[BACK]Return to subr_pool.c CVS log [TXT][DIR] Up to [cvs.NetBSD.org] / src / sys / kern

Annotation of src/sys/kern/subr_pool.c, Revision 1.42

1.42    ! thorpej     1: /*     $NetBSD: subr_pool.c,v 1.41 2000/11/19 00:29:51 sommerfeld 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:
1.42    ! thorpej   109: static int     pool_catchup(struct pool *);
        !           110: static void    pool_prime_page(struct pool *, caddr_t);
        !           111: static void    *pool_page_alloc(unsigned long, int, int);
        !           112: static void    pool_page_free(void *, unsigned long, int);
1.3       pk        113:
1.42    ! thorpej   114: static void pool_print1(struct pool *, const char *,
        !           115:        void (*)(const char *, ...));
1.3       pk        116:
                    117: /*
                    118:  * Pool log entry. An array of these is allocated in pool_create().
                    119:  */
                    120: struct pool_log {
                    121:        const char      *pl_file;
                    122:        long            pl_line;
                    123:        int             pl_action;
1.25      thorpej   124: #define        PRLOG_GET       1
                    125: #define        PRLOG_PUT       2
1.3       pk        126:        void            *pl_addr;
1.1       pk        127: };
                    128:
1.3       pk        129: /* Number of entries in pool log buffers */
1.17      thorpej   130: #ifndef POOL_LOGSIZE
                    131: #define        POOL_LOGSIZE    10
                    132: #endif
                    133:
                    134: int pool_logsize = POOL_LOGSIZE;
1.1       pk        135:
1.25      thorpej   136: #ifdef DIAGNOSTIC
1.42    ! thorpej   137: static __inline void
        !           138: pr_log(struct pool *pp, void *v, int action, const char *file, long line)
1.3       pk        139: {
                    140:        int n = pp->pr_curlogentry;
                    141:        struct pool_log *pl;
                    142:
1.20      thorpej   143:        if ((pp->pr_roflags & PR_LOGGING) == 0)
1.3       pk        144:                return;
                    145:
                    146:        /*
                    147:         * Fill in the current entry. Wrap around and overwrite
                    148:         * the oldest entry if necessary.
                    149:         */
                    150:        pl = &pp->pr_log[n];
                    151:        pl->pl_file = file;
                    152:        pl->pl_line = line;
                    153:        pl->pl_action = action;
                    154:        pl->pl_addr = v;
                    155:        if (++n >= pp->pr_logsize)
                    156:                n = 0;
                    157:        pp->pr_curlogentry = n;
                    158: }
                    159:
                    160: static void
1.42    ! thorpej   161: pr_printlog(struct pool *pp, struct pool_item *pi,
        !           162:     void (*pr)(const char *, ...))
1.3       pk        163: {
                    164:        int i = pp->pr_logsize;
                    165:        int n = pp->pr_curlogentry;
                    166:
1.20      thorpej   167:        if ((pp->pr_roflags & PR_LOGGING) == 0)
1.3       pk        168:                return;
                    169:
                    170:        /*
                    171:         * Print all entries in this pool's log.
                    172:         */
                    173:        while (i-- > 0) {
                    174:                struct pool_log *pl = &pp->pr_log[n];
                    175:                if (pl->pl_action != 0) {
1.25      thorpej   176:                        if (pi == NULL || pi == pl->pl_addr) {
                    177:                                (*pr)("\tlog entry %d:\n", i);
                    178:                                (*pr)("\t\taction = %s, addr = %p\n",
                    179:                                    pl->pl_action == PRLOG_GET ? "get" : "put",
                    180:                                    pl->pl_addr);
                    181:                                (*pr)("\t\tfile: %s at line %lu\n",
                    182:                                    pl->pl_file, pl->pl_line);
                    183:                        }
1.3       pk        184:                }
                    185:                if (++n >= pp->pr_logsize)
                    186:                        n = 0;
                    187:        }
                    188: }
1.25      thorpej   189:
1.42    ! thorpej   190: static __inline void
        !           191: pr_enter(struct pool *pp, const char *file, long line)
1.25      thorpej   192: {
                    193:
1.34      thorpej   194:        if (__predict_false(pp->pr_entered_file != NULL)) {
1.25      thorpej   195:                printf("pool %s: reentrancy at file %s line %ld\n",
                    196:                    pp->pr_wchan, file, line);
                    197:                printf("         previous entry at file %s line %ld\n",
                    198:                    pp->pr_entered_file, pp->pr_entered_line);
                    199:                panic("pr_enter");
                    200:        }
                    201:
                    202:        pp->pr_entered_file = file;
                    203:        pp->pr_entered_line = line;
                    204: }
                    205:
1.42    ! thorpej   206: static __inline void
        !           207: pr_leave(struct pool *pp)
1.25      thorpej   208: {
                    209:
1.34      thorpej   210:        if (__predict_false(pp->pr_entered_file == NULL)) {
1.25      thorpej   211:                printf("pool %s not entered?\n", pp->pr_wchan);
                    212:                panic("pr_leave");
                    213:        }
                    214:
                    215:        pp->pr_entered_file = NULL;
                    216:        pp->pr_entered_line = 0;
                    217: }
                    218:
1.42    ! thorpej   219: static __inline void
        !           220: pr_enter_check(struct pool *pp, void (*pr)(const char *, ...))
1.25      thorpej   221: {
                    222:
                    223:        if (pp->pr_entered_file != NULL)
                    224:                (*pr)("\n\tcurrently entered from file %s line %ld\n",
                    225:                    pp->pr_entered_file, pp->pr_entered_line);
                    226: }
1.3       pk        227: #else
1.25      thorpej   228: #define        pr_log(pp, v, action, file, line)
                    229: #define        pr_printlog(pp, pi, pr)
                    230: #define        pr_enter(pp, file, line)
                    231: #define        pr_leave(pp)
                    232: #define        pr_enter_check(pp, pr)
                    233: #endif /* DIAGNOSTIC */
1.3       pk        234:
                    235: /*
                    236:  * Return the pool page header based on page address.
                    237:  */
1.42    ! thorpej   238: static __inline struct pool_item_header *
        !           239: pr_find_pagehead(struct pool *pp, caddr_t page)
1.3       pk        240: {
                    241:        struct pool_item_header *ph;
                    242:
1.20      thorpej   243:        if ((pp->pr_roflags & PR_PHINPAGE) != 0)
1.3       pk        244:                return ((struct pool_item_header *)(page + pp->pr_phoffset));
                    245:
                    246:        for (ph = LIST_FIRST(&pp->pr_hashtab[PR_HASH_INDEX(pp, page)]);
                    247:             ph != NULL;
                    248:             ph = LIST_NEXT(ph, ph_hashlist)) {
                    249:                if (ph->ph_page == page)
                    250:                        return (ph);
                    251:        }
                    252:        return (NULL);
                    253: }
                    254:
                    255: /*
                    256:  * Remove a page from the pool.
                    257:  */
1.42    ! thorpej   258: static __inline void
        !           259: pr_rmpage(struct pool *pp, struct pool_item_header *ph)
1.3       pk        260: {
                    261:
                    262:        /*
1.7       thorpej   263:         * If the page was idle, decrement the idle page count.
1.3       pk        264:         */
1.6       thorpej   265:        if (ph->ph_nmissing == 0) {
                    266: #ifdef DIAGNOSTIC
                    267:                if (pp->pr_nidle == 0)
                    268:                        panic("pr_rmpage: nidle inconsistent");
1.20      thorpej   269:                if (pp->pr_nitems < pp->pr_itemsperpage)
                    270:                        panic("pr_rmpage: nitems inconsistent");
1.6       thorpej   271: #endif
                    272:                pp->pr_nidle--;
                    273:        }
1.7       thorpej   274:
1.20      thorpej   275:        pp->pr_nitems -= pp->pr_itemsperpage;
                    276:
1.7       thorpej   277:        /*
                    278:         * Unlink a page from the pool and release it.
