xref: /freebsd/sys/kern/kern_malloc.c (revision 2ad872c5794e4c26fdf6ed219ad3f09ca0d5304a)
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  * $Id: kern_malloc.c,v 1.50 1999/01/08 17:31:09 eivind Exp $
35  */
36 
37 #include "opt_vm.h"
38 
39 #include <sys/param.h>
40 #include <sys/systm.h>
41 #include <sys/kernel.h>
42 #define MALLOC_INSTANTIATE
43 #include <sys/malloc.h>
44 #include <sys/mbuf.h>
45 #include <sys/vmmeter.h>
46 #include <sys/lock.h>
47 
48 #include <vm/vm.h>
49 #include <vm/vm_param.h>
50 #include <vm/vm_kern.h>
51 #include <vm/vm_extern.h>
52 #include <vm/pmap.h>
53 #include <vm/vm_map.h>
54 
55 static void kmeminit __P((void *));
56 SYSINIT(kmem, SI_SUB_KMEM, SI_ORDER_FIRST, kmeminit, NULL)
57 
58 static MALLOC_DEFINE(M_FREE, "free", "should be on free list");
59 
60 static struct malloc_type *kmemstatistics;
61 static struct kmembuckets bucket[MINBUCKET + 16];
62 static struct kmemusage *kmemusage;
63 static char *kmembase;
64 static char *kmemlimit;
65 static int vm_kmem_size;
66 
67 #ifdef INVARIANTS
68 /*
69  * This structure provides a set of masks to catch unaligned frees.
70  */
71 static long addrmask[] = { 0,
72 	0x00000001, 0x00000003, 0x00000007, 0x0000000f,
73 	0x0000001f, 0x0000003f, 0x0000007f, 0x000000ff,
74 	0x000001ff, 0x000003ff, 0x000007ff, 0x00000fff,
75 	0x00001fff, 0x00003fff, 0x00007fff, 0x0000ffff,
76 };
77 
78 /*
79  * The WEIRD_ADDR is used as known text to copy into free objects so
80  * that modifications after frees can be detected.
81  */
82 #define WEIRD_ADDR	0xdeadc0de
83 #define MAX_COPY	64
84 
85 /*
86  * Normally the first word of the structure is used to hold the list
87  * pointer for free objects. However, when running with diagnostics,
88  * we use the third and fourth fields, so as to catch modifications
89  * in the most commonly trashed first two words.
90  */
91 struct freelist {
92 	long	spare0;
93 	struct malloc_type *type;
94 	long	spare1;
95 	caddr_t	next;
96 };
97 #else /* !INVARIANTS */
98 struct freelist {
99 	caddr_t	next;
100 };
101 #endif /* INVARIANTS */
102 
103 /*
104  * Allocate a block of memory
105  */
106 void *
107 malloc(size, type, flags)
108 	unsigned long size;
109 	struct malloc_type *type;
110 	int flags;
111 {
112 	register struct kmembuckets *kbp;
113 	register struct kmemusage *kup;
114 	register struct freelist *freep;
115 	long indx, npg, allocsize;
116 	int s;
117 	caddr_t va, cp, savedlist;
118 #ifdef INVARIANTS
119 	long *end, *lp;
120 	int copysize;
121 	char *savedtype;
122 #endif
123 	register struct malloc_type *ksp = type;
124 
125 	if (!type->ks_next)
126 		malloc_init(type);
127 
128 	indx = BUCKETINDX(size);
129 	kbp = &bucket[indx];
130 	s = splmem();
131 	while (ksp->ks_memuse >= ksp->ks_limit) {
132 		if (flags & M_NOWAIT) {
133 			splx(s);
134 			return ((void *) NULL);
135 		}
136 		if (ksp->ks_limblocks < 65535)
137 			ksp->ks_limblocks++;
138 		tsleep((caddr_t)ksp, PSWP+2, type->ks_shortdesc, 0);
139 	}
140 	ksp->ks_size |= 1 << indx;
141 #ifdef INVARIANTS
142 	copysize = 1 << indx < MAX_COPY ? 1 << indx : MAX_COPY;
143 #endif
144 	if (kbp->kb_next == NULL) {
145 		kbp->kb_last = NULL;
146 		if (size > MAXALLOCSAVE)
147 			allocsize = roundup(size, PAGE_SIZE);
148 		else
149 			allocsize = 1 << indx;
150 		npg = btoc(allocsize);
151 		va = (caddr_t) kmem_malloc(kmem_map, (vm_size_t)ctob(npg), flags);
152 		if (va == NULL) {
153 			splx(s);
154 			return ((void *) NULL);
155 		}
156 		kbp->kb_total += kbp->kb_elmpercl;
157 		kup = btokup(va);
158 		kup->ku_indx = indx;
159 		if (allocsize > MAXALLOCSAVE) {
160 			if (npg > 65535)
161 				panic("malloc: allocation too large");
162 			kup->ku_pagecnt = npg;
163 			ksp->ks_memuse += allocsize;
164 			goto out;
165 		}
166 		kup->ku_freecnt = kbp->kb_elmpercl;
167 		kbp->kb_totalfree += kbp->kb_elmpercl;
168 		/*
169 		 * Just in case we blocked while allocating memory,
170 		 * and someone else also allocated memory for this
171 		 * bucket, don't assume the list is still empty.
172 		 */
173 		savedlist = kbp->kb_next;
174 		kbp->kb_next = cp = va + (npg * PAGE_SIZE) - allocsize;
175 		for (;;) {
176 			freep = (struct freelist *)cp;
177 #ifdef INVARIANTS
178 			/*
179 			 * Copy in known text to detect modification
180 			 * after freeing.
181 			 */
182 			end = (long *)&cp[copysize];
183 			for (lp = (long *)cp; lp < end; lp++)
184 				*lp = WEIRD_ADDR;
185 			freep->type = M_FREE;
186 #endif /* INVARIANTS */
187 			if (cp <= va)
188 				break;
189 			cp -= allocsize;
190 			freep->next = cp;
191 		}
192 		freep->next = savedlist;
193 		if (kbp->kb_last == NULL)
194 			kbp->kb_last = (caddr_t)freep;
195 	}
196 	va = kbp->kb_next;
197 	kbp->kb_next = ((struct freelist *)va)->next;
198 #ifdef INVARIANTS
199 	freep = (struct freelist *)va;
200 	savedtype = (char *) type->ks_shortdesc;
201 #if BYTE_ORDER == BIG_ENDIAN
202 	freep->type = (struct malloc_type *)WEIRD_ADDR >> 16;
203 #endif
204 #if BYTE_ORDER == LITTLE_ENDIAN
205 	freep->type = (struct malloc_type *)WEIRD_ADDR;
206 #endif
207 	if ((intptr_t)(void *)&freep->next & 0x2)
208 		freep->next = (caddr_t)((WEIRD_ADDR >> 16)|(WEIRD_ADDR << 16));
209 	else
210 		freep->next = (caddr_t)WEIRD_ADDR;
211 	end = (long *)&va[copysize];
212 	for (lp = (long *)va; lp < end; lp++) {
213 		if (*lp == WEIRD_ADDR)
214 			continue;
215 		printf("%s %ld of object %p size %lu %s %s (0x%lx != 0x%lx)\n",
216 			"Data modified on freelist: word",
217 			(long)(lp - (long *)va), (void *)va, size,
218 			"previous type", savedtype, *lp, (u_long)WEIRD_ADDR);
219 		break;
220 	}
221 	freep->spare0 = 0;
222 #endif /* INVARIANTS */
223 	kup = btokup(va);
224 	if (kup->ku_indx != indx)
225 		panic("malloc: wrong bucket");
226 	if (kup->ku_freecnt == 0)
227 		panic("malloc: lost data");
228 	kup->ku_freecnt--;
229 	kbp->kb_totalfree--;
230 	ksp->ks_memuse += 1 << indx;
231 out:
232 	kbp->kb_calls++;
233 	ksp->ks_inuse++;
234 	ksp->ks_calls++;
235 	if (ksp->ks_memuse > ksp->ks_maxused)
236 		ksp->ks_maxused = ksp->ks_memuse;
237 	splx(s);
238 	return ((void *) va);
239 }
240 
241 /*
242  * Free a block of memory allocated by malloc.
243  */
244 void
245 free(addr, type)
246 	void *addr;
247 	struct malloc_type *type;
248 {
249 	register struct kmembuckets *kbp;
250 	register struct kmemusage *kup;
251 	register struct freelist *freep;
252 	long size;
253 	int s;
254 #ifdef INVARIANTS
255 	struct freelist *fp;
256 	long *end, *lp, alloc, copysize;
257 #endif
258 	register struct malloc_type *ksp = type;
259 
260 	if (!type->ks_next)
261 		panic("freeing with unknown type (%s)", type->ks_shortdesc);
262 
263 	KASSERT(kmembase <= (char *)addr && (char *)addr < kmemlimit,
264 	    ("free: address %p out of range", (void *)addr));
265 	kup = btokup(addr);
266 	size = 1 << kup->ku_indx;
267 	kbp = &bucket[kup->ku_indx];
268 	s = splmem();
269 #ifdef INVARIANTS
270 	/*
271 	 * Check for returns of data that do not point to the
272 	 * beginning of the allocation.
273 	 */
274 	if (size > PAGE_SIZE)
275 		alloc = addrmask[BUCKETINDX(PAGE_SIZE)];
276 	else
277 		alloc = addrmask[kup->ku_indx];
278 	if (((uintptr_t)(void *)addr & alloc) != 0)
279 		panic("free: unaligned addr %p, size %ld, type %s, mask %ld",
280 		    (void *)addr, size, type->ks_shortdesc, alloc);
281 #endif /* INVARIANTS */
282 	if (size > MAXALLOCSAVE) {
283 		kmem_free(kmem_map, (vm_offset_t)addr, ctob(kup->ku_pagecnt));
284 		size = kup->ku_pagecnt << PAGE_SHIFT;
285 		ksp->ks_memuse -= size;
286 		kup->ku_indx = 0;
287 		kup->ku_pagecnt = 0;
288 		if (ksp->ks_memuse + size >= ksp->ks_limit &&
289 		    ksp->ks_memuse < ksp->ks_limit)
290 			wakeup((caddr_t)ksp);
291 		ksp->ks_inuse--;
292 		kbp->kb_total -= 1;
293 		splx(s);
294 		return;
295 	}
296 	freep = (struct freelist *)addr;
297 #ifdef INVARIANTS
298 	/*
299 	 * Check for multiple frees. Use a quick check to see if
300 	 * it looks free before laboriously searching the freelist.
301 	 */
302 	if (freep->spare0 == WEIRD_ADDR) {
303 		fp = (struct freelist *)kbp->kb_next;
304 		while (fp) {
305 			if (fp->spare0 != WEIRD_ADDR)
306 				panic("free: free item %p modified", fp);
307 			else if (addr == (caddr_t)fp)
308 				panic("free: multiple freed item %p", addr);
309 			fp = (struct freelist *)fp->next;
310 		}
311 	}
312 	/*
313 	 * Copy in known text to detect modification after freeing
314 	 * and to make it look free. Also, save the type being freed
315 	 * so we can list likely culprit if modification is detected
316 	 * when the object is reallocated.
317 	 */
318 	copysize = size < MAX_COPY ? size : MAX_COPY;
319 	end = (long *)&((caddr_t)addr)[copysize];
320 	for (lp = (long *)addr; lp < end; lp++)
321 		*lp = WEIRD_ADDR;
322 	freep->type = type;
323 #endif /* INVARIANTS */
324 	kup->ku_freecnt++;
325 	if (kup->ku_freecnt >= kbp->kb_elmpercl)
326 		if (kup->ku_freecnt > kbp->kb_elmpercl)
327 			panic("free: multiple frees");
328 		else if (kbp->kb_totalfree > kbp->kb_highwat)
329 			kbp->kb_couldfree++;
330 	kbp->kb_totalfree++;
331 	ksp->ks_memuse -= size;
332 	if (ksp->ks_memuse + size >= ksp->ks_limit &&
333 	    ksp->ks_memuse < ksp->ks_limit)
334 		wakeup((caddr_t)ksp);
335 	ksp->ks_inuse--;
336 #ifdef OLD_MALLOC_MEMORY_POLICY
337 	if (kbp->kb_next == NULL)
338 		kbp->kb_next = addr;
339 	else
340 		((struct freelist *)kbp->kb_last)->next = addr;
341 	freep->next = NULL;
342 	kbp->kb_last = addr;
343 #else
344 	/*
345 	 * Return memory to the head of the queue for quick reuse.  This
346 	 * can improve performance by improving the probability of the
347 	 * item being in the cache when it is reused.
348 	 */
349 	if (kbp->kb_next == NULL) {
350 		kbp->kb_next = addr;
351 		kbp->kb_last = addr;
352 		freep->next = NULL;
353 	} else {
354 		freep->next = kbp->kb_next;
355 		kbp->kb_next = addr;
356 	}
357 #endif
358 	splx(s);
359 }
360 
361 /*
362  * Initialize the kernel memory allocator
363  */
364 /* ARGSUSED*/
365 static void
366 kmeminit(dummy)
367 	void *dummy;
368 {
369 	register long indx;
370 	int npg;
371 	int mem_size;
372 
373 #if	((MAXALLOCSAVE & (MAXALLOCSAVE - 1)) != 0)
374 #error "kmeminit: MAXALLOCSAVE not power of 2"
375 #endif
376 #if	(MAXALLOCSAVE > MINALLOCSIZE * 32768)
377 #error "kmeminit: MAXALLOCSAVE too big"
378 #endif
379 #if	(MAXALLOCSAVE < PAGE_SIZE)
380 #error "kmeminit: MAXALLOCSAVE too small"
381 #endif
382 
383 	/*
384 	 * Try to auto-tune the kernel memory size, so that it is
385 	 * more applicable for a wider range of machine sizes.
386 	 * On an X86, a VM_KMEM_SIZE_SCALE value of 4 is good, while
387 	 * a VM_KMEM_SIZE of 12MB is a fair compromise.  The
388 	 * VM_KMEM_SIZE_MAX is dependent on the maximum KVA space
389 	 * available, and on an X86 with a total KVA space of 256MB,
390 	 * try to keep VM_KMEM_SIZE_MAX at 80MB or below.
391 	 *
392 	 * Note that the kmem_map is also used by the zone allocator,
393 	 * so make sure that there is enough space.
394 	 */
395 	vm_kmem_size = VM_KMEM_SIZE;
396 	mem_size = cnt.v_page_count * PAGE_SIZE;
397 
398 #if defined(VM_KMEM_SIZE_SCALE)
399 	if ((mem_size / VM_KMEM_SIZE_SCALE) > vm_kmem_size)
400 		vm_kmem_size = mem_size / VM_KMEM_SIZE_SCALE;
401 #endif
402 
403 #if defined(VM_KMEM_SIZE_MAX)
404 	if (vm_kmem_size >= VM_KMEM_SIZE_MAX)
405 		vm_kmem_size = VM_KMEM_SIZE_MAX;
406 #endif
407 
408 	if (vm_kmem_size > 2 * (cnt.v_page_count * PAGE_SIZE))
409 		vm_kmem_size = 2 * (cnt.v_page_count * PAGE_SIZE);
410 
411 	npg = (nmbufs * MSIZE + nmbclusters * MCLBYTES + vm_kmem_size)
412 		/ PAGE_SIZE;
413 
414 	kmemusage = (struct kmemusage *) kmem_alloc(kernel_map,
415 		(vm_size_t)(npg * sizeof(struct kmemusage)));
416 	kmem_map = kmem_suballoc(kernel_map, (vm_offset_t *)&kmembase,
417 		(vm_offset_t *)&kmemlimit, (vm_size_t)(npg * PAGE_SIZE));
418 	kmem_map->system_map = 1;
419 	for (indx = 0; indx < MINBUCKET + 16; indx++) {
420 		if (1 << indx >= PAGE_SIZE)
421 			bucket[indx].kb_elmpercl = 1;
422 		else
423 			bucket[indx].kb_elmpercl = PAGE_SIZE / (1 << indx);
424 		bucket[indx].kb_highwat = 5 * bucket[indx].kb_elmpercl;
425 	}
426 }
427 
428 void
429 malloc_init(data)
430 	void *data;
431 {
432 	struct malloc_type *type = (struct malloc_type *)data;
433 
434 	if (type->ks_magic != M_MAGIC)
435 		panic("malloc type lacks magic");
436 
437 	if (type->ks_next)
438 		return;
439 
440 	if (cnt.v_page_count == 0)
441 		panic("malloc_init not allowed before vm init");
442 
443 	/*
444 	 * The default limits for each malloc region is 1/2 of the
445 	 * malloc portion of the kmem map size.
446 	 */
447 	type->ks_limit = vm_kmem_size / 2;
448 	type->ks_next = kmemstatistics;
449 	kmemstatistics = type;
450 }
451 
452 void
453 malloc_uninit(data)
454 	void *data;
455 {
456 	struct malloc_type *type = (struct malloc_type *)data;
457 	struct malloc_type *t;
458 
459 	if (type->ks_magic != M_MAGIC)
460 		panic("malloc type lacks magic");
461 
462 	if (cnt.v_page_count == 0)
463 		panic("malloc_uninit not allowed before vm init");
464 
465 	if (type == kmemstatistics)
466 		kmemstatistics = type->ks_next;
467 	else {
468 		for (t = kmemstatistics; t->ks_next != NULL; t = t->ks_next) {
469 			if (t->ks_next == type) {
470 				t->ks_next = type->ks_next;
471 				break;
472 			}
473 		}
474 	}
475 }
476