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 PAGE_SIZE
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];
158 #if PAGE_SIZE > 16384
160 #if PAGE_SIZE > 32768
162 #if PAGE_SIZE > 65536
163 #error "Unsupported PAGE_SIZE"
173 * Zone to allocate malloc type descriptions from. For ABI reasons, memory
174 * types are described by a data structure passed by the declaring code, but
175 * the malloc(9) implementation has its own data structure describing the
176 * type and statistics. This permits the malloc(9)-internal data structures
177 * to be modified without breaking binary-compiled kernel modules that
178 * declare malloc types.
180 static uma_zone_t mt_zone;
183 SYSCTL_ULONG(_vm, OID_AUTO, kmem_size, CTLFLAG_RDTUN, &vm_kmem_size, 0,
184 "Size of kernel memory");
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");
209 * The malloc_mtx protects the kmemstatistics linked list.
211 struct mtx malloc_mtx;
213 #ifdef MALLOC_PROFILE
214 uint64_t krequests[KMEM_ZSIZE + 1];
216 static int sysctl_kern_mprof(SYSCTL_HANDLER_ARGS);
219 static int sysctl_kern_malloc_stats(SYSCTL_HANDLER_ARGS);
222 * time_uptime of the last malloc(9) failure (induced or real).
224 static time_t t_malloc_fail;
226 #if defined(MALLOC_MAKE_FAILURES) || (MALLOC_DEBUG_MAXZONES > 1)
227 static SYSCTL_NODE(_debug, OID_AUTO, malloc, CTLFLAG_RD, 0,
228 "Kernel malloc debugging options");
232 * malloc(9) fault injection -- cause malloc failures every (n) mallocs when
233 * the caller specifies M_NOWAIT. If set to 0, no failures are caused.
235 #ifdef MALLOC_MAKE_FAILURES
236 static int malloc_failure_rate;
237 static int malloc_nowait_count;
238 static int malloc_failure_count;
239 SYSCTL_INT(_debug_malloc, OID_AUTO, failure_rate, CTLFLAG_RW,
240 &malloc_failure_rate, 0, "Every (n) mallocs with M_NOWAIT will fail");
241 TUNABLE_INT("debug.malloc.failure_rate", &malloc_failure_rate);
242 SYSCTL_INT(_debug_malloc, OID_AUTO, failure_count, CTLFLAG_RD,
243 &malloc_failure_count, 0, "Number of imposed M_NOWAIT malloc failures");
247 sysctl_kmem_map_size(SYSCTL_HANDLER_ARGS)
251 size = vmem_size(kmem_arena, VMEM_ALLOC);
252 return (sysctl_handle_long(oidp, &size, 0, req));
256 sysctl_kmem_map_free(SYSCTL_HANDLER_ARGS)
260 size = vmem_size(kmem_arena, VMEM_FREE);
261 return (sysctl_handle_long(oidp, &size, 0, req));
265 * malloc(9) uma zone separation -- sub-page buffer overruns in one
266 * malloc type will affect only a subset of other malloc types.
268 #if MALLOC_DEBUG_MAXZONES > 1
270 tunable_set_numzones(void)
273 TUNABLE_INT_FETCH("debug.malloc.numzones",
276 /* Sanity check the number of malloc uma zones. */
279 if (numzones > MALLOC_DEBUG_MAXZONES)
280 numzones = MALLOC_DEBUG_MAXZONES;
282 SYSINIT(numzones, SI_SUB_TUNABLES, SI_ORDER_ANY, tunable_set_numzones, NULL);
283 SYSCTL_INT(_debug_malloc, OID_AUTO, numzones, CTLFLAG_RDTUN,
284 &numzones, 0, "Number of malloc uma subzones");
287 * Any number that changes regularly is an okay choice for the
288 * offset. Build numbers are pretty good of you have them.