                    279:         */
                    280:        TAILQ_REMOVE(&pp->pr_pagelist, ph, ph_pagelist);
                    281:        (*pp->pr_free)(ph->ph_page, pp->pr_pagesz, pp->pr_mtype);
                    282:        pp->pr_npages--;
                    283:        pp->pr_npagefree++;
1.6       thorpej   284:
1.22      chs       285:        if ((pp->pr_roflags & PR_PHINPAGE) == 0) {
1.27      pk        286:                int s;
1.22      chs       287:                LIST_REMOVE(ph, ph_hashlist);
1.27      pk        288:                s = splhigh();
1.22      chs       289:                pool_put(&phpool, ph);
1.27      pk        290:                splx(s);
1.22      chs       291:        }
                    292:
1.3       pk        293:        if (pp->pr_curpage == ph) {
                    294:                /*
                    295:                 * Find a new non-empty page header, if any.
                    296:                 * Start search from the page head, to increase the
                    297:                 * chance for "high water" pages to be freed.
                    298:                 */
                    299:                for (ph = TAILQ_FIRST(&pp->pr_pagelist); ph != NULL;
                    300:                     ph = TAILQ_NEXT(ph, ph_pagelist))
                    301:                        if (TAILQ_FIRST(&ph->ph_itemlist) != NULL)
                    302:                                break;
                    303:
                    304:                pp->pr_curpage = ph;
1.21      thorpej   305:        }
1.3       pk        306: }
                    307:
                    308: /*
                    309:  * Allocate and initialize a pool.
                    310:  */
1.1       pk        311: struct pool *
1.42    ! thorpej   312: pool_create(size_t size, u_int align, u_int ioff, int nitems,
        !           313:     const char *wchan, size_t pagesz,
        !           314:     void *(*alloc)(unsigned long, int, int),
        !           315:     void (*release)(void *, unsigned long, int),
        !           316:     int mtype)
1.1       pk        317: {
                    318:        struct pool *pp;
1.3       pk        319:        int flags;
1.1       pk        320:
1.3       pk        321:        pp = (struct pool *)malloc(sizeof(*pp), M_POOL, M_NOWAIT);
                    322:        if (pp == NULL)
1.1       pk        323:                return (NULL);
1.3       pk        324:
                    325:        flags = PR_FREEHEADER;
                    326:        pool_init(pp, size, align, ioff, flags, wchan, pagesz,
                    327:                  alloc, release, mtype);
                    328:
                    329:        if (nitems != 0) {
                    330:                if (pool_prime(pp, nitems, NULL) != 0) {
                    331:                        pool_destroy(pp);
                    332:                        return (NULL);
                    333:                }
1.1       pk        334:        }
                    335:
1.3       pk        336:        return (pp);
                    337: }
                    338:
                    339: /*
                    340:  * Initialize the given pool resource structure.
                    341:  *
                    342:  * We export this routine to allow other kernel parts to declare
                    343:  * static pools that must be initialized before malloc() is available.
                    344:  */
                    345: void
1.42    ! thorpej   346: pool_init(struct pool *pp, size_t size, u_int align, u_int ioff, int flags,
        !           347:     const char *wchan, size_t pagesz,
        !           348:     void *(*alloc)(unsigned long, int, int),
        !           349:     void (*release)(void *, unsigned long, int),
        !           350:     int mtype)
1.3       pk        351: {
1.16      briggs    352:        int off, slack, i;
1.3       pk        353:
1.25      thorpej   354: #ifdef POOL_DIAGNOSTIC
                    355:        /*
                    356:         * Always log if POOL_DIAGNOSTIC is defined.
                    357:         */
                    358:        if (pool_logsize != 0)
                    359:                flags |= PR_LOGGING;
                    360: #endif
                    361:
1.3       pk        362:        /*
                    363:         * Check arguments and construct default values.
                    364:         */
1.36      pk        365:        if (!powerof2(pagesz))
1.3       pk        366:                panic("pool_init: page size invalid (%lx)\n", (u_long)pagesz);
                    367:
1.4       thorpej   368:        if (alloc == NULL && release == NULL) {
1.3       pk        369:                alloc = pool_page_alloc;
                    370:                release = pool_page_free;
1.4       thorpej   371:                pagesz = PAGE_SIZE;     /* Rounds to PAGE_SIZE anyhow. */
                    372:        } else if ((alloc != NULL && release != NULL) == 0) {
                    373:                /* If you specifiy one, must specify both. */
                    374:                panic("pool_init: must specify alloc and release together");
                    375:        }
                    376:
1.3       pk        377:        if (pagesz == 0)
                    378:                pagesz = PAGE_SIZE;
                    379:
                    380:        if (align == 0)
                    381:                align = ALIGN(1);
1.14      thorpej   382:
                    383:        if (size < sizeof(struct pool_item))
                    384:                size = sizeof(struct pool_item);
1.3       pk        385:
1.35      pk        386:        size = ALIGN(size);
                    387:        if (size >= pagesz)
                    388:                panic("pool_init: pool item size (%lu) too large",
                    389:                      (u_long)size);
                    390:
1.3       pk        391:        /*
                    392:         * Initialize the pool structure.
                    393:         */
                    394:        TAILQ_INIT(&pp->pr_pagelist);
                    395:        pp->pr_curpage = NULL;
                    396:        pp->pr_npages = 0;
                    397:        pp->pr_minitems = 0;
                    398:        pp->pr_minpages = 0;
                    399:        pp->pr_maxpages = UINT_MAX;
1.20      thorpej   400:        pp->pr_roflags = flags;
                    401:        pp->pr_flags = 0;
1.35      pk        402:        pp->pr_size = size;
1.3       pk        403:        pp->pr_align = align;
                    404:        pp->pr_wchan = wchan;
                    405:        pp->pr_mtype = mtype;
                    406:        pp->pr_alloc = alloc;
                    407:        pp->pr_free = release;
                    408:        pp->pr_pagesz = pagesz;
                    409:        pp->pr_pagemask = ~(pagesz - 1);
                    410:        pp->pr_pageshift = ffs(pagesz) - 1;
1.20      thorpej   411:        pp->pr_nitems = 0;
                    412:        pp->pr_nout = 0;
                    413:        pp->pr_hardlimit = UINT_MAX;
                    414:        pp->pr_hardlimit_warning = NULL;
1.31      thorpej   415:        pp->pr_hardlimit_ratecap.tv_sec = 0;
                    416:        pp->pr_hardlimit_ratecap.tv_usec = 0;
                    417:        pp->pr_hardlimit_warning_last.tv_sec = 0;
                    418:        pp->pr_hardlimit_warning_last.tv_usec = 0;
1.3       pk        419:
                    420:        /*
                    421:         * Decide whether to put the page header off page to avoid
                    422:         * wasting too large a part of the page. Off-page page headers
                    423:         * go on a hash table, so we can match a returned item
                    424:         * with its header based on the page address.
                    425:         * We use 1/16 of the page size as the threshold (XXX: tune)
                    426:         */
                    427:        if (pp->pr_size < pagesz/16) {
                    428:                /* Use the end of the page for the page header */
1.20      thorpej   429:                pp->pr_roflags |= PR_PHINPAGE;
1.3       pk        430:                pp->pr_phoffset = off =
                    431:                        pagesz - ALIGN(sizeof(struct pool_item_header));
1.2       pk        432:        } else {
1.3       pk        433:                /* The page header will be taken from our page header pool */
                    434:                pp->pr_phoffset = 0;
                    435:                off = pagesz;
1.16      briggs    436:                for (i = 0; i < PR_HASHTABSIZE; i++) {
                    437:                        LIST_INIT(&pp->pr_hashtab[i]);
                    438:                }
1.2       pk        439:        }
1.1       pk        440:
1.3       pk        441:        /*
                    442:         * Alignment is to take place at `ioff' within the item. This means
                    443:         * we must reserve up to `align - 1' bytes on the page to allow
                    444:         * appropriate positioning of each item.
                    445:         *
                    446:         * Silently enforce `0 <= ioff < align'.
