[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 1.50.2.1

1.50.2.1! nathanw     1: /*     $NetBSD: subr_pool.c,v 1.50 2001/01/29 02:38:02 enami Exp $     */
1.1       pk          2:
                      3: /*-
1.43      thorpej     4:  * Copyright (c) 1997, 1999, 2000 The NetBSD Foundation, Inc.
1.1       pk          5:  * All rights reserved.
                      6:  *
                      7:  * This code is derived from software contributed to The NetBSD Foundation
1.20      thorpej     8:  * by Paul Kranenburg; by Jason R. Thorpe of the Numerical Aerospace
                      9:  * Simulation Facility, NASA Ames Research Center.
1.1       pk         10:  *
                     11:  * Redistribution and use in source and binary forms, with or without
                     12:  * modification, are permitted provided that the following conditions
                     13:  * are met:
                     14:  * 1. Redistributions of source code must retain the above copyright
                     15:  *    notice, this list of conditions and the following disclaimer.
                     16:  * 2. Redistributions in binary form must reproduce the above copyright
                     17:  *    notice, this list of conditions and the following disclaimer in the
                     18:  *    documentation and/or other materials provided with the distribution.
                     19:  * 3. All advertising materials mentioning features or use of this software
                     20:  *    must display the following acknowledgement:
1.13      christos   21:  *     This product includes software developed by the NetBSD
                     22:  *     Foundation, Inc. and its contributors.
1.1       pk         23:  * 4. Neither the name of The NetBSD Foundation nor the names of its
                     24:  *    contributors may be used to endorse or promote products derived
                     25:  *    from this software without specific prior written permission.
                     26:  *
                     27:  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
                     28:  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
                     29:  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
                     30:  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
                     31:  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
                     32:  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
                     33:  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
                     34:  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
                     35:  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
                     36:  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
                     37:  * POSSIBILITY OF SUCH DAMAGE.
                     38:  */
1.24      scottr     39:
1.25      thorpej    40: #include "opt_pool.h"
1.24      scottr     41: #include "opt_poollog.h"
1.28      thorpej    42: #include "opt_lockdebug.h"
1.1       pk         43:
                     44: #include <sys/param.h>
                     45: #include <sys/systm.h>
                     46: #include <sys/proc.h>
                     47: #include <sys/errno.h>
                     48: #include <sys/kernel.h>
                     49: #include <sys/malloc.h>
                     50: #include <sys/lock.h>
                     51: #include <sys/pool.h>
1.20      thorpej    52: #include <sys/syslog.h>
1.3       pk         53:
                     54: #include <uvm/uvm.h>
                     55:
1.1       pk         56: /*
                     57:  * Pool resource management utility.
1.3       pk         58:  *
                     59:  * Memory is allocated in pages which are split into pieces according
                     60:  * to the pool item size. Each page is kept on a list headed by `pr_pagelist'
                     61:  * in the pool structure and the individual pool items are on a linked list
                     62:  * headed by `ph_itemlist' in each page header. The memory for building
                     63:  * the page list is either taken from the allocated pages themselves (for
                     64:  * small pool items) or taken from an internal pool of page headers (`phpool').
1.1       pk         65:  */
                     66:
1.3       pk         67: /* List of all pools */
1.5       thorpej    68: TAILQ_HEAD(,pool) pool_head = TAILQ_HEAD_INITIALIZER(pool_head);
1.3       pk         69:
                     70: /* Private pool for page header structures */
                     71: static struct pool phpool;
                     72:
                     73: /* # of seconds to retain page after last use */
                     74: int pool_inactive_time = 10;
                     75:
                     76: /* Next candidate for drainage (see pool_drain()) */
1.23      thorpej    77: static struct pool     *drainpp;
                     78:
                     79: /* This spin lock protects both pool_head and drainpp. */
                     80: struct simplelock pool_head_slock = SIMPLELOCK_INITIALIZER;
1.3       pk         81:
                     82: struct pool_item_header {
                     83:        /* Page headers */
                     84:        TAILQ_ENTRY(pool_item_header)
                     85:                                ph_pagelist;    /* pool page list */
                     86:        TAILQ_HEAD(,pool_item)  ph_itemlist;    /* chunk list for this page */
                     87:        LIST_ENTRY(pool_item_header)
                     88:                                ph_hashlist;    /* Off-page page headers */
                     89:        int                     ph_nmissing;    /* # of chunks in use */
                     90:        caddr_t                 ph_page;        /* this page's address */
                     91:        struct timeval          ph_time;        /* last referenced */
                     92: };
                     93:
1.1       pk         94: struct pool_item {
1.3       pk         95: #ifdef DIAGNOSTIC
                     96:        int pi_magic;
1.33      chs        97: #endif
1.25      thorpej    98: #define        PI_MAGIC 0xdeadbeef
1.3       pk         99:        /* Other entries use only this list entry */
                    100:        TAILQ_ENTRY(pool_item)  pi_list;
                    101: };
                    102:
1.25      thorpej   103: #define        PR_HASH_INDEX(pp,addr) \
1.3       pk        104:        (((u_long)(addr) >> (pp)->pr_pageshift) & (PR_HASHTABSIZE - 1))
                    105:
1.50.2.1! nathanw   106: #define        POOL_NEEDS_CATCHUP(pp)                                          \
        !           107:        ((pp)->pr_nitems < (pp)->pr_minitems)
        !           108:
1.43      thorpej   109: /*
                    110:  * Pool cache management.
                    111:  *
                    112:  * Pool caches provide a way for constructed objects to be cached by the
                    113:  * pool subsystem.  This can lead to performance improvements by avoiding
                    114:  * needless object construction/destruction; it is deferred until absolutely
                    115:  * necessary.
                    116:  *
                    117:  * Caches are grouped into cache groups.  Each cache group references
                    118:  * up to 16 constructed objects.  When a cache allocates an object
                    119:  * from the pool, it calls the object's constructor and places it into
                    120:  * a cache group.  When a cache group frees an object back to the pool,
                    121:  * it first calls the object's destructor.  This allows the object to
                    122:  * persist in constructed form while freed to the cache.
                    123:  *
                    124:  * Multiple caches may exist for each pool.  This allows a single
                    125:  * object type to have multiple constructed forms.  The pool references
                    126:  * each cache, so that when a pool is drained by the pagedaemon, it can
                    127:  * drain each individual cache as well.  Each time a cache is drained,
                    128:  * the most idle cache group is freed to the pool in its entirety.
                    129:  *
                    130:  * Pool caches are layed on top of pools.  By layering them, we can avoid
                    131:  * the complexity of cache management for pools which would not benefit
                    132:  * from it.
                    133:  */
                    134:
                    135: /* The cache group pool. */
                    136: static struct pool pcgpool;
                    137:
                    138: /* The pool cache group. */
                    139: #define        PCG_NOBJECTS            16
                    140: struct pool_cache_group {
                    141:        TAILQ_ENTRY(pool_cache_group)
                    142:                pcg_list;       /* link in the pool cache's group list */
                    143:        u_int   pcg_avail;      /* # available objects */
                    144:                                /* pointers to the objects */
                    145:        void    *pcg_objects[PCG_NOBJECTS];
                    146: };
1.3       pk        147:
1.43      thorpej   148: static void    pool_cache_reclaim(struct pool_cache *);
1.3       pk        149:
1.42      thorpej   150: static int     pool_catchup(struct pool *);
1.50.2.1! nathanw   151: static void    pool_prime_page(struct pool *, caddr_t,
        !           152:                    struct pool_item_header *);
1.42      thorpej   153: static void    *pool_page_alloc(unsigned long, int, int);
                    154: static void    pool_page_free(void *, unsigned long, int);
1.3       pk        155:
1.42      thorpej   156: static void pool_print1(struct pool *, const char *,
                    157:        void (*)(const char *, ...));
1.3       pk        158:
                    159: /*
1.50.2.1! nathanw   160:  * Pool log entry. An array of these is allocated in pool_init().
1.3       pk        161:  */
                    162: struct pool_log {
                    163:        const char      *pl_file;
                    164:        long            pl_line;
                    165:        int             pl_action;
1.25      thorpej   166: #define        PRLOG_GET       1
                    167: #define        PRLOG_PUT       2
1.3       pk        168:        void            *pl_addr;
1.1       pk        169: };
                    170:
1.3       pk        171: /* Number of entries in pool log buffers */
1.17      thorpej   172: #ifndef POOL_LOGSIZE
                    173: #define        POOL_LOGSIZE    10
                    174: #endif
                    175:
                    176: int pool_logsize = POOL_LOGSIZE;
1.1       pk        177:
1.50.2.1! nathanw   178: #ifdef POOL_DIAGNOSTIC
1.42      thorpej   179: static __inline void
                    180: pr_log(struct pool *pp, void *v, int action, const char *file, long line)
1.3       pk        181: {
                    182:        int n = pp->pr_curlogentry;
                    183:        struct pool_log *pl;
                    184:
1.20      thorpej   185:        if ((pp->pr_roflags & PR_LOGGING) == 0)
1.3       pk        186:                return;
                    187:
                    188:        /*
                    189:         * Fill in the current entry. Wrap around and overwrite
                    190:         * the oldest entry if necessary.
                    191:         */
                    192:        pl = &pp->pr_log[n];
                    193:        pl->pl_file = file;
                    194:        pl->pl_line = line;
                    195:        pl->pl_action = action;
                    196:        pl->pl_addr = v;
                    197:        if (++n >= pp->pr_logsize)
                    198:                n = 0;
                    199:        pp->pr_curlogentry = n;
                    200: }
                    201:
                    202: static void
1.42      thorpej   203: pr_printlog(struct pool *pp, struct pool_item *pi,
                    204:     void (*pr)(const char *, ...))
1.3       pk        205: {
                    206:        int i = pp->pr_logsize;
                    207:        int n = pp->pr_curlogentry;
                    208:
1.20      thorpej   209:        if ((pp->pr_roflags & PR_LOGGING) == 0)
1.3       pk        210:                return;
                    211:
                    212:        /*
                    213:         * Print all entries in this pool's log.