290 static u_int zone_offset = __FreeBSD_version;
291 TUNABLE_INT("debug.malloc.zone_offset", &zone_offset);
292 SYSCTL_UINT(_debug_malloc, OID_AUTO, zone_offset, CTLFLAG_RDTUN,
293 &zone_offset, 0, "Separate malloc types by examining the "
294 "Nth character in the malloc type short description.");
297 mtp_get_subzone(const char *desc)
302 if (desc == NULL || (len = strlen(desc)) == 0)
304 val = desc[zone_offset % len];
305 return (val % numzones);
307 #elif MALLOC_DEBUG_MAXZONES == 0
308 #error "MALLOC_DEBUG_MAXZONES must be positive."
311 mtp_get_subzone(const char *desc)
316 #endif /* MALLOC_DEBUG_MAXZONES > 1 */
319 malloc_last_fail(void)
322 return (time_uptime - t_malloc_fail);
326 * An allocation has succeeded -- update malloc type statistics for the
327 * amount of bucket size. Occurs within a critical section so that the
328 * thread isn't preempted and doesn't migrate while updating per-PCU
332 malloc_type_zone_allocated(struct malloc_type *mtp, unsigned long size,
335 struct malloc_type_internal *mtip;
336 struct malloc_type_stats *mtsp;
339 mtip = mtp->ks_handle;
340 mtsp = &mtip->mti_stats[curcpu];
342 mtsp->mts_memalloced += size;
343 mtsp->mts_numallocs++;
346 mtsp->mts_size |= 1 << zindx;
349 if (dtrace_malloc_probe != NULL) {
350 uint32_t probe_id = mtip->mti_probes[DTMALLOC_PROBE_MALLOC];
352 (dtrace_malloc_probe)(probe_id,
353 (uintptr_t) mtp, (uintptr_t) mtip,
354 (uintptr_t) mtsp, size, zindx);
362 malloc_type_allocated(struct malloc_type *mtp, unsigned long size)
366 malloc_type_zone_allocated(mtp, size, -1);
370 * A free operation has occurred -- update malloc type statistics for the
371 * amount of the bucket size. Occurs within a critical section so that the
372 * thread isn't preempted and doesn't migrate while updating per-CPU
376 malloc_type_freed(struct malloc_type *mtp, unsigned long size)
378 struct malloc_type_internal *mtip;
379 struct malloc_type_stats *mtsp;
382 mtip = mtp->ks_handle;
383 mtsp = &mtip->mti_stats[curcpu];
384 mtsp->mts_memfreed += size;
385 mtsp->mts_numfrees++;
388 if (dtrace_malloc_probe != NULL) {
389 uint32_t probe_id = mtip->mti_probes[DTMALLOC_PROBE_FREE];
391 (dtrace_malloc_probe)(probe_id,
392 (uintptr_t) mtp, (uintptr_t) mtip,
393 (uintptr_t) mtsp, size, 0);
403 * Allocate a block of physically contiguous memory.
405 * If M_NOWAIT is set, this routine will not block and return NULL if
406 * the allocation fails.
409 contigmalloc(unsigned long size, struct malloc_type *type, int flags,
410 vm_paddr_t low, vm_paddr_t high, unsigned long alignment,
415 ret = (void *)kmem_alloc_contig(kernel_arena, size, flags, low, high,
416 alignment, boundary, VM_MEMATTR_DEFAULT);
418 malloc_type_allocated(type, round_page(size));
425 * Free a block of memory allocated by contigmalloc.
427 * This routine may not block.
430 contigfree(void *addr, unsigned long size, struct malloc_type *type)
433 kmem_free(kernel_arena, (vm_offset_t)addr, size);
434 malloc_type_freed(type, round_page(size));
440 * Allocate a block of memory.
442 * If M_NOWAIT is set, this routine will not block and return NULL if
443 * the allocation fails.
446 malloc(unsigned long size, struct malloc_type *mtp, int flags)
449 struct malloc_type_internal *mtip;
452 #if defined(DIAGNOSTIC) || defined(DEBUG_REDZONE)
453 unsigned long osize = size;
457 KASSERT(mtp->ks_magic == M_MAGIC, ("malloc: bad malloc type magic"));
459 * Check that exactly one of M_WAITOK or M_NOWAIT is specified.