                    447:         */
                    448:        pp->pr_itemoffset = ioff = ioff % align;
                    449:        pp->pr_itemsperpage = (off - ((align - ioff) % align)) / pp->pr_size;
                    450:
                    451:        /*
                    452:         * Use the slack between the chunks and the page header
                    453:         * for "cache coloring".
                    454:         */
                    455:        slack = off - pp->pr_itemsperpage * pp->pr_size;
                    456:        pp->pr_maxcolor = (slack / align) * align;
                    457:        pp->pr_curcolor = 0;
                    458:
                    459:        pp->pr_nget = 0;
                    460:        pp->pr_nfail = 0;
                    461:        pp->pr_nput = 0;
                    462:        pp->pr_npagealloc = 0;
                    463:        pp->pr_npagefree = 0;
1.1       pk        464:        pp->pr_hiwat = 0;
1.8       thorpej   465:        pp->pr_nidle = 0;
1.3       pk        466:
1.25      thorpej   467:        if (flags & PR_LOGGING) {
                    468:                if (kmem_map == NULL ||
                    469:                    (pp->pr_log = malloc(pool_logsize * sizeof(struct pool_log),
                    470:                     M_TEMP, M_NOWAIT)) == NULL)
1.20      thorpej   471:                        pp->pr_roflags &= ~PR_LOGGING;
1.3       pk        472:                pp->pr_curlogentry = 0;
                    473:                pp->pr_logsize = pool_logsize;
                    474:        }
1.25      thorpej   475:
                    476:        pp->pr_entered_file = NULL;
                    477:        pp->pr_entered_line = 0;
1.3       pk        478:
1.21      thorpej   479:        simple_lock_init(&pp->pr_slock);
1.1       pk        480:
1.3       pk        481:        /*
                    482:         * Initialize private page header pool if we haven't done so yet.
1.23      thorpej   483:         * XXX LOCKING.
1.3       pk        484:         */
                    485:        if (phpool.pr_size == 0) {
                    486:                pool_init(&phpool, sizeof(struct pool_item_header), 0, 0,
                    487:                          0, "phpool", 0, 0, 0, 0);
1.1       pk        488:        }
                    489:
1.23      thorpej   490:        /* Insert into the list of all pools. */
                    491:        simple_lock(&pool_head_slock);
                    492:        TAILQ_INSERT_TAIL(&pool_head, pp, pr_poollist);
                    493:        simple_unlock(&pool_head_slock);
1.1       pk        494: }
                    495:
                    496: /*
                    497:  * De-commision a pool resource.
                    498:  */
                    499: void
1.42    ! thorpej   500: pool_destroy(struct pool *pp)
1.1       pk        501: {
1.3       pk        502:        struct pool_item_header *ph;
                    503:
                    504: #ifdef DIAGNOSTIC
1.20      thorpej   505:        if (pp->pr_nout != 0) {
1.25      thorpej   506:                pr_printlog(pp, NULL, printf);
1.20      thorpej   507:                panic("pool_destroy: pool busy: still out: %u\n",
                    508:                    pp->pr_nout);
1.3       pk        509:        }
                    510: #endif
1.1       pk        511:
1.3       pk        512:        /* Remove all pages */
1.20      thorpej   513:        if ((pp->pr_roflags & PR_STATIC) == 0)
1.3       pk        514:                while ((ph = pp->pr_pagelist.tqh_first) != NULL)
                    515:                        pr_rmpage(pp, ph);
                    516:
                    517:        /* Remove from global pool list */
1.23      thorpej   518:        simple_lock(&pool_head_slock);
1.3       pk        519:        TAILQ_REMOVE(&pool_head, pp, pr_poollist);
1.23      thorpej   520:        /* XXX Only clear this if we were drainpp? */
1.3       pk        521:        drainpp = NULL;
1.23      thorpej   522:        simple_unlock(&pool_head_slock);
1.3       pk        523:
1.20      thorpej   524:        if ((pp->pr_roflags & PR_LOGGING) != 0)
1.3       pk        525:                free(pp->pr_log, M_TEMP);
1.2       pk        526:
1.20      thorpej   527:        if (pp->pr_roflags & PR_FREEHEADER)
1.3       pk        528:                free(pp, M_POOL);
1.1       pk        529: }
                    530:
                    531:
                    532: /*
1.3       pk        533:  * Grab an item from the pool; must be called at appropriate spl level
1.1       pk        534:  */
1.3       pk        535: void *
1.42    ! thorpej   536: _pool_get(struct pool *pp, int flags, const char *file, long line)
1.1       pk        537: {
                    538:        void *v;
                    539:        struct pool_item *pi;
1.3       pk        540:        struct pool_item_header *ph;
1.1       pk        541:
1.2       pk        542: #ifdef DIAGNOSTIC
1.34      thorpej   543:        if (__predict_false((pp->pr_roflags & PR_STATIC) &&
                    544:                            (flags & PR_MALLOCOK))) {
1.25      thorpej   545:                pr_printlog(pp, NULL, printf);
1.2       pk        546:                panic("pool_get: static");
1.3       pk        547:        }
1.2       pk        548: #endif
                    549:
1.37      sommerfe  550:        if (__predict_false(curproc == NULL && doing_shutdown == 0 &&
                    551:                            (flags & PR_WAITOK) != 0))
1.3       pk        552:                panic("pool_get: must have NOWAIT");
1.1       pk        553:
1.21      thorpej   554:        simple_lock(&pp->pr_slock);
1.25      thorpej   555:        pr_enter(pp, file, line);
1.20      thorpej   556:
                    557:  startover:
                    558:        /*
                    559:         * Check to see if we've reached the hard limit.  If we have,
                    560:         * and we can wait, then wait until an item has been returned to
                    561:         * the pool.
                    562:         */
                    563: #ifdef DIAGNOSTIC
1.34      thorpej   564:        if (__predict_false(pp->pr_nout > pp->pr_hardlimit)) {
1.25      thorpej   565:                pr_leave(pp);
1.21      thorpej   566:                simple_unlock(&pp->pr_slock);
1.20      thorpej   567:                panic("pool_get: %s: crossed hard limit", pp->pr_wchan);
                    568:        }
                    569: #endif
1.34      thorpej   570:        if (__predict_false(pp->pr_nout == pp->pr_hardlimit)) {
1.29      sommerfe  571:                if ((flags & PR_WAITOK) && !(flags & PR_LIMITFAIL)) {
1.20      thorpej   572:                        /*
                    573:                         * XXX: A warning isn't logged in this case.  Should
                    574:                         * it be?
                    575:                         */
                    576:                        pp->pr_flags |= PR_WANTED;
1.25      thorpej   577:                        pr_leave(pp);
1.40      sommerfe  578:                        ltsleep(pp, PSWP, pp->pr_wchan, 0, &pp->pr_slock);
1.25      thorpej   579:                        pr_enter(pp, file, line);
1.20      thorpej   580:                        goto startover;
                    581:                }
1.31      thorpej   582:
                    583:                /*
                    584:                 * Log a message that the hard limit has been hit.
                    585:                 */
                    586:                if (pp->pr_hardlimit_warning != NULL &&
                    587:                    ratecheck(&pp->pr_hardlimit_warning_last,
                    588:                              &pp->pr_hardlimit_ratecap))
                    589:                        log(LOG_ERR, "%s\n", pp->pr_hardlimit_warning);
1.21      thorpej   590:
                    591:                if (flags & PR_URGENT)
                    592:                        panic("pool_get: urgent");
                    593:
                    594:                pp->pr_nfail++;
                    595:
1.25      thorpej   596:                pr_leave(pp);
1.21      thorpej   597:                simple_unlock(&pp->pr_slock);
1.20      thorpej   598:                return (NULL);
                    599:        }
                    600:
1.3       pk        601:        /*
                    602:         * The convention we use is that if `curpage' is not NULL, then
                    603:         * it points at a non-empty bucket. In particular, `curpage'
                    604:         * never points at a page header which has PR_PHINPAGE set and
                    605:         * has no items in its bucket.
