[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 1.30.2.5

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

CVSweb <webmaster@jp.NetBSD.org>