[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 1.60.4.1

1.60.4.1! fvdl        1: /*     $NetBSD: subr_pool.c,v 1.61 2001/09/26 07:14:56 chs Exp $       */
1.1       pk          2:
                      3: /*-
1.43      thorpej     4:  * Copyright (c) 1997, 1999, 2000 The NetBSD Foundation, Inc.
1.1       pk          5:  * All rights reserved.
                      6:  *
                      7:  * This code is derived from software contributed to The NetBSD Foundation
1.20      thorpej     8:  * by Paul Kranenburg; by Jason R. Thorpe of the Numerical Aerospace
                      9:  * Simulation Facility, NASA Ames Research Center.
1.1       pk         10:  *
                     11:  * Redistribution and use in source and binary forms, with or without
                     12:  * modification, are permitted provided that the following conditions
                     13:  * are met:
                     14:  * 1. Redistributions of source code must retain the above copyright
                     15:  *    notice, this list of conditions and the following disclaimer.
                     16:  * 2. Redistributions in binary form must reproduce the above copyright
                     17:  *    notice, this list of conditions and the following disclaimer in the
                     18:  *    documentation and/or other materials provided with the distribution.
                     19:  * 3. All advertising materials mentioning features or use of this software
                     20:  *    must display the following acknowledgement:
1.13      christos   21:  *     This product includes software developed by the NetBSD
                     22:  *     Foundation, Inc. and its contributors.
1.1       pk         23:  * 4. Neither the name of The NetBSD Foundation nor the names of its
                     24:  *    contributors may be used to endorse or promote products derived
                     25:  *    from this software without specific prior written permission.
                     26:  *
                     27:  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
                     28:  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
                     29:  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
                     30:  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
                     31:  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
                     32:  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
                     33:  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
                     34:  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
                     35:  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
                     36:  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
                     37:  * POSSIBILITY OF SUCH DAMAGE.
                     38:  */
1.24      scottr     39:
1.25      thorpej    40: #include "opt_pool.h"
1.24      scottr     41: #include "opt_poollog.h"
1.28      thorpej    42: #include "opt_lockdebug.h"
1.1       pk         43:
                     44: #include <sys/param.h>
                     45: #include <sys/systm.h>
                     46: #include <sys/proc.h>
                     47: #include <sys/errno.h>
                     48: #include <sys/kernel.h>
                     49: #include <sys/malloc.h>
                     50: #include <sys/lock.h>
                     51: #include <sys/pool.h>
1.20      thorpej    52: #include <sys/syslog.h>
1.3       pk         53:
                     54: #include <uvm/uvm.h>
                     55:
1.1       pk         56: /*
                     57:  * Pool resource management utility.
1.3       pk         58:  *
                     59:  * Memory is allocated in pages which are split into pieces according
                     60:  * to the pool item size. Each page is kept on a list headed by `pr_pagelist'
                     61:  * in the pool structure and the individual pool items are on a linked list
                     62:  * headed by `ph_itemlist' in each page header. The memory for building
                     63:  * the page list is either taken from the allocated pages themselves (for
                     64:  * small pool items) or taken from an internal pool of page headers (`phpool').
1.1       pk         65:  */
                     66:
1.3       pk         67: /* List of all pools */
1.5       thorpej    68: TAILQ_HEAD(,pool) pool_head = TAILQ_HEAD_INITIALIZER(pool_head);
1.3       pk         69:
                     70: /* Private pool for page header structures */
                     71: static struct pool phpool;
                     72:
                     73: /* # of seconds to retain page after last use */
                     74: int pool_inactive_time = 10;
                     75:
                     76: /* Next candidate for drainage (see pool_drain()) */
1.23      thorpej    77: static struct pool     *drainpp;
                     78:
                     79: /* This spin lock protects both pool_head and drainpp. */
                     80: struct simplelock pool_head_slock = SIMPLELOCK_INITIALIZER;
1.3       pk         81:
                     82: struct pool_item_header {
                     83:        /* Page headers */
                     84:        TAILQ_ENTRY(pool_item_header)
                     85:                                ph_pagelist;    /* pool page list */
                     86:        TAILQ_HEAD(,pool_item)  ph_itemlist;    /* chunk list for this page */
                     87:        LIST_ENTRY(pool_item_header)
                     88:                                ph_hashlist;    /* Off-page page headers */
                     89:        int                     ph_nmissing;    /* # of chunks in use */
                     90:        caddr_t                 ph_page;        /* this page's address */
                     91:        struct timeval          ph_time;        /* last referenced */
                     92: };
1.60.4.1! fvdl       93: TAILQ_HEAD(pool_pagelist,pool_item_header);
1.3       pk         94:
1.1       pk         95: struct pool_item {
1.3       pk         96: #ifdef DIAGNOSTIC
                     97:        int pi_magic;
1.33      chs        98: #endif
1.25      thorpej    99: #define        PI_MAGIC 0xdeadbeef
1.3       pk        100:        /* Other entries use only this list entry */
                    101:        TAILQ_ENTRY(pool_item)  pi_list;
                    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:
1.53      thorpej   107: #define        POOL_NEEDS_CATCHUP(pp)                                          \
                    108:        ((pp)->pr_nitems < (pp)->pr_minitems)
                    109:
1.43      thorpej   110: /*
                    111:  * Pool cache management.
                    112:  *
                    113:  * Pool caches provide a way for constructed objects to be cached by the
                    114:  * pool subsystem.  This can lead to performance improvements by avoiding
                    115:  * needless object construction/destruction; it is deferred until absolutely
                    116:  * necessary.
                    117:  *
                    118:  * Caches are grouped into cache groups.  Each cache group references
                    119:  * up to 16 constructed objects.  When a cache allocates an object
                    120:  * from the pool, it calls the object's constructor and places it into
                    121:  * a cache group.  When a cache group frees an object back to the pool,
                    122:  * it first calls the object's destructor.  This allows the object to
                    123:  * persist in constructed form while freed to the cache.
                    124:  *
                    125:  * Multiple caches may exist for each pool.  This allows a single
                    126:  * object type to have multiple constructed forms.  The pool references
                    127:  * each cache, so that when a pool is drained by the pagedaemon, it can
                    128:  * drain each individual cache as well.  Each time a cache is drained,
                    129:  * the most idle cache group is freed to the pool in its entirety.
                    130:  *
                    131:  * Pool caches are layed on top of pools.  By layering them, we can avoid
                    132:  * the complexity of cache management for pools which would not benefit
                    133:  * from it.
                    134:  */
                    135:
                    136: /* The cache group pool. */
                    137: static struct pool pcgpool;
                    138:
                    139: /* The pool cache group. */
                    140: #define        PCG_NOBJECTS            16
                    141: struct pool_cache_group {
                    142:        TAILQ_ENTRY(pool_cache_group)
                    143:                pcg_list;       /* link in the pool cache's group list */
                    144:        u_int   pcg_avail;      /* # available objects */
                    145:                                /* pointers to the objects */
                    146:        void    *pcg_objects[PCG_NOBJECTS];
                    147: };
1.3       pk        148:
1.43      thorpej   149: static void    pool_cache_reclaim(struct pool_cache *);
1.3       pk        150:
1.42      thorpej   151: static int     pool_catchup(struct pool *);
1.55      thorpej   152: static void    pool_prime_page(struct pool *, caddr_t,
                    153:                    struct pool_item_header *);
1.42      thorpej   154: static void    *pool_page_alloc(unsigned long, int, int);
                    155: static void    pool_page_free(void *, unsigned long, int);
1.3       pk        156:
1.42      thorpej   157: static void pool_print1(struct pool *, const char *,
                    158:        void (*)(const char *, ...));
1.3       pk        159:
                    160: /*
1.52      thorpej   161:  * Pool log entry. An array of these is allocated in pool_init().
1.3       pk        162:  */
                    163: struct pool_log {
                    164:        const char      *pl_file;
                    165:        long            pl_line;
                    166:        int             pl_action;
1.25      thorpej   167: #define        PRLOG_GET       1
                    168: #define        PRLOG_PUT       2
1.3       pk        169:        void            *pl_addr;
1.1       pk        170: };
                    171:
1.3       pk        172: /* Number of entries in pool log buffers */
1.17      thorpej   173: #ifndef POOL_LOGSIZE
                    174: #define        POOL_LOGSIZE    10
                    175: #endif
                    176:
                    177: int pool_logsize = POOL_LOGSIZE;
1.1       pk        178:
1.59      thorpej   179: #ifdef POOL_DIAGNOSTIC
1.42      thorpej   180: static __inline void
                    181: pr_log(struct pool *pp, void *v, int action, const char *file, long line)
1.3       pk        182: {
                    183:        int n = pp->pr_curlogentry;
                    184:        struct pool_log *pl;
                    185:
1.20      thorpej   186:        if ((pp->pr_roflags & PR_LOGGING) == 0)
1.3       pk        187:                return;
                    188:
                    189:        /*
                    190:         * Fill in the current entry. Wrap around and overwrite
                    191:         * the oldest entry if necessary.
                    192:         */
                    193:        pl = &pp->pr_log[n];
                    194:        pl->pl_file = file;
                    195:        pl->pl_line = line;
                    196:        pl->pl_action = action;
                    197:        pl->pl_addr = v;
                    198:        if (++n >= pp->pr_logsize)
                    199:                n = 0;
                    200:        pp->pr_curlogentry = n;
                    201: }
                    202:
                    203: static void
1.42      thorpej   204: pr_printlog(struct pool *pp, struct pool_item *pi,
                    205:     void (*pr)(const char *, ...))
1.3       pk        206: {
                    207:        int i = pp->pr_logsize;
                    208:        int n = pp->pr_curlogentry;
                    209:
1.20      thorpej   210:        if ((pp->pr_roflags & PR_LOGGING) == 0)
1.3       pk        211:                return;
                    212:
                    213:        /*
                    214:         * Print all entries in this pool's log.
                    215:         */
                    216:        while (i-- > 0) {
                    217:                struct pool_log *pl = &pp->pr_log[n];
                    218:                if (pl->pl_action != 0) {
1.25      thorpej   219:                        if (pi == NULL || pi == pl->pl_addr) {
                    220:                                (*pr)("\tlog entry %d:\n", i);
                    221:                                (*pr)("\t\taction = %s, addr = %p\n",
                    222:                                    pl->pl_action == PRLOG_GET ? "get" : "put",
                    223:                                    pl->pl_addr);
                    224:                                (*pr)("\t\tfile: %s at line %lu\n",
                    225:                                    pl->pl_file, pl->pl_line);
                    226:                        }
1.3       pk        227:                }
                    228:                if (++n >= pp->pr_logsize)
                    229:                        n = 0;
                    230:        }
                    231: }
1.25      thorpej   232:
1.42      thorpej   233: static __inline void
                    234: pr_enter(struct pool *pp, const char *file, long line)
1.25      thorpej   235: {
                    236:
1.34      thorpej   237:        if (__predict_false(pp->pr_entered_file != NULL)) {
1.25      thorpej   238:                printf("pool %s: reentrancy at file %s line %ld\n",
                    239:                    pp->pr_wchan, file, line);
                    240:                printf("         previous entry at file %s line %ld\n",
                    241:                    pp->pr_entered_file, pp->pr_entered_line);
                    242:                panic("pr_enter");
                    243:        }
                    244:
                    245:        pp->pr_entered_file = file;
                    246:        pp->pr_entered_line = line;
                    247: }
                    248:
1.42      thorpej   249: static __inline void
                    250: pr_leave(struct pool *pp)
1.25      thorpej   251: {
                    252:
1.34      thorpej   253:        if (__predict_false(pp->pr_entered_file == NULL)) {
1.25      thorpej   254:                printf("pool %s not entered?\n", pp->pr_wchan);
                    255:                panic("pr_leave");
                    256:        }
                    257:
                    258:        pp->pr_entered_file = NULL;
                    259:        pp->pr_entered_line = 0;
                    260: }
                    261:
1.42      thorpej   262: static __inline void
                    263: pr_enter_check(struct pool *pp, void (*pr)(const char *, ...))