                    606:         */
1.20      thorpej   607:        if ((ph = pp->pr_curpage) == NULL) {
1.15      pk        608:                void *v;
                    609:
1.20      thorpej   610: #ifdef DIAGNOSTIC
                    611:                if (pp->pr_nitems != 0) {
1.21      thorpej   612:                        simple_unlock(&pp->pr_slock);
1.20      thorpej   613:                        printf("pool_get: %s: curpage NULL, nitems %u\n",
                    614:                            pp->pr_wchan, pp->pr_nitems);
                    615:                        panic("pool_get: nitems inconsistent\n");
                    616:                }
                    617: #endif
                    618:
1.21      thorpej   619:                /*
                    620:                 * Call the back-end page allocator for more memory.
                    621:                 * Release the pool lock, as the back-end page allocator
                    622:                 * may block.
                    623:                 */
1.25      thorpej   624:                pr_leave(pp);
1.21      thorpej   625:                simple_unlock(&pp->pr_slock);
                    626:                v = (*pp->pr_alloc)(pp->pr_pagesz, flags, pp->pr_mtype);
                    627:                simple_lock(&pp->pr_slock);
1.25      thorpej   628:                pr_enter(pp, file, line);
1.15      pk        629:
1.21      thorpej   630:                if (v == NULL) {
                    631:                        /*
                    632:                         * We were unable to allocate a page, but
                    633:                         * we released the lock during allocation,
                    634:                         * so perhaps items were freed back to the
                    635:                         * pool.  Check for this case.
                    636:                         */
                    637:                        if (pp->pr_curpage != NULL)
                    638:                                goto startover;
1.15      pk        639:
1.3       pk        640:                        if (flags & PR_URGENT)
                    641:                                panic("pool_get: urgent");
1.21      thorpej   642:
1.3       pk        643:                        if ((flags & PR_WAITOK) == 0) {
                    644:                                pp->pr_nfail++;
1.25      thorpej   645:                                pr_leave(pp);
1.21      thorpej   646:                                simple_unlock(&pp->pr_slock);
1.1       pk        647:                                return (NULL);
1.3       pk        648:                        }
                    649:
1.15      pk        650:                        /*
                    651:                         * Wait for items to be returned to this pool.
1.21      thorpej   652:                         *
1.15      pk        653:                         * XXX: we actually want to wait just until
                    654:                         * the page allocator has memory again. Depending
                    655:                         * on this pool's usage, we might get stuck here
                    656:                         * for a long time.
1.20      thorpej   657:                         *
                    658:                         * XXX: maybe we should wake up once a second and
                    659:                         * try again?
1.15      pk        660:                         */
1.1       pk        661:                        pp->pr_flags |= PR_WANTED;
1.25      thorpej   662:                        pr_leave(pp);
1.40      sommerfe  663:                        ltsleep(pp, PSWP, pp->pr_wchan, 0, &pp->pr_slock);
1.25      thorpej   664:                        pr_enter(pp, file, line);
1.20      thorpej   665:                        goto startover;
1.1       pk        666:                }
1.3       pk        667:
1.15      pk        668:                /* We have more memory; add it to the pool */
                    669:                pp->pr_npagealloc++;
                    670:                pool_prime_page(pp, v);
                    671:
1.20      thorpej   672:                /* Start the allocation process over. */
                    673:                goto startover;
1.3       pk        674:        }
                    675:
1.34      thorpej   676:        if (__predict_false((v = pi = TAILQ_FIRST(&ph->ph_itemlist)) == NULL)) {
1.25      thorpej   677:                pr_leave(pp);
1.21      thorpej   678:                simple_unlock(&pp->pr_slock);
1.3       pk        679:                panic("pool_get: %s: page empty", pp->pr_wchan);
1.21      thorpej   680:        }
1.20      thorpej   681: #ifdef DIAGNOSTIC
1.34      thorpej   682:        if (__predict_false(pp->pr_nitems == 0)) {
1.25      thorpej   683:                pr_leave(pp);
1.21      thorpej   684:                simple_unlock(&pp->pr_slock);
1.20      thorpej   685:                printf("pool_get: %s: items on itemlist, nitems %u\n",
                    686:                    pp->pr_wchan, pp->pr_nitems);
                    687:                panic("pool_get: nitems inconsistent\n");
                    688:        }
                    689: #endif
1.3       pk        690:        pr_log(pp, v, PRLOG_GET, file, line);
                    691:
                    692: #ifdef DIAGNOSTIC
1.34      thorpej   693:        if (__predict_false(pi->pi_magic != PI_MAGIC)) {
1.25      thorpej   694:                pr_printlog(pp, pi, printf);
1.3       pk        695:                panic("pool_get(%s): free list modified: magic=%x; page %p;"
                    696:                       " item addr %p\n",
                    697:                        pp->pr_wchan, pi->pi_magic, ph->ph_page, pi);
                    698:        }
                    699: #endif
                    700:
                    701:        /*
                    702:         * Remove from item list.
                    703:         */
                    704:        TAILQ_REMOVE(&ph->ph_itemlist, pi, pi_list);
1.20      thorpej   705:        pp->pr_nitems--;
                    706:        pp->pr_nout++;
1.6       thorpej   707:        if (ph->ph_nmissing == 0) {
                    708: #ifdef DIAGNOSTIC
1.34      thorpej   709:                if (__predict_false(pp->pr_nidle == 0))
1.6       thorpej   710:                        panic("pool_get: nidle inconsistent");
                    711: #endif
                    712:                pp->pr_nidle--;
                    713:        }
1.3       pk        714:        ph->ph_nmissing++;
                    715:        if (TAILQ_FIRST(&ph->ph_itemlist) == NULL) {
1.21      thorpej   716: #ifdef DIAGNOSTIC
1.34      thorpej   717:                if (__predict_false(ph->ph_nmissing != pp->pr_itemsperpage)) {
1.25      thorpej   718:                        pr_leave(pp);
1.21      thorpej   719:                        simple_unlock(&pp->pr_slock);
                    720:                        panic("pool_get: %s: nmissing inconsistent",
                    721:                            pp->pr_wchan);
                    722:                }
                    723: #endif
1.3       pk        724:                /*
                    725:                 * Find a new non-empty page header, if any.
                    726:                 * Start search from the page head, to increase
                    727:                 * the chance for "high water" pages to be freed.
                    728:                 *
1.21      thorpej   729:                 * Migrate empty pages to the end of the list.  This
                    730:                 * will speed the update of curpage as pages become
                    731:                 * idle.  Empty pages intermingled with idle pages
                    732:                 * is no big deal.  As soon as a page becomes un-empty,
                    733:                 * it will move back to the head of the list.
1.3       pk        734:                 */
                    735:                TAILQ_REMOVE(&pp->pr_pagelist, ph, ph_pagelist);
1.21      thorpej   736:                TAILQ_INSERT_TAIL(&pp->pr_pagelist, ph, ph_pagelist);
                    737:                for (ph = TAILQ_FIRST(&pp->pr_pagelist); ph != NULL;
                    738:                     ph = TAILQ_NEXT(ph, ph_pagelist))
1.3       pk        739:                        if (TAILQ_FIRST(&ph->ph_itemlist) != NULL)
                    740:                                break;
                    741:
                    742:                pp->pr_curpage = ph;
1.1       pk        743:        }
1.3       pk        744:
                    745:        pp->pr_nget++;
1.20      thorpej   746:
                    747:        /*
                    748:         * If we have a low water mark and we are now below that low
                    749:         * water mark, add more items to the pool.
                    750:         */
                    751:        if (pp->pr_nitems < pp->pr_minitems && pool_catchup(pp) != 0) {
                    752:                /*
                    753:                 * XXX: Should we log a warning?  Should we set up a timeout
                    754:                 * to try again in a second or so?  The latter could break
                    755:                 * a caller's assumptions about interrupt protection, etc.
