[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 1.120

1.119     yamt        1: /*     $NetBSD$        */
1.1       pk          2:
                      3: /*-
1.43      thorpej     4:  * Copyright (c) 1997, 1999, 2000 The NetBSD Foundation, Inc.
1.1       pk          5:  * All rights reserved.
                      6:  *
                      7:  * This code is derived from software contributed to The NetBSD Foundation
1.20      thorpej     8:  * by Paul Kranenburg; by Jason R. Thorpe of the Numerical Aerospace
                      9:  * Simulation Facility, NASA Ames Research Center.
1.1       pk         10:  *
                     11:  * Redistribution and use in source and binary forms, with or without
                     12:  * modification, are permitted provided that the following conditions
                     13:  * are met:
                     14:  * 1. Redistributions of source code must retain the above copyright
                     15:  *    notice, this list of conditions and the following disclaimer.
                     16:  * 2. Redistributions in binary form must reproduce the above copyright
                     17:  *    notice, this list of conditions and the following disclaimer in the
                     18:  *    documentation and/or other materials provided with the distribution.
                     19:  * 3. All advertising materials mentioning features or use of this software
                     20:  *    must display the following acknowledgement:
1.13      christos   21:  *     This product includes software developed by the NetBSD
                     22:  *     Foundation, Inc. and its contributors.
1.1       pk         23:  * 4. Neither the name of The NetBSD Foundation nor the names of its
                     24:  *    contributors may be used to endorse or promote products derived
                     25:  *    from this software without specific prior written permission.
                     26:  *
                     27:  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
                     28:  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
                     29:  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
                     30:  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
                     31:  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
                     32:  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
                     33:  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
                     34:  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
                     35:  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
                     36:  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
                     37:  * POSSIBILITY OF SUCH DAMAGE.
                     38:  */
1.64      lukem      39:
                     40: #include <sys/cdefs.h>
1.119     yamt       41: __KERNEL_RCSID(0, "$NetBSD$");
1.24      scottr     42:
1.25      thorpej    43: #include "opt_pool.h"
1.24      scottr     44: #include "opt_poollog.h"
1.28      thorpej    45: #include "opt_lockdebug.h"
1.1       pk         46:
                     47: #include <sys/param.h>
                     48: #include <sys/systm.h>
                     49: #include <sys/proc.h>
                     50: #include <sys/errno.h>
                     51: #include <sys/kernel.h>
                     52: #include <sys/malloc.h>
                     53: #include <sys/lock.h>
                     54: #include <sys/pool.h>
1.20      thorpej    55: #include <sys/syslog.h>
1.3       pk         56:
                     57: #include <uvm/uvm.h>
                     58:
1.1       pk         59: /*
                     60:  * Pool resource management utility.
1.3       pk         61:  *
1.88      chs        62:  * Memory is allocated in pages which are split into pieces according to
                     63:  * the pool item size. Each page is kept on one of three lists in the
                     64:  * pool structure: `pr_emptypages', `pr_fullpages' and `pr_partpages',
                     65:  * for empty, full and partially-full pages respectively. The individual
                     66:  * pool items are on a linked list headed by `ph_itemlist' in each page
                     67:  * header. The memory for building the page list is either taken from
                     68:  * the allocated pages themselves (for small pool items) or taken from
                     69:  * an internal pool of page headers (`phpool').
1.1       pk         70:  */
                     71:
1.3       pk         72: /* List of all pools */
1.102     chs        73: LIST_HEAD(,pool) pool_head = LIST_HEAD_INITIALIZER(pool_head);
1.3       pk         74:
                     75: /* Private pool for page header structures */
1.97      yamt       76: #define        PHPOOL_MAX      8
                     77: static struct pool phpool[PHPOOL_MAX];
                     78: #define        PHPOOL_FREELIST_NELEM(idx)      (((idx) == 0) ? 0 : (1 << (idx)))
1.3       pk         79:
1.62      bjh21      80: #ifdef POOL_SUBPAGE
                     81: /* Pool of subpages for use by normal pools. */
                     82: static struct pool psppool;
                     83: #endif
                     84:
1.117     yamt       85: static SLIST_HEAD(, pool_allocator) pa_deferinitq =
                     86:     SLIST_HEAD_INITIALIZER(pa_deferinitq);
                     87:
1.98      yamt       88: static void *pool_page_alloc_meta(struct pool *, int);
                     89: static void pool_page_free_meta(struct pool *, void *);
                     90:
                     91: /* allocator for pool metadata */
                     92: static struct pool_allocator pool_allocator_meta = {
1.117     yamt       93:        pool_page_alloc_meta, pool_page_free_meta,
                     94:        .pa_backingmapptr = &kmem_map,
1.98      yamt       95: };
                     96:
1.3       pk         97: /* # of seconds to retain page after last use */
                     98: int pool_inactive_time = 10;
                     99:
                    100: /* Next candidate for drainage (see pool_drain()) */
1.23      thorpej   101: static struct pool     *drainpp;
                    102:
                    103: /* This spin lock protects both pool_head and drainpp. */
                    104: struct simplelock pool_head_slock = SIMPLELOCK_INITIALIZER;
1.3       pk        105:
1.99      yamt      106: typedef uint8_t pool_item_freelist_t;
                    107:
1.3       pk        108: struct pool_item_header {
                    109:        /* Page headers */
1.88      chs       110:        LIST_ENTRY(pool_item_header)
1.3       pk        111:                                ph_pagelist;    /* pool page list */
1.88      chs       112:        SPLAY_ENTRY(pool_item_header)
                    113:                                ph_node;        /* Off-page page headers */
1.3       pk        114:        caddr_t                 ph_page;        /* this page's address */
                    115:        struct timeval          ph_time;        /* last referenced */
1.97      yamt      116:        union {
                    117:                /* !PR_NOTOUCH */
                    118:                struct {
1.102     chs       119:                        LIST_HEAD(, pool_item)
1.97      yamt      120:                                phu_itemlist;   /* chunk list for this page */
                    121:                } phu_normal;
                    122:                /* PR_NOTOUCH */
                    123:                struct {
                    124:                        uint16_t
                    125:                                phu_off;        /* start offset in page */
1.99      yamt      126:                        pool_item_freelist_t
1.97      yamt      127:                                phu_firstfree;  /* first free item */
1.99      yamt      128:                        /*
                    129:                         * XXX it might be better to use
                    130:                         * a simple bitmap and ffs(3)
                    131:                         */
1.97      yamt      132:                } phu_notouch;
                    133:        } ph_u;
                    134:        uint16_t                ph_nmissing;    /* # of chunks in use */
1.3       pk        135: };
1.97      yamt      136: #define        ph_itemlist     ph_u.phu_normal.phu_itemlist
                    137: #define        ph_off          ph_u.phu_notouch.phu_off
                    138: #define        ph_firstfree    ph_u.phu_notouch.phu_firstfree
1.3       pk        139:
1.1       pk        140: struct pool_item {
1.3       pk        141: #ifdef DIAGNOSTIC
1.82      thorpej   142:        u_int pi_magic;
1.33      chs       143: #endif
1.82      thorpej   144: #define        PI_MAGIC 0xdeadbeefU
1.3       pk        145:        /* Other entries use only this list entry */
1.102     chs       146:        LIST_ENTRY(pool_item)   pi_list;
1.3       pk        147: };
                    148:
1.53      thorpej   149: #define        POOL_NEEDS_CATCHUP(pp)                                          \
                    150:        ((pp)->pr_nitems < (pp)->pr_minitems)
                    151:
1.43      thorpej   152: /*
                    153:  * Pool cache management.
                    154:  *
                    155:  * Pool caches provide a way for constructed objects to be cached by the
                    156:  * pool subsystem.  This can lead to performance improvements by avoiding
                    157:  * needless object construction/destruction; it is deferred until absolutely
                    158:  * necessary.
                    159:  *
                    160:  * Caches are grouped into cache groups.  Each cache group references
                    161:  * up to 16 constructed objects.  When a cache allocates an object
                    162:  * from the pool, it calls the object's constructor and places it into
                    163:  * a cache group.  When a cache group frees an object back to the pool,
                    164:  * it first calls the object's destructor.  This allows the object to
                    165:  * persist in constructed form while freed to the cache.
                    166:  *
                    167:  * Multiple caches may exist for each pool.  This allows a single
                    168:  * object type to have multiple constructed forms.  The pool references
                    169:  * each cache, so that when a pool is drained by the pagedaemon, it can
                    170:  * drain each individual cache as well.  Each time a cache is drained,
                    171:  * the most idle cache group is freed to the pool in its entirety.
                    172:  *
                    173:  * Pool caches are layed on top of pools.  By layering them, we can avoid
                    174:  * the complexity of cache management for pools which would not benefit
                    175:  * from it.
                    176:  */
                    177:
                    178: /* The cache group pool. */
                    179: static struct pool pcgpool;
1.3       pk        180:
1.102     chs       181: static void    pool_cache_reclaim(struct pool_cache *, struct pool_pagelist *,
                    182:                                   struct pool_cache_grouplist *);
                    183: static void    pcg_grouplist_free(struct pool_cache_grouplist *);
1.3       pk        184:
1.42      thorpej   185: static int     pool_catchup(struct pool *);
1.55      thorpej   186: static void    pool_prime_page(struct pool *, caddr_t,
                    187:                    struct pool_item_header *);
1.88      chs       188: static void    pool_update_curpage(struct pool *);
1.66      thorpej   189:
1.113     yamt      190: static int     pool_grow(struct pool *, int);
1.117     yamt      191: static void    *pool_allocator_alloc(struct pool *, int);
                    192: static void    pool_allocator_free(struct pool *, void *);
1.3       pk        193:
1.97      yamt      194: static void pool_print_pagelist(struct pool *, struct pool_pagelist *,
1.88      chs       195:        void (*)(const char *, ...));
1.42      thorpej   196: static void pool_print1(struct pool *, const char *,
                    197:        void (*)(const char *, ...));
1.3       pk        198:
1.88      chs       199: static int pool_chk_page(struct pool *, const char *,
                    200:                         struct pool_item_header *);
                    201:
1.3       pk        202: /*
1.52      thorpej   203:  * Pool log entry. An array of these is allocated in pool_init().
1.3       pk        204:  */
                    205: struct pool_log {
                    206:        const char      *pl_file;
                    207:        long            pl_line;
                    208:        int             pl_action;
1.25      thorpej   209: #define        PRLOG_GET       1
                    210: #define        PRLOG_PUT       2
1.3       pk        211:        void            *pl_addr;
1.1       pk        212: };
                    213:
1.86      matt      214: #ifdef POOL_DIAGNOSTIC
1.3       pk        215: /* Number of entries in pool log buffers */
1.17      thorpej   216: #ifndef POOL_LOGSIZE
                    217: #define        POOL_LOGSIZE    10
                    218: #endif
                    219:
                    220: int pool_logsize = POOL_LOGSIZE;
1.1       pk        221:
1.110     perry     222: static inline void
1.42      thorpej   223: pr_log(struct pool *pp, void *v, int action, const char *file, long line)
1.3       pk        224: {
                    225:        int n = pp->pr_curlogentry;
                    226:        struct pool_log *pl;
                    227:
1.20      thorpej   228:        if ((pp->pr_roflags & PR_LOGGING) == 0)
1.3       pk        229:                return;
                    230:
                    231:        /*
                    232:         * Fill in the current entry. Wrap around and overwrite
                    233:         * the oldest entry if necessary.
                    234:         */
                    235:        pl = &pp->pr_log[n];
                    236:        pl->pl_file = file;
                    237:        pl->pl_line = line;
                    238:        pl->pl_action = action;
                    239:        pl->pl_addr = v;
                    240:        if (++n >= pp->pr_logsize)
                    241:                n = 0;
                    242:        pp->pr_curlogentry = n;
                    243: }
                    244:
                    245: static void
1.42      thorpej   246: pr_printlog(struct pool *pp, struct pool_item *pi,
                    247:     void (*pr)(const char *, ...))
1.3       pk        248: {
                    249:        int i = pp->pr_logsize;
                    250:        int n = pp->pr_curlogentry;
                    251:
1.20      thorpej   252:        if ((pp->pr_roflags & PR_LOGGING) == 0)
1.3       pk        253:                return;
                    254:
                    255:        /*
                    256:         * Print all entries in this pool's log.