1.25      thorpej   264: {
                    265:
                    266:        if (pp->pr_entered_file != NULL)
                    267:                (*pr)("\n\tcurrently entered from file %s line %ld\n",
                    268:                    pp->pr_entered_file, pp->pr_entered_line);
                    269: }
1.3       pk        270: #else
1.25      thorpej   271: #define        pr_log(pp, v, action, file, line)
                    272: #define        pr_printlog(pp, pi, pr)
                    273: #define        pr_enter(pp, file, line)
                    274: #define        pr_leave(pp)
                    275: #define        pr_enter_check(pp, pr)
1.59      thorpej   276: #endif /* POOL_DIAGNOSTIC */
1.3       pk        277:
                    278: /*
                    279:  * Return the pool page header based on page address.
                    280:  */
1.42      thorpej   281: static __inline struct pool_item_header *
                    282: pr_find_pagehead(struct pool *pp, caddr_t page)
1.3       pk        283: {
                    284:        struct pool_item_header *ph;
                    285:
1.20      thorpej   286:        if ((pp->pr_roflags & PR_PHINPAGE) != 0)
1.3       pk        287:                return ((struct pool_item_header *)(page + pp->pr_phoffset));
                    288:
                    289:        for (ph = LIST_FIRST(&pp->pr_hashtab[PR_HASH_INDEX(pp, page)]);
                    290:             ph != NULL;
                    291:             ph = LIST_NEXT(ph, ph_hashlist)) {
                    292:                if (ph->ph_page == page)
                    293:                        return (ph);
                    294:        }
                    295:        return (NULL);
                    296: }
                    297:
                    298: /*
                    299:  * Remove a page from the pool.
                    300:  */
1.42      thorpej   301: static __inline void
1.60.4.1! fvdl      302: pr_rmpage(struct pool *pp, struct pool_item_header *ph,
        !           303:      struct pool_pagelist *pq)
1.3       pk        304: {
1.60.4.1! fvdl      305:        int s;
1.3       pk        306:
                    307:        /*
1.7       thorpej   308:         * If the page was idle, decrement the idle page count.
1.3       pk        309:         */
1.6       thorpej   310:        if (ph->ph_nmissing == 0) {
                    311: #ifdef DIAGNOSTIC
                    312:                if (pp->pr_nidle == 0)
                    313:                        panic("pr_rmpage: nidle inconsistent");
1.20      thorpej   314:                if (pp->pr_nitems < pp->pr_itemsperpage)
                    315:                        panic("pr_rmpage: nitems inconsistent");
1.6       thorpej   316: #endif
                    317:                pp->pr_nidle--;
                    318:        }
1.7       thorpej   319:
1.20      thorpej   320:        pp->pr_nitems -= pp->pr_itemsperpage;
                    321:
1.7       thorpej   322:        /*
1.60.4.1! fvdl      323:         * Unlink a page from the pool and release it (or queue it for release).
1.7       thorpej   324:         */
                    325:        TAILQ_REMOVE(&pp->pr_pagelist, ph, ph_pagelist);
1.60.4.1! fvdl      326:        if (pq) {
        !           327:                TAILQ_INSERT_HEAD(pq, ph, ph_pagelist);
        !           328:        } else {
        !           329:                (*pp->pr_free)(ph->ph_page, pp->pr_pagesz, pp->pr_mtype);
        !           330:                if ((pp->pr_roflags & PR_PHINPAGE) == 0) {
        !           331:                        LIST_REMOVE(ph, ph_hashlist);
        !           332:                        s = splhigh();
        !           333:                        pool_put(&phpool, ph);
        !           334:                        splx(s);
        !           335:                }
        !           336:        }
1.7       thorpej   337:        pp->pr_npages--;
                    338:        pp->pr_npagefree++;
1.6       thorpej   339:
1.3       pk        340:        if (pp->pr_curpage == ph) {
                    341:                /*
                    342:                 * Find a new non-empty page header, if any.
                    343:                 * Start search from the page head, to increase the
                    344:                 * chance for "high water" pages to be freed.
                    345:                 */
1.60.4.1! fvdl      346:                TAILQ_FOREACH(ph, &pp->pr_pagelist, ph_pagelist)
1.3       pk        347:                        if (TAILQ_FIRST(&ph->ph_itemlist) != NULL)
                    348:                                break;
                    349:
                    350:                pp->pr_curpage = ph;
1.21      thorpej   351:        }
1.3       pk        352: }
                    353:
                    354: /*
                    355:  * Initialize the given pool resource structure.
                    356:  *
                    357:  * We export this routine to allow other kernel parts to declare
                    358:  * static pools that must be initialized before malloc() is available.
                    359:  */
                    360: void
1.42      thorpej   361: pool_init(struct pool *pp, size_t size, u_int align, u_int ioff, int flags,
                    362:     const char *wchan, size_t pagesz,
                    363:     void *(*alloc)(unsigned long, int, int),
                    364:     void (*release)(void *, unsigned long, int),
                    365:     int mtype)
1.3       pk        366: {
1.16      briggs    367:        int off, slack, i;
1.3       pk        368:
1.25      thorpej   369: #ifdef POOL_DIAGNOSTIC
                    370:        /*
                    371:         * Always log if POOL_DIAGNOSTIC is defined.
                    372:         */
                    373:        if (pool_logsize != 0)
                    374:                flags |= PR_LOGGING;
                    375: #endif
                    376:
1.3       pk        377:        /*
                    378:         * Check arguments and construct default values.
                    379:         */
1.36      pk        380:        if (!powerof2(pagesz))
1.3       pk        381:                panic("pool_init: page size invalid (%lx)\n", (u_long)pagesz);
                    382:
1.4       thorpej   383:        if (alloc == NULL && release == NULL) {
1.3       pk        384:                alloc = pool_page_alloc;
                    385:                release = pool_page_free;
1.4       thorpej   386:                pagesz = PAGE_SIZE;     /* Rounds to PAGE_SIZE anyhow. */
                    387:        } else if ((alloc != NULL && release != NULL) == 0) {
                    388:                /* If you specifiy one, must specify both. */
                    389:                panic("pool_init: must specify alloc and release together");
                    390:        }
                    391:
1.3       pk        392:        if (pagesz == 0)
                    393:                pagesz = PAGE_SIZE;
                    394:
                    395:        if (align == 0)
                    396:                align = ALIGN(1);
1.14      thorpej   397:
                    398:        if (size < sizeof(struct pool_item))
                    399:                size = sizeof(struct pool_item);
1.3       pk        400:
1.35      pk        401:        size = ALIGN(size);
1.43      thorpej   402:        if (size > pagesz)
1.35      pk        403:                panic("pool_init: pool item size (%lu) too large",
                    404:                      (u_long)size);
                    405:
1.3       pk        406:        /*
                    407:         * Initialize the pool structure.
                    408:         */
                    409:        TAILQ_INIT(&pp->pr_pagelist);
1.43      thorpej   410:        TAILQ_INIT(&pp->pr_cachelist);
1.3       pk        411:        pp->pr_curpage = NULL;
                    412:        pp->pr_npages = 0;
                    413:        pp->pr_minitems = 0;
                    414:        pp->pr_minpages = 0;
                    415:        pp->pr_maxpages = UINT_MAX;
1.20      thorpej   416:        pp->pr_roflags = flags;
                    417:        pp->pr_flags = 0;
1.35      pk        418:        pp->pr_size = size;
1.3       pk        419:        pp->pr_align = align;
                    420:        pp->pr_wchan = wchan;
                    421:        pp->pr_mtype = mtype;
                    422:        pp->pr_alloc = alloc;
                    423:        pp->pr_free = release;
                    424:        pp->pr_pagesz = pagesz;
                    425:        pp->pr_pagemask = ~(pagesz - 1);
                    426:        pp->pr_pageshift = ffs(pagesz) - 1;
1.20      thorpej   427:        pp->pr_nitems = 0;
                    428:        pp->pr_nout = 0;
                    429:        pp->pr_hardlimit = UINT_MAX;
                    430:        pp->pr_hardlimit_warning = NULL;
1.31      thorpej   431:        pp->pr_hardlimit_ratecap.tv_sec = 0;
                    432:        pp->pr_hardlimit_ratecap.tv_usec = 0;
                    433:        pp->pr_hardlimit_warning_last.tv_sec = 0;
                    434:        pp->pr_hardlimit_warning_last.tv_usec = 0;
1.3       pk        435:
                    436:        /*
                    437:         * Decide whether to put the page header off page to avoid
                    438:         * wasting too large a part of the page. Off-page page headers
                    439:         * go on a hash table, so we can match a returned item
                    440:         * with its header based on the page address.
                    441:         * We use 1/16 of the page size as the threshold (XXX: tune)
                    442:         */
                    443:        if (pp->pr_size < pagesz/16) {
                    444:                /* Use the end of the page for the page header */
1.20      thorpej   445:                pp->pr_roflags |= PR_PHINPAGE;
1.3       pk        446:                pp->pr_phoffset = off =
                    447:                        pagesz - ALIGN(sizeof(struct pool_item_header));
1.2       pk        448:        } else {
1.3       pk        449:                /* The page header will be taken from our page header pool */
                    450:                pp->pr_phoffset = 0;
                    451:                off = pagesz;
1.16      briggs    452:                for (i = 0; i < PR_HASHTABSIZE; i++) {
                    453:                        LIST_INIT(&pp->pr_hashtab[i]);
                    454:                }
1.2       pk        455:        }
1.1       pk        456:
1.3       pk        457:        /*
                    458:         * Alignment is to take place at `ioff' within the item. This means
                    459:         * we must reserve up to `align - 1' bytes on the page to allow
                    460:         * appropriate positioning of each item.
                    461:         *
                    462:         * Silently enforce `0 <= ioff < align'.
