Annotation of src/sys/kern/subr_pool.c, Revision 1.52
1.52 ! thorpej 1: /* $NetBSD: subr_pool.c,v 1.51 2001/05/04 19:41:26 thorpej Exp $ */
1.1 pk 2:
3: /*-
1.43 thorpej 4: * Copyright (c) 1997, 1999, 2000 The NetBSD Foundation, Inc.
1.1 pk 5: * All rights reserved.
6: *
7: * This code is derived from software contributed to The NetBSD Foundation
1.20 thorpej 8: * by Paul Kranenburg; by Jason R. Thorpe of the Numerical Aerospace
9: * Simulation Facility, NASA Ames Research Center.
1.1 pk 10: *
11: * Redistribution and use in source and binary forms, with or without
12: * modification, are permitted provided that the following conditions
13: * are met:
14: * 1. Redistributions of source code must retain the above copyright
15: * notice, this list of conditions and the following disclaimer.
16: * 2. Redistributions in binary form must reproduce the above copyright
17: * notice, this list of conditions and the following disclaimer in the
18: * documentation and/or other materials provided with the distribution.
19: * 3. All advertising materials mentioning features or use of this software
20: * must display the following acknowledgement:
1.13 christos 21: * This product includes software developed by the NetBSD
22: * Foundation, Inc. and its contributors.
1.1 pk 23: * 4. Neither the name of The NetBSD Foundation nor the names of its
24: * contributors may be used to endorse or promote products derived
25: * from this software without specific prior written permission.
26: *
27: * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
28: * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
29: * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
30: * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
31: * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
32: * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
33: * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
34: * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
35: * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
36: * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
37: * POSSIBILITY OF SUCH DAMAGE.
38: */
1.24 scottr 39:
1.25 thorpej 40: #include "opt_pool.h"
1.24 scottr 41: #include "opt_poollog.h"
1.28 thorpej 42: #include "opt_lockdebug.h"
1.1 pk 43:
44: #include <sys/param.h>
45: #include <sys/systm.h>
46: #include <sys/proc.h>
47: #include <sys/errno.h>
48: #include <sys/kernel.h>
49: #include <sys/malloc.h>
50: #include <sys/lock.h>
51: #include <sys/pool.h>
1.20 thorpej 52: #include <sys/syslog.h>
1.3 pk 53:
54: #include <uvm/uvm.h>
55:
1.1 pk 56: /*
57: * Pool resource management utility.
1.3 pk 58: *
59: * Memory is allocated in pages which are split into pieces according
60: * to the pool item size. Each page is kept on a list headed by `pr_pagelist'
61: * in the pool structure and the individual pool items are on a linked list
62: * headed by `ph_itemlist' in each page header. The memory for building
63: * the page list is either taken from the allocated pages themselves (for
64: * small pool items) or taken from an internal pool of page headers (`phpool').
1.1 pk 65: */
66:
1.3 pk 67: /* List of all pools */
1.5 thorpej 68: TAILQ_HEAD(,pool) pool_head = TAILQ_HEAD_INITIALIZER(pool_head);
1.3 pk 69:
70: /* Private pool for page header structures */
71: static struct pool phpool;
72:
73: /* # of seconds to retain page after last use */
74: int pool_inactive_time = 10;
75:
76: /* Next candidate for drainage (see pool_drain()) */
1.23 thorpej 77: static struct pool *drainpp;
78:
79: /* This spin lock protects both pool_head and drainpp. */
80: struct simplelock pool_head_slock = SIMPLELOCK_INITIALIZER;
1.3 pk 81:
82: struct pool_item_header {
83: /* Page headers */
84: TAILQ_ENTRY(pool_item_header)
85: ph_pagelist; /* pool page list */
86: TAILQ_HEAD(,pool_item) ph_itemlist; /* chunk list for this page */
87: LIST_ENTRY(pool_item_header)
88: ph_hashlist; /* Off-page page headers */
89: int ph_nmissing; /* # of chunks in use */
90: caddr_t ph_page; /* this page's address */
91: struct timeval ph_time; /* last referenced */
92: };
93:
1.1 pk 94: struct pool_item {
1.3 pk 95: #ifdef DIAGNOSTIC
96: int pi_magic;
1.33 chs 97: #endif
1.25 thorpej 98: #define PI_MAGIC 0xdeadbeef
1.3 pk 99: /* Other entries use only this list entry */
100: TAILQ_ENTRY(pool_item) pi_list;
101: };
102:
1.25 thorpej 103: #define PR_HASH_INDEX(pp,addr) \
1.3 pk 104: (((u_long)(addr) >> (pp)->pr_pageshift) & (PR_HASHTABSIZE - 1))
105:
1.43 thorpej 106: /*
107: * Pool cache management.
108: *
109: * Pool caches provide a way for constructed objects to be cached by the
110: * pool subsystem. This can lead to performance improvements by avoiding
111: * needless object construction/destruction; it is deferred until absolutely
112: * necessary.
113: *
114: * Caches are grouped into cache groups. Each cache group references
115: * up to 16 constructed objects. When a cache allocates an object
116: * from the pool, it calls the object's constructor and places it into
117: * a cache group. When a cache group frees an object back to the pool,
118: * it first calls the object's destructor. This allows the object to
119: * persist in constructed form while freed to the cache.
120: *
121: * Multiple caches may exist for each pool. This allows a single
122: * object type to have multiple constructed forms. The pool references
123: * each cache, so that when a pool is drained by the pagedaemon, it can
124: * drain each individual cache as well. Each time a cache is drained,
125: * the most idle cache group is freed to the pool in its entirety.
126: *
127: * Pool caches are layed on top of pools. By layering them, we can avoid
128: * the complexity of cache management for pools which would not benefit
129: * from it.
130: */
131:
132: /* The cache group pool. */
133: static struct pool pcgpool;
134:
135: /* The pool cache group. */
136: #define PCG_NOBJECTS 16
137: struct pool_cache_group {
138: TAILQ_ENTRY(pool_cache_group)
139: pcg_list; /* link in the pool cache's group list */
140: u_int pcg_avail; /* # available objects */
141: /* pointers to the objects */
142: void *pcg_objects[PCG_NOBJECTS];
143: };
1.3 pk 144:
1.43 thorpej 145: static void pool_cache_reclaim(struct pool_cache *);
1.3 pk 146:
1.42 thorpej 147: static int pool_catchup(struct pool *);
1.50 enami 148: static int pool_prime_page(struct pool *, caddr_t, int);
1.42 thorpej 149: static void *pool_page_alloc(unsigned long, int, int);
150: static void pool_page_free(void *, unsigned long, int);
1.3 pk 151:
1.42 thorpej 152: static void pool_print1(struct pool *, const char *,
153: void (*)(const char *, ...));
1.3 pk 154:
155: /*
1.52 ! thorpej 156: * Pool log entry. An array of these is allocated in pool_init().
1.3 pk 157: */
158: struct pool_log {
159: const char *pl_file;
160: long pl_line;
161: int pl_action;
1.25 thorpej 162: #define PRLOG_GET 1
163: #define PRLOG_PUT 2
1.3 pk 164: void *pl_addr;
1.1 pk 165: };
166:
1.3 pk 167: /* Number of entries in pool log buffers */
1.17 thorpej 168: #ifndef POOL_LOGSIZE
169: #define POOL_LOGSIZE 10
170: #endif
171:
172: int pool_logsize = POOL_LOGSIZE;
1.1 pk 173:
1.25 thorpej 174: #ifdef DIAGNOSTIC
1.42 thorpej 175: static __inline void
176: pr_log(struct pool *pp, void *v, int action, const char *file, long line)
1.3 pk 177: {
178: int n = pp->pr_curlogentry;
179: struct pool_log *pl;
180:
1.20 thorpej 181: if ((pp->pr_roflags & PR_LOGGING) == 0)
1.3 pk 182: return;
183:
184: /*
185: * Fill in the current entry. Wrap around and overwrite
186: * the oldest entry if necessary.
187: */
188: pl = &pp->pr_log[n];
189: pl->pl_file = file;
190: pl->pl_line = line;
191: pl->pl_action = action;
192: pl->pl_addr = v;
193: if (++n >= pp->pr_logsize)
194: n = 0;
195: pp->pr_curlogentry = n;
196: }
197:
198: static void
1.42 thorpej 199: pr_printlog(struct pool *pp, struct pool_item *pi,
200: void (*pr)(const char *, ...))
1.3 pk 201: {
202: int i = pp->pr_logsize;
203: int n = pp->pr_curlogentry;
204:
1.20 thorpej 205: if ((pp->pr_roflags & PR_LOGGING) == 0)
1.3 pk 206: return;
207:
208: /*
209: * Print all entries in this pool's log.
