;
883 temp_bytes += mtsp->mts_memalloced;
884 temp_bytes -= mtsp->mts_memfreed;
886 if (temp_allocs > 0 || temp_bytes > 0) {
887 printf("Warning: memory type %s leaked memory on destroy "
888 "(%ld allocations, %ld bytes leaked).\n", mtp->ks_shortdesc,
889 temp_allocs, temp_bytes);
892 slab = vtoslab((vm_offset_t) mtip & (~UMA_SLAB_MASK));
893 uma_zfree_arg(mt_zone, mtip, slab);
897 malloc_desc2type(const char *desc)
899 struct malloc_type *mtp;
901 mtx_assert(&malloc_mtx, MA_OWNED);
902 for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) {
903 if (strcmp(mtp->ks_shortdesc, desc) == 0)
910 sysctl_kern_malloc_stats(SYSCTL_HANDLER_ARGS)
912 struct malloc_type_stream_header mtsh;
913 struct malloc_type_internal *mtip;
914 struct malloc_type_header mth;
915 struct malloc_type *mtp;
919 error = sysctl_wire_old_buffer(req, 0);
922 sbuf_new_for_sysctl(&sbuf, NULL, 128, req);
923 sbuf_clear_flags(&sbuf, SBUF_INCLUDENUL);
924 mtx_lock(&malloc_mtx);
927 * Insert stream header.
929 bzero(&mtsh, sizeof(mtsh));
930 mtsh.mtsh_version = MALLOC_TYPE_STREAM_VERSION;
931 mtsh.mtsh_maxcpus = MAXCPU;
932 mtsh.mtsh_count = kmemcount;
933 (void)sbuf_bcat(&sbuf, &mtsh, sizeof(mtsh));
936 * Insert alternating sequence of type headers and type statistics.
938 for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) {
939 mtip = (struct malloc_type_internal *)mtp->ks_handle;
942 * Insert type header.
944 bzero(&mth, sizeof(mth));
945 strlcpy(mth.mth_name, mtp->ks_shortdesc, MALLOC_MAX_NAME);
946 (void)sbuf_bcat(&sbuf, &mth, sizeof(mth));
949 * Insert type statistics for each CPU.
951 for (i = 0; i < MAXCPU; i++) {
952 (void)sbuf_bcat(&sbuf, &mtip->mti_stats[i],
953 sizeof(mtip->mti_stats[i]));
956 mtx_unlock(&malloc_mtx);
957 error = sbuf_finish(&sbuf);
962 SYSCTL_PROC(_kern, OID_AUTO, malloc_stats, CTLFLAG_RD|CTLTYPE_STRUCT,
963 0, 0, sysctl_kern_malloc_stats, "s,malloc_type_ustats",
964 "Return malloc types");
966 SYSCTL_INT(_kern, OID_AUTO, malloc_count, CTLFLAG_RD, &kmemcount, 0,
967 "Count of kernel malloc types");
970 malloc_type_list(malloc_type_list_func_t *func, void *arg)
972 struct malloc_type *mtp, **bufmtp;
976 mtx_lock(&malloc_mtx);
978 mtx_assert(&malloc_mtx, MA_OWNED);
980 mtx_unlock(&malloc_mtx);
982 buflen = sizeof(struct malloc_type *) * count;
983 bufmtp = malloc(buflen, M_TEMP, M_WAITOK);
985 mtx_lock(&malloc_mtx);
987 if (count < kmemcount) {
988 free(bufmtp, M_TEMP);
992 for (mtp = kmemstatistics, i = 0; mtp != NULL; mtp = mtp->ks_next, i++)
995 mtx_unlock(&malloc_mtx);
997 for (i = 0; i < count; i++)
998 (func)(bufmtp[i], arg);
1000 free(bufmtp, M_TEMP);
1004 DB_SHOW_COMMAND(malloc, db_show_malloc)
1006 struct malloc_type_internal *mtip;
1007 struct malloc_type *mtp;
1008 uint64_t allocs, frees;
1009 uint64_t alloced, freed;
1012 db_printf("%18s %12s %12s %12s\n", "Type", "InUse", "MemUse",
1014 for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) {
1015 mtip = (struct malloc_type_internal *)mtp->ks_handle;
1020 for (i = 0; i < MAXCPU; i++) {
1021 allocs += mtip->mti_stats[i].mts_numallocs;
1022 frees += mtip->mti_stats[i].mts_numfrees;
1023 alloced += mtip->mti_stats[i].mts_memalloced;
1024 freed += mtip->mti_stats[i].mts_memfreed;
1026 db_printf("%18s %12ju %12juK %12ju\n",
1027 mtp->ks_shortdesc, allocs - frees,
1028 (alloced - freed + 1023) / 1024, allocs);
1034 #if MALLOC_DEBUG_MAXZONES > 1
1035 DB_SHOW_COMMAND(multizone_matches, db_show_multizone_matches)
1037 struct malloc_type_internal *mtip;
1038 struct malloc_type *mtp;
1042 db_printf("Usage: show multizone_matches <malloc type/addr>\n");
1046 if (mtp->ks_magic != M_MAGIC) {
1047 db_printf("Magic %lx does not match expected %x\n",
1048 mtp->ks_magic, M_MAGIC);
1052 mtip = mtp->ks_handle;
1053 subzone = mtip->mti_zone;
1055 for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) {
1056 mtip = mtp->ks_handle;
1057 if (mtip->mti_zone != subzone)
1059 db_printf("%s\n", mtp->ks_shortdesc);
1064 #endif /* MALLOC_DEBUG_MAXZONES > 1 */
1067 #ifdef MALLOC_PROFILE
1070 sysctl_kern_mprof(SYSCTL_HANDLER_ARGS)
1084 error = sysctl_wire_old_buffer(req, 0);
1087 sbuf_new_for_sysctl(&sbuf, NULL, 128, req);
1089 "\n Size Requests Real Size\n");
1090 for (i = 0; i < KMEM_ZSIZE; i++) {
1091 size = i << KMEM_ZSHIFT;
1092 rsize = kmemzones[kmemsize[i]].kz_size;
1093 count = (long long unsigned)krequests[i];
1095 sbuf_printf(&sbuf, "%6d%28llu%11d\n", size,
1096 (unsigned long long)count, rsize);
1098 if ((rsize * count) > (size * count))
1099 waste += (rsize * count) - (size * count);
1100 mem += (rsize * count);
1103 "\nTotal memory used:\t%30llu\nTotal Memory wasted:\t%30llu\n",
1104 (unsigned long long)mem, (unsigned long long)waste);
1105 error = sbuf_finish(&sbuf);
1110 SYSCTL_OID(_kern, OID_AUTO, mprof, CTLTYPE_STRING|CTLFLAG_RD,
1111 NULL, 0, sysctl_kern_mprof, "A", "Malloc Profiling");
1112 #endif /* MALLOC_PROFILE */