461 indx = flags & (M_WAITOK | M_NOWAIT);
462 if (indx != M_NOWAIT && indx != M_WAITOK) {
463 static struct timeval lasterr;
464 static int curerr, once;
465 if (once == 0 && ppsratecheck(&lasterr, &curerr, 1)) {
466 printf("Bad malloc flags: %x\n", indx);
473 #ifdef MALLOC_MAKE_FAILURES
474 if ((flags & M_NOWAIT) && (malloc_failure_rate != 0)) {
475 atomic_add_int(&malloc_nowait_count, 1);
476 if ((malloc_nowait_count % malloc_failure_rate) == 0) {
477 atomic_add_int(&malloc_failure_count, 1);
478 t_malloc_fail = time_uptime;
483 if (flags & M_WAITOK)
484 KASSERT(curthread->td_intr_nesting_level == 0,
485 ("malloc(M_WAITOK) in interrupt context"));
487 #ifdef DEBUG_MEMGUARD
488 if (memguard_cmp_mtp(mtp, size)) {
489 va = memguard_alloc(size, flags);
492 /* This is unfortunate but should not be fatal. */
497 size = redzone_size_ntor(size);
500 if (size <= KMEM_ZMAX) {
501 mtip = mtp->ks_handle;
502 if (size & KMEM_ZMASK)
503 size = (size & ~KMEM_ZMASK) + KMEM_ZBASE;
504 indx = kmemsize[size >> KMEM_ZSHIFT];
505 KASSERT(mtip->mti_zone < numzones,
506 ("mti_zone %u out of range %d",
507 mtip->mti_zone, numzones));
508 zone = kmemzones[indx].kz_zone[mtip->mti_zone];
509 #ifdef MALLOC_PROFILE
510 krequests[size >> KMEM_ZSHIFT]++;
512 va = uma_zalloc(zone, flags);
514 size = zone->uz_size;
515 malloc_type_zone_allocated(mtp, va == NULL ? 0 : size, indx);
517 size = roundup(size, PAGE_SIZE);
519 va = uma_large_malloc(size, flags);
520 malloc_type_allocated(mtp, va == NULL ? 0 : size);
522 if (flags & M_WAITOK)
523 KASSERT(va != NULL, ("malloc(M_WAITOK) returned NULL"));
525 t_malloc_fail = time_uptime;
527 if (va != NULL && !(flags & M_ZERO)) {
528 memset(va, 0x70, osize);
533 va = redzone_setup(va, osize);
535 return ((void *) va);
541 * Free a block of memory allocated by malloc.
543 * This routine may not block.
546 free(void *addr, struct malloc_type *mtp)
551 KASSERT(mtp->ks_magic == M_MAGIC, ("free: bad malloc type magic"));
553 /* free(NULL, ...) does nothing */
557 #ifdef DEBUG_MEMGUARD
558 if (is_memguard_addr(addr)) {
566 addr = redzone_addr_ntor(addr);
569 slab = vtoslab((vm_offset_t)addr & (~UMA_SLAB_MASK));
572 panic("free: address %p(%p) has not been allocated.\n",
573 addr, (void *)((u_long)addr & (~UMA_SLAB_MASK)));
575 if (!(slab->us_flags & UMA_SLAB_MALLOC)) {
577 struct malloc_type **mtpp = addr;
579 size = slab->us_keg->uk_size;
582 * Cache a pointer to the malloc_type that most recently freed
583 * this memory here. This way we know who is most likely to
584 * have stepped on it later.