                    257:         */
                    258:        while (i-- > 0) {
                    259:                struct pool_log *pl = &pp->pr_log[n];
                    260:                if (pl->pl_action != 0) {
1.25      thorpej   261:                        if (pi == NULL || pi == pl->pl_addr) {
                    262:                                (*pr)("\tlog entry %d:\n", i);
                    263:                                (*pr)("\t\taction = %s, addr = %p\n",
                    264:                                    pl->pl_action == PRLOG_GET ? "get" : "put",
                    265:                                    pl->pl_addr);
                    266:                                (*pr)("\t\tfile: %s at line %lu\n",
                    267:                                    pl->pl_file, pl->pl_line);
                    268:                        }
1.3       pk        269:                }
                    270:                if (++n >= pp->pr_logsize)
                    271:                        n = 0;
                    272:        }
                    273: }
1.25      thorpej   274:
1.110     perry     275: static inline void
1.42      thorpej   276: pr_enter(struct pool *pp, const char *file, long line)
1.25      thorpej   277: {
                    278:
1.34      thorpej   279:        if (__predict_false(pp->pr_entered_file != NULL)) {
1.25      thorpej   280:                printf("pool %s: reentrancy at file %s line %ld\n",
                    281:                    pp->pr_wchan, file, line);
                    282:                printf("         previous entry at file %s line %ld\n",
                    283:                    pp->pr_entered_file, pp->pr_entered_line);
                    284:                panic("pr_enter");
                    285:        }
                    286:
                    287:        pp->pr_entered_file = file;
                    288:        pp->pr_entered_line = line;
                    289: }
                    290:
1.110     perry     291: static inline void
1.42      thorpej   292: pr_leave(struct pool *pp)
1.25      thorpej   293: {
                    294:
1.34      thorpej   295:        if (__predict_false(pp->pr_entered_file == NULL)) {
1.25      thorpej   296:                printf("pool %s not entered?\n", pp->pr_wchan);
                    297:                panic("pr_leave");
                    298:        }
                    299:
                    300:        pp->pr_entered_file = NULL;
                    301:        pp->pr_entered_line = 0;
                    302: }
                    303:
1.110     perry     304: static inline void
1.42      thorpej   305: pr_enter_check(struct pool *pp, void (*pr)(const char *, ...))
1.25      thorpej   306: {
                    307:
                    308:        if (pp->pr_entered_file != NULL)
                    309:                (*pr)("\n\tcurrently entered from file %s line %ld\n",
                    310:                    pp->pr_entered_file, pp->pr_entered_line);
                    311: }
1.3       pk        312: #else
1.25      thorpej   313: #define        pr_log(pp, v, action, file, line)
                    314: #define        pr_printlog(pp, pi, pr)
                    315: #define        pr_enter(pp, file, line)
                    316: #define        pr_leave(pp)
                    317: #define        pr_enter_check(pp, pr)
1.59      thorpej   318: #endif /* POOL_DIAGNOSTIC */
1.3       pk        319:
1.110     perry     320: static inline int
1.97      yamt      321: pr_item_notouch_index(const struct pool *pp, const struct pool_item_header *ph,
                    322:     const void *v)
                    323: {
                    324:        const char *cp = v;
                    325:        int idx;
                    326:
                    327:        KASSERT(pp->pr_roflags & PR_NOTOUCH);
                    328:        idx = (cp - ph->ph_page - ph->ph_off) / pp->pr_size;
                    329:        KASSERT(idx < pp->pr_itemsperpage);
                    330:        return idx;
                    331: }
                    332:
1.99      yamt      333: #define        PR_FREELIST_ALIGN(p) \
                    334:        roundup((uintptr_t)(p), sizeof(pool_item_freelist_t))
                    335: #define        PR_FREELIST(ph) ((pool_item_freelist_t *)PR_FREELIST_ALIGN((ph) + 1))
                    336: #define        PR_INDEX_USED   ((pool_item_freelist_t)-1)
                    337: #define        PR_INDEX_EOL    ((pool_item_freelist_t)-2)
1.97      yamt      338:
1.110     perry     339: static inline void
1.97      yamt      340: pr_item_notouch_put(const struct pool *pp, struct pool_item_header *ph,
                    341:     void *obj)
                    342: {
                    343:        int idx = pr_item_notouch_index(pp, ph, obj);
1.99      yamt      344:        pool_item_freelist_t *freelist = PR_FREELIST(ph);
1.97      yamt      345:
                    346:        KASSERT(freelist[idx] == PR_INDEX_USED);
                    347:        freelist[idx] = ph->ph_firstfree;
                    348:        ph->ph_firstfree = idx;
                    349: }
                    350:
1.110     perry     351: static inline void *
1.97      yamt      352: pr_item_notouch_get(const struct pool *pp, struct pool_item_header *ph)
                    353: {
                    354:        int idx = ph->ph_firstfree;
1.99      yamt      355:        pool_item_freelist_t *freelist = PR_FREELIST(ph);
1.97      yamt      356:
                    357:        KASSERT(freelist[idx] != PR_INDEX_USED);
                    358:        ph->ph_firstfree = freelist[idx];
                    359:        freelist[idx] = PR_INDEX_USED;
                    360:
                    361:        return ph->ph_page + ph->ph_off + idx * pp->pr_size;
                    362: }
                    363:
1.110     perry     364: static inline int
1.88      chs       365: phtree_compare(struct pool_item_header *a, struct pool_item_header *b)
                    366: {
                    367:        if (a->ph_page < b->ph_page)
                    368:                return (-1);
                    369:        else if (a->ph_page > b->ph_page)
                    370:                return (1);
                    371:        else
                    372:                return (0);
                    373: }
                    374:
                    375: SPLAY_PROTOTYPE(phtree, pool_item_header, ph_node, phtree_compare);
                    376: SPLAY_GENERATE(phtree, pool_item_header, ph_node, phtree_compare);
                    377:
1.3       pk        378: /*
                    379:  * Return the pool page header based on page address.
                    380:  */
1.110     perry     381: static inline struct pool_item_header *
1.42      thorpej   382: pr_find_pagehead(struct pool *pp, caddr_t page)
1.3       pk        383: {
1.88      chs       384:        struct pool_item_header *ph, tmp;
1.3       pk        385:
1.20      thorpej   386:        if ((pp->pr_roflags & PR_PHINPAGE) != 0)
1.3       pk        387:                return ((struct pool_item_header *)(page + pp->pr_phoffset));
                    388:
1.88      chs       389:        tmp.ph_page = page;
                    390:        ph = SPLAY_FIND(phtree, &pp->pr_phtree, &tmp);
                    391:        return ph;
1.3       pk        392: }
                    393:
1.101     thorpej   394: static void
                    395: pr_pagelist_free(struct pool *pp, struct pool_pagelist *pq)
                    396: {
                    397:        struct pool_item_header *ph;
                    398:        int s;
                    399:
                    400:        while ((ph = LIST_FIRST(pq)) != NULL) {
                    401:                LIST_REMOVE(ph, ph_pagelist);
                    402:                pool_allocator_free(pp, ph->ph_page);
                    403:                if ((pp->pr_roflags & PR_PHINPAGE) == 0) {
                    404:                        s = splvm();
                    405:                        pool_put(pp->pr_phpool, ph);
                    406:                        splx(s);
                    407:                }
                    408:        }
                    409: }
                    410:
1.3       pk        411: /*
                    412:  * Remove a page from the pool.
                    413:  */
1.110     perry     414: static inline void
1.61      chs       415: pr_rmpage(struct pool *pp, struct pool_item_header *ph,
                    416:      struct pool_pagelist *pq)
1.3       pk        417: {
                    418:
1.101     thorpej   419:        LOCK_ASSERT(simple_lock_held(&pp->pr_slock));
1.91      yamt      420:
1.3       pk        421:        /*
1.7       thorpej   422:         * If the page was idle, decrement the idle page count.
1.3       pk        423:         */
1.6       thorpej   424:        if (ph->ph_nmissing == 0) {
                    425: #ifdef DIAGNOSTIC
                    426:                if (pp->pr_nidle == 0)
                    427:                        panic("pr_rmpage: nidle inconsistent");
1.20      thorpej   428:                if (pp->pr_nitems < pp->pr_itemsperpage)
                    429:                        panic("pr_rmpage: nitems inconsistent");
1.6       thorpej   430: #endif
                    431:                pp->pr_nidle--;
                    432:        }
1.7       thorpej   433:
1.20      thorpej   434:        pp->pr_nitems -= pp->pr_itemsperpage;
                    435:
1.7       thorpej   436:        /*
1.101     thorpej   437:         * Unlink the page from the pool and queue it for release.
1.7       thorpej   438:         */
1.88      chs       439:        LIST_REMOVE(ph, ph_pagelist);
1.91      yamt      440:        if ((pp->pr_roflags & PR_PHINPAGE) == 0)
                    441:                SPLAY_REMOVE(phtree, &pp->pr_phtree, ph);
1.101     thorpej   442:        LIST_INSERT_HEAD(pq, ph, ph_pagelist);
                    443:
1.7       thorpej   444:        pp->pr_npages--;
                    445:        pp->pr_npagefree++;
1.6       thorpej   446:
1.88      chs       447:        pool_update_curpage(pp);
1.3       pk        448: }
                    449:
1.117     yamt      450: static boolean_t
                    451: pa_starved_p(struct pool_allocator *pa)
                    452: {
                    453:
                    454:        if (pa->pa_backingmap != NULL) {
                    455:                return vm_map_starved_p(pa->pa_backingmap);
                    456:        }
                    457:        return FALSE;
                    458: }
                    459:
                    460: static int
                    461: pool_reclaim_callback(struct callback_entry *ce, void *obj, void *arg)
                    462: {
                    463:        struct pool *pp = obj;
                    464:        struct pool_allocator *pa = pp->pr_alloc;
                    465:
                    466:        KASSERT(&pp->pr_reclaimerentry == ce);
                    467:        pool_reclaim(pp);
                    468:        if (!pa_starved_p(pa)) {
                    469:                return CALLBACK_CHAIN_ABORT;
                    470:        }
                    471:        return CALLBACK_CHAIN_CONTINUE;
                    472: }
                    473:
                    474: static void
                    475: pool_reclaim_register(struct pool *pp)
                    476: {
                    477:        struct vm_map *map = pp->pr_alloc->pa_backingmap;
                    478:        int s;
                    479:
                    480:        if (map == NULL) {
                    481:                return;
                    482:        }
                    483:
                    484:        s = splvm(); /* not necessary for INTRSAFE maps, but don't care. */
                    485:        callback_register(&vm_map_to_kernel(map)->vmk_reclaim_callback,
                    486:            &pp->pr_reclaimerentry, pp, pool_reclaim_callback);
                    487:        splx(s);
                    488: }
                    489:
                    490: static void
                    491: pool_reclaim_unregister(struct pool *pp)
                    492: {
                    493:        struct vm_map *map = pp->pr_alloc->pa_backingmap;
                    494:        int s;
                    495:
                    496:        if (map == NULL) {
                    497:                return;
                    498:        }
                    499:
                    500:        s = splvm(); /* not necessary for INTRSAFE maps, but don't care. */
                    501:        callback_unregister(&vm_map_to_kernel(map)->vmk_reclaim_callback,
                    502:            &pp->pr_reclaimerentry);
                    503:        splx(s);
                    504: }
                    505:
                    506: static void
                    507: pa_reclaim_register(struct pool_allocator *pa)
                    508: {
                    509:        struct vm_map *map = *pa->pa_backingmapptr;
                    510:        struct pool *pp;
                    511:
                    512:        KASSERT(pa->pa_backingmap == NULL);
                    513:        if (map == NULL) {
                    514:                SLIST_INSERT_HEAD(&pa_deferinitq, pa, pa_q);
                    515:                return;
                    516:        }
                    517:        pa->pa_backingmap = map;
                    518:        TAILQ_FOREACH(pp, &pa->pa_list, pr_alloc_list) {
                    519:                pool_reclaim_register(pp);
                    520:        }
                    521: }
                    522:
1.3       pk        523: /*
1.94      simonb    524:  * Initialize all the pools listed in the "pools" link set.
