Annotation of src/sys/kern/subr_vmem.c, Revision 1.21
1.21 ! yamt 1: /* $NetBSD: subr_vmem.c,v 1.20 2006/11/09 10:08:53 yamt Exp $ */
1.1 yamt 2:
3: /*-
4: * Copyright (c)2006 YAMAMOTO Takashi,
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: *
16: * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
17: * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18: * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19: * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
20: * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21: * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22: * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23: * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24: * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25: * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26: * SUCH DAMAGE.
27: */
28:
29: /*
30: * reference:
31: * - Magazines and Vmem: Extending the Slab Allocator
32: * to Many CPUs and Arbitrary Resources
33: * http://www.usenix.org/event/usenix01/bonwick.html
1.18 yamt 34: *
35: * todo:
36: * - decide how to import segments for vmem_xalloc.
37: * - don't rely on malloc(9).
1.1 yamt 38: */
39:
40: #include <sys/cdefs.h>
1.21 ! yamt 41: __KERNEL_RCSID(0, "$NetBSD: subr_vmem.c,v 1.20 2006/11/09 10:08:53 yamt Exp $");
1.1 yamt 42:
43: #define VMEM_DEBUG
1.5 yamt 44: #if defined(_KERNEL)
45: #define QCACHE
46: #endif /* defined(_KERNEL) */
1.1 yamt 47:
48: #include <sys/param.h>
49: #include <sys/hash.h>
50: #include <sys/queue.h>
51:
52: #if defined(_KERNEL)
53: #include <sys/systm.h>
54: #include <sys/lock.h>
55: #include <sys/malloc.h>
56: #include <sys/once.h>
57: #include <sys/pool.h>
1.3 yamt 58: #include <sys/proc.h>
1.1 yamt 59: #include <sys/vmem.h>
60: #else /* defined(_KERNEL) */
61: #include "../sys/vmem.h"
62: #endif /* defined(_KERNEL) */
63:
64: #if defined(_KERNEL)
65: #define SIMPLELOCK_DECL(name) struct simplelock name
66: #else /* defined(_KERNEL) */
67: #include <errno.h>
68: #include <assert.h>
69: #include <stdlib.h>
70:
71: #define KASSERT(a) assert(a)
72: #define SIMPLELOCK_DECL(name) /* nothing */
73: #define LOCK_ASSERT(a) /* nothing */
74: #define simple_lock_init(a) /* nothing */
75: #define simple_lock(a) /* nothing */
76: #define simple_unlock(a) /* nothing */
1.3 yamt 77: #define ASSERT_SLEEPABLE(lk, msg) /* nothing */
1.1 yamt 78: #endif /* defined(_KERNEL) */
79:
80: struct vmem;
81: struct vmem_btag;
82:
83: #if defined(VMEM_DEBUG)
84: void vmem_dump(const vmem_t *);
85: #endif /* defined(VMEM_DEBUG) */
86:
1.4 yamt 87: #define VMEM_MAXORDER (sizeof(vmem_size_t) * CHAR_BIT)
1.1 yamt 88: #define VMEM_HASHSIZE_INIT 4096 /* XXX */
89:
90: #define VM_FITMASK (VM_BESTFIT | VM_INSTANTFIT)
91:
92: CIRCLEQ_HEAD(vmem_seglist, vmem_btag);
93: LIST_HEAD(vmem_freelist, vmem_btag);
94: LIST_HEAD(vmem_hashlist, vmem_btag);
95:
1.5 yamt 96: #if defined(QCACHE)
97: #define VMEM_QCACHE_IDX_MAX 32
98:
99: #define QC_NAME_MAX 16
100:
101: struct qcache {
102: struct pool qc_pool;
103: struct pool_cache qc_cache;
104: vmem_t *qc_vmem;
105: char qc_name[QC_NAME_MAX];
106: };
107: typedef struct qcache qcache_t;
108: #define QC_POOL_TO_QCACHE(pool) ((qcache_t *)(pool))
109: #endif /* defined(QCACHE) */
110:
1.1 yamt 111: /* vmem arena */
112: struct vmem {
113: SIMPLELOCK_DECL(vm_lock);
114: vmem_addr_t (*vm_allocfn)(vmem_t *, vmem_size_t, vmem_size_t *,
115: vm_flag_t);
116: void (*vm_freefn)(vmem_t *, vmem_addr_t, vmem_size_t);
117: vmem_t *vm_source;
118: struct vmem_seglist vm_seglist;
119: struct vmem_freelist vm_freelist[VMEM_MAXORDER];
120: size_t vm_hashsize;
121: size_t vm_nbusytag;
122: struct vmem_hashlist *vm_hashlist;
123: size_t vm_quantum_mask;
124: int vm_quantum_shift;
125: const char *vm_name;
1.5 yamt 126:
127: #if defined(QCACHE)
128: /* quantum cache */
129: size_t vm_qcache_max;
130: struct pool_allocator vm_qcache_allocator;
131: qcache_t vm_qcache[VMEM_QCACHE_IDX_MAX];
132: #endif /* defined(QCACHE) */
1.1 yamt 133: };
134:
135: #define VMEM_LOCK(vm) simple_lock(&vm->vm_lock)
136: #define VMEM_UNLOCK(vm) simple_unlock(&vm->vm_lock)
137: #define VMEM_LOCK_INIT(vm) simple_lock_init(&vm->vm_lock);
138: #define VMEM_ASSERT_LOCKED(vm) \
139: LOCK_ASSERT(simple_lock_held(&vm->vm_lock))
140: #define VMEM_ASSERT_UNLOCKED(vm) \
141: LOCK_ASSERT(!simple_lock_held(&vm->vm_lock))
142:
143: /* boundary tag */
144: struct vmem_btag {
145: CIRCLEQ_ENTRY(vmem_btag) bt_seglist;
146: union {
147: LIST_ENTRY(vmem_btag) u_freelist; /* BT_TYPE_FREE */
148: LIST_ENTRY(vmem_btag) u_hashlist; /* BT_TYPE_BUSY */
149: } bt_u;
150: #define bt_hashlist bt_u.u_hashlist
151: #define bt_freelist bt_u.u_freelist
152: vmem_addr_t bt_start;
153: vmem_size_t bt_size;
154: int bt_type;
155: };
156:
157: #define BT_TYPE_SPAN 1
158: #define BT_TYPE_SPAN_STATIC 2
159: #define BT_TYPE_FREE 3
160: #define BT_TYPE_BUSY 4
161: #define BT_ISSPAN_P(bt) ((bt)->bt_type <= BT_TYPE_SPAN_STATIC)
162:
163: #define BT_END(bt) ((bt)->bt_start + (bt)->bt_size)
164:
165: typedef struct vmem_btag bt_t;
166:
167: /* ---- misc */
168:
1.19 yamt 169: #define VMEM_ALIGNUP(addr, align) \
170: (-(-(addr) & -(align)))
171: #define VMEM_CROSS_P(addr1, addr2, boundary) \
172: ((((addr1) ^ (addr2)) & -(boundary)) != 0)
173:
1.4 yamt 174: #define ORDER2SIZE(order) ((vmem_size_t)1 << (order))
175:
1.1 yamt 176: static int
177: calc_order(vmem_size_t size)
178: {
1.4 yamt 179: vmem_size_t target;
1.1 yamt 180: int i;
181:
182: KASSERT(size != 0);
183:
184: i = 0;
1.4 yamt 185: target = size >> 1;
186: while (ORDER2SIZE(i) <= target) {
1.1 yamt 187: i++;
188: }
189:
1.4 yamt 190: KASSERT(ORDER2SIZE(i) <= size);
191: KASSERT(size < ORDER2SIZE(i + 1) || ORDER2SIZE(i + 1) < ORDER2SIZE(i));
1.1 yamt 192:
193: return i;
194: }
195:
196: #if defined(_KERNEL)
197: static MALLOC_DEFINE(M_VMEM, "vmem", "vmem");
198: #endif /* defined(_KERNEL) */
199:
200: static void *
201: xmalloc(size_t sz, vm_flag_t flags)
202: {
203:
204: #if defined(_KERNEL)
205: return malloc(sz, M_VMEM,
206: M_CANFAIL | ((flags & VM_SLEEP) ? M_WAITOK : M_NOWAIT));
207: #else /* defined(_KERNEL) */
208: return malloc(sz);
209: #endif /* defined(_KERNEL) */
210: }
211:
212: static void
213: xfree(void *p)
214: {
215:
216: #if defined(_KERNEL)
217: return free(p, M_VMEM);
218: #else /* defined(_KERNEL) */
219: return free(p);
220: #endif /* defined(_KERNEL) */
221: }
222:
223: /* ---- boundary tag */
224:
225: #if defined(_KERNEL)
226: static struct pool_cache bt_poolcache;
227: static POOL_INIT(bt_pool, sizeof(bt_t), 0, 0, 0, "vmembtpl", NULL);
228: #endif /* defined(_KERNEL) */
229:
230: static bt_t *
1.17 yamt 231: bt_alloc(vmem_t *vm, vm_flag_t flags)
1.1 yamt 232: {
233: bt_t *bt;
234:
235: #if defined(_KERNEL)
1.21 ! yamt 236: int s;
! 237:
1.1 yamt 238: /* XXX bootstrap */
1.21 ! yamt 239: s = splvm();
1.1 yamt 240: bt = pool_cache_get(&bt_poolcache,
241: (flags & VM_SLEEP) != 0 ? PR_WAITOK : PR_NOWAIT);
1.21 ! yamt 242: splx(s);
1.1 yamt 243: #else /* defined(_KERNEL) */
244: bt = malloc(sizeof *bt);
245: #endif /* defined(_KERNEL) */
246:
247: return bt;
248: }
249:
250: static void
1.17 yamt 251: bt_free(vmem_t *vm, bt_t *bt)
1.1 yamt 252: {
253:
254: #if defined(_KERNEL)
1.21 ! yamt 255: int s;
! 256:
1.1 yamt 257: /* XXX bootstrap */
1.21 ! yamt 258: s = splvm();
1.1 yamt 259: pool_cache_put(&bt_poolcache, bt);
1.21 ! yamt 260: splx(s);
1.1 yamt 261: #else /* defined(_KERNEL) */
262: free(bt);
263: #endif /* defined(_KERNEL) */
264: }
265:
266: /*
267: * freelist[0] ... [1, 1]
268: * freelist[1] ... [2, 3]
269: * freelist[2] ... [4, 7]
270: * freelist[3] ... [8, 15]
271: * :
272: * freelist[n] ... [(1 << n), (1 << (n + 1)) - 1]
273: * :
274: */
275:
276: static struct vmem_freelist *
277: bt_freehead_tofree(vmem_t *vm, vmem_size_t size)
278: {
279: const vmem_size_t qsize = size >> vm->vm_quantum_shift;
280: int idx;
281:
282: KASSERT((size & vm->vm_quantum_mask) == 0);
283: KASSERT(size != 0);
284:
285: idx = calc_order(qsize);
286: KASSERT(idx >= 0);
287: KASSERT(idx < VMEM_MAXORDER);
288:
289: return &vm->vm_freelist[idx];
290: }
291:
292: static struct vmem_freelist *
293: bt_freehead_toalloc(vmem_t *vm, vmem_size_t size, vm_flag_t strat)
294: {
295: const vmem_size_t qsize = size >> vm->vm_quantum_shift;
296: int idx;
297:
298: KASSERT((size & vm->vm_quantum_mask) == 0);
299: KASSERT(size != 0);
300:
301: idx = calc_order(qsize);
1.4 yamt 302: if (strat == VM_INSTANTFIT && ORDER2SIZE(idx) != qsize) {
1.1 yamt 303: idx++;
304: /* check too large request? */
305: }
306: KASSERT(idx >= 0);
307: KASSERT(idx < VMEM_MAXORDER);
308:
309: return &vm->vm_freelist[idx];
310: }
311:
312: /* ---- boundary tag hash */
313:
314: static struct vmem_hashlist *
315: bt_hashhead(vmem_t *vm, vmem_addr_t addr)
316: {
317: struct vmem_hashlist *list;
318: unsigned int hash;
319:
320: hash = hash32_buf(&addr, sizeof(addr), HASH32_BUF_INIT);
321: list = &vm->vm_hashlist[hash % vm->vm_hashsize];
322:
323: return list;
324: }
325:
326: static bt_t *
327: bt_lookupbusy(vmem_t *vm, vmem_addr_t addr)
328: {
329: struct vmem_hashlist *list;
330: bt_t *bt;
331:
332: list = bt_hashhead(vm, addr);
333: LIST_FOREACH(bt, list, bt_hashlist) {
334: if (bt->bt_start == addr) {
335: break;
336: }
337: }
338:
339: return bt;
340: }
341:
342: static void
343: bt_rembusy(vmem_t *vm, bt_t *bt)
344: {
345:
346: KASSERT(vm->vm_nbusytag > 0);
347: vm->vm_nbusytag--;
348: LIST_REMOVE(bt, bt_hashlist);
349: }
350:
351: static void
352: bt_insbusy(vmem_t *vm, bt_t *bt)
353: {
354: struct vmem_hashlist *list;
355:
356: KASSERT(bt->bt_type == BT_TYPE_BUSY);
357:
358: list = bt_hashhead(vm, bt->bt_start);
359: LIST_INSERT_HEAD(list, bt, bt_hashlist);
360: vm->vm_nbusytag++;
361: }
362:
363: /* ---- boundary tag list */
364:
365: static void
366: bt_remseg(vmem_t *vm, bt_t *bt)
367: {
368:
369: CIRCLEQ_REMOVE(&vm->vm_seglist, bt, bt_seglist);
370: }
371:
372: static void
373: bt_insseg(vmem_t *vm, bt_t *bt, bt_t *prev)
374: {
375:
376: CIRCLEQ_INSERT_AFTER(&vm->vm_seglist, prev, bt, bt_seglist);
377: }
378:
379: static void
380: bt_insseg_tail(vmem_t *vm, bt_t *bt)
381: {
382:
383: CIRCLEQ_INSERT_TAIL(&vm->vm_seglist, bt, bt_seglist);
384: }
385:
386: static void
1.17 yamt 387: bt_remfree(vmem_t *vm, bt_t *bt)
1.1 yamt 388: {
389:
390: KASSERT(bt->bt_type == BT_TYPE_FREE);
391:
392: LIST_REMOVE(bt, bt_freelist);
393: }
394:
395: static void
396: bt_insfree(vmem_t *vm, bt_t *bt)
397: {
398: struct vmem_freelist *list;
399:
400: list = bt_freehead_tofree(vm, bt->bt_size);
401: LIST_INSERT_HEAD(list, bt, bt_freelist);
402: }
403:
404: /* ---- vmem internal functions */
405:
1.5 yamt 406: #if defined(QCACHE)
407: static inline vm_flag_t
408: prf_to_vmf(int prflags)
409: {
410: vm_flag_t vmflags;
411:
412: KASSERT((prflags & ~(PR_LIMITFAIL | PR_WAITOK | PR_NOWAIT)) == 0);
413: if ((prflags & PR_WAITOK) != 0) {
414: vmflags = VM_SLEEP;
415: } else {
416: vmflags = VM_NOSLEEP;
417: }
418: return vmflags;
419: }
420:
421: static inline int
422: vmf_to_prf(vm_flag_t vmflags)
423: {
424: int prflags;
425:
1.7 yamt 426: if ((vmflags & VM_SLEEP) != 0) {
1.5 yamt 427: prflags = PR_WAITOK;
1.7 yamt 428: } else {
1.5 yamt 429: prflags = PR_NOWAIT;
430: }
431: return prflags;
432: }
433:
434: static size_t
435: qc_poolpage_size(size_t qcache_max)
436: {
437: int i;
438:
439: for (i = 0; ORDER2SIZE(i) <= qcache_max * 3; i++) {
440: /* nothing */
441: }
442: return ORDER2SIZE(i);
443: }
444:
445: static void *
446: qc_poolpage_alloc(struct pool *pool, int prflags)
447: {
448: qcache_t *qc = QC_POOL_TO_QCACHE(pool);
449: vmem_t *vm = qc->qc_vmem;
450:
451: return (void *)vmem_alloc(vm, pool->pr_alloc->pa_pagesz,
452: prf_to_vmf(prflags) | VM_INSTANTFIT);
453: }
454:
455: static void
456: qc_poolpage_free(struct pool *pool, void *addr)
457: {
458: qcache_t *qc = QC_POOL_TO_QCACHE(pool);
459: vmem_t *vm = qc->qc_vmem;
460:
461: vmem_free(vm, (vmem_addr_t)addr, pool->pr_alloc->pa_pagesz);
462: }
463:
464: static void
465: qc_init(vmem_t *vm, size_t qcache_max)
466: {
467: struct pool_allocator *pa;
468: int qcache_idx_max;
469: int i;
470:
471: KASSERT((qcache_max & vm->vm_quantum_mask) == 0);
472: if (qcache_max > (VMEM_QCACHE_IDX_MAX << vm->vm_quantum_shift)) {
473: qcache_max = VMEM_QCACHE_IDX_MAX << vm->vm_quantum_shift;
474: }
475: vm->vm_qcache_max = qcache_max;
476: pa = &vm->vm_qcache_allocator;
477: memset(pa, 0, sizeof(*pa));
478: pa->pa_alloc = qc_poolpage_alloc;
479: pa->pa_free = qc_poolpage_free;
480: pa->pa_pagesz = qc_poolpage_size(qcache_max);
481:
482: qcache_idx_max = qcache_max >> vm->vm_quantum_shift;
483: for (i = 1; i <= qcache_idx_max; i++) {
484: qcache_t *qc = &vm->vm_qcache[i - 1];
485: size_t size = i << vm->vm_quantum_shift;
486:
487: qc->qc_vmem = vm;
1.8 martin 488: snprintf(qc->qc_name, sizeof(qc->qc_name), "%s-%zu",
1.5 yamt 489: vm->vm_name, size);
1.15 yamt 490: pool_init(&qc->qc_pool, size, ORDER2SIZE(vm->vm_quantum_shift),
491: 0, PR_NOALIGN | PR_NOTOUCH /* XXX */, qc->qc_name, pa);
1.5 yamt 492: pool_cache_init(&qc->qc_cache, &qc->qc_pool, NULL, NULL, NULL);
493: }
494: }
1.6 yamt 495:
496: static boolean_t
497: qc_reap(vmem_t *vm)
498: {
499: int i;
500: int qcache_idx_max;
501: boolean_t didsomething = FALSE;
502:
503: qcache_idx_max = vm->vm_qcache_max >> vm->vm_quantum_shift;
504: for (i = 1; i <= qcache_idx_max; i++) {
505: qcache_t *qc = &vm->vm_qcache[i - 1];
506:
507: if (pool_reclaim(&qc->qc_pool) != 0) {
508: didsomething = TRUE;
509: }
510: }
511:
512: return didsomething;
513: }
1.5 yamt 514: #endif /* defined(QCACHE) */
515:
1.1 yamt 516: #if defined(_KERNEL)
517: static int
518: vmem_init(void)
519: {
520:
521: pool_cache_init(&bt_poolcache, &bt_pool, NULL, NULL, NULL);
522: return 0;
523: }
524: #endif /* defined(_KERNEL) */
525:
526: static vmem_addr_t
527: vmem_add1(vmem_t *vm, vmem_addr_t addr, vmem_size_t size, vm_flag_t flags,
528: int spanbttype)
529: {
530: bt_t *btspan;
531: bt_t *btfree;
532:
533: KASSERT((flags & (VM_SLEEP|VM_NOSLEEP)) != 0);
534: KASSERT((~flags & (VM_SLEEP|VM_NOSLEEP)) != 0);
535: VMEM_ASSERT_UNLOCKED(vm);
536:
537: btspan = bt_alloc(vm, flags);
538: if (btspan == NULL) {
539: return VMEM_ADDR_NULL;
540: }
541: btfree = bt_alloc(vm, flags);
542: if (btfree == NULL) {
543: bt_free(vm, btspan);
544: return VMEM_ADDR_NULL;
545: }
546:
547: btspan->bt_type = spanbttype;
548: btspan->bt_start = addr;
549: btspan->bt_size = size;
550:
551: btfree->bt_type = BT_TYPE_FREE;
552: btfree->bt_start = addr;
553: btfree->bt_size = size;
554:
555: VMEM_LOCK(vm);
556: bt_insseg_tail(vm, btspan);
557: bt_insseg(vm, btfree, btspan);
558: bt_insfree(vm, btfree);
559: VMEM_UNLOCK(vm);
560:
561: return addr;
562: }
563:
564: static int
565: vmem_import(vmem_t *vm, vmem_size_t size, vm_flag_t flags)
566: {
567: vmem_addr_t addr;
568:
569: VMEM_ASSERT_UNLOCKED(vm);
570:
571: if (vm->vm_allocfn == NULL) {
572: return EINVAL;
573: }
574:
575: addr = (*vm->vm_allocfn)(vm->vm_source, size, &size, flags);
576: if (addr == VMEM_ADDR_NULL) {
577: return ENOMEM;
578: }
579:
580: if (vmem_add1(vm, addr, size, flags, BT_TYPE_SPAN) == VMEM_ADDR_NULL) {
581: (*vm->vm_freefn)(vm->vm_source, addr, size);
582: return ENOMEM;
583: }
584:
585: return 0;
586: }
587:
588: static int
589: vmem_rehash(vmem_t *vm, size_t newhashsize, vm_flag_t flags)
590: {
591: bt_t *bt;
592: int i;
593: struct vmem_hashlist *newhashlist;
594: struct vmem_hashlist *oldhashlist;
595: size_t oldhashsize;
596:
597: KASSERT(newhashsize > 0);
598: VMEM_ASSERT_UNLOCKED(vm);
599:
600: newhashlist =
601: xmalloc(sizeof(struct vmem_hashlist *) * newhashsize, flags);
602: if (newhashlist == NULL) {
603: return ENOMEM;
604: }
605: for (i = 0; i < newhashsize; i++) {
606: LIST_INIT(&newhashlist[i]);
607: }
608:
609: VMEM_LOCK(vm);
610: oldhashlist = vm->vm_hashlist;
611: oldhashsize = vm->vm_hashsize;
612: vm->vm_hashlist = newhashlist;
613: vm->vm_hashsize = newhashsize;
614: if (oldhashlist == NULL) {
615: VMEM_UNLOCK(vm);
616: return 0;
617: }
618: for (i = 0; i < oldhashsize; i++) {
619: while ((bt = LIST_FIRST(&oldhashlist[i])) != NULL) {
620: bt_rembusy(vm, bt); /* XXX */
621: bt_insbusy(vm, bt);
622: }
623: }
624: VMEM_UNLOCK(vm);
625:
626: xfree(oldhashlist);
627:
628: return 0;
629: }
630:
1.10 yamt 631: /*
632: * vmem_fit: check if a bt can satisfy the given restrictions.
633: */
634:
635: static vmem_addr_t
636: vmem_fit(const bt_t *bt, vmem_size_t size, vmem_size_t align, vmem_size_t phase,
637: vmem_size_t nocross, vmem_addr_t minaddr, vmem_addr_t maxaddr)
638: {
639: vmem_addr_t start;
640: vmem_addr_t end;
641:
642: KASSERT(bt->bt_size >= size);
643:
644: /*
645: * XXX assumption: vmem_addr_t and vmem_size_t are
646: * unsigned integer of the same size.
647: */
648:
649: start = bt->bt_start;
650: if (start < minaddr) {
651: start = minaddr;
652: }
653: end = BT_END(bt);
654: if (end > maxaddr - 1) {
655: end = maxaddr - 1;
656: }
657: if (start >= end) {
658: return VMEM_ADDR_NULL;
659: }
1.19 yamt 660:
661: start = VMEM_ALIGNUP(start - phase, align) + phase;
1.10 yamt 662: if (start < bt->bt_start) {
663: start += align;
664: }
1.19 yamt 665: if (VMEM_CROSS_P(start, start + size - 1, nocross)) {
1.10 yamt 666: KASSERT(align < nocross);
1.19 yamt 667: start = VMEM_ALIGNUP(start - phase, nocross) + phase;
1.10 yamt 668: }
669: if (start < end && end - start >= size) {
670: KASSERT((start & (align - 1)) == phase);
1.19 yamt 671: KASSERT(!VMEM_CROSS_P(start, start + size - 1, nocross));
1.10 yamt 672: KASSERT(minaddr <= start);
673: KASSERT(maxaddr == 0 || start + size <= maxaddr);
674: KASSERT(bt->bt_start <= start);
675: KASSERT(start + size <= BT_END(bt));
676: return start;
677: }
678: return VMEM_ADDR_NULL;
679: }
680:
1.1 yamt 681: /* ---- vmem API */
682:
683: /*
684: * vmem_create: create an arena.
685: *
686: * => must not be called from interrupt context.
687: */
688:
689: vmem_t *
690: vmem_create(const char *name, vmem_addr_t base, vmem_size_t size,
691: vmem_size_t quantum,
692: vmem_addr_t (*allocfn)(vmem_t *, vmem_size_t, vmem_size_t *, vm_flag_t),
693: void (*freefn)(vmem_t *, vmem_addr_t, vmem_size_t),
694: vmem_t *source, vmem_size_t qcache_max, vm_flag_t flags)
695: {
696: vmem_t *vm;
697: int i;
698: #if defined(_KERNEL)
699: static ONCE_DECL(control);
700: #endif /* defined(_KERNEL) */
701:
702: KASSERT((flags & (VM_SLEEP|VM_NOSLEEP)) != 0);
703: KASSERT((~flags & (VM_SLEEP|VM_NOSLEEP)) != 0);
704:
705: #if defined(_KERNEL)
706: if (RUN_ONCE(&control, vmem_init)) {
707: return NULL;
708: }
709: #endif /* defined(_KERNEL) */
710: vm = xmalloc(sizeof(*vm), flags);
711: if (vm == NULL) {
712: return NULL;
713: }
714:
715: VMEM_LOCK_INIT(vm);
716: vm->vm_name = name;
717: vm->vm_quantum_mask = quantum - 1;
718: vm->vm_quantum_shift = calc_order(quantum);
1.4 yamt 719: KASSERT(ORDER2SIZE(vm->vm_quantum_shift) == quantum);
1.1 yamt 720: vm->vm_allocfn = allocfn;
721: vm->vm_freefn = freefn;
722: vm->vm_source = source;
723: vm->vm_nbusytag = 0;
1.5 yamt 724: #if defined(QCACHE)
725: qc_init(vm, qcache_max);
726: #endif /* defined(QCACHE) */
1.1 yamt 727:
728: CIRCLEQ_INIT(&vm->vm_seglist);
729: for (i = 0; i < VMEM_MAXORDER; i++) {
730: LIST_INIT(&vm->vm_freelist[i]);
731: }
732: vm->vm_hashlist = NULL;
733: if (vmem_rehash(vm, VMEM_HASHSIZE_INIT, flags)) {
734: vmem_destroy(vm);
735: return NULL;
736: }
737:
738: if (size != 0) {
739: if (vmem_add(vm, base, size, flags) == 0) {
740: vmem_destroy(vm);
741: return NULL;
742: }
743: }
744:
745: return vm;
746: }
747:
748: void
749: vmem_destroy(vmem_t *vm)
750: {
751:
752: VMEM_ASSERT_UNLOCKED(vm);
753:
754: if (vm->vm_hashlist != NULL) {
755: int i;
756:
757: for (i = 0; i < vm->vm_hashsize; i++) {
758: bt_t *bt;
759:
760: while ((bt = LIST_FIRST(&vm->vm_hashlist[i])) != NULL) {
761: KASSERT(bt->bt_type == BT_TYPE_SPAN_STATIC);
762: bt_free(vm, bt);
763: }
764: }
765: xfree(vm->vm_hashlist);
766: }
767: xfree(vm);
768: }
769:
770: vmem_size_t
771: vmem_roundup_size(vmem_t *vm, vmem_size_t size)
772: {
773:
774: return (size + vm->vm_quantum_mask) & ~vm->vm_quantum_mask;
775: }
776:
777: /*
778: * vmem_alloc:
779: *
780: * => caller must ensure appropriate spl,
781: * if the arena can be accessed from interrupt context.
782: */
783:
784: vmem_addr_t
785: vmem_alloc(vmem_t *vm, vmem_size_t size0, vm_flag_t flags)
786: {
1.12 yamt 787: const vmem_size_t size __unused = vmem_roundup_size(vm, size0);
788: const vm_flag_t strat __unused = flags & VM_FITMASK;
1.1 yamt 789:
790: KASSERT((flags & (VM_SLEEP|VM_NOSLEEP)) != 0);
791: KASSERT((~flags & (VM_SLEEP|VM_NOSLEEP)) != 0);
792: VMEM_ASSERT_UNLOCKED(vm);
793:
794: KASSERT(size0 > 0);
795: KASSERT(size > 0);
796: KASSERT(strat == VM_BESTFIT || strat == VM_INSTANTFIT);
1.3 yamt 797: if ((flags & VM_SLEEP) != 0) {
1.16 yamt 798: ASSERT_SLEEPABLE(NULL, __func__);
1.3 yamt 799: }
1.1 yamt 800:
1.5 yamt 801: #if defined(QCACHE)
802: if (size <= vm->vm_qcache_max) {
803: int qidx = size >> vm->vm_quantum_shift;
804: qcache_t *qc = &vm->vm_qcache[qidx - 1];
805:
806: return (vmem_addr_t)pool_cache_get(&qc->qc_cache,
807: vmf_to_prf(flags));
808: }
809: #endif /* defined(QCACHE) */
810:
1.10 yamt 811: return vmem_xalloc(vm, size0, 0, 0, 0, 0, 0, flags);
812: }
813:
814: vmem_addr_t
815: vmem_xalloc(vmem_t *vm, vmem_size_t size0, vmem_size_t align, vmem_size_t phase,
816: vmem_size_t nocross, vmem_addr_t minaddr, vmem_addr_t maxaddr,
817: vm_flag_t flags)
818: {
819: struct vmem_freelist *list;
820: struct vmem_freelist *first;
821: struct vmem_freelist *end;
822: bt_t *bt;
823: bt_t *btnew;
824: bt_t *btnew2;
825: const vmem_size_t size = vmem_roundup_size(vm, size0);
826: vm_flag_t strat = flags & VM_FITMASK;
827: vmem_addr_t start;
828:
829: KASSERT(size0 > 0);
830: KASSERT(size > 0);
831: KASSERT(strat == VM_BESTFIT || strat == VM_INSTANTFIT);
832: if ((flags & VM_SLEEP) != 0) {
1.16 yamt 833: ASSERT_SLEEPABLE(NULL, __func__);
1.10 yamt 834: }
835: KASSERT((align & vm->vm_quantum_mask) == 0);
836: KASSERT((align & (align - 1)) == 0);
837: KASSERT((phase & vm->vm_quantum_mask) == 0);
838: KASSERT((nocross & vm->vm_quantum_mask) == 0);
839: KASSERT((nocross & (nocross - 1)) == 0);
840: KASSERT((align == 0 && phase == 0) || phase < align);
841: KASSERT(nocross == 0 || nocross >= size);
842: KASSERT(maxaddr == 0 || minaddr < maxaddr);
1.19 yamt 843: KASSERT(!VMEM_CROSS_P(phase, phase + size - 1, nocross));
1.10 yamt 844:
845: if (align == 0) {
846: align = vm->vm_quantum_mask + 1;
847: }
1.1 yamt 848: btnew = bt_alloc(vm, flags);
849: if (btnew == NULL) {
850: return VMEM_ADDR_NULL;
851: }
1.10 yamt 852: btnew2 = bt_alloc(vm, flags); /* XXX not necessary if no restrictions */
853: if (btnew2 == NULL) {
854: bt_free(vm, btnew);
855: return VMEM_ADDR_NULL;
856: }
1.1 yamt 857:
858: retry_strat:
859: first = bt_freehead_toalloc(vm, size, strat);
860: end = &vm->vm_freelist[VMEM_MAXORDER];
861: retry:
862: bt = NULL;
863: VMEM_LOCK(vm);
1.2 yamt 864: if (strat == VM_INSTANTFIT) {
865: for (list = first; list < end; list++) {
866: bt = LIST_FIRST(list);
867: if (bt != NULL) {
1.10 yamt 868: start = vmem_fit(bt, size, align, phase,
869: nocross, minaddr, maxaddr);
870: if (start != VMEM_ADDR_NULL) {
871: goto gotit;
872: }
1.2 yamt 873: }
874: }
875: } else { /* VM_BESTFIT */
876: for (list = first; list < end; list++) {
877: LIST_FOREACH(bt, list, bt_freelist) {
878: if (bt->bt_size >= size) {
1.10 yamt 879: start = vmem_fit(bt, size, align, phase,
880: nocross, minaddr, maxaddr);
881: if (start != VMEM_ADDR_NULL) {
882: goto gotit;
883: }
1.2 yamt 884: }
1.1 yamt 885: }
886: }
887: }
1.2 yamt 888: VMEM_UNLOCK(vm);
1.1 yamt 889: #if 1
1.2 yamt 890: if (strat == VM_INSTANTFIT) {
891: strat = VM_BESTFIT;
892: goto retry_strat;
893: }
1.1 yamt 894: #endif
1.10 yamt 895: if (align != vm->vm_quantum_mask + 1 || phase != 0 ||
896: nocross != 0 || minaddr != 0 || maxaddr != 0) {
897:
898: /*
899: * XXX should try to import a region large enough to
900: * satisfy restrictions?
901: */
902:
1.20 yamt 903: goto fail;
1.10 yamt 904: }
1.2 yamt 905: if (vmem_import(vm, size, flags) == 0) {
906: goto retry;
1.1 yamt 907: }
1.2 yamt 908: /* XXX */
1.20 yamt 909: fail:
910: bt_free(vm, btnew);
911: bt_free(vm, btnew2);
1.2 yamt 912: return VMEM_ADDR_NULL;
913:
914: gotit:
1.1 yamt 915: KASSERT(bt->bt_type == BT_TYPE_FREE);
916: KASSERT(bt->bt_size >= size);
917: bt_remfree(vm, bt);
1.10 yamt 918: if (bt->bt_start != start) {
919: btnew2->bt_type = BT_TYPE_FREE;
920: btnew2->bt_start = bt->bt_start;
921: btnew2->bt_size = start - bt->bt_start;
922: bt->bt_start = start;
923: bt->bt_size -= btnew2->bt_size;
924: bt_insfree(vm, btnew2);
925: bt_insseg(vm, btnew2, CIRCLEQ_PREV(bt, bt_seglist));
926: btnew2 = NULL;
927: }
928: KASSERT(bt->bt_start == start);
1.1 yamt 929: if (bt->bt_size != size && bt->bt_size - size > vm->vm_quantum_mask) {
930: /* split */
931: btnew->bt_type = BT_TYPE_BUSY;
932: btnew->bt_start = bt->bt_start;
933: btnew->bt_size = size;
934: bt->bt_start = bt->bt_start + size;
935: bt->bt_size -= size;
936: bt_insfree(vm, bt);
937: bt_insseg(vm, btnew, CIRCLEQ_PREV(bt, bt_seglist));
938: bt_insbusy(vm, btnew);
939: VMEM_UNLOCK(vm);
940: } else {
941: bt->bt_type = BT_TYPE_BUSY;
942: bt_insbusy(vm, bt);
943: VMEM_UNLOCK(vm);
944: bt_free(vm, btnew);
945: btnew = bt;
946: }
1.10 yamt 947: if (btnew2 != NULL) {
948: bt_free(vm, btnew2);
949: }
1.1 yamt 950: KASSERT(btnew->bt_size >= size);
951: btnew->bt_type = BT_TYPE_BUSY;
952:
953: return btnew->bt_start;
954: }
955:
956: /*
957: * vmem_free:
958: *
959: * => caller must ensure appropriate spl,
960: * if the arena can be accessed from interrupt context.
961: */
962:
963: void
964: vmem_free(vmem_t *vm, vmem_addr_t addr, vmem_size_t size)
965: {
966:
967: VMEM_ASSERT_UNLOCKED(vm);
968: KASSERT(addr != VMEM_ADDR_NULL);
969: KASSERT(size > 0);
970:
1.5 yamt 971: #if defined(QCACHE)
972: if (size <= vm->vm_qcache_max) {
973: int qidx = (size + vm->vm_quantum_mask) >> vm->vm_quantum_shift;
974: qcache_t *qc = &vm->vm_qcache[qidx - 1];
975:
976: return pool_cache_put(&qc->qc_cache, (void *)addr);
977: }
978: #endif /* defined(QCACHE) */
979:
1.10 yamt 980: vmem_xfree(vm, addr, size);
981: }
982:
983: void
1.17 yamt 984: vmem_xfree(vmem_t *vm, vmem_addr_t addr, vmem_size_t size)
1.10 yamt 985: {
986: bt_t *bt;
987: bt_t *t;
988:
989: VMEM_ASSERT_UNLOCKED(vm);
990: KASSERT(addr != VMEM_ADDR_NULL);
991: KASSERT(size > 0);
992:
1.1 yamt 993: VMEM_LOCK(vm);
994:
995: bt = bt_lookupbusy(vm, addr);
996: KASSERT(bt != NULL);
997: KASSERT(bt->bt_start == addr);
998: KASSERT(bt->bt_size == vmem_roundup_size(vm, size) ||
999: bt->bt_size - vmem_roundup_size(vm, size) <= vm->vm_quantum_mask);
1000: KASSERT(bt->bt_type == BT_TYPE_BUSY);
1001: bt_rembusy(vm, bt);
1002: bt->bt_type = BT_TYPE_FREE;
1003:
1004: /* coalesce */
1005: t = CIRCLEQ_NEXT(bt, bt_seglist);
1006: if (t != NULL && t->bt_type == BT_TYPE_FREE) {
1007: KASSERT(BT_END(bt) == t->bt_start);
1008: bt_remfree(vm, t);
1009: bt_remseg(vm, t);
1010: bt->bt_size += t->bt_size;
1011: bt_free(vm, t);
1012: }
1013: t = CIRCLEQ_PREV(bt, bt_seglist);
1014: if (t != NULL && t->bt_type == BT_TYPE_FREE) {
1015: KASSERT(BT_END(t) == bt->bt_start);
1016: bt_remfree(vm, t);
1017: bt_remseg(vm, t);
1018: bt->bt_size += t->bt_size;
1019: bt->bt_start = t->bt_start;
1020: bt_free(vm, t);
1021: }
1022:
1023: t = CIRCLEQ_PREV(bt, bt_seglist);
1024: KASSERT(t != NULL);
1025: KASSERT(BT_ISSPAN_P(t) || t->bt_type == BT_TYPE_BUSY);
1026: if (vm->vm_freefn != NULL && t->bt_type == BT_TYPE_SPAN &&
1027: t->bt_size == bt->bt_size) {
1028: vmem_addr_t spanaddr;
1029: vmem_size_t spansize;
1030:
1031: KASSERT(t->bt_start == bt->bt_start);
1032: spanaddr = bt->bt_start;
1033: spansize = bt->bt_size;
1034: bt_remseg(vm, bt);
1035: bt_free(vm, bt);
1036: bt_remseg(vm, t);
1037: bt_free(vm, t);
1038: VMEM_UNLOCK(vm);
1039: (*vm->vm_freefn)(vm->vm_source, spanaddr, spansize);
1040: } else {
1041: bt_insfree(vm, bt);
1042: VMEM_UNLOCK(vm);
1043: }
1044: }
1045:
1046: /*
1047: * vmem_add:
1048: *
1049: * => caller must ensure appropriate spl,
1050: * if the arena can be accessed from interrupt context.
1051: */
1052:
1053: vmem_addr_t
1054: vmem_add(vmem_t *vm, vmem_addr_t addr, vmem_size_t size, vm_flag_t flags)
1055: {
1056:
1057: return vmem_add1(vm, addr, size, flags, BT_TYPE_SPAN_STATIC);
1058: }
1059:
1.6 yamt 1060: /*
1061: * vmem_reap: reap unused resources.
1062: *
1063: * => return TRUE if we successfully reaped something.
1064: */
1065:
1066: boolean_t
1067: vmem_reap(vmem_t *vm)
1068: {
1069: boolean_t didsomething = FALSE;
1070:
1071: VMEM_ASSERT_UNLOCKED(vm);
1072:
1073: #if defined(QCACHE)
1074: didsomething = qc_reap(vm);
1075: #endif /* defined(QCACHE) */
1076: return didsomething;
1077: }
1078:
1.1 yamt 1079: /* ---- debug */
1080:
1081: #if defined(VMEM_DEBUG)
1082:
1083: #if !defined(_KERNEL)
1084: #include <stdio.h>
1085: #endif /* !defined(_KERNEL) */
1086:
1087: void bt_dump(const bt_t *);
1088:
1089: void
1090: bt_dump(const bt_t *bt)
1091: {
1092:
1093: printf("\t%p: %" PRIu64 ", %" PRIu64 ", %d\n",
1094: bt, (uint64_t)bt->bt_start, (uint64_t)bt->bt_size,
1095: bt->bt_type);
1096: }
1097:
1098: void
1099: vmem_dump(const vmem_t *vm)
1100: {
1101: const bt_t *bt;
1102: int i;
1103:
1104: printf("vmem %p '%s'\n", vm, vm->vm_name);
1105: CIRCLEQ_FOREACH(bt, &vm->vm_seglist, bt_seglist) {
1106: bt_dump(bt);
1107: }
1108:
1109: for (i = 0; i < VMEM_MAXORDER; i++) {
1110: const struct vmem_freelist *fl = &vm->vm_freelist[i];
1111:
1112: if (LIST_EMPTY(fl)) {
1113: continue;
1114: }
1115:
1116: printf("freelist[%d]\n", i);
1117: LIST_FOREACH(bt, fl, bt_freelist) {
1118: bt_dump(bt);
1119: if (bt->bt_size) {
1120: }
1121: }
1122: }
1123: }
1124:
1125: #if !defined(_KERNEL)
1126:
1127: #include <stdlib.h>
1128:
1129: int
1130: main()
1131: {
1132: vmem_t *vm;
1133: vmem_addr_t p;
1134: struct reg {
1135: vmem_addr_t p;
1136: vmem_size_t sz;
1.10 yamt 1137: boolean_t x;
1.1 yamt 1138: } *reg = NULL;
1139: int nreg = 0;
1140: int nalloc = 0;
1141: int nfree = 0;
1142: vmem_size_t total = 0;
1143: #if 1
1144: vm_flag_t strat = VM_INSTANTFIT;
1145: #else
1146: vm_flag_t strat = VM_BESTFIT;
1147: #endif
1148:
1149: vm = vmem_create("test", VMEM_ADDR_NULL, 0, 1,
1150: NULL, NULL, NULL, 0, VM_NOSLEEP);
1151: if (vm == NULL) {
1152: printf("vmem_create\n");
1153: exit(EXIT_FAILURE);
1154: }
1155: vmem_dump(vm);
1156:
1157: p = vmem_add(vm, 100, 200, VM_SLEEP);
1158: p = vmem_add(vm, 2000, 1, VM_SLEEP);
1159: p = vmem_add(vm, 40000, 0x10000000>>12, VM_SLEEP);
1160: p = vmem_add(vm, 10000, 10000, VM_SLEEP);
1161: p = vmem_add(vm, 500, 1000, VM_SLEEP);
1162: vmem_dump(vm);
1163: for (;;) {
1164: struct reg *r;
1.10 yamt 1165: int t = rand() % 100;
1.1 yamt 1166:
1.10 yamt 1167: if (t > 45) {
1168: /* alloc */
1.1 yamt 1169: vmem_size_t sz = rand() % 500 + 1;
1.10 yamt 1170: boolean_t x;
1171: vmem_size_t align, phase, nocross;
1172: vmem_addr_t minaddr, maxaddr;
1173:
1174: if (t > 70) {
1175: x = TRUE;
1176: /* XXX */
1177: align = 1 << (rand() % 15);
1178: phase = rand() % 65536;
1179: nocross = 1 << (rand() % 15);
1180: if (align <= phase) {
1181: phase = 0;
1182: }
1.19 yamt 1183: if (VMEM_CROSS_P(phase, phase + sz - 1,
1184: nocross)) {
1.10 yamt 1185: nocross = 0;
1186: }
1187: minaddr = rand() % 50000;
1188: maxaddr = rand() % 70000;
1189: if (minaddr > maxaddr) {
1190: minaddr = 0;
1191: maxaddr = 0;
1192: }
1193: printf("=== xalloc %" PRIu64
1194: " align=%" PRIu64 ", phase=%" PRIu64
1195: ", nocross=%" PRIu64 ", min=%" PRIu64
1196: ", max=%" PRIu64 "\n",
1197: (uint64_t)sz,
1198: (uint64_t)align,
1199: (uint64_t)phase,
1200: (uint64_t)nocross,
1201: (uint64_t)minaddr,
1202: (uint64_t)maxaddr);
1203: p = vmem_xalloc(vm, sz, align, phase, nocross,
1204: minaddr, maxaddr, strat|VM_SLEEP);
1205: } else {
1206: x = FALSE;
1207: printf("=== alloc %" PRIu64 "\n", (uint64_t)sz);
1208: p = vmem_alloc(vm, sz, strat|VM_SLEEP);
1209: }
1.1 yamt 1210: printf("-> %" PRIu64 "\n", (uint64_t)p);
1211: vmem_dump(vm);
1212: if (p == VMEM_ADDR_NULL) {
1.10 yamt 1213: if (x) {
1214: continue;
1215: }
1.1 yamt 1216: break;
1217: }
1218: nreg++;
1219: reg = realloc(reg, sizeof(*reg) * nreg);
1220: r = ®[nreg - 1];
1221: r->p = p;
1222: r->sz = sz;
1.10 yamt 1223: r->x = x;
1.1 yamt 1224: total += sz;
1225: nalloc++;
1226: } else if (nreg != 0) {
1.10 yamt 1227: /* free */
1.1 yamt 1228: r = ®[rand() % nreg];
1229: printf("=== free %" PRIu64 ", %" PRIu64 "\n",
1230: (uint64_t)r->p, (uint64_t)r->sz);
1.10 yamt 1231: if (r->x) {
1232: vmem_xfree(vm, r->p, r->sz);
1233: } else {
1234: vmem_free(vm, r->p, r->sz);
1235: }
1.1 yamt 1236: total -= r->sz;
1237: vmem_dump(vm);
1238: *r = reg[nreg - 1];
1239: nreg--;
1240: nfree++;
1241: }
1242: printf("total=%" PRIu64 "\n", (uint64_t)total);
1243: }
1244: fprintf(stderr, "total=%" PRIu64 ", nalloc=%d, nfree=%d\n",
1245: (uint64_t)total, nalloc, nfree);
1246: exit(EXIT_SUCCESS);
1247: }
1248: #endif /* !defined(_KERNEL) */
1249: #endif /* defined(VMEM_DEBUG) */
CVSweb <webmaster@jp.NetBSD.org>
![[BACK]](/icons/back.gif){kind=icon}
Return to subr_vmem.c CVS log ![[TXT]](/icons/text.gif){kind=icon}
![[DIR]](/icons/dir.gif){kind=icon}
Up to [cvs.NetBSD.org] / src / sys / kern