586 * This code assumes that size is a multiple of 8 bytes for
589 mtpp = (struct malloc_type **)
590 ((unsigned long)mtpp & ~UMA_ALIGN_PTR);
591 mtpp += (size - sizeof(struct malloc_type *)) /
592 sizeof(struct malloc_type *);
595 uma_zfree_arg(LIST_FIRST(&slab->us_keg->uk_zones), addr, slab);
597 size = slab->us_size;
598 uma_large_free(slab);
600 malloc_type_freed(mtp, size);
604 * realloc: change the size of a memory block
607 realloc(void *addr, unsigned long size, struct malloc_type *mtp, int flags)
613 KASSERT(mtp->ks_magic == M_MAGIC,
614 ("realloc: bad malloc type magic"));
616 /* realloc(NULL, ...) is equivalent to malloc(...) */
618 return (malloc(size, mtp, flags));
621 * XXX: Should report free of old memory and alloc of new memory to
625 #ifdef DEBUG_MEMGUARD
626 if (is_memguard_addr(addr))
627 return (memguard_realloc(addr, size, mtp, flags));
632 alloc = redzone_get_size(addr);
634 slab = vtoslab((vm_offset_t)addr & ~(UMA_SLAB_MASK));
637 KASSERT(slab != NULL,
638 ("realloc: address %p out of range", (void *)addr));
640 /* Get the size of the original block */
641 if (!(slab->us_flags & UMA_SLAB_MALLOC))
642 alloc = slab->us_keg->uk_size;
644 alloc = slab->us_size;
646 /* Reuse the original block if appropriate */
648 && (size > (alloc >> REALLOC_FRACTION) || alloc == MINALLOCSIZE))
650 #endif /* !DEBUG_REDZONE */
652 /* Allocate a new, bigger (or smaller) block */
653 if ((newaddr = malloc(size, mtp, flags)) == NULL)
656 /* Copy over original contents */
657 bcopy(addr, newaddr, min(size, alloc));
663 * reallocf: same as realloc() but free memory on failure.
666 reallocf(void *addr, unsigned long size, struct malloc_type *mtp, int flags)
670 if ((mem = realloc(addr, size, mtp, flags)) == NULL)
676 * Wake the page daemon when we exhaust KVA. It will call the lowmem handler
677 * and uma_reclaim() callbacks in a context that is safe.
680 kmem_reclaim(vmem_t *vm, int flags)
686 CTASSERT(VM_KMEM_SIZE_SCALE >= 1);
689 * Initialize the kernel memory (kmem) arena.
694 u_long mem_size, tmp;
697 * Calculate the amount of kernel virtual address (KVA) space that is
698 * preallocated to the kmem arena. In order to support a wide range
699 * of machines, it is a function of the physical memory size,
702 * min(max(physical memory size / VM_KMEM_SIZE_SCALE,
703 * VM_KMEM_SIZE_MIN), VM_KMEM_SIZE_MAX)
705 * Every architecture must define an integral value for
706 * VM_KMEM_SIZE_SCALE. However, the definitions of VM_KMEM_SIZE_MIN
707 * and VM_KMEM_SIZE_MAX, which represent respectively the floor and
708 * ceiling on this preallocation, are optional. Typically,
709 * VM_KMEM_SIZE_MAX is itself a function of the available KVA space on
710 * a given architecture.
712 mem_size = cnt.v_page_count;
714 vm_kmem_size_scale = VM_KMEM_SIZE_SCALE;
715 TUNABLE_INT_FETCH("vm.kmem_size_scale", &vm_kmem_size_scale);
716 if (vm_kmem_size_scale < 1)
717 vm_kmem_size_scale = VM_KMEM_SIZE_SCALE;
719 vm_kmem_size = (mem_size / vm_kmem_size_scale) * PAGE_SIZE;
721 #if defined(VM_KMEM_SIZE_MIN)
722 vm_kmem_size_min = VM_KMEM_SIZE_MIN;
724 TUNABLE_ULONG_FETCH("vm.kmem_size_min", &vm_kmem_size_min);
725 if (vm_kmem_size_min > 0 && vm_kmem_size < vm_kmem_size_min)
726 vm_kmem_size = vm_kmem_size_min;
728 #if defined(VM_KMEM_SIZE_MAX)
729 vm_kmem_size_max = VM_KMEM_SIZE_MAX;
731 TUNABLE_ULONG_FETCH("vm.kmem_size_max", &vm_kmem_size_max);
732 if (vm_kmem_size_max > 0 && vm_kmem_size >= vm_kmem_size_max)
733 vm_kmem_size = vm_kmem_size_max;
736 * Alternatively, the amount of KVA space that is preallocated to the
737 * kmem arena can be set statically at compile-time or manually
738 * through the kernel environment. However, it is still limited to
739 * twice the physical memory size, which has been sufficient to handle
740 * the most severe cases of external fragmentation in the kmem arena.