210: */
211: while (i-- > 0) {
212: struct pool_log *pl = &pp->pr_log[n];
213: if (pl->pl_action != 0) {
1.25 thorpej 214: if (pi == NULL || pi == pl->pl_addr) {
215: (*pr)("\tlog entry %d:\n", i);
216: (*pr)("\t\taction = %s, addr = %p\n",
217: pl->pl_action == PRLOG_GET ? "get" : "put",
218: pl->pl_addr);
219: (*pr)("\t\tfile: %s at line %lu\n",
220: pl->pl_file, pl->pl_line);
221: }
1.3 pk 222: }
223: if (++n >= pp->pr_logsize)
224: n = 0;
225: }
226: }
1.25 thorpej 227:
1.42 thorpej 228: static __inline void
229: pr_enter(struct pool *pp, const char *file, long line)
1.25 thorpej 230: {
231:
1.34 thorpej 232: if (__predict_false(pp->pr_entered_file != NULL)) {
1.25 thorpej 233: printf("pool %s: reentrancy at file %s line %ld\n",
234: pp->pr_wchan, file, line);
235: printf(" previous entry at file %s line %ld\n",
236: pp->pr_entered_file, pp->pr_entered_line);
237: panic("pr_enter");
238: }
239:
240: pp->pr_entered_file = file;
241: pp->pr_entered_line = line;
242: }
243:
1.42 thorpej 244: static __inline void
245: pr_leave(struct pool *pp)
1.25 thorpej 246: {
247:
1.34 thorpej 248: if (__predict_false(pp->pr_entered_file == NULL)) {
1.25 thorpej 249: printf("pool %s not entered?\n", pp->pr_wchan);
250: panic("pr_leave");
251: }
252:
253: pp->pr_entered_file = NULL;
254: pp->pr_entered_line = 0;
255: }
256:
1.42 thorpej 257: static __inline void
258: pr_enter_check(struct pool *pp, void (*pr)(const char *, ...))
1.25 thorpej 259: {
260:
261: if (pp->pr_entered_file != NULL)
262: (*pr)("\n\tcurrently entered from file %s line %ld\n",
263: pp->pr_entered_file, pp->pr_entered_line);
264: }
1.3 pk 265: #else
1.25 thorpej 266: #define pr_log(pp, v, action, file, line)
267: #define pr_printlog(pp, pi, pr)
268: #define pr_enter(pp, file, line)
269: #define pr_leave(pp)
270: #define pr_enter_check(pp, pr)
271: #endif /* DIAGNOSTIC */
1.3 pk 272:
273: /*
274: * Return the pool page header based on page address.
275: */
1.42 thorpej 276: static __inline struct pool_item_header *
277: pr_find_pagehead(struct pool *pp, caddr_t page)
1.3 pk 278: {
279: struct pool_item_header *ph;
280:
1.20 thorpej 281: if ((pp->pr_roflags & PR_PHINPAGE) != 0)
1.3 pk 282: return ((struct pool_item_header *)(page + pp->pr_phoffset));
283:
284: for (ph = LIST_FIRST(&pp->pr_hashtab[PR_HASH_INDEX(pp, page)]);
285: ph != NULL;
286: ph = LIST_NEXT(ph, ph_hashlist)) {
287: if (ph->ph_page == page)
288: return (ph);
289: }
290: return (NULL);
291: }
292:
293: /*
294: * Remove a page from the pool.
295: */
1.42 thorpej 296: static __inline void
297: pr_rmpage(struct pool *pp, struct pool_item_header *ph)
1.3 pk 298: {
299:
300: /*
1.7 thorpej 301: * If the page was idle, decrement the idle page count.
1.3 pk 302: */
1.6 thorpej 303: if (ph->ph_nmissing == 0) {
304: #ifdef DIAGNOSTIC
305: if (pp->pr_nidle == 0)
306: panic("pr_rmpage: nidle inconsistent");
1.20 thorpej 307: if (pp->pr_nitems < pp->pr_itemsperpage)
308: panic("pr_rmpage: nitems inconsistent");
1.6 thorpej 309: #endif
310: pp->pr_nidle--;
311: }
1.7 thorpej 312:
1.20 thorpej 313: pp->pr_nitems -= pp->pr_itemsperpage;
314:
1.7 thorpej 315: /*
316: * Unlink a page from the pool and release it.
317: */
318: TAILQ_REMOVE(&pp->pr_pagelist, ph, ph_pagelist);
319: (*pp->pr_free)(ph->ph_page, pp->pr_pagesz, pp->pr_mtype);
320: pp->pr_npages--;
321: pp->pr_npagefree++;
1.6 thorpej 322:
1.22 chs 323: if ((pp->pr_roflags & PR_PHINPAGE) == 0) {
1.27 pk 324: int s;
1.22 chs 325: LIST_REMOVE(ph, ph_hashlist);
1.27 pk 326: s = splhigh();
1.22 chs 327: pool_put(&phpool, ph);
1.27 pk 328: splx(s);
1.22 chs 329: }
330:
1.3 pk 331: if (pp->pr_curpage == ph) {
332: /*
333: * Find a new non-empty page header, if any.
334: * Start search from the page head, to increase the
335: * chance for "high water" pages to be freed.
336: */
337: for (ph = TAILQ_FIRST(&pp->pr_pagelist); ph != NULL;
338: ph = TAILQ_NEXT(ph, ph_pagelist))
339: if (TAILQ_FIRST(&ph->ph_itemlist) != NULL)
340: break;
341:
342: pp->pr_curpage = ph;
1.21 thorpej 343: }
1.3 pk 344: }
345:
346: /*
347: * Initialize the given pool resource structure.
348: *
349: * We export this routine to allow other kernel parts to declare
350: * static pools that must be initialized before malloc() is available.
351: */
352: void
1.42 thorpej 353: pool_init(struct pool *pp, size_t size, u_int align, u_int ioff, int flags,
354: const char *wchan, size_t pagesz,
355: void *(*alloc)(unsigned long, int, int),
356: void (*release)(void *, unsigned long, int),
357: int mtype)
1.3 pk 358: {
1.16 briggs 359: int off, slack, i;
1.3 pk 360:
1.25 thorpej 361: #ifdef POOL_DIAGNOSTIC
362: /*
363: * Always log if POOL_DIAGNOSTIC is defined.
364: */
365: if (pool_logsize != 0)
366: flags |= PR_LOGGING;
367: #endif
368:
1.3 pk 369: /*
370: * Check arguments and construct default values.
371: */
1.36 pk 372: if (!powerof2(pagesz))
1.3 pk 373: panic("pool_init: page size invalid (%lx)\n", (u_long)pagesz);
374:
1.4 thorpej 375: if (alloc == NULL && release == NULL) {
1.3 pk 376: alloc = pool_page_alloc;
377: release = pool_page_free;
1.4 thorpej 378: pagesz = PAGE_SIZE; /* Rounds to PAGE_SIZE anyhow. */
379: } else if ((alloc != NULL && release != NULL) == 0) {
380: /* If you specifiy one, must specify both. */
381: panic("pool_init: must specify alloc and release together");
382: }
383:
1.3 pk 384: if (pagesz == 0)
385: pagesz = PAGE_SIZE;
386:
387: if (align == 0)
388: align = ALIGN(1);
1.14 thorpej 389:
390: if (size < sizeof(struct pool_item))
391: size = sizeof(struct pool_item);
1.3 pk 392:
1.35 pk 393: size = ALIGN(size);
1.43 thorpej 394: if (size > pagesz)
1.35 pk 395: panic("pool_init: pool item size (%lu) too large",
396: (u_long)size);
397:
1.3 pk 398: /*
399: * Initialize the pool structure.
400: */
401: TAILQ_INIT(&pp->pr_pagelist);
1.43 thorpej 402: TAILQ_INIT(&pp->pr_cachelist);
1.3 pk 403: pp->pr_curpage = NULL;
404: pp->pr_npages = 0;
405: pp->pr_minitems = 0;
406: pp->pr_minpages = 0;
407: pp->pr_maxpages = UINT_MAX;
1.20 thorpej 408: pp->pr_roflags = flags;
409: pp->pr_flags = 0;
1.35 pk 410: pp->pr_size = size;
1.3 pk 411: pp->pr_align = align;
412: pp->pr_wchan = wchan;
413: pp->pr_mtype = mtype;
414: pp->pr_alloc = alloc;
415: pp->pr_free = release;
416: pp->pr_pagesz = pagesz;
417: pp->pr_pagemask = ~(pagesz - 1);
418: pp->pr_pageshift = ffs(pagesz) - 1;
1.20 thorpej 419: pp->pr_nitems = 0;
420: pp->pr_nout = 0;
421: pp->pr_hardlimit = UINT_MAX;
422: pp->pr_hardlimit_warning = NULL;
1.31 thorpej 423: pp->pr_hardlimit_ratecap.tv_sec = 0;
424: pp->pr_hardlimit_ratecap.tv_usec = 0;
425: pp->pr_hardlimit_warning_last.tv_sec = 0;
426: pp->pr_hardlimit_warning_last.tv_usec = 0;
1.3 pk 427:
428: /*
429: * Decide whether to put the page header off page to avoid
430: * wasting too large a part of the page. Off-page page headers
431: * go on a hash table, so we can match a returned item
432: * with its header based on the page address.