                    214:         */
                    215:        while (i-- > 0) {
                    216:                struct pool_log *pl = &pp->pr_log[n];
                    217:                if (pl->pl_action != 0) {
1.25      thorpej   218:                        if (pi == NULL || pi == pl->pl_addr) {
                    219:                                (*pr)("\tlog entry %d:\n", i);
                    220:                                (*pr)("\t\taction = %s, addr = %p\n",
                    221:                                    pl->pl_action == PRLOG_GET ? "get" : "put",
                    222:                                    pl->pl_addr);
                    223:                                (*pr)("\t\tfile: %s at line %lu\n",
                    224:                                    pl->pl_file, pl->pl_line);
                    225:                        }
1.3       pk        226:                }
                    227:                if (++n >= pp->pr_logsize)
                    228:                        n = 0;
                    229:        }
                    230: }
1.25      thorpej   231:
1.42      thorpej   232: static __inline void
                    233: pr_enter(struct pool *pp, const char *file, long line)
1.25      thorpej   234: {
                    235:
1.34      thorpej   236:        if (__predict_false(pp->pr_entered_file != NULL)) {
1.25      thorpej   237:                printf("pool %s: reentrancy at file %s line %ld\n",
                    238:                    pp->pr_wchan, file, line);
                    239:                printf("         previous entry at file %s line %ld\n",
                    240:                    pp->pr_entered_file, pp->pr_entered_line);
                    241:                panic("pr_enter");
                    242:        }
                    243:
                    244:        pp->pr_entered_file = file;
                    245:        pp->pr_entered_line = line;
                    246: }
                    247:
1.42      thorpej   248: static __inline void
                    249: pr_leave(struct pool *pp)
1.25      thorpej   250: {
                    251:
1.34      thorpej   252:        if (__predict_false(pp->pr_entered_file == NULL)) {
1.25      thorpej   253:                printf("pool %s not entered?\n", pp->pr_wchan);
                    254:                panic("pr_leave");
                    255:        }
                    256:
                    257:        pp->pr_entered_file = NULL;
                    258:        pp->pr_entered_line = 0;
                    259: }
                    260:
1.42      thorpej   261: static __inline void
                    262: pr_enter_check(struct pool *pp, void (*pr)(const char *, ...))
1.25      thorpej   263: {
                    264:
                    265:        if (pp->pr_entered_file != NULL)
                    266:                (*pr)("\n\tcurrently entered from file %s line %ld\n",
                    267:                    pp->pr_entered_file, pp->pr_entered_line);
                    268: }
1.3       pk        269: #else
1.25      thorpej   270: #define        pr_log(pp, v, action, file, line)
                    271: #define        pr_printlog(pp, pi, pr)
                    272: #define        pr_enter(pp, file, line)
                    273: #define        pr_leave(pp)
                    274: #define        pr_enter_check(pp, pr)
1.50.2.1! nathanw   275: #endif /* POOL_DIAGNOSTIC */
1.3       pk        276:
                    277: /*
                    278:  * Return the pool page header based on page address.
                    279:  */
1.42      thorpej   280: static __inline struct pool_item_header *
                    281: pr_find_pagehead(struct pool *pp, caddr_t page)
1.3       pk        282: {
                    283:        struct pool_item_header *ph;
                    284:
1.20      thorpej   285:        if ((pp->pr_roflags & PR_PHINPAGE) != 0)
1.3       pk        286:                return ((struct pool_item_header *)(page + pp->pr_phoffset));
                    287:
                    288:        for (ph = LIST_FIRST(&pp->pr_hashtab[PR_HASH_INDEX(pp, page)]);
                    289:             ph != NULL;
                    290:             ph = LIST_NEXT(ph, ph_hashlist)) {
                    291:                if (ph->ph_page == page)
                    292:                        return (ph);
                    293:        }
                    294:        return (NULL);
                    295: }
                    296:
                    297: /*
                    298:  * Remove a page from the pool.
                    299:  */
1.42      thorpej   300: static __inline void
                    301: pr_rmpage(struct pool *pp, struct pool_item_header *ph)
1.3       pk        302: {
                    303:
                    304:        /*
1.7       thorpej   305:         * If the page was idle, decrement the idle page count.
1.3       pk        306:         */
1.6       thorpej   307:        if (ph->ph_nmissing == 0) {
                    308: #ifdef DIAGNOSTIC
                    309:                if (pp->pr_nidle == 0)
                    310:                        panic("pr_rmpage: nidle inconsistent");
1.20      thorpej   311:                if (pp->pr_nitems < pp->pr_itemsperpage)
                    312:                        panic("pr_rmpage: nitems inconsistent");
1.6       thorpej   313: #endif
                    314:                pp->pr_nidle--;
                    315:        }
1.7       thorpej   316:
1.20      thorpej   317:        pp->pr_nitems -= pp->pr_itemsperpage;
                    318:
1.7       thorpej   319:        /*
                    320:         * Unlink a page from the pool and release it.
                    321:         */
                    322:        TAILQ_REMOVE(&pp->pr_pagelist, ph, ph_pagelist);
                    323:        (*pp->pr_free)(ph->ph_page, pp->pr_pagesz, pp->pr_mtype);
                    324:        pp->pr_npages--;
                    325:        pp->pr_npagefree++;
1.6       thorpej   326:
1.22      chs       327:        if ((pp->pr_roflags & PR_PHINPAGE) == 0) {
1.27      pk        328:                int s;
1.22      chs       329:                LIST_REMOVE(ph, ph_hashlist);
1.27      pk        330:                s = splhigh();
1.22      chs       331:                pool_put(&phpool, ph);
1.27      pk        332:                splx(s);
1.22      chs       333:        }
                    334:
1.3       pk        335:        if (pp->pr_curpage == ph) {
                    336:                /*
                    337:                 * Find a new non-empty page header, if any.
                    338:                 * Start search from the page head, to increase the
                    339:                 * chance for "high water" pages to be freed.
                    340:                 */
                    341:                for (ph = TAILQ_FIRST(&pp->pr_pagelist); ph != NULL;
                    342:                     ph = TAILQ_NEXT(ph, ph_pagelist))
                    343:                        if (TAILQ_FIRST(&ph->ph_itemlist) != NULL)
                    344:                                break;
                    345:
                    346:                pp->pr_curpage = ph;
1.21      thorpej   347:        }
1.3       pk        348: }
                    349:
                    350: /*
                    351:  * Initialize the given pool resource structure.
                    352:  *
                    353:  * We export this routine to allow other kernel parts to declare
                    354:  * static pools that must be initialized before malloc() is available.
                    355:  */
                    356: void
1.42      thorpej   357: pool_init(struct pool *pp, size_t size, u_int align, u_int ioff, int flags,
                    358:     const char *wchan, size_t pagesz,
                    359:     void *(*alloc)(unsigned long, int, int),
                    360:     void (*release)(void *, unsigned long, int),
                    361:     int mtype)
1.3       pk        362: {
1.16      briggs    363:        int off, slack, i;
1.3       pk        364:
1.25      thorpej   365: #ifdef POOL_DIAGNOSTIC
                    366:        /*
                    367:         * Always log if POOL_DIAGNOSTIC is defined.
                    368:         */
                    369:        if (pool_logsize != 0)
                    370:                flags |= PR_LOGGING;
                    371: #endif
                    372:
1.3       pk        373:        /*
                    374:         * Check arguments and construct default values.
                    375:         */
1.36      pk        376:        if (!powerof2(pagesz))
1.3       pk        377:                panic("pool_init: page size invalid (%lx)\n", (u_long)pagesz);
                    378:
1.4       thorpej   379:        if (alloc == NULL && release == NULL) {
1.3       pk        380:                alloc = pool_page_alloc;
                    381:                release = pool_page_free;
1.4       thorpej   382:                pagesz = PAGE_SIZE;     /* Rounds to PAGE_SIZE anyhow. */
                    383:        } else if ((alloc != NULL && release != NULL) == 0) {
                    384:                /* If you specifiy one, must specify both. */
                    385:                panic("pool_init: must specify alloc and release together");
                    386:        }
                    387:
1.3       pk        388:        if (pagesz == 0)
                    389:                pagesz = PAGE_SIZE;
                    390:
                    391:        if (align == 0)
                    392:                align = ALIGN(1);
1.14      thorpej   393:
                    394:        if (size < sizeof(struct pool_item))
                    395:                size = sizeof(struct pool_item);
1.3       pk        396:
1.35      pk        397:        size = ALIGN(size);
1.43      thorpej   398:        if (size > pagesz)
1.35      pk        399:                panic("pool_init: pool item size (%lu) too large",
                    400:                      (u_long)size);
                    401:
1.3       pk        402:        /*
                    403:         * Initialize the pool structure.
                    404:         */
                    405:        TAILQ_INIT(&pp->pr_pagelist);
1.43      thorpej   406:        TAILQ_INIT(&pp->pr_cachelist);
1.3       pk        407:        pp->pr_curpage = NULL;
                    408:        pp->pr_npages = 0;
                    409:        pp->pr_minitems = 0;
                    410:        pp->pr_minpages = 0;
                    411:        pp->pr_maxpages = UINT_MAX;
1.20      thorpej   412:        pp->pr_roflags = flags;
                    413:        pp->pr_flags = 0;
1.35      pk        414:        pp->pr_size = size;
1.3       pk        415:        pp->pr_align = align;
                    416:        pp->pr_wchan = wchan;
                    417:        pp->pr_mtype = mtype;
                    418:        pp->pr_alloc = alloc;
                    419:        pp->pr_free = release;
                    420:        pp->pr_pagesz = pagesz;
                    421:        pp->pr_pagemask = ~(pagesz - 1);
                    422:        pp->pr_pageshift = ffs(pagesz) - 1;
1.20      thorpej   423:        pp->pr_nitems = 0;
                    424:        pp->pr_nout = 0;
                    425:        pp->pr_hardlimit = UINT_MAX;
                    426:        pp->pr_hardlimit_warning = NULL;
1.31      thorpej   427:        pp->pr_hardlimit_ratecap.tv_sec = 0;
                    428:        pp->pr_hardlimit_ratecap.tv_usec = 0;
                    429:        pp->pr_hardlimit_warning_last.tv_sec = 0;
                    430:        pp->pr_hardlimit_warning_last.tv_usec = 0;
1.3       pk        431:
                    432:        /*
                    433:         * Decide whether to put the page header off page to avoid
                    434:         * wasting too large a part of the page. Off-page page headers
                    435:         * go on a hash table, so we can match a returned item
                    436:         * with its header based on the page address.
                    437:         * We use 1/16 of the page size as the threshold (XXX: tune)
                    438:         */
                    439:        if (pp->pr_size < pagesz/16) {
                    440:                /* Use the end of the page for the page header */
1.20      thorpej   441:                pp->pr_roflags |= PR_PHINPAGE;
1.3       pk        442:                pp->pr_phoffset = off =
                    443:                        pagesz - ALIGN(sizeof(struct pool_item_header));
1.2       pk        444:        } else {
1.3       pk        445:                /* The page header will be taken from our page header pool */
                    446:                pp->pr_phoffset = 0;
                    447:                off = pagesz;
1.16      briggs    448:                for (i = 0; i < PR_HASHTABSIZE; i++) {
                    449:                        LIST_INIT(&pp->pr_hashtab[i]);
                    450:                }
1.2       pk        451:        }
1.1       pk        452:
1.3       pk        453:        /*
                    454:         * Alignment is to take place at `ioff' within the item. This means
                    455:         * we must reserve up to `align - 1' bytes on the page to allow
                    456:         * appropriate positioning of each item.
                    457:         *
                    458:         * Silently enforce `0 <= ioff < align'.
                    459:         */
                    460:        pp->pr_itemoffset = ioff = ioff % align;
                    461:        pp->pr_itemsperpage = (off - ((align - ioff) % align)) / pp->pr_size;
1.43      thorpej   462:        KASSERT(pp->pr_itemsperpage != 0);
1.3       pk        463:
                    464:        /*
                    465:         * Use the slack between the chunks and the page header
                    466:         * for "cache coloring".