                    756:                 */
                    757:        }
                    758:
1.25      thorpej   759:        pr_leave(pp);
1.21      thorpej   760:        simple_unlock(&pp->pr_slock);
1.1       pk        761:        return (v);
                    762: }
                    763:
                    764: /*
1.3       pk        765:  * Return resource to the pool; must be called at appropriate spl level
1.1       pk        766:  */
1.3       pk        767: void
1.42    ! thorpej   768: _pool_put(struct pool *pp, void *v, const char *file, long line)
1.1       pk        769: {
                    770:        struct pool_item *pi = v;
1.3       pk        771:        struct pool_item_header *ph;
                    772:        caddr_t page;
1.21      thorpej   773:        int s;
1.3       pk        774:
                    775:        page = (caddr_t)((u_long)v & pp->pr_pagemask);
1.1       pk        776:
1.21      thorpej   777:        simple_lock(&pp->pr_slock);
1.25      thorpej   778:        pr_enter(pp, file, line);
1.30      thorpej   779:
                    780: #ifdef DIAGNOSTIC
1.34      thorpej   781:        if (__predict_false(pp->pr_nout == 0)) {
1.30      thorpej   782:                printf("pool %s: putting with none out\n",
                    783:                    pp->pr_wchan);
                    784:                panic("pool_put");
                    785:        }
                    786: #endif
1.3       pk        787:
                    788:        pr_log(pp, v, PRLOG_PUT, file, line);
                    789:
1.34      thorpej   790:        if (__predict_false((ph = pr_find_pagehead(pp, page)) == NULL)) {
1.25      thorpej   791:                pr_printlog(pp, NULL, printf);
1.3       pk        792:                panic("pool_put: %s: page header missing", pp->pr_wchan);
                    793:        }
1.28      thorpej   794:
                    795: #ifdef LOCKDEBUG
                    796:        /*
                    797:         * Check if we're freeing a locked simple lock.
                    798:         */
                    799:        simple_lock_freecheck((caddr_t)pi, ((caddr_t)pi) + pp->pr_size);
                    800: #endif
1.3       pk        801:
                    802:        /*
                    803:         * Return to item list.
                    804:         */
1.2       pk        805: #ifdef DIAGNOSTIC
1.3       pk        806:        pi->pi_magic = PI_MAGIC;
                    807: #endif
1.32      chs       808: #ifdef DEBUG
                    809:        {
                    810:                int i, *ip = v;
                    811:
                    812:                for (i = 0; i < pp->pr_size / sizeof(int); i++) {
                    813:                        *ip++ = PI_MAGIC;
                    814:                }
                    815:        }
                    816: #endif
                    817:
1.3       pk        818:        TAILQ_INSERT_HEAD(&ph->ph_itemlist, pi, pi_list);
                    819:        ph->ph_nmissing--;
                    820:        pp->pr_nput++;
1.20      thorpej   821:        pp->pr_nitems++;
                    822:        pp->pr_nout--;
1.3       pk        823:
                    824:        /* Cancel "pool empty" condition if it exists */
                    825:        if (pp->pr_curpage == NULL)
                    826:                pp->pr_curpage = ph;
                    827:
                    828:        if (pp->pr_flags & PR_WANTED) {
                    829:                pp->pr_flags &= ~PR_WANTED;
1.15      pk        830:                if (ph->ph_nmissing == 0)
                    831:                        pp->pr_nidle++;
1.25      thorpej   832:                pr_leave(pp);
1.21      thorpej   833:                simple_unlock(&pp->pr_slock);
1.3       pk        834:                wakeup((caddr_t)pp);
                    835:                return;
                    836:        }
                    837:
                    838:        /*
1.21      thorpej   839:         * If this page is now complete, do one of two things:
                    840:         *
                    841:         *      (1) If we have more pages than the page high water
                    842:         *          mark, free the page back to the system.
                    843:         *
                    844:         *      (2) Move it to the end of the page list, so that
                    845:         *          we minimize our chances of fragmenting the
                    846:         *          pool.  Idle pages migrate to the end (along with
                    847:         *          completely empty pages, so that we find un-empty
                    848:         *          pages more quickly when we update curpage) of the
                    849:         *          list so they can be more easily swept up by
                    850:         *          the pagedaemon when pages are scarce.
1.3       pk        851:         */
                    852:        if (ph->ph_nmissing == 0) {
1.6       thorpej   853:                pp->pr_nidle++;
1.3       pk        854:                if (pp->pr_npages > pp->pr_maxpages) {
                    855:                        pr_rmpage(pp, ph);
                    856:                } else {
                    857:                        TAILQ_REMOVE(&pp->pr_pagelist, ph, ph_pagelist);
                    858:                        TAILQ_INSERT_TAIL(&pp->pr_pagelist, ph, ph_pagelist);
                    859:
1.21      thorpej   860:                        /*
                    861:                         * Update the timestamp on the page.  A page must
                    862:                         * be idle for some period of time before it can
                    863:                         * be reclaimed by the pagedaemon.  This minimizes
                    864:                         * ping-pong'ing for memory.
                    865:                         */
                    866:                        s = splclock();
                    867:                        ph->ph_time = mono_time;
                    868:                        splx(s);
                    869:
                    870:                        /*
                    871:                         * Update the current page pointer.  Just look for
                    872:                         * the first page with any free items.
                    873:                         *
                    874:                         * XXX: Maybe we want an option to look for the
                    875:                         * page with the fewest available items, to minimize
                    876:                         * fragmentation?
                    877:                         */
1.3       pk        878:                        for (ph = TAILQ_FIRST(&pp->pr_pagelist); ph != NULL;
                    879:                             ph = TAILQ_NEXT(ph, ph_pagelist))
                    880:                                if (TAILQ_FIRST(&ph->ph_itemlist) != NULL)
                    881:                                        break;
1.1       pk        882:
1.3       pk        883:                        pp->pr_curpage = ph;
1.1       pk        884:                }
                    885:        }
1.21      thorpej   886:        /*
                    887:         * If the page has just become un-empty, move it to the head of
                    888:         * the list, and make it the current page.  The next allocation
                    889:         * will get the item from this page, instead of further fragmenting
                    890:         * the pool.
                    891:         */
                    892:        else if (ph->ph_nmissing == (pp->pr_itemsperpage - 1)) {
                    893:                TAILQ_REMOVE(&pp->pr_pagelist, ph, ph_pagelist);
                    894:                TAILQ_INSERT_HEAD(&pp->pr_pagelist, ph, ph_pagelist);
                    895:                pp->pr_curpage = ph;
                    896:        }
                    897:
1.25      thorpej   898:        pr_leave(pp);
1.21      thorpej   899:        simple_unlock(&pp->pr_slock);
1.3       pk        900:
1.1       pk        901: }
                    902:
                    903: /*
1.3       pk        904:  * Add N items to the pool.
1.1       pk        905:  */
                    906: int
1.42    ! thorpej   907: pool_prime(struct pool *pp, int n, caddr_t storage)
1.1       pk        908: {
1.3       pk        909:        caddr_t cp;
                    910:        int newnitems, newpages;
1.2       pk        911:
                    912: #ifdef DIAGNOSTIC
1.34      thorpej   913:        if (__predict_false(storage && !(pp->pr_roflags & PR_STATIC)))
1.2       pk        914:                panic("pool_prime: static");
                    915:        /* !storage && static caught below */
                    916: #endif
1.1       pk        917:
1.21      thorpej   918:        simple_lock(&pp->pr_slock);
                    919:
1.3       pk        920:        newnitems = pp->pr_minitems + n;
                    921:        newpages =
1.18      thorpej   922:                roundup(newnitems, pp->pr_itemsperpage) / pp->pr_itemsperpage
1.3       pk        923:                - pp->pr_minpages;
                    924:
                    925:        while (newpages-- > 0) {
1.20      thorpej   926:                if (pp->pr_roflags & PR_STATIC) {
1.3       pk        927:                        cp = storage;
                    928:                        storage += pp->pr_pagesz;
                    929:                } else {
1.21      thorpej   930:                        simple_unlock(&pp->pr_slock);
1.3       pk        931:                        cp = (*pp->pr_alloc)(pp->pr_pagesz, 0, pp->pr_mtype);
1.21      thorpej   932:                        simple_lock(&pp->pr_slock);
1.3       pk        933:                }
1.2       pk        934:
1.3       pk        935:                if (cp == NULL) {
1.21      thorpej   936:                        simple_unlock(&pp->pr_slock);
1.1       pk        937:                        return (ENOMEM);
                    938:                }
                    939:
1.26      thorpej   940:                pp->pr_npagealloc++;
1.3       pk        941:                pool_prime_page(pp, cp);
                    942:                pp->pr_minpages++;
1.1       pk        943:        }
1.3       pk        944:
                    945:        pp->pr_minitems = newnitems;
                    946:
                    947:        if (pp->pr_minpages >= pp->pr_maxpages)
                    948:                pp->pr_maxpages = pp->pr_minpages + 1;  /* XXX */
                    949:
1.21      thorpej   950:        simple_unlock(&pp->pr_slock);
1.1       pk        951:        return (0);
                    952: }
1.3       pk        953:
                    954: /*
                    955:  * Add a page worth of items to the pool.
1.21      thorpej   956:  *
                    957:  * Note, we must be called with the pool descriptor LOCKED.
1.3       pk        958:  */
1.21      thorpej   959: static void
1.42    ! thorpej   960: pool_prime_page(struct pool *pp, caddr_t storage)
1.3       pk        961: {
                    962:        struct pool_item *pi;
                    963:        struct pool_item_header *ph;
                    964:        caddr_t cp = storage;
                    965:        unsigned int align = pp->pr_align;
                    966:        unsigned int ioff = pp->pr_itemoffset;
1.27      pk        967:        int s, n;
1.36      pk        968:
                    969:        if (((u_long)cp & (pp->pr_pagesz - 1)) != 0)
                    970:                panic("pool_prime_page: %s: unaligned page", pp->pr_wchan);
1.3       pk        971:
1.20      thorpej   972:        if ((pp->pr_roflags & PR_PHINPAGE) != 0) {
1.3       pk        973:                ph = (struct pool_item_header *)(cp + pp->pr_phoffset);
                    974:        } else {
1.27      pk        975:                s = splhigh();
1.3       pk        976:                ph = pool_get(&phpool, PR_URGENT);
1.27      pk        977:                splx(s);
1.3       pk        978:                LIST_INSERT_HEAD(&pp->pr_hashtab[PR_HASH_INDEX(pp, cp)],
                    979:                                 ph, ph_hashlist);
                    980:        }
                    981:
                    982:        /*
                    983:         * Insert page header.
                    984:         */
                    985:        TAILQ_INSERT_HEAD(&pp->pr_pagelist, ph, ph_pagelist);
                    986:        TAILQ_INIT(&ph->ph_itemlist);
                    987:        ph->ph_page = storage;
                    988:        ph->ph_nmissing = 0;
1.21      thorpej   989:        memset(&ph->ph_time, 0, sizeof(ph->ph_time));
1.3       pk        990:
1.6       thorpej   991:        pp->pr_nidle++;
                    992:
1.3       pk        993:        /*
                    994:         * Color this page.
                    995:         */
                    996:        cp = (caddr_t)(cp + pp->pr_curcolor);
                    997:        if ((pp->pr_curcolor += align) > pp->pr_maxcolor)
                    998:                pp->pr_curcolor = 0;
                    999:
                   1000:        /*
                   1001:         * Adjust storage to apply aligment to `pr_itemoffset' in each item.
                   1002:         */
                   1003:        if (ioff != 0)
                   1004:                cp = (caddr_t)(cp + (align - ioff));
                   1005:
                   1006:        /*
                   1007:         * Insert remaining chunks on the bucket list.
                   1008:         */
                   1009:        n = pp->pr_itemsperpage;
1.20      thorpej  1010:        pp->pr_nitems += n;
1.3       pk       1011:
                   1012:        while (n--) {
                   1013:                pi = (struct pool_item *)cp;
                   1014:
                   1015:                /* Insert on page list */
                   1016:                TAILQ_INSERT_TAIL(&ph->ph_itemlist, pi, pi_list);
                   1017: #ifdef DIAGNOSTIC
                   1018:                pi->pi_magic = PI_MAGIC;
                   1019: #endif
                   1020:                cp = (caddr_t)(cp + pp->pr_size);
                   1021:        }
                   1022:
                   1023:        /*
                   1024:         * If the pool was depleted, point at the new page.
                   1025:         */
                   1026:        if (pp->pr_curpage == NULL)
                   1027:                pp->pr_curpage = ph;
                   1028:
                   1029:        if (++pp->pr_npages > pp->pr_hiwat)
                   1030:                pp->pr_hiwat = pp->pr_npages;
                   1031: }
                   1032:
1.20      thorpej  1033: /*
                   1034:  * Like pool_prime(), except this is used by pool_get() when nitems
                   1035:  * drops below the low water mark.  This is used to catch up nitmes
                   1036:  * with the low water mark.
                   1037:  *
1.21      thorpej  1038:  * Note 1, we never wait for memory here, we let the caller decide what to do.
1.20      thorpej  1039:  *
                   1040:  * Note 2, this doesn't work with static pools.
                   1041:  *
                   1042:  * Note 3, we must be called with the pool already locked, and we return
                   1043:  * with it locked.
                   1044:  */
                   1045: static int
1.42    ! thorpej  1046: pool_catchup(struct pool *pp)
1.20      thorpej  1047: {
                   1048:        caddr_t cp;
                   1049:        int error = 0;
                   1050:
                   1051:        if (pp->pr_roflags & PR_STATIC) {
                   1052:                /*
                   1053:                 * We dropped below the low water mark, and this is not a
                   1054:                 * good thing.  Log a warning.
1.21      thorpej  1055:                 *
                   1056:                 * XXX: rate-limit this?
1.20      thorpej  1057:                 */
                   1058:                printf("WARNING: static pool `%s' dropped below low water "
                   1059:                    "mark\n", pp->pr_wchan);
                   1060:                return (0);
                   1061:        }
                   1062:
1.21      thorpej  1063:        while (pp->pr_nitems < pp->pr_minitems) {
1.20      thorpej  1064:                /*
1.21      thorpej  1065:                 * Call the page back-end allocator for more memory.
                   1066:                 *
                   1067:                 * XXX: We never wait, so should we bother unlocking
                   1068:                 * the pool descriptor?
1.20      thorpej  1069:                 */
1.21      thorpej  1070:                simple_unlock(&pp->pr_slock);
1.20      thorpej  1071:                cp = (*pp->pr_alloc)(pp->pr_pagesz, 0, pp->pr_mtype);
1.21      thorpej  1072:                simple_lock(&pp->pr_slock);
1.34      thorpej  1073:                if (__predict_false(cp == NULL)) {
1.20      thorpej  1074:                        error = ENOMEM;
                   1075:                        break;
                   1076:                }
1.26      thorpej  1077:                pp->pr_npagealloc++;
1.20      thorpej  1078:                pool_prime_page(pp, cp);
                   1079:        }
                   1080:
                   1081:        return (error);
                   1082: }
                   1083:
1.3       pk       1084: void
1.42    ! thorpej  1085: pool_setlowat(struct pool *pp, int n)
1.3       pk       1086: {
1.20      thorpej  1087:        int error;
1.15      pk       1088:
1.21      thorpej  1089:        simple_lock(&pp->pr_slock);
                   1090:
1.3       pk       1091:        pp->pr_minitems = n;
1.15      pk       1092:        pp->pr_minpages = (n == 0)
                   1093:                ? 0
1.18      thorpej  1094:                : roundup(n, pp->pr_itemsperpage) / pp->pr_itemsperpage;
1.20      thorpej  1095:
                   1096:        /* Make sure we're caught up with the newly-set low water mark. */
1.41      sommerfe 1097:        if ((pp->pr_nitems < pp->pr_minitems) &&
                   1098:            (error = pool_catchup(pp)) != 0) {
1.20      thorpej  1099:                /*
                   1100:                 * XXX: Should we log a warning?  Should we set up a timeout
                   1101:                 * to try again in a second or so?  The latter could break
                   1102:                 * a caller's assumptions about interrupt protection, etc.