                    525:  */
                    526: void
1.117     yamt      527: pool_subsystem_init(void)
1.94      simonb    528: {
1.117     yamt      529:        struct pool_allocator *pa;
1.94      simonb    530:        __link_set_decl(pools, struct link_pool_init);
                    531:        struct link_pool_init * const *pi;
                    532:
                    533:        __link_set_foreach(pi, pools)
                    534:                pool_init((*pi)->pp, (*pi)->size, (*pi)->align,
                    535:                    (*pi)->align_offset, (*pi)->flags, (*pi)->wchan,
                    536:                    (*pi)->palloc);
1.117     yamt      537:
                    538:        while ((pa = SLIST_FIRST(&pa_deferinitq)) != NULL) {
                    539:                KASSERT(pa->pa_backingmapptr != NULL);
                    540:                KASSERT(*pa->pa_backingmapptr != NULL);
                    541:                SLIST_REMOVE_HEAD(&pa_deferinitq, pa_q);
                    542:                pa_reclaim_register(pa);
                    543:        }
1.94      simonb    544: }
                    545:
                    546: /*
1.3       pk        547:  * Initialize the given pool resource structure.
                    548:  *
                    549:  * We export this routine to allow other kernel parts to declare
                    550:  * static pools that must be initialized before malloc() is available.
                    551:  */
                    552: void
1.42      thorpej   553: pool_init(struct pool *pp, size_t size, u_int align, u_int ioff, int flags,
1.66      thorpej   554:     const char *wchan, struct pool_allocator *palloc)
1.3       pk        555: {
1.116     simonb    556: #ifdef DEBUG
                    557:        struct pool *pp1;
                    558: #endif
1.92      enami     559:        size_t trysize, phsize;
1.116     simonb    560:        int off, slack, s;
1.3       pk        561:
1.99      yamt      562:        KASSERT((1UL << (CHAR_BIT * sizeof(pool_item_freelist_t))) - 2 >=
                    563:            PHPOOL_FREELIST_NELEM(PHPOOL_MAX - 1));
                    564:
1.116     simonb    565: #ifdef DEBUG
                    566:        /*
                    567:         * Check that the pool hasn't already been initialised and
                    568:         * added to the list of all pools.
                    569:         */
                    570:        LIST_FOREACH(pp1, &pool_head, pr_poollist) {
                    571:                if (pp == pp1)
                    572:                        panic("pool_init: pool %s already initialised",
                    573:                            wchan);
                    574:        }
                    575: #endif
                    576:
1.25      thorpej   577: #ifdef POOL_DIAGNOSTIC
                    578:        /*
                    579:         * Always log if POOL_DIAGNOSTIC is defined.
                    580:         */
                    581:        if (pool_logsize != 0)
                    582:                flags |= PR_LOGGING;
                    583: #endif
                    584:
1.66      thorpej   585:        if (palloc == NULL)
                    586:                palloc = &pool_allocator_kmem;
1.112     bjh21     587: #ifdef POOL_SUBPAGE
                    588:        if (size > palloc->pa_pagesz) {
                    589:                if (palloc == &pool_allocator_kmem)
                    590:                        palloc = &pool_allocator_kmem_fullpage;
                    591:                else if (palloc == &pool_allocator_nointr)
                    592:                        palloc = &pool_allocator_nointr_fullpage;
                    593:        }
1.66      thorpej   594: #endif /* POOL_SUBPAGE */
                    595:        if ((palloc->pa_flags & PA_INITIALIZED) == 0) {
1.112     bjh21     596:                if (palloc->pa_pagesz == 0)
1.66      thorpej   597:                        palloc->pa_pagesz = PAGE_SIZE;
                    598:
                    599:                TAILQ_INIT(&palloc->pa_list);
                    600:
                    601:                simple_lock_init(&palloc->pa_slock);
                    602:                palloc->pa_pagemask = ~(palloc->pa_pagesz - 1);
                    603:                palloc->pa_pageshift = ffs(palloc->pa_pagesz) - 1;
1.117     yamt      604:
                    605:                if (palloc->pa_backingmapptr != NULL) {
                    606:                        pa_reclaim_register(palloc);
                    607:                }
1.66      thorpej   608:                palloc->pa_flags |= PA_INITIALIZED;
1.4       thorpej   609:        }
1.3       pk        610:
                    611:        if (align == 0)
                    612:                align = ALIGN(1);
1.14      thorpej   613:
1.120   ! yamt      614:        if ((flags & PR_NOTOUCH) == 0 && size < sizeof(struct pool_item))
1.14      thorpej   615:                size = sizeof(struct pool_item);
1.3       pk        616:
1.78      thorpej   617:        size = roundup(size, align);
1.66      thorpej   618: #ifdef DIAGNOSTIC
                    619:        if (size > palloc->pa_pagesz)
1.35      pk        620:                panic("pool_init: pool item size (%lu) too large",
                    621:                      (u_long)size);
1.66      thorpej   622: #endif
1.35      pk        623:
1.3       pk        624:        /*
                    625:         * Initialize the pool structure.
                    626:         */
1.88      chs       627:        LIST_INIT(&pp->pr_emptypages);
                    628:        LIST_INIT(&pp->pr_fullpages);
                    629:        LIST_INIT(&pp->pr_partpages);
1.102     chs       630:        LIST_INIT(&pp->pr_cachelist);
1.3       pk        631:        pp->pr_curpage = NULL;
                    632:        pp->pr_npages = 0;
                    633:        pp->pr_minitems = 0;
                    634:        pp->pr_minpages = 0;
                    635:        pp->pr_maxpages = UINT_MAX;
1.20      thorpej   636:        pp->pr_roflags = flags;
                    637:        pp->pr_flags = 0;
1.35      pk        638:        pp->pr_size = size;
1.3       pk        639:        pp->pr_align = align;
                    640:        pp->pr_wchan = wchan;
1.66      thorpej   641:        pp->pr_alloc = palloc;
1.20      thorpej   642:        pp->pr_nitems = 0;
                    643:        pp->pr_nout = 0;
                    644:        pp->pr_hardlimit = UINT_MAX;
                    645:        pp->pr_hardlimit_warning = NULL;
1.31      thorpej   646:        pp->pr_hardlimit_ratecap.tv_sec = 0;
                    647:        pp->pr_hardlimit_ratecap.tv_usec = 0;
                    648:        pp->pr_hardlimit_warning_last.tv_sec = 0;
                    649:        pp->pr_hardlimit_warning_last.tv_usec = 0;
1.68      thorpej   650:        pp->pr_drain_hook = NULL;
                    651:        pp->pr_drain_hook_arg = NULL;
1.3       pk        652:
                    653:        /*
                    654:         * Decide whether to put the page header off page to avoid
1.92      enami     655:         * wasting too large a part of the page or too big item.
                    656:         * Off-page page headers go on a hash table, so we can match
                    657:         * a returned item with its header based on the page address.
                    658:         * We use 1/16 of the page size and about 8 times of the item
                    659:         * size as the threshold (XXX: tune)
                    660:         *
                    661:         * However, we'll put the header into the page if we can put
                    662:         * it without wasting any items.
                    663:         *
                    664:         * Silently enforce `0 <= ioff < align'.
1.3       pk        665:         */
1.92      enami     666:        pp->pr_itemoffset = ioff %= align;
                    667:        /* See the comment below about reserved bytes. */
                    668:        trysize = palloc->pa_pagesz - ((align - ioff) % align);
                    669:        phsize = ALIGN(sizeof(struct pool_item_header));
1.97      yamt      670:        if ((pp->pr_roflags & PR_NOTOUCH) == 0 &&
                    671:            (pp->pr_size < MIN(palloc->pa_pagesz / 16, phsize << 3) ||
                    672:            trysize / pp->pr_size == (trysize - phsize) / pp->pr_size)) {
1.3       pk        673:                /* Use the end of the page for the page header */
1.20      thorpej   674:                pp->pr_roflags |= PR_PHINPAGE;
1.92      enami     675:                pp->pr_phoffset = off = palloc->pa_pagesz - phsize;
1.2       pk        676:        } else {
1.3       pk        677:                /* The page header will be taken from our page header pool */
                    678:                pp->pr_phoffset = 0;
1.66      thorpej   679:                off = palloc->pa_pagesz;
1.88      chs       680:                SPLAY_INIT(&pp->pr_phtree);
1.2       pk        681:        }
1.1       pk        682:
1.3       pk        683:        /*
                    684:         * Alignment is to take place at `ioff' within the item. This means
                    685:         * we must reserve up to `align - 1' bytes on the page to allow
                    686:         * appropriate positioning of each item.
                    687:         */
                    688:        pp->pr_itemsperpage = (off - ((align - ioff) % align)) / pp->pr_size;
1.43      thorpej   689:        KASSERT(pp->pr_itemsperpage != 0);
1.97      yamt      690:        if ((pp->pr_roflags & PR_NOTOUCH)) {
                    691:                int idx;
                    692:
                    693:                for (idx = 0; pp->pr_itemsperpage > PHPOOL_FREELIST_NELEM(idx);
                    694:                    idx++) {
                    695:                        /* nothing */
                    696:                }
                    697:                if (idx >= PHPOOL_MAX) {
                    698:                        /*
                    699:                         * if you see this panic, consider to tweak
                    700:                         * PHPOOL_MAX and PHPOOL_FREELIST_NELEM.
                    701:                         */
                    702:                        panic("%s: too large itemsperpage(%d) for PR_NOTOUCH",
                    703:                            pp->pr_wchan, pp->pr_itemsperpage);
                    704:                }
                    705:                pp->pr_phpool = &phpool[idx];
                    706:        } else if ((pp->pr_roflags & PR_PHINPAGE) == 0) {
                    707:                pp->pr_phpool = &phpool[0];
                    708:        }
                    709: #if defined(DIAGNOSTIC)
                    710:        else {
                    711:                pp->pr_phpool = NULL;
                    712:        }
                    713: #endif
1.3       pk        714:
                    715:        /*
                    716:         * Use the slack between the chunks and the page header
                    717:         * for "cache coloring".
                    718:         */
                    719:        slack = off - pp->pr_itemsperpage * pp->pr_size;
                    720:        pp->pr_maxcolor = (slack / align) * align;
                    721:        pp->pr_curcolor = 0;
                    722:
                    723:        pp->pr_nget = 0;
                    724:        pp->pr_nfail = 0;
                    725:        pp->pr_nput = 0;
                    726:        pp->pr_npagealloc = 0;
                    727:        pp->pr_npagefree = 0;
1.1       pk        728:        pp->pr_hiwat = 0;
1.8       thorpej   729:        pp->pr_nidle = 0;
1.3       pk        730:
1.59      thorpej   731: #ifdef POOL_DIAGNOSTIC
1.25      thorpej   732:        if (flags & PR_LOGGING) {
                    733:                if (kmem_map == NULL ||
                    734:                    (pp->pr_log = malloc(pool_logsize * sizeof(struct pool_log),
                    735:                     M_TEMP, M_NOWAIT)) == NULL)
1.20      thorpej   736:                        pp->pr_roflags &= ~PR_LOGGING;
1.3       pk        737:                pp->pr_curlogentry = 0;
                    738:                pp->pr_logsize = pool_logsize;
                    739:        }
1.59      thorpej   740: #endif
1.25      thorpej   741:
                    742:        pp->pr_entered_file = NULL;
                    743:        pp->pr_entered_line = 0;
1.3       pk        744:
1.21      thorpej   745:        simple_lock_init(&pp->pr_slock);
1.1       pk        746:
1.3       pk        747:        /*
1.43      thorpej   748:         * Initialize private page header pool and cache magazine pool if we
                    749:         * haven't done so yet.
1.23      thorpej   750:         * XXX LOCKING.