                    467:         */
                    468:        slack = off - pp->pr_itemsperpage * pp->pr_size;
                    469:        pp->pr_maxcolor = (slack / align) * align;
                    470:        pp->pr_curcolor = 0;
                    471:
                    472:        pp->pr_nget = 0;
                    473:        pp->pr_nfail = 0;
                    474:        pp->pr_nput = 0;
                    475:        pp->pr_npagealloc = 0;
                    476:        pp->pr_npagefree = 0;
1.1       pk        477:        pp->pr_hiwat = 0;
1.8       thorpej   478:        pp->pr_nidle = 0;
1.3       pk        479:
1.50.2.1! nathanw   480: #ifdef POOL_DIAGNOSTIC
1.25      thorpej   481:        if (flags & PR_LOGGING) {
                    482:                if (kmem_map == NULL ||
                    483:                    (pp->pr_log = malloc(pool_logsize * sizeof(struct pool_log),
                    484:                     M_TEMP, M_NOWAIT)) == NULL)
1.20      thorpej   485:                        pp->pr_roflags &= ~PR_LOGGING;
1.3       pk        486:                pp->pr_curlogentry = 0;
                    487:                pp->pr_logsize = pool_logsize;
                    488:        }
1.50.2.1! nathanw   489: #endif
1.25      thorpej   490:
                    491:        pp->pr_entered_file = NULL;
                    492:        pp->pr_entered_line = 0;
1.3       pk        493:
1.21      thorpej   494:        simple_lock_init(&pp->pr_slock);
1.1       pk        495:
1.3       pk        496:        /*
1.43      thorpej   497:         * Initialize private page header pool and cache magazine pool if we
                    498:         * haven't done so yet.
1.23      thorpej   499:         * XXX LOCKING.
1.3       pk        500:         */
                    501:        if (phpool.pr_size == 0) {
                    502:                pool_init(&phpool, sizeof(struct pool_item_header), 0, 0,
1.43      thorpej   503:                    0, "phpool", 0, 0, 0, 0);
                    504:                pool_init(&pcgpool, sizeof(struct pool_cache_group), 0, 0,
                    505:                    0, "pcgpool", 0, 0, 0, 0);
1.1       pk        506:        }
                    507:
1.23      thorpej   508:        /* Insert into the list of all pools. */
                    509:        simple_lock(&pool_head_slock);
                    510:        TAILQ_INSERT_TAIL(&pool_head, pp, pr_poollist);
                    511:        simple_unlock(&pool_head_slock);
1.1       pk        512: }
                    513:
                    514: /*
                    515:  * De-commision a pool resource.
                    516:  */
                    517: void
1.42      thorpej   518: pool_destroy(struct pool *pp)
1.1       pk        519: {
1.3       pk        520:        struct pool_item_header *ph;
1.43      thorpej   521:        struct pool_cache *pc;
                    522:
                    523:        /* Destroy all caches for this pool. */
                    524:        while ((pc = TAILQ_FIRST(&pp->pr_cachelist)) != NULL)
                    525:                pool_cache_destroy(pc);
1.3       pk        526:
                    527: #ifdef DIAGNOSTIC
1.20      thorpej   528:        if (pp->pr_nout != 0) {
1.25      thorpej   529:                pr_printlog(pp, NULL, printf);
1.20      thorpej   530:                panic("pool_destroy: pool busy: still out: %u\n",
                    531:                    pp->pr_nout);
1.3       pk        532:        }
                    533: #endif
1.1       pk        534:
1.3       pk        535:        /* Remove all pages */
1.20      thorpej   536:        if ((pp->pr_roflags & PR_STATIC) == 0)
1.3       pk        537:                while ((ph = pp->pr_pagelist.tqh_first) != NULL)
                    538:                        pr_rmpage(pp, ph);
                    539:
                    540:        /* Remove from global pool list */
1.23      thorpej   541:        simple_lock(&pool_head_slock);
1.3       pk        542:        TAILQ_REMOVE(&pool_head, pp, pr_poollist);
1.23      thorpej   543:        /* XXX Only clear this if we were drainpp? */
1.3       pk        544:        drainpp = NULL;
1.23      thorpej   545:        simple_unlock(&pool_head_slock);
1.3       pk        546:
1.50.2.1! nathanw   547: #ifdef POOL_DIAGNOSTIC
1.20      thorpej   548:        if ((pp->pr_roflags & PR_LOGGING) != 0)
1.3       pk        549:                free(pp->pr_log, M_TEMP);
1.50.2.1! nathanw   550: #endif
1.2       pk        551:
1.20      thorpej   552:        if (pp->pr_roflags & PR_FREEHEADER)
1.3       pk        553:                free(pp, M_POOL);
1.1       pk        554: }
                    555:
1.50.2.1! nathanw   556: static __inline struct pool_item_header *
        !           557: pool_alloc_item_header(struct pool *pp, caddr_t storage, int flags)
        !           558: {
        !           559:        struct pool_item_header *ph;
        !           560:        int s;
        !           561:
        !           562:        LOCK_ASSERT(simple_lock_held(&pp->pr_slock) == 0);
        !           563:
        !           564:        if ((pp->pr_roflags & PR_PHINPAGE) != 0)
        !           565:                ph = (struct pool_item_header *) (storage + pp->pr_phoffset);
        !           566:        else {
        !           567:                s = splhigh();
        !           568:                ph = pool_get(&phpool, flags);
        !           569:                splx(s);
        !           570:        }
        !           571:
        !           572:        return (ph);
        !           573: }
1.1       pk        574:
                    575: /*
1.3       pk        576:  * Grab an item from the pool; must be called at appropriate spl level
1.1       pk        577:  */
1.3       pk        578: void *
1.50.2.1! nathanw   579: #ifdef POOL_DIAGNOSTIC
1.42      thorpej   580: _pool_get(struct pool *pp, int flags, const char *file, long line)
1.50.2.1! nathanw   581: #else
        !           582: pool_get(struct pool *pp, int flags)
        !           583: #endif
1.1       pk        584: {
                    585:        struct pool_item *pi;
1.3       pk        586:        struct pool_item_header *ph;
1.50.2.1! nathanw   587:        void *v;
1.1       pk        588:
1.2       pk        589: #ifdef DIAGNOSTIC
1.34      thorpej   590:        if (__predict_false((pp->pr_roflags & PR_STATIC) &&
                    591:                            (flags & PR_MALLOCOK))) {
1.25      thorpej   592:                pr_printlog(pp, NULL, printf);
1.2       pk        593:                panic("pool_get: static");
1.3       pk        594:        }
1.2       pk        595:
1.37      sommerfe  596:        if (__predict_false(curproc == NULL && doing_shutdown == 0 &&
                    597:                            (flags & PR_WAITOK) != 0))
1.3       pk        598:                panic("pool_get: must have NOWAIT");
1.1       pk        599:
1.50.2.1! nathanw   600: #ifdef LOCKDEBUG
        !           601:        if (flags & PR_WAITOK)
        !           602:                simple_lock_only_held(NULL, "pool_get(PR_WAITOK)");
        !           603: #endif
        !           604: #endif /* DIAGNOSTIC */
        !           605:
1.21      thorpej   606:        simple_lock(&pp->pr_slock);
1.25      thorpej   607:        pr_enter(pp, file, line);
1.20      thorpej   608:
                    609:  startover:
                    610:        /*
                    611:         * Check to see if we've reached the hard limit.  If we have,
                    612:         * and we can wait, then wait until an item has been returned to
                    613:         * the pool.
                    614:         */
                    615: #ifdef DIAGNOSTIC
1.34      thorpej   616:        if (__predict_false(pp->pr_nout > pp->pr_hardlimit)) {
1.25      thorpej   617:                pr_leave(pp);
1.21      thorpej   618:                simple_unlock(&pp->pr_slock);
1.20      thorpej   619:                panic("pool_get: %s: crossed hard limit", pp->pr_wchan);
                    620:        }
                    621: #endif
1.34      thorpej   622:        if (__predict_false(pp->pr_nout == pp->pr_hardlimit)) {
1.29      sommerfe  623:                if ((flags & PR_WAITOK) && !(flags & PR_LIMITFAIL)) {
1.20      thorpej   624:                        /*
                    625:                         * XXX: A warning isn't logged in this case.  Should
                    626:                         * it be?
                    627:                         */
                    628:                        pp->pr_flags |= PR_WANTED;
1.25      thorpej   629:                        pr_leave(pp);
1.40      sommerfe  630:                        ltsleep(pp, PSWP, pp->pr_wchan, 0, &pp->pr_slock);
1.25      thorpej   631:                        pr_enter(pp, file, line);
1.20      thorpej   632:                        goto startover;
                    633:                }
1.31      thorpej   634:
                    635:                /*
                    636:                 * Log a message that the hard limit has been hit.
                    637:                 */
                    638:                if (pp->pr_hardlimit_warning != NULL &&
                    639:                    ratecheck(&pp->pr_hardlimit_warning_last,
                    640:                              &pp->pr_hardlimit_ratecap))
                    641:                        log(LOG_ERR, "%s\n", pp->pr_hardlimit_warning);
1.21      thorpej   642:
                    643:                if (flags & PR_URGENT)
                    644:                        panic("pool_get: urgent");
                    645:
                    646:                pp->pr_nfail++;
                    647:
1.25      thorpej   648:                pr_leave(pp);
1.21      thorpej   649:                simple_unlock(&pp->pr_slock);
1.20      thorpej   650:                return (NULL);
                    651:        }
                    652:
1.3       pk        653:        /*
                    654:         * The convention we use is that if `curpage' is not NULL, then
                    655:         * it points at a non-empty bucket. In particular, `curpage'
                    656:         * never points at a page header which has PR_PHINPAGE set and
                    657:         * has no items in its bucket.
                    658:         */
1.20      thorpej   659:        if ((ph = pp->pr_curpage) == NULL) {
                    660: #ifdef DIAGNOSTIC
                    661:                if (pp->pr_nitems != 0) {
1.21      thorpej   662:                        simple_unlock(&pp->pr_slock);
1.20      thorpej   663:                        printf("pool_get: %s: curpage NULL, nitems %u\n",
                    664:                            pp->pr_wchan, pp->pr_nitems);
                    665:                        panic("pool_get: nitems inconsistent\n");
                    666:                }
                    667: #endif
                    668:
1.21      thorpej   669:                /*
                    670:                 * Call the back-end page allocator for more memory.
                    671:                 * Release the pool lock, as the back-end page allocator
                    672:                 * may block.
                    673:                 */
1.25      thorpej   674:                pr_leave(pp);
1.21      thorpej   675:                simple_unlock(&pp->pr_slock);
                    676:                v = (*pp->pr_alloc)(pp->pr_pagesz, flags, pp->pr_mtype);
1.50.2.1! nathanw   677:                if (__predict_true(v != NULL))
        !           678:                        ph = pool_alloc_item_header(pp, v, flags);
1.21      thorpej   679:                simple_lock(&pp->pr_slock);
1.25      thorpej   680:                pr_enter(pp, file, line);
1.15      pk        681:
1.50.2.1! nathanw   682:                if (__predict_false(v == NULL || ph == NULL)) {
        !           683:                        if (v != NULL)
        !           684:                                (*pp->pr_free)(v, pp->pr_pagesz, pp->pr_mtype);
        !           685:
1.21      thorpej   686:                        /*
1.50.2.1! nathanw   687:                         * We were unable to allocate a page or item
        !           688:                         * header, but we released the lock during
        !           689:                         * allocation, so perhaps items were freed
        !           690:                         * back to the pool.  Check for this case.
1.21      thorpej   691:                         */
                    692:                        if (pp->pr_curpage != NULL)
                    693:                                goto startover;
1.15      pk        694:
1.3       pk        695:                        if (flags & PR_URGENT)
                    696:                                panic("pool_get: urgent");
1.21      thorpej   697:
1.3       pk        698:                        if ((flags & PR_WAITOK) == 0) {
                    699:                                pp->pr_nfail++;
1.25      thorpej   700:                                pr_leave(pp);
1.21      thorpej   701:                                simple_unlock(&pp->pr_slock);
1.1       pk        702:                                return (NULL);
1.3       pk        703:                        }
                    704:
1.15      pk        705:                        /*
                    706:                         * Wait for items to be returned to this pool.
1.21      thorpej   707:                         *
1.15      pk        708:                         * XXX: we actually want to wait just until
                    709:                         * the page allocator has memory again. Depending
                    710:                         * on this pool's usage, we might get stuck here
                    711:                         * for a long time.
1.20      thorpej   712:                         *
                    713:                         * XXX: maybe we should wake up once a second and
                    714:                         * try again?
1.15      pk        715:                         */
1.1       pk        716:                        pp->pr_flags |= PR_WANTED;
1.25      thorpej   717:                        pr_leave(pp);
1.40      sommerfe  718:                        ltsleep(pp, PSWP, pp->pr_wchan, 0, &pp->pr_slock);
1.25      thorpej   719:                        pr_enter(pp, file, line);
1.20      thorpej   720:                        goto startover;
1.1       pk        721:                }
1.3       pk        722:
1.15      pk        723:                /* We have more memory; add it to the pool */
1.50.2.1! nathanw   724:                pool_prime_page(pp, v, ph);
1.15      pk        725:                pp->pr_npagealloc++;
                    726:
1.20      thorpej   727:                /* Start the allocation process over. */
                    728:                goto startover;
1.3       pk        729:        }
                    730:
1.34      thorpej   731:        if (__predict_false((v = pi = TAILQ_FIRST(&ph->ph_itemlist)) == NULL)) {
1.25      thorpej   732:                pr_leave(pp);
1.21      thorpej   733:                simple_unlock(&pp->pr_slock);
1.3       pk        734:                panic("pool_get: %s: page empty", pp->pr_wchan);
1.21      thorpej   735:        }
1.20      thorpej   736: #ifdef DIAGNOSTIC
1.34      thorpej   737:        if (__predict_false(pp->pr_nitems == 0)) {
1.25      thorpej   738:                pr_leave(pp);
1.21      thorpej   739:                simple_unlock(&pp->pr_slock);
1.20      thorpej   740:                printf("pool_get: %s: items on itemlist, nitems %u\n",
                    741:                    pp->pr_wchan, pp->pr_nitems);
                    742:                panic("pool_get: nitems inconsistent\n");
                    743:        }
1.50.2.1! nathanw   744:
1.3       pk        745:        pr_log(pp, v, PRLOG_GET, file, line);
                    746:
1.34      thorpej   747:        if (__predict_false(pi->pi_magic != PI_MAGIC)) {
1.25      thorpej   748:                pr_printlog(pp, pi, printf);
1.3       pk        749:                panic("pool_get(%s): free list modified: magic=%x; page %p;"
                    750:                       " item addr %p\n",
                    751:                        pp->pr_wchan, pi->pi_magic, ph->ph_page, pi);
                    752:        }
                    753: #endif
                    754:
                    755:        /*
                    756:         * Remove from item list.
                    757:         */
                    758:        TAILQ_REMOVE(&ph->ph_itemlist, pi, pi_list);
1.20      thorpej   759:        pp->pr_nitems--;
                    760:        pp->pr_nout++;
1.6       thorpej   761:        if (ph->ph_nmissing == 0) {
                    762: #ifdef DIAGNOSTIC
1.34      thorpej   763:                if (__predict_false(pp->pr_nidle == 0))
1.6       thorpej   764:                        panic("pool_get: nidle inconsistent");
                    765: #endif
                    766:                pp->pr_nidle--;
                    767:        }
1.3       pk        768:        ph->ph_nmissing++;
                    769:        if (TAILQ_FIRST(&ph->ph_itemlist) == NULL) {
1.21      thorpej   770: #ifdef DIAGNOSTIC
1.34      thorpej   771:                if (__predict_false(ph->ph_nmissing != pp->pr_itemsperpage)) {
1.25      thorpej   772:                        pr_leave(pp);
1.21      thorpej   773:                        simple_unlock(&pp->pr_slock);
                    774:                        panic("pool_get: %s: nmissing inconsistent",
                    775:                            pp->pr_wchan);
                    776:                }
                    777: #endif
1.3       pk        778:                /*
                    779:                 * Find a new non-empty page header, if any.
                    780:                 * Start search from the page head, to increase
                    781:                 * the chance for "high water" pages to be freed.
                    782:                 *
1.21      thorpej   783:                 * Migrate empty pages to the end of the list.  This
                    784:                 * will speed the update of curpage as pages become
                    785:                 * idle.  Empty pages intermingled with idle pages
                    786:                 * is no big deal.  As soon as a page becomes un-empty,
                    787:                 * it will move back to the head of the list.
1.3       pk        788:                 */
                    789:                TAILQ_REMOVE(&pp->pr_pagelist, ph, ph_pagelist);
1.21      thorpej   790:                TAILQ_INSERT_TAIL(&pp->pr_pagelist, ph, ph_pagelist);
                    791:                for (ph = TAILQ_FIRST(&pp->pr_pagelist); ph != NULL;
                    792:                     ph = TAILQ_NEXT(ph, ph_pagelist))
1.3       pk        793:                        if (TAILQ_FIRST(&ph->ph_itemlist) != NULL)
                    794:                                break;
                    795:
                    796:                pp->pr_curpage = ph;
1.1       pk        797:        }
1.3       pk        798:
                    799:        pp->pr_nget++;
1.20      thorpej   800:
                    801:        /*
                    802:         * If we have a low water mark and we are now below that low
                    803:         * water mark, add more items to the pool.
                    804:         */
1.50.2.1! nathanw   805:        if (POOL_NEEDS_CATCHUP(pp) && pool_catchup(pp) != 0) {
1.20      thorpej   806:                /*
                    807:                 * XXX: Should we log a warning?  Should we set up a timeout
                    808:                 * to try again in a second or so?  The latter could break
                    809:                 * a caller's assumptions about interrupt protection, etc.
                    810:                 */
                    811:        }
                    812:
1.25      thorpej   813:        pr_leave(pp);
1.21      thorpej   814:        simple_unlock(&pp->pr_slock);
1.1       pk        815:        return (v);
                    816: }
                    817:
                    818: /*
1.43      thorpej   819:  * Internal version of pool_put().  Pool is already locked/entered.
1.1       pk        820:  */
1.43      thorpej   821: static void
1.50.2.1! nathanw   822: pool_do_put(struct pool *pp, void *v)
1.1       pk        823: {
                    824:        struct pool_item *pi = v;
1.3       pk        825:        struct pool_item_header *ph;
                    826:        caddr_t page;
1.21      thorpej   827:        int s;
1.3       pk        828:
                    829:        page = (caddr_t)((u_long)v & pp->pr_pagemask);
1.1       pk        830:
1.30      thorpej   831: #ifdef DIAGNOSTIC
1.34      thorpej   832:        if (__predict_false(pp->pr_nout == 0)) {
1.30      thorpej   833:                printf("pool %s: putting with none out\n",
                    834:                    pp->pr_wchan);
                    835:                panic("pool_put");
                    836:        }
                    837: #endif
1.3       pk        838:
1.34      thorpej   839:        if (__predict_false((ph = pr_find_pagehead(pp, page)) == NULL)) {
1.25      thorpej   840:                pr_printlog(pp, NULL, printf);
1.3       pk        841:                panic("pool_put: %s: page header missing", pp->pr_wchan);
                    842:        }
1.28      thorpej   843:
                    844: #ifdef LOCKDEBUG
                    845:        /*
                    846:         * Check if we're freeing a locked simple lock.
                    847:         */
                    848:        simple_lock_freecheck((caddr_t)pi, ((caddr_t)pi) + pp->pr_size);
                    849: #endif
1.3       pk        850:
                    851:        /*
                    852:         * Return to item list.
                    853:         */
1.2       pk        854: #ifdef DIAGNOSTIC
1.3       pk        855:        pi->pi_magic = PI_MAGIC;
                    856: #endif
1.32      chs       857: #ifdef DEBUG
                    858:        {
                    859:                int i, *ip = v;
                    860:
                    861:                for (i = 0; i < pp->pr_size / sizeof(int); i++) {
                    862:                        *ip++ = PI_MAGIC;
                    863:                }
                    864:        }
                    865: #endif
                    866:
1.3       pk        867:        TAILQ_INSERT_HEAD(&ph->ph_itemlist, pi, pi_list);
                    868:        ph->ph_nmissing--;
                    869:        pp->pr_nput++;
1.20      thorpej   870:        pp->pr_nitems++;
                    871:        pp->pr_nout--;
1.3       pk        872:
                    873:        /* Cancel "pool empty" condition if it exists */
                    874:        if (pp->pr_curpage == NULL)
                    875:                pp->pr_curpage = ph;
                    876:
                    877:        if (pp->pr_flags & PR_WANTED) {
                    878:                pp->pr_flags &= ~PR_WANTED;
1.15      pk        879:                if (ph->ph_nmissing == 0)
                    880:                        pp->pr_nidle++;
1.3       pk        881:                wakeup((caddr_t)pp);
                    882:                return;
                    883:        }
                    884:
                    885:        /*
1.21      thorpej   886:         * If this page is now complete, do one of two things:
                    887:         *
                    888:         *      (1) If we have more pages than the page high water
                    889:         *          mark, free the page back to the system.
                    890:         *
                    891:         *      (2) Move it to the end of the page list, so that
                    892:         *          we minimize our chances of fragmenting the
                    893:         *          pool.  Idle pages migrate to the end (along with
                    894:         *          completely empty pages, so that we find un-empty
                    895:         *          pages more quickly when we update curpage) of the
                    896:         *          list so they can be more easily swept up by
                    897:         *          the pagedaemon when pages are scarce.
1.3       pk        898:         */
                    899:        if (ph->ph_nmissing == 0) {
1.6       thorpej   900:                pp->pr_nidle++;
1.3       pk        901:                if (pp->pr_npages > pp->pr_maxpages) {
                    902:                        pr_rmpage(pp, ph);
                    903:                } else {
                    904:                        TAILQ_REMOVE(&pp->pr_pagelist, ph, ph_pagelist);
                    905:                        TAILQ_INSERT_TAIL(&pp->pr_pagelist, ph, ph_pagelist);
                    906:
1.21      thorpej   907:                        /*
                    908:                         * Update the timestamp on the page.  A page must
                    909:                         * be idle for some period of time before it can
                    910:                         * be reclaimed by the pagedaemon.  This minimizes
                    911:                         * ping-pong'ing for memory.
                    912:                         */
                    913:                        s = splclock();
                    914:                        ph->ph_time = mono_time;
                    915:                        splx(s);
                    916:
                    917:                        /*
                    918:                         * Update the current page pointer.  Just look for
                    919:                         * the first page with any free items.
                    920:                         *
                    921:                         * XXX: Maybe we want an option to look for the
                    922:                         * page with the fewest available items, to minimize
                    923:                         * fragmentation?
                    924:                         */
1.3       pk        925:                        for (ph = TAILQ_FIRST(&pp->pr_pagelist); ph != NULL;
                    926:                             ph = TAILQ_NEXT(ph, ph_pagelist))
                    927:                                if (TAILQ_FIRST(&ph->ph_itemlist) != NULL)
                    928:                                        break;
1.1       pk        929:
1.3       pk        930:                        pp->pr_curpage = ph;
1.1       pk        931:                }
                    932:        }
1.21      thorpej   933:        /*
                    934:         * If the page has just become un-empty, move it to the head of
                    935:         * the list, and make it the current page.  The next allocation
                    936:         * will get the item from this page, instead of further fragmenting
                    937:         * the pool.
                    938:         */
                    939:        else if (ph->ph_nmissing == (pp->pr_itemsperpage - 1)) {
                    940:                TAILQ_REMOVE(&pp->pr_pagelist, ph, ph_pagelist);
                    941:                TAILQ_INSERT_HEAD(&pp->pr_pagelist, ph, ph_pagelist);
                    942:                pp->pr_curpage = ph;
                    943:        }
1.43      thorpej   944: }
                    945:
                    946: /*
                    947:  * Return resource to the pool; must be called at appropriate spl level
                    948:  */
1.50.2.1! nathanw   949: #ifdef POOL_DIAGNOSTIC
1.43      thorpej   950: void
                    951: _pool_put(struct pool *pp, void *v, const char *file, long line)
                    952: {
                    953:
                    954:        simple_lock(&pp->pr_slock);
                    955:        pr_enter(pp, file, line);
                    956:
1.50.2.1! nathanw   957:        pr_log(pp, v, PRLOG_PUT, file, line);
        !           958:
        !           959:        pool_do_put(pp, v);
1.21      thorpej   960:
1.25      thorpej   961:        pr_leave(pp);
1.21      thorpej   962:        simple_unlock(&pp->pr_slock);
1.1       pk        963: }
1.50.2.1! nathanw   964: #undef pool_put
        !           965: #endif /* POOL_DIAGNOSTIC */
        !           966:
        !           967: void
        !           968: pool_put(struct pool *pp, void *v)
        !           969: {
        !           970:
        !           971:        simple_lock(&pp->pr_slock);
        !           972:
        !           973:        pool_do_put(pp, v);
        !           974:
        !           975:        simple_unlock(&pp->pr_slock);
        !           976: }
        !           977:
        !           978: #ifdef POOL_DIAGNOSTIC
        !           979: #define                pool_put(h, v)  _pool_put((h), (v), __FILE__, __LINE__)
        !           980: #endif
1.1       pk        981:
                    982: /*
1.3       pk        983:  * Add N items to the pool.
1.1       pk        984:  */
                    985: int
1.50.2.1! nathanw   986: pool_prime(struct pool *pp, int n)
1.1       pk        987: {
1.50.2.1! nathanw   988:        struct pool_item_header *ph;
1.3       pk        989:        caddr_t cp;
1.50.2.1! nathanw   990:        int newpages, error = 0;
1.1       pk        991:
1.21      thorpej   992:        simple_lock(&pp->pr_slock);
                    993:
1.50.2.1! nathanw   994:        newpages = roundup(n, pp->pr_itemsperpage) / pp->pr_itemsperpage;
1.3       pk        995:
                    996:        while (newpages-- > 0) {
1.50.2.1! nathanw   997:                simple_unlock(&pp->pr_slock);
        !           998:                cp = (*pp->pr_alloc)(pp->pr_pagesz, PR_NOWAIT, pp->pr_mtype);
        !           999:                if (__predict_true(cp != NULL))
        !          1000:                        ph = pool_alloc_item_header(pp, cp, PR_NOWAIT);
        !          1001:                simple_lock(&pp->pr_slock);
1.2       pk       1002:
1.50.2.1! nathanw  1003:                if (__predict_false(cp == NULL || ph == NULL)) {
        !          1004:                        error = ENOMEM;
        !          1005:                        if (cp != NULL)
        !          1006:                                (*pp->pr_free)(cp, pp->pr_pagesz, pp->pr_mtype);
        !          1007:                        break;
1.1       pk       1008:                }
                   1009:
1.50.2.1! nathanw  1010:                pool_prime_page(pp, cp, ph);
1.26      thorpej  1011:                pp->pr_npagealloc++;
1.3       pk       1012:                pp->pr_minpages++;
1.1       pk       1013:        }
1.3       pk       1014:
                   1015:        if (pp->pr_minpages >= pp->pr_maxpages)
                   1016:                pp->pr_maxpages = pp->pr_minpages + 1;  /* XXX */
                   1017:
1.21      thorpej  1018:        simple_unlock(&pp->pr_slock);
1.1       pk       1019:        return (0);
                   1020: }
1.3       pk       1021:
                   1022: /*
                   1023:  * Add a page worth of items to the pool.
1.21      thorpej  1024:  *
                   1025:  * Note, we must be called with the pool descriptor LOCKED.
1.3       pk       1026:  */
1.50.2.1! nathanw  1027: static void
        !          1028: pool_prime_page(struct pool *pp, caddr_t storage, struct pool_item_header *ph)
1.3       pk       1029: {
                   1030:        struct pool_item *pi;
                   1031:        caddr_t cp = storage;
                   1032:        unsigned int align = pp->pr_align;
                   1033:        unsigned int ioff = pp->pr_itemoffset;
1.50.2.1! nathanw  1034:        int n;
1.36      pk       1035:
                   1036:        if (((u_long)cp & (pp->pr_pagesz - 1)) != 0)
                   1037:                panic("pool_prime_page: %s: unaligned page", pp->pr_wchan);
1.3       pk       1038:
1.50.2.1! nathanw  1039:        if ((pp->pr_roflags & PR_PHINPAGE) == 0)
1.3       pk       1040:                LIST_INSERT_HEAD(&pp->pr_hashtab[PR_HASH_INDEX(pp, cp)],
1.50.2.1! nathanw  1041:                    ph, ph_hashlist);
1.3       pk       1042:
                   1043:        /*
                   1044:         * Insert page header.
                   1045:         */
                   1046:        TAILQ_INSERT_HEAD(&pp->pr_pagelist, ph, ph_pagelist);
                   1047:        TAILQ_INIT(&ph->ph_itemlist);
                   1048:        ph->ph_page = storage;
                   1049:        ph->ph_nmissing = 0;
1.21      thorpej  1050:        memset(&ph->ph_time, 0, sizeof(ph->ph_time));
1.3       pk       1051:
1.6       thorpej  1052:        pp->pr_nidle++;
                   1053:
1.3       pk       1054:        /*
                   1055:         * Color this page.
                   1056:         */
                   1057:        cp = (caddr_t)(cp + pp->pr_curcolor);
                   1058:        if ((pp->pr_curcolor += align) > pp->pr_maxcolor)
                   1059:                pp->pr_curcolor = 0;
                   1060:
                   1061:        /*
                   1062:         * Adjust storage to apply aligment to `pr_itemoffset' in each item.
                   1063:         */
                   1064:        if (ioff != 0)
                   1065:                cp = (caddr_t)(cp + (align - ioff));
                   1066:
                   1067:        /*
                   1068:         * Insert remaining chunks on the bucket list.
                   1069:         */
                   1070:        n = pp->pr_itemsperpage;
1.20      thorpej  1071:        pp->pr_nitems += n;
1.3       pk       1072:
                   1073:        while (n--) {
                   1074:                pi = (struct pool_item *)cp;
                   1075:
                   1076:                /* Insert on page list */
                   1077:                TAILQ_INSERT_TAIL(&ph->ph_itemlist, pi, pi_list);
                   1078: #ifdef DIAGNOSTIC
                   1079:                pi->pi_magic = PI_MAGIC;
                   1080: #endif
                   1081:                cp = (caddr_t)(cp + pp->pr_size);
                   1082:        }
                   1083:
                   1084:        /*
                   1085:         * If the pool was depleted, point at the new page.
                   1086:         */
                   1087:        if (pp->pr_curpage == NULL)
                   1088:                pp->pr_curpage = ph;
                   1089:
                   1090:        if (++pp->pr_npages > pp->pr_hiwat)
                   1091:                pp->pr_hiwat = pp->pr_npages;
                   1092: }
                   1093:
1.20      thorpej  1094: /*
1.50.2.1! nathanw  1095:  * Used by pool_get() when nitems drops below the low water mark.  This
        !          1096:  * is used to catch up nitmes with the low water mark.
1.20      thorpej  1097:  *
1.21      thorpej  1098:  * Note 1, we never wait for memory here, we let the caller decide what to do.
1.20      thorpej  1099:  *
                   1100:  * Note 2, this doesn't work with static pools.
                   1101:  *
                   1102:  * Note 3, we must be called with the pool already locked, and we return
                   1103:  * with it locked.
                   1104:  */
                   1105: static int
1.42      thorpej  1106: pool_catchup(struct pool *pp)
1.20      thorpej  1107: {
1.50.2.1! nathanw  1108:        struct pool_item_header *ph;
1.20      thorpej  1109:        caddr_t cp;
                   1110:        int error = 0;
                   1111:
                   1112:        if (pp->pr_roflags & PR_STATIC) {
                   1113:                /*
                   1114:                 * We dropped below the low water mark, and this is not a
                   1115:                 * good thing.  Log a warning.
1.21      thorpej  1116:                 *
                   1117:                 * XXX: rate-limit this?
1.20      thorpej  1118:                 */
                   1119:                printf("WARNING: static pool `%s' dropped below low water "
                   1120:                    "mark\n", pp->pr_wchan);
                   1121:                return (0);
                   1122:        }
                   1123:
1.50.2.1! nathanw  1124:        while (POOL_NEEDS_CATCHUP(pp)) {
1.20      thorpej  1125:                /*
1.21      thorpej  1126:                 * Call the page back-end allocator for more memory.
                   1127:                 *
                   1128:                 * XXX: We never wait, so should we bother unlocking
                   1129:                 * the pool descriptor?
