xref: /freebsd/sys/kern/kern_malloc.c (revision a3e8fd0b7f663db7eafff527d5c3ca3bcfa8a537)
1 /*
2  * Copyright (c) 1987, 1991, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  * 3. All advertising materials mentioning features or use of this software
14  *    must display the following acknowledgement:
15  *	This product includes software developed by the University of
16  *	California, Berkeley and its contributors.
17  * 4. Neither the name of the University nor the names of its contributors
18  *    may be used to endorse or promote products derived from this software
19  *    without specific prior written permission.
20  *
21  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
22  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31  * SUCH DAMAGE.
32  *
33  *	@(#)kern_malloc.c	8.3 (Berkeley) 1/4/94
34  * $FreeBSD$
35  */
36 
37 #include "opt_vm.h"
38 
39 #include <sys/param.h>
40 #include <sys/systm.h>
41 #include <sys/kernel.h>
42 #include <sys/lock.h>
43 #include <sys/malloc.h>
44 #include <sys/mbuf.h>
45 #include <sys/mutex.h>
46 #include <sys/vmmeter.h>
47 #include <sys/proc.h>
48 #include <sys/sysctl.h>
49 
50 #include <vm/vm.h>
51 #include <vm/pmap.h>
52 #include <vm/vm_param.h>
53 #include <vm/vm_kern.h>
54 #include <vm/vm_extern.h>
55 #include <vm/vm_map.h>
56 #include <vm/vm_page.h>
57 #include <vm/uma.h>
58 #include <vm/uma_int.h>
59 #include <vm/uma_dbg.h>
60 
61 #if defined(INVARIANTS) && defined(__i386__)
62 #include <machine/cpu.h>
63 #endif
64 
65 /*
66  * When realloc() is called, if the new size is sufficiently smaller than
67  * the old size, realloc() will allocate a new, smaller block to avoid
68  * wasting memory. 'Sufficiently smaller' is defined as: newsize <=
69  * oldsize / 2^n, where REALLOC_FRACTION defines the value of 'n'.
70  */
71 #ifndef REALLOC_FRACTION
72 #define	REALLOC_FRACTION	1	/* new block if <= half the size */
73 #endif
74 
75 MALLOC_DEFINE(M_CACHE, "cache", "Various Dynamically allocated caches");
76 MALLOC_DEFINE(M_DEVBUF, "devbuf", "device driver memory");
77 MALLOC_DEFINE(M_TEMP, "temp", "misc temporary data buffers");
78 
79 MALLOC_DEFINE(M_IP6OPT, "ip6opt", "IPv6 options");
80 MALLOC_DEFINE(M_IP6NDP, "ip6ndp", "IPv6 Neighbor Discovery");
81 
82 static void kmeminit(void *);
83 SYSINIT(kmem, SI_SUB_KMEM, SI_ORDER_FIRST, kmeminit, NULL)
84 
85 static MALLOC_DEFINE(M_FREE, "free", "should be on free list");
86 
87 static struct malloc_type *kmemstatistics;
88 static char *kmembase;
89 static char *kmemlimit;
90 
91 #define KMEM_ZSHIFT	4
92 #define KMEM_ZBASE	16
93 #define KMEM_ZMASK	(KMEM_ZBASE - 1)
94 
95 #define KMEM_ZMAX	65536
96 #define KMEM_ZSIZE	(KMEM_ZMAX >> KMEM_ZSHIFT)
97 static u_int8_t kmemsize[KMEM_ZSIZE + 1];
98 
99 /* These won't be powers of two for long */
100 struct {
101 	int kz_size;
102 	char *kz_name;
103 	uma_zone_t kz_zone;
104 } kmemzones[] = {
105 	{16, "16", NULL},
106 	{32, "32", NULL},
107 	{64, "64", NULL},
108 	{128, "128", NULL},
109 	{256, "256", NULL},
110 	{512, "512", NULL},
111 	{1024, "1024", NULL},
112 	{2048, "2048", NULL},
113 	{4096, "4096", NULL},
114 	{8192, "8192", NULL},
115 	{16384, "16384", NULL},
116 	{32768, "32768", NULL},
117 	{65536, "65536", NULL},
118 	{0, NULL},
119 };
120 
121 u_int vm_kmem_size;
122 
123 /*
124  * The malloc_mtx protects the kmemstatistics linked list.
125  */
126 
127 struct mtx malloc_mtx;
128 
129 #ifdef MALLOC_PROFILE
130 uint64_t krequests[KMEM_ZSIZE + 1];
131 
132 static int sysctl_kern_mprof(SYSCTL_HANDLER_ARGS);
133 #endif
134 
135 static int sysctl_kern_malloc(SYSCTL_HANDLER_ARGS);
136 
137 /*
138  *	malloc:
139  *
140  *	Allocate a block of memory.
141  *
142  *	If M_NOWAIT is set, this routine will not block and return NULL if
143  *	the allocation fails.
144  */
145 void *
146 malloc(size, type, flags)
147 	unsigned long size;
148 	struct malloc_type *type;
149 	int flags;
150 {
151 	int indx;
152 	caddr_t va;
153 	uma_zone_t zone;
154 	register struct malloc_type *ksp = type;
155 
156 #if 0
157 	if (size == 0)
158 		Debugger("zero size malloc");
159 #endif
160 	if (!(flags & M_NOWAIT))
161 		KASSERT(curthread->td_intr_nesting_level == 0,
162 		   ("malloc(M_WAITOK) in interrupt context"));
163 	if (size <= KMEM_ZMAX) {
164 		if (size & KMEM_ZMASK)
165 			size = (size & ~KMEM_ZMASK) + KMEM_ZBASE;
166 		indx = kmemsize[size >> KMEM_ZSHIFT];
167 		zone = kmemzones[indx].kz_zone;
168 #ifdef MALLOC_PROFILE
169 		krequests[size >> KMEM_ZSHIFT]++;
170 #endif
171 		va = uma_zalloc(zone, flags);
172 		mtx_lock(&ksp->ks_mtx);
173 		if (va == NULL)
174 			goto out;
175 
176 		ksp->ks_size |= 1 << indx;
177 		size = zone->uz_size;
178 	} else {
179 		size = roundup(size, PAGE_SIZE);
180 		zone = NULL;
181 		va = uma_large_malloc(size, flags);
182 		mtx_lock(&ksp->ks_mtx);
183 		if (va == NULL)
184 			goto out;
185 	}
186 	ksp->ks_memuse += size;
187 	ksp->ks_inuse++;
188 out:
189 	ksp->ks_calls++;
190 	if (ksp->ks_memuse > ksp->ks_maxused)
191 		ksp->ks_maxused = ksp->ks_memuse;
192 
193 	mtx_unlock(&ksp->ks_mtx);
194 	return ((void *) va);
195 }
196 
197 /*
198  *	free:
199  *
200  *	Free a block of memory allocated by malloc.
201  *
202  *	This routine may not block.
203  */
204 void
205 free(addr, type)
206 	void *addr;
207 	struct malloc_type *type;
208 {
209 	register struct malloc_type *ksp = type;
210 	uma_slab_t slab;
211 	u_long size;
212 
213 	/* free(NULL, ...) does nothing */
214 	if (addr == NULL)
215 		return;
216 
217 	size = 0;
218 
219 	slab = vtoslab((vm_offset_t)addr & (~UMA_SLAB_MASK));
220 
221 	if (slab == NULL)
222 		panic("free: address %p(%p) has not been allocated.\n",
223 		    addr, (void *)((u_long)addr & (~UMA_SLAB_MASK)));
224 
225 
226 	if (!(slab->us_flags & UMA_SLAB_MALLOC)) {
227 #ifdef INVARIANTS
228 		struct malloc_type **mtp = addr;
229 #endif
230 		size = slab->us_zone->uz_size;
231 #ifdef INVARIANTS
232 		/*
233 		 * Cache a pointer to the malloc_type that most recently freed
234 		 * this memory here.  This way we know who is most likely to
235 		 * have stepped on it later.
236 		 *
237 		 * This code assumes that size is a multiple of 8 bytes for
238 		 * 64 bit machines
239 		 */
240 		mtp = (struct malloc_type **)
241 		    ((unsigned long)mtp & ~UMA_ALIGN_PTR);
242 		mtp += (size - sizeof(struct malloc_type *)) /
243 		    sizeof(struct malloc_type *);
244 		*mtp = type;
245 #endif
246 		uma_zfree_arg(slab->us_zone, addr, slab);
247 	} else {
248 		size = slab->us_size;
249 		uma_large_free(slab);
250 	}
251 	mtx_lock(&ksp->ks_mtx);
252 	ksp->ks_memuse -= size;
253 	ksp->ks_inuse--;
254 	mtx_unlock(&ksp->ks_mtx);
255 }
256 
257 /*
258  *	realloc: change the size of a memory block
259  */
260 void *
261 realloc(addr, size, type, flags)
262 	void *addr;
263 	unsigned long size;
264 	struct malloc_type *type;
265 	int flags;
266 {
267 	uma_slab_t slab;
268 	unsigned long alloc;
269 	void *newaddr;
270 
271 	/* realloc(NULL, ...) is equivalent to malloc(...) */
272 	if (addr == NULL)
273 		return (malloc(size, type, flags));
274 
275 	slab = vtoslab((vm_offset_t)addr & ~(UMA_SLAB_MASK));
276 
277 	/* Sanity check */
278 	KASSERT(slab != NULL,
279 	    ("realloc: address %p out of range", (void *)addr));
280 
281 	/* Get the size of the original block */
282 	if (slab->us_zone)
283 		alloc = slab->us_zone->uz_size;
284 	else
285 		alloc = slab->us_size;
286 
287 	/* Reuse the original block if appropriate */
288 	if (size <= alloc
289 	    && (size > (alloc >> REALLOC_FRACTION) || alloc == MINALLOCSIZE))
290 		return (addr);
291 
292 	/* Allocate a new, bigger (or smaller) block */
293 	if ((newaddr = malloc(size, type, flags)) == NULL)
294 		return (NULL);
295 
296 	/* Copy over original contents */
297 	bcopy(addr, newaddr, min(size, alloc));
298 	free(addr, type);
299 	return (newaddr);
300 }
301 
302 /*
303  *	reallocf: same as realloc() but free memory on failure.
304  */
305 void *
306 reallocf(addr, size, type, flags)
307 	void *addr;
308 	unsigned long size;
309 	struct malloc_type *type;
310 	int flags;
311 {
312 	void *mem;
313 
314 	if ((mem = realloc(addr, size, type, flags)) == NULL)
315 		free(addr, type);
316 	return (mem);
317 }
318 
319 /*
320  * Initialize the kernel memory allocator
321  */
322 /* ARGSUSED*/
323 static void
324 kmeminit(dummy)
325 	void *dummy;
326 {
327 	u_int8_t indx;
328 	u_long npg;
329 	u_long mem_size;
330 	int i;
331 
332 	mtx_init(&malloc_mtx, "malloc", NULL, MTX_DEF);
333 
334 	/*
335 	 * Try to auto-tune the kernel memory size, so that it is
336 	 * more applicable for a wider range of machine sizes.
337 	 * On an X86, a VM_KMEM_SIZE_SCALE value of 4 is good, while
338 	 * a VM_KMEM_SIZE of 12MB is a fair compromise.  The
339 	 * VM_KMEM_SIZE_MAX is dependent on the maximum KVA space
340 	 * available, and on an X86 with a total KVA space of 256MB,
341 	 * try to keep VM_KMEM_SIZE_MAX at 80MB or below.
342 	 *
343 	 * Note that the kmem_map is also used by the zone allocator,
344 	 * so make sure that there is enough space.
345 	 */
346 	vm_kmem_size = VM_KMEM_SIZE;
347 	mem_size = cnt.v_page_count * PAGE_SIZE;
348 
349 #if defined(VM_KMEM_SIZE_SCALE)
350 	if ((mem_size / VM_KMEM_SIZE_SCALE) > vm_kmem_size)
351 		vm_kmem_size = mem_size / VM_KMEM_SIZE_SCALE;
352 #endif
353 
354 #if defined(VM_KMEM_SIZE_MAX)
355 	if (vm_kmem_size >= VM_KMEM_SIZE_MAX)
356 		vm_kmem_size = VM_KMEM_SIZE_MAX;
357 #endif
358 
359 	/* Allow final override from the kernel environment */
360 	TUNABLE_INT_FETCH("kern.vm.kmem.size", &vm_kmem_size);
361 
362 	/*
363 	 * Limit kmem virtual size to twice the physical memory.
364 	 * This allows for kmem map sparseness, but limits the size
365 	 * to something sane. Be careful to not overflow the 32bit
366 	 * ints while doing the check.
367 	 */
368 	if ((vm_kmem_size / 2) > (cnt.v_page_count * PAGE_SIZE))
369 		vm_kmem_size = 2 * cnt.v_page_count * PAGE_SIZE;
370 
371 	/*
372 	 * In mbuf_init(), we set up submaps for mbufs and clusters, in which
373 	 * case we rounddown() (nmbufs * MSIZE) and (nmbclusters * MCLBYTES),
374 	 * respectively. Mathematically, this means that what we do here may
375 	 * amount to slightly more address space than we need for the submaps,
376 	 * but it never hurts to have an extra page in kmem_map.
377 	 */
378 	npg = (nmbufs * MSIZE + nmbclusters * MCLBYTES + nmbcnt *
379 	    sizeof(u_int) + vm_kmem_size) / PAGE_SIZE;
380 
381 	kmem_map = kmem_suballoc(kernel_map, (vm_offset_t *)&kmembase,
382 		(vm_offset_t *)&kmemlimit, (vm_size_t)(npg * PAGE_SIZE));
383 	kmem_map->system_map = 1;
384 
385 	uma_startup2();
386 
387 	for (i = 0, indx = 0; kmemzones[indx].kz_size != 0; indx++) {
388 		int size = kmemzones[indx].kz_size;
389 		char *name = kmemzones[indx].kz_name;
390 
391 		kmemzones[indx].kz_zone = uma_zcreate(name, size,
392 #ifdef INVARIANTS
393 		    mtrash_ctor, mtrash_dtor, mtrash_init, mtrash_fini,
394 #else
395 		    NULL, NULL, NULL, NULL,
396 #endif
397 		    UMA_ALIGN_PTR, UMA_ZONE_MALLOC);
398 
399 		for (;i <= size; i+= KMEM_ZBASE)
400 			kmemsize[i >> KMEM_ZSHIFT] = indx;
401 
402 	}
403 }
404 
405 void
406 malloc_init(data)
407 	void *data;
408 {
409 	struct malloc_type *type = (struct malloc_type *)data;
410 
411 	mtx_lock(&malloc_mtx);
412 	if (type->ks_magic != M_MAGIC)
413 		panic("malloc type lacks magic");
414 
415 	if (cnt.v_page_count == 0)
416 		panic("malloc_init not allowed before vm init");
417 
418 	if (type->ks_next != NULL)
419 		return;
420 
421 	type->ks_next = kmemstatistics;
422 	kmemstatistics = type;
423 	mtx_init(&type->ks_mtx, type->ks_shortdesc, "Malloc Stats", MTX_DEF);
424 	mtx_unlock(&malloc_mtx);
425 }
426 
427 void
428 malloc_uninit(data)
429 	void *data;
430 {
431 	struct malloc_type *type = (struct malloc_type *)data;
432 	struct malloc_type *t;
433 
434 	mtx_lock(&malloc_mtx);
435 	mtx_lock(&type->ks_mtx);
436 	if (type->ks_magic != M_MAGIC)
437 		panic("malloc type lacks magic");
438 
439 	if (cnt.v_page_count == 0)
440 		panic("malloc_uninit not allowed before vm init");
441 
442 	if (type == kmemstatistics)
443 		kmemstatistics = type->ks_next;
444 	else {
445 		for (t = kmemstatistics; t->ks_next != NULL; t = t->ks_next) {
446 			if (t->ks_next == type) {
447 				t->ks_next = type->ks_next;
448 				break;
449 			}
450 		}
451 	}
452 	type->ks_next = NULL;
453 	mtx_destroy(&type->ks_mtx);
454 	mtx_unlock(&malloc_mtx);
455 }
456 
457 static int
458 sysctl_kern_malloc(SYSCTL_HANDLER_ARGS)
459 {
460 	struct malloc_type *type;
461 	int linesize = 128;
462 	int curline;
463 	int bufsize;
464 	int first;
465 	int error;
466 	char *buf;
467 	char *p;
468 	int cnt;
469 	int len;
470 	int i;
471 
472 	cnt = 0;
473 
474 	mtx_lock(&malloc_mtx);
475 	for (type = kmemstatistics; type != NULL; type = type->ks_next)
476 		cnt++;
477 
478 	mtx_unlock(&malloc_mtx);
479 	bufsize = linesize * (cnt + 1);
480 	p = buf = (char *)malloc(bufsize, M_TEMP, M_WAITOK|M_ZERO);
481 	mtx_lock(&malloc_mtx);
482 
483 	len = snprintf(p, linesize,
484 	    "\n        Type  InUse MemUse HighUse Requests  Size(s)\n");
485 	p += len;
486 
487 	for (type = kmemstatistics; cnt != 0 && type != NULL;
488 	    type = type->ks_next, cnt--) {
489 		if (type->ks_calls == 0)
490 			continue;
491 
492 		curline = linesize - 2;	/* Leave room for the \n */
493 		len = snprintf(p, curline, "%13s%6lu%6luK%7luK%9llu",
494 			type->ks_shortdesc,
495 			type->ks_inuse,
496 			(type->ks_memuse + 1023) / 1024,
497 			(type->ks_maxused + 1023) / 1024,
498 			(long long unsigned)type->ks_calls);
499 		curline -= len;
500 		p += len;
501 
502 		first = 1;
503 		for (i = 0; i < sizeof(kmemzones) / sizeof(kmemzones[0]) - 1;
504 		    i++) {
505 			if (type->ks_size & (1 << i)) {
506 				if (first)
507 					len = snprintf(p, curline, "  ");
508 				else
509 					len = snprintf(p, curline, ",");
510 				curline -= len;
511 				p += len;
512 
513 				len = snprintf(p, curline,
514 				    "%s", kmemzones[i].kz_name);
515 				curline -= len;
516 				p += len;
517 
518 				first = 0;
519 			}
520 		}
521 
522 		len = snprintf(p, 2, "\n");
523 		p += len;
524 	}
525 
526 	mtx_unlock(&malloc_mtx);
527 	error = SYSCTL_OUT(req, buf, p - buf);
528 
529 	free(buf, M_TEMP);
530 	return (error);
531 }
532 
533 SYSCTL_OID(_kern, OID_AUTO, malloc, CTLTYPE_STRING|CTLFLAG_RD,
534     NULL, 0, sysctl_kern_malloc, "A", "Malloc Stats");
535 
536 #ifdef MALLOC_PROFILE
537 
538 static int
539 sysctl_kern_mprof(SYSCTL_HANDLER_ARGS)
540 {
541 	int linesize = 64;
542 	uint64_t count;
543 	uint64_t waste;
544 	uint64_t mem;
545 	int bufsize;
546 	int error;
547 	char *buf;
548 	int rsize;
549 	int size;
550 	char *p;
551 	int len;
552 	int i;
553 
554 	bufsize = linesize * (KMEM_ZSIZE + 1);
555 	bufsize += 128; 	/* For the stats line */
556 	bufsize += 128; 	/* For the banner line */
557 	waste = 0;
558 	mem = 0;
559 
560 	p = buf = (char *)malloc(bufsize, M_TEMP, M_WAITOK|M_ZERO);
561 	len = snprintf(p, bufsize,
562 	    "\n  Size                    Requests  Real Size\n");
563 	bufsize -= len;
564 	p += len;
565 
566 	for (i = 0; i < KMEM_ZSIZE; i++) {
567 		size = i << KMEM_ZSHIFT;
568 		rsize = kmemzones[kmemsize[i]].kz_size;
569 		count = (long long unsigned)krequests[i];
570 
571 		len = snprintf(p, bufsize, "%6d%28llu%11d\n",
572 		    size, (unsigned long long)count, rsize);
573 		bufsize -= len;
574 		p += len;
575 
576 		if ((rsize * count) > (size * count))
577 			waste += (rsize * count) - (size * count);
578 		mem += (rsize * count);
579 	}
580 
581 	len = snprintf(p, bufsize,
582 	    "\nTotal memory used:\t%30llu\nTotal Memory wasted:\t%30llu\n",
583 	    (unsigned long long)mem, (unsigned long long)waste);
584 	p += len;
585 
586 	error = SYSCTL_OUT(req, buf, p - buf);
587 
588 	free(buf, M_TEMP);
589 	return (error);
590 }
591 
592 SYSCTL_OID(_kern, OID_AUTO, mprof, CTLTYPE_STRING|CTLFLAG_RD,
593     NULL, 0, sysctl_kern_mprof, "A", "Malloc Profiling");
594 #endif /* MALLOC_PROFILE */
595