1.3       pk        751:         */
1.97      yamt      752:        if (phpool[0].pr_size == 0) {
                    753:                int idx;
                    754:                for (idx = 0; idx < PHPOOL_MAX; idx++) {
                    755:                        static char phpool_names[PHPOOL_MAX][6+1+6+1];
                    756:                        int nelem;
                    757:                        size_t sz;
                    758:
                    759:                        nelem = PHPOOL_FREELIST_NELEM(idx);
                    760:                        snprintf(phpool_names[idx], sizeof(phpool_names[idx]),
                    761:                            "phpool-%d", nelem);
                    762:                        sz = sizeof(struct pool_item_header);
                    763:                        if (nelem) {
                    764:                                sz = PR_FREELIST_ALIGN(sz)
1.99      yamt      765:                                    + nelem * sizeof(pool_item_freelist_t);
1.97      yamt      766:                        }
                    767:                        pool_init(&phpool[idx], sz, 0, 0, 0,
1.98      yamt      768:                            phpool_names[idx], &pool_allocator_meta);
1.97      yamt      769:                }
1.62      bjh21     770: #ifdef POOL_SUBPAGE
                    771:                pool_init(&psppool, POOL_SUBPAGE, POOL_SUBPAGE, 0,
1.98      yamt      772:                    PR_RECURSIVE, "psppool", &pool_allocator_meta);
1.62      bjh21     773: #endif
1.43      thorpej   774:                pool_init(&pcgpool, sizeof(struct pool_cache_group), 0, 0,
1.98      yamt      775:                    0, "pcgpool", &pool_allocator_meta);
1.1       pk        776:        }
                    777:
1.23      thorpej   778:        /* Insert into the list of all pools. */
                    779:        simple_lock(&pool_head_slock);
1.102     chs       780:        LIST_INSERT_HEAD(&pool_head, pp, pr_poollist);
1.23      thorpej   781:        simple_unlock(&pool_head_slock);
1.66      thorpej   782:
                    783:        /* Insert this into the list of pools using this allocator. */
1.93      dbj       784:        s = splvm();
1.66      thorpej   785:        simple_lock(&palloc->pa_slock);
                    786:        TAILQ_INSERT_TAIL(&palloc->pa_list, pp, pr_alloc_list);
                    787:        simple_unlock(&palloc->pa_slock);
1.93      dbj       788:        splx(s);
1.117     yamt      789:        pool_reclaim_register(pp);
1.1       pk        790: }
                    791:
                    792: /*
                    793:  * De-commision a pool resource.
                    794:  */
                    795: void
1.42      thorpej   796: pool_destroy(struct pool *pp)
1.1       pk        797: {
1.101     thorpej   798:        struct pool_pagelist pq;
1.3       pk        799:        struct pool_item_header *ph;
1.93      dbj       800:        int s;
1.43      thorpej   801:
1.101     thorpej   802:        /* Remove from global pool list */
                    803:        simple_lock(&pool_head_slock);
1.102     chs       804:        LIST_REMOVE(pp, pr_poollist);
1.101     thorpej   805:        if (drainpp == pp)
                    806:                drainpp = NULL;
                    807:        simple_unlock(&pool_head_slock);
                    808:
                    809:        /* Remove this pool from its allocator's list of pools. */
1.117     yamt      810:        pool_reclaim_unregister(pp);
1.93      dbj       811:        s = splvm();
1.66      thorpej   812:        simple_lock(&pp->pr_alloc->pa_slock);
                    813:        TAILQ_REMOVE(&pp->pr_alloc->pa_list, pp, pr_alloc_list);
                    814:        simple_unlock(&pp->pr_alloc->pa_slock);
1.93      dbj       815:        splx(s);
1.66      thorpej   816:
1.101     thorpej   817:        s = splvm();
                    818:        simple_lock(&pp->pr_slock);
                    819:
1.102     chs       820:        KASSERT(LIST_EMPTY(&pp->pr_cachelist));
1.3       pk        821:
                    822: #ifdef DIAGNOSTIC
1.20      thorpej   823:        if (pp->pr_nout != 0) {
1.25      thorpej   824:                pr_printlog(pp, NULL, printf);
1.80      provos    825:                panic("pool_destroy: pool busy: still out: %u",
1.20      thorpej   826:                    pp->pr_nout);
1.3       pk        827:        }
                    828: #endif
1.1       pk        829:
1.101     thorpej   830:        KASSERT(LIST_EMPTY(&pp->pr_fullpages));
                    831:        KASSERT(LIST_EMPTY(&pp->pr_partpages));
                    832:
1.3       pk        833:        /* Remove all pages */
1.101     thorpej   834:        LIST_INIT(&pq);
1.88      chs       835:        while ((ph = LIST_FIRST(&pp->pr_emptypages)) != NULL)
1.101     thorpej   836:                pr_rmpage(pp, ph, &pq);
                    837:
                    838:        simple_unlock(&pp->pr_slock);
                    839:        splx(s);
1.3       pk        840:
1.101     thorpej   841:        pr_pagelist_free(pp, &pq);
1.3       pk        842:
1.59      thorpej   843: #ifdef POOL_DIAGNOSTIC
1.20      thorpej   844:        if ((pp->pr_roflags & PR_LOGGING) != 0)
1.3       pk        845:                free(pp->pr_log, M_TEMP);
1.59      thorpej   846: #endif
1.1       pk        847: }
                    848:
1.68      thorpej   849: void
                    850: pool_set_drain_hook(struct pool *pp, void (*fn)(void *, int), void *arg)
                    851: {
                    852:
                    853:        /* XXX no locking -- must be used just after pool_init() */
                    854: #ifdef DIAGNOSTIC
                    855:        if (pp->pr_drain_hook != NULL)
                    856:                panic("pool_set_drain_hook(%s): already set", pp->pr_wchan);
                    857: #endif
                    858:        pp->pr_drain_hook = fn;
                    859:        pp->pr_drain_hook_arg = arg;
                    860: }
                    861:
1.88      chs       862: static struct pool_item_header *
1.55      thorpej   863: pool_alloc_item_header(struct pool *pp, caddr_t storage, int flags)
                    864: {
                    865:        struct pool_item_header *ph;
                    866:        int s;
                    867:
                    868:        LOCK_ASSERT(simple_lock_held(&pp->pr_slock) == 0);
                    869:
                    870:        if ((pp->pr_roflags & PR_PHINPAGE) != 0)
                    871:                ph = (struct pool_item_header *) (storage + pp->pr_phoffset);
                    872:        else {
1.85      pk        873:                s = splvm();
1.97      yamt      874:                ph = pool_get(pp->pr_phpool, flags);
1.55      thorpej   875:                splx(s);
                    876:        }
                    877:
                    878:        return (ph);
                    879: }
1.1       pk        880:
                    881: /*
1.3       pk        882:  * Grab an item from the pool; must be called at appropriate spl level
1.1       pk        883:  */
1.3       pk        884: void *
1.59      thorpej   885: #ifdef POOL_DIAGNOSTIC
1.42      thorpej   886: _pool_get(struct pool *pp, int flags, const char *file, long line)
1.56      sommerfe  887: #else
                    888: pool_get(struct pool *pp, int flags)
                    889: #endif
1.1       pk        890: {
                    891:        struct pool_item *pi;
1.3       pk        892:        struct pool_item_header *ph;
1.55      thorpej   893:        void *v;
1.1       pk        894:
1.2       pk        895: #ifdef DIAGNOSTIC
1.95      atatat    896:        if (__predict_false(pp->pr_itemsperpage == 0))
                    897:                panic("pool_get: pool %p: pr_itemsperpage is zero, "
                    898:                    "pool not initialized?", pp);
1.84      thorpej   899:        if (__predict_false(curlwp == NULL && doing_shutdown == 0 &&
1.37      sommerfe  900:                            (flags & PR_WAITOK) != 0))
1.77      matt      901:                panic("pool_get: %s: must have NOWAIT", pp->pr_wchan);
1.58      thorpej   902:
1.102     chs       903: #endif /* DIAGNOSTIC */
1.58      thorpej   904: #ifdef LOCKDEBUG
                    905:        if (flags & PR_WAITOK)
1.119     yamt      906:                ASSERT_SLEEPABLE(NULL, "pool_get(PR_WAITOK)");
1.102     chs       907:        SCHED_ASSERT_UNLOCKED();
1.56      sommerfe  908: #endif
1.1       pk        909:
1.21      thorpej   910:        simple_lock(&pp->pr_slock);
1.25      thorpej   911:        pr_enter(pp, file, line);
1.20      thorpej   912:
                    913:  startover:
                    914:        /*
                    915:         * Check to see if we've reached the hard limit.  If we have,
                    916:         * and we can wait, then wait until an item has been returned to
                    917:         * the pool.
                    918:         */
                    919: #ifdef DIAGNOSTIC
1.34      thorpej   920:        if (__predict_false(pp->pr_nout > pp->pr_hardlimit)) {
1.25      thorpej   921:                pr_leave(pp);
1.21      thorpej   922:                simple_unlock(&pp->pr_slock);
1.20      thorpej   923:                panic("pool_get: %s: crossed hard limit", pp->pr_wchan);
                    924:        }
                    925: #endif
1.34      thorpej   926:        if (__predict_false(pp->pr_nout == pp->pr_hardlimit)) {
1.68      thorpej   927:                if (pp->pr_drain_hook != NULL) {
                    928:                        /*
                    929:                         * Since the drain hook is going to free things
                    930:                         * back to the pool, unlock, call the hook, re-lock,
                    931:                         * and check the hardlimit condition again.
                    932:                         */
                    933:                        pr_leave(pp);
                    934:                        simple_unlock(&pp->pr_slock);
                    935:                        (*pp->pr_drain_hook)(pp->pr_drain_hook_arg, flags);
                    936:                        simple_lock(&pp->pr_slock);
                    937:                        pr_enter(pp, file, line);
                    938:                        if (pp->pr_nout < pp->pr_hardlimit)
                    939:                                goto startover;
                    940:                }
                    941:
1.29      sommerfe  942:                if ((flags & PR_WAITOK) && !(flags & PR_LIMITFAIL)) {
1.20      thorpej   943:                        /*
                    944:                         * XXX: A warning isn't logged in this case.  Should
                    945:                         * it be?
                    946:                         */
                    947:                        pp->pr_flags |= PR_WANTED;
1.25      thorpej   948:                        pr_leave(pp);
1.40      sommerfe  949:                        ltsleep(pp, PSWP, pp->pr_wchan, 0, &pp->pr_slock);
1.25      thorpej   950:                        pr_enter(pp, file, line);
1.20      thorpej   951:                        goto startover;
                    952:                }
1.31      thorpej   953:
                    954:                /*
                    955:                 * Log a message that the hard limit has been hit.
                    956:                 */
                    957:                if (pp->pr_hardlimit_warning != NULL &&
                    958:                    ratecheck(&pp->pr_hardlimit_warning_last,
                    959:                              &pp->pr_hardlimit_ratecap))
                    960:                        log(LOG_ERR, "%s\n", pp->pr_hardlimit_warning);
1.21      thorpej   961:
                    962:                pp->pr_nfail++;
                    963:
1.25      thorpej   964:                pr_leave(pp);
1.21      thorpej   965:                simple_unlock(&pp->pr_slock);
1.20      thorpej   966:                return (NULL);
                    967:        }
                    968:
1.3       pk        969:        /*
                    970:         * The convention we use is that if `curpage' is not NULL, then
                    971:         * it points at a non-empty bucket. In particular, `curpage'
                    972:         * never points at a page header which has PR_PHINPAGE set and
                    973:         * has no items in its bucket.
                    974:         */
1.20      thorpej   975:        if ((ph = pp->pr_curpage) == NULL) {
1.113     yamt      976:                int error;
                    977:
1.20      thorpej   978: #ifdef DIAGNOSTIC
                    979:                if (pp->pr_nitems != 0) {
1.21      thorpej   980:                        simple_unlock(&pp->pr_slock);
1.20      thorpej   981:                        printf("pool_get: %s: curpage NULL, nitems %u\n",
                    982:                            pp->pr_wchan, pp->pr_nitems);
1.80      provos    983:                        panic("pool_get: nitems inconsistent");
1.20      thorpej   984:                }
                    985: #endif
                    986:
1.21      thorpej   987:                /*
                    988:                 * Call the back-end page allocator for more memory.
                    989:                 * Release the pool lock, as the back-end page allocator
                    990:                 * may block.
                    991:                 */
1.25      thorpej   992:                pr_leave(pp);
1.113     yamt      993:                error = pool_grow(pp, flags);
                    994:                pr_enter(pp, file, line);
                    995:                if (error != 0) {
1.21      thorpej   996:                        /*
1.55      thorpej   997:                         * We were unable to allocate a page or item
                    998:                         * header, but we released the lock during
                    999:                         * allocation, so perhaps items were freed
                   1000:                         * back to the pool.  Check for this case.
