xref: /netbsd-src/sys/kern/subr_kmem.c (revision 6a493d6bc668897c91594964a732d38505b70cbb)
1 /*	$NetBSD: subr_kmem.c,v 1.51 2013/10/25 16:09:29 martin Exp $	*/
2 
3 /*-
4  * Copyright (c) 2009 The NetBSD Foundation, Inc.
5  * All rights reserved.
6  *
7  * This code is derived from software contributed to The NetBSD Foundation
8  * by Andrew Doran.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
20  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
21  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
22  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
23  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
24  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
25  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
26  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
27  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
28  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
29  * POSSIBILITY OF SUCH DAMAGE.
30  */
31 
32 /*-
33  * Copyright (c)2006 YAMAMOTO Takashi,
34  * All rights reserved.
35  *
36  * Redistribution and use in source and binary forms, with or without
37  * modification, are permitted provided that the following conditions
38  * are met:
39  * 1. Redistributions of source code must retain the above copyright
40  *    notice, this list of conditions and the following disclaimer.
41  * 2. Redistributions in binary form must reproduce the above copyright
42  *    notice, this list of conditions and the following disclaimer in the
43  *    documentation and/or other materials provided with the distribution.
44  *
45  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
46  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
47  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
48  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
49  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
50  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
51  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
52  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
53  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
54  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
55  * SUCH DAMAGE.
56  */
57 
58 /*
59  * allocator of kernel wired memory.
60  */
61 
62 /*
63  * This allocator has some debug features enabled with "option DEBUG".
64  *
65  * KMEM_POISON
66  *	Try to detect modify-after-free bugs.
67  *
68  *	Fill freed (in the sense of kmem_free) memory with a garbage pattern.
69  *	Check the pattern on allocation.
70  *
71  * KMEM_REDZONE
72  *	Try to detect overrun bugs.
73  *
74  *	Allocate some more bytes for each allocation.
75  *	The extra bytes are checked by KMEM_POISON on kmem_free.
76  *
77  * KMEM_SIZE
78  *	Try to detect alloc/free size mismatch bugs.
79  *
80  *	Prefix each allocations with a fixed-sized header and record
81  *	the exact user-requested allocation size in it.
82  *	When freeing, compare it with kmem_free's "size" argument.
83  *
84  * KMEM_GUARD
85  *	See the below "kmguard" section.
86  */
87 
88 /*
89  * kmguard
90  *
91  * A kernel with "option DEBUG" has "kmguard" debugging feature compiled in.
92  * See the comment in uvm/uvm_kmguard.c for what kind of bugs it tries to
93  * detect.  Even if compiled in, it's disabled by default because it's very
94  * expensive.  You can enable it on boot by:
95  *
96  * 	boot -d
97  * 	db> w kmem_guard_depth 0t30000
98  * 	db> c
99  *
100  * The default value of kmem_guard_depth is 0, which means disabled.
101  * It can be changed by KMEM_GUARD_DEPTH kernel config option.
102  */
103 
104 #include <sys/cdefs.h>
105 __KERNEL_RCSID(0, "$NetBSD: subr_kmem.c,v 1.51 2013/10/25 16:09:29 martin Exp $");
106 
107 #include <sys/param.h>
108 #include <sys/callback.h>
109 #include <sys/kmem.h>
110 #include <sys/pool.h>
111 #include <sys/debug.h>
112 #include <sys/lockdebug.h>
113 #include <sys/cpu.h>
114 
115 #include <uvm/uvm_extern.h>
116 #include <uvm/uvm_map.h>
117 #include <uvm/uvm_kmguard.h>
118 
119 #include <lib/libkern/libkern.h>
120 
121 struct kmem_cache_info {
122 	size_t		kc_size;
123 	const char *	kc_name;
124 };
125 
126 static const struct kmem_cache_info kmem_cache_sizes[] = {
127 	{  8, "kmem-8" },
128 	{ 16, "kmem-16" },
129 	{ 24, "kmem-24" },
130 	{ 32, "kmem-32" },
131 	{ 40, "kmem-40" },
132 	{ 48, "kmem-48" },
133 	{ 56, "kmem-56" },
134 	{ 64, "kmem-64" },
135 	{ 80, "kmem-80" },
136 	{ 96, "kmem-96" },
137 	{ 112, "kmem-112" },
138 	{ 128, "kmem-128" },
139 	{ 160, "kmem-160" },
140 	{ 192, "kmem-192" },
141 	{ 224, "kmem-224" },
142 	{ 256, "kmem-256" },
143 	{ 320, "kmem-320" },
144 	{ 384, "kmem-384" },
145 	{ 448, "kmem-448" },
146 	{ 512, "kmem-512" },
147 	{ 768, "kmem-768" },
148 	{ 1024, "kmem-1024" },
149 	{ 0, NULL }
150 };
151 
152 static const struct kmem_cache_info kmem_cache_big_sizes[] = {
153 	{ 2048, "kmem-2048" },
154 	{ 4096, "kmem-4096" },
155 	{ 8192, "kmem-8192" },
156 	{ 16384, "kmem-16384" },
157 	{ 0, NULL }
158 };
159 
160 /*
161  * KMEM_ALIGN is the smallest guaranteed alignment and also the
162  * smallest allocateable quantum.
163  * Every cache size >= CACHE_LINE_SIZE gets CACHE_LINE_SIZE alignment.
164  */
165 #define	KMEM_ALIGN		8
166 #define	KMEM_SHIFT		3
167 #define	KMEM_MAXSIZE		1024
168 #define	KMEM_CACHE_COUNT	(KMEM_MAXSIZE >> KMEM_SHIFT)
169 
170 static pool_cache_t kmem_cache[KMEM_CACHE_COUNT] __cacheline_aligned;
171 static size_t kmem_cache_maxidx __read_mostly;
172 
173 #define	KMEM_BIG_ALIGN		2048
174 #define	KMEM_BIG_SHIFT		11
175 #define	KMEM_BIG_MAXSIZE	16384
176 #define	KMEM_CACHE_BIG_COUNT	(KMEM_BIG_MAXSIZE >> KMEM_BIG_SHIFT)
177 
178 static pool_cache_t kmem_cache_big[KMEM_CACHE_BIG_COUNT] __cacheline_aligned;
179 static size_t kmem_cache_big_maxidx __read_mostly;
180 
181 
182 #if defined(DEBUG) && defined(_HARDKERNEL)
183 #ifndef KMEM_GUARD_DEPTH
184 #define KMEM_GUARD_DEPTH 0
185 #endif
186 int kmem_guard_depth = KMEM_GUARD_DEPTH;
187 size_t kmem_guard_size;
188 static struct uvm_kmguard kmem_guard;
189 static void *kmem_freecheck;
190 #define	KMEM_POISON
191 #define	KMEM_REDZONE
192 #define	KMEM_SIZE
193 #define	KMEM_GUARD
194 #endif /* defined(DEBUG) */
195 
196 #if defined(KMEM_POISON)
197 static int kmem_poison_ctor(void *, void *, int);
198 static void kmem_poison_fill(void *, size_t);
199 static void kmem_poison_check(void *, size_t);
200 #else /* defined(KMEM_POISON) */
201 #define	kmem_poison_fill(p, sz)		/* nothing */
202 #define	kmem_poison_check(p, sz)	/* nothing */
203 #endif /* defined(KMEM_POISON) */
204 
205 #if defined(KMEM_REDZONE)
206 #define	REDZONE_SIZE	1
207 #else /* defined(KMEM_REDZONE) */
208 #define	REDZONE_SIZE	0
209 #endif /* defined(KMEM_REDZONE) */
210 
211 #if defined(KMEM_SIZE)
212 #define	SIZE_SIZE	(MAX(KMEM_ALIGN, sizeof(size_t)))
213 static void kmem_size_set(void *, size_t);
214 static void kmem_size_check(void *, size_t);
215 #else
216 #define	SIZE_SIZE	0
217 #define	kmem_size_set(p, sz)	/* nothing */
218 #define	kmem_size_check(p, sz)	/* nothing */
219 #endif
220 
221 CTASSERT(KM_SLEEP == PR_WAITOK);
222 CTASSERT(KM_NOSLEEP == PR_NOWAIT);
223 
224 /*
225  * kmem_intr_alloc: allocate wired memory.
226  */
227 
228 void *
229 kmem_intr_alloc(size_t requested_size, km_flag_t kmflags)
230 {
231 	size_t allocsz, index;
232 	size_t size;
233 	pool_cache_t pc;
234 	uint8_t *p;
235 
236 	KASSERT(requested_size > 0);
237 
238 #ifdef KMEM_GUARD
239 	if (requested_size <= kmem_guard_size) {
240 		return uvm_kmguard_alloc(&kmem_guard, requested_size,
241 		    (kmflags & KM_SLEEP) != 0);
242 	}
243 #endif
244 	size = kmem_roundup_size(requested_size);
245 	allocsz = size + REDZONE_SIZE + SIZE_SIZE;
246 
247 	if ((index = ((allocsz -1) >> KMEM_SHIFT))
248 	    < kmem_cache_maxidx) {
249 		pc = kmem_cache[index];
250 	} else if ((index = ((allocsz - 1) >> KMEM_BIG_SHIFT))
251             < kmem_cache_big_maxidx) {
252 		pc = kmem_cache_big[index];
253 	} else {
254 		int ret = uvm_km_kmem_alloc(kmem_va_arena,
255 		    (vsize_t)round_page(size),
256 		    ((kmflags & KM_SLEEP) ? VM_SLEEP : VM_NOSLEEP)
257 		     | VM_INSTANTFIT, (vmem_addr_t *)&p);
258 		if (ret) {
259 			return NULL;
260 		}
261 		FREECHECK_OUT(&kmem_freecheck, p);
262 		return p;
263 	}
264 
265 	p = pool_cache_get(pc, kmflags);
266 
267 	if (__predict_true(p != NULL)) {
268 		kmem_poison_check(p, size);
269 		FREECHECK_OUT(&kmem_freecheck, p);
270 		kmem_size_set(p, requested_size);
271 
272 		return p + SIZE_SIZE;
273 	}
274 	return p;
275 }
276 
277 /*
278  * kmem_intr_zalloc: allocate zeroed wired memory.
279  */
280 
281 void *
282 kmem_intr_zalloc(size_t size, km_flag_t kmflags)
283 {
284 	void *p;
285 
286 	p = kmem_intr_alloc(size, kmflags);
287 	if (p != NULL) {
288 		memset(p, 0, size);
289 	}
290 	return p;
291 }
292 
293 /*
294  * kmem_intr_free: free wired memory allocated by kmem_alloc.
295  */
296 
297 void
298 kmem_intr_free(void *p, size_t requested_size)
299 {
300 	size_t allocsz, index;
301 	size_t size;
302 	pool_cache_t pc;
303 
304 	KASSERT(p != NULL);
305 	KASSERT(requested_size > 0);
306 
307 #ifdef KMEM_GUARD
308 	if (requested_size <= kmem_guard_size) {
309 		uvm_kmguard_free(&kmem_guard, requested_size, p);
310 		return;
311 	}
312 #endif
313 	size = kmem_roundup_size(requested_size);
314 	allocsz = size + REDZONE_SIZE + SIZE_SIZE;
315 
316 	if ((index = ((allocsz -1) >> KMEM_SHIFT))
317 	    < kmem_cache_maxidx) {
318 		pc = kmem_cache[index];
319 	} else if ((index = ((allocsz - 1) >> KMEM_BIG_SHIFT))
320             < kmem_cache_big_maxidx) {
321 		pc = kmem_cache_big[index];
322 	} else {
323 		FREECHECK_IN(&kmem_freecheck, p);
324 		uvm_km_kmem_free(kmem_va_arena, (vaddr_t)p,
325 		    round_page(size));
326 		return;
327 	}
328 
329 	p = (uint8_t *)p - SIZE_SIZE;
330 	kmem_size_check(p, requested_size);
331 	FREECHECK_IN(&kmem_freecheck, p);
332 	LOCKDEBUG_MEM_CHECK(p, size);
333 	kmem_poison_check((uint8_t *)p + SIZE_SIZE + size,
334       	    allocsz - (SIZE_SIZE + size));
335 	kmem_poison_fill(p, allocsz);
336 
337 	pool_cache_put(pc, p);
338 }
339 
340 /* ---- kmem API */
341 
342 /*
343  * kmem_alloc: allocate wired memory.
344  * => must not be called from interrupt context.
345  */
346 
347 void *
348 kmem_alloc(size_t size, km_flag_t kmflags)
349 {
350 
351 	KASSERTMSG((!cpu_intr_p() && !cpu_softintr_p()),
352 	    "kmem(9) should not be used from the interrupt context");
353 	return kmem_intr_alloc(size, kmflags);
354 }
355 
356 /*
357  * kmem_zalloc: allocate zeroed wired memory.
358  * => must not be called from interrupt context.
359  */
360 
361 void *
362 kmem_zalloc(size_t size, km_flag_t kmflags)
363 {
364 
365 	KASSERTMSG((!cpu_intr_p() && !cpu_softintr_p()),
366 	    "kmem(9) should not be used from the interrupt context");
367 	return kmem_intr_zalloc(size, kmflags);
368 }
369 
370 /*
371  * kmem_free: free wired memory allocated by kmem_alloc.
372  * => must not be called from interrupt context.
373  */
374 
375 void
376 kmem_free(void *p, size_t size)
377 {
378 
379 	KASSERT(!cpu_intr_p());
380 	KASSERT(!cpu_softintr_p());
381 	kmem_intr_free(p, size);
382 }
383 
384 static size_t
385 kmem_create_caches(const struct kmem_cache_info *array,
386     pool_cache_t alloc_table[], size_t maxsize, int shift, int ipl)
387 {
388 	size_t maxidx = 0;
389 	size_t table_unit = (1 << shift);
390 	size_t size = table_unit;
391 	int i;
392 
393 	for (i = 0; array[i].kc_size != 0 ; i++) {
394 		const char *name = array[i].kc_name;
395 		size_t cache_size = array[i].kc_size;
396 		struct pool_allocator *pa;
397 		int flags = PR_NOALIGN;
398 		pool_cache_t pc;
399 		size_t align;
400 
401 		if ((cache_size & (CACHE_LINE_SIZE - 1)) == 0)
402 			align = CACHE_LINE_SIZE;
403 		else if ((cache_size & (PAGE_SIZE - 1)) == 0)
404 			align = PAGE_SIZE;
405 		else
406 			align = KMEM_ALIGN;
407 
408 		if (cache_size < CACHE_LINE_SIZE)
409 			flags |= PR_NOTOUCH;
410 
411 		/* check if we reached the requested size */
412 		if (cache_size > maxsize || cache_size > PAGE_SIZE) {
413 			break;
414 		}
415 		if ((cache_size >> shift) > maxidx) {
416 			maxidx = cache_size >> shift;
417 		}
418 
419 		if ((cache_size >> shift) > maxidx) {
420 			maxidx = cache_size >> shift;
421 		}
422 
423 		pa = &pool_allocator_kmem;
424 #if defined(KMEM_POISON)
425 		pc = pool_cache_init(cache_size, align, 0, flags,
426 		    name, pa, ipl, kmem_poison_ctor,
427 		    NULL, (void *)cache_size);
428 #else /* defined(KMEM_POISON) */
429 		pc = pool_cache_init(cache_size, align, 0, flags,
430 		    name, pa, ipl, NULL, NULL, NULL);
431 #endif /* defined(KMEM_POISON) */
432 
433 		while (size <= cache_size) {
434 			alloc_table[(size - 1) >> shift] = pc;
435 			size += table_unit;
436 		}
437 	}
438 	return maxidx;
439 }
440 
441 void
442 kmem_init(void)
443 {
444 
445 #ifdef KMEM_GUARD
446 	uvm_kmguard_init(&kmem_guard, &kmem_guard_depth, &kmem_guard_size,
447 	    kmem_va_arena);
448 #endif
449 	kmem_cache_maxidx = kmem_create_caches(kmem_cache_sizes,
450 	    kmem_cache, KMEM_MAXSIZE, KMEM_SHIFT, IPL_VM);
451        	kmem_cache_big_maxidx = kmem_create_caches(kmem_cache_big_sizes,
452 	    kmem_cache_big, PAGE_SIZE, KMEM_BIG_SHIFT, IPL_VM);
453 }
454 
455 size_t
456 kmem_roundup_size(size_t size)
457 {
458 
459 	return (size + (KMEM_ALIGN - 1)) & ~(KMEM_ALIGN - 1);
460 }
461 
462 /* ---- debug */
463 
464 #if defined(KMEM_POISON)
465 
466 #if defined(_LP64)
467 #define PRIME 0x9e37fffffffc0000UL
468 #else /* defined(_LP64) */
469 #define PRIME 0x9e3779b1
470 #endif /* defined(_LP64) */
471 
472 static inline uint8_t
473 kmem_poison_pattern(const void *p)
474 {
475 
476 	return (uint8_t)(((uintptr_t)p) * PRIME
477 	   >> ((sizeof(uintptr_t) - sizeof(uint8_t))) * CHAR_BIT);
478 }
479 
480 static int
481 kmem_poison_ctor(void *arg, void *obj, int flag)
482 {
483 	size_t sz = (size_t)arg;
484 
485 	kmem_poison_fill(obj, sz);
486 
487 	return 0;
488 }
489 
490 static void
491 kmem_poison_fill(void *p, size_t sz)
492 {
493 	uint8_t *cp;
494 	const uint8_t *ep;
495 
496 	cp = p;
497 	ep = cp + sz;
498 	while (cp < ep) {
499 		*cp = kmem_poison_pattern(cp);
500 		cp++;
501 	}
502 }
503 
504 static void
505 kmem_poison_check(void *p, size_t sz)
506 {
507 	uint8_t *cp;
508 	const uint8_t *ep;
509 
510 	cp = p;
511 	ep = cp + sz;
512 	while (cp < ep) {
513 		const uint8_t expected = kmem_poison_pattern(cp);
514 
515 		if (*cp != expected) {
516 			panic("%s: %p: 0x%02x != 0x%02x\n",
517 			   __func__, cp, *cp, expected);
518 		}
519 		cp++;
520 	}
521 }
522 
523 #endif /* defined(KMEM_POISON) */
524 
525 #if defined(KMEM_SIZE)
526 static void
527 kmem_size_set(void *p, size_t sz)
528 {
529 
530 	memcpy(p, &sz, sizeof(sz));
531 }
532 
533 static void
534 kmem_size_check(void *p, size_t sz)
535 {
536 	size_t psz;
537 
538 	memcpy(&psz, p, sizeof(psz));
539 	if (psz != sz) {
540 		panic("kmem_free(%p, %zu) != allocated size %zu",
541 		    (const uint8_t *)p + SIZE_SIZE, sz, psz);
542 	}
543 }
544 #endif	/* defined(KMEM_SIZE) */
545 
546 /*
547  * Used to dynamically allocate string with kmem accordingly to format.
548  */
549 char *
550 kmem_asprintf(const char *fmt, ...)
551 {
552 	int size __diagused, len;
553 	va_list va;
554 	char *str;
555 
556 	va_start(va, fmt);
557 	len = vsnprintf(NULL, 0, fmt, va);
558 	va_end(va);
559 
560 	str = kmem_alloc(len + 1, KM_SLEEP);
561 
562 	va_start(va, fmt);
563 	size = vsnprintf(str, len + 1, fmt, va);
564 	va_end(va);
565 
566 	KASSERT(size == len);
567 
568 	return str;
569 }
570