                   1103:                 */
                   1104:        }
1.21      thorpej  1105:
                   1106:        simple_unlock(&pp->pr_slock);
1.3       pk       1107: }
                   1108:
                   1109: void
1.42    ! thorpej  1110: pool_sethiwat(struct pool *pp, int n)
1.3       pk       1111: {
1.15      pk       1112:
1.21      thorpej  1113:        simple_lock(&pp->pr_slock);
                   1114:
1.15      pk       1115:        pp->pr_maxpages = (n == 0)
                   1116:                ? 0
1.18      thorpej  1117:                : roundup(n, pp->pr_itemsperpage) / pp->pr_itemsperpage;
1.21      thorpej  1118:
                   1119:        simple_unlock(&pp->pr_slock);
1.3       pk       1120: }
                   1121:
1.20      thorpej  1122: void
1.42    ! thorpej  1123: pool_sethardlimit(struct pool *pp, int n, const char *warnmess, int ratecap)
1.20      thorpej  1124: {
                   1125:
1.21      thorpej  1126:        simple_lock(&pp->pr_slock);
1.20      thorpej  1127:
                   1128:        pp->pr_hardlimit = n;
                   1129:        pp->pr_hardlimit_warning = warnmess;
1.31      thorpej  1130:        pp->pr_hardlimit_ratecap.tv_sec = ratecap;
                   1131:        pp->pr_hardlimit_warning_last.tv_sec = 0;
                   1132:        pp->pr_hardlimit_warning_last.tv_usec = 0;
1.20      thorpej  1133:
                   1134:        /*
1.21      thorpej  1135:         * In-line version of pool_sethiwat(), because we don't want to
                   1136:         * release the lock.
1.20      thorpej  1137:         */
                   1138:        pp->pr_maxpages = (n == 0)
                   1139:                ? 0
                   1140:                : roundup(n, pp->pr_itemsperpage) / pp->pr_itemsperpage;
1.21      thorpej  1141:
                   1142:        simple_unlock(&pp->pr_slock);
1.20      thorpej  1143: }
1.3       pk       1144:
                   1145: /*
                   1146:  * Default page allocator.
                   1147:  */
                   1148: static void *
1.42    ! thorpej  1149: pool_page_alloc(unsigned long sz, int flags, int mtype)
1.3       pk       1150: {
1.11      thorpej  1151:        boolean_t waitok = (flags & PR_WAITOK) ? TRUE : FALSE;
1.3       pk       1152:
1.11      thorpej  1153:        return ((void *)uvm_km_alloc_poolpage(waitok));
1.3       pk       1154: }
                   1155:
                   1156: static void
1.42    ! thorpej  1157: pool_page_free(void *v, unsigned long sz, int mtype)
1.3       pk       1158: {
                   1159:
1.10      eeh      1160:        uvm_km_free_poolpage((vaddr_t)v);
1.3       pk       1161: }
1.12      thorpej  1162:
                   1163: /*
                   1164:  * Alternate pool page allocator for pools that know they will
                   1165:  * never be accessed in interrupt context.
                   1166:  */
                   1167: void *
1.42    ! thorpej  1168: pool_page_alloc_nointr(unsigned long sz, int flags, int mtype)
1.12      thorpej  1169: {
                   1170:        boolean_t waitok = (flags & PR_WAITOK) ? TRUE : FALSE;
                   1171:
                   1172:        return ((void *)uvm_km_alloc_poolpage1(kernel_map, uvm.kernel_object,
                   1173:            waitok));
                   1174: }
                   1175:
                   1176: void
1.42    ! thorpej  1177: pool_page_free_nointr(void *v, unsigned long sz, int mtype)
1.12      thorpej  1178: {
                   1179:
                   1180:        uvm_km_free_poolpage1(kernel_map, (vaddr_t)v);
                   1181: }
                   1182:
1.3       pk       1183:
                   1184: /*
                   1185:  * Release all complete pages that have not been used recently.
                   1186:  */
                   1187: void
1.42    ! thorpej  1188: _pool_reclaim(struct pool *pp, const char *file, long line)
1.3       pk       1189: {
                   1190:        struct pool_item_header *ph, *phnext;
1.21      thorpej  1191:        struct timeval curtime;
                   1192:        int s;
1.3       pk       1193:
1.20      thorpej  1194:        if (pp->pr_roflags & PR_STATIC)
1.3       pk       1195:                return;
                   1196:
1.21      thorpej  1197:        if (simple_lock_try(&pp->pr_slock) == 0)
1.3       pk       1198:                return;
1.25      thorpej  1199:        pr_enter(pp, file, line);
1.3       pk       1200:
1.21      thorpej  1201:        s = splclock();
                   1202:        curtime = mono_time;
                   1203:        splx(s);
                   1204:
1.3       pk       1205:        for (ph = TAILQ_FIRST(&pp->pr_pagelist); ph != NULL; ph = phnext) {
                   1206:                phnext = TAILQ_NEXT(ph, ph_pagelist);
                   1207:
                   1208:                /* Check our minimum page claim */
                   1209:                if (pp->pr_npages <= pp->pr_minpages)
                   1210:                        break;
                   1211:
                   1212:                if (ph->ph_nmissing == 0) {
                   1213:                        struct timeval diff;
                   1214:                        timersub(&curtime, &ph->ph_time, &diff);
                   1215:                        if (diff.tv_sec < pool_inactive_time)
                   1216:                                continue;
1.21      thorpej  1217:
                   1218:                        /*
                   1219:                         * If freeing this page would put us below
                   1220:                         * the low water mark, stop now.
                   1221:                         */
                   1222:                        if ((pp->pr_nitems - pp->pr_itemsperpage) <
                   1223:                            pp->pr_minitems)
                   1224:                                break;
                   1225:
1.3       pk       1226:                        pr_rmpage(pp, ph);
                   1227:                }
                   1228:        }
                   1229:
1.25      thorpej  1230:        pr_leave(pp);
1.21      thorpej  1231:        simple_unlock(&pp->pr_slock);
1.3       pk       1232: }
                   1233:
                   1234:
                   1235: /*
                   1236:  * Drain pools, one at a time.
1.21      thorpej  1237:  *
                   1238:  * Note, we must never be called from an interrupt context.
1.3       pk       1239:  */
                   1240: void
1.42    ! thorpej  1241: pool_drain(void *arg)
1.3       pk       1242: {
                   1243:        struct pool *pp;
1.23      thorpej  1244:        int s;
1.3       pk       1245:
1.23      thorpej  1246:        s = splimp();
                   1247:        simple_lock(&pool_head_slock);
                   1248:
                   1249:        if (drainpp == NULL && (drainpp = TAILQ_FIRST(&pool_head)) == NULL)
                   1250:                goto out;
1.3       pk       1251:
                   1252:        pp = drainpp;
                   1253:        drainpp = TAILQ_NEXT(pp, pr_poollist);
                   1254:
                   1255:        pool_reclaim(pp);
1.23      thorpej  1256:
                   1257:  out:
                   1258:        simple_unlock(&pool_head_slock);
1.3       pk       1259:        splx(s);
                   1260: }
                   1261:
                   1262:
                   1263: /*
                   1264:  * Diagnostic helpers.
                   1265:  */
                   1266: void
1.42    ! thorpej  1267: pool_print(struct pool *pp, const char *modif)
1.21      thorpej  1268: {
                   1269:        int s;
                   1270:
                   1271:        s = splimp();
1.25      thorpej  1272:        if (simple_lock_try(&pp->pr_slock) == 0) {
                   1273:                printf("pool %s is locked; try again later\n",
                   1274:                    pp->pr_wchan);
                   1275:                splx(s);
                   1276:                return;
                   1277:        }
                   1278:        pool_print1(pp, modif, printf);
1.21      thorpej  1279:        simple_unlock(&pp->pr_slock);
                   1280:        splx(s);
                   1281: }
                   1282:
1.25      thorpej  1283: void
1.42    ! thorpej  1284: pool_printit(struct pool *pp, const char *modif, void (*pr)(const char *, ...))
1.25      thorpej  1285: {
                   1286:        int didlock = 0;
                   1287:
                   1288:        if (pp == NULL) {
                   1289:                (*pr)("Must specify a pool to print.\n");
                   1290:                return;
                   1291:        }
                   1292:
                   1293:        /*
                   1294:         * Called from DDB; interrupts should be blocked, and all
                   1295:         * other processors should be paused.  We can skip locking
                   1296:         * the pool in this case.
                   1297:         *
                   1298:         * We do a simple_lock_try() just to print the lock
                   1299:         * status, however.
                   1300:         */
                   1301:
                   1302:        if (simple_lock_try(&pp->pr_slock) == 0)
                   1303:                (*pr)("WARNING: pool %s is locked\n", pp->pr_wchan);
                   1304:        else
                   1305:                didlock = 1;
                   1306:
                   1307:        pool_print1(pp, modif, pr);
                   1308:
                   1309:        if (didlock)
                   1310:                simple_unlock(&pp->pr_slock);
                   1311: }
                   1312:
1.21      thorpej  1313: static void
1.42    ! thorpej  1314: pool_print1(struct pool *pp, const char *modif, void (*pr)(const char *, ...))
1.3       pk       1315: {
1.25      thorpej  1316:        struct pool_item_header *ph;
                   1317: #ifdef DIAGNOSTIC
                   1318:        struct pool_item *pi;
                   1319: #endif
                   1320:        int print_log = 0, print_pagelist = 0;
                   1321:        char c;
                   1322:
                   1323:        while ((c = *modif++) != '\0') {
                   1324:                if (c == 'l')
                   1325:                        print_log = 1;
                   1326:                if (c == 'p')
                   1327:                        print_pagelist = 1;
                   1328:                modif++;
                   1329:        }
                   1330:
                   1331:        (*pr)("POOL %s: size %u, align %u, ioff %u, roflags 0x%08x\n",
                   1332:            pp->pr_wchan, pp->pr_size, pp->pr_align, pp->pr_itemoffset,
                   1333:            pp->pr_roflags);
                   1334:        (*pr)("\tpagesz %u, mtype %d\n", pp->pr_pagesz, pp->pr_mtype);
                   1335:        (*pr)("\talloc %p, release %p\n", pp->pr_alloc, pp->pr_free);
                   1336:        (*pr)("\tminitems %u, minpages %u, maxpages %u, npages %u\n",
                   1337:            pp->pr_minitems, pp->pr_minpages, pp->pr_maxpages, pp->pr_npages);
                   1338:        (*pr)("\titemsperpage %u, nitems %u, nout %u, hardlimit %u\n",
                   1339:            pp->pr_itemsperpage, pp->pr_nitems, pp->pr_nout, pp->pr_hardlimit);
                   1340:
                   1341:        (*pr)("\n\tnget %lu, nfail %lu, nput %lu\n",
                   1342:            pp->pr_nget, pp->pr_nfail, pp->pr_nput);
                   1343:        (*pr)("\tnpagealloc %lu, npagefree %lu, hiwat %u, nidle %lu\n",
                   1344:            pp->pr_npagealloc, pp->pr_npagefree, pp->pr_hiwat, pp->pr_nidle);
                   1345:
                   1346:        if (print_pagelist == 0)
                   1347:                goto skip_pagelist;
                   1348:
                   1349:        if ((ph = TAILQ_FIRST(&pp->pr_pagelist)) != NULL)
                   1350:                (*pr)("\n\tpage list:\n");
                   1351:        for (; ph != NULL; ph = TAILQ_NEXT(ph, ph_pagelist)) {
                   1352:                (*pr)("\t\tpage %p, nmissing %d, time %lu,%lu\n",
                   1353:                    ph->ph_page, ph->ph_nmissing,
                   1354:                    (u_long)ph->ph_time.tv_sec,
                   1355:                    (u_long)ph->ph_time.tv_usec);
                   1356: #ifdef DIAGNOSTIC
                   1357:                for (pi = TAILQ_FIRST(&ph->ph_itemlist); pi != NULL;
                   1358:                     pi = TAILQ_NEXT(pi, pi_list)) {
                   1359:                        if (pi->pi_magic != PI_MAGIC) {
                   1360:                                (*pr)("\t\t\titem %p, magic 0x%x\n",
                   1361:                                    pi, pi->pi_magic);
                   1362:                        }
                   1363:                }
                   1364: #endif
                   1365:        }
                   1366:        if (pp->pr_curpage == NULL)
                   1367:                (*pr)("\tno current page\n");
                   1368:        else
                   1369:                (*pr)("\tcurpage %p\n", pp->pr_curpage->ph_page);
                   1370:
                   1371:  skip_pagelist:
                   1372:
                   1373:        if (print_log == 0)
                   1374:                goto skip_log;
                   1375:
                   1376:        (*pr)("\n");
                   1377:        if ((pp->pr_roflags & PR_LOGGING) == 0)
                   1378:                (*pr)("\tno log\n");
                   1379:        else
                   1380:                pr_printlog(pp, NULL, pr);
1.3       pk       1381:
1.25      thorpej  1382:  skip_log:
1.3       pk       1383:
1.25      thorpej  1384:        pr_enter_check(pp, pr);
1.3       pk       1385: }
                   1386:
                   1387: int
1.42    ! thorpej  1388: pool_chk(struct pool *pp, const char *label)
1.3       pk       1389: {
                   1390:        struct pool_item_header *ph;
                   1391:        int r = 0;
                   1392:
1.21      thorpej  1393:        simple_lock(&pp->pr_slock);
1.3       pk       1394:
                   1395:        for (ph = TAILQ_FIRST(&pp->pr_pagelist); ph != NULL;
                   1396:             ph = TAILQ_NEXT(ph, ph_pagelist)) {
                   1397:
                   1398:                struct pool_item *pi;
                   1399:                int n;
                   1400:                caddr_t page;
                   1401:
                   1402:                page = (caddr_t)((u_long)ph & pp->pr_pagemask);
1.20      thorpej  1403:                if (page != ph->ph_page &&
                   1404:                    (pp->pr_roflags & PR_PHINPAGE) != 0) {
1.3       pk       1405:                        if (label != NULL)
                   1406:                                printf("%s: ", label);
1.16      briggs   1407:                        printf("pool(%p:%s): page inconsistency: page %p;"
                   1408:                               " at page head addr %p (p %p)\n", pp,
1.3       pk       1409:                                pp->pr_wchan, ph->ph_page,
                   1410:                                ph, page);
                   1411:                        r++;
                   1412:                        goto out;
                   1413:                }
                   1414:
                   1415:                for (pi = TAILQ_FIRST(&ph->ph_itemlist), n = 0;
                   1416:                     pi != NULL;
                   1417:                     pi = TAILQ_NEXT(pi,pi_list), n++) {
                   1418:
                   1419: #ifdef DIAGNOSTIC
                   1420:                        if (pi->pi_magic != PI_MAGIC) {
                   1421:                                if (label != NULL)
                   1422:                                        printf("%s: ", label);
                   1423:                                printf("pool(%s): free list modified: magic=%x;"
                   1424:                                       " page %p; item ordinal %d;"
                   1425:                                       " addr %p (p %p)\n",
                   1426:                                        pp->pr_wchan, pi->pi_magic, ph->ph_page,
                   1427:                                        n, pi, page);
                   1428:                                panic("pool");
                   1429:                        }
                   1430: #endif
                   1431:                        page = (caddr_t)((u_long)pi & pp->pr_pagemask);
                   1432:                        if (page == ph->ph_page)
                   1433:                                continue;
                   1434:
                   1435:                        if (label != NULL)
                   1436:                                printf("%s: ", label);
1.16      briggs   1437:                        printf("pool(%p:%s): page inconsistency: page %p;"
                   1438:                               " item ordinal %d; addr %p (p %p)\n", pp,
1.3       pk       1439:                                pp->pr_wchan, ph->ph_page,
                   1440:                                n, pi, page);
                   1441:                        r++;
                   1442:                        goto out;
                   1443:                }
                   1444:        }
                   1445: out:
1.21      thorpej  1446:        simple_unlock(&pp->pr_slock);
1.3       pk       1447:        return (r);
                   1448: }

CVSweb <webmaster@jp.NetBSD.org>