1.20      thorpej  1130:                 */
1.21      thorpej  1131:                simple_unlock(&pp->pr_slock);
1.50.2.1! nathanw  1132:                cp = (*pp->pr_alloc)(pp->pr_pagesz, PR_NOWAIT, pp->pr_mtype);
        !          1133:                if (__predict_true(cp != NULL))
        !          1134:                        ph = pool_alloc_item_header(pp, cp, PR_NOWAIT);
1.21      thorpej  1135:                simple_lock(&pp->pr_slock);
1.50.2.1! nathanw  1136:                if (__predict_false(cp == NULL || ph == NULL)) {
        !          1137:                        if (cp != NULL)
        !          1138:                                (*pp->pr_free)(cp, pp->pr_pagesz, pp->pr_mtype);
1.20      thorpej  1139:                        error = ENOMEM;
                   1140:                        break;
                   1141:                }
1.50.2.1! nathanw  1142:                pool_prime_page(pp, cp, ph);
1.26      thorpej  1143:                pp->pr_npagealloc++;
1.20      thorpej  1144:        }
                   1145:
                   1146:        return (error);
                   1147: }
                   1148:
1.3       pk       1149: void
1.42      thorpej  1150: pool_setlowat(struct pool *pp, int n)
1.3       pk       1151: {
1.20      thorpej  1152:        int error;
1.15      pk       1153:
1.21      thorpej  1154:        simple_lock(&pp->pr_slock);
                   1155:
1.3       pk       1156:        pp->pr_minitems = n;
1.15      pk       1157:        pp->pr_minpages = (n == 0)
                   1158:                ? 0
1.18      thorpej  1159:                : roundup(n, pp->pr_itemsperpage) / pp->pr_itemsperpage;
1.20      thorpej  1160:
                   1161:        /* Make sure we're caught up with the newly-set low water mark. */
1.50.2.1! nathanw  1162:        if (POOL_NEEDS_CATCHUP(pp) && (error = pool_catchup(pp) != 0)) {
1.20      thorpej  1163:                /*
                   1164:                 * XXX: Should we log a warning?  Should we set up a timeout
                   1165:                 * to try again in a second or so?  The latter could break
                   1166:                 * a caller's assumptions about interrupt protection, etc.
                   1167:                 */
                   1168:        }
1.21      thorpej  1169:
                   1170:        simple_unlock(&pp->pr_slock);
1.3       pk       1171: }
                   1172:
                   1173: void
1.42      thorpej  1174: pool_sethiwat(struct pool *pp, int n)
1.3       pk       1175: {
1.15      pk       1176:
1.21      thorpej  1177:        simple_lock(&pp->pr_slock);
                   1178:
1.15      pk       1179:        pp->pr_maxpages = (n == 0)
                   1180:                ? 0
1.18      thorpej  1181:                : roundup(n, pp->pr_itemsperpage) / pp->pr_itemsperpage;
1.21      thorpej  1182:
                   1183:        simple_unlock(&pp->pr_slock);
1.3       pk       1184: }
                   1185:
1.20      thorpej  1186: void
1.42      thorpej  1187: pool_sethardlimit(struct pool *pp, int n, const char *warnmess, int ratecap)
1.20      thorpej  1188: {
                   1189:
1.21      thorpej  1190:        simple_lock(&pp->pr_slock);
1.20      thorpej  1191:
                   1192:        pp->pr_hardlimit = n;
                   1193:        pp->pr_hardlimit_warning = warnmess;
1.31      thorpej  1194:        pp->pr_hardlimit_ratecap.tv_sec = ratecap;
                   1195:        pp->pr_hardlimit_warning_last.tv_sec = 0;
                   1196:        pp->pr_hardlimit_warning_last.tv_usec = 0;
1.20      thorpej  1197:
                   1198:        /*
1.21      thorpej  1199:         * In-line version of pool_sethiwat(), because we don't want to
                   1200:         * release the lock.
1.20      thorpej  1201:         */
                   1202:        pp->pr_maxpages = (n == 0)
                   1203:                ? 0
                   1204:                : roundup(n, pp->pr_itemsperpage) / pp->pr_itemsperpage;
1.21      thorpej  1205:
                   1206:        simple_unlock(&pp->pr_slock);
1.20      thorpej  1207: }
1.3       pk       1208:
                   1209: /*
                   1210:  * Default page allocator.
                   1211:  */
                   1212: static void *
1.42      thorpej  1213: pool_page_alloc(unsigned long sz, int flags, int mtype)
1.3       pk       1214: {
1.11      thorpej  1215:        boolean_t waitok = (flags & PR_WAITOK) ? TRUE : FALSE;
1.3       pk       1216:
1.11      thorpej  1217:        return ((void *)uvm_km_alloc_poolpage(waitok));
1.3       pk       1218: }
                   1219:
                   1220: static void
1.42      thorpej  1221: pool_page_free(void *v, unsigned long sz, int mtype)
1.3       pk       1222: {
                   1223:
1.10      eeh      1224:        uvm_km_free_poolpage((vaddr_t)v);
1.3       pk       1225: }
1.12      thorpej  1226:
                   1227: /*
                   1228:  * Alternate pool page allocator for pools that know they will
                   1229:  * never be accessed in interrupt context.
                   1230:  */
                   1231: void *
1.42      thorpej  1232: pool_page_alloc_nointr(unsigned long sz, int flags, int mtype)
1.12      thorpej  1233: {
                   1234:        boolean_t waitok = (flags & PR_WAITOK) ? TRUE : FALSE;
                   1235:
                   1236:        return ((void *)uvm_km_alloc_poolpage1(kernel_map, uvm.kernel_object,
                   1237:            waitok));
                   1238: }
                   1239:
                   1240: void
1.42      thorpej  1241: pool_page_free_nointr(void *v, unsigned long sz, int mtype)
1.12      thorpej  1242: {
                   1243:
                   1244:        uvm_km_free_poolpage1(kernel_map, (vaddr_t)v);
                   1245: }
                   1246:
1.3       pk       1247:
                   1248: /*
                   1249:  * Release all complete pages that have not been used recently.
                   1250:  */
                   1251: void
1.50.2.1! nathanw  1252: #ifdef POOL_DIAGNOSTIC
1.42      thorpej  1253: _pool_reclaim(struct pool *pp, const char *file, long line)
1.50.2.1! nathanw  1254: #else
        !          1255: pool_reclaim(struct pool *pp)
        !          1256: #endif
1.3       pk       1257: {
                   1258:        struct pool_item_header *ph, *phnext;
1.43      thorpej  1259:        struct pool_cache *pc;
1.21      thorpej  1260:        struct timeval curtime;
                   1261:        int s;
1.3       pk       1262:
1.20      thorpej  1263:        if (pp->pr_roflags & PR_STATIC)
1.3       pk       1264:                return;
                   1265:
1.21      thorpej  1266:        if (simple_lock_try(&pp->pr_slock) == 0)
1.3       pk       1267:                return;
1.25      thorpej  1268:        pr_enter(pp, file, line);
1.3       pk       1269:
1.43      thorpej  1270:        /*
                   1271:         * Reclaim items from the pool's caches.
                   1272:         */
                   1273:        for (pc = TAILQ_FIRST(&pp->pr_cachelist); pc != NULL;
                   1274:             pc = TAILQ_NEXT(pc, pc_poollist))
                   1275:                pool_cache_reclaim(pc);
                   1276:
1.21      thorpej  1277:        s = splclock();
                   1278:        curtime = mono_time;
                   1279:        splx(s);
                   1280:
1.3       pk       1281:        for (ph = TAILQ_FIRST(&pp->pr_pagelist); ph != NULL; ph = phnext) {
                   1282:                phnext = TAILQ_NEXT(ph, ph_pagelist);
                   1283:
                   1284:                /* Check our minimum page claim */
                   1285:                if (pp->pr_npages <= pp->pr_minpages)
                   1286:                        break;
                   1287:
                   1288:                if (ph->ph_nmissing == 0) {
                   1289:                        struct timeval diff;
                   1290:                        timersub(&curtime, &ph->ph_time, &diff);
                   1291:                        if (diff.tv_sec < pool_inactive_time)
                   1292:                                continue;
1.21      thorpej  1293:
                   1294:                        /*
                   1295:                         * If freeing this page would put us below
                   1296:                         * the low water mark, stop now.
                   1297:                         */
                   1298:                        if ((pp->pr_nitems - pp->pr_itemsperpage) <
                   1299:                            pp->pr_minitems)
                   1300:                                break;
                   1301:
1.3       pk       1302:                        pr_rmpage(pp, ph);
                   1303:                }
                   1304:        }
                   1305:
1.25      thorpej  1306:        pr_leave(pp);
1.21      thorpej  1307:        simple_unlock(&pp->pr_slock);
1.3       pk       1308: }
                   1309:
                   1310:
                   1311: /*
                   1312:  * Drain pools, one at a time.
1.21      thorpej  1313:  *
                   1314:  * Note, we must never be called from an interrupt context.
1.3       pk       1315:  */
                   1316: void
1.42      thorpej  1317: pool_drain(void *arg)
1.3       pk       1318: {
                   1319:        struct pool *pp;
1.23      thorpej  1320:        int s;
1.3       pk       1321:
1.49      thorpej  1322:        s = splvm();
1.23      thorpej  1323:        simple_lock(&pool_head_slock);
                   1324:
                   1325:        if (drainpp == NULL && (drainpp = TAILQ_FIRST(&pool_head)) == NULL)
                   1326:                goto out;
1.3       pk       1327:
                   1328:        pp = drainpp;
                   1329:        drainpp = TAILQ_NEXT(pp, pr_poollist);
                   1330:
                   1331:        pool_reclaim(pp);
1.23      thorpej  1332:
                   1333:  out:
                   1334:        simple_unlock(&pool_head_slock);
1.3       pk       1335:        splx(s);
                   1336: }
                   1337:
                   1338:
                   1339: /*
                   1340:  * Diagnostic helpers.
                   1341:  */
                   1342: void
1.42      thorpej  1343: pool_print(struct pool *pp, const char *modif)
1.21      thorpej  1344: {
                   1345:        int s;
                   1346:
1.49      thorpej  1347:        s = splvm();
1.25      thorpej  1348:        if (simple_lock_try(&pp->pr_slock) == 0) {
                   1349:                printf("pool %s is locked; try again later\n",
                   1350:                    pp->pr_wchan);
                   1351:                splx(s);
                   1352:                return;
                   1353:        }
                   1354:        pool_print1(pp, modif, printf);
1.21      thorpej  1355:        simple_unlock(&pp->pr_slock);
                   1356:        splx(s);
                   1357: }
                   1358:
1.25      thorpej  1359: void
1.42      thorpej  1360: pool_printit(struct pool *pp, const char *modif, void (*pr)(const char *, ...))
1.25      thorpej  1361: {
                   1362:        int didlock = 0;
                   1363:
                   1364:        if (pp == NULL) {
                   1365:                (*pr)("Must specify a pool to print.\n");
                   1366:                return;
                   1367:        }
                   1368:
                   1369:        /*
                   1370:         * Called from DDB; interrupts should be blocked, and all
                   1371:         * other processors should be paused.  We can skip locking
                   1372:         * the pool in this case.
                   1373:         *
                   1374:         * We do a simple_lock_try() just to print the lock
                   1375:         * status, however.
                   1376:         */
                   1377:
                   1378:        if (simple_lock_try(&pp->pr_slock) == 0)
                   1379:                (*pr)("WARNING: pool %s is locked\n", pp->pr_wchan);
                   1380:        else
                   1381:                didlock = 1;
                   1382:
                   1383:        pool_print1(pp, modif, pr);
                   1384:
                   1385:        if (didlock)
                   1386:                simple_unlock(&pp->pr_slock);
                   1387: }
                   1388:
1.21      thorpej  1389: static void
1.42      thorpej  1390: pool_print1(struct pool *pp, const char *modif, void (*pr)(const char *, ...))
1.3       pk       1391: {
1.25      thorpej  1392:        struct pool_item_header *ph;
1.44      thorpej  1393:        struct pool_cache *pc;
                   1394:        struct pool_cache_group *pcg;
1.25      thorpej  1395: #ifdef DIAGNOSTIC
                   1396:        struct pool_item *pi;
                   1397: #endif
1.44      thorpej  1398:        int i, print_log = 0, print_pagelist = 0, print_cache = 0;
1.25      thorpej  1399:        char c;
                   1400:
                   1401:        while ((c = *modif++) != '\0') {
                   1402:                if (c == 'l')
                   1403:                        print_log = 1;
                   1404:                if (c == 'p')
                   1405:                        print_pagelist = 1;
1.44      thorpej  1406:                if (c == 'c')
                   1407:                        print_cache = 1;
1.25      thorpej  1408:                modif++;
                   1409:        }
                   1410:
                   1411:        (*pr)("POOL %s: size %u, align %u, ioff %u, roflags 0x%08x\n",
                   1412:            pp->pr_wchan, pp->pr_size, pp->pr_align, pp->pr_itemoffset,
                   1413:            pp->pr_roflags);
                   1414:        (*pr)("\tpagesz %u, mtype %d\n", pp->pr_pagesz, pp->pr_mtype);
                   1415:        (*pr)("\talloc %p, release %p\n", pp->pr_alloc, pp->pr_free);
                   1416:        (*pr)("\tminitems %u, minpages %u, maxpages %u, npages %u\n",
                   1417:            pp->pr_minitems, pp->pr_minpages, pp->pr_maxpages, pp->pr_npages);
                   1418:        (*pr)("\titemsperpage %u, nitems %u, nout %u, hardlimit %u\n",
                   1419:            pp->pr_itemsperpage, pp->pr_nitems, pp->pr_nout, pp->pr_hardlimit);
                   1420:
                   1421:        (*pr)("\n\tnget %lu, nfail %lu, nput %lu\n",
                   1422:            pp->pr_nget, pp->pr_nfail, pp->pr_nput);
                   1423:        (*pr)("\tnpagealloc %lu, npagefree %lu, hiwat %u, nidle %lu\n",
                   1424:            pp->pr_npagealloc, pp->pr_npagefree, pp->pr_hiwat, pp->pr_nidle);
                   1425:
                   1426:        if (print_pagelist == 0)
                   1427:                goto skip_pagelist;
                   1428:
                   1429:        if ((ph = TAILQ_FIRST(&pp->pr_pagelist)) != NULL)
                   1430:                (*pr)("\n\tpage list:\n");
                   1431:        for (; ph != NULL; ph = TAILQ_NEXT(ph, ph_pagelist)) {
                   1432:                (*pr)("\t\tpage %p, nmissing %d, time %lu,%lu\n",
                   1433:                    ph->ph_page, ph->ph_nmissing,
                   1434:                    (u_long)ph->ph_time.tv_sec,
                   1435:                    (u_long)ph->ph_time.tv_usec);
                   1436: #ifdef DIAGNOSTIC
                   1437:                for (pi = TAILQ_FIRST(&ph->ph_itemlist); pi != NULL;
                   1438:                     pi = TAILQ_NEXT(pi, pi_list)) {
                   1439:                        if (pi->pi_magic != PI_MAGIC) {
                   1440:                                (*pr)("\t\t\titem %p, magic 0x%x\n",
                   1441:                                    pi, pi->pi_magic);
                   1442:                        }
                   1443:                }
                   1444: #endif
                   1445:        }
                   1446:        if (pp->pr_curpage == NULL)
                   1447:                (*pr)("\tno current page\n");
                   1448:        else
                   1449:                (*pr)("\tcurpage %p\n", pp->pr_curpage->ph_page);
                   1450:
                   1451:  skip_pagelist:
                   1452:
                   1453:        if (print_log == 0)
                   1454:                goto skip_log;
                   1455:
                   1456:        (*pr)("\n");
                   1457:        if ((pp->pr_roflags & PR_LOGGING) == 0)
                   1458:                (*pr)("\tno log\n");
                   1459:        else
                   1460:                pr_printlog(pp, NULL, pr);
1.3       pk       1461:
1.25      thorpej  1462:  skip_log:
1.44      thorpej  1463:
                   1464:        if (print_cache == 0)
                   1465:                goto skip_cache;
                   1466:
                   1467:        for (pc = TAILQ_FIRST(&pp->pr_cachelist); pc != NULL;
                   1468:             pc = TAILQ_NEXT(pc, pc_poollist)) {
                   1469:                (*pr)("\tcache %p: allocfrom %p freeto %p\n", pc,
                   1470:                    pc->pc_allocfrom, pc->pc_freeto);
1.48      thorpej  1471:                (*pr)("\t    hits %lu misses %lu ngroups %lu nitems %lu\n",
                   1472:                    pc->pc_hits, pc->pc_misses, pc->pc_ngroups, pc->pc_nitems);
1.44      thorpej  1473:                for (pcg = TAILQ_FIRST(&pc->pc_grouplist); pcg != NULL;
                   1474:                     pcg = TAILQ_NEXT(pcg, pcg_list)) {
                   1475:                        (*pr)("\t\tgroup %p: avail %d\n", pcg, pcg->pcg_avail);
                   1476:                        for (i = 0; i < PCG_NOBJECTS; i++)
                   1477:                                (*pr)("\t\t\t%p\n", pcg->pcg_objects[i]);
                   1478:                }
                   1479:        }
                   1480:
                   1481:  skip_cache:
1.3       pk       1482:
1.25      thorpej  1483:        pr_enter_check(pp, pr);
1.3       pk       1484: }
                   1485:
                   1486: int
1.42      thorpej  1487: pool_chk(struct pool *pp, const char *label)
1.3       pk       1488: {
                   1489:        struct pool_item_header *ph;
                   1490:        int r = 0;
                   1491:
1.21      thorpej  1492:        simple_lock(&pp->pr_slock);
1.3       pk       1493:
                   1494:        for (ph = TAILQ_FIRST(&pp->pr_pagelist); ph != NULL;
                   1495:             ph = TAILQ_NEXT(ph, ph_pagelist)) {
                   1496:
                   1497:                struct pool_item *pi;
                   1498:                int n;
                   1499:                caddr_t page;
                   1500:
                   1501:                page = (caddr_t)((u_long)ph & pp->pr_pagemask);
1.20      thorpej  1502:                if (page != ph->ph_page &&
                   1503:                    (pp->pr_roflags & PR_PHINPAGE) != 0) {
1.3       pk       1504:                        if (label != NULL)
                   1505:                                printf("%s: ", label);
1.16      briggs   1506:                        printf("pool(%p:%s): page inconsistency: page %p;"
                   1507:                               " at page head addr %p (p %p)\n", pp,
1.3       pk       1508:                                pp->pr_wchan, ph->ph_page,
                   1509:                                ph, page);
                   1510:                        r++;
                   1511:                        goto out;
                   1512:                }
                   1513:
                   1514:                for (pi = TAILQ_FIRST(&ph->ph_itemlist), n = 0;
                   1515:                     pi != NULL;
                   1516:                     pi = TAILQ_NEXT(pi,pi_list), n++) {
                   1517:
                   1518: #ifdef DIAGNOSTIC
                   1519:                        if (pi->pi_magic != PI_MAGIC) {
                   1520:                                if (label != NULL)
                   1521:                                        printf("%s: ", label);
                   1522:                                printf("pool(%s): free list modified: magic=%x;"
                   1523:                                       " page %p; item ordinal %d;"
                   1524:                                       " addr %p (p %p)\n",
                   1525:                                        pp->pr_wchan, pi->pi_magic, ph->ph_page,
                   1526:                                        n, pi, page);
                   1527:                                panic("pool");
                   1528:                        }
                   1529: #endif
                   1530:                        page = (caddr_t)((u_long)pi & pp->pr_pagemask);
                   1531:                        if (page == ph->ph_page)
                   1532:                                continue;
                   1533:
                   1534:                        if (label != NULL)
                   1535:                                printf("%s: ", label);
1.16      briggs   1536:                        printf("pool(%p:%s): page inconsistency: page %p;"
                   1537:                               " item ordinal %d; addr %p (p %p)\n", pp,
1.3       pk       1538:                                pp->pr_wchan, ph->ph_page,
                   1539:                                n, pi, page);
                   1540:                        r++;
                   1541:                        goto out;
                   1542:                }
                   1543:        }
                   1544: out:
1.21      thorpej  1545:        simple_unlock(&pp->pr_slock);
1.3       pk       1546:        return (r);
1.43      thorpej  1547: }
                   1548:
                   1549: /*
                   1550:  * pool_cache_init:
                   1551:  *
                   1552:  *     Initialize a pool cache.
                   1553:  *
                   1554:  *     NOTE: If the pool must be protected from interrupts, we expect
                   1555:  *     to be called at the appropriate interrupt priority level.
                   1556:  */
                   1557: void
                   1558: pool_cache_init(struct pool_cache *pc, struct pool *pp,
                   1559:     int (*ctor)(void *, void *, int),
                   1560:     void (*dtor)(void *, void *),
                   1561:     void *arg)
                   1562: {
                   1563:
                   1564:        TAILQ_INIT(&pc->pc_grouplist);
                   1565:        simple_lock_init(&pc->pc_slock);
                   1566:
                   1567:        pc->pc_allocfrom = NULL;
                   1568:        pc->pc_freeto = NULL;
                   1569:        pc->pc_pool = pp;
                   1570:
                   1571:        pc->pc_ctor = ctor;
                   1572:        pc->pc_dtor = dtor;
                   1573:        pc->pc_arg  = arg;
                   1574:
1.48      thorpej  1575:        pc->pc_hits   = 0;
                   1576:        pc->pc_misses = 0;
                   1577:
                   1578:        pc->pc_ngroups = 0;
                   1579:
                   1580:        pc->pc_nitems = 0;
                   1581:
1.43      thorpej  1582:        simple_lock(&pp->pr_slock);
                   1583:        TAILQ_INSERT_TAIL(&pp->pr_cachelist, pc, pc_poollist);
                   1584:        simple_unlock(&pp->pr_slock);
                   1585: }
                   1586:
                   1587: /*
                   1588:  * pool_cache_destroy:
                   1589:  *
                   1590:  *     Destroy a pool cache.
                   1591:  */
                   1592: void
                   1593: pool_cache_destroy(struct pool_cache *pc)
                   1594: {
                   1595:        struct pool *pp = pc->pc_pool;
                   1596:
                   1597:        /* First, invalidate the entire cache. */
                   1598:        pool_cache_invalidate(pc);
                   1599:
                   1600:        /* ...and remove it from the pool's cache list. */
                   1601:        simple_lock(&pp->pr_slock);
                   1602:        TAILQ_REMOVE(&pp->pr_cachelist, pc, pc_poollist);
                   1603:        simple_unlock(&pp->pr_slock);
                   1604: }
                   1605:
                   1606: static __inline void *
                   1607: pcg_get(struct pool_cache_group *pcg)
                   1608: {
                   1609:        void *object;
                   1610:        u_int idx;
                   1611:
                   1612:        KASSERT(pcg->pcg_avail <= PCG_NOBJECTS);
1.45      thorpej  1613:        KASSERT(pcg->pcg_avail != 0);
1.43      thorpej  1614:        idx = --pcg->pcg_avail;
                   1615:
                   1616:        KASSERT(pcg->pcg_objects[idx] != NULL);
                   1617:        object = pcg->pcg_objects[idx];
                   1618:        pcg->pcg_objects[idx] = NULL;
                   1619:
                   1620:        return (object);
                   1621: }
                   1622:
                   1623: static __inline void
                   1624: pcg_put(struct pool_cache_group *pcg, void *object)
                   1625: {
                   1626:        u_int idx;
                   1627:
                   1628:        KASSERT(pcg->pcg_avail < PCG_NOBJECTS);
                   1629:        idx = pcg->pcg_avail++;
                   1630:
                   1631:        KASSERT(pcg->pcg_objects[idx] == NULL);
                   1632:        pcg->pcg_objects[idx] = object;
                   1633: }
                   1634:
                   1635: /*
                   1636:  * pool_cache_get:
                   1637:  *
                   1638:  *     Get an object from a pool cache.
                   1639:  */
                   1640: void *
                   1641: pool_cache_get(struct pool_cache *pc, int flags)
                   1642: {
                   1643:        struct pool_cache_group *pcg;
                   1644:        void *object;
                   1645:
1.50.2.1! nathanw  1646: #ifdef LOCKDEBUG
        !          1647:        if (flags & PR_WAITOK)
        !          1648:                simple_lock_only_held(NULL, "pool_cache_get(PR_WAITOK)");
        !          1649: #endif
        !          1650:
1.43      thorpej  1651:        simple_lock(&pc->pc_slock);
                   1652:
                   1653:        if ((pcg = pc->pc_allocfrom) == NULL) {
                   1654:                for (pcg = TAILQ_FIRST(&pc->pc_grouplist); pcg != NULL;
                   1655:                     pcg = TAILQ_NEXT(pcg, pcg_list)) {
                   1656:                        if (pcg->pcg_avail != 0) {
                   1657:                                pc->pc_allocfrom = pcg;
                   1658:                                goto have_group;
                   1659:                        }
                   1660:                }
                   1661:
                   1662:                /*
                   1663:                 * No groups with any available objects.  Allocate
                   1664:                 * a new object, construct it, and return it to
                   1665:                 * the caller.  We will allocate a group, if necessary,
                   1666:                 * when the object is freed back to the cache.
                   1667:                 */
1.48      thorpej  1668:                pc->pc_misses++;
1.43      thorpej  1669:                simple_unlock(&pc->pc_slock);
                   1670:                object = pool_get(pc->pc_pool, flags);
                   1671:                if (object != NULL && pc->pc_ctor != NULL) {
                   1672:                        if ((*pc->pc_ctor)(pc->pc_arg, object, flags) != 0) {
                   1673:                                pool_put(pc->pc_pool, object);
                   1674:                                return (NULL);
                   1675:                        }
                   1676:                }
                   1677:                return (object);
                   1678:        }
                   1679:
                   1680:  have_group:
1.48      thorpej  1681:        pc->pc_hits++;
                   1682:        pc->pc_nitems--;
1.43      thorpej  1683:        object = pcg_get(pcg);
                   1684:
                   1685:        if (pcg->pcg_avail == 0)
                   1686:                pc->pc_allocfrom = NULL;
1.45      thorpej  1687:
1.43      thorpej  1688:        simple_unlock(&pc->pc_slock);
                   1689:
                   1690:        return (object);
                   1691: }
                   1692:
                   1693: /*
                   1694:  * pool_cache_put:
                   1695:  *
                   1696:  *     Put an object back to the pool cache.
                   1697:  */
                   1698: void
                   1699: pool_cache_put(struct pool_cache *pc, void *object)
                   1700: {
                   1701:        struct pool_cache_group *pcg;
                   1702:
                   1703:        simple_lock(&pc->pc_slock);
                   1704:
                   1705:        if ((pcg = pc->pc_freeto) == NULL) {
                   1706:                for (pcg = TAILQ_FIRST(&pc->pc_grouplist); pcg != NULL;
                   1707:                     pcg = TAILQ_NEXT(pcg, pcg_list)) {
                   1708:                        if (pcg->pcg_avail != PCG_NOBJECTS) {
                   1709:                                pc->pc_freeto = pcg;
                   1710:                                goto have_group;
                   1711:                        }
                   1712:                }
                   1713:
                   1714:                /*
                   1715:                 * No empty groups to free the object to.  Attempt to
1.47      thorpej  1716:                 * allocate one.
1.43      thorpej  1717:                 */
1.47      thorpej  1718:                simple_unlock(&pc->pc_slock);
1.43      thorpej  1719:                pcg = pool_get(&pcgpool, PR_NOWAIT);
                   1720:                if (pcg != NULL) {
                   1721:                        memset(pcg, 0, sizeof(*pcg));
1.47      thorpej  1722:                        simple_lock(&pc->pc_slock);
1.48      thorpej  1723:                        pc->pc_ngroups++;
1.43      thorpej  1724:                        TAILQ_INSERT_TAIL(&pc->pc_grouplist, pcg, pcg_list);
1.47      thorpej  1725:                        if (pc->pc_freeto == NULL)
                   1726:                                pc->pc_freeto = pcg;
1.43      thorpej  1727:                        goto have_group;
                   1728:                }
                   1729:
                   1730:                /*
                   1731:                 * Unable to allocate a cache group; destruct the object
                   1732:                 * and free it back to the pool.
                   1733:                 */
1.50.2.1! nathanw  1734:                pool_cache_destruct_object(pc, object);
1.43      thorpej  1735:                return;
                   1736:        }
                   1737:
                   1738:  have_group:
1.48      thorpej  1739:        pc->pc_nitems++;
1.43      thorpej  1740:        pcg_put(pcg, object);
                   1741:
                   1742:        if (pcg->pcg_avail == PCG_NOBJECTS)
                   1743:                pc->pc_freeto = NULL;
                   1744:
                   1745:        simple_unlock(&pc->pc_slock);
                   1746: }
                   1747:
                   1748: /*
1.50.2.1! nathanw  1749:  * pool_cache_destruct_object:
        !          1750:  *
        !          1751:  *     Force destruction of an object and its release back into
        !          1752:  *     the pool.
        !          1753:  */
        !          1754: void
        !          1755: pool_cache_destruct_object(struct pool_cache *pc, void *object)
        !          1756: {
        !          1757:
        !          1758:        if (pc->pc_dtor != NULL)
        !          1759:                (*pc->pc_dtor)(pc->pc_arg, object);
        !          1760:        pool_put(pc->pc_pool, object);
        !          1761: }
        !          1762:
        !          1763: /*
1.43      thorpej  1764:  * pool_cache_do_invalidate:
                   1765:  *
                   1766:  *     This internal function implements pool_cache_invalidate() and
                   1767:  *     pool_cache_reclaim().
                   1768:  */
                   1769: static void
                   1770: pool_cache_do_invalidate(struct pool_cache *pc, int free_groups,
1.50.2.1! nathanw  1771:     void (*putit)(struct pool *, void *))
1.43      thorpej  1772: {
                   1773:        struct pool_cache_group *pcg, *npcg;
                   1774:        void *object;
                   1775:
                   1776:        for (pcg = TAILQ_FIRST(&pc->pc_grouplist); pcg != NULL;
                   1777:             pcg = npcg) {
                   1778:                npcg = TAILQ_NEXT(pcg, pcg_list);
                   1779:                while (pcg->pcg_avail != 0) {
1.48      thorpej  1780:                        pc->pc_nitems--;
1.43      thorpej  1781:                        object = pcg_get(pcg);
1.45      thorpej  1782:                        if (pcg->pcg_avail == 0 && pc->pc_allocfrom == pcg)
                   1783:                                pc->pc_allocfrom = NULL;
1.43      thorpej  1784:                        if (pc->pc_dtor != NULL)
                   1785:                                (*pc->pc_dtor)(pc->pc_arg, object);
1.50.2.1! nathanw  1786:                        (*putit)(pc->pc_pool, object);
1.43      thorpej  1787:                }
                   1788:                if (free_groups) {
1.48      thorpej  1789:                        pc->pc_ngroups--;
1.43      thorpej  1790:                        TAILQ_REMOVE(&pc->pc_grouplist, pcg, pcg_list);
1.46      thorpej  1791:                        if (pc->pc_freeto == pcg)
                   1792:                                pc->pc_freeto = NULL;
1.43      thorpej  1793:                        pool_put(&pcgpool, pcg);
                   1794:                }
                   1795:        }
                   1796: }
                   1797:
                   1798: /*
                   1799:  * pool_cache_invalidate:
                   1800:  *
                   1801:  *     Invalidate a pool cache (destruct and release all of the
                   1802:  *     cached objects).
                   1803:  */
                   1804: void
                   1805: pool_cache_invalidate(struct pool_cache *pc)
                   1806: {
                   1807:
                   1808:        simple_lock(&pc->pc_slock);
1.50.2.1! nathanw  1809:        pool_cache_do_invalidate(pc, 0, pool_put);
1.43      thorpej  1810:        simple_unlock(&pc->pc_slock);
                   1811: }
                   1812:
                   1813: /*
                   1814:  * pool_cache_reclaim:
                   1815:  *
                   1816:  *     Reclaim a pool cache for pool_reclaim().
                   1817:  */
                   1818: static void
                   1819: pool_cache_reclaim(struct pool_cache *pc)
                   1820: {
                   1821:
1.47      thorpej  1822:        simple_lock(&pc->pc_slock);
1.43      thorpej  1823:        pool_cache_do_invalidate(pc, 1, pool_do_put);
                   1824:        simple_unlock(&pc->pc_slock);
1.3       pk       1825: }

CVSweb <webmaster@jp.NetBSD.org>