742 #if defined(VM_KMEM_SIZE)
743 vm_kmem_size = VM_KMEM_SIZE;
745 TUNABLE_ULONG_FETCH("vm.kmem_size", &vm_kmem_size);
746 if (vm_kmem_size / 2 / PAGE_SIZE > mem_size)
747 vm_kmem_size = 2 * mem_size * PAGE_SIZE;
749 vm_kmem_size = round_page(vm_kmem_size);
750 #ifdef DEBUG_MEMGUARD
751 tmp = memguard_fudge(vm_kmem_size, kernel_map);
755 vmem_init(kmem_arena, "kmem arena", kva_alloc(tmp), tmp, PAGE_SIZE,
757 vmem_set_reclaim(kmem_arena, kmem_reclaim);
759 #ifdef DEBUG_MEMGUARD
761 * Initialize MemGuard if support compiled in. MemGuard is a
762 * replacement allocator used for detecting tamper-after-free
763 * scenarios as they occur. It is only used for debugging.
765 memguard_init(kmem_arena);
770 * Initialize the kernel memory allocator
774 mallocinit(void *dummy)
779 mtx_init(&malloc_mtx, "malloc", NULL, MTX_DEF);
785 mt_zone = uma_zcreate("mt_zone", sizeof(struct malloc_type_internal),
787 mtrash_ctor, mtrash_dtor, mtrash_init, mtrash_fini,
789 NULL, NULL, NULL, NULL,
791 UMA_ALIGN_PTR, UMA_ZONE_MALLOC);
792 for (i = 0, indx = 0; kmemzones[indx].kz_size != 0; indx++) {
793 int size = kmemzones[indx].kz_size;
794 char *name = kmemzones[indx].kz_name;
797 for (subzone = 0; subzone < numzones; subzone++) {
798 kmemzones[indx].kz_zone[subzone] =
799 uma_zcreate(name, size,
801 mtrash_ctor, mtrash_dtor, mtrash_init, mtrash_fini,
803 NULL, NULL, NULL, NULL,
805 UMA_ALIGN_PTR, UMA_ZONE_MALLOC);
807 for (;i <= size; i+= KMEM_ZBASE)
808 kmemsize[i >> KMEM_ZSHIFT] = indx;
812 SYSINIT(kmem, SI_SUB_KMEM, SI_ORDER_FIRST, mallocinit, NULL);
815 malloc_init(void *data)
817 struct malloc_type_internal *mtip;
818 struct malloc_type *mtp;
820 KASSERT(cnt.v_page_count != 0, ("malloc_register before vm_init"));
823 if (mtp->ks_magic != M_MAGIC)
824 panic("malloc_init: bad malloc type magic");
826 mtip = uma_zalloc(mt_zone, M_WAITOK | M_ZERO);
827 mtp->ks_handle = mtip;
828 mtip->mti_zone = mtp_get_subzone(mtp->ks_shortdesc);
830 mtx_lock(&malloc_mtx);
831 mtp->ks_next = kmemstatistics;
832 kmemstatistics = mtp;
834 mtx_unlock(&malloc_mtx);
838 malloc_uninit(void *data)
840 struct malloc_type_internal *mtip;
841 struct malloc_type_stats *mtsp;
842 struct malloc_type *mtp, *temp;
844 long temp_allocs, temp_bytes;
848 KASSERT(mtp->ks_magic == M_MAGIC,
849 ("malloc_uninit: bad malloc type magic"));
850 KASSERT(mtp->ks_handle != NULL, ("malloc_deregister: cookie NULL"));
852 mtx_lock(&malloc_mtx);
853 mtip = mtp->ks_handle;
854 mtp->ks_handle = NULL;
855 if (mtp != kmemstatistics) {
856 for (temp = kmemstatistics; temp != NULL;
857 temp = temp->ks_next) {
858 if (temp->ks_next == mtp) {
859 temp->ks_next = mtp->ks_next;
864 ("malloc_uninit: type '%s' not found", mtp->ks_shortdesc));
866 kmemstatistics = mtp->ks_next;
868 mtx_unlock(&malloc_mtx);
871 * Look for memory leaks.
873 temp_allocs = temp_bytes = 0;
874 for (i = 0; i < MAXCPU; i++) {
875 mtsp = &mtip->mti_stats[i];
876 temp_allocs += mtsp->mts_numallocs;
877 temp_allocs -= mtsp->mts_numfrees;
878 temp_bytes += mtsp->mts_memalloced;
879 temp_bytes -= mtsp->mts_memfreed;
881 if (temp_allocs > 0 || temp_bytes > 0) {
882 printf("Warning: memory type %s leaked memory on destroy "
883 "(%ld allocations, %ld bytes leaked).\n", mtp->ks_shortdesc,
884 temp_allocs, temp_bytes);
887 slab = vtoslab((vm_offset_t) mtip & (~UMA_SLAB_MASK));
888 uma_zfree_arg(mt_zone, mtip, slab);
892 malloc_desc2type(const char *desc)
894 struct malloc_type *mtp;
896 mtx_assert(&malloc_mtx, MA_OWNED);
897 for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) {
898 if (strcmp(mtp->ks_shortdesc, desc) == 0)
905 sysctl_kern_malloc_stats(SYSCTL_HANDLER_ARGS)
907 struct malloc_type_stream_header mtsh;
908 struct malloc_type_internal *mtip;
909 struct malloc_type_header mth;
910 struct malloc_type *mtp;
914 error = sysctl_wire_old_buffer(req, 0);
917 sbuf_new_for_sysctl(&sbuf, NULL, 128, req);
918 mtx_lock(&malloc_mtx);
921 * Insert stream header.
923 bzero(&mtsh, sizeof(mtsh));
924 mtsh.mtsh_version = MALLOC_TYPE_STREAM_VERSION;
925 mtsh.mtsh_maxcpus = MAXCPU;
926 mtsh.mtsh_count = kmemcount;
927 (void)sbuf_bcat(&sbuf, &mtsh, sizeof(mtsh));
930 * Insert alternating sequence of type headers and type statistics.
932 for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) {
933 mtip = (struct malloc_type_internal *)mtp->ks_handle;
936 * Insert type header.
938 bzero(&mth, sizeof(mth));
939 strlcpy(mth.mth_name, mtp->ks_shortdesc, MALLOC_MAX_NAME);
940 (void)sbuf_bcat(&sbuf, &mth, sizeof(mth));
943 * Insert type statistics for each CPU.
945 for (i = 0; i < MAXCPU; i++) {
946 (void)sbuf_bcat(&sbuf, &mtip->mti_stats[i],
947 sizeof(mtip->mti_stats[i]));
950 mtx_unlock(&malloc_mtx);
951 error = sbuf_finish(&sbuf);
956 SYSCTL_PROC(_kern, OID_AUTO, malloc_stats, CTLFLAG_RD|CTLTYPE_STRUCT,
957 0, 0, sysctl_kern_malloc_stats, "s,malloc_type_ustats",
958 "Return malloc types");
960 SYSCTL_INT(_kern, OID_AUTO, malloc_count, CTLFLAG_RD, &kmemcount, 0,
961 "Count of kernel malloc types");
964 malloc_type_list(malloc_type_list_func_t *func, void *arg)
966 struct malloc_type *mtp, **bufmtp;
970 mtx_lock(&malloc_mtx);
972 mtx_assert(&malloc_mtx, MA_OWNED);
974 mtx_unlock(&malloc_mtx);
976 buflen = sizeof(struct malloc_type *) * count;
977 bufmtp = malloc(buflen, M_TEMP, M_WAITOK);
979 mtx_lock(&malloc_mtx);
981 if (count < kmemcount) {
982 free(bufmtp, M_TEMP);
986 for (mtp = kmemstatistics, i = 0; mtp != NULL; mtp = mtp->ks_next, i++)
989 mtx_unlock(&malloc_mtx);
991 for (i = 0; i < count; i++)
992 (func)(bufmtp[i], arg);
994 free(bufmtp, M_TEMP);
998 DB_SHOW_COMMAND(malloc, db_show_malloc)
1000 struct malloc_type_internal *mtip;
1001 struct malloc_type *mtp;
1002 uint64_t allocs, frees;
1003 uint64_t alloced, freed;
1006 db_printf("%18s %12s %12s %12s\n", "Type", "InUse", "MemUse",
1008 for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) {
1009 mtip = (struct malloc_type_internal *)mtp->ks_handle;
1014 for (i = 0; i < MAXCPU; i++) {
1015 allocs += mtip->mti_stats[i].mts_numallocs;
1016 frees += mtip->mti_stats[i].mts_numfrees;
1017 alloced += mtip->mti_stats[i].mts_memalloced;
1018 freed += mtip->mti_stats[i].mts_memfreed;
1020 db_printf("%18s %12ju %12juK %12ju\n",
1021 mtp->ks_shortdesc, allocs - frees,
1022 (alloced - freed + 1023) / 1024, allocs);
1028 #if MALLOC_DEBUG_MAXZONES > 1
1029 DB_SHOW_COMMAND(multizone_matches, db_show_multizone_matches)
1031 struct malloc_type_internal *mtip;
1032 struct malloc_type *mtp;
1036 db_printf("Usage: show multizone_matches <malloc type/addr>\n");
1040 if (mtp->ks_magic != M_MAGIC) {
1041 db_printf("Magic %lx does not match expected %x\n",
1042 mtp->ks_magic, M_MAGIC);
1046 mtip = mtp->ks_handle;
1047 subzone = mtip->mti_zone;
1049 for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) {
1050 mtip = mtp->ks_handle;
1051 if (mtip->mti_zone != subzone)
1053 db_printf("%s\n", mtp->ks_shortdesc);
1058 #endif /* MALLOC_DEBUG_MAXZONES > 1 */
1061 #ifdef MALLOC_PROFILE
1064 sysctl_kern_mprof(SYSCTL_HANDLER_ARGS)
1078 error = sysctl_wire_old_buffer(req, 0);
1081 sbuf_new_for_sysctl(&sbuf, NULL, 128, req);
1083 "\n Size Requests Real Size\n");
1084 for (i = 0; i < KMEM_ZSIZE; i++) {
1085 size = i << KMEM_ZSHIFT;
1086 rsize = kmemzones[kmemsize[i]].kz_size;
1087 count = (long long unsigned)krequests[i];
1089 sbuf_printf(&sbuf, "%6d%28llu%11d\n", size,
1090 (unsigned long long)count, rsize);
1092 if ((rsize * count) > (size * count))
1093 waste += (rsize * count) - (size * count);
1094 mem += (rsize * count);
1097 "\nTotal memory used:\t%30llu\nTotal Memory wasted:\t%30llu\n",
1098 (unsigned long long)mem, (unsigned long long)waste);
1099 error = sbuf_finish(&sbuf);
1104 SYSCTL_OID(_kern, OID_AUTO, mprof, CTLTYPE_STRING|CTLFLAG_RD,
1105 NULL, 0, sysctl_kern_mprof, "A", "Malloc Profiling");
1106 #endif /* MALLOC_PROFILE */