                    463:         */
                    464:        pp->pr_itemoffset = ioff = ioff % align;
                    465:        pp->pr_itemsperpage = (off - ((align - ioff) % align)) / pp->pr_size;
1.43      thorpej   466:        KASSERT(pp->pr_itemsperpage != 0);
1.3       pk        467:
                    468:        /*
                    469:         * Use the slack between the chunks and the page header
                    470:         * for "cache coloring".
                    471:         */
                    472:        slack = off - pp->pr_itemsperpage * pp->pr_size;
                    473:        pp->pr_maxcolor = (slack / align) * align;
                    474:        pp->pr_curcolor = 0;
                    475:
                    476:        pp->pr_nget = 0;
                    477:        pp->pr_nfail = 0;
                    478:        pp->pr_nput = 0;
                    479:        pp->pr_npagealloc = 0;
                    480:        pp->pr_npagefree = 0;
1.1       pk        481:        pp->pr_hiwat = 0;
1.8       thorpej   482:        pp->pr_nidle = 0;
1.3       pk        483:
1.59      thorpej   484: #ifdef POOL_DIAGNOSTIC
1.25      thorpej   485:        if (flags & PR_LOGGING) {
                    486:                if (kmem_map == NULL ||
                    487:                    (pp->pr_log = malloc(pool_logsize * sizeof(struct pool_log),
                    488:                     M_TEMP, M_NOWAIT)) == NULL)
1.20      thorpej   489:                        pp->pr_roflags &= ~PR_LOGGING;
1.3       pk        490:                pp->pr_curlogentry = 0;
                    491:                pp->pr_logsize = pool_logsize;
                    492:        }
1.59      thorpej   493: #endif
1.25      thorpej   494:
                    495:        pp->pr_entered_file = NULL;
                    496:        pp->pr_entered_line = 0;
1.3       pk        497:
1.21      thorpej   498:        simple_lock_init(&pp->pr_slock);
1.1       pk        499:
1.3       pk        500:        /*
1.43      thorpej   501:         * Initialize private page header pool and cache magazine pool if we
                    502:         * haven't done so yet.
1.23      thorpej   503:         * XXX LOCKING.
1.3       pk        504:         */
                    505:        if (phpool.pr_size == 0) {
                    506:                pool_init(&phpool, sizeof(struct pool_item_header), 0, 0,
1.43      thorpej   507:                    0, "phpool", 0, 0, 0, 0);
                    508:                pool_init(&pcgpool, sizeof(struct pool_cache_group), 0, 0,
                    509:                    0, "pcgpool", 0, 0, 0, 0);
1.1       pk        510:        }
                    511:
1.23      thorpej   512:        /* Insert into the list of all pools. */
                    513:        simple_lock(&pool_head_slock);
                    514:        TAILQ_INSERT_TAIL(&pool_head, pp, pr_poollist);
                    515:        simple_unlock(&pool_head_slock);
1.1       pk        516: }
                    517:
                    518: /*
                    519:  * De-commision a pool resource.
                    520:  */
                    521: void
1.42      thorpej   522: pool_destroy(struct pool *pp)
1.1       pk        523: {
1.3       pk        524:        struct pool_item_header *ph;
1.43      thorpej   525:        struct pool_cache *pc;
                    526:
                    527:        /* Destroy all caches for this pool. */
                    528:        while ((pc = TAILQ_FIRST(&pp->pr_cachelist)) != NULL)
                    529:                pool_cache_destroy(pc);
1.3       pk        530:
                    531: #ifdef DIAGNOSTIC
1.20      thorpej   532:        if (pp->pr_nout != 0) {
1.25      thorpej   533:                pr_printlog(pp, NULL, printf);
1.20      thorpej   534:                panic("pool_destroy: pool busy: still out: %u\n",
                    535:                    pp->pr_nout);
1.3       pk        536:        }
                    537: #endif
1.1       pk        538:
1.3       pk        539:        /* Remove all pages */
1.20      thorpej   540:        if ((pp->pr_roflags & PR_STATIC) == 0)
1.60.4.1! fvdl      541:                while ((ph = TAILQ_FIRST(&pp->pr_pagelist)) != NULL)
        !           542:                        pr_rmpage(pp, ph, NULL);
1.3       pk        543:
                    544:        /* Remove from global pool list */
1.23      thorpej   545:        simple_lock(&pool_head_slock);
1.3       pk        546:        TAILQ_REMOVE(&pool_head, pp, pr_poollist);
1.60.4.1! fvdl      547:        if (drainpp == pp) {
        !           548:                drainpp = NULL;
        !           549:        }
1.23      thorpej   550:        simple_unlock(&pool_head_slock);
1.3       pk        551:
1.59      thorpej   552: #ifdef POOL_DIAGNOSTIC
1.20      thorpej   553:        if ((pp->pr_roflags & PR_LOGGING) != 0)
1.3       pk        554:                free(pp->pr_log, M_TEMP);
1.59      thorpej   555: #endif
1.2       pk        556:
1.20      thorpej   557:        if (pp->pr_roflags & PR_FREEHEADER)
1.3       pk        558:                free(pp, M_POOL);
1.1       pk        559: }
                    560:
1.55      thorpej   561: static __inline struct pool_item_header *
                    562: pool_alloc_item_header(struct pool *pp, caddr_t storage, int flags)
                    563: {
                    564:        struct pool_item_header *ph;
                    565:        int s;
                    566:
                    567:        LOCK_ASSERT(simple_lock_held(&pp->pr_slock) == 0);
                    568:
                    569:        if ((pp->pr_roflags & PR_PHINPAGE) != 0)
                    570:                ph = (struct pool_item_header *) (storage + pp->pr_phoffset);
                    571:        else {
                    572:                s = splhigh();
                    573:                ph = pool_get(&phpool, flags);
                    574:                splx(s);
                    575:        }
                    576:
                    577:        return (ph);
                    578: }
1.1       pk        579:
                    580: /*
1.3       pk        581:  * Grab an item from the pool; must be called at appropriate spl level
1.1       pk        582:  */
1.3       pk        583: void *
1.59      thorpej   584: #ifdef POOL_DIAGNOSTIC
1.42      thorpej   585: _pool_get(struct pool *pp, int flags, const char *file, long line)
1.56      sommerfe  586: #else
                    587: pool_get(struct pool *pp, int flags)
                    588: #endif
1.1       pk        589: {
                    590:        struct pool_item *pi;
1.3       pk        591:        struct pool_item_header *ph;
1.55      thorpej   592:        void *v;
1.1       pk        593:
1.2       pk        594: #ifdef DIAGNOSTIC
1.34      thorpej   595:        if (__predict_false((pp->pr_roflags & PR_STATIC) &&
                    596:                            (flags & PR_MALLOCOK))) {
1.25      thorpej   597:                pr_printlog(pp, NULL, printf);
1.2       pk        598:                panic("pool_get: static");
1.3       pk        599:        }
1.2       pk        600:
1.37      sommerfe  601:        if (__predict_false(curproc == NULL && doing_shutdown == 0 &&
                    602:                            (flags & PR_WAITOK) != 0))
1.3       pk        603:                panic("pool_get: must have NOWAIT");
1.58      thorpej   604:
                    605: #ifdef LOCKDEBUG
                    606:        if (flags & PR_WAITOK)
                    607:                simple_lock_only_held(NULL, "pool_get(PR_WAITOK)");
1.56      sommerfe  608: #endif
1.58      thorpej   609: #endif /* DIAGNOSTIC */
1.1       pk        610:
1.21      thorpej   611:        simple_lock(&pp->pr_slock);
1.25      thorpej   612:        pr_enter(pp, file, line);
1.20      thorpej   613:
                    614:  startover:
                    615:        /*
                    616:         * Check to see if we've reached the hard limit.  If we have,
                    617:         * and we can wait, then wait until an item has been returned to
                    618:         * the pool.
                    619:         */
                    620: #ifdef DIAGNOSTIC
1.34      thorpej   621:        if (__predict_false(pp->pr_nout > pp->pr_hardlimit)) {
1.25      thorpej   622:                pr_leave(pp);
1.21      thorpej   623:                simple_unlock(&pp->pr_slock);
1.20      thorpej   624:                panic("pool_get: %s: crossed hard limit", pp->pr_wchan);
                    625:        }
                    626: #endif
1.34      thorpej   627:        if (__predict_false(pp->pr_nout == pp->pr_hardlimit)) {
1.29      sommerfe  628:                if ((flags & PR_WAITOK) && !(flags & PR_LIMITFAIL)) {
1.20      thorpej   629:                        /*
                    630:                         * XXX: A warning isn't logged in this case.  Should
                    631:                         * it be?
                    632:                         */
                    633:                        pp->pr_flags |= PR_WANTED;
1.25      thorpej   634:                        pr_leave(pp);
1.40      sommerfe  635:                        ltsleep(pp, PSWP, pp->pr_wchan, 0, &pp->pr_slock);
1.25      thorpej   636:                        pr_enter(pp, file, line);
1.20      thorpej   637:                        goto startover;
                    638:                }
1.31      thorpej   639:
                    640:                /*
                    641:                 * Log a message that the hard limit has been hit.
                    642:                 */
                    643:                if (pp->pr_hardlimit_warning != NULL &&
                    644:                    ratecheck(&pp->pr_hardlimit_warning_last,
                    645:                              &pp->pr_hardlimit_ratecap))
                    646:                        log(LOG_ERR, "%s\n", pp->pr_hardlimit_warning);
1.21      thorpej   647:
                    648:                if (flags & PR_URGENT)
                    649:                        panic("pool_get: urgent");
                    650:
                    651:                pp->pr_nfail++;
                    652:
1.25      thorpej   653:                pr_leave(pp);
1.21      thorpej   654:                simple_unlock(&pp->pr_slock);
1.20      thorpej   655:                return (NULL);
                    656:        }
                    657:
1.3       pk        658:        /*
                    659:         * The convention we use is that if `curpage' is not NULL, then
                    660:         * it points at a non-empty bucket. In particular, `curpage'
                    661:         * never points at a page header which has PR_PHINPAGE set and
                    662:         * has no items in its bucket.
                    663:         */
1.20      thorpej   664:        if ((ph = pp->pr_curpage) == NULL) {
                    665: #ifdef DIAGNOSTIC
                    666:                if (pp->pr_nitems != 0) {
1.21      thorpej   667:                        simple_unlock(&pp->pr_slock);
1.20      thorpej   668:                        printf("pool_get: %s: curpage NULL, nitems %u\n",
                    669:                            pp->pr_wchan, pp->pr_nitems);
                    670:                        panic("pool_get: nitems inconsistent\n");
                    671:                }
                    672: #endif
                    673:
1.21      thorpej   674:                /*
                    675:                 * Call the back-end page allocator for more memory.
                    676:                 * Release the pool lock, as the back-end page allocator
                    677:                 * may block.
                    678:                 */
1.25      thorpej   679:                pr_leave(pp);
1.21      thorpej   680:                simple_unlock(&pp->pr_slock);
                    681:                v = (*pp->pr_alloc)(pp->pr_pagesz, flags, pp->pr_mtype);
1.55      thorpej   682:                if (__predict_true(v != NULL))
                    683:                        ph = pool_alloc_item_header(pp, v, flags);
1.21      thorpej   684:                simple_lock(&pp->pr_slock);
1.25      thorpej   685:                pr_enter(pp, file, line);
1.15      pk        686:
1.55      thorpej   687:                if (__predict_false(v == NULL || ph == NULL)) {
                    688:                        if (v != NULL)
                    689:                                (*pp->pr_free)(v, pp->pr_pagesz, pp->pr_mtype);
                    690:
1.21      thorpej   691:                        /*
1.55      thorpej   692:                         * We were unable to allocate a page or item
                    693:                         * header, but we released the lock during
                    694:                         * allocation, so perhaps items were freed
                    695:                         * back to the pool.  Check for this case.
1.21      thorpej   696:                         */
                    697:                        if (pp->pr_curpage != NULL)
                    698:                                goto startover;
1.15      pk        699:
1.3       pk        700:                        if (flags & PR_URGENT)
                    701:                                panic("pool_get: urgent");
1.21      thorpej   702:
1.3       pk        703:                        if ((flags & PR_WAITOK) == 0) {
                    704:                                pp->pr_nfail++;
1.25      thorpej   705:                                pr_leave(pp);
1.21      thorpej   706:                                simple_unlock(&pp->pr_slock);
1.1       pk        707:                                return (NULL);
1.3       pk        708:                        }
                    709:
1.15      pk        710:                        /*
                    711:                         * Wait for items to be returned to this pool.
1.21      thorpej   712:                         *
1.15      pk        713:                         * XXX: we actually want to wait just until
                    714:                         * the page allocator has memory again. Depending
                    715:                         * on this pool's usage, we might get stuck here
                    716:                         * for a long time.
1.20      thorpej   717:                         *
                    718:                         * XXX: maybe we should wake up once a second and
                    719:                         * try again?
1.15      pk        720:                         */
1.1       pk        721:                        pp->pr_flags |= PR_WANTED;
1.25      thorpej   722:                        pr_leave(pp);
1.40      sommerfe  723:                        ltsleep(pp, PSWP, pp->pr_wchan, 0, &pp->pr_slock);
1.25      thorpej   724:                        pr_enter(pp, file, line);
1.20      thorpej   725:                        goto startover;
1.1       pk        726:                }
1.3       pk        727:
1.15      pk        728:                /* We have more memory; add it to the pool */
1.55      thorpej   729:                pool_prime_page(pp, v, ph);
1.15      pk        730:                pp->pr_npagealloc++;
                    731:
1.20      thorpej   732:                /* Start the allocation process over. */
                    733:                goto startover;
1.3       pk        734:        }
                    735:
1.34      thorpej   736:        if (__predict_false((v = pi = TAILQ_FIRST(&ph->ph_itemlist)) == NULL)) {
1.25      thorpej   737:                pr_leave(pp);
1.21      thorpej   738:                simple_unlock(&pp->pr_slock);
1.3       pk        739:                panic("pool_get: %s: page empty", pp->pr_wchan);
1.21      thorpej   740:        }
1.20      thorpej   741: #ifdef DIAGNOSTIC
1.34      thorpej   742:        if (__predict_false(pp->pr_nitems == 0)) {
1.25      thorpej   743:                pr_leave(pp);
1.21      thorpej   744:                simple_unlock(&pp->pr_slock);
1.20      thorpej   745:                printf("pool_get: %s: items on itemlist, nitems %u\n",
                    746:                    pp->pr_wchan, pp->pr_nitems);
                    747:                panic("pool_get: nitems inconsistent\n");
                    748:        }
1.56      sommerfe  749:
1.3       pk        750:        pr_log(pp, v, PRLOG_GET, file, line);
                    751:
1.34      thorpej   752:        if (__predict_false(pi->pi_magic != PI_MAGIC)) {
1.25      thorpej   753:                pr_printlog(pp, pi, printf);
1.3       pk        754:                panic("pool_get(%s): free list modified: magic=%x; page %p;"
                    755:                       " item addr %p\n",
                    756:                        pp->pr_wchan, pi->pi_magic, ph->ph_page, pi);
                    757:        }
                    758: #endif
                    759:
                    760:        /*
                    761:         * Remove from item list.
                    762:         */
                    763:        TAILQ_REMOVE(&ph->ph_itemlist, pi, pi_list);
1.20      thorpej   764:        pp->pr_nitems--;
                    765:        pp->pr_nout++;
1.6       thorpej   766:        if (ph->ph_nmissing == 0) {
                    767: #ifdef DIAGNOSTIC
1.34      thorpej   768:                if (__predict_false(pp->pr_nidle == 0))
1.6       thorpej   769:                        panic("pool_get: nidle inconsistent");
                    770: #endif
                    771:                pp->pr_nidle--;
                    772:        }
1.3       pk        773:        ph->ph_nmissing++;
                    774:        if (TAILQ_FIRST(&ph->ph_itemlist) == NULL) {
1.21      thorpej   775: #ifdef DIAGNOSTIC
1.34      thorpej   776:                if (__predict_false(ph->ph_nmissing != pp->pr_itemsperpage)) {
1.25      thorpej   777:                        pr_leave(pp);
1.21      thorpej   778:                        simple_unlock(&pp->pr_slock);
                    779:                        panic("pool_get: %s: nmissing inconsistent",
                    780:                            pp->pr_wchan);
                    781:                }
                    782: #endif
1.3       pk        783:                /*
                    784:                 * Find a new non-empty page header, if any.
                    785:                 * Start search from the page head, to increase
                    786:                 * the chance for "high water" pages to be freed.
                    787:                 *
1.21      thorpej   788:                 * Migrate empty pages to the end of the list.  This
                    789:                 * will speed the update of curpage as pages become
                    790:                 * idle.  Empty pages intermingled with idle pages
                    791:                 * is no big deal.  As soon as a page becomes un-empty,
                    792:                 * it will move back to the head of the list.
1.3       pk        793:                 */
                    794:                TAILQ_REMOVE(&pp->pr_pagelist, ph, ph_pagelist);
1.21      thorpej   795:                TAILQ_INSERT_TAIL(&pp->pr_pagelist, ph, ph_pagelist);
1.60.4.1! fvdl      796:                TAILQ_FOREACH(ph, &pp->pr_pagelist, ph_pagelist)
1.3       pk        797:                        if (TAILQ_FIRST(&ph->ph_itemlist) != NULL)
                    798:                                break;
                    799:
                    800:                pp->pr_curpage = ph;
1.1       pk        801:        }
1.3       pk        802:
                    803:        pp->pr_nget++;
1.20      thorpej   804:
                    805:        /*
                    806:         * If we have a low water mark and we are now below that low
                    807:         * water mark, add more items to the pool.
                    808:         */
1.53      thorpej   809:        if (POOL_NEEDS_CATCHUP(pp) && pool_catchup(pp) != 0) {
1.20      thorpej   810:                /*
                    811:                 * XXX: Should we log a warning?  Should we set up a timeout
                    812:                 * to try again in a second or so?  The latter could break
                    813:                 * a caller's assumptions about interrupt protection, etc.
                    814:                 */
                    815:        }
                    816:
1.25      thorpej   817:        pr_leave(pp);
1.21      thorpej   818:        simple_unlock(&pp->pr_slock);
1.1       pk        819:        return (v);
                    820: }
                    821:
                    822: /*
1.43      thorpej   823:  * Internal version of pool_put().  Pool is already locked/entered.
1.1       pk        824:  */
1.43      thorpej   825: static void
1.56      sommerfe  826: pool_do_put(struct pool *pp, void *v)
1.1       pk        827: {
                    828:        struct pool_item *pi = v;
1.3       pk        829:        struct pool_item_header *ph;
                    830:        caddr_t page;
1.21      thorpej   831:        int s;
1.3       pk        832:
1.60.4.1! fvdl      833:        LOCK_ASSERT(simple_lock_held(&pp->pr_slock));
        !           834:
1.3       pk        835:        page = (caddr_t)((u_long)v & pp->pr_pagemask);
1.1       pk        836:
1.30      thorpej   837: #ifdef DIAGNOSTIC
1.34      thorpej   838:        if (__predict_false(pp->pr_nout == 0)) {
1.30      thorpej   839:                printf("pool %s: putting with none out\n",
                    840:                    pp->pr_wchan);
                    841:                panic("pool_put");
                    842:        }
                    843: #endif
1.3       pk        844:
1.34      thorpej   845:        if (__predict_false((ph = pr_find_pagehead(pp, page)) == NULL)) {
1.25      thorpej   846:                pr_printlog(pp, NULL, printf);
1.3       pk        847:                panic("pool_put: %s: page header missing", pp->pr_wchan);
                    848:        }
1.28      thorpej   849:
                    850: #ifdef LOCKDEBUG
                    851:        /*
                    852:         * Check if we're freeing a locked simple lock.
                    853:         */
                    854:        simple_lock_freecheck((caddr_t)pi, ((caddr_t)pi) + pp->pr_size);
                    855: #endif
1.3       pk        856:
                    857:        /*
                    858:         * Return to item list.
                    859:         */
1.2       pk        860: #ifdef DIAGNOSTIC
1.3       pk        861:        pi->pi_magic = PI_MAGIC;
                    862: #endif
1.32      chs       863: #ifdef DEBUG
                    864:        {
                    865:                int i, *ip = v;
                    866:
                    867:                for (i = 0; i < pp->pr_size / sizeof(int); i++) {
                    868:                        *ip++ = PI_MAGIC;
                    869:                }
                    870:        }
                    871: #endif
                    872:
1.3       pk        873:        TAILQ_INSERT_HEAD(&ph->ph_itemlist, pi, pi_list);
                    874:        ph->ph_nmissing--;
                    875:        pp->pr_nput++;
1.20      thorpej   876:        pp->pr_nitems++;
                    877:        pp->pr_nout--;
1.3       pk        878:
                    879:        /* Cancel "pool empty" condition if it exists */
                    880:        if (pp->pr_curpage == NULL)
                    881:                pp->pr_curpage = ph;
                    882:
                    883:        if (pp->pr_flags & PR_WANTED) {
                    884:                pp->pr_flags &= ~PR_WANTED;
1.15      pk        885:                if (ph->ph_nmissing == 0)
                    886:                        pp->pr_nidle++;
1.3       pk        887:                wakeup((caddr_t)pp);
                    888:                return;
                    889:        }
                    890:
                    891:        /*
1.21      thorpej   892:         * If this page is now complete, do one of two things:
                    893:         *
                    894:         *      (1) If we have more pages than the page high water
                    895:         *          mark, free the page back to the system.
                    896:         *
                    897:         *      (2) Move it to the end of the page list, so that
                    898:         *          we minimize our chances of fragmenting the
                    899:         *          pool.  Idle pages migrate to the end (along with
                    900:         *          completely empty pages, so that we find un-empty
                    901:         *          pages more quickly when we update curpage) of the
                    902:         *          list so they can be more easily swept up by
                    903:         *          the pagedaemon when pages are scarce.
1.3       pk        904:         */
                    905:        if (ph->ph_nmissing == 0) {
1.6       thorpej   906:                pp->pr_nidle++;
1.3       pk        907:                if (pp->pr_npages > pp->pr_maxpages) {
1.60.4.1! fvdl      908:                        pr_rmpage(pp, ph, NULL);
1.3       pk        909:                } else {
                    910:                        TAILQ_REMOVE(&pp->pr_pagelist, ph, ph_pagelist);
                    911:                        TAILQ_INSERT_TAIL(&pp->pr_pagelist, ph, ph_pagelist);
                    912:
1.21      thorpej   913:                        /*
                    914:                         * Update the timestamp on the page.  A page must
                    915:                         * be idle for some period of time before it can
                    916:                         * be reclaimed by the pagedaemon.  This minimizes
                    917:                         * ping-pong'ing for memory.
                    918:                         */
                    919:                        s = splclock();
                    920:                        ph->ph_time = mono_time;
                    921:                        splx(s);
                    922:
                    923:                        /*
                    924:                         * Update the current page pointer.  Just look for
                    925:                         * the first page with any free items.
                    926:                         *
                    927:                         * XXX: Maybe we want an option to look for the
                    928:                         * page with the fewest available items, to minimize
                    929:                         * fragmentation?
                    930:                         */
1.60.4.1! fvdl      931:                        TAILQ_FOREACH(ph, &pp->pr_pagelist, ph_pagelist)
1.3       pk        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:        }
1.43      thorpej   949: }
                    950:
                    951: /*
                    952:  * Return resource to the pool; must be called at appropriate spl level
                    953:  */
1.59      thorpej   954: #ifdef POOL_DIAGNOSTIC
1.43      thorpej   955: void
                    956: _pool_put(struct pool *pp, void *v, const char *file, long line)
                    957: {
                    958:
                    959:        simple_lock(&pp->pr_slock);
                    960:        pr_enter(pp, file, line);
                    961:
1.56      sommerfe  962:        pr_log(pp, v, PRLOG_PUT, file, line);
                    963:
                    964:        pool_do_put(pp, v);
1.21      thorpej   965:
1.25      thorpej   966:        pr_leave(pp);
1.21      thorpej   967:        simple_unlock(&pp->pr_slock);
1.1       pk        968: }
1.57      sommerfe  969: #undef pool_put
1.59      thorpej   970: #endif /* POOL_DIAGNOSTIC */
1.1       pk        971:
1.56      sommerfe  972: void
                    973: pool_put(struct pool *pp, void *v)
                    974: {
                    975:
                    976:        simple_lock(&pp->pr_slock);
                    977:
                    978:        pool_do_put(pp, v);
                    979:
                    980:        simple_unlock(&pp->pr_slock);
                    981: }
1.57      sommerfe  982:
1.59      thorpej   983: #ifdef POOL_DIAGNOSTIC
1.57      sommerfe  984: #define                pool_put(h, v)  _pool_put((h), (v), __FILE__, __LINE__)
1.56      sommerfe  985: #endif
                    986:
1.1       pk        987: /*
1.55      thorpej   988:  * Add N items to the pool.
                    989:  */
                    990: int
                    991: pool_prime(struct pool *pp, int n)
                    992: {
                    993:        struct pool_item_header *ph;
                    994:        caddr_t cp;
                    995:        int newpages, error = 0;
                    996:
                    997:        simple_lock(&pp->pr_slock);
                    998:
                    999:        newpages = roundup(n, pp->pr_itemsperpage) / pp->pr_itemsperpage;
                   1000:
                   1001:        while (newpages-- > 0) {
                   1002:                simple_unlock(&pp->pr_slock);
                   1003:                cp = (*pp->pr_alloc)(pp->pr_pagesz, PR_NOWAIT, pp->pr_mtype);
                   1004:                if (__predict_true(cp != NULL))
                   1005:                        ph = pool_alloc_item_header(pp, cp, PR_NOWAIT);
                   1006:                simple_lock(&pp->pr_slock);
                   1007:
                   1008:                if (__predict_false(cp == NULL || ph == NULL)) {
                   1009:                        error = ENOMEM;
                   1010:                        if (cp != NULL)
                   1011:                                (*pp->pr_free)(cp, pp->pr_pagesz, pp->pr_mtype);
                   1012:                        break;
                   1013:                }
                   1014:
                   1015:                pool_prime_page(pp, cp, ph);
                   1016:                pp->pr_npagealloc++;
                   1017:                pp->pr_minpages++;
                   1018:        }
                   1019:
                   1020:        if (pp->pr_minpages >= pp->pr_maxpages)
                   1021:                pp->pr_maxpages = pp->pr_minpages + 1;  /* XXX */
                   1022:
                   1023:        simple_unlock(&pp->pr_slock);
                   1024:        return (0);
                   1025: }
                   1026:
                   1027: /*
1.3       pk       1028:  * Add a page worth of items to the pool.
1.21      thorpej  1029:  *
                   1030:  * Note, we must be called with the pool descriptor LOCKED.
1.3       pk       1031:  */
1.55      thorpej  1032: static void
                   1033: pool_prime_page(struct pool *pp, caddr_t storage, struct pool_item_header *ph)
1.3       pk       1034: {
                   1035:        struct pool_item *pi;
                   1036:        caddr_t cp = storage;
                   1037:        unsigned int align = pp->pr_align;
                   1038:        unsigned int ioff = pp->pr_itemoffset;
1.55      thorpej  1039:        int n;
1.36      pk       1040:
                   1041:        if (((u_long)cp & (pp->pr_pagesz - 1)) != 0)
                   1042:                panic("pool_prime_page: %s: unaligned page", pp->pr_wchan);
1.3       pk       1043:
1.55      thorpej  1044:        if ((pp->pr_roflags & PR_PHINPAGE) == 0)
1.3       pk       1045:                LIST_INSERT_HEAD(&pp->pr_hashtab[PR_HASH_INDEX(pp, cp)],
1.55      thorpej  1046:                    ph, ph_hashlist);
1.3       pk       1047:
                   1048:        /*
                   1049:         * Insert page header.
                   1050:         */
                   1051:        TAILQ_INSERT_HEAD(&pp->pr_pagelist, ph, ph_pagelist);
                   1052:        TAILQ_INIT(&ph->ph_itemlist);
                   1053:        ph->ph_page = storage;
                   1054:        ph->ph_nmissing = 0;
1.21      thorpej  1055:        memset(&ph->ph_time, 0, sizeof(ph->ph_time));
1.3       pk       1056:
1.6       thorpej  1057:        pp->pr_nidle++;
                   1058:
1.3       pk       1059:        /*
                   1060:         * Color this page.
                   1061:         */
                   1062:        cp = (caddr_t)(cp + pp->pr_curcolor);
                   1063:        if ((pp->pr_curcolor += align) > pp->pr_maxcolor)
                   1064:                pp->pr_curcolor = 0;
                   1065:
                   1066:        /*
                   1067:         * Adjust storage to apply aligment to `pr_itemoffset' in each item.
                   1068:         */
                   1069:        if (ioff != 0)
                   1070:                cp = (caddr_t)(cp + (align - ioff));
                   1071:
                   1072:        /*
                   1073:         * Insert remaining chunks on the bucket list.
                   1074:         */
                   1075:        n = pp->pr_itemsperpage;
1.20      thorpej  1076:        pp->pr_nitems += n;
1.3       pk       1077:
                   1078:        while (n--) {
                   1079:                pi = (struct pool_item *)cp;
                   1080:
                   1081:                /* Insert on page list */
                   1082:                TAILQ_INSERT_TAIL(&ph->ph_itemlist, pi, pi_list);
                   1083: #ifdef DIAGNOSTIC
                   1084:                pi->pi_magic = PI_MAGIC;
                   1085: #endif
                   1086:                cp = (caddr_t)(cp + pp->pr_size);
                   1087:        }
                   1088:
                   1089:        /*
                   1090:         * If the pool was depleted, point at the new page.
                   1091:         */
                   1092:        if (pp->pr_curpage == NULL)
                   1093:                pp->pr_curpage = ph;
                   1094:
                   1095:        if (++pp->pr_npages > pp->pr_hiwat)
                   1096:                pp->pr_hiwat = pp->pr_npages;
                   1097: }
                   1098:
1.20      thorpej  1099: /*
1.52      thorpej  1100:  * Used by pool_get() when nitems drops below the low water mark.  This
                   1101:  * is used to catch up nitmes with the low water mark.
1.20      thorpej  1102:  *
1.21      thorpej  1103:  * Note 1, we never wait for memory here, we let the caller decide what to do.
1.20      thorpej  1104:  *
                   1105:  * Note 2, this doesn't work with static pools.
                   1106:  *
                   1107:  * Note 3, we must be called with the pool already locked, and we return
                   1108:  * with it locked.
                   1109:  */
                   1110: static int
1.42      thorpej  1111: pool_catchup(struct pool *pp)
1.20      thorpej  1112: {
1.55      thorpej  1113:        struct pool_item_header *ph;
1.20      thorpej  1114:        caddr_t cp;
                   1115:        int error = 0;
                   1116:
                   1117:        if (pp->pr_roflags & PR_STATIC) {
                   1118:                /*
                   1119:                 * We dropped below the low water mark, and this is not a
                   1120:                 * good thing.  Log a warning.
1.21      thorpej  1121:                 *
                   1122:                 * XXX: rate-limit this?
1.20      thorpej  1123:                 */
                   1124:                printf("WARNING: static pool `%s' dropped below low water "
                   1125:                    "mark\n", pp->pr_wchan);
                   1126:                return (0);
                   1127:        }
                   1128:
1.54      thorpej  1129:        while (POOL_NEEDS_CATCHUP(pp)) {
1.20      thorpej  1130:                /*
1.21      thorpej  1131:                 * Call the page back-end allocator for more memory.
                   1132:                 *
                   1133:                 * XXX: We never wait, so should we bother unlocking
                   1134:                 * the pool descriptor?
1.20      thorpej  1135:                 */
1.21      thorpej  1136:                simple_unlock(&pp->pr_slock);
1.55      thorpej  1137:                cp = (*pp->pr_alloc)(pp->pr_pagesz, PR_NOWAIT, pp->pr_mtype);
                   1138:                if (__predict_true(cp != NULL))
                   1139:                        ph = pool_alloc_item_header(pp, cp, PR_NOWAIT);
1.21      thorpej  1140:                simple_lock(&pp->pr_slock);
1.55      thorpej  1141:                if (__predict_false(cp == NULL || ph == NULL)) {
                   1142:                        if (cp != NULL)
                   1143:                                (*pp->pr_free)(cp, pp->pr_pagesz, pp->pr_mtype);
1.20      thorpej  1144:                        error = ENOMEM;
                   1145:                        break;
                   1146:                }
1.55      thorpej  1147:                pool_prime_page(pp, cp, ph);
1.26      thorpej  1148:                pp->pr_npagealloc++;
1.20      thorpej  1149:        }
                   1150:
                   1151:        return (error);
                   1152: }
                   1153:
1.3       pk       1154: void
1.42      thorpej  1155: pool_setlowat(struct pool *pp, int n)
1.3       pk       1156: {
1.20      thorpej  1157:        int error;
1.15      pk       1158:
1.21      thorpej  1159:        simple_lock(&pp->pr_slock);
                   1160:
1.3       pk       1161:        pp->pr_minitems = n;
1.15      pk       1162:        pp->pr_minpages = (n == 0)
                   1163:                ? 0
1.18      thorpej  1164:                : roundup(n, pp->pr_itemsperpage) / pp->pr_itemsperpage;
1.20      thorpej  1165:
                   1166:        /* Make sure we're caught up with the newly-set low water mark. */
1.53      thorpej  1167:        if (POOL_NEEDS_CATCHUP(pp) && (error = pool_catchup(pp) != 0)) {
1.20      thorpej  1168:                /*
                   1169:                 * XXX: Should we log a warning?  Should we set up a timeout
                   1170:                 * to try again in a second or so?  The latter could break
                   1171:                 * a caller's assumptions about interrupt protection, etc.
                   1172:                 */
                   1173:        }
1.21      thorpej  1174:
                   1175:        simple_unlock(&pp->pr_slock);
1.3       pk       1176: }
                   1177:
                   1178: void
1.42      thorpej  1179: pool_sethiwat(struct pool *pp, int n)
1.3       pk       1180: {
1.15      pk       1181:
1.21      thorpej  1182:        simple_lock(&pp->pr_slock);
                   1183:
1.15      pk       1184:        pp->pr_maxpages = (n == 0)
                   1185:                ? 0
1.18      thorpej  1186:                : roundup(n, pp->pr_itemsperpage) / pp->pr_itemsperpage;
1.21      thorpej  1187:
                   1188:        simple_unlock(&pp->pr_slock);
1.3       pk       1189: }
                   1190:
1.20      thorpej  1191: void
1.42      thorpej  1192: pool_sethardlimit(struct pool *pp, int n, const char *warnmess, int ratecap)
1.20      thorpej  1193: {
                   1194:
1.21      thorpej  1195:        simple_lock(&pp->pr_slock);
1.20      thorpej  1196:
                   1197:        pp->pr_hardlimit = n;
                   1198:        pp->pr_hardlimit_warning = warnmess;
1.31      thorpej  1199:        pp->pr_hardlimit_ratecap.tv_sec = ratecap;
                   1200:        pp->pr_hardlimit_warning_last.tv_sec = 0;
                   1201:        pp->pr_hardlimit_warning_last.tv_usec = 0;
1.20      thorpej  1202:
                   1203:        /*
1.21      thorpej  1204:         * In-line version of pool_sethiwat(), because we don't want to
                   1205:         * release the lock.
1.20      thorpej  1206:         */
                   1207:        pp->pr_maxpages = (n == 0)
                   1208:                ? 0
                   1209:                : roundup(n, pp->pr_itemsperpage) / pp->pr_itemsperpage;
1.21      thorpej  1210:
                   1211:        simple_unlock(&pp->pr_slock);
1.20      thorpej  1212: }
1.3       pk       1213:
                   1214: /*
                   1215:  * Default page allocator.
                   1216:  */
                   1217: static void *
1.42      thorpej  1218: pool_page_alloc(unsigned long sz, int flags, int mtype)
1.3       pk       1219: {
1.11      thorpej  1220:        boolean_t waitok = (flags & PR_WAITOK) ? TRUE : FALSE;
1.3       pk       1221:
1.11      thorpej  1222:        return ((void *)uvm_km_alloc_poolpage(waitok));
1.3       pk       1223: }
                   1224:
                   1225: static void
1.42      thorpej  1226: pool_page_free(void *v, unsigned long sz, int mtype)
1.3       pk       1227: {
                   1228:
1.10      eeh      1229:        uvm_km_free_poolpage((vaddr_t)v);
1.3       pk       1230: }
1.12      thorpej  1231:
                   1232: /*
                   1233:  * Alternate pool page allocator for pools that know they will
                   1234:  * never be accessed in interrupt context.
                   1235:  */
                   1236: void *
1.42      thorpej  1237: pool_page_alloc_nointr(unsigned long sz, int flags, int mtype)
1.12      thorpej  1238: {
                   1239:        boolean_t waitok = (flags & PR_WAITOK) ? TRUE : FALSE;
                   1240:
                   1241:        return ((void *)uvm_km_alloc_poolpage1(kernel_map, uvm.kernel_object,
                   1242:            waitok));
                   1243: }
                   1244:
                   1245: void
1.42      thorpej  1246: pool_page_free_nointr(void *v, unsigned long sz, int mtype)
1.12      thorpej  1247: {
                   1248:
                   1249:        uvm_km_free_poolpage1(kernel_map, (vaddr_t)v);
                   1250: }
                   1251:
1.3       pk       1252:
                   1253: /*
                   1254:  * Release all complete pages that have not been used recently.
                   1255:  */
                   1256: void
1.59      thorpej  1257: #ifdef POOL_DIAGNOSTIC
1.42      thorpej  1258: _pool_reclaim(struct pool *pp, const char *file, long line)
1.56      sommerfe 1259: #else
                   1260: pool_reclaim(struct pool *pp)
                   1261: #endif
1.3       pk       1262: {
                   1263:        struct pool_item_header *ph, *phnext;
1.43      thorpej  1264:        struct pool_cache *pc;
1.21      thorpej  1265:        struct timeval curtime;
1.60.4.1! fvdl     1266:        struct pool_pagelist pq;
1.21      thorpej  1267:        int s;
1.3       pk       1268:
1.20      thorpej  1269:        if (pp->pr_roflags & PR_STATIC)
1.3       pk       1270:                return;
                   1271:
1.21      thorpej  1272:        if (simple_lock_try(&pp->pr_slock) == 0)
1.3       pk       1273:                return;
1.25      thorpej  1274:        pr_enter(pp, file, line);
1.60.4.1! fvdl     1275:        TAILQ_INIT(&pq);
1.3       pk       1276:
1.43      thorpej  1277:        /*
                   1278:         * Reclaim items from the pool's caches.
                   1279:         */
1.60.4.1! fvdl     1280:        TAILQ_FOREACH(pc, &pp->pr_cachelist, pc_poollist)
1.43      thorpej  1281:                pool_cache_reclaim(pc);
                   1282:
1.21      thorpej  1283:        s = splclock();
                   1284:        curtime = mono_time;
                   1285:        splx(s);
                   1286:
1.3       pk       1287:        for (ph = TAILQ_FIRST(&pp->pr_pagelist); ph != NULL; ph = phnext) {
                   1288:                phnext = TAILQ_NEXT(ph, ph_pagelist);
                   1289:
                   1290:                /* Check our minimum page claim */
                   1291:                if (pp->pr_npages <= pp->pr_minpages)
                   1292:                        break;
                   1293:
                   1294:                if (ph->ph_nmissing == 0) {
                   1295:                        struct timeval diff;
                   1296:                        timersub(&curtime, &ph->ph_time, &diff);
                   1297:                        if (diff.tv_sec < pool_inactive_time)
                   1298:                                continue;
1.21      thorpej  1299:
                   1300:                        /*
                   1301:                         * If freeing this page would put us below
                   1302:                         * the low water mark, stop now.
                   1303:                         */
                   1304:                        if ((pp->pr_nitems - pp->pr_itemsperpage) <
                   1305:                            pp->pr_minitems)
                   1306:                                break;
                   1307:
1.60.4.1! fvdl     1308:                        pr_rmpage(pp, ph, &pq);
1.3       pk       1309:                }
                   1310:        }
                   1311:
1.25      thorpej  1312:        pr_leave(pp);
1.21      thorpej  1313:        simple_unlock(&pp->pr_slock);
1.60.4.1! fvdl     1314:        if (TAILQ_EMPTY(&pq)) {
        !          1315:                return;
        !          1316:        }
        !          1317:        while ((ph = TAILQ_FIRST(&pq)) != NULL) {
        !          1318:                TAILQ_REMOVE(&pq, ph, ph_pagelist);
        !          1319:                (*pp->pr_free)(ph->ph_page, pp->pr_pagesz, pp->pr_mtype);
        !          1320:                if (pp->pr_roflags & PR_PHINPAGE) {
        !          1321:                        continue;
        !          1322:                }
        !          1323:                LIST_REMOVE(ph, ph_hashlist);
        !          1324:                s = splhigh();
        !          1325:                pool_put(&phpool, ph);
        !          1326:                splx(s);
        !          1327:        }
1.3       pk       1328: }
                   1329:
                   1330:
                   1331: /*
                   1332:  * Drain pools, one at a time.
1.21      thorpej  1333:  *
                   1334:  * Note, we must never be called from an interrupt context.
1.3       pk       1335:  */
                   1336: void
1.42      thorpej  1337: pool_drain(void *arg)
1.3       pk       1338: {
                   1339:        struct pool *pp;
1.23      thorpej  1340:        int s;
1.3       pk       1341:
1.60.4.1! fvdl     1342:        pp = NULL;
1.49      thorpej  1343:        s = splvm();
1.23      thorpej  1344:        simple_lock(&pool_head_slock);
1.60.4.1! fvdl     1345:        if (drainpp == NULL) {
        !          1346:                drainpp = TAILQ_FIRST(&pool_head);
        !          1347:        }
        !          1348:        if (drainpp) {
        !          1349:                pp = drainpp;
        !          1350:                drainpp = TAILQ_NEXT(pp, pr_poollist);
        !          1351:        }
1.23      thorpej  1352:        simple_unlock(&pool_head_slock);
1.3       pk       1353:        splx(s);
1.60.4.1! fvdl     1354:
        !          1355:        pool_reclaim(pp);
1.3       pk       1356: }
                   1357:
                   1358:
                   1359: /*
                   1360:  * Diagnostic helpers.
                   1361:  */
                   1362: void
1.42      thorpej  1363: pool_print(struct pool *pp, const char *modif)
1.21      thorpej  1364: {
                   1365:        int s;
                   1366:
1.49      thorpej  1367:        s = splvm();
1.25      thorpej  1368:        if (simple_lock_try(&pp->pr_slock) == 0) {
                   1369:                printf("pool %s is locked; try again later\n",
                   1370:                    pp->pr_wchan);
                   1371:                splx(s);
                   1372:                return;
                   1373:        }
                   1374:        pool_print1(pp, modif, printf);
1.21      thorpej  1375:        simple_unlock(&pp->pr_slock);
                   1376:        splx(s);
                   1377: }
                   1378:
1.25      thorpej  1379: void
1.42      thorpej  1380: pool_printit(struct pool *pp, const char *modif, void (*pr)(const char *, ...))
1.25      thorpej  1381: {
                   1382:        int didlock = 0;
                   1383:
                   1384:        if (pp == NULL) {
                   1385:                (*pr)("Must specify a pool to print.\n");
                   1386:                return;
                   1387:        }
                   1388:
                   1389:        /*
                   1390:         * Called from DDB; interrupts should be blocked, and all
                   1391:         * other processors should be paused.  We can skip locking
                   1392:         * the pool in this case.
                   1393:         *
                   1394:         * We do a simple_lock_try() just to print the lock
                   1395:         * status, however.
                   1396:         */
                   1397:
                   1398:        if (simple_lock_try(&pp->pr_slock) == 0)
                   1399:                (*pr)("WARNING: pool %s is locked\n", pp->pr_wchan);
                   1400:        else
                   1401:                didlock = 1;
                   1402:
                   1403:        pool_print1(pp, modif, pr);
                   1404:
                   1405:        if (didlock)
                   1406:                simple_unlock(&pp->pr_slock);
                   1407: }
                   1408:
1.21      thorpej  1409: static void
1.42      thorpej  1410: pool_print1(struct pool *pp, const char *modif, void (*pr)(const char *, ...))
1.3       pk       1411: {
1.25      thorpej  1412:        struct pool_item_header *ph;
1.44      thorpej  1413:        struct pool_cache *pc;
                   1414:        struct pool_cache_group *pcg;
1.25      thorpej  1415: #ifdef DIAGNOSTIC
                   1416:        struct pool_item *pi;
                   1417: #endif
1.44      thorpej  1418:        int i, print_log = 0, print_pagelist = 0, print_cache = 0;
1.25      thorpej  1419:        char c;
                   1420:
                   1421:        while ((c = *modif++) != '\0') {
                   1422:                if (c == 'l')
                   1423:                        print_log = 1;
                   1424:                if (c == 'p')
                   1425:                        print_pagelist = 1;
1.44      thorpej  1426:                if (c == 'c')
                   1427:                        print_cache = 1;
1.25      thorpej  1428:                modif++;
                   1429:        }
                   1430:
                   1431:        (*pr)("POOL %s: size %u, align %u, ioff %u, roflags 0x%08x\n",
                   1432:            pp->pr_wchan, pp->pr_size, pp->pr_align, pp->pr_itemoffset,
                   1433:            pp->pr_roflags);
                   1434:        (*pr)("\tpagesz %u, mtype %d\n", pp->pr_pagesz, pp->pr_mtype);
                   1435:        (*pr)("\talloc %p, release %p\n", pp->pr_alloc, pp->pr_free);
                   1436:        (*pr)("\tminitems %u, minpages %u, maxpages %u, npages %u\n",
                   1437:            pp->pr_minitems, pp->pr_minpages, pp->pr_maxpages, pp->pr_npages);
                   1438:        (*pr)("\titemsperpage %u, nitems %u, nout %u, hardlimit %u\n",
                   1439:            pp->pr_itemsperpage, pp->pr_nitems, pp->pr_nout, pp->pr_hardlimit);
                   1440:
                   1441:        (*pr)("\n\tnget %lu, nfail %lu, nput %lu\n",
                   1442:            pp->pr_nget, pp->pr_nfail, pp->pr_nput);
                   1443:        (*pr)("\tnpagealloc %lu, npagefree %lu, hiwat %u, nidle %lu\n",
                   1444:            pp->pr_npagealloc, pp->pr_npagefree, pp->pr_hiwat, pp->pr_nidle);
                   1445:
                   1446:        if (print_pagelist == 0)
                   1447:                goto skip_pagelist;
                   1448:
                   1449:        if ((ph = TAILQ_FIRST(&pp->pr_pagelist)) != NULL)
                   1450:                (*pr)("\n\tpage list:\n");
                   1451:        for (; ph != NULL; ph = TAILQ_NEXT(ph, ph_pagelist)) {
                   1452:                (*pr)("\t\tpage %p, nmissing %d, time %lu,%lu\n",
                   1453:                    ph->ph_page, ph->ph_nmissing,
                   1454:                    (u_long)ph->ph_time.tv_sec,
                   1455:                    (u_long)ph->ph_time.tv_usec);
                   1456: #ifdef DIAGNOSTIC
1.60.4.1! fvdl     1457:                TAILQ_FOREACH(pi, &ph->ph_itemlist, pi_list) {
1.25      thorpej  1458:                        if (pi->pi_magic != PI_MAGIC) {
                   1459:                                (*pr)("\t\t\titem %p, magic 0x%x\n",
                   1460:                                    pi, pi->pi_magic);
                   1461:                        }
                   1462:                }
                   1463: #endif
                   1464:        }
                   1465:        if (pp->pr_curpage == NULL)
                   1466:                (*pr)("\tno current page\n");
                   1467:        else
                   1468:                (*pr)("\tcurpage %p\n", pp->pr_curpage->ph_page);
                   1469:
                   1470:  skip_pagelist:
                   1471:
                   1472:        if (print_log == 0)
                   1473:                goto skip_log;
                   1474:
                   1475:        (*pr)("\n");
                   1476:        if ((pp->pr_roflags & PR_LOGGING) == 0)
                   1477:                (*pr)("\tno log\n");
                   1478:        else
                   1479:                pr_printlog(pp, NULL, pr);
1.3       pk       1480:
1.25      thorpej  1481:  skip_log:
1.44      thorpej  1482:
                   1483:        if (print_cache == 0)
                   1484:                goto skip_cache;
                   1485:
1.60.4.1! fvdl     1486:        TAILQ_FOREACH(pc, &pp->pr_cachelist, pc_poollist) {
1.44      thorpej  1487:                (*pr)("\tcache %p: allocfrom %p freeto %p\n", pc,
                   1488:                    pc->pc_allocfrom, pc->pc_freeto);
1.48      thorpej  1489:                (*pr)("\t    hits %lu misses %lu ngroups %lu nitems %lu\n",
                   1490:                    pc->pc_hits, pc->pc_misses, pc->pc_ngroups, pc->pc_nitems);
1.60.4.1! fvdl     1491:                TAILQ_FOREACH(pcg, &pc->pc_grouplist, pcg_list) {
1.44      thorpej  1492:                        (*pr)("\t\tgroup %p: avail %d\n", pcg, pcg->pcg_avail);
                   1493:                        for (i = 0; i < PCG_NOBJECTS; i++)
                   1494:                                (*pr)("\t\t\t%p\n", pcg->pcg_objects[i]);
                   1495:                }
                   1496:        }
                   1497:
                   1498:  skip_cache:
1.3       pk       1499:
1.25      thorpej  1500:        pr_enter_check(pp, pr);
1.3       pk       1501: }
                   1502:
                   1503: int
1.42      thorpej  1504: pool_chk(struct pool *pp, const char *label)
1.3       pk       1505: {
                   1506:        struct pool_item_header *ph;
                   1507:        int r = 0;
                   1508:
1.21      thorpej  1509:        simple_lock(&pp->pr_slock);
1.3       pk       1510:
1.60.4.1! fvdl     1511:        TAILQ_FOREACH(ph, &pp->pr_pagelist, ph_pagelist) {
1.3       pk       1512:                struct pool_item *pi;
                   1513:                int n;
                   1514:                caddr_t page;
                   1515:
                   1516:                page = (caddr_t)((u_long)ph & pp->pr_pagemask);
1.20      thorpej  1517:                if (page != ph->ph_page &&
                   1518:                    (pp->pr_roflags & PR_PHINPAGE) != 0) {
1.3       pk       1519:                        if (label != NULL)
                   1520:                                printf("%s: ", label);
1.16      briggs   1521:                        printf("pool(%p:%s): page inconsistency: page %p;"
                   1522:                               " at page head addr %p (p %p)\n", pp,
1.3       pk       1523:                                pp->pr_wchan, ph->ph_page,
                   1524:                                ph, page);
                   1525:                        r++;
                   1526:                        goto out;
                   1527:                }
                   1528:
                   1529:                for (pi = TAILQ_FIRST(&ph->ph_itemlist), n = 0;
                   1530:                     pi != NULL;
                   1531:                     pi = TAILQ_NEXT(pi,pi_list), n++) {
                   1532:
                   1533: #ifdef DIAGNOSTIC
                   1534:                        if (pi->pi_magic != PI_MAGIC) {
                   1535:                                if (label != NULL)
                   1536:                                        printf("%s: ", label);
                   1537:                                printf("pool(%s): free list modified: magic=%x;"
                   1538:                                       " page %p; item ordinal %d;"
                   1539:                                       " addr %p (p %p)\n",
                   1540:                                        pp->pr_wchan, pi->pi_magic, ph->ph_page,
                   1541:                                        n, pi, page);
                   1542:                                panic("pool");
                   1543:                        }
                   1544: #endif
                   1545:                        page = (caddr_t)((u_long)pi & pp->pr_pagemask);
                   1546:                        if (page == ph->ph_page)
                   1547:                                continue;
                   1548:
                   1549:                        if (label != NULL)
                   1550:                                printf("%s: ", label);
1.16      briggs   1551:                        printf("pool(%p:%s): page inconsistency: page %p;"
                   1552:                               " item ordinal %d; addr %p (p %p)\n", pp,
1.3       pk       1553:                                pp->pr_wchan, ph->ph_page,
                   1554:                                n, pi, page);
                   1555:                        r++;
                   1556:                        goto out;
                   1557:                }
                   1558:        }
                   1559: out:
1.21      thorpej  1560:        simple_unlock(&pp->pr_slock);
1.3       pk       1561:        return (r);
1.43      thorpej  1562: }
                   1563:
                   1564: /*
                   1565:  * pool_cache_init:
                   1566:  *
                   1567:  *     Initialize a pool cache.
                   1568:  *
                   1569:  *     NOTE: If the pool must be protected from interrupts, we expect
                   1570:  *     to be called at the appropriate interrupt priority level.
                   1571:  */
                   1572: void
                   1573: pool_cache_init(struct pool_cache *pc, struct pool *pp,
                   1574:     int (*ctor)(void *, void *, int),
                   1575:     void (*dtor)(void *, void *),
                   1576:     void *arg)
                   1577: {
                   1578:
                   1579:        TAILQ_INIT(&pc->pc_grouplist);
                   1580:        simple_lock_init(&pc->pc_slock);
                   1581:
                   1582:        pc->pc_allocfrom = NULL;
                   1583:        pc->pc_freeto = NULL;
                   1584:        pc->pc_pool = pp;
                   1585:
                   1586:        pc->pc_ctor = ctor;
                   1587:        pc->pc_dtor = dtor;
                   1588:        pc->pc_arg  = arg;
                   1589:
1.48      thorpej  1590:        pc->pc_hits   = 0;
                   1591:        pc->pc_misses = 0;
                   1592:
                   1593:        pc->pc_ngroups = 0;
                   1594:
                   1595:        pc->pc_nitems = 0;
                   1596:
1.43      thorpej  1597:        simple_lock(&pp->pr_slock);
                   1598:        TAILQ_INSERT_TAIL(&pp->pr_cachelist, pc, pc_poollist);
                   1599:        simple_unlock(&pp->pr_slock);
                   1600: }
                   1601:
                   1602: /*
                   1603:  * pool_cache_destroy:
                   1604:  *
                   1605:  *     Destroy a pool cache.
                   1606:  */
                   1607: void
                   1608: pool_cache_destroy(struct pool_cache *pc)
                   1609: {
                   1610:        struct pool *pp = pc->pc_pool;
                   1611:
                   1612:        /* First, invalidate the entire cache. */
                   1613:        pool_cache_invalidate(pc);
                   1614:
                   1615:        /* ...and remove it from the pool's cache list. */
                   1616:        simple_lock(&pp->pr_slock);
                   1617:        TAILQ_REMOVE(&pp->pr_cachelist, pc, pc_poollist);
                   1618:        simple_unlock(&pp->pr_slock);
                   1619: }
                   1620:
                   1621: static __inline void *
                   1622: pcg_get(struct pool_cache_group *pcg)
                   1623: {
                   1624:        void *object;
                   1625:        u_int idx;
                   1626:
                   1627:        KASSERT(pcg->pcg_avail <= PCG_NOBJECTS);
1.45      thorpej  1628:        KASSERT(pcg->pcg_avail != 0);
1.43      thorpej  1629:        idx = --pcg->pcg_avail;
                   1630:
                   1631:        KASSERT(pcg->pcg_objects[idx] != NULL);
                   1632:        object = pcg->pcg_objects[idx];
                   1633:        pcg->pcg_objects[idx] = NULL;
                   1634:
                   1635:        return (object);
                   1636: }
                   1637:
                   1638: static __inline void
                   1639: pcg_put(struct pool_cache_group *pcg, void *object)
                   1640: {
                   1641:        u_int idx;
                   1642:
                   1643:        KASSERT(pcg->pcg_avail < PCG_NOBJECTS);
                   1644:        idx = pcg->pcg_avail++;
                   1645:
                   1646:        KASSERT(pcg->pcg_objects[idx] == NULL);
                   1647:        pcg->pcg_objects[idx] = object;
                   1648: }
                   1649:
                   1650: /*
                   1651:  * pool_cache_get:
                   1652:  *
                   1653:  *     Get an object from a pool cache.
                   1654:  */
                   1655: void *
                   1656: pool_cache_get(struct pool_cache *pc, int flags)
                   1657: {
                   1658:        struct pool_cache_group *pcg;
                   1659:        void *object;
1.58      thorpej  1660:
                   1661: #ifdef LOCKDEBUG
                   1662:        if (flags & PR_WAITOK)
                   1663:                simple_lock_only_held(NULL, "pool_cache_get(PR_WAITOK)");
                   1664: #endif
1.43      thorpej  1665:
                   1666:        simple_lock(&pc->pc_slock);
                   1667:
                   1668:        if ((pcg = pc->pc_allocfrom) == NULL) {
1.60.4.1! fvdl     1669:                TAILQ_FOREACH(pcg, &pc->pc_grouplist, pcg_list) {
1.43      thorpej  1670:                        if (pcg->pcg_avail != 0) {
                   1671:                                pc->pc_allocfrom = pcg;
                   1672:                                goto have_group;
                   1673:                        }
                   1674:                }
                   1675:
                   1676:                /*
                   1677:                 * No groups with any available objects.  Allocate
                   1678:                 * a new object, construct it, and return it to
                   1679:                 * the caller.  We will allocate a group, if necessary,
                   1680:                 * when the object is freed back to the cache.
                   1681:                 */
1.48      thorpej  1682:                pc->pc_misses++;
1.43      thorpej  1683:                simple_unlock(&pc->pc_slock);
                   1684:                object = pool_get(pc->pc_pool, flags);
                   1685:                if (object != NULL && pc->pc_ctor != NULL) {
                   1686:                        if ((*pc->pc_ctor)(pc->pc_arg, object, flags) != 0) {
                   1687:                                pool_put(pc->pc_pool, object);
                   1688:                                return (NULL);
                   1689:                        }
                   1690:                }
                   1691:                return (object);
                   1692:        }
                   1693:
                   1694:  have_group:
1.48      thorpej  1695:        pc->pc_hits++;
                   1696:        pc->pc_nitems--;
1.43      thorpej  1697:        object = pcg_get(pcg);
                   1698:
                   1699:        if (pcg->pcg_avail == 0)
                   1700:                pc->pc_allocfrom = NULL;
1.45      thorpej  1701:
1.43      thorpej  1702:        simple_unlock(&pc->pc_slock);
                   1703:
                   1704:        return (object);
                   1705: }
                   1706:
                   1707: /*
                   1708:  * pool_cache_put:
                   1709:  *
                   1710:  *     Put an object back to the pool cache.
                   1711:  */
                   1712: void
                   1713: pool_cache_put(struct pool_cache *pc, void *object)
                   1714: {
                   1715:        struct pool_cache_group *pcg;
1.60      thorpej  1716:        int s;
1.43      thorpej  1717:
                   1718:        simple_lock(&pc->pc_slock);
                   1719:
                   1720:        if ((pcg = pc->pc_freeto) == NULL) {
1.60.4.1! fvdl     1721:                TAILQ_FOREACH(pcg, &pc->pc_grouplist, pcg_list) {
1.43      thorpej  1722:                        if (pcg->pcg_avail != PCG_NOBJECTS) {
                   1723:                                pc->pc_freeto = pcg;
                   1724:                                goto have_group;
                   1725:                        }
                   1726:                }
                   1727:
                   1728:                /*
                   1729:                 * No empty groups to free the object to.  Attempt to
1.47      thorpej  1730:                 * allocate one.
1.43      thorpej  1731:                 */
1.47      thorpej  1732:                simple_unlock(&pc->pc_slock);
1.60      thorpej  1733:                s = splvm();
1.43      thorpej  1734:                pcg = pool_get(&pcgpool, PR_NOWAIT);
1.60      thorpej  1735:                splx(s);
1.43      thorpej  1736:                if (pcg != NULL) {
                   1737:                        memset(pcg, 0, sizeof(*pcg));
1.47      thorpej  1738:                        simple_lock(&pc->pc_slock);
1.48      thorpej  1739:                        pc->pc_ngroups++;
1.43      thorpej  1740:                        TAILQ_INSERT_TAIL(&pc->pc_grouplist, pcg, pcg_list);
1.47      thorpej  1741:                        if (pc->pc_freeto == NULL)
                   1742:                                pc->pc_freeto = pcg;
1.43      thorpej  1743:                        goto have_group;
                   1744:                }
                   1745:
                   1746:                /*
                   1747:                 * Unable to allocate a cache group; destruct the object
                   1748:                 * and free it back to the pool.
                   1749:                 */
1.51      thorpej  1750:                pool_cache_destruct_object(pc, object);
1.43      thorpej  1751:                return;
                   1752:        }
                   1753:
                   1754:  have_group:
1.48      thorpej  1755:        pc->pc_nitems++;
1.43      thorpej  1756:        pcg_put(pcg, object);
                   1757:
                   1758:        if (pcg->pcg_avail == PCG_NOBJECTS)
                   1759:                pc->pc_freeto = NULL;
                   1760:
                   1761:        simple_unlock(&pc->pc_slock);
1.51      thorpej  1762: }
                   1763:
                   1764: /*
                   1765:  * pool_cache_destruct_object:
                   1766:  *
                   1767:  *     Force destruction of an object and its release back into
                   1768:  *     the pool.
                   1769:  */
                   1770: void
                   1771: pool_cache_destruct_object(struct pool_cache *pc, void *object)
                   1772: {
                   1773:
                   1774:        if (pc->pc_dtor != NULL)
                   1775:                (*pc->pc_dtor)(pc->pc_arg, object);
                   1776:        pool_put(pc->pc_pool, object);
1.43      thorpej  1777: }
                   1778:
                   1779: /*
                   1780:  * pool_cache_do_invalidate:
                   1781:  *
                   1782:  *     This internal function implements pool_cache_invalidate() and
                   1783:  *     pool_cache_reclaim().
                   1784:  */
                   1785: static void
                   1786: pool_cache_do_invalidate(struct pool_cache *pc, int free_groups,
1.56      sommerfe 1787:     void (*putit)(struct pool *, void *))
1.43      thorpej  1788: {
                   1789:        struct pool_cache_group *pcg, *npcg;
                   1790:        void *object;
1.60      thorpej  1791:        int s;
1.43      thorpej  1792:
                   1793:        for (pcg = TAILQ_FIRST(&pc->pc_grouplist); pcg != NULL;
                   1794:             pcg = npcg) {
                   1795:                npcg = TAILQ_NEXT(pcg, pcg_list);
                   1796:                while (pcg->pcg_avail != 0) {
1.48      thorpej  1797:                        pc->pc_nitems--;
1.43      thorpej  1798:                        object = pcg_get(pcg);
1.45      thorpej  1799:                        if (pcg->pcg_avail == 0 && pc->pc_allocfrom == pcg)
                   1800:                                pc->pc_allocfrom = NULL;
1.43      thorpej  1801:                        if (pc->pc_dtor != NULL)
                   1802:                                (*pc->pc_dtor)(pc->pc_arg, object);
1.56      sommerfe 1803:                        (*putit)(pc->pc_pool, object);
1.43      thorpej  1804:                }
                   1805:                if (free_groups) {
1.48      thorpej  1806:                        pc->pc_ngroups--;
1.43      thorpej  1807:                        TAILQ_REMOVE(&pc->pc_grouplist, pcg, pcg_list);
1.46      thorpej  1808:                        if (pc->pc_freeto == pcg)
                   1809:                                pc->pc_freeto = NULL;
1.60      thorpej  1810:                        s = splvm();
1.43      thorpej  1811:                        pool_put(&pcgpool, pcg);
1.60      thorpej  1812:                        splx(s);
1.43      thorpej  1813:                }
                   1814:        }
                   1815: }
                   1816:
                   1817: /*
                   1818:  * pool_cache_invalidate:
                   1819:  *
                   1820:  *     Invalidate a pool cache (destruct and release all of the
                   1821:  *     cached objects).
                   1822:  */
                   1823: void
                   1824: pool_cache_invalidate(struct pool_cache *pc)
                   1825: {
                   1826:
                   1827:        simple_lock(&pc->pc_slock);
1.56      sommerfe 1828:        pool_cache_do_invalidate(pc, 0, pool_put);
1.43      thorpej  1829:        simple_unlock(&pc->pc_slock);
                   1830: }
                   1831:
                   1832: /*
                   1833:  * pool_cache_reclaim:
                   1834:  *
                   1835:  *     Reclaim a pool cache for pool_reclaim().
                   1836:  */
                   1837: static void
                   1838: pool_cache_reclaim(struct pool_cache *pc)
                   1839: {
                   1840:
1.47      thorpej  1841:        simple_lock(&pc->pc_slock);
1.43      thorpej  1842:        pool_cache_do_invalidate(pc, 1, pool_do_put);
                   1843:        simple_unlock(&pc->pc_slock);
1.3       pk       1844: }

CVSweb <webmaster@jp.NetBSD.org>