1 /*-
2 * Copyright (c) 1987, 1991, 1993
3 * The Regents of the University of California.
4 * Copyright (c) 2005-2009 Robert N. M. Watson
5 * All rights reserved.
6 *
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
9 * are met:
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.
18 *
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
29 * SUCH DAMAGE.
30 *
31 * @(#)kern_malloc.c 8.3 (Berkeley) 1/4/94
32 */
33
34 /*
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
42 * description.
43 */
44
45 #include <sys/cdefs.h>
46 __FBSDID("$FreeBSD: releng/8.3/sys/kern/kern_malloc.c 230419 2012-01-21 07:21:44Z alc $");
47
48 #include "opt_ddb.h"
49 #include "opt_kdtrace.h"
50 #include "opt_vm.h"
51
52 #include <sys/param.h>
53 #include <sys/systm.h>
54 #include <sys/kdb.h>
55 #include <sys/kernel.h>
56 #include <sys/lock.h>
57 #include <sys/malloc.h>
58 #include <sys/mbuf.h>
59 #include <sys/mutex.h>
60 #include <sys/vmmeter.h>
61 #include <sys/proc.h>
62 #include <sys/sbuf.h>
63 #include <sys/sysctl.h>
64 #include <sys/time.h>
65
66 #include <vm/vm.h>
67 #include <vm/pmap.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>
73 #include <vm/uma.h>
74 #include <vm/uma_int.h>
75 #include <vm/uma_dbg.h>
76
77 #ifdef DEBUG_MEMGUARD
78 #include <vm/memguard.h>
79 #endif
80 #ifdef DEBUG_REDZONE
81 #include <vm/redzone.h>
82 #endif
83
84 #if defined(INVARIANTS) && defined(__i386__)
85 #include <machine/cpu.h>
86 #endif
87
88 #include <ddb/ddb.h>
89
90 #ifdef KDTRACE_HOOKS
91 #include <sys/dtrace_bsd.h>
92
93 dtrace_malloc_probe_func_t dtrace_malloc_probe;
94 #endif
95
96 /*
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'.
101 */
102 #ifndef REALLOC_FRACTION
103 #define REALLOC_FRACTION 1 /* new block if <= half the size */
104 #endif
105
106 /*
107 * Centrally define some common malloc types.
108 */
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");
112
113 MALLOC_DEFINE(M_IP6OPT, "ip6opt", "IPv6 options");
114 MALLOC_DEFINE(M_IP6NDP, "ip6ndp", "IPv6 Neighbor Discovery");
115
116 static void kmeminit(void *);
117 SYSINIT(kmem, SI_SUB_KMEM, SI_ORDER_FIRST, kmeminit, NULL);
118
119 static MALLOC_DEFINE(M_FREE, "free", "should be on free list");
120
121 static struct malloc_type *kmemstatistics;
122 static vm_offset_t kmembase;
123 static vm_offset_t kmemlimit;
124 static int kmemcount;
125
126 #define KMEM_ZSHIFT 4
127 #define KMEM_ZBASE 16
128 #define KMEM_ZMASK (KMEM_ZBASE - 1)
129
130 #define KMEM_ZMAX PAGE_SIZE
131 #define KMEM_ZSIZE (KMEM_ZMAX >> KMEM_ZSHIFT)
132 static u_int8_t kmemsize[KMEM_ZSIZE + 1];
133
134 /*
135 * Small malloc(9) memory allocations are allocated from a set of UMA buckets
136 * of various sizes.
137 *
138 * XXX: The comment here used to read "These won't be powers of two for
139 * long." It's possible that a significant amount of wasted memory could be
140 * recovered by tuning the sizes of these buckets.
141 */
142 struct {
143 int kz_size;
144 char *kz_name;
145 uma_zone_t kz_zone;
146 } kmemzones[] = {
147 {16, "16", NULL},
148 {32, "32", NULL},
149 {64, "64", NULL},
150 {128, "128", NULL},
151 {256, "256", NULL},
152 {512, "512", NULL},
153 {1024, "1024", NULL},
154 {2048, "2048", NULL},
155 {4096, "4096", NULL},
156 #if PAGE_SIZE > 4096
157 {8192, "8192", NULL},
158 #if PAGE_SIZE > 8192
159 {16384, "16384", NULL},
160 #if PAGE_SIZE > 16384
161 {32768, "32768", NULL},
162 #if PAGE_SIZE > 32768
163 {65536, "65536", NULL},
164 #if PAGE_SIZE > 65536
165 #error "Unsupported PAGE_SIZE"
166 #endif /* 65536 */
167 #endif /* 32768 */
168 #endif /* 16384 */
169 #endif /* 8192 */
170 #endif /* 4096 */
171 {0, NULL},
172 };
173
174 /*
175 * Zone to allocate malloc type descriptions from. For ABI reasons, memory
176 * types are described by a data structure passed by the declaring code, but
177 * the malloc(9) implementation has its own data structure describing the
178 * type and statistics. This permits the malloc(9)-internal data structures
179 * to be modified without breaking binary-compiled kernel modules that
180 * declare malloc types.
181 */
182 static uma_zone_t mt_zone;
183
184 u_long vm_kmem_size;
185 SYSCTL_ULONG(_vm, OID_AUTO, kmem_size, CTLFLAG_RDTUN, &vm_kmem_size, 0,
186 "Size of kernel memory");
187
188 static u_long vm_kmem_size_min;
189 SYSCTL_ULONG(_vm, OID_AUTO, kmem_size_min, CTLFLAG_RDTUN, &vm_kmem_size_min, 0,
190 "Minimum size of kernel memory");
191
192 static u_long vm_kmem_size_max;
193 SYSCTL_ULONG(_vm, OID_AUTO, kmem_size_max, CTLFLAG_RDTUN, &vm_kmem_size_max, 0,
194 "Maximum size of kernel memory");
195
196 static u_int vm_kmem_size_scale;
197 SYSCTL_UINT(_vm, OID_AUTO, kmem_size_scale, CTLFLAG_RDTUN, &vm_kmem_size_scale, 0,
198 "Scale factor for kernel memory size");
199
200 static int sysctl_kmem_map_size(SYSCTL_HANDLER_ARGS);
201 SYSCTL_PROC(_vm, OID_AUTO, kmem_map_size,
202 CTLFLAG_RD | CTLTYPE_ULONG | CTLFLAG_MPSAFE, NULL, 0,
203 sysctl_kmem_map_size, "LU", "Current kmem_map allocation size");
204
205 static int sysctl_kmem_map_free(SYSCTL_HANDLER_ARGS);
206 SYSCTL_PROC(_vm, OID_AUTO, kmem_map_free,
207 CTLFLAG_RD | CTLTYPE_ULONG | CTLFLAG_MPSAFE, NULL, 0,
208 sysctl_kmem_map_free, "LU", "Largest contiguous free range in kmem_map");
209
210 /*
211 * The malloc_mtx protects the kmemstatistics linked list.
212 */
213 struct mtx malloc_mtx;
214
215 #ifdef MALLOC_PROFILE
216 uint64_t krequests[KMEM_ZSIZE + 1];
217
218 static int sysctl_kern_mprof(SYSCTL_HANDLER_ARGS);
219 #endif
220
221 static int sysctl_kern_malloc_stats(SYSCTL_HANDLER_ARGS);
222
223 /*
224 * time_uptime of the last malloc(9) failure (induced or real).
225 */
226 static time_t t_malloc_fail;
227
228 /*
229 * malloc(9) fault injection -- cause malloc failures every (n) mallocs when
230 * the caller specifies M_NOWAIT. If set to 0, no failures are caused.
231 */
232 #ifdef MALLOC_MAKE_FAILURES
233 SYSCTL_NODE(_debug, OID_AUTO, malloc, CTLFLAG_RD, 0,
234 "Kernel malloc debugging options");
235
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");
244 #endif
245
246 static int
247 sysctl_kmem_map_size(SYSCTL_HANDLER_ARGS)
248 {
249 u_long size;
250
251 size = kmem_map->size;
252 return (sysctl_handle_long(oidp, &size, 0, req));
253 }
254
255 static int
256 sysctl_kmem_map_free(SYSCTL_HANDLER_ARGS)
257 {
258 u_long size;
259
260 vm_map_lock_read(kmem_map);
261 size = kmem_map->root != NULL ? kmem_map->root->max_free :
262 kmem_map->max_offset - kmem_map->min_offset;
263 vm_map_unlock_read(kmem_map);
264 return (sysctl_handle_long(oidp, &size, 0, req));
265 }
266
267 int
268 malloc_last_fail(void)
269 {
270
271 return (time_uptime - t_malloc_fail);
272 }
273
274 /*
275 * An allocation has succeeded -- update malloc type statistics for the
276 * amount of bucket size. Occurs within a critical section so that the
277 * thread isn't preempted and doesn't migrate while updating per-PCU
278 * statistics.
279 */
280 static void
281 malloc_type_zone_allocated(struct malloc_type *mtp, unsigned long size,
282 int zindx)
283 {
284 struct malloc_type_internal *mtip;
285 struct malloc_type_stats *mtsp;
286
287 critical_enter();
288 mtip = mtp->ks_handle;
289 mtsp = &mtip->mti_stats[curcpu];
290 if (size > 0) {
291 mtsp->mts_memalloced += size;
292 mtsp->mts_numallocs++;
293 }
294 if (zindx != -1)
295 mtsp->mts_size |= 1 << zindx;
296
297 #ifdef KDTRACE_HOOKS
298 if (dtrace_malloc_probe != NULL) {
299 uint32_t probe_id = mtip->mti_probes[DTMALLOC_PROBE_MALLOC];
300 if (probe_id != 0)
301 (dtrace_malloc_probe)(probe_id,
302 (uintptr_t) mtp, (uintptr_t) mtip,
303 (uintptr_t) mtsp, size, zindx);
304 }
305 #endif
306
307 critical_exit();
308 }
309
310 void
311 malloc_type_allocated(struct malloc_type *mtp, unsigned long size)
312 {
313
314 if (size > 0)
315 malloc_type_zone_allocated(mtp, size, -1);
316 }
317
318 /*
319 * A free operation has occurred -- update malloc type statistics for the
320 * amount of the bucket size. Occurs within a critical section so that the
321 * thread isn't preempted and doesn't migrate while updating per-CPU
322 * statistics.
323 */
324 void
325 malloc_type_freed(struct malloc_type *mtp, unsigned long size)
326 {
327 struct malloc_type_internal *mtip;
328 struct malloc_type_stats *mtsp;
329
330 critical_enter();
331 mtip = mtp->ks_handle;
332 mtsp = &mtip->mti_stats[curcpu];
333 mtsp->mts_memfreed += size;
334 mtsp->mts_numfrees++;
335
336 #ifdef KDTRACE_HOOKS
337 if (dtrace_malloc_probe != NULL) {
338 uint32_t probe_id = mtip->mti_probes[DTMALLOC_PROBE_FREE];
339 if (probe_id != 0)
340 (dtrace_malloc_probe)(probe_id,
341 (uintptr_t) mtp, (uintptr_t) mtip,
342 (uintptr_t) mtsp, size, 0);
343 }
344 #endif
345
346 critical_exit();
347 }
348
349 /*
350 * malloc:
351 *
352 * Allocate a block of memory.
353 *
354 * If M_NOWAIT is set, this routine will not block and return NULL if
355 * the allocation fails.
356 */
357 void *
358 malloc(unsigned long size, struct malloc_type *mtp, int flags)
359 {
360 int indx;
361 caddr_t va;
362 uma_zone_t zone;
363 #if defined(DIAGNOSTIC) || defined(DEBUG_REDZONE)
364 unsigned long osize = size;
365 #endif
366
367 #ifdef INVARIANTS
368 KASSERT(mtp->ks_magic == M_MAGIC, ("malloc: bad malloc type magic"));
369 /*
370 * Check that exactly one of M_WAITOK or M_NOWAIT is specified.
371 */
372 indx = flags & (M_WAITOK | M_NOWAIT);
373 if (indx != M_NOWAIT && indx != M_WAITOK) {
374 static struct timeval lasterr;
375 static int curerr, once;
376 if (once == 0 && ppsratecheck(&lasterr, &curerr, 1)) {
377 printf("Bad malloc flags: %x\n", indx);
378 kdb_backtrace();
379 flags |= M_WAITOK;
380 once++;
381 }
382 }
383 #endif
384 #ifdef MALLOC_MAKE_FAILURES
385 if ((flags & M_NOWAIT) && (malloc_failure_rate != 0)) {
386 atomic_add_int(&malloc_nowait_count, 1);
387 if ((malloc_nowait_count % malloc_failure_rate) == 0) {
388 atomic_add_int(&malloc_failure_count, 1);
389 t_malloc_fail = time_uptime;
390 return (NULL);
391 }
392 }
393 #endif
394 if (flags & M_WAITOK)
395 KASSERT(curthread->td_intr_nesting_level == 0,
396 ("malloc(M_WAITOK) in interrupt context"));
397
398 #ifdef DEBUG_MEMGUARD
399 if (memguard_cmp(mtp, size)) {
400 va = memguard_alloc(size, flags);
401 if (va != NULL)
402 return (va);
403 /* This is unfortunate but should not be fatal. */
404 }
405 #endif
406
407 #ifdef DEBUG_REDZONE
408 size = redzone_size_ntor(size);
409 #endif
410
411 if (size <= KMEM_ZMAX) {
412 if (size & KMEM_ZMASK)
413 size = (size & ~KMEM_ZMASK) + KMEM_ZBASE;
414 indx = kmemsize[size >> KMEM_ZSHIFT];
415 zone = kmemzones[indx].kz_zone;
416 #ifdef MALLOC_PROFILE
417 krequests[size >> KMEM_ZSHIFT]++;
418 #endif
419 va = uma_zalloc(zone, flags);
420 if (va != NULL)
421 size = zone->uz_size;
422 malloc_type_zone_allocated(mtp, va == NULL ? 0 : size, indx);
423 } else {
424 size = roundup(size, PAGE_SIZE);
425 zone = NULL;
426 va = uma_large_malloc(size, flags);
427 malloc_type_allocated(mtp, va == NULL ? 0 : size);
428 }
429 if (flags & M_WAITOK)
430 KASSERT(va != NULL, ("malloc(M_WAITOK) returned NULL"));
431 else if (va == NULL)
432 t_malloc_fail = time_uptime;
433 #ifdef DIAGNOSTIC
434 if (va != NULL && !(flags & M_ZERO)) {
435 memset(va, 0x70, osize);
436 }
437 #endif
438 #ifdef DEBUG_REDZONE
439 if (va != NULL)
440 va = redzone_setup(va, osize);
441 #endif
442 return ((void *) va);
443 }
444
445 /*
446 * free:
447 *
448 * Free a block of memory allocated by malloc.
449 *
450 * This routine may not block.
451 */
452 void
453 free(void *addr, struct malloc_type *mtp)
454 {
455 uma_slab_t slab;
456 u_long size;
457
458 KASSERT(mtp->ks_magic == M_MAGIC, ("free: bad malloc type magic"));
459
460 /* free(NULL, ...) does nothing */
461 if (addr == NULL)
462 return;
463
464 #ifdef DEBUG_MEMGUARD
465 if (is_memguard_addr(addr)) {
466 memguard_free(addr);
467 return;
468 }
469 #endif
470
471 #ifdef DEBUG_REDZONE
472 redzone_check(addr);
473 addr = redzone_addr_ntor(addr);
474 #endif
475
476 slab = vtoslab((vm_offset_t)addr & (~UMA_SLAB_MASK));
477
478 if (slab == NULL)
479 panic("free: address %p(%p) has not been allocated.\n",
480 addr, (void *)((u_long)addr & (~UMA_SLAB_MASK)));
481
482
483 if (!(slab->us_flags & UMA_SLAB_MALLOC)) {
484 #ifdef INVARIANTS
485 struct malloc_type **mtpp = addr;
486 #endif
487 size = slab->us_keg->uk_size;
488 #ifdef INVARIANTS
489 /*
490 * Cache a pointer to the malloc_type that most recently freed
491 * this memory here. This way we know who is most likely to
492 * have stepped on it later.
493 *
494 * This code assumes that size is a multiple of 8 bytes for
495 * 64 bit machines
496 */
497 mtpp = (struct malloc_type **)
498 ((unsigned long)mtpp & ~UMA_ALIGN_PTR);
499 mtpp += (size - sizeof(struct malloc_type *)) /
500 sizeof(struct malloc_type *);
501 *mtpp = mtp;
502 #endif
503 uma_zfree_arg(LIST_FIRST(&slab->us_keg->uk_zones), addr, slab);
504 } else {
505 size = slab->us_size;
506 uma_large_free(slab);
507 }
508 malloc_type_freed(mtp, size);
509 }
510
511 /*
512 * realloc: change the size of a memory block
513 */
514 void *
515 realloc(void *addr, unsigned long size, struct malloc_type *mtp, int flags)
516 {
517 uma_slab_t slab;
518 unsigned long alloc;
519 void *newaddr;
520
521 KASSERT(mtp->ks_magic == M_MAGIC,
522 ("realloc: bad malloc type magic"));
523
524 /* realloc(NULL, ...) is equivalent to malloc(...) */
525 if (addr == NULL)
526 return (malloc(size, mtp, flags));
527
528 /*
529 * XXX: Should report free of old memory and alloc of new memory to
530 * per-CPU stats.
531 */
532
533 #ifdef DEBUG_MEMGUARD
534 if (is_memguard_addr(addr))
535 return (memguard_realloc(addr, size, mtp, flags));
536 #endif
537
538 #ifdef DEBUG_REDZONE
539 slab = NULL;
540 alloc = redzone_get_size(addr);
541 #else
542 slab = vtoslab((vm_offset_t)addr & ~(UMA_SLAB_MASK));
543
544 /* Sanity check */
545 KASSERT(slab != NULL,
546 ("realloc: address %p out of range", (void *)addr));
547
548 /* Get the size of the original block */
549 if (!(slab->us_flags & UMA_SLAB_MALLOC))
550 alloc = slab->us_keg->uk_size;
551 else
552 alloc = slab->us_size;
553
554 /* Reuse the original block if appropriate */
555 if (size <= alloc
556 && (size > (alloc >> REALLOC_FRACTION) || alloc == MINALLOCSIZE))
557 return (addr);
558 #endif /* !DEBUG_REDZONE */
559
560 /* Allocate a new, bigger (or smaller) block */
561 if ((newaddr = malloc(size, mtp, flags)) == NULL)
562 return (NULL);
563
564 /* Copy over original contents */
565 bcopy(addr, newaddr, min(size, alloc));
566 free(addr, mtp);
567 return (newaddr);
568 }
569
570 /*
571 * reallocf: same as realloc() but free memory on failure.
572 */
573 void *
574 reallocf(void *addr, unsigned long size, struct malloc_type *mtp, int flags)
575 {
576 void *mem;
577
578 if ((mem = realloc(addr, size, mtp, flags)) == NULL)
579 free(addr, mtp);
580 return (mem);
581 }
582
583 /*
584 * Initialize the kernel memory allocator
585 */
586 /* ARGSUSED*/
587 static void
588 kmeminit(void *dummy)
589 {
590 u_int8_t indx;
591 u_long mem_size, tmp;
592 int i;
593
594 mtx_init(&malloc_mtx, "malloc", NULL, MTX_DEF);
595
596 /*
597 * Try to auto-tune the kernel memory size, so that it is
598 * more applicable for a wider range of machine sizes. The
599 * VM_KMEM_SIZE_MAX is dependent on the maximum KVA space
600 * available.
601 *
602 * Note that the kmem_map is also used by the zone allocator,
603 * so make sure that there is enough space.
604 */
605 vm_kmem_size = VM_KMEM_SIZE + nmbclusters * PAGE_SIZE;
606 mem_size = cnt.v_page_count;
607
608 #if defined(VM_KMEM_SIZE_SCALE)
609 vm_kmem_size_scale = VM_KMEM_SIZE_SCALE;
610 #endif
611 TUNABLE_INT_FETCH("vm.kmem_size_scale", &vm_kmem_size_scale);
612 if (vm_kmem_size_scale > 0 &&
613 (mem_size / vm_kmem_size_scale) > (vm_kmem_size / PAGE_SIZE))
614 vm_kmem_size = (mem_size / vm_kmem_size_scale) * PAGE_SIZE;
615
616 #if defined(VM_KMEM_SIZE_MIN)
617 vm_kmem_size_min = VM_KMEM_SIZE_MIN;
618 #endif
619 TUNABLE_ULONG_FETCH("vm.kmem_size_min", &vm_kmem_size_min);
620 if (vm_kmem_size_min > 0 && vm_kmem_size < vm_kmem_size_min) {
621 vm_kmem_size = vm_kmem_size_min;
622 }
623
624 #if defined(VM_KMEM_SIZE_MAX)
625 vm_kmem_size_max = VM_KMEM_SIZE_MAX;
626 #endif
627 TUNABLE_ULONG_FETCH("vm.kmem_size_max", &vm_kmem_size_max);
628 if (vm_kmem_size_max > 0 && vm_kmem_size >= vm_kmem_size_max)
629 vm_kmem_size = vm_kmem_size_max;
630
631 /* Allow final override from the kernel environment */
632 TUNABLE_ULONG_FETCH("vm.kmem_size", &vm_kmem_size);
633
634 /*
635 * Limit kmem virtual size to twice the physical memory.
636 * This allows for kmem map sparseness, but limits the size
637 * to something sane. Be careful to not overflow the 32bit
638 * ints while doing the check or the adjustment.
639 */
640 if (vm_kmem_size / 2 / PAGE_SIZE > mem_size)
641 vm_kmem_size = 2 * mem_size * PAGE_SIZE;
642
643 /*
644 * Tune settings based on the kmem map's size at this time.
645 */
646 init_param3(vm_kmem_size / PAGE_SIZE);
647
648 #ifdef DEBUG_MEMGUARD
649 tmp = memguard_fudge(vm_kmem_size, vm_kmem_size_max);
650 #else
651 tmp = vm_kmem_size;
652 #endif
653 kmem_map = kmem_suballoc(kernel_map, &kmembase, &kmemlimit,
654 tmp, TRUE);
655 kmem_map->system_map = 1;
656
657 #ifdef DEBUG_MEMGUARD
658 /*
659 * Initialize MemGuard if support compiled in. MemGuard is a
660 * replacement allocator used for detecting tamper-after-free
661 * scenarios as they occur. It is only used for debugging.
662 */
663 memguard_init(kmem_map);
664 #endif
665
666 uma_startup2();
667
668 mt_zone = uma_zcreate("mt_zone", sizeof(struct malloc_type_internal),
669 #ifdef INVARIANTS
670 mtrash_ctor, mtrash_dtor, mtrash_init, mtrash_fini,
671 #else
672 NULL, NULL, NULL, NULL,
673 #endif
674 UMA_ALIGN_PTR, UMA_ZONE_MALLOC);
675 for (i = 0, indx = 0; kmemzones[indx].kz_size != 0; indx++) {
676 int size = kmemzones[indx].kz_size;
677 char *name = kmemzones[indx].kz_name;
678
679 kmemzones[indx].kz_zone = uma_zcreate(name, size,
680 #ifdef INVARIANTS
681 mtrash_ctor, mtrash_dtor, mtrash_init, mtrash_fini,
682 #else
683 NULL, NULL, NULL, NULL,
684 #endif
685 UMA_ALIGN_PTR, UMA_ZONE_MALLOC);
686
687 for (;i <= size; i+= KMEM_ZBASE)
688 kmemsize[i >> KMEM_ZSHIFT] = indx;
689
690 }
691 }
692
693 void
694 malloc_init(void *data)
695 {
696 struct malloc_type_internal *mtip;
697 struct malloc_type *mtp;
698
699 KASSERT(cnt.v_page_count != 0, ("malloc_register before vm_init"));
700
701 mtp = data;
702 if (mtp->ks_magic != M_MAGIC)
703 panic("malloc_init: bad malloc type magic");
704
705 mtip = uma_zalloc(mt_zone, M_WAITOK | M_ZERO);
706 mtp->ks_handle = mtip;
707
708 mtx_lock(&malloc_mtx);
709 mtp->ks_next = kmemstatistics;
710 kmemstatistics = mtp;
711 kmemcount++;
712 mtx_unlock(&malloc_mtx);
713 }
714
715 void
716 malloc_uninit(void *data)
717 {
718 struct malloc_type_internal *mtip;
719 struct malloc_type_stats *mtsp;
720 struct malloc_type *mtp, *temp;
721 uma_slab_t slab;
722 long temp_allocs, temp_bytes;
723 int i;
724
725 mtp = data;
726 KASSERT(mtp->ks_magic == M_MAGIC,
727 ("malloc_uninit: bad malloc type magic"));
728 KASSERT(mtp->ks_handle != NULL, ("malloc_deregister: cookie NULL"));
729
730 mtx_lock(&malloc_mtx);
731 mtip = mtp->ks_handle;
732 mtp->ks_handle = NULL;
733 if (mtp != kmemstatistics) {
734 for (temp = kmemstatistics; temp != NULL;
735 temp = temp->ks_next) {
736 if (temp->ks_next == mtp) {
737 temp->ks_next = mtp->ks_next;
738 break;
739 }
740 }
741 KASSERT(temp,
742 ("malloc_uninit: type '%s' not found", mtp->ks_shortdesc));
743 } else
744 kmemstatistics = mtp->ks_next;
745 kmemcount--;
746 mtx_unlock(&malloc_mtx);
747
748 /*
749 * Look for memory leaks.
750 */
751 temp_allocs = temp_bytes = 0;
752 for (i = 0; i < MAXCPU; i++) {
753 mtsp = &mtip->mti_stats[i];
754 temp_allocs += mtsp->mts_numallocs;
755 temp_allocs -= mtsp->mts_numfrees;
756 temp_bytes += mtsp->mts_memalloced;
757 temp_bytes -= mtsp->mts_memfreed;
758 }
759 if (temp_allocs > 0 || temp_bytes > 0) {
760 printf("Warning: memory type %s leaked memory on destroy "
761 "(%ld allocations, %ld bytes leaked).\n", mtp->ks_shortdesc,
762 temp_allocs, temp_bytes);
763 }
764
765 slab = vtoslab((vm_offset_t) mtip & (~UMA_SLAB_MASK));
766 uma_zfree_arg(mt_zone, mtip, slab);
767 }
768
769 struct malloc_type *
770 malloc_desc2type(const char *desc)
771 {
772 struct malloc_type *mtp;
773
774 mtx_assert(&malloc_mtx, MA_OWNED);
775 for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) {
776 if (strcmp(mtp->ks_shortdesc, desc) == 0)
777 return (mtp);
778 }
779 return (NULL);
780 }
781
782 static int
783 sysctl_kern_malloc_stats(SYSCTL_HANDLER_ARGS)
784 {
785 struct malloc_type_stream_header mtsh;
786 struct malloc_type_internal *mtip;
787 struct malloc_type_header mth;
788 struct malloc_type *mtp;
789 int buflen, count, error, i;
790 struct sbuf sbuf;
791 char *buffer;
792
793 mtx_lock(&malloc_mtx);
794 restart:
795 mtx_assert(&malloc_mtx, MA_OWNED);
796 count = kmemcount;
797 mtx_unlock(&malloc_mtx);
798 buflen = sizeof(mtsh) + count * (sizeof(mth) +
799 sizeof(struct malloc_type_stats) * MAXCPU) + 1;
800 buffer = malloc(buflen, M_TEMP, M_WAITOK | M_ZERO);
801 mtx_lock(&malloc_mtx);
802 if (count < kmemcount) {
803 free(buffer, M_TEMP);
804 goto restart;
805 }
806
807 sbuf_new(&sbuf, buffer, buflen, SBUF_FIXEDLEN);
808
809 /*
810 * Insert stream header.
811 */
812 bzero(&mtsh, sizeof(mtsh));
813 mtsh.mtsh_version = MALLOC_TYPE_STREAM_VERSION;
814 mtsh.mtsh_maxcpus = MAXCPU;
815 mtsh.mtsh_count = kmemcount;
816 if (sbuf_bcat(&sbuf, &mtsh, sizeof(mtsh)) < 0) {
817 mtx_unlock(&malloc_mtx);
818 error = ENOMEM;
819 goto out;
820 }
821
822 /*
823 * Insert alternating sequence of type headers and type statistics.
824 */
825 for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) {
826 mtip = (struct malloc_type_internal *)mtp->ks_handle;
827
828 /*
829 * Insert type header.
830 */
831 bzero(&mth, sizeof(mth));
832 strlcpy(mth.mth_name, mtp->ks_shortdesc, MALLOC_MAX_NAME);
833 if (sbuf_bcat(&sbuf, &mth, sizeof(mth)) < 0) {
834 mtx_unlock(&malloc_mtx);
835 error = ENOMEM;
836 goto out;
837 }
838
839 /*
840 * Insert type statistics for each CPU.
841 */
842 for (i = 0; i < MAXCPU; i++) {
843 if (sbuf_bcat(&sbuf, &mtip->mti_stats[i],
844 sizeof(mtip->mti_stats[i])) < 0) {
845 mtx_unlock(&malloc_mtx);
846 error = ENOMEM;
847 goto out;
848 }
849 }
850 }
851 mtx_unlock(&malloc_mtx);
852 sbuf_finish(&sbuf);
853 error = SYSCTL_OUT(req, sbuf_data(&sbuf), sbuf_len(&sbuf));
854 out:
855 sbuf_delete(&sbuf);
856 free(buffer, M_TEMP);
857 return (error);
858 }
859
860 SYSCTL_PROC(_kern, OID_AUTO, malloc_stats, CTLFLAG_RD|CTLTYPE_STRUCT,
861 0, 0, sysctl_kern_malloc_stats, "s,malloc_type_ustats",
862 "Return malloc types");
863
864 SYSCTL_INT(_kern, OID_AUTO, malloc_count, CTLFLAG_RD, &kmemcount, 0,
865 "Count of kernel malloc types");
866
867 void
868 malloc_type_list(malloc_type_list_func_t *func, void *arg)
869 {
870 struct malloc_type *mtp, **bufmtp;
871 int count, i;
872 size_t buflen;
873
874 mtx_lock(&malloc_mtx);
875 restart:
876 mtx_assert(&malloc_mtx, MA_OWNED);
877 count = kmemcount;
878 mtx_unlock(&malloc_mtx);
879
880 buflen = sizeof(struct malloc_type *) * count;
881 bufmtp = malloc(buflen, M_TEMP, M_WAITOK);
882
883 mtx_lock(&malloc_mtx);
884
885 if (count < kmemcount) {
886 free(bufmtp, M_TEMP);
887 goto restart;
888 }
889
890 for (mtp = kmemstatistics, i = 0; mtp != NULL; mtp = mtp->ks_next, i++)
891 bufmtp[i] = mtp;
892
893 mtx_unlock(&malloc_mtx);
894
895 for (i = 0; i < count; i++)
896 (func)(bufmtp[i], arg);
897
898 free(bufmtp, M_TEMP);
899 }
900
901 #ifdef DDB
902 DB_SHOW_COMMAND(malloc, db_show_malloc)
903 {
904 struct malloc_type_internal *mtip;
905 struct malloc_type *mtp;
906 u_int64_t allocs, frees;
907 u_int64_t alloced, freed;
908 int i;
909
910 db_printf("%18s %12s %12s %12s\n", "Type", "InUse", "MemUse",
911 "Requests");
912 for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) {
913 mtip = (struct malloc_type_internal *)mtp->ks_handle;
914 allocs = 0;
915 frees = 0;
916 alloced = 0;
917 freed = 0;
918 for (i = 0; i < MAXCPU; i++) {
919 allocs += mtip->mti_stats[i].mts_numallocs;
920 frees += mtip->mti_stats[i].mts_numfrees;
921 alloced += mtip->mti_stats[i].mts_memalloced;
922 freed += mtip->mti_stats[i].mts_memfreed;
923 }
924 db_printf("%18s %12ju %12juK %12ju\n",
925 mtp->ks_shortdesc, allocs - frees,
926 (alloced - freed + 1023) / 1024, allocs);
927 }
928 }
929 #endif
930
931 #ifdef MALLOC_PROFILE
932
933 static int
934 sysctl_kern_mprof(SYSCTL_HANDLER_ARGS)
935 {
936 int linesize = 64;
937 struct sbuf sbuf;
938 uint64_t count;
939 uint64_t waste;
940 uint64_t mem;
941 int bufsize;
942 int error;
943 char *buf;
944 int rsize;
945 int size;
946 int i;
947
948 bufsize = linesize * (KMEM_ZSIZE + 1);
949 bufsize += 128; /* For the stats line */
950 bufsize += 128; /* For the banner line */
951 waste = 0;
952 mem = 0;
953
954 buf = malloc(bufsize, M_TEMP, M_WAITOK|M_ZERO);
955 sbuf_new(&sbuf, buf, bufsize, SBUF_FIXEDLEN);
956 sbuf_printf(&sbuf,
957 "\n Size Requests Real Size\n");
958 for (i = 0; i < KMEM_ZSIZE; i++) {
959 size = i << KMEM_ZSHIFT;
960 rsize = kmemzones[kmemsize[i]].kz_size;
961 count = (long long unsigned)krequests[i];
962
963 sbuf_printf(&sbuf, "%6d%28llu%11d\n", size,
964 (unsigned long long)count, rsize);
965
966 if ((rsize * count) > (size * count))
967 waste += (rsize * count) - (size * count);
968 mem += (rsize * count);
969 }
970 sbuf_printf(&sbuf,
971 "\nTotal memory used:\t%30llu\nTotal Memory wasted:\t%30llu\n",
972 (unsigned long long)mem, (unsigned long long)waste);
973 sbuf_finish(&sbuf);
974
975 error = SYSCTL_OUT(req, sbuf_data(&sbuf), sbuf_len(&sbuf));
976
977 sbuf_delete(&sbuf);
978 free(buf, M_TEMP);
979 return (error);
980 }
981
982 SYSCTL_OID(_kern, OID_AUTO, mprof, CTLTYPE_STRING|CTLFLAG_RD,
983 NULL, 0, sysctl_kern_mprof, "A", "Malloc Profiling");
984 #endif /* MALLOC_PROFILE */
Cache object: e478c2619efd9d4fba3988504504705c
|