1.21      thorpej  1001:                         */
                   1002:                        if (pp->pr_curpage != NULL)
                   1003:                                goto startover;
1.15      pk       1004:
1.117     yamt     1005:                        pp->pr_nfail++;
1.25      thorpej  1006:                        pr_leave(pp);
1.117     yamt     1007:                        simple_unlock(&pp->pr_slock);
                   1008:                        return (NULL);
1.1       pk       1009:                }
1.3       pk       1010:
1.20      thorpej  1011:                /* Start the allocation process over. */
                   1012:                goto startover;
1.3       pk       1013:        }
1.97      yamt     1014:        if (pp->pr_roflags & PR_NOTOUCH) {
                   1015: #ifdef DIAGNOSTIC
                   1016:                if (__predict_false(ph->ph_nmissing == pp->pr_itemsperpage)) {
                   1017:                        pr_leave(pp);
                   1018:                        simple_unlock(&pp->pr_slock);
                   1019:                        panic("pool_get: %s: page empty", pp->pr_wchan);
                   1020:                }
                   1021: #endif
                   1022:                v = pr_item_notouch_get(pp, ph);
                   1023: #ifdef POOL_DIAGNOSTIC
                   1024:                pr_log(pp, v, PRLOG_GET, file, line);
                   1025: #endif
                   1026:        } else {
1.102     chs      1027:                v = pi = LIST_FIRST(&ph->ph_itemlist);
1.97      yamt     1028:                if (__predict_false(v == NULL)) {
                   1029:                        pr_leave(pp);
                   1030:                        simple_unlock(&pp->pr_slock);
                   1031:                        panic("pool_get: %s: page empty", pp->pr_wchan);
                   1032:                }
1.20      thorpej  1033: #ifdef DIAGNOSTIC
1.97      yamt     1034:                if (__predict_false(pp->pr_nitems == 0)) {
                   1035:                        pr_leave(pp);
                   1036:                        simple_unlock(&pp->pr_slock);
                   1037:                        printf("pool_get: %s: items on itemlist, nitems %u\n",
                   1038:                            pp->pr_wchan, pp->pr_nitems);
                   1039:                        panic("pool_get: nitems inconsistent");
                   1040:                }
1.65      enami    1041: #endif
1.56      sommerfe 1042:
1.65      enami    1043: #ifdef POOL_DIAGNOSTIC
1.97      yamt     1044:                pr_log(pp, v, PRLOG_GET, file, line);
1.65      enami    1045: #endif
1.3       pk       1046:
1.65      enami    1047: #ifdef DIAGNOSTIC
1.97      yamt     1048:                if (__predict_false(pi->pi_magic != PI_MAGIC)) {
                   1049:                        pr_printlog(pp, pi, printf);
                   1050:                        panic("pool_get(%s): free list modified: "
                   1051:                            "magic=%x; page %p; item addr %p\n",
                   1052:                            pp->pr_wchan, pi->pi_magic, ph->ph_page, pi);
                   1053:                }
1.3       pk       1054: #endif
                   1055:
1.97      yamt     1056:                /*
                   1057:                 * Remove from item list.
                   1058:                 */
1.102     chs      1059:                LIST_REMOVE(pi, pi_list);
1.97      yamt     1060:        }
1.20      thorpej  1061:        pp->pr_nitems--;
                   1062:        pp->pr_nout++;
1.6       thorpej  1063:        if (ph->ph_nmissing == 0) {
                   1064: #ifdef DIAGNOSTIC
1.34      thorpej  1065:                if (__predict_false(pp->pr_nidle == 0))
1.6       thorpej  1066:                        panic("pool_get: nidle inconsistent");
                   1067: #endif
                   1068:                pp->pr_nidle--;
1.88      chs      1069:
                   1070:                /*
                   1071:                 * This page was previously empty.  Move it to the list of
                   1072:                 * partially-full pages.  This page is already curpage.
                   1073:                 */
                   1074:                LIST_REMOVE(ph, ph_pagelist);
                   1075:                LIST_INSERT_HEAD(&pp->pr_partpages, ph, ph_pagelist);
1.6       thorpej  1076:        }
1.3       pk       1077:        ph->ph_nmissing++;
1.97      yamt     1078:        if (ph->ph_nmissing == pp->pr_itemsperpage) {
1.21      thorpej  1079: #ifdef DIAGNOSTIC
1.97      yamt     1080:                if (__predict_false((pp->pr_roflags & PR_NOTOUCH) == 0 &&
1.102     chs      1081:                    !LIST_EMPTY(&ph->ph_itemlist))) {
1.25      thorpej  1082:                        pr_leave(pp);
1.21      thorpej  1083:                        simple_unlock(&pp->pr_slock);
                   1084:                        panic("pool_get: %s: nmissing inconsistent",
                   1085:                            pp->pr_wchan);
                   1086:                }
                   1087: #endif
1.3       pk       1088:                /*
1.88      chs      1089:                 * This page is now full.  Move it to the full list
                   1090:                 * and select a new current page.
1.3       pk       1091:                 */
1.88      chs      1092:                LIST_REMOVE(ph, ph_pagelist);
                   1093:                LIST_INSERT_HEAD(&pp->pr_fullpages, ph, ph_pagelist);
                   1094:                pool_update_curpage(pp);
1.1       pk       1095:        }
1.3       pk       1096:
                   1097:        pp->pr_nget++;
1.111     christos 1098:        pr_leave(pp);
1.20      thorpej  1099:
                   1100:        /*
                   1101:         * If we have a low water mark and we are now below that low
                   1102:         * water mark, add more items to the pool.
                   1103:         */
1.53      thorpej  1104:        if (POOL_NEEDS_CATCHUP(pp) && pool_catchup(pp) != 0) {
1.20      thorpej  1105:                /*
                   1106:                 * XXX: Should we log a warning?  Should we set up a timeout
                   1107:                 * to try again in a second or so?  The latter could break
                   1108:                 * a caller's assumptions about interrupt protection, etc.
                   1109:                 */
                   1110:        }
                   1111:
1.21      thorpej  1112:        simple_unlock(&pp->pr_slock);
1.1       pk       1113:        return (v);
                   1114: }
                   1115:
                   1116: /*
1.43      thorpej  1117:  * Internal version of pool_put().  Pool is already locked/entered.
1.1       pk       1118:  */
1.43      thorpej  1119: static void
1.101     thorpej  1120: pool_do_put(struct pool *pp, void *v, struct pool_pagelist *pq)
1.1       pk       1121: {
                   1122:        struct pool_item *pi = v;
1.3       pk       1123:        struct pool_item_header *ph;
                   1124:        caddr_t page;
                   1125:
1.61      chs      1126:        LOCK_ASSERT(simple_lock_held(&pp->pr_slock));
1.102     chs      1127:        SCHED_ASSERT_UNLOCKED();
1.61      chs      1128:
1.66      thorpej  1129:        page = (caddr_t)((u_long)v & pp->pr_alloc->pa_pagemask);
1.1       pk       1130:
1.30      thorpej  1131: #ifdef DIAGNOSTIC
1.34      thorpej  1132:        if (__predict_false(pp->pr_nout == 0)) {
1.30      thorpej  1133:                printf("pool %s: putting with none out\n",
                   1134:                    pp->pr_wchan);
                   1135:                panic("pool_put");
                   1136:        }
                   1137: #endif
1.3       pk       1138:
1.34      thorpej  1139:        if (__predict_false((ph = pr_find_pagehead(pp, page)) == NULL)) {
1.25      thorpej  1140:                pr_printlog(pp, NULL, printf);
1.3       pk       1141:                panic("pool_put: %s: page header missing", pp->pr_wchan);
                   1142:        }
1.28      thorpej  1143:
                   1144: #ifdef LOCKDEBUG
                   1145:        /*
                   1146:         * Check if we're freeing a locked simple lock.
                   1147:         */
                   1148:        simple_lock_freecheck((caddr_t)pi, ((caddr_t)pi) + pp->pr_size);
                   1149: #endif
1.3       pk       1150:
                   1151:        /*
                   1152:         * Return to item list.
                   1153:         */
1.97      yamt     1154:        if (pp->pr_roflags & PR_NOTOUCH) {
                   1155:                pr_item_notouch_put(pp, ph, v);
                   1156:        } else {
1.2       pk       1157: #ifdef DIAGNOSTIC
1.97      yamt     1158:                pi->pi_magic = PI_MAGIC;
1.3       pk       1159: #endif
1.32      chs      1160: #ifdef DEBUG
1.97      yamt     1161:                {
                   1162:                        int i, *ip = v;
1.32      chs      1163:
1.97      yamt     1164:                        for (i = 0; i < pp->pr_size / sizeof(int); i++) {
                   1165:                                *ip++ = PI_MAGIC;
                   1166:                        }
1.32      chs      1167:                }
                   1168: #endif
                   1169:
1.102     chs      1170:                LIST_INSERT_HEAD(&ph->ph_itemlist, pi, pi_list);
1.97      yamt     1171:        }
1.79      thorpej  1172:        KDASSERT(ph->ph_nmissing != 0);
1.3       pk       1173:        ph->ph_nmissing--;
                   1174:        pp->pr_nput++;
1.20      thorpej  1175:        pp->pr_nitems++;
                   1176:        pp->pr_nout--;
1.3       pk       1177:
                   1178:        /* Cancel "pool empty" condition if it exists */
                   1179:        if (pp->pr_curpage == NULL)
                   1180:                pp->pr_curpage = ph;
                   1181:
                   1182:        if (pp->pr_flags & PR_WANTED) {
                   1183:                pp->pr_flags &= ~PR_WANTED;
1.15      pk       1184:                if (ph->ph_nmissing == 0)
                   1185:                        pp->pr_nidle++;
1.3       pk       1186:                wakeup((caddr_t)pp);
                   1187:                return;
                   1188:        }
                   1189:
                   1190:        /*
1.88      chs      1191:         * If this page is now empty, do one of two things:
1.21      thorpej  1192:         *
1.88      chs      1193:         *      (1) If we have more pages than the page high water mark,
1.96      thorpej  1194:         *          free the page back to the system.  ONLY CONSIDER
1.90      thorpej  1195:         *          FREEING BACK A PAGE IF WE HAVE MORE THAN OUR MINIMUM PAGE
                   1196:         *          CLAIM.
1.21      thorpej  1197:         *
1.88      chs      1198:         *      (2) Otherwise, move the page to the empty page list.
                   1199:         *
                   1200:         * Either way, select a new current page (so we use a partially-full
                   1201:         * page if one is available).
1.3       pk       1202:         */
                   1203:        if (ph->ph_nmissing == 0) {
1.6       thorpej  1204:                pp->pr_nidle++;
1.90      thorpej  1205:                if (pp->pr_npages > pp->pr_minpages &&
                   1206:                    (pp->pr_npages > pp->pr_maxpages ||
1.117     yamt     1207:                     pa_starved_p(pp->pr_alloc))) {
1.101     thorpej  1208:                        pr_rmpage(pp, ph, pq);
1.3       pk       1209:                } else {
1.88      chs      1210:                        LIST_REMOVE(ph, ph_pagelist);
                   1211:                        LIST_INSERT_HEAD(&pp->pr_emptypages, ph, ph_pagelist);
1.3       pk       1212:
1.21      thorpej  1213:                        /*
                   1214:                         * Update the timestamp on the page.  A page must
                   1215:                         * be idle for some period of time before it can
                   1216:                         * be reclaimed by the pagedaemon.  This minimizes
                   1217:                         * ping-pong'ing for memory.
                   1218:                         */
1.118     kardel   1219:                        getmicrotime(&ph->ph_time);
1.1       pk       1220:                }
1.88      chs      1221:                pool_update_curpage(pp);
1.1       pk       1222:        }
1.88      chs      1223:
1.21      thorpej  1224:        /*
1.88      chs      1225:         * If the page was previously completely full, move it to the
                   1226:         * partially-full list and make it the current page.  The next
                   1227:         * allocation will get the item from this page, instead of
                   1228:         * further fragmenting the pool.
1.21      thorpej  1229:         */
                   1230:        else if (ph->ph_nmissing == (pp->pr_itemsperpage - 1)) {
1.88      chs      1231:                LIST_REMOVE(ph, ph_pagelist);
                   1232:                LIST_INSERT_HEAD(&pp->pr_partpages, ph, ph_pagelist);
1.21      thorpej  1233:                pp->pr_curpage = ph;
                   1234:        }
1.43      thorpej  1235: }
                   1236:
                   1237: /*
                   1238:  * Return resource to the pool; must be called at appropriate spl level
                   1239:  */
1.59      thorpej  1240: #ifdef POOL_DIAGNOSTIC
1.43      thorpej  1241: void
                   1242: _pool_put(struct pool *pp, void *v, const char *file, long line)
                   1243: {
1.101     thorpej  1244:        struct pool_pagelist pq;
                   1245:
                   1246:        LIST_INIT(&pq);
1.43      thorpej  1247:
                   1248:        simple_lock(&pp->pr_slock);
                   1249:        pr_enter(pp, file, line);
                   1250:
1.56      sommerfe 1251:        pr_log(pp, v, PRLOG_PUT, file, line);
                   1252:
1.101     thorpej  1253:        pool_do_put(pp, v, &pq);
1.21      thorpej  1254:
1.25      thorpej  1255:        pr_leave(pp);
1.21      thorpej  1256:        simple_unlock(&pp->pr_slock);
1.101     thorpej  1257:
1.102     chs      1258:        pr_pagelist_free(pp, &pq);
1.1       pk       1259: }
1.57      sommerfe 1260: #undef pool_put
1.59      thorpej  1261: #endif /* POOL_DIAGNOSTIC */
1.1       pk       1262:
1.56      sommerfe 1263: void
                   1264: pool_put(struct pool *pp, void *v)
                   1265: {
1.101     thorpej  1266:        struct pool_pagelist pq;
                   1267:
                   1268:        LIST_INIT(&pq);
1.56      sommerfe 1269:
                   1270:        simple_lock(&pp->pr_slock);
1.101     thorpej  1271:        pool_do_put(pp, v, &pq);
                   1272:        simple_unlock(&pp->pr_slock);
1.56      sommerfe 1273:
1.102     chs      1274:        pr_pagelist_free(pp, &pq);
1.56      sommerfe 1275: }
1.57      sommerfe 1276:
1.59      thorpej  1277: #ifdef POOL_DIAGNOSTIC
1.57      sommerfe 1278: #define                pool_put(h, v)  _pool_put((h), (v), __FILE__, __LINE__)
1.56      sommerfe 1279: #endif
1.74      thorpej  1280:
                   1281: /*
1.113     yamt     1282:  * pool_grow: grow a pool by a page.
                   1283:  *
                   1284:  * => called with pool locked.
                   1285:  * => unlock and relock the pool.
                   1286:  * => return with pool locked.
                   1287:  */
                   1288:
                   1289: static int
                   1290: pool_grow(struct pool *pp, int flags)
                   1291: {
                   1292:        struct pool_item_header *ph = NULL;
                   1293:        char *cp;
                   1294:
                   1295:        simple_unlock(&pp->pr_slock);
                   1296:        cp = pool_allocator_alloc(pp, flags);
                   1297:        if (__predict_true(cp != NULL)) {
                   1298:                ph = pool_alloc_item_header(pp, cp, flags);
                   1299:        }
                   1300:        if (__predict_false(cp == NULL || ph == NULL)) {
                   1301:                if (cp != NULL) {
                   1302:                        pool_allocator_free(pp, cp);
                   1303:                }
                   1304:                simple_lock(&pp->pr_slock);
                   1305:                return ENOMEM;
                   1306:        }
                   1307:
                   1308:        simple_lock(&pp->pr_slock);
                   1309:        pool_prime_page(pp, cp, ph);
                   1310:        pp->pr_npagealloc++;
                   1311:        return 0;
                   1312: }
                   1313:
                   1314: /*
1.74      thorpej  1315:  * Add N items to the pool.
                   1316:  */
                   1317: int
                   1318: pool_prime(struct pool *pp, int n)
                   1319: {
1.75      simonb   1320:        int newpages;
1.113     yamt     1321:        int error = 0;
1.74      thorpej  1322:
                   1323:        simple_lock(&pp->pr_slock);
                   1324:
                   1325:        newpages = roundup(n, pp->pr_itemsperpage) / pp->pr_itemsperpage;
                   1326:
                   1327:        while (newpages-- > 0) {
1.113     yamt     1328:                error = pool_grow(pp, PR_NOWAIT);
                   1329:                if (error) {
1.74      thorpej  1330:                        break;
                   1331:                }
                   1332:                pp->pr_minpages++;
                   1333:        }
                   1334:
                   1335:        if (pp->pr_minpages >= pp->pr_maxpages)
                   1336:                pp->pr_maxpages = pp->pr_minpages + 1;  /* XXX */
                   1337:
                   1338:        simple_unlock(&pp->pr_slock);
1.113     yamt     1339:        return error;
1.74      thorpej  1340: }
1.55      thorpej  1341:
                   1342: /*
1.3       pk       1343:  * Add a page worth of items to the pool.
1.21      thorpej  1344:  *
                   1345:  * Note, we must be called with the pool descriptor LOCKED.
1.3       pk       1346:  */
1.55      thorpej  1347: static void
                   1348: pool_prime_page(struct pool *pp, caddr_t storage, struct pool_item_header *ph)
1.3       pk       1349: {
                   1350:        struct pool_item *pi;
                   1351:        caddr_t cp = storage;
                   1352:        unsigned int align = pp->pr_align;
                   1353:        unsigned int ioff = pp->pr_itemoffset;
1.55      thorpej  1354:        int n;
1.36      pk       1355:
1.91      yamt     1356:        LOCK_ASSERT(simple_lock_held(&pp->pr_slock));
                   1357:
1.66      thorpej  1358: #ifdef DIAGNOSTIC
                   1359:        if (((u_long)cp & (pp->pr_alloc->pa_pagesz - 1)) != 0)
1.36      pk       1360:                panic("pool_prime_page: %s: unaligned page", pp->pr_wchan);
1.66      thorpej  1361: #endif
1.3       pk       1362:
                   1363:        /*
                   1364:         * Insert page header.
                   1365:         */
1.88      chs      1366:        LIST_INSERT_HEAD(&pp->pr_emptypages, ph, ph_pagelist);
1.102     chs      1367:        LIST_INIT(&ph->ph_itemlist);
1.3       pk       1368:        ph->ph_page = storage;
                   1369:        ph->ph_nmissing = 0;
1.118     kardel   1370:        getmicrotime(&ph->ph_time);
1.88      chs      1371:        if ((pp->pr_roflags & PR_PHINPAGE) == 0)
                   1372:                SPLAY_INSERT(phtree, &pp->pr_phtree, ph);
1.3       pk       1373:
1.6       thorpej  1374:        pp->pr_nidle++;
                   1375:
1.3       pk       1376:        /*
                   1377:         * Color this page.
                   1378:         */
                   1379:        cp = (caddr_t)(cp + pp->pr_curcolor);
                   1380:        if ((pp->pr_curcolor += align) > pp->pr_maxcolor)
                   1381:                pp->pr_curcolor = 0;
                   1382:
                   1383:        /*
                   1384:         * Adjust storage to apply aligment to `pr_itemoffset' in each item.
                   1385:         */
                   1386:        if (ioff != 0)
                   1387:                cp = (caddr_t)(cp + (align - ioff));
                   1388:
                   1389:        /*
                   1390:         * Insert remaining chunks on the bucket list.
                   1391:         */
                   1392:        n = pp->pr_itemsperpage;
1.20      thorpej  1393:        pp->pr_nitems += n;
1.3       pk       1394:
1.97      yamt     1395:        if (pp->pr_roflags & PR_NOTOUCH) {
1.99      yamt     1396:                pool_item_freelist_t *freelist = PR_FREELIST(ph);
1.97      yamt     1397:                int i;
                   1398:
1.99      yamt     1399:                ph->ph_off = cp - storage;
1.97      yamt     1400:                ph->ph_firstfree = 0;
                   1401:                for (i = 0; i < n - 1; i++)
                   1402:                        freelist[i] = i + 1;
                   1403:                freelist[n - 1] = PR_INDEX_EOL;
                   1404:        } else {
                   1405:                while (n--) {
                   1406:                        pi = (struct pool_item *)cp;
1.78      thorpej  1407:
1.97      yamt     1408:                        KASSERT(((((vaddr_t)pi) + ioff) & (align - 1)) == 0);
1.3       pk       1409:
1.97      yamt     1410:                        /* Insert on page list */
1.102     chs      1411:                        LIST_INSERT_HEAD(&ph->ph_itemlist, pi, pi_list);
1.3       pk       1412: #ifdef DIAGNOSTIC
1.97      yamt     1413:                        pi->pi_magic = PI_MAGIC;
1.3       pk       1414: #endif
1.97      yamt     1415:                        cp = (caddr_t)(cp + pp->pr_size);
                   1416:                }
1.3       pk       1417:        }
                   1418:
                   1419:        /*
                   1420:         * If the pool was depleted, point at the new page.
                   1421:         */
                   1422:        if (pp->pr_curpage == NULL)
                   1423:                pp->pr_curpage = ph;
                   1424:
                   1425:        if (++pp->pr_npages > pp->pr_hiwat)
                   1426:                pp->pr_hiwat = pp->pr_npages;
                   1427: }
                   1428:
1.20      thorpej  1429: /*
1.52      thorpej  1430:  * Used by pool_get() when nitems drops below the low water mark.  This
1.88      chs      1431:  * is used to catch up pr_nitems with the low water mark.
1.20      thorpej  1432:  *
1.21      thorpej  1433:  * Note 1, we never wait for memory here, we let the caller decide what to do.
1.20      thorpej  1434:  *
1.73      thorpej  1435:  * Note 2, we must be called with the pool already locked, and we return
1.20      thorpej  1436:  * with it locked.
                   1437:  */
                   1438: static int
1.42      thorpej  1439: pool_catchup(struct pool *pp)
1.20      thorpej  1440: {
                   1441:        int error = 0;
                   1442:
1.54      thorpej  1443:        while (POOL_NEEDS_CATCHUP(pp)) {
1.113     yamt     1444:                error = pool_grow(pp, PR_NOWAIT);
                   1445:                if (error) {
1.20      thorpej  1446:                        break;
                   1447:                }
                   1448:        }
1.113     yamt     1449:        return error;
1.20      thorpej  1450: }
                   1451:
1.88      chs      1452: static void
                   1453: pool_update_curpage(struct pool *pp)
                   1454: {
                   1455:
                   1456:        pp->pr_curpage = LIST_FIRST(&pp->pr_partpages);
                   1457:        if (pp->pr_curpage == NULL) {
                   1458:                pp->pr_curpage = LIST_FIRST(&pp->pr_emptypages);
                   1459:        }
                   1460: }
                   1461:
1.3       pk       1462: void
1.42      thorpej  1463: pool_setlowat(struct pool *pp, int n)
1.3       pk       1464: {
1.15      pk       1465:
1.21      thorpej  1466:        simple_lock(&pp->pr_slock);
                   1467:
1.3       pk       1468:        pp->pr_minitems = n;
1.15      pk       1469:        pp->pr_minpages = (n == 0)
                   1470:                ? 0
1.18      thorpej  1471:                : roundup(n, pp->pr_itemsperpage) / pp->pr_itemsperpage;
1.20      thorpej  1472:
                   1473:        /* Make sure we're caught up with the newly-set low water mark. */
1.75      simonb   1474:        if (POOL_NEEDS_CATCHUP(pp) && pool_catchup(pp) != 0) {
1.20      thorpej  1475:                /*
                   1476:                 * XXX: Should we log a warning?  Should we set up a timeout
                   1477:                 * to try again in a second or so?  The latter could break
                   1478:                 * a caller's assumptions about interrupt protection, etc.
                   1479:                 */
                   1480:        }
1.21      thorpej  1481:
                   1482:        simple_unlock(&pp->pr_slock);
1.3       pk       1483: }
                   1484:
                   1485: void
1.42      thorpej  1486: pool_sethiwat(struct pool *pp, int n)
1.3       pk       1487: {
1.15      pk       1488:
1.21      thorpej  1489:        simple_lock(&pp->pr_slock);
                   1490:
1.15      pk       1491:        pp->pr_maxpages = (n == 0)
                   1492:                ? 0
1.18      thorpej  1493:                : roundup(n, pp->pr_itemsperpage) / pp->pr_itemsperpage;
1.21      thorpej  1494:
                   1495:        simple_unlock(&pp->pr_slock);
1.3       pk       1496: }
                   1497:
1.20      thorpej  1498: void
1.42      thorpej  1499: pool_sethardlimit(struct pool *pp, int n, const char *warnmess, int ratecap)
1.20      thorpej  1500: {
                   1501:
1.21      thorpej  1502:        simple_lock(&pp->pr_slock);
1.20      thorpej  1503:
                   1504:        pp->pr_hardlimit = n;
                   1505:        pp->pr_hardlimit_warning = warnmess;
1.31      thorpej  1506:        pp->pr_hardlimit_ratecap.tv_sec = ratecap;
                   1507:        pp->pr_hardlimit_warning_last.tv_sec = 0;
                   1508:        pp->pr_hardlimit_warning_last.tv_usec = 0;
1.20      thorpej  1509:
                   1510:        /*
1.21      thorpej  1511:         * In-line version of pool_sethiwat(), because we don't want to
                   1512:         * release the lock.
1.20      thorpej  1513:         */
                   1514:        pp->pr_maxpages = (n == 0)
                   1515:                ? 0
                   1516:                : roundup(n, pp->pr_itemsperpage) / pp->pr_itemsperpage;
1.21      thorpej  1517:
                   1518:        simple_unlock(&pp->pr_slock);
1.20      thorpej  1519: }
1.3       pk       1520:
                   1521: /*
                   1522:  * Release all complete pages that have not been used recently.
                   1523:  */
1.66      thorpej  1524: int
1.59      thorpej  1525: #ifdef POOL_DIAGNOSTIC
1.42      thorpej  1526: _pool_reclaim(struct pool *pp, const char *file, long line)
1.56      sommerfe 1527: #else
                   1528: pool_reclaim(struct pool *pp)
                   1529: #endif
1.3       pk       1530: {
                   1531:        struct pool_item_header *ph, *phnext;
1.43      thorpej  1532:        struct pool_cache *pc;
1.61      chs      1533:        struct pool_pagelist pq;
1.102     chs      1534:        struct pool_cache_grouplist pcgl;
                   1535:        struct timeval curtime, diff;
1.3       pk       1536:
1.68      thorpej  1537:        if (pp->pr_drain_hook != NULL) {
                   1538:                /*
                   1539:                 * The drain hook must be called with the pool unlocked.
                   1540:                 */
                   1541:                (*pp->pr_drain_hook)(pp->pr_drain_hook_arg, PR_NOWAIT);
                   1542:        }
                   1543:
1.21      thorpej  1544:        if (simple_lock_try(&pp->pr_slock) == 0)
1.66      thorpej  1545:                return (0);
1.25      thorpej  1546:        pr_enter(pp, file, line);
1.68      thorpej  1547:
1.88      chs      1548:        LIST_INIT(&pq);
1.102     chs      1549:        LIST_INIT(&pcgl);
1.3       pk       1550:
1.43      thorpej  1551:        /*
                   1552:         * Reclaim items from the pool's caches.
                   1553:         */
1.102     chs      1554:        LIST_FOREACH(pc, &pp->pr_cachelist, pc_poollist)
                   1555:                pool_cache_reclaim(pc, &pq, &pcgl);
1.43      thorpej  1556:
1.118     kardel   1557:        getmicrotime(&curtime);
1.21      thorpej  1558:
1.88      chs      1559:        for (ph = LIST_FIRST(&pp->pr_emptypages); ph != NULL; ph = phnext) {
                   1560:                phnext = LIST_NEXT(ph, ph_pagelist);
1.3       pk       1561:
                   1562:                /* Check our minimum page claim */
                   1563:                if (pp->pr_npages <= pp->pr_minpages)
                   1564:                        break;
                   1565:
1.88      chs      1566:                KASSERT(ph->ph_nmissing == 0);
                   1567:                timersub(&curtime, &ph->ph_time, &diff);
1.117     yamt     1568:                if (diff.tv_sec < pool_inactive_time
                   1569:                    && !pa_starved_p(pp->pr_alloc))
1.88      chs      1570:                        continue;
1.21      thorpej  1571:
1.88      chs      1572:                /*
                   1573:                 * If freeing this page would put us below
                   1574:                 * the low water mark, stop now.
                   1575:                 */
                   1576:                if ((pp->pr_nitems - pp->pr_itemsperpage) <
                   1577:                    pp->pr_minitems)
                   1578:                        break;
1.21      thorpej  1579:
1.88      chs      1580:                pr_rmpage(pp, ph, &pq);
1.3       pk       1581:        }
                   1582:
1.25      thorpej  1583:        pr_leave(pp);
1.21      thorpej  1584:        simple_unlock(&pp->pr_slock);
1.102     chs      1585:        if (LIST_EMPTY(&pq) && LIST_EMPTY(&pcgl))
                   1586:                return 0;
1.66      thorpej  1587:
1.101     thorpej  1588:        pr_pagelist_free(pp, &pq);
1.102     chs      1589:        pcg_grouplist_free(&pcgl);
1.66      thorpej  1590:        return (1);
1.3       pk       1591: }
                   1592:
                   1593: /*
                   1594:  * Drain pools, one at a time.
1.21      thorpej  1595:  *
                   1596:  * Note, we must never be called from an interrupt context.
1.3       pk       1597:  */
                   1598: void
1.42      thorpej  1599: pool_drain(void *arg)
1.3       pk       1600: {
                   1601:        struct pool *pp;
1.23      thorpej  1602:        int s;
1.3       pk       1603:
1.61      chs      1604:        pp = NULL;
1.49      thorpej  1605:        s = splvm();
1.23      thorpej  1606:        simple_lock(&pool_head_slock);
1.61      chs      1607:        if (drainpp == NULL) {
1.102     chs      1608:                drainpp = LIST_FIRST(&pool_head);
1.61      chs      1609:        }
                   1610:        if (drainpp) {
                   1611:                pp = drainpp;
1.102     chs      1612:                drainpp = LIST_NEXT(pp, pr_poollist);
1.61      chs      1613:        }
                   1614:        simple_unlock(&pool_head_slock);
1.115     christos 1615:        if (pp)
                   1616:                pool_reclaim(pp);
1.61      chs      1617:        splx(s);
1.3       pk       1618: }
                   1619:
                   1620: /*
                   1621:  * Diagnostic helpers.
                   1622:  */
                   1623: void
1.42      thorpej  1624: pool_print(struct pool *pp, const char *modif)
1.21      thorpej  1625: {
                   1626:        int s;
                   1627:
1.49      thorpej  1628:        s = splvm();
1.25      thorpej  1629:        if (simple_lock_try(&pp->pr_slock) == 0) {
                   1630:                printf("pool %s is locked; try again later\n",
                   1631:                    pp->pr_wchan);
                   1632:                splx(s);
                   1633:                return;
                   1634:        }
                   1635:        pool_print1(pp, modif, printf);
1.21      thorpej  1636:        simple_unlock(&pp->pr_slock);
                   1637:        splx(s);
                   1638: }
                   1639:
1.25      thorpej  1640: void
1.108     yamt     1641: pool_printall(const char *modif, void (*pr)(const char *, ...))
                   1642: {
                   1643:        struct pool *pp;
                   1644:
                   1645:        if (simple_lock_try(&pool_head_slock) == 0) {
                   1646:                (*pr)("WARNING: pool_head_slock is locked\n");
                   1647:        } else {
                   1648:                simple_unlock(&pool_head_slock);
                   1649:        }
                   1650:
                   1651:        LIST_FOREACH(pp, &pool_head, pr_poollist) {
                   1652:                pool_printit(pp, modif, pr);
                   1653:        }
                   1654: }
                   1655:
                   1656: void
1.42      thorpej  1657: pool_printit(struct pool *pp, const char *modif, void (*pr)(const char *, ...))
1.25      thorpej  1658: {
                   1659:
                   1660:        if (pp == NULL) {
                   1661:                (*pr)("Must specify a pool to print.\n");
                   1662:                return;
                   1663:        }
                   1664:
                   1665:        /*
                   1666:         * Called from DDB; interrupts should be blocked, and all
                   1667:         * other processors should be paused.  We can skip locking
                   1668:         * the pool in this case.
                   1669:         *
                   1670:         * We do a simple_lock_try() just to print the lock
                   1671:         * status, however.
                   1672:         */
                   1673:
                   1674:        if (simple_lock_try(&pp->pr_slock) == 0)
                   1675:                (*pr)("WARNING: pool %s is locked\n", pp->pr_wchan);
                   1676:        else
1.107     yamt     1677:                simple_unlock(&pp->pr_slock);
1.25      thorpej  1678:
                   1679:        pool_print1(pp, modif, pr);
                   1680: }
                   1681:
1.21      thorpej  1682: static void
1.97      yamt     1683: pool_print_pagelist(struct pool *pp, struct pool_pagelist *pl,
                   1684:     void (*pr)(const char *, ...))
1.88      chs      1685: {
                   1686:        struct pool_item_header *ph;
                   1687: #ifdef DIAGNOSTIC
                   1688:        struct pool_item *pi;
                   1689: #endif
                   1690:
                   1691:        LIST_FOREACH(ph, pl, ph_pagelist) {
                   1692:                (*pr)("\t\tpage %p, nmissing %d, time %lu,%lu\n",
                   1693:                    ph->ph_page, ph->ph_nmissing,
                   1694:                    (u_long)ph->ph_time.tv_sec,
                   1695:                    (u_long)ph->ph_time.tv_usec);
                   1696: #ifdef DIAGNOSTIC
1.97      yamt     1697:                if (!(pp->pr_roflags & PR_NOTOUCH)) {
1.102     chs      1698:                        LIST_FOREACH(pi, &ph->ph_itemlist, pi_list) {
1.97      yamt     1699:                                if (pi->pi_magic != PI_MAGIC) {
                   1700:                                        (*pr)("\t\t\titem %p, magic 0x%x\n",
                   1701:                                            pi, pi->pi_magic);
                   1702:                                }
1.88      chs      1703:                        }
                   1704:                }
                   1705: #endif
                   1706:        }
                   1707: }
                   1708:
                   1709: static void
1.42      thorpej  1710: pool_print1(struct pool *pp, const char *modif, void (*pr)(const char *, ...))
1.3       pk       1711: {
1.25      thorpej  1712:        struct pool_item_header *ph;
1.44      thorpej  1713:        struct pool_cache *pc;
                   1714:        struct pool_cache_group *pcg;
                   1715:        int i, print_log = 0, print_pagelist = 0, print_cache = 0;
1.25      thorpej  1716:        char c;
                   1717:
                   1718:        while ((c = *modif++) != '\0') {
                   1719:                if (c == 'l')
                   1720:                        print_log = 1;
                   1721:                if (c == 'p')
                   1722:                        print_pagelist = 1;
1.44      thorpej  1723:                if (c == 'c')
                   1724:                        print_cache = 1;
1.25      thorpej  1725:        }
                   1726:
                   1727:        (*pr)("POOL %s: size %u, align %u, ioff %u, roflags 0x%08x\n",
                   1728:            pp->pr_wchan, pp->pr_size, pp->pr_align, pp->pr_itemoffset,
                   1729:            pp->pr_roflags);
1.66      thorpej  1730:        (*pr)("\talloc %p\n", pp->pr_alloc);
1.25      thorpej  1731:        (*pr)("\tminitems %u, minpages %u, maxpages %u, npages %u\n",
                   1732:            pp->pr_minitems, pp->pr_minpages, pp->pr_maxpages, pp->pr_npages);
                   1733:        (*pr)("\titemsperpage %u, nitems %u, nout %u, hardlimit %u\n",
                   1734:            pp->pr_itemsperpage, pp->pr_nitems, pp->pr_nout, pp->pr_hardlimit);
                   1735:
                   1736:        (*pr)("\n\tnget %lu, nfail %lu, nput %lu\n",
                   1737:            pp->pr_nget, pp->pr_nfail, pp->pr_nput);
                   1738:        (*pr)("\tnpagealloc %lu, npagefree %lu, hiwat %u, nidle %lu\n",
                   1739:            pp->pr_npagealloc, pp->pr_npagefree, pp->pr_hiwat, pp->pr_nidle);
                   1740:
                   1741:        if (print_pagelist == 0)
                   1742:                goto skip_pagelist;
                   1743:
1.88      chs      1744:        if ((ph = LIST_FIRST(&pp->pr_emptypages)) != NULL)
                   1745:                (*pr)("\n\tempty page list:\n");
1.97      yamt     1746:        pool_print_pagelist(pp, &pp->pr_emptypages, pr);
1.88      chs      1747:        if ((ph = LIST_FIRST(&pp->pr_fullpages)) != NULL)
                   1748:                (*pr)("\n\tfull page list:\n");
1.97      yamt     1749:        pool_print_pagelist(pp, &pp->pr_fullpages, pr);
1.88      chs      1750:        if ((ph = LIST_FIRST(&pp->pr_partpages)) != NULL)
                   1751:                (*pr)("\n\tpartial-page list:\n");
1.97      yamt     1752:        pool_print_pagelist(pp, &pp->pr_partpages, pr);
1.88      chs      1753:
1.25      thorpej  1754:        if (pp->pr_curpage == NULL)
                   1755:                (*pr)("\tno current page\n");
                   1756:        else
                   1757:                (*pr)("\tcurpage %p\n", pp->pr_curpage->ph_page);
                   1758:
                   1759:  skip_pagelist:
                   1760:        if (print_log == 0)
                   1761:                goto skip_log;
                   1762:
                   1763:        (*pr)("\n");
                   1764:        if ((pp->pr_roflags & PR_LOGGING) == 0)
                   1765:                (*pr)("\tno log\n");
                   1766:        else
                   1767:                pr_printlog(pp, NULL, pr);
1.3       pk       1768:
1.25      thorpej  1769:  skip_log:
1.44      thorpej  1770:        if (print_cache == 0)
                   1771:                goto skip_cache;
                   1772:
1.102     chs      1773: #define PR_GROUPLIST(pcg)                                              \
                   1774:        (*pr)("\t\tgroup %p: avail %d\n", pcg, pcg->pcg_avail);         \
                   1775:        for (i = 0; i < PCG_NOBJECTS; i++) {                            \
                   1776:                if (pcg->pcg_objects[i].pcgo_pa !=                      \
                   1777:                    POOL_PADDR_INVALID) {                               \
                   1778:                        (*pr)("\t\t\t%p, 0x%llx\n",                     \
                   1779:                            pcg->pcg_objects[i].pcgo_va,                \
                   1780:                            (unsigned long long)                        \
                   1781:                            pcg->pcg_objects[i].pcgo_pa);               \
                   1782:                } else {                                                \
                   1783:                        (*pr)("\t\t\t%p\n",                             \
                   1784:                            pcg->pcg_objects[i].pcgo_va);               \
                   1785:                }                                                       \
                   1786:        }
                   1787:
                   1788:        LIST_FOREACH(pc, &pp->pr_cachelist, pc_poollist) {
1.103     chs      1789:                (*pr)("\tcache %p\n", pc);
1.48      thorpej  1790:                (*pr)("\t    hits %lu misses %lu ngroups %lu nitems %lu\n",
                   1791:                    pc->pc_hits, pc->pc_misses, pc->pc_ngroups, pc->pc_nitems);
1.102     chs      1792:                (*pr)("\t    full groups:\n");
1.103     chs      1793:                LIST_FOREACH(pcg, &pc->pc_fullgroups, pcg_list) {
1.102     chs      1794:                        PR_GROUPLIST(pcg);
1.103     chs      1795:                }
1.102     chs      1796:                (*pr)("\t    partial groups:\n");
1.103     chs      1797:                LIST_FOREACH(pcg, &pc->pc_partgroups, pcg_list) {
1.102     chs      1798:                        PR_GROUPLIST(pcg);
1.103     chs      1799:                }
1.102     chs      1800:                (*pr)("\t    empty groups:\n");
1.103     chs      1801:                LIST_FOREACH(pcg, &pc->pc_emptygroups, pcg_list) {
1.102     chs      1802:                        PR_GROUPLIST(pcg);
1.103     chs      1803:                }
1.44      thorpej  1804:        }
1.102     chs      1805: #undef PR_GROUPLIST
1.44      thorpej  1806:
                   1807:  skip_cache:
1.88      chs      1808:        pr_enter_check(pp, pr);
                   1809: }
                   1810:
                   1811: static int
                   1812: pool_chk_page(struct pool *pp, const char *label, struct pool_item_header *ph)
                   1813: {
                   1814:        struct pool_item *pi;
                   1815:        caddr_t page;
                   1816:        int n;
                   1817:
                   1818:        page = (caddr_t)((u_long)ph & pp->pr_alloc->pa_pagemask);
                   1819:        if (page != ph->ph_page &&
                   1820:            (pp->pr_roflags & PR_PHINPAGE) != 0) {
                   1821:                if (label != NULL)
                   1822:                        printf("%s: ", label);
                   1823:                printf("pool(%p:%s): page inconsistency: page %p;"
                   1824:                       " at page head addr %p (p %p)\n", pp,
                   1825:                        pp->pr_wchan, ph->ph_page,
                   1826:                        ph, page);
                   1827:                return 1;
                   1828:        }
1.3       pk       1829:
1.97      yamt     1830:        if ((pp->pr_roflags & PR_NOTOUCH) != 0)
                   1831:                return 0;
                   1832:
1.102     chs      1833:        for (pi = LIST_FIRST(&ph->ph_itemlist), n = 0;
1.88      chs      1834:             pi != NULL;
1.102     chs      1835:             pi = LIST_NEXT(pi,pi_list), n++) {
1.88      chs      1836:
                   1837: #ifdef DIAGNOSTIC
                   1838:                if (pi->pi_magic != PI_MAGIC) {
                   1839:                        if (label != NULL)
                   1840:                                printf("%s: ", label);
                   1841:                        printf("pool(%s): free list modified: magic=%x;"
                   1842:                               " page %p; item ordinal %d;"
                   1843:                               " addr %p (p %p)\n",
                   1844:                                pp->pr_wchan, pi->pi_magic, ph->ph_page,
                   1845:                                n, pi, page);
                   1846:                        panic("pool");
                   1847:                }
                   1848: #endif
                   1849:                page =
                   1850:                    (caddr_t)((u_long)pi & pp->pr_alloc->pa_pagemask);
                   1851:                if (page == ph->ph_page)
                   1852:                        continue;
                   1853:
                   1854:                if (label != NULL)
                   1855:                        printf("%s: ", label);
                   1856:                printf("pool(%p:%s): page inconsistency: page %p;"
                   1857:                       " item ordinal %d; addr %p (p %p)\n", pp,
                   1858:                        pp->pr_wchan, ph->ph_page,
                   1859:                        n, pi, page);
                   1860:                return 1;
                   1861:        }
                   1862:        return 0;
1.3       pk       1863: }
                   1864:
1.88      chs      1865:
1.3       pk       1866: int
1.42      thorpej  1867: pool_chk(struct pool *pp, const char *label)
1.3       pk       1868: {
                   1869:        struct pool_item_header *ph;
                   1870:        int r = 0;
                   1871:
1.21      thorpej  1872:        simple_lock(&pp->pr_slock);
1.88      chs      1873:        LIST_FOREACH(ph, &pp->pr_emptypages, ph_pagelist) {
                   1874:                r = pool_chk_page(pp, label, ph);
                   1875:                if (r) {
                   1876:                        goto out;
                   1877:                }
                   1878:        }
                   1879:        LIST_FOREACH(ph, &pp->pr_fullpages, ph_pagelist) {
                   1880:                r = pool_chk_page(pp, label, ph);
                   1881:                if (r) {
1.3       pk       1882:                        goto out;
                   1883:                }
1.88      chs      1884:        }
                   1885:        LIST_FOREACH(ph, &pp->pr_partpages, ph_pagelist) {
                   1886:                r = pool_chk_page(pp, label, ph);
                   1887:                if (r) {
1.3       pk       1888:                        goto out;
                   1889:                }
                   1890:        }
1.88      chs      1891:
1.3       pk       1892: out:
1.21      thorpej  1893:        simple_unlock(&pp->pr_slock);
1.3       pk       1894:        return (r);
1.43      thorpej  1895: }
                   1896:
                   1897: /*
                   1898:  * pool_cache_init:
                   1899:  *
                   1900:  *     Initialize a pool cache.
                   1901:  *
                   1902:  *     NOTE: If the pool must be protected from interrupts, we expect
                   1903:  *     to be called at the appropriate interrupt priority level.
                   1904:  */
                   1905: void
                   1906: pool_cache_init(struct pool_cache *pc, struct pool *pp,
                   1907:     int (*ctor)(void *, void *, int),
                   1908:     void (*dtor)(void *, void *),
                   1909:     void *arg)
                   1910: {
                   1911:
1.102     chs      1912:        LIST_INIT(&pc->pc_emptygroups);
                   1913:        LIST_INIT(&pc->pc_fullgroups);
                   1914:        LIST_INIT(&pc->pc_partgroups);
1.43      thorpej  1915:        simple_lock_init(&pc->pc_slock);
                   1916:
                   1917:        pc->pc_pool = pp;
                   1918:
                   1919:        pc->pc_ctor = ctor;
                   1920:        pc->pc_dtor = dtor;
                   1921:        pc->pc_arg  = arg;
                   1922:
1.48      thorpej  1923:        pc->pc_hits   = 0;
                   1924:        pc->pc_misses = 0;
                   1925:
                   1926:        pc->pc_ngroups = 0;
                   1927:
                   1928:        pc->pc_nitems = 0;
                   1929:
1.43      thorpej  1930:        simple_lock(&pp->pr_slock);
1.102     chs      1931:        LIST_INSERT_HEAD(&pp->pr_cachelist, pc, pc_poollist);
1.43      thorpej  1932:        simple_unlock(&pp->pr_slock);
                   1933: }
                   1934:
                   1935: /*
                   1936:  * pool_cache_destroy:
                   1937:  *
                   1938:  *     Destroy a pool cache.
                   1939:  */
                   1940: void
                   1941: pool_cache_destroy(struct pool_cache *pc)
                   1942: {
                   1943:        struct pool *pp = pc->pc_pool;
                   1944:
                   1945:        /* First, invalidate the entire cache. */
                   1946:        pool_cache_invalidate(pc);
                   1947:
                   1948:        /* ...and remove it from the pool's cache list. */
                   1949:        simple_lock(&pp->pr_slock);
1.102     chs      1950:        LIST_REMOVE(pc, pc_poollist);
1.43      thorpej  1951:        simple_unlock(&pp->pr_slock);
                   1952: }
                   1953:
1.110     perry    1954: static inline void *
1.87      thorpej  1955: pcg_get(struct pool_cache_group *pcg, paddr_t *pap)
1.43      thorpej  1956: {
                   1957:        void *object;
                   1958:        u_int idx;
                   1959:
                   1960:        KASSERT(pcg->pcg_avail <= PCG_NOBJECTS);
1.45      thorpej  1961:        KASSERT(pcg->pcg_avail != 0);
1.43      thorpej  1962:        idx = --pcg->pcg_avail;
                   1963:
1.87      thorpej  1964:        KASSERT(pcg->pcg_objects[idx].pcgo_va != NULL);
                   1965:        object = pcg->pcg_objects[idx].pcgo_va;
                   1966:        if (pap != NULL)
                   1967:                *pap = pcg->pcg_objects[idx].pcgo_pa;
                   1968:        pcg->pcg_objects[idx].pcgo_va = NULL;
1.43      thorpej  1969:
                   1970:        return (object);
                   1971: }
                   1972:
1.110     perry    1973: static inline void
1.87      thorpej  1974: pcg_put(struct pool_cache_group *pcg, void *object, paddr_t pa)
1.43      thorpej  1975: {
                   1976:        u_int idx;
                   1977:
                   1978:        KASSERT(pcg->pcg_avail < PCG_NOBJECTS);
                   1979:        idx = pcg->pcg_avail++;
                   1980:
1.87      thorpej  1981:        KASSERT(pcg->pcg_objects[idx].pcgo_va == NULL);
                   1982:        pcg->pcg_objects[idx].pcgo_va = object;
                   1983:        pcg->pcg_objects[idx].pcgo_pa = pa;
1.43      thorpej  1984: }
                   1985:
1.102     chs      1986: static void
                   1987: pcg_grouplist_free(struct pool_cache_grouplist *pcgl)
                   1988: {
                   1989:        struct pool_cache_group *pcg;
                   1990:        int s;
                   1991:
                   1992:        s = splvm();
                   1993:        while ((pcg = LIST_FIRST(pