Annotation of src/sys/kern/subr_pool.c, Revision 1.39
1.39 ! mrg 1: /* $NetBSD: subr_pool.c,v 1.38 2000/06/26 14:21:14 mrg Exp $ */
1.1 pk 2:
3: /*-
1.20 thorpej 4: * Copyright (c) 1997, 1999 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:
103:
1.25 thorpej 104: #define PR_HASH_INDEX(pp,addr) \
1.3 pk 105: (((u_long)(addr) >> (pp)->pr_pageshift) & (PR_HASHTABSIZE - 1))
106:
107:
108:
109: static struct pool_item_header
110: *pr_find_pagehead __P((struct pool *, caddr_t));
111: static void pr_rmpage __P((struct pool *, struct pool_item_header *));
1.20 thorpej 112: static int pool_catchup __P((struct pool *));
1.21 thorpej 113: static void pool_prime_page __P((struct pool *, caddr_t));
1.3 pk 114: static void *pool_page_alloc __P((unsigned long, int, int));
115: static void pool_page_free __P((void *, unsigned long, int));
116:
1.25 thorpej 117: static void pool_print1 __P((struct pool *, const char *,
118: void (*)(const char *, ...)));
1.3 pk 119:
120: /*
121: * Pool log entry. An array of these is allocated in pool_create().
122: */
123: struct pool_log {
124: const char *pl_file;
125: long pl_line;
126: int pl_action;
1.25 thorpej 127: #define PRLOG_GET 1
128: #define PRLOG_PUT 2
1.3 pk 129: void *pl_addr;
1.1 pk 130: };
131:
1.3 pk 132: /* Number of entries in pool log buffers */
1.17 thorpej 133: #ifndef POOL_LOGSIZE
134: #define POOL_LOGSIZE 10
135: #endif
136:
137: int pool_logsize = POOL_LOGSIZE;
1.1 pk 138:
1.25 thorpej 139: #ifdef DIAGNOSTIC
1.3 pk 140: static void pr_log __P((struct pool *, void *, int, const char *, long));
1.25 thorpej 141: static void pr_printlog __P((struct pool *, struct pool_item *,
142: void (*)(const char *, ...)));
143: static void pr_enter __P((struct pool *, const char *, long));
144: static void pr_leave __P((struct pool *));
145: static void pr_enter_check __P((struct pool *,
146: void (*)(const char *, ...)));
1.3 pk 147:
148: static __inline__ void
149: pr_log(pp, v, action, file, line)
150: struct pool *pp;
151: void *v;
152: int action;
153: const char *file;
154: long line;
155: {
156: int n = pp->pr_curlogentry;
157: struct pool_log *pl;
158:
1.20 thorpej 159: if ((pp->pr_roflags & PR_LOGGING) == 0)
1.3 pk 160: return;
161:
162: /*
163: * Fill in the current entry. Wrap around and overwrite
164: * the oldest entry if necessary.
165: */
166: pl = &pp->pr_log[n];
167: pl->pl_file = file;
168: pl->pl_line = line;
169: pl->pl_action = action;
170: pl->pl_addr = v;
171: if (++n >= pp->pr_logsize)
172: n = 0;
173: pp->pr_curlogentry = n;
174: }
175:
176: static void
1.25 thorpej 177: pr_printlog(pp, pi, pr)
1.3 pk 178: struct pool *pp;
1.25 thorpej 179: struct pool_item *pi;
180: void (*pr) __P((const char *, ...));
1.3 pk 181: {
182: int i = pp->pr_logsize;
183: int n = pp->pr_curlogentry;
184:
1.20 thorpej 185: if ((pp->pr_roflags & PR_LOGGING) == 0)
1.3 pk 186: return;
187:
188: /*
189: * Print all entries in this pool's log.
190: */
191: while (i-- > 0) {
192: struct pool_log *pl = &pp->pr_log[n];
193: if (pl->pl_action != 0) {
1.25 thorpej 194: if (pi == NULL || pi == pl->pl_addr) {
195: (*pr)("\tlog entry %d:\n", i);
196: (*pr)("\t\taction = %s, addr = %p\n",
197: pl->pl_action == PRLOG_GET ? "get" : "put",
198: pl->pl_addr);
199: (*pr)("\t\tfile: %s at line %lu\n",
200: pl->pl_file, pl->pl_line);
201: }
1.3 pk 202: }
203: if (++n >= pp->pr_logsize)
204: n = 0;
205: }
206: }
1.25 thorpej 207:
208: static __inline__ void
209: pr_enter(pp, file, line)
210: struct pool *pp;
211: const char *file;
212: long line;
213: {
214:
1.34 thorpej 215: if (__predict_false(pp->pr_entered_file != NULL)) {
1.25 thorpej 216: printf("pool %s: reentrancy at file %s line %ld\n",
217: pp->pr_wchan, file, line);
218: printf(" previous entry at file %s line %ld\n",
219: pp->pr_entered_file, pp->pr_entered_line);
220: panic("pr_enter");
221: }
222:
223: pp->pr_entered_file = file;
224: pp->pr_entered_line = line;
225: }
226:
227: static __inline__ void
228: pr_leave(pp)
229: struct pool *pp;
230: {
231:
1.34 thorpej 232: if (__predict_false(pp->pr_entered_file == NULL)) {
1.25 thorpej 233: printf("pool %s not entered?\n", pp->pr_wchan);
234: panic("pr_leave");
235: }
236:
237: pp->pr_entered_file = NULL;
238: pp->pr_entered_line = 0;
239: }
240:
241: static __inline__ void
242: pr_enter_check(pp, pr)
243: struct pool *pp;
244: void (*pr) __P((const char *, ...));
245: {
246:
247: if (pp->pr_entered_file != NULL)
248: (*pr)("\n\tcurrently entered from file %s line %ld\n",
249: pp->pr_entered_file, pp->pr_entered_line);
250: }
1.3 pk 251: #else
1.25 thorpej 252: #define pr_log(pp, v, action, file, line)
253: #define pr_printlog(pp, pi, pr)
254: #define pr_enter(pp, file, line)
255: #define pr_leave(pp)
256: #define pr_enter_check(pp, pr)
257: #endif /* DIAGNOSTIC */
1.3 pk 258:
259: /*
260: * Return the pool page header based on page address.
261: */
262: static __inline__ struct pool_item_header *
263: pr_find_pagehead(pp, page)
264: struct pool *pp;
265: caddr_t page;
266: {
267: struct pool_item_header *ph;
268:
1.20 thorpej 269: if ((pp->pr_roflags & PR_PHINPAGE) != 0)
1.3 pk 270: return ((struct pool_item_header *)(page + pp->pr_phoffset));
271:
272: for (ph = LIST_FIRST(&pp->pr_hashtab[PR_HASH_INDEX(pp, page)]);
273: ph != NULL;
274: ph = LIST_NEXT(ph, ph_hashlist)) {
275: if (ph->ph_page == page)
276: return (ph);
277: }
278: return (NULL);
279: }
280:
281: /*
282: * Remove a page from the pool.
283: */
284: static __inline__ void
285: pr_rmpage(pp, ph)
286: struct pool *pp;
287: struct pool_item_header *ph;
288: {
289:
290: /*
1.7 thorpej 291: * If the page was idle, decrement the idle page count.
1.3 pk 292: */
1.6 thorpej 293: if (ph->ph_nmissing == 0) {
294: #ifdef DIAGNOSTIC
295: if (pp->pr_nidle == 0)
296: panic("pr_rmpage: nidle inconsistent");
1.20 thorpej 297: if (pp->pr_nitems < pp->pr_itemsperpage)
298: panic("pr_rmpage: nitems inconsistent");
1.6 thorpej 299: #endif
300: pp->pr_nidle--;
301: }
1.7 thorpej 302:
1.20 thorpej 303: pp->pr_nitems -= pp->pr_itemsperpage;
304:
1.7 thorpej 305: /*
306: * Unlink a page from the pool and release it.
307: */
308: TAILQ_REMOVE(&pp->pr_pagelist, ph, ph_pagelist);
309: (*pp->pr_free)(ph->ph_page, pp->pr_pagesz, pp->pr_mtype);
310: pp->pr_npages--;
311: pp->pr_npagefree++;
1.6 thorpej 312:
1.22 chs 313: if ((pp->pr_roflags & PR_PHINPAGE) == 0) {
1.27 pk 314: int s;
1.22 chs 315: LIST_REMOVE(ph, ph_hashlist);
1.27 pk 316: s = splhigh();
1.22 chs 317: pool_put(&phpool, ph);
1.27 pk 318: splx(s);
1.22 chs 319: }
320:
1.3 pk 321: if (pp->pr_curpage == ph) {
322: /*
323: * Find a new non-empty page header, if any.
324: * Start search from the page head, to increase the
325: * chance for "high water" pages to be freed.
326: */
327: for (ph = TAILQ_FIRST(&pp->pr_pagelist); ph != NULL;
328: ph = TAILQ_NEXT(ph, ph_pagelist))
329: if (TAILQ_FIRST(&ph->ph_itemlist) != NULL)
330: break;
331:
332: pp->pr_curpage = ph;
1.21 thorpej 333: }
1.3 pk 334: }
335:
336: /*
337: * Allocate and initialize a pool.
338: */
1.1 pk 339: struct pool *
1.3 pk 340: pool_create(size, align, ioff, nitems, wchan, pagesz, alloc, release, mtype)
1.1 pk 341: size_t size;
1.3 pk 342: u_int align;
343: u_int ioff;
1.1 pk 344: int nitems;
1.21 thorpej 345: const char *wchan;
1.3 pk 346: size_t pagesz;
347: void *(*alloc) __P((unsigned long, int, int));
348: void (*release) __P((void *, unsigned long, int));
1.1 pk 349: int mtype;
350: {
351: struct pool *pp;
1.3 pk 352: int flags;
1.1 pk 353:
1.3 pk 354: pp = (struct pool *)malloc(sizeof(*pp), M_POOL, M_NOWAIT);
355: if (pp == NULL)
1.1 pk 356: return (NULL);
1.3 pk 357:
358: flags = PR_FREEHEADER;
359: pool_init(pp, size, align, ioff, flags, wchan, pagesz,
360: alloc, release, mtype);
361:
362: if (nitems != 0) {
363: if (pool_prime(pp, nitems, NULL) != 0) {
364: pool_destroy(pp);
365: return (NULL);
366: }
1.1 pk 367: }
368:
1.3 pk 369: return (pp);
370: }
371:
372: /*
373: * Initialize the given pool resource structure.
374: *
375: * We export this routine to allow other kernel parts to declare
376: * static pools that must be initialized before malloc() is available.
377: */
378: void
379: pool_init(pp, size, align, ioff, flags, wchan, pagesz, alloc, release, mtype)
380: struct pool *pp;
381: size_t size;
382: u_int align;
383: u_int ioff;
384: int flags;
1.21 thorpej 385: const char *wchan;
1.3 pk 386: size_t pagesz;
387: void *(*alloc) __P((unsigned long, int, int));
388: void (*release) __P((void *, unsigned long, int));
389: int mtype;
390: {
1.16 briggs 391: int off, slack, i;
1.3 pk 392:
1.25 thorpej 393: #ifdef POOL_DIAGNOSTIC
394: /*
395: * Always log if POOL_DIAGNOSTIC is defined.
396: */
397: if (pool_logsize != 0)
398: flags |= PR_LOGGING;
399: #endif
400:
1.3 pk 401: /*
402: * Check arguments and construct default values.
403: */
1.36 pk 404: if (!powerof2(pagesz))
1.3 pk 405: panic("pool_init: page size invalid (%lx)\n", (u_long)pagesz);
406:
1.4 thorpej 407: if (alloc == NULL && release == NULL) {
1.3 pk 408: alloc = pool_page_alloc;
409: release = pool_page_free;
1.4 thorpej 410: pagesz = PAGE_SIZE; /* Rounds to PAGE_SIZE anyhow. */
411: } else if ((alloc != NULL && release != NULL) == 0) {
412: /* If you specifiy one, must specify both. */
413: panic("pool_init: must specify alloc and release together");
414: }
415:
1.3 pk 416: if (pagesz == 0)
417: pagesz = PAGE_SIZE;
418:
419: if (align == 0)
420: align = ALIGN(1);
1.14 thorpej 421:
422: if (size < sizeof(struct pool_item))
423: size = sizeof(struct pool_item);
1.3 pk 424:
1.35 pk 425: size = ALIGN(size);
426: if (size >= pagesz)
427: panic("pool_init: pool item size (%lu) too large",
428: (u_long)size);
429:
1.3 pk 430: /*
431: * Initialize the pool structure.
432: */
433: TAILQ_INIT(&pp->pr_pagelist);
434: pp->pr_curpage = NULL;
435: pp->pr_npages = 0;
436: pp->pr_minitems = 0;
437: pp->pr_minpages = 0;
438: pp->pr_maxpages = UINT_MAX;
1.20 thorpej 439: pp->pr_roflags = flags;
440: pp->pr_flags = 0;
1.35 pk 441: pp->pr_size = size;
1.3 pk 442: pp->pr_align = align;
443: pp->pr_wchan = wchan;
444: pp->pr_mtype = mtype;
445: pp->pr_alloc = alloc;
446: pp->pr_free = release;
447: pp->pr_pagesz = pagesz;
448: pp->pr_pagemask = ~(pagesz - 1);
449: pp->pr_pageshift = ffs(pagesz) - 1;
1.20 thorpej 450: pp->pr_nitems = 0;
451: pp->pr_nout = 0;
452: pp->pr_hardlimit = UINT_MAX;
453: pp->pr_hardlimit_warning = NULL;
1.31 thorpej 454: pp->pr_hardlimit_ratecap.tv_sec = 0;
455: pp->pr_hardlimit_ratecap.tv_usec = 0;
456: pp->pr_hardlimit_warning_last.tv_sec = 0;
457: pp->pr_hardlimit_warning_last.tv_usec = 0;
1.3 pk 458:
459: /*
460: * Decide whether to put the page header off page to avoid
461: * wasting too large a part of the page. Off-page page headers
462: * go on a hash table, so we can match a returned item
463: * with its header based on the page address.
464: * We use 1/16 of the page size as the threshold (XXX: tune)
465: */
466: if (pp->pr_size < pagesz/16) {
467: /* Use the end of the page for the page header */
1.20 thorpej 468: pp->pr_roflags |= PR_PHINPAGE;
1.3 pk 469: pp->pr_phoffset = off =
470: pagesz - ALIGN(sizeof(struct pool_item_header));
1.2 pk 471: } else {
1.3 pk 472: /* The page header will be taken from our page header pool */
473: pp->pr_phoffset = 0;
474: off = pagesz;
1.16 briggs 475: for (i = 0; i < PR_HASHTABSIZE; i++) {
476: LIST_INIT(&pp->pr_hashtab[i]);
477: }
1.2 pk 478: }
1.1 pk 479:
1.3 pk 480: /*
481: * Alignment is to take place at `ioff' within the item. This means
482: * we must reserve up to `align - 1' bytes on the page to allow
483: * appropriate positioning of each item.
484: *
485: * Silently enforce `0 <= ioff < align'.
486: */
487: pp->pr_itemoffset = ioff = ioff % align;
488: pp->pr_itemsperpage = (off - ((align - ioff) % align)) / pp->pr_size;
489:
490: /*
491: * Use the slack between the chunks and the page header
492: * for "cache coloring".
493: */
494: slack = off - pp->pr_itemsperpage * pp->pr_size;
495: pp->pr_maxcolor = (slack / align) * align;
496: pp->pr_curcolor = 0;
497:
498: pp->pr_nget = 0;
499: pp->pr_nfail = 0;
500: pp->pr_nput = 0;
501: pp->pr_npagealloc = 0;
502: pp->pr_npagefree = 0;
1.1 pk 503: pp->pr_hiwat = 0;
1.8 thorpej 504: pp->pr_nidle = 0;
1.3 pk 505:
1.25 thorpej 506: if (flags & PR_LOGGING) {
507: if (kmem_map == NULL ||
508: (pp->pr_log = malloc(pool_logsize * sizeof(struct pool_log),
509: M_TEMP, M_NOWAIT)) == NULL)
1.20 thorpej 510: pp->pr_roflags &= ~PR_LOGGING;
1.3 pk 511: pp->pr_curlogentry = 0;
512: pp->pr_logsize = pool_logsize;
513: }
1.25 thorpej 514:
515: pp->pr_entered_file = NULL;
516: pp->pr_entered_line = 0;
1.3 pk 517:
1.21 thorpej 518: simple_lock_init(&pp->pr_slock);
1.1 pk 519:
1.3 pk 520: /*
521: * Initialize private page header pool if we haven't done so yet.
1.23 thorpej 522: * XXX LOCKING.
1.3 pk 523: */
524: if (phpool.pr_size == 0) {
525: pool_init(&phpool, sizeof(struct pool_item_header), 0, 0,
526: 0, "phpool", 0, 0, 0, 0);
1.1 pk 527: }
528:
1.23 thorpej 529: /* Insert into the list of all pools. */
530: simple_lock(&pool_head_slock);
531: TAILQ_INSERT_TAIL(&pool_head, pp, pr_poollist);
532: simple_unlock(&pool_head_slock);
1.1 pk 533: }
534:
535: /*
536: * De-commision a pool resource.
537: */
538: void
539: pool_destroy(pp)
540: struct pool *pp;
541: {
1.3 pk 542: struct pool_item_header *ph;
543:
544: #ifdef DIAGNOSTIC
1.20 thorpej 545: if (pp->pr_nout != 0) {
1.25 thorpej 546: pr_printlog(pp, NULL, printf);
1.20 thorpej 547: panic("pool_destroy: pool busy: still out: %u\n",
548: pp->pr_nout);
1.3 pk 549: }
550: #endif
1.1 pk 551:
1.3 pk 552: /* Remove all pages */
1.20 thorpej 553: if ((pp->pr_roflags & PR_STATIC) == 0)
1.3 pk 554: while ((ph = pp->pr_pagelist.tqh_first) != NULL)
555: pr_rmpage(pp, ph);
556:
557: /* Remove from global pool list */
1.23 thorpej 558: simple_lock(&pool_head_slock);
1.3 pk 559: TAILQ_REMOVE(&pool_head, pp, pr_poollist);
1.23 thorpej 560: /* XXX Only clear this if we were drainpp? */
1.3 pk 561: drainpp = NULL;
1.23 thorpej 562: simple_unlock(&pool_head_slock);
1.3 pk 563:
1.20 thorpej 564: if ((pp->pr_roflags & PR_LOGGING) != 0)
1.3 pk 565: free(pp->pr_log, M_TEMP);
1.2 pk 566:
1.20 thorpej 567: if (pp->pr_roflags & PR_FREEHEADER)
1.3 pk 568: free(pp, M_POOL);
1.1 pk 569: }
570:
571:
572: /*
1.3 pk 573: * Grab an item from the pool; must be called at appropriate spl level
1.1 pk 574: */
1.3 pk 575: void *
576: _pool_get(pp, flags, file, line)
577: struct pool *pp;
578: int flags;
579: const char *file;
580: long line;
1.1 pk 581: {
582: void *v;
583: struct pool_item *pi;
1.3 pk 584: struct pool_item_header *ph;
1.1 pk 585:
1.2 pk 586: #ifdef DIAGNOSTIC
1.34 thorpej 587: if (__predict_false((pp->pr_roflags & PR_STATIC) &&
588: (flags & PR_MALLOCOK))) {
1.25 thorpej 589: pr_printlog(pp, NULL, printf);
1.2 pk 590: panic("pool_get: static");
1.3 pk 591: }
1.2 pk 592: #endif
593:
1.37 sommerfe 594: if (__predict_false(curproc == NULL && doing_shutdown == 0 &&
595: (flags & PR_WAITOK) != 0))
1.3 pk 596: panic("pool_get: must have NOWAIT");
1.1 pk 597:
1.21 thorpej 598: simple_lock(&pp->pr_slock);
1.25 thorpej 599: pr_enter(pp, file, line);
1.20 thorpej 600:
601: startover:
602: /*
603: * Check to see if we've reached the hard limit. If we have,
604: * and we can wait, then wait until an item has been returned to
605: * the pool.
606: */
607: #ifdef DIAGNOSTIC
1.34 thorpej 608: if (__predict_false(pp->pr_nout > pp->pr_hardlimit)) {
1.25 thorpej 609: pr_leave(pp);
1.21 thorpej 610: simple_unlock(&pp->pr_slock);
1.20 thorpej 611: panic("pool_get: %s: crossed hard limit", pp->pr_wchan);
612: }
613: #endif
1.34 thorpej 614: if (__predict_false(pp->pr_nout == pp->pr_hardlimit)) {
1.29 sommerfe 615: if ((flags & PR_WAITOK) && !(flags & PR_LIMITFAIL)) {
1.20 thorpej 616: /*
617: * XXX: A warning isn't logged in this case. Should
618: * it be?
619: */
620: pp->pr_flags |= PR_WANTED;
1.25 thorpej 621: pr_leave(pp);
1.21 thorpej 622: simple_unlock(&pp->pr_slock);
1.20 thorpej 623: tsleep((caddr_t)pp, PSWP, pp->pr_wchan, 0);
1.21 thorpej 624: simple_lock(&pp->pr_slock);
1.25 thorpej 625: pr_enter(pp, file, line);
1.20 thorpej 626: goto startover;
627: }
1.31 thorpej 628:
629: /*
630: * Log a message that the hard limit has been hit.
631: */
632: if (pp->pr_hardlimit_warning != NULL &&
633: ratecheck(&pp->pr_hardlimit_warning_last,
634: &pp->pr_hardlimit_ratecap))
635: log(LOG_ERR, "%s\n", pp->pr_hardlimit_warning);
1.21 thorpej 636:
637: if (flags & PR_URGENT)
638: panic("pool_get: urgent");
639:
640: pp->pr_nfail++;
641:
1.25 thorpej 642: pr_leave(pp);
1.21 thorpej 643: simple_unlock(&pp->pr_slock);
1.20 thorpej 644: return (NULL);
645: }
646:
1.3 pk 647: /*
648: * The convention we use is that if `curpage' is not NULL, then
649: * it points at a non-empty bucket. In particular, `curpage'
650: * never points at a page header which has PR_PHINPAGE set and
651: * has no items in its bucket.
652: */
1.20 thorpej 653: if ((ph = pp->pr_curpage) == NULL) {
1.15 pk 654: void *v;
655:
1.20 thorpej 656: #ifdef DIAGNOSTIC
657: if (pp->pr_nitems != 0) {
1.21 thorpej 658: simple_unlock(&pp->pr_slock);
1.20 thorpej 659: printf("pool_get: %s: curpage NULL, nitems %u\n",
660: pp->pr_wchan, pp->pr_nitems);
661: panic("pool_get: nitems inconsistent\n");
662: }
663: #endif
664:
1.21 thorpej 665: /*
666: * Call the back-end page allocator for more memory.
667: * Release the pool lock, as the back-end page allocator
668: * may block.
669: */
1.25 thorpej 670: pr_leave(pp);
1.21 thorpej 671: simple_unlock(&pp->pr_slock);
672: v = (*pp->pr_alloc)(pp->pr_pagesz, flags, pp->pr_mtype);
673: simple_lock(&pp->pr_slock);
1.25 thorpej 674: pr_enter(pp, file, line);
1.15 pk 675:
1.21 thorpej 676: if (v == NULL) {
677: /*
678: * We were unable to allocate a page, but
679: * we released the lock during allocation,
680: * so perhaps items were freed back to the
681: * pool. Check for this case.
682: */
683: if (pp->pr_curpage != NULL)
684: goto startover;
1.15 pk 685:
1.3 pk 686: if (flags & PR_URGENT)
687: panic("pool_get: urgent");
1.21 thorpej 688:
1.3 pk 689: if ((flags & PR_WAITOK) == 0) {
690: pp->pr_nfail++;
1.25 thorpej 691: pr_leave(pp);
1.21 thorpej 692: simple_unlock(&pp->pr_slock);
1.1 pk 693: return (NULL);
1.3 pk 694: }
695:
1.15 pk 696: /*
697: * Wait for items to be returned to this pool.
1.21 thorpej 698: *
1.15 pk 699: * XXX: we actually want to wait just until
700: * the page allocator has memory again. Depending
701: * on this pool's usage, we might get stuck here
702: * for a long time.
1.20 thorpej 703: *
704: * XXX: maybe we should wake up once a second and
705: * try again?
1.15 pk 706: */
1.1 pk 707: pp->pr_flags |= PR_WANTED;
1.25 thorpej 708: pr_leave(pp);
1.21 thorpej 709: simple_unlock(&pp->pr_slock);
1.1 pk 710: tsleep((caddr_t)pp, PSWP, pp->pr_wchan, 0);
1.21 thorpej 711: simple_lock(&pp->pr_slock);
1.25 thorpej 712: pr_enter(pp, file, line);
1.20 thorpej 713: goto startover;
1.1 pk 714: }
1.3 pk 715:
1.15 pk 716: /* We have more memory; add it to the pool */
717: pp->pr_npagealloc++;
718: pool_prime_page(pp, v);
719:
1.20 thorpej 720: /* Start the allocation process over. */
721: goto startover;
1.3 pk 722: }
723:
1.34 thorpej 724: if (__predict_false((v = pi = TAILQ_FIRST(&ph->ph_itemlist)) == NULL)) {
1.25 thorpej 725: pr_leave(pp);
1.21 thorpej 726: simple_unlock(&pp->pr_slock);
1.3 pk 727: panic("pool_get: %s: page empty", pp->pr_wchan);
1.21 thorpej 728: }
1.20 thorpej 729: #ifdef DIAGNOSTIC
1.34 thorpej 730: if (__predict_false(pp->pr_nitems == 0)) {
1.25 thorpej 731: pr_leave(pp);
1.21 thorpej 732: simple_unlock(&pp->pr_slock);
1.20 thorpej 733: printf("pool_get: %s: items on itemlist, nitems %u\n",
734: pp->pr_wchan, pp->pr_nitems);
735: panic("pool_get: nitems inconsistent\n");
736: }
737: #endif
1.3 pk 738: pr_log(pp, v, PRLOG_GET, file, line);
739:
740: #ifdef DIAGNOSTIC
1.34 thorpej 741: if (__predict_false(pi->pi_magic != PI_MAGIC)) {
1.25 thorpej 742: pr_printlog(pp, pi, printf);
1.3 pk 743: panic("pool_get(%s): free list modified: magic=%x; page %p;"
744: " item addr %p\n",
745: pp->pr_wchan, pi->pi_magic, ph->ph_page, pi);
746: }
747: #endif
748:
749: /*
750: * Remove from item list.
751: */
752: TAILQ_REMOVE(&ph->ph_itemlist, pi, pi_list);
1.20 thorpej 753: pp->pr_nitems--;
754: pp->pr_nout++;
1.6 thorpej 755: if (ph->ph_nmissing == 0) {
756: #ifdef DIAGNOSTIC
1.34 thorpej 757: if (__predict_false(pp->pr_nidle == 0))
1.6 thorpej 758: panic("pool_get: nidle inconsistent");
759: #endif
760: pp->pr_nidle--;
761: }
1.3 pk 762: ph->ph_nmissing++;
763: if (TAILQ_FIRST(&ph->ph_itemlist) == NULL) {
1.21 thorpej 764: #ifdef DIAGNOSTIC
1.34 thorpej 765: if (__predict_false(ph->ph_nmissing != pp->pr_itemsperpage)) {
1.25 thorpej 766: pr_leave(pp);
1.21 thorpej 767: simple_unlock(&pp->pr_slock);
768: panic("pool_get: %s: nmissing inconsistent",
769: pp->pr_wchan);
770: }
771: #endif
1.3 pk 772: /*
773: * Find a new non-empty page header, if any.
774: * Start search from the page head, to increase
775: * the chance for "high water" pages to be freed.
776: *
1.21 thorpej 777: * Migrate empty pages to the end of the list. This
778: * will speed the update of curpage as pages become
779: * idle. Empty pages intermingled with idle pages
780: * is no big deal. As soon as a page becomes un-empty,
781: * it will move back to the head of the list.
1.3 pk 782: */
783: TAILQ_REMOVE(&pp->pr_pagelist, ph, ph_pagelist);
1.21 thorpej 784: TAILQ_INSERT_TAIL(&pp->pr_pagelist, ph, ph_pagelist);
785: for (ph = TAILQ_FIRST(&pp->pr_pagelist); ph != NULL;
786: ph = TAILQ_NEXT(ph, ph_pagelist))
1.3 pk 787: if (TAILQ_FIRST(&ph->ph_itemlist) != NULL)
788: break;
789:
790: pp->pr_curpage = ph;
1.1 pk 791: }
1.3 pk 792:
793: pp->pr_nget++;
1.20 thorpej 794:
795: /*
796: * If we have a low water mark and we are now below that low
797: * water mark, add more items to the pool.
798: */
799: if (pp->pr_nitems < pp->pr_minitems && pool_catchup(pp) != 0) {
800: /*
801: * XXX: Should we log a warning? Should we set up a timeout
802: * to try again in a second or so? The latter could break
803: * a caller's assumptions about interrupt protection, etc.
804: */
805: }
806:
1.25 thorpej 807: pr_leave(pp);
1.21 thorpej 808: simple_unlock(&pp->pr_slock);
1.1 pk 809: return (v);
810: }
811:
812: /*
1.3 pk 813: * Return resource to the pool; must be called at appropriate spl level
1.1 pk 814: */
1.3 pk 815: void
816: _pool_put(pp, v, file, line)
817: struct pool *pp;
818: void *v;
819: const char *file;
820: long line;
1.1 pk 821: {
822: struct pool_item *pi = v;
1.3 pk 823: struct pool_item_header *ph;
824: caddr_t page;
1.21 thorpej 825: int s;
1.3 pk 826:
827: page = (caddr_t)((u_long)v & pp->pr_pagemask);
1.1 pk 828:
1.21 thorpej 829: simple_lock(&pp->pr_slock);
1.25 thorpej 830: pr_enter(pp, file, line);
1.30 thorpej 831:
832: #ifdef DIAGNOSTIC
1.34 thorpej 833: if (__predict_false(pp->pr_nout == 0)) {
1.30 thorpej 834: printf("pool %s: putting with none out\n",
835: pp->pr_wchan);
836: panic("pool_put");
837: }
838: #endif
1.3 pk 839:
840: pr_log(pp, v, PRLOG_PUT, file, line);
841:
1.34 thorpej 842: if (__predict_false((ph = pr_find_pagehead(pp, page)) == NULL)) {
1.25 thorpej 843: pr_printlog(pp, NULL, printf);
1.3 pk 844: panic("pool_put: %s: page header missing", pp->pr_wchan);
845: }
1.28 thorpej 846:
847: #ifdef LOCKDEBUG
848: /*
849: * Check if we're freeing a locked simple lock.
850: */
851: simple_lock_freecheck((caddr_t)pi, ((caddr_t)pi) + pp->pr_size);
852: #endif
1.3 pk 853:
854: /*
855: * Return to item list.
856: */
1.2 pk 857: #ifdef DIAGNOSTIC
1.3 pk 858: pi->pi_magic = PI_MAGIC;
859: #endif
1.32 chs 860: #ifdef DEBUG
861: {
862: int i, *ip = v;
863:
864: for (i = 0; i < pp->pr_size / sizeof(int); i++) {
865: *ip++ = PI_MAGIC;
866: }
867: }
868: #endif
869:
1.3 pk 870: TAILQ_INSERT_HEAD(&ph->ph_itemlist, pi, pi_list);
871: ph->ph_nmissing--;
872: pp->pr_nput++;
1.20 thorpej 873: pp->pr_nitems++;
874: pp->pr_nout--;
1.3 pk 875:
876: /* Cancel "pool empty" condition if it exists */
877: if (pp->pr_curpage == NULL)
878: pp->pr_curpage = ph;
879:
880: if (pp->pr_flags & PR_WANTED) {
881: pp->pr_flags &= ~PR_WANTED;
1.15 pk 882: if (ph->ph_nmissing == 0)
883: pp->pr_nidle++;
1.25 thorpej 884: pr_leave(pp);
1.21 thorpej 885: simple_unlock(&pp->pr_slock);
1.3 pk 886: wakeup((caddr_t)pp);
887: return;
888: }
889:
890: /*
1.21 thorpej 891: * If this page is now complete, do one of two things:
892: *
893: * (1) If we have more pages than the page high water
894: * mark, free the page back to the system.
895: *
896: * (2) Move it to the end of the page list, so that
897: * we minimize our chances of fragmenting the
898: * pool. Idle pages migrate to the end (along with
899: * completely empty pages, so that we find un-empty
900: * pages more quickly when we update curpage) of the
901: * list so they can be more easily swept up by
902: * the pagedaemon when pages are scarce.
1.3 pk 903: */
904: if (ph->ph_nmissing == 0) {
1.6 thorpej 905: pp->pr_nidle++;
1.3 pk 906: if (pp->pr_npages > pp->pr_maxpages) {
907: pr_rmpage(pp, ph);
908: } else {
909: TAILQ_REMOVE(&pp->pr_pagelist, ph, ph_pagelist);
910: TAILQ_INSERT_TAIL(&pp->pr_pagelist, ph, ph_pagelist);
911:
1.21 thorpej 912: /*
913: * Update the timestamp on the page. A page must
914: * be idle for some period of time before it can
915: * be reclaimed by the pagedaemon. This minimizes
916: * ping-pong'ing for memory.
917: */
918: s = splclock();
919: ph->ph_time = mono_time;
920: splx(s);
921:
922: /*
923: * Update the current page pointer. Just look for
924: * the first page with any free items.
925: *
926: * XXX: Maybe we want an option to look for the
927: * page with the fewest available items, to minimize
928: * fragmentation?
929: */
1.3 pk 930: for (ph = TAILQ_FIRST(&pp->pr_pagelist); ph != NULL;
931: ph = TAILQ_NEXT(ph, ph_pagelist))
932: if (TAILQ_FIRST(&ph->ph_itemlist) != NULL)
933: break;
1.1 pk 934:
1.3 pk 935: pp->pr_curpage = ph;
1.1 pk 936: }
937: }
1.21 thorpej 938: /*
939: * If the page has just become un-empty, move it to the head of
940: * the list, and make it the current page. The next allocation
941: * will get the item from this page, instead of further fragmenting
942: * the pool.
943: */
944: else if (ph->ph_nmissing == (pp->pr_itemsperpage - 1)) {
945: TAILQ_REMOVE(&pp->pr_pagelist, ph, ph_pagelist);
946: TAILQ_INSERT_HEAD(&pp->pr_pagelist, ph, ph_pagelist);
947: pp->pr_curpage = ph;
948: }
949:
1.25 thorpej 950: pr_leave(pp);
1.21 thorpej 951: simple_unlock(&pp->pr_slock);
1.3 pk 952:
1.1 pk 953: }
954:
955: /*
1.3 pk 956: * Add N items to the pool.
1.1 pk 957: */
958: int
1.2 pk 959: pool_prime(pp, n, storage)
1.1 pk 960: struct pool *pp;
961: int n;
1.2 pk 962: caddr_t storage;
1.1 pk 963: {
1.3 pk 964: caddr_t cp;
965: int newnitems, newpages;
1.2 pk 966:
967: #ifdef DIAGNOSTIC
1.34 thorpej 968: if (__predict_false(storage && !(pp->pr_roflags & PR_STATIC)))
1.2 pk 969: panic("pool_prime: static");
970: /* !storage && static caught below */
971: #endif
1.1 pk 972:
1.21 thorpej 973: simple_lock(&pp->pr_slock);
974:
1.3 pk 975: newnitems = pp->pr_minitems + n;
976: newpages =
1.18 thorpej 977: roundup(newnitems, pp->pr_itemsperpage) / pp->pr_itemsperpage
1.3 pk 978: - pp->pr_minpages;
979:
980: while (newpages-- > 0) {
1.20 thorpej 981: if (pp->pr_roflags & PR_STATIC) {
1.3 pk 982: cp = storage;
983: storage += pp->pr_pagesz;
984: } else {
1.21 thorpej 985: simple_unlock(&pp->pr_slock);
1.3 pk 986: cp = (*pp->pr_alloc)(pp->pr_pagesz, 0, pp->pr_mtype);
1.21 thorpej 987: simple_lock(&pp->pr_slock);
1.3 pk 988: }
1.2 pk 989:
1.3 pk 990: if (cp == NULL) {
1.21 thorpej 991: simple_unlock(&pp->pr_slock);
1.1 pk 992: return (ENOMEM);
993: }
994:
1.26 thorpej 995: pp->pr_npagealloc++;
1.3 pk 996: pool_prime_page(pp, cp);
997: pp->pr_minpages++;
1.1 pk 998: }
1.3 pk 999:
1000: pp->pr_minitems = newnitems;
1001:
1002: if (pp->pr_minpages >= pp->pr_maxpages)
1003: pp->pr_maxpages = pp->pr_minpages + 1; /* XXX */
1004:
1.21 thorpej 1005: simple_unlock(&pp->pr_slock);
1.1 pk 1006: return (0);
1007: }
1.3 pk 1008:
1009: /*
1010: * Add a page worth of items to the pool.
1.21 thorpej 1011: *
1012: * Note, we must be called with the pool descriptor LOCKED.
1.3 pk 1013: */
1.21 thorpej 1014: static void
1.3 pk 1015: pool_prime_page(pp, storage)
1016: struct pool *pp;
1017: caddr_t storage;
1018: {
1019: struct pool_item *pi;
1020: struct pool_item_header *ph;
1021: caddr_t cp = storage;
1022: unsigned int align = pp->pr_align;
1023: unsigned int ioff = pp->pr_itemoffset;
1.27 pk 1024: int s, n;
1.36 pk 1025:
1026: if (((u_long)cp & (pp->pr_pagesz - 1)) != 0)
1027: panic("pool_prime_page: %s: unaligned page", pp->pr_wchan);
1.3 pk 1028:
1.20 thorpej 1029: if ((pp->pr_roflags & PR_PHINPAGE) != 0) {
1.3 pk 1030: ph = (struct pool_item_header *)(cp + pp->pr_phoffset);
1031: } else {
1.27 pk 1032: s = splhigh();
1.3 pk 1033: ph = pool_get(&phpool, PR_URGENT);
1.27 pk 1034: splx(s);
1.3 pk 1035: LIST_INSERT_HEAD(&pp->pr_hashtab[PR_HASH_INDEX(pp, cp)],
1036: ph, ph_hashlist);
1037: }
1038:
1039: /*
1040: * Insert page header.
1041: */
1042: TAILQ_INSERT_HEAD(&pp->pr_pagelist, ph, ph_pagelist);
1043: TAILQ_INIT(&ph->ph_itemlist);
1044: ph->ph_page = storage;
1045: ph->ph_nmissing = 0;
1.21 thorpej 1046: memset(&ph->ph_time, 0, sizeof(ph->ph_time));
1.3 pk 1047:
1.6 thorpej 1048: pp->pr_nidle++;
1049:
1.3 pk 1050: /*
1051: * Color this page.
1052: */
1053: cp = (caddr_t)(cp + pp->pr_curcolor);
1054: if ((pp->pr_curcolor += align) > pp->pr_maxcolor)
1055: pp->pr_curcolor = 0;
1056:
1057: /*
1058: * Adjust storage to apply aligment to `pr_itemoffset' in each item.
1059: */
1060: if (ioff != 0)
1061: cp = (caddr_t)(cp + (align - ioff));
1062:
1063: /*
1064: * Insert remaining chunks on the bucket list.
1065: */
1066: n = pp->pr_itemsperpage;
1.20 thorpej 1067: pp->pr_nitems += n;
1.3 pk 1068:
1069: while (n--) {
1070: pi = (struct pool_item *)cp;
1071:
1072: /* Insert on page list */
1073: TAILQ_INSERT_TAIL(&ph->ph_itemlist, pi, pi_list);
1074: #ifdef DIAGNOSTIC
1075: pi->pi_magic = PI_MAGIC;
1076: #endif
1077: cp = (caddr_t)(cp + pp->pr_size);
1078: }
1079:
1080: /*
1081: * If the pool was depleted, point at the new page.
1082: */
1083: if (pp->pr_curpage == NULL)
1084: pp->pr_curpage = ph;
1085:
1086: if (++pp->pr_npages > pp->pr_hiwat)
1087: pp->pr_hiwat = pp->pr_npages;
1088: }
1089:
1.20 thorpej 1090: /*
1091: * Like pool_prime(), except this is used by pool_get() when nitems
1092: * drops below the low water mark. This is used to catch up nitmes
1093: * with the low water mark.
1094: *
1.21 thorpej 1095: * Note 1, we never wait for memory here, we let the caller decide what to do.
1.20 thorpej 1096: *
1097: * Note 2, this doesn't work with static pools.
1098: *
1099: * Note 3, we must be called with the pool already locked, and we return
1100: * with it locked.
1101: */
1102: static int
1103: pool_catchup(pp)
1104: struct pool *pp;
1105: {
1106: caddr_t cp;
1107: int error = 0;
1108:
1109: if (pp->pr_roflags & PR_STATIC) {
1110: /*
1111: * We dropped below the low water mark, and this is not a
1112: * good thing. Log a warning.
1.21 thorpej 1113: *
1114: * XXX: rate-limit this?
1.20 thorpej 1115: */
1116: printf("WARNING: static pool `%s' dropped below low water "
1117: "mark\n", pp->pr_wchan);
1118: return (0);
1119: }
1120:
1.21 thorpej 1121: while (pp->pr_nitems < pp->pr_minitems) {
1.20 thorpej 1122: /*
1.21 thorpej 1123: * Call the page back-end allocator for more memory.
1124: *
1125: * XXX: We never wait, so should we bother unlocking
1126: * the pool descriptor?
1.20 thorpej 1127: */
1.21 thorpej 1128: simple_unlock(&pp->pr_slock);
1.20 thorpej 1129: cp = (*pp->pr_alloc)(pp->pr_pagesz, 0, pp->pr_mtype);
1.21 thorpej 1130: simple_lock(&pp->pr_slock);
1.34 thorpej 1131: if (__predict_false(cp == NULL)) {
1.20 thorpej 1132: error = ENOMEM;
1133: break;
1134: }
1.26 thorpej 1135: pp->pr_npagealloc++;
1.20 thorpej 1136: pool_prime_page(pp, cp);
1137: }
1138:
1139: return (error);
1140: }
1141:
1.3 pk 1142: void
1143: pool_setlowat(pp, n)
1144: pool_handle_t pp;
1145: int n;
1146: {
1.20 thorpej 1147: int error;
1.15 pk 1148:
1.21 thorpej 1149: simple_lock(&pp->pr_slock);
1150:
1.3 pk 1151: pp->pr_minitems = n;
1.15 pk 1152: pp->pr_minpages = (n == 0)
1153: ? 0
1.18 thorpej 1154: : roundup(n, pp->pr_itemsperpage) / pp->pr_itemsperpage;
1.20 thorpej 1155:
1156: /* Make sure we're caught up with the newly-set low water mark. */
1.21 thorpej 1157: if ((error = pool_catchup(pp)) != 0) {
1.20 thorpej 1158: /*
1159: * XXX: Should we log a warning? Should we set up a timeout
1160: * to try again in a second or so? The latter could break
1161: * a caller's assumptions about interrupt protection, etc.
1162: */
1163: }
1.21 thorpej 1164:
1165: simple_unlock(&pp->pr_slock);
1.3 pk 1166: }
1167:
1168: void
1169: pool_sethiwat(pp, n)
1170: pool_handle_t pp;
1171: int n;
1172: {
1.15 pk 1173:
1.21 thorpej 1174: simple_lock(&pp->pr_slock);
1175:
1.15 pk 1176: pp->pr_maxpages = (n == 0)
1177: ? 0
1.18 thorpej 1178: : roundup(n, pp->pr_itemsperpage) / pp->pr_itemsperpage;
1.21 thorpej 1179:
1180: simple_unlock(&pp->pr_slock);
1.3 pk 1181: }
1182:
1.20 thorpej 1183: void
1184: pool_sethardlimit(pp, n, warnmess, ratecap)
1185: pool_handle_t pp;
1186: int n;
1187: const char *warnmess;
1188: int ratecap;
1189: {
1190:
1.21 thorpej 1191: simple_lock(&pp->pr_slock);
1.20 thorpej 1192:
1193: pp->pr_hardlimit = n;
1194: pp->pr_hardlimit_warning = warnmess;
1.31 thorpej 1195: pp->pr_hardlimit_ratecap.tv_sec = ratecap;
1196: pp->pr_hardlimit_warning_last.tv_sec = 0;
1197: pp->pr_hardlimit_warning_last.tv_usec = 0;
1.20 thorpej 1198:
1199: /*
1.21 thorpej 1200: * In-line version of pool_sethiwat(), because we don't want to
1201: * release the lock.
1.20 thorpej 1202: */
1203: pp->pr_maxpages = (n == 0)
1204: ? 0
1205: : roundup(n, pp->pr_itemsperpage) / pp->pr_itemsperpage;
1.21 thorpej 1206:
1207: simple_unlock(&pp->pr_slock);
1.20 thorpej 1208: }
1.3 pk 1209:
1210: /*
1211: * Default page allocator.
1212: */
1213: static void *
1214: pool_page_alloc(sz, flags, mtype)
1215: unsigned long sz;
1216: int flags;
1217: int mtype;
1218: {
1.11 thorpej 1219: boolean_t waitok = (flags & PR_WAITOK) ? TRUE : FALSE;
1.3 pk 1220:
1.11 thorpej 1221: return ((void *)uvm_km_alloc_poolpage(waitok));
1.3 pk 1222: }
1223:
1224: static void
1225: pool_page_free(v, sz, mtype)
1226: void *v;
1227: unsigned long sz;
1228: int mtype;
1229: {
1230:
1.10 eeh 1231: uvm_km_free_poolpage((vaddr_t)v);
1.3 pk 1232: }
1.12 thorpej 1233:
1234: /*
1235: * Alternate pool page allocator for pools that know they will
1236: * never be accessed in interrupt context.
1237: */
1238: void *
1239: pool_page_alloc_nointr(sz, flags, mtype)
1240: unsigned long sz;
1241: int flags;
1242: int mtype;
1243: {
1244: boolean_t waitok = (flags & PR_WAITOK) ? TRUE : FALSE;
1245:
1246: return ((void *)uvm_km_alloc_poolpage1(kernel_map, uvm.kernel_object,
1247: waitok));
1248: }
1249:
1250: void
1251: pool_page_free_nointr(v, sz, mtype)
1252: void *v;
1253: unsigned long sz;
1254: int mtype;
1255: {
1256:
1257: uvm_km_free_poolpage1(kernel_map, (vaddr_t)v);
1258: }
1259:
1.3 pk 1260:
1261: /*
1262: * Release all complete pages that have not been used recently.
1263: */
1264: void
1.25 thorpej 1265: _pool_reclaim(pp, file, line)
1.3 pk 1266: pool_handle_t pp;
1.25 thorpej 1267: const char *file;
1268: long line;
1.3 pk 1269: {
1270: struct pool_item_header *ph, *phnext;
1.21 thorpej 1271: struct timeval curtime;
1272: int s;
1.3 pk 1273:
1.20 thorpej 1274: if (pp->pr_roflags & PR_STATIC)
1.3 pk 1275: return;
1276:
1.21 thorpej 1277: if (simple_lock_try(&pp->pr_slock) == 0)
1.3 pk 1278: return;
1.25 thorpej 1279: pr_enter(pp, file, line);
1.3 pk 1280:
1.21 thorpej 1281: s = splclock();
1282: curtime = mono_time;
1283: splx(s);
1284:
1.3 pk 1285: for (ph = TAILQ_FIRST(&pp->pr_pagelist); ph != NULL; ph = phnext) {
1286: phnext = TAILQ_NEXT(ph, ph_pagelist);
1287:
1288: /* Check our minimum page claim */
1289: if (pp->pr_npages <= pp->pr_minpages)
1290: break;
1291:
1292: if (ph->ph_nmissing == 0) {
1293: struct timeval diff;
1294: timersub(&curtime, &ph->ph_time, &diff);
1295: if (diff.tv_sec < pool_inactive_time)
1296: continue;
1.21 thorpej 1297:
1298: /*
1299: * If freeing this page would put us below
1300: * the low water mark, stop now.
1301: */
1302: if ((pp->pr_nitems - pp->pr_itemsperpage) <
1303: pp->pr_minitems)
1304: break;
1305:
1.3 pk 1306: pr_rmpage(pp, ph);
1307: }
1308: }
1309:
1.25 thorpej 1310: pr_leave(pp);
1.21 thorpej 1311: simple_unlock(&pp->pr_slock);
1.3 pk 1312: }
1313:
1314:
1315: /*
1316: * Drain pools, one at a time.
1.21 thorpej 1317: *
1318: * Note, we must never be called from an interrupt context.
1.3 pk 1319: */
1320: void
1321: pool_drain(arg)
1322: void *arg;
1323: {
1324: struct pool *pp;
1.23 thorpej 1325: int s;
1.3 pk 1326:
1.23 thorpej 1327: s = splimp();
1328: simple_lock(&pool_head_slock);
1329:
1330: if (drainpp == NULL && (drainpp = TAILQ_FIRST(&pool_head)) == NULL)
1331: goto out;
1.3 pk 1332:
1333: pp = drainpp;
1334: drainpp = TAILQ_NEXT(pp, pr_poollist);
1335:
1336: pool_reclaim(pp);
1.23 thorpej 1337:
1338: out:
1339: simple_unlock(&pool_head_slock);
1.3 pk 1340: splx(s);
1341: }
1342:
1343:
1344: /*
1345: * Diagnostic helpers.
1346: */
1347: void
1.25 thorpej 1348: pool_print(pp, modif)
1.3 pk 1349: struct pool *pp;
1.25 thorpej 1350: const char *modif;
1.21 thorpej 1351: {
1352: int s;
1353:
1354: s = splimp();
1.25 thorpej 1355: if (simple_lock_try(&pp->pr_slock) == 0) {
1356: printf("pool %s is locked; try again later\n",
1357: pp->pr_wchan);
1358: splx(s);
1359: return;
1360: }
1361: pool_print1(pp, modif, printf);
1.21 thorpej 1362: simple_unlock(&pp->pr_slock);
1363: splx(s);
1364: }
1365:
1.25 thorpej 1366: void
1367: pool_printit(pp, modif, pr)
1368: struct pool *pp;
1369: const char *modif;
1370: void (*pr) __P((const char *, ...));
1371: {
1372: int didlock = 0;
1373:
1374: if (pp == NULL) {
1375: (*pr)("Must specify a pool to print.\n");
1376: return;
1377: }
1378:
1379: /*
1380: * Called from DDB; interrupts should be blocked, and all
1381: * other processors should be paused. We can skip locking
1382: * the pool in this case.
1383: *
1384: * We do a simple_lock_try() just to print the lock
1385: * status, however.
1386: */
1387:
1388: if (simple_lock_try(&pp->pr_slock) == 0)
1389: (*pr)("WARNING: pool %s is locked\n", pp->pr_wchan);
1390: else
1391: didlock = 1;
1392:
1393: pool_print1(pp, modif, pr);
1394:
1395: if (didlock)
1396: simple_unlock(&pp->pr_slock);
1397: }
1398:
1.21 thorpej 1399: static void
1.25 thorpej 1400: pool_print1(pp, modif, pr)
1.21 thorpej 1401: struct pool *pp;
1.25 thorpej 1402: const char *modif;
1403: void (*pr) __P((const char *, ...));
1.3 pk 1404: {
1.25 thorpej 1405: struct pool_item_header *ph;
1406: #ifdef DIAGNOSTIC
1407: struct pool_item *pi;
1408: #endif
1409: int print_log = 0, print_pagelist = 0;
1410: char c;
1411:
1412: while ((c = *modif++) != '\0') {
1413: if (c == 'l')
1414: print_log = 1;
1415: if (c == 'p')
1416: print_pagelist = 1;
1417: modif++;
1418: }
1419:
1420: (*pr)("POOL %s: size %u, align %u, ioff %u, roflags 0x%08x\n",
1421: pp->pr_wchan, pp->pr_size, pp->pr_align, pp->pr_itemoffset,
1422: pp->pr_roflags);
1423: (*pr)("\tpagesz %u, mtype %d\n", pp->pr_pagesz, pp->pr_mtype);
1424: (*pr)("\talloc %p, release %p\n", pp->pr_alloc, pp->pr_free);
1425: (*pr)("\tminitems %u, minpages %u, maxpages %u, npages %u\n",
1426: pp->pr_minitems, pp->pr_minpages, pp->pr_maxpages, pp->pr_npages);
1427: (*pr)("\titemsperpage %u, nitems %u, nout %u, hardlimit %u\n",
1428: pp->pr_itemsperpage, pp->pr_nitems, pp->pr_nout, pp->pr_hardlimit);
1429:
1430: (*pr)("\n\tnget %lu, nfail %lu, nput %lu\n",
1431: pp->pr_nget, pp->pr_nfail, pp->pr_nput);
1432: (*pr)("\tnpagealloc %lu, npagefree %lu, hiwat %u, nidle %lu\n",
1433: pp->pr_npagealloc, pp->pr_npagefree, pp->pr_hiwat, pp->pr_nidle);
1434:
1435: if (print_pagelist == 0)
1436: goto skip_pagelist;
1437:
1438: if ((ph = TAILQ_FIRST(&pp->pr_pagelist)) != NULL)
1439: (*pr)("\n\tpage list:\n");
1440: for (; ph != NULL; ph = TAILQ_NEXT(ph, ph_pagelist)) {
1441: (*pr)("\t\tpage %p, nmissing %d, time %lu,%lu\n",
1442: ph->ph_page, ph->ph_nmissing,
1443: (u_long)ph->ph_time.tv_sec,
1444: (u_long)ph->ph_time.tv_usec);
1445: #ifdef DIAGNOSTIC
1446: for (pi = TAILQ_FIRST(&ph->ph_itemlist); pi != NULL;
1447: pi = TAILQ_NEXT(pi, pi_list)) {
1448: if (pi->pi_magic != PI_MAGIC) {
1449: (*pr)("\t\t\titem %p, magic 0x%x\n",
1450: pi, pi->pi_magic);
1451: }
1452: }
1453: #endif
1454: }
1455: if (pp->pr_curpage == NULL)
1456: (*pr)("\tno current page\n");
1457: else
1458: (*pr)("\tcurpage %p\n", pp->pr_curpage->ph_page);
1459:
1460: skip_pagelist:
1461:
1462: if (print_log == 0)
1463: goto skip_log;
1464:
1465: (*pr)("\n");
1466: if ((pp->pr_roflags & PR_LOGGING) == 0)
1467: (*pr)("\tno log\n");
1468: else
1469: pr_printlog(pp, NULL, pr);
1.3 pk 1470:
1.25 thorpej 1471: skip_log:
1.3 pk 1472:
1.25 thorpej 1473: pr_enter_check(pp, pr);
1.3 pk 1474: }
1475:
1476: int
1477: pool_chk(pp, label)
1478: struct pool *pp;
1479: char *label;
1480: {
1481: struct pool_item_header *ph;
1482: int r = 0;
1483:
1.21 thorpej 1484: simple_lock(&pp->pr_slock);
1.3 pk 1485:
1486: for (ph = TAILQ_FIRST(&pp->pr_pagelist); ph != NULL;
1487: ph = TAILQ_NEXT(ph, ph_pagelist)) {
1488:
1489: struct pool_item *pi;
1490: int n;
1491: caddr_t page;
1492:
1493: page = (caddr_t)((u_long)ph & pp->pr_pagemask);
1.20 thorpej 1494: if (page != ph->ph_page &&
1495: (pp->pr_roflags & PR_PHINPAGE) != 0) {
1.3 pk 1496: if (label != NULL)
1497: printf("%s: ", label);
1.16 briggs 1498: printf("pool(%p:%s): page inconsistency: page %p;"
1499: " at page head addr %p (p %p)\n", pp,
1.3 pk 1500: pp->pr_wchan, ph->ph_page,
1501: ph, page);
1502: r++;
1503: goto out;
1504: }
1505:
1506: for (pi = TAILQ_FIRST(&ph->ph_itemlist), n = 0;
1507: pi != NULL;
1508: pi = TAILQ_NEXT(pi,pi_list), n++) {
1509:
1510: #ifdef DIAGNOSTIC
1511: if (pi->pi_magic != PI_MAGIC) {
1512: if (label != NULL)
1513: printf("%s: ", label);
1514: printf("pool(%s): free list modified: magic=%x;"
1515: " page %p; item ordinal %d;"
1516: " addr %p (p %p)\n",
1517: pp->pr_wchan, pi->pi_magic, ph->ph_page,
1518: n, pi, page);
1519: panic("pool");
1520: }
1521: #endif
1522: page = (caddr_t)((u_long)pi & pp->pr_pagemask);
1523: if (page == ph->ph_page)
1524: continue;
1525:
1526: if (label != NULL)
1527: printf("%s: ", label);
1.16 briggs 1528: printf("pool(%p:%s): page inconsistency: page %p;"
1529: " item ordinal %d; addr %p (p %p)\n", pp,
1.3 pk 1530: pp->pr_wchan, ph->ph_page,
1531: n, pi, page);
1532: r++;
1533: goto out;
1534: }
1535: }
1536: out:
1.21 thorpej 1537: simple_unlock(&pp->pr_slock);
1.3 pk 1538: return (r);
1539: }
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