433: * We use 1/16 of the page size as the threshold (XXX: tune)
434: */
435: if (pp->pr_size < pagesz/16) {
436: /* Use the end of the page for the page header */
1.20 thorpej 437: pp->pr_roflags |= PR_PHINPAGE;
1.3 pk 438: pp->pr_phoffset = off =
439: pagesz - ALIGN(sizeof(struct pool_item_header));
1.2 pk 440: } else {
1.3 pk 441: /* The page header will be taken from our page header pool */
442: pp->pr_phoffset = 0;
443: off = pagesz;
1.16 briggs 444: for (i = 0; i < PR_HASHTABSIZE; i++) {
445: LIST_INIT(&pp->pr_hashtab[i]);
446: }
1.2 pk 447: }
1.1 pk 448:
1.3 pk 449: /*
450: * Alignment is to take place at `ioff' within the item. This means
451: * we must reserve up to `align - 1' bytes on the page to allow
452: * appropriate positioning of each item.
453: *
454: * Silently enforce `0 <= ioff < align'.
455: */
456: pp->pr_itemoffset = ioff = ioff % align;
457: pp->pr_itemsperpage = (off - ((align - ioff) % align)) / pp->pr_size;
1.43 thorpej 458: KASSERT(pp->pr_itemsperpage != 0);
1.3 pk 459:
460: /*
461: * Use the slack between the chunks and the page header
462: * for "cache coloring".
463: */
464: slack = off - pp->pr_itemsperpage * pp->pr_size;
465: pp->pr_maxcolor = (slack / align) * align;
466: pp->pr_curcolor = 0;
467:
468: pp->pr_nget = 0;
469: pp->pr_nfail = 0;
470: pp->pr_nput = 0;
471: pp->pr_npagealloc = 0;
472: pp->pr_npagefree = 0;
1.1 pk 473: pp->pr_hiwat = 0;
1.8 thorpej 474: pp->pr_nidle = 0;
1.3 pk 475:
1.25 thorpej 476: if (flags & PR_LOGGING) {
477: if (kmem_map == NULL ||
478: (pp->pr_log = malloc(pool_logsize * sizeof(struct pool_log),
479: M_TEMP, M_NOWAIT)) == NULL)
1.20 thorpej 480: pp->pr_roflags &= ~PR_LOGGING;
1.3 pk 481: pp->pr_curlogentry = 0;
482: pp->pr_logsize = pool_logsize;
483: }
1.25 thorpej 484:
485: pp->pr_entered_file = NULL;
486: pp->pr_entered_line = 0;
1.3 pk 487:
1.21 thorpej 488: simple_lock_init(&pp->pr_slock);
1.1 pk 489:
1.3 pk 490: /*
1.43 thorpej 491: * Initialize private page header pool and cache magazine pool if we
492: * haven't done so yet.
1.23 thorpej 493: * XXX LOCKING.
1.3 pk 494: */
495: if (phpool.pr_size == 0) {
496: pool_init(&phpool, sizeof(struct pool_item_header), 0, 0,
1.43 thorpej 497: 0, "phpool", 0, 0, 0, 0);
498: pool_init(&pcgpool, sizeof(struct pool_cache_group), 0, 0,
499: 0, "pcgpool", 0, 0, 0, 0);
1.1 pk 500: }
501:
1.23 thorpej 502: /* Insert into the list of all pools. */
503: simple_lock(&pool_head_slock);
504: TAILQ_INSERT_TAIL(&pool_head, pp, pr_poollist);
505: simple_unlock(&pool_head_slock);
1.1 pk 506: }
507:
508: /*
509: * De-commision a pool resource.
510: */
511: void
1.42 thorpej 512: pool_destroy(struct pool *pp)
1.1 pk 513: {
1.3 pk 514: struct pool_item_header *ph;
1.43 thorpej 515: struct pool_cache *pc;
516:
517: /* Destroy all caches for this pool. */
518: while ((pc = TAILQ_FIRST(&pp->pr_cachelist)) != NULL)
519: pool_cache_destroy(pc);
1.3 pk 520:
521: #ifdef DIAGNOSTIC
1.20 thorpej 522: if (pp->pr_nout != 0) {
1.25 thorpej 523: pr_printlog(pp, NULL, printf);
1.20 thorpej 524: panic("pool_destroy: pool busy: still out: %u\n",
525: pp->pr_nout);
1.3 pk 526: }
527: #endif
1.1 pk 528:
1.3 pk 529: /* Remove all pages */
1.20 thorpej 530: if ((pp->pr_roflags & PR_STATIC) == 0)
1.3 pk 531: while ((ph = pp->pr_pagelist.tqh_first) != NULL)
532: pr_rmpage(pp, ph);
533:
534: /* Remove from global pool list */
1.23 thorpej 535: simple_lock(&pool_head_slock);
1.3 pk 536: TAILQ_REMOVE(&pool_head, pp, pr_poollist);
1.23 thorpej 537: /* XXX Only clear this if we were drainpp? */
1.3 pk 538: drainpp = NULL;
1.23 thorpej 539: simple_unlock(&pool_head_slock);
1.3 pk 540:
1.20 thorpej 541: if ((pp->pr_roflags & PR_LOGGING) != 0)
1.3 pk 542: free(pp->pr_log, M_TEMP);
1.2 pk 543:
1.20 thorpej 544: if (pp->pr_roflags & PR_FREEHEADER)
1.3 pk 545: free(pp, M_POOL);
1.1 pk 546: }
547:
548:
549: /*
1.3 pk 550: * Grab an item from the pool; must be called at appropriate spl level
1.1 pk 551: */
1.3 pk 552: void *
1.42 thorpej 553: _pool_get(struct pool *pp, int flags, const char *file, long line)
1.1 pk 554: {
555: void *v;
556: struct pool_item *pi;
1.3 pk 557: struct pool_item_header *ph;
1.1 pk 558:
1.2 pk 559: #ifdef DIAGNOSTIC
1.34 thorpej 560: if (__predict_false((pp->pr_roflags & PR_STATIC) &&
561: (flags & PR_MALLOCOK))) {
1.25 thorpej 562: pr_printlog(pp, NULL, printf);
1.2 pk 563: panic("pool_get: static");
1.3 pk 564: }
1.2 pk 565: #endif
566:
1.37 sommerfe 567: if (__predict_false(curproc == NULL && doing_shutdown == 0 &&
568: (flags & PR_WAITOK) != 0))
1.3 pk 569: panic("pool_get: must have NOWAIT");
1.1 pk 570:
1.21 thorpej 571: simple_lock(&pp->pr_slock);
1.25 thorpej 572: pr_enter(pp, file, line);
1.20 thorpej 573:
574: startover:
575: /*
576: * Check to see if we've reached the hard limit. If we have,
577: * and we can wait, then wait until an item has been returned to
578: * the pool.
579: */
580: #ifdef DIAGNOSTIC
1.34 thorpej 581: if (__predict_false(pp->pr_nout > pp->pr_hardlimit)) {
1.25 thorpej 582: pr_leave(pp);
1.21 thorpej 583: simple_unlock(&pp->pr_slock);
1.20 thorpej 584: panic("pool_get: %s: crossed hard limit", pp->pr_wchan);
585: }
586: #endif
1.34 thorpej 587: if (__predict_false(pp->pr_nout == pp->pr_hardlimit)) {
1.29 sommerfe 588: if ((flags & PR_WAITOK) && !(flags & PR_LIMITFAIL)) {
1.20 thorpej 589: /*
590: * XXX: A warning isn't logged in this case. Should
591: * it be?
592: */
593: pp->pr_flags |= PR_WANTED;
1.25 thorpej 594: pr_leave(pp);
1.40 sommerfe 595: ltsleep(pp, PSWP, pp->pr_wchan, 0, &pp->pr_slock);
1.25 thorpej 596: pr_enter(pp, file, line);
1.20 thorpej 597: goto startover;
598: }
1.31 thorpej 599:
600: /*
601: * Log a message that the hard limit has been hit.
602: */
603: if (pp->pr_hardlimit_warning != NULL &&
604: ratecheck(&pp->pr_hardlimit_warning_last,
605: &pp->pr_hardlimit_ratecap))
606: log(LOG_ERR, "%s\n", pp->pr_hardlimit_warning);
1.21 thorpej 607:
608: if (flags & PR_URGENT)
609: panic("pool_get: urgent");
610:
611: pp->pr_nfail++;
612:
1.25 thorpej 613: pr_leave(pp);
1.21 thorpej 614: simple_unlock(&pp->pr_slock);
1.20 thorpej 615: return (NULL);
616: }
617:
1.3 pk 618: /*
619: * The convention we use is that if `curpage' is not NULL, then
620: * it points at a non-empty bucket. In particular, `curpage'
621: * never points at a page header which has PR_PHINPAGE set and
622: * has no items in its bucket.
623: */
1.20 thorpej 624: if ((ph = pp->pr_curpage) == NULL) {
1.15 pk 625: void *v;
626:
1.20 thorpej 627: #ifdef DIAGNOSTIC
628: if (pp->pr_nitems != 0) {
1.21 thorpej 629: simple_unlock(&pp->pr_slock);
1.20 thorpej 630: printf("pool_get: %s: curpage NULL, nitems %u\n",
631: pp->pr_wchan, pp->pr_nitems);
632: panic("pool_get: nitems inconsistent\n");
633: }
634: #endif
635:
1.21 thorpej 636: /*
637: * Call the back-end page allocator for more memory.
638: * Release the pool lock, as the back-end page allocator
639: * may block.
640: */
1.25 thorpej 641: pr_leave(pp);
1.21 thorpej 642: simple_unlock(&pp->pr_slock);
643: v = (*pp->pr_alloc)(pp->pr_pagesz, flags, pp->pr_mtype);
644: simple_lock(&pp->pr_slock);
1.25 thorpej 645: pr_enter(pp, file, line);
1.15 pk 646:
1.21 thorpej 647: if (v == NULL) {
648: /*
649: * We were unable to allocate a page, but
650: * we released the lock during allocation,
651: * so perhaps items were freed back to the
652: * pool. Check for this case.
653: */
654: if (pp->pr_curpage != NULL)
655: goto startover;
1.15 pk 656:
1.3 pk 657: if (flags & PR_URGENT)
658: panic("pool_get: urgent");
1.21 thorpej 659:
1.3 pk 660: if ((flags & PR_WAITOK) == 0) {
661: pp->pr_nfail++;
1.25 thorpej 662: pr_leave(pp);
1.21 thorpej 663: simple_unlock(&pp->pr_slock);
1.1 pk 664: return (NULL);
1.3 pk 665: }
666:
1.15 pk 667: /*
668: * Wait for items to be returned to this pool.
1.21 thorpej 669: *
1.15 pk 670: * XXX: we actually want to wait just until
671: * the page allocator has memory again. Depending
672: * on this pool's usage, we might get stuck here
673: * for a long time.
1.20 thorpej 674: *
675: * XXX: maybe we should wake up once a second and
676: * try again?
1.15 pk 677: */
1.1 pk 678: pp->pr_flags |= PR_WANTED;
1.25 thorpej 679: pr_leave(pp);
1.40 sommerfe 680: ltsleep(pp, PSWP, pp->pr_wchan, 0, &pp->pr_slock);
1.25 thorpej 681: pr_enter(pp, file, line);
1.20 thorpej 682: goto startover;
1.1 pk 683: }
1.3 pk 684:
1.15 pk 685: /* We have more memory; add it to the pool */
1.50 enami 686: if (pool_prime_page(pp, v, flags & PR_WAITOK) != 0) {
687: /*
688: * Probably, we don't allowed to wait and
689: * couldn't allocate a page header.
690: */
691: (*pp->pr_free)(v, pp->pr_pagesz, pp->pr_mtype);
692: pp->pr_nfail++;
693: pr_leave(pp);
694: simple_unlock(&pp->pr_slock);
695: return (NULL);
696: }
1.15 pk 697: pp->pr_npagealloc++;
698:
1.20 thorpej 699: /* Start the allocation process over. */
700: goto startover;
1.3 pk 701: }
702:
1.34 thorpej 703: if (__predict_false((v = pi = TAILQ_FIRST(&ph->ph_itemlist)) == NULL)) {
1.25 thorpej 704: pr_leave(pp);
1.21 thorpej 705: simple_unlock(&pp->pr_slock);
1.3 pk 706: panic("pool_get: %s: page empty", pp->pr_wchan);
1.21 thorpej 707: }
1.20 thorpej 708: #ifdef DIAGNOSTIC
1.34 thorpej 709: if (__predict_false(pp->pr_nitems == 0)) {
1.25 thorpej 710: pr_leave(pp);
1.21 thorpej 711: simple_unlock(&pp->pr_slock);
1.20 thorpej 712: printf("pool_get: %s: items on itemlist, nitems %u\n",
713: pp->pr_wchan, pp->pr_nitems);
714: panic("pool_get: nitems inconsistent\n");
715: }
716: #endif
1.3 pk 717: pr_log(pp, v, PRLOG_GET, file, line);
718:
719: #ifdef DIAGNOSTIC
1.34 thorpej 720: if (__predict_false(pi->pi_magic != PI_MAGIC)) {
1.25 thorpej 721: pr_printlog(pp, pi, printf);
1.3 pk 722: panic("pool_get(%s): free list modified: magic=%x; page %p;"
723: " item addr %p\n",
724: pp->pr_wchan, pi->pi_magic, ph->ph_page, pi);
725: }
726: #endif
727:
728: /*
729: * Remove from item list.
730: */
731: TAILQ_REMOVE(&ph->ph_itemlist, pi, pi_list);
1.20 thorpej 732: pp->pr_nitems--;
733: pp->pr_nout++;
1.6 thorpej 734: if (ph->ph_nmissing == 0) {
735: #ifdef DIAGNOSTIC
1.34 thorpej 736: if (__predict_false(pp->pr_nidle == 0))
1.6 thorpej 737: panic("pool_get: nidle inconsistent");
738: #endif
739: pp->pr_nidle--;
740: }
1.3 pk 741: ph->ph_nmissing++;
742: if (TAILQ_FIRST(&ph->ph_itemlist) == NULL) {
1.21 thorpej 743: #ifdef DIAGNOSTIC
1.34 thorpej 744: if (__predict_false(ph->ph_nmissing != pp->pr_itemsperpage)) {
1.25 thorpej 745: pr_leave(pp);
1.21 thorpej 746: simple_unlock(&pp->pr_slock);
747: panic("pool_get: %s: nmissing inconsistent",
748: pp->pr_wchan);
749: }
750: #endif
1.3 pk 751: /*
752: * Find a new non-empty page header, if any.
753: * Start search from the page head, to increase
754: * the chance for "high water" pages to be freed.
755: *
1.21 thorpej 756: * Migrate empty pages to the end of the list. This
757: * will speed the update of curpage as pages become
758: * idle. Empty pages intermingled with idle pages
759: * is no big deal. As soon as a page becomes un-empty,
760: * it will move back to the head of the list.
1.3 pk 761: */
762: TAILQ_REMOVE(&pp->pr_pagelist, ph, ph_pagelist);
1.21 thorpej 763: TAILQ_INSERT_TAIL(&pp->pr_pagelist, ph, ph_pagelist);
764: for (ph = TAILQ_FIRST(&pp->pr_pagelist); ph != NULL;
765: ph = TAILQ_NEXT(ph, ph_pagelist))
1.3 pk 766: if (TAILQ_FIRST(&ph->ph_itemlist) != NULL)
767: break;
768:
769: pp->pr_curpage = ph;
1.1 pk 770: }
1.3 pk 771:
772: pp->pr_nget++;
1.20 thorpej 773:
774: /*
775: * If we have a low water mark and we are now below that low
776: * water mark, add more items to the pool.
777: */
778: if (pp->pr_nitems < pp->pr_minitems && pool_catchup(pp) != 0) {
779: /*
780: * XXX: Should we log a warning? Should we set up a timeout
781: * to try again in a second or so? The latter could break
782: * a caller's assumptions about interrupt protection, etc.
783: */
784: }
785:
1.25 thorpej 786: pr_leave(pp);
1.21 thorpej 787: simple_unlock(&pp->pr_slock);
1.1 pk 788: return (v);
789: }
790:
791: /*
1.43 thorpej 792: * Internal version of pool_put(). Pool is already locked/entered.
1.1 pk 793: */
1.43 thorpej 794: static void
795: pool_do_put(struct pool *pp, void *v, const char *file, long line)
1.1 pk 796: {
797: struct pool_item *pi = v;
1.3 pk 798: struct pool_item_header *ph;
799: caddr_t page;
1.21 thorpej 800: int s;
1.3 pk 801:
802: page = (caddr_t)((u_long)v & pp->pr_pagemask);
1.1 pk 803:
1.30 thorpej 804: #ifdef DIAGNOSTIC
1.34 thorpej 805: if (__predict_false(pp->pr_nout == 0)) {
1.30 thorpej 806: printf("pool %s: putting with none out\n",
807: pp->pr_wchan);
808: panic("pool_put");
809: }
810: #endif
1.3 pk 811:
812: pr_log(pp, v, PRLOG_PUT, file, line);
813:
1.34 thorpej 814: if (__predict_false((ph = pr_find_pagehead(pp, page)) == NULL)) {
1.25 thorpej 815: pr_printlog(pp, NULL, printf);
1.3 pk 816: panic("pool_put: %s: page header missing", pp->pr_wchan);
817: }
1.28 thorpej 818:
819: #ifdef LOCKDEBUG
820: /*
821: * Check if we're freeing a locked simple lock.
822: */
823: simple_lock_freecheck((caddr_t)pi, ((caddr_t)pi) + pp->pr_size);
824: #endif
1.3 pk 825:
826: /*
827: * Return to item list.
828: */
1.2 pk 829: #ifdef DIAGNOSTIC
1.3 pk 830: pi->pi_magic = PI_MAGIC;
831: #endif
1.32 chs 832: #ifdef DEBUG
833: {
834: int i, *ip = v;
835:
836: for (i = 0; i < pp->pr_size / sizeof(int); i++) {
837: *ip++ = PI_MAGIC;
838: }
839: }
840: #endif
841:
1.3 pk 842: TAILQ_INSERT_HEAD(&ph->ph_itemlist, pi, pi_list);
843: ph->ph_nmissing--;
844: pp->pr_nput++;
1.20 thorpej 845: pp->pr_nitems++;
846: pp->pr_nout--;
1.3 pk 847:
848: /* Cancel "pool empty" condition if it exists */
849: if (pp->pr_curpage == NULL)
850: pp->pr_curpage = ph;
851:
852: if (pp->pr_flags & PR_WANTED) {
853: pp->pr_flags &= ~PR_WANTED;
1.15 pk 854: if (ph->ph_nmissing == 0)
855: pp->pr_nidle++;
1.3 pk 856: wakeup((caddr_t)pp);
857: return;
858: }
859:
860: /*
1.21 thorpej 861: * If this page is now complete, do one of two things:
862: *
863: * (1) If we have more pages than the page high water
864: * mark, free the page back to the system.
865: *
866: * (2) Move it to the end of the page list, so that
867: * we minimize our chances of fragmenting the
868: * pool. Idle pages migrate to the end (along with
869: * completely empty pages, so that we find un-empty
870: * pages more quickly when we update curpage) of the
871: * list so they can be more easily swept up by
872: * the pagedaemon when pages are scarce.
1.3 pk 873: */
874: if (ph->ph_nmissing == 0) {
1.6 thorpej 875: pp->pr_nidle++;
1.3 pk 876: if (pp->pr_npages > pp->pr_maxpages) {
877: pr_rmpage(pp, ph);
878: } else {
879: TAILQ_REMOVE(&pp->pr_pagelist, ph, ph_pagelist);
880: TAILQ_INSERT_TAIL(&pp->pr_pagelist, ph, ph_pagelist);
881:
1.21 thorpej 882: /*
883: * Update the timestamp on the page. A page must
884: * be idle for some period of time before it can
885: * be reclaimed by the pagedaemon. This minimizes
886: * ping-pong'ing for memory.
887: */
888: s = splclock();
889: ph->ph_time = mono_time;
890: splx(s);
891:
892: /*
893: * Update the current page pointer. Just look for
894: * the first page with any free items.
895: *
896: * XXX: Maybe we want an option to look for the
897: * page with the fewest available items, to minimize
898: * fragmentation?
899: */
1.3 pk 900: for (ph = TAILQ_FIRST(&pp->pr_pagelist); ph != NULL;
901: ph = TAILQ_NEXT(ph, ph_pagelist))
902: if (TAILQ_FIRST(&ph->ph_itemlist) != NULL)
903: break;
1.1 pk 904:
1.3 pk 905: pp->pr_curpage = ph;
1.1 pk 906: }
907: }
1.21 thorpej 908: /*
909: * If the page has just become un-empty, move it to the head of
910: * the list, and make it the current page. The next allocation
911: * will get the item from this page, instead of further fragmenting
912: * the pool.
913: */
914: else if (ph->ph_nmissing == (pp->pr_itemsperpage - 1)) {
915: TAILQ_REMOVE(&pp->pr_pagelist, ph, ph_pagelist);
916: TAILQ_INSERT_HEAD(&pp->pr_pagelist, ph, ph_pagelist);
917: pp->pr_curpage = ph;
918: }
1.43 thorpej 919: }
920:
921: /*
922: * Return resource to the pool; must be called at appropriate spl level
923: */
924: void
925: _pool_put(struct pool *pp, void *v, const char *file, long line)
926: {
927:
928: simple_lock(&pp->pr_slock);
929: pr_enter(pp, file, line);
930:
931: pool_do_put(pp, v, file, line);
1.21 thorpej 932:
1.25 thorpej 933: pr_leave(pp);
1.21 thorpej 934: simple_unlock(&pp->pr_slock);
1.1 pk 935: }
936:
937: /*
1.3 pk 938: * Add a page worth of items to the pool.
1.21 thorpej 939: *
940: * Note, we must be called with the pool descriptor LOCKED.
1.3 pk 941: */
1.50 enami 942: static int
943: pool_prime_page(struct pool *pp, caddr_t storage, int flags)
1.3 pk 944: {
945: struct pool_item *pi;
946: struct pool_item_header *ph;
947: caddr_t cp = storage;
948: unsigned int align = pp->pr_align;
949: unsigned int ioff = pp->pr_itemoffset;
1.27 pk 950: int s, n;
1.36 pk 951:
952: if (((u_long)cp & (pp->pr_pagesz - 1)) != 0)
953: panic("pool_prime_page: %s: unaligned page", pp->pr_wchan);
1.3 pk 954:
1.20 thorpej 955: if ((pp->pr_roflags & PR_PHINPAGE) != 0) {
1.3 pk 956: ph = (struct pool_item_header *)(cp + pp->pr_phoffset);
957: } else {
1.27 pk 958: s = splhigh();
1.50 enami 959: ph = pool_get(&phpool, flags);
1.27 pk 960: splx(s);
1.50 enami 961: if (ph == NULL)
962: return (ENOMEM);
1.3 pk 963: LIST_INSERT_HEAD(&pp->pr_hashtab[PR_HASH_INDEX(pp, cp)],
964: ph, ph_hashlist);
965: }
966:
967: /*
968: * Insert page header.
969: */
970: TAILQ_INSERT_HEAD(&pp->pr_pagelist, ph, ph_pagelist);
971: TAILQ_INIT(&ph->ph_itemlist);
972: ph->ph_page = storage;
973: ph->ph_nmissing = 0;
1.21 thorpej 974: memset(&ph->ph_time, 0, sizeof(ph->ph_time));
1.3 pk 975:
1.6 thorpej 976: pp->pr_nidle++;
977:
1.3 pk 978: /*
979: * Color this page.
980: */
981: cp = (caddr_t)(cp + pp->pr_curcolor);
982: if ((pp->pr_curcolor += align) > pp->pr_maxcolor)
983: pp->pr_curcolor = 0;
984:
985: /*
986: * Adjust storage to apply aligment to `pr_itemoffset' in each item.
987: */
988: if (ioff != 0)
989: cp = (caddr_t)(cp + (align - ioff));
990:
991: /*
992: * Insert remaining chunks on the bucket list.
993: */
994: n = pp->pr_itemsperpage;
1.20 thorpej 995: pp->pr_nitems += n;
1.3 pk 996:
997: while (n--) {
998: pi = (struct pool_item *)cp;
999:
1000: /* Insert on page list */
1001: TAILQ_INSERT_TAIL(&ph->ph_itemlist, pi, pi_list);
1002: #ifdef DIAGNOSTIC
1003: pi->pi_magic = PI_MAGIC;
1004: #endif
1005: cp = (caddr_t)(cp + pp->pr_size);
1006: }
1007:
1008: /*
1009: * If the pool was depleted, point at the new page.
1010: */
1011: if (pp->pr_curpage == NULL)
1012: pp->pr_curpage = ph;
1013:
1014: if (++pp->pr_npages > pp->pr_hiwat)
1015: pp->pr_hiwat = pp->pr_npages;
1.50 enami 1016:
1017: return (0);
1.3 pk 1018: }
1019:
1.20 thorpej 1020: /*
1.52 ! thorpej 1021: * Used by pool_get() when nitems drops below the low water mark. This
! 1022: * is used to catch up nitmes with the low water mark.
1.20 thorpej 1023: *
1.21 thorpej 1024: * Note 1, we never wait for memory here, we let the caller decide what to do.
1.20 thorpej 1025: *
1026: * Note 2, this doesn't work with static pools.
1027: *
1028: * Note 3, we must be called with the pool already locked, and we return
1029: * with it locked.
1030: */
1031: static int
1.42 thorpej 1032: pool_catchup(struct pool *pp)
1.20 thorpej 1033: {
1034: caddr_t cp;
1035: int error = 0;
1036:
1037: if (pp->pr_roflags & PR_STATIC) {
1038: /*
1039: * We dropped below the low water mark, and this is not a
1040: * good thing. Log a warning.
1.21 thorpej 1041: *
1042: * XXX: rate-limit this?
1.20 thorpej 1043: */
1044: printf("WARNING: static pool `%s' dropped below low water "
1045: "mark\n", pp->pr_wchan);
1046: return (0);
1047: }
1048:
1.21 thorpej 1049: while (pp->pr_nitems < pp->pr_minitems) {
1.20 thorpej 1050: /*
1.21 thorpej 1051: * Call the page back-end allocator for more memory.
1052: *
1053: * XXX: We never wait, so should we bother unlocking
1054: * the pool descriptor?
1.20 thorpej 1055: */
1.21 thorpej 1056: simple_unlock(&pp->pr_slock);
1.20 thorpej 1057: cp = (*pp->pr_alloc)(pp->pr_pagesz, 0, pp->pr_mtype);
1.21 thorpej 1058: simple_lock(&pp->pr_slock);
1.34 thorpej 1059: if (__predict_false(cp == NULL)) {
1.20 thorpej 1060: error = ENOMEM;
1061: break;
1062: }
1.50 enami 1063: if ((error = pool_prime_page(pp, cp, PR_NOWAIT)) != 0) {
1064: (*pp->pr_free)(cp, pp->pr_pagesz, pp->pr_mtype);
1065: break;
1066: }
1.26 thorpej 1067: pp->pr_npagealloc++;
1.20 thorpej 1068: }
1069:
1070: return (error);
1071: }
1072:
1.3 pk 1073: void
1.42 thorpej 1074: pool_setlowat(struct pool *pp, int n)
1.3 pk 1075: {
1.20 thorpej 1076: int error;
1.15 pk 1077:
1.21 thorpej 1078: simple_lock(&pp->pr_slock);
1079:
1.3 pk 1080: pp->pr_minitems = n;
1.15 pk 1081: pp->pr_minpages = (n == 0)
1082: ? 0
1.18 thorpej 1083: : roundup(n, pp->pr_itemsperpage) / pp->pr_itemsperpage;
1.20 thorpej 1084:
1085: /* Make sure we're caught up with the newly-set low water mark. */
1.41 sommerfe 1086: if ((pp->pr_nitems < pp->pr_minitems) &&
1087: (error = pool_catchup(pp)) != 0) {
1.20 thorpej 1088: /*
1089: * XXX: Should we log a warning? Should we set up a timeout
1090: * to try again in a second or so? The latter could break
1091: * a caller's assumptions about interrupt protection, etc.
1092: */
1093: }
1.21 thorpej 1094:
1095: simple_unlock(&pp->pr_slock);
1.3 pk 1096: }
1097:
1098: void
1.42 thorpej 1099: pool_sethiwat(struct pool *pp, int n)
1.3 pk 1100: {
1.15 pk 1101:
1.21 thorpej 1102: simple_lock(&pp->pr_slock);
1103:
1.15 pk 1104: pp->pr_maxpages = (n == 0)
1105: ? 0
1.18 thorpej 1106: : roundup(n, pp->pr_itemsperpage) / pp->pr_itemsperpage;
1.21 thorpej 1107:
1108: simple_unlock(&pp->pr_slock);
1.3 pk 1109: }
1110:
1.20 thorpej 1111: void
1.42 thorpej 1112: pool_sethardlimit(struct pool *pp, int n, const char *warnmess, int ratecap)
1.20 thorpej 1113: {
1114:
1.21 thorpej 1115: simple_lock(&pp->pr_slock);
1.20 thorpej 1116:
1117: pp->pr_hardlimit = n;
1118: pp->pr_hardlimit_warning = warnmess;
1.31 thorpej 1119: pp->pr_hardlimit_ratecap.tv_sec = ratecap;
1120: pp->pr_hardlimit_warning_last.tv_sec = 0;
1121: pp->pr_hardlimit_warning_last.tv_usec = 0;
1.20 thorpej 1122:
1123: /*
1.21 thorpej 1124: * In-line version of pool_sethiwat(), because we don't want to
1125: * release the lock.
1.20 thorpej 1126: */
1127: pp->pr_maxpages = (n == 0)
1128: ? 0
1129: : roundup(n, pp->pr_itemsperpage) / pp->pr_itemsperpage;
1.21 thorpej 1130:
1131: simple_unlock(&pp->pr_slock);
1.20 thorpej 1132: }
1.3 pk 1133:
1134: /*
1135: * Default page allocator.
1136: */
1137: static void *
1.42 thorpej 1138: pool_page_alloc(unsigned long sz, int flags, int mtype)
1.3 pk 1139: {
1.11 thorpej 1140: boolean_t waitok = (flags & PR_WAITOK) ? TRUE : FALSE;
1.3 pk 1141:
1.11 thorpej 1142: return ((void *)uvm_km_alloc_poolpage(waitok));
1.3 pk 1143: }
1144:
1145: static void
1.42 thorpej 1146: pool_page_free(void *v, unsigned long sz, int mtype)
1.3 pk 1147: {
1148:
1.10 eeh 1149: uvm_km_free_poolpage((vaddr_t)v);
1.3 pk 1150: }
1.12 thorpej 1151:
1152: /*
1153: * Alternate pool page allocator for pools that know they will
1154: * never be accessed in interrupt context.
1155: */
1156: void *
1.42 thorpej 1157: pool_page_alloc_nointr(unsigned long sz, int flags, int mtype)
1.12 thorpej 1158: {
1159: boolean_t waitok = (flags & PR_WAITOK) ? TRUE : FALSE;
1160:
1161: return ((void *)uvm_km_alloc_poolpage1(kernel_map, uvm.kernel_object,
1162: waitok));
1163: }
1164:
1165: void
1.42 thorpej 1166: pool_page_free_nointr(void *v, unsigned long sz, int mtype)
1.12 thorpej 1167: {
1168:
1169: uvm_km_free_poolpage1(kernel_map, (vaddr_t)v);
1170: }
1171:
1.3 pk 1172:
1173: /*
1174: * Release all complete pages that have not been used recently.
1175: */
1176: void
1.42 thorpej 1177: _pool_reclaim(struct pool *pp, const char *file, long line)
1.3 pk 1178: {
1179: struct pool_item_header *ph, *phnext;
1.43 thorpej 1180: struct pool_cache *pc;
1.21 thorpej 1181: struct timeval curtime;
1182: int s;
1.3 pk 1183:
1.20 thorpej 1184: if (pp->pr_roflags & PR_STATIC)
1.3 pk 1185: return;
1186:
1.21 thorpej 1187: if (simple_lock_try(&pp->pr_slock) == 0)
1.3 pk 1188: return;
1.25 thorpej 1189: pr_enter(pp, file, line);
1.3 pk 1190:
1.43 thorpej 1191: /*
1192: * Reclaim items from the pool's caches.
1193: */
1194: for (pc = TAILQ_FIRST(&pp->pr_cachelist); pc != NULL;
1195: pc = TAILQ_NEXT(pc, pc_poollist))
1196: pool_cache_reclaim(pc);
1197:
1.21 thorpej 1198: s = splclock();
1199: curtime = mono_time;
1200: splx(s);
1201:
1.3 pk 1202: for (ph = TAILQ_FIRST(&pp->pr_pagelist); ph != NULL; ph = phnext) {
1203: phnext = TAILQ_NEXT(ph, ph_pagelist);
1204:
1205: /* Check our minimum page claim */
1206: if (pp->pr_npages <= pp->pr_minpages)
1207: break;
1208:
1209: if (ph->ph_nmissing == 0) {
1210: struct timeval diff;
1211: timersub(&curtime, &ph->ph_time, &diff);
1212: if (diff.tv_sec < pool_inactive_time)
1213: continue;
1.21 thorpej 1214:
1215: /*
1216: * If freeing this page would put us below
1217: * the low water mark, stop now.
1218: */
1219: if ((pp->pr_nitems - pp->pr_itemsperpage) <
1220: pp->pr_minitems)
1221: break;
1222:
1.3 pk 1223: pr_rmpage(pp, ph);
1224: }
1225: }
1226:
1.25 thorpej 1227: pr_leave(pp);
1.21 thorpej 1228: simple_unlock(&pp->pr_slock);
1.3 pk 1229: }
1230:
1231:
1232: /*
1233: * Drain pools, one at a time.
1.21 thorpej 1234: *
1235: * Note, we must never be called from an interrupt context.
1.3 pk 1236: */
1237: void
1.42 thorpej 1238: pool_drain(void *arg)
1.3 pk 1239: {
1240: struct pool *pp;
1.23 thorpej 1241: int s;
1.3 pk 1242:
1.49 thorpej 1243: s = splvm();
1.23 thorpej 1244: simple_lock(&pool_head_slock);
1245:
1246: if (drainpp == NULL && (drainpp = TAILQ_FIRST(&pool_head)) == NULL)
1247: goto out;
1.3 pk 1248:
1249: pp = drainpp;
1250: drainpp = TAILQ_NEXT(pp, pr_poollist);
1251:
1252: pool_reclaim(pp);
1.23 thorpej 1253:
1254: out:
1255: simple_unlock(&pool_head_slock);
1.3 pk 1256: splx(s);
1257: }
1258:
1259:
1260: /*
1261: * Diagnostic helpers.
1262: */
1263: void
1.42 thorpej 1264: pool_print(struct pool *pp, const char *modif)
1.21 thorpej 1265: {
1266: int s;
1267:
1.49 thorpej 1268: s = splvm();
1.25 thorpej 1269: if (simple_lock_try(&pp->pr_slock) == 0) {
1270: printf("pool %s is locked; try again later\n",
1271: pp->pr_wchan);
1272: splx(s);
1273: return;
1274: }
1275: pool_print1(pp, modif, printf);
1.21 thorpej 1276: simple_unlock(&pp->pr_slock);
1277: splx(s);
1278: }
1279:
1.25 thorpej 1280: void
1.42 thorpej 1281: pool_printit(struct pool *pp, const char *modif, void (*pr)(const char *, ...))
1.25 thorpej 1282: {
1283: int didlock = 0;
1284:
1285: if (pp == NULL) {
1286: (*pr)("Must specify a pool to print.\n");
1287: return;
1288: }
1289:
1290: /*
1291: * Called from DDB; interrupts should be blocked, and all
1292: * other processors should be paused. We can skip locking
1293: * the pool in this case.
1294: *
1295: * We do a simple_lock_try() just to print the lock
1296: * status, however.
1297: */
1298:
1299: if (simple_lock_try(&pp->pr_slock) == 0)
1300: (*pr)("WARNING: pool %s is locked\n", pp->pr_wchan);
1301: else
1302: didlock = 1;
1303:
1304: pool_print1(pp, modif, pr);
1305:
1306: if (didlock)
1307: simple_unlock(&pp->pr_slock);
1308: }
1309:
1.21 thorpej 1310: static void
1.42 thorpej 1311: pool_print1(struct pool *pp, const char *modif, void (*pr)(const char *, ...))
1.3 pk 1312: {
1.25 thorpej 1313: struct pool_item_header *ph;
1.44 thorpej 1314: struct pool_cache *pc;
1315: struct pool_cache_group *pcg;
1.25 thorpej 1316: #ifdef DIAGNOSTIC
1317: struct pool_item *pi;
1318: #endif
1.44 thorpej 1319: int i, print_log = 0, print_pagelist = 0, print_cache = 0;
1.25 thorpej 1320: char c;
1321:
1322: while ((c = *modif++) != '\0') {
1323: if (c == 'l')
1324: print_log = 1;
1325: if (c == 'p')
1326: print_pagelist = 1;
1.44 thorpej 1327: if (c == 'c')
1328: print_cache = 1;
1.25 thorpej 1329: modif++;
1330: }
1331:
1332: (*pr)("POOL %s: size %u, align %u, ioff %u, roflags 0x%08x\n",
1333: pp->pr_wchan, pp->pr_size, pp->pr_align, pp->pr_itemoffset,
1334: pp->pr_roflags);
1335: (*pr)("\tpagesz %u, mtype %d\n", pp->pr_pagesz, pp->pr_mtype);
1336: (*pr)("\talloc %p, release %p\n", pp->pr_alloc, pp->pr_free);
1337: (*pr)("\tminitems %u, minpages %u, maxpages %u, npages %u\n",
1338: pp->pr_minitems, pp->pr_minpages, pp->pr_maxpages, pp->pr_npages);
1339: (*pr)("\titemsperpage %u, nitems %u, nout %u, hardlimit %u\n",
1340: pp->pr_itemsperpage, pp->pr_nitems, pp->pr_nout, pp->pr_hardlimit);
1341:
1342: (*pr)("\n\tnget %lu, nfail %lu, nput %lu\n",
1343: pp->pr_nget, pp->pr_nfail, pp->pr_nput);
1344: (*pr)("\tnpagealloc %lu, npagefree %lu, hiwat %u, nidle %lu\n",
1345: pp->pr_npagealloc, pp->pr_npagefree, pp->pr_hiwat, pp->pr_nidle);
1346:
1347: if (print_pagelist == 0)
1348: goto skip_pagelist;
1349:
1350: if ((ph = TAILQ_FIRST(&pp->pr_pagelist)) != NULL)
1351: (*pr)("\n\tpage list:\n");
1352: for (; ph != NULL; ph = TAILQ_NEXT(ph, ph_pagelist)) {
1353: (*pr)("\t\tpage %p, nmissing %d, time %lu,%lu\n",
1354: ph->ph_page, ph->ph_nmissing,
1355: (u_long)ph->ph_time.tv_sec,
1356: (u_long)ph->ph_time.tv_usec);
1357: #ifdef DIAGNOSTIC
1358: for (pi = TAILQ_FIRST(&ph->ph_itemlist); pi != NULL;
1359: pi = TAILQ_NEXT(pi, pi_list)) {
1360: if (pi->pi_magic != PI_MAGIC) {
1361: (*pr)("\t\t\titem %p, magic 0x%x\n",
1362: pi, pi->pi_magic);
1363: }
1364: }
1365: #endif
1366: }
1367: if (pp->pr_curpage == NULL)
1368: (*pr)("\tno current page\n");
1369: else
1370: (*pr)("\tcurpage %p\n", pp->pr_curpage->ph_page);
1371:
1372: skip_pagelist:
1373:
1374: if (print_log == 0)
1375: goto skip_log;
1376:
1377: (*pr)("\n");
1378: if ((pp->pr_roflags & PR_LOGGING) == 0)
1379: (*pr)("\tno log\n");
1380: else
1381: pr_printlog(pp, NULL, pr);
1.3 pk 1382:
1.25 thorpej 1383: skip_log:
1.44 thorpej 1384:
1385: if (print_cache == 0)
1386: goto skip_cache;
1387:
1388: for (pc = TAILQ_FIRST(&pp->pr_cachelist); pc != NULL;
1389: pc = TAILQ_NEXT(pc, pc_poollist)) {
1390: (*pr)("\tcache %p: allocfrom %p freeto %p\n", pc,
1391: pc->pc_allocfrom, pc->pc_freeto);
1.48 thorpej 1392: (*pr)("\t hits %lu misses %lu ngroups %lu nitems %lu\n",
1393: pc->pc_hits, pc->pc_misses, pc->pc_ngroups, pc->pc_nitems);
1.44 thorpej 1394: for (pcg = TAILQ_FIRST(&pc->pc_grouplist); pcg != NULL;
1395: pcg = TAILQ_NEXT(pcg, pcg_list)) {
1396: (*pr)("\t\tgroup %p: avail %d\n", pcg, pcg->pcg_avail);
1397: for (i = 0; i < PCG_NOBJECTS; i++)
1398: (*pr)("\t\t\t%p\n", pcg->pcg_objects[i]);
1399: }
1400: }
1401:
1402: skip_cache:
1.3 pk 1403:
1.25 thorpej 1404: pr_enter_check(pp, pr);
1.3 pk 1405: }
1406:
1407: int
1.42 thorpej 1408: pool_chk(struct pool *pp, const char *label)
1.3 pk 1409: {
1410: struct pool_item_header *ph;
1411: int r = 0;
1412:
1.21 thorpej 1413: simple_lock(&pp->pr_slock);
1.3 pk 1414:
1415: for (ph = TAILQ_FIRST(&pp->pr_pagelist); ph != NULL;
1416: ph = TAILQ_NEXT(ph, ph_pagelist)) {
1417:
1418: struct pool_item *pi;
1419: int n;
1420: caddr_t page;
1421:
1422: page = (caddr_t)((u_long)ph & pp->pr_pagemask);
1.20 thorpej 1423: if (page != ph->ph_page &&
1424: (pp->pr_roflags & PR_PHINPAGE) != 0) {
1.3 pk 1425: if (label != NULL)
1426: printf("%s: ", label);
1.16 briggs 1427: printf("pool(%p:%s): page inconsistency: page %p;"
1428: " at page head addr %p (p %p)\n", pp,
1.3 pk 1429: pp->pr_wchan, ph->ph_page,
1430: ph, page);
1431: r++;
1432: goto out;
1433: }
1434:
1435: for (pi = TAILQ_FIRST(&ph->ph_itemlist), n = 0;
1436: pi != NULL;
1437: pi = TAILQ_NEXT(pi,pi_list), n++) {
1438:
1439: #ifdef DIAGNOSTIC
1440: if (pi->pi_magic != PI_MAGIC) {
1441: if (label != NULL)
1442: printf("%s: ", label);
1443: printf("pool(%s): free list modified: magic=%x;"
1444: " page %p; item ordinal %d;"
1445: " addr %p (p %p)\n",
1446: pp->pr_wchan, pi->pi_magic, ph->ph_page,
1447: n, pi, page);
1448: panic("pool");
1449: }
1450: #endif
1451: page = (caddr_t)((u_long)pi & pp->pr_pagemask);
1452: if (page == ph->ph_page)
1453: continue;
1454:
1455: if (label != NULL)
1456: printf("%s: ", label);
1.16 briggs 1457: printf("pool(%p:%s): page inconsistency: page %p;"
1458: " item ordinal %d; addr %p (p %p)\n", pp,
1.3 pk 1459: pp->pr_wchan, ph->ph_page,
1460: n, pi, page);
1461: r++;
1462: goto out;
1463: }
1464: }
1465: out:
1.21 thorpej 1466: simple_unlock(&pp->pr_slock);
1.3 pk 1467: return (r);
1.43 thorpej 1468: }
1469:
1470: /*
1471: * pool_cache_init:
1472: *
1473: * Initialize a pool cache.
1474: *
1475: * NOTE: If the pool must be protected from interrupts, we expect
1476: * to be called at the appropriate interrupt priority level.
1477: */
1478: void
1479: pool_cache_init(struct pool_cache *pc, struct pool *pp,
1480: int (*ctor)(void *, void *, int),
1481: void (*dtor)(void *, void *),
1482: void *arg)
1483: {
1484:
1485: TAILQ_INIT(&pc->pc_grouplist);
1486: simple_lock_init(&pc->pc_slock);
1487:
1488: pc->pc_allocfrom = NULL;
1489: pc->pc_freeto = NULL;
1490: pc->pc_pool = pp;
1491:
1492: pc->pc_ctor = ctor;
1493: pc->pc_dtor = dtor;
1494: pc->pc_arg = arg;
1495:
1.48 thorpej 1496: pc->pc_hits = 0;
1497: pc->pc_misses = 0;
1498:
1499: pc->pc_ngroups = 0;
1500:
1501: pc->pc_nitems = 0;
1502:
1.43 thorpej 1503: simple_lock(&pp->pr_slock);
1504: TAILQ_INSERT_TAIL(&pp->pr_cachelist, pc, pc_poollist);
1505: simple_unlock(&pp->pr_slock);
1506: }
1507:
1508: /*
1509: * pool_cache_destroy:
1510: *
1511: * Destroy a pool cache.
1512: */
1513: void
1514: pool_cache_destroy(struct pool_cache *pc)
1515: {
1516: struct pool *pp = pc->pc_pool;
1517:
1518: /* First, invalidate the entire cache. */
1519: pool_cache_invalidate(pc);
1520:
1521: /* ...and remove it from the pool's cache list. */
1522: simple_lock(&pp->pr_slock);
1523: TAILQ_REMOVE(&pp->pr_cachelist, pc, pc_poollist);
1524: simple_unlock(&pp->pr_slock);
1525: }
1526:
1527: static __inline void *
1528: pcg_get(struct pool_cache_group *pcg)
1529: {
1530: void *object;
1531: u_int idx;
1532:
1533: KASSERT(pcg->pcg_avail <= PCG_NOBJECTS);
1.45 thorpej 1534: KASSERT(pcg->pcg_avail != 0);
1.43 thorpej 1535: idx = --pcg->pcg_avail;
1536:
1537: KASSERT(pcg->pcg_objects[idx] != NULL);
1538: object = pcg->pcg_objects[idx];
1539: pcg->pcg_objects[idx] = NULL;
1540:
1541: return (object);
1542: }
1543:
1544: static __inline void
1545: pcg_put(struct pool_cache_group *pcg, void *object)
1546: {
1547: u_int idx;
1548:
1549: KASSERT(pcg->pcg_avail < PCG_NOBJECTS);
1550: idx = pcg->pcg_avail++;
1551:
1552: KASSERT(pcg->pcg_objects[idx] == NULL);
1553: pcg->pcg_objects[idx] = object;
1554: }
1555:
1556: /*
1557: * pool_cache_get:
1558: *
1559: * Get an object from a pool cache.
1560: */
1561: void *
1562: pool_cache_get(struct pool_cache *pc, int flags)
1563: {
1564: struct pool_cache_group *pcg;
1565: void *object;
1566:
1567: simple_lock(&pc->pc_slock);
1568:
1569: if ((pcg = pc->pc_allocfrom) == NULL) {
1570: for (pcg = TAILQ_FIRST(&pc->pc_grouplist); pcg != NULL;
1571: pcg = TAILQ_NEXT(pcg, pcg_list)) {
1572: if (pcg->pcg_avail != 0) {
1573: pc->pc_allocfrom = pcg;
1574: goto have_group;
1575: }
1576: }
1577:
1578: /*
1579: * No groups with any available objects. Allocate
1580: * a new object, construct it, and return it to
1581: * the caller. We will allocate a group, if necessary,
1582: * when the object is freed back to the cache.
1583: */
1.48 thorpej 1584: pc->pc_misses++;
1.43 thorpej 1585: simple_unlock(&pc->pc_slock);
1586: object = pool_get(pc->pc_pool, flags);
1587: if (object != NULL && pc->pc_ctor != NULL) {
1588: if ((*pc->pc_ctor)(pc->pc_arg, object, flags) != 0) {
1589: pool_put(pc->pc_pool, object);
1590: return (NULL);
1591: }
1592: }
1593: return (object);
1594: }
1595:
1596: have_group:
1.48 thorpej 1597: pc->pc_hits++;
1598: pc->pc_nitems--;
1.43 thorpej 1599: object = pcg_get(pcg);
1600:
1601: if (pcg->pcg_avail == 0)
1602: pc->pc_allocfrom = NULL;
1.45 thorpej 1603:
1.43 thorpej 1604: simple_unlock(&pc->pc_slock);
1605:
1606: return (object);
1607: }
1608:
1609: /*
1610: * pool_cache_put:
1611: *
1612: * Put an object back to the pool cache.
1613: */
1614: void
1615: pool_cache_put(struct pool_cache *pc, void *object)
1616: {
1617: struct pool_cache_group *pcg;
1618:
1619: simple_lock(&pc->pc_slock);
1620:
1621: if ((pcg = pc->pc_freeto) == NULL) {
1622: for (pcg = TAILQ_FIRST(&pc->pc_grouplist); pcg != NULL;
1623: pcg = TAILQ_NEXT(pcg, pcg_list)) {
1624: if (pcg->pcg_avail != PCG_NOBJECTS) {
1625: pc->pc_freeto = pcg;
1626: goto have_group;
1627: }
1628: }
1629:
1630: /*
1631: * No empty groups to free the object to. Attempt to
1.47 thorpej 1632: * allocate one.
1.43 thorpej 1633: */
1.47 thorpej 1634: simple_unlock(&pc->pc_slock);
1.43 thorpej 1635: pcg = pool_get(&pcgpool, PR_NOWAIT);
1636: if (pcg != NULL) {
1637: memset(pcg, 0, sizeof(*pcg));
1.47 thorpej 1638: simple_lock(&pc->pc_slock);
1.48 thorpej 1639: pc->pc_ngroups++;
1.43 thorpej 1640: TAILQ_INSERT_TAIL(&pc->pc_grouplist, pcg, pcg_list);
1.47 thorpej 1641: if (pc->pc_freeto == NULL)
1642: pc->pc_freeto = pcg;
1.43 thorpej 1643: goto have_group;
1644: }
1645:
1646: /*
1647: * Unable to allocate a cache group; destruct the object
1648: * and free it back to the pool.
1649: */
1.51 thorpej 1650: pool_cache_destruct_object(pc, object);
1.43 thorpej 1651: return;
1652: }
1653:
1654: have_group:
1.48 thorpej 1655: pc->pc_nitems++;
1.43 thorpej 1656: pcg_put(pcg, object);
1657:
1658: if (pcg->pcg_avail == PCG_NOBJECTS)
1659: pc->pc_freeto = NULL;
1660:
1661: simple_unlock(&pc->pc_slock);
1.51 thorpej 1662: }
1663:
1664: /*
1665: * pool_cache_destruct_object:
1666: *
1667: * Force destruction of an object and its release back into
1668: * the pool.
1669: */
1670: void
1671: pool_cache_destruct_object(struct pool_cache *pc, void *object)
1672: {
1673:
1674: if (pc->pc_dtor != NULL)
1675: (*pc->pc_dtor)(pc->pc_arg, object);
1676: pool_put(pc->pc_pool, object);
1.43 thorpej 1677: }
1678:
1679: /*
1680: * pool_cache_do_invalidate:
1681: *
1682: * This internal function implements pool_cache_invalidate() and
1683: * pool_cache_reclaim().
1684: */
1685: static void
1686: pool_cache_do_invalidate(struct pool_cache *pc, int free_groups,
1687: void (*putit)(struct pool *, void *, const char *, long))
1688: {
1689: struct pool_cache_group *pcg, *npcg;
1690: void *object;
1691:
1692: for (pcg = TAILQ_FIRST(&pc->pc_grouplist); pcg != NULL;
1693: pcg = npcg) {
1694: npcg = TAILQ_NEXT(pcg, pcg_list);
1695: while (pcg->pcg_avail != 0) {
1.48 thorpej 1696: pc->pc_nitems--;
1.43 thorpej 1697: object = pcg_get(pcg);
1.45 thorpej 1698: if (pcg->pcg_avail == 0 && pc->pc_allocfrom == pcg)
1699: pc->pc_allocfrom = NULL;
1.43 thorpej 1700: if (pc->pc_dtor != NULL)
1701: (*pc->pc_dtor)(pc->pc_arg, object);
1702: (*putit)(pc->pc_pool, object, __FILE__, __LINE__);
1703: }
1704: if (free_groups) {
1.48 thorpej 1705: pc->pc_ngroups--;
1.43 thorpej 1706: TAILQ_REMOVE(&pc->pc_grouplist, pcg, pcg_list);
1.46 thorpej 1707: if (pc->pc_freeto == pcg)
1708: pc->pc_freeto = NULL;
1.43 thorpej 1709: pool_put(&pcgpool, pcg);
1710: }
1711: }
1712: }
1713:
1714: /*
1715: * pool_cache_invalidate:
1716: *
1717: * Invalidate a pool cache (destruct and release all of the
1718: * cached objects).
1719: */
1720: void
1721: pool_cache_invalidate(struct pool_cache *pc)
1722: {
1723:
1724: simple_lock(&pc->pc_slock);
1725: pool_cache_do_invalidate(pc, 0, _pool_put);
1726: simple_unlock(&pc->pc_slock);
1727: }
1728:
1729: /*
1730: * pool_cache_reclaim:
1731: *
1732: * Reclaim a pool cache for pool_reclaim().
1733: */
1734: static void
1735: pool_cache_reclaim(struct pool_cache *pc)
1736: {
1737:
1.47 thorpej 1738: simple_lock(&pc->pc_slock);
1.43 thorpej 1739: pool_cache_do_invalidate(pc, 1, pool_do_put);
1740: simple_unlock(&pc->pc_slock);
1.3 pk 1741: }
CVSweb <webmaster@jp.NetBSD.org>
![[BACK]](/icons/back.gif){kind=icon}
Return to subr_pool.c CVS log ![[TXT]](/icons/text.gif){kind=icon}
![[DIR]](/icons/dir.gif){kind=icon}
Up to [cvs.NetBSD.org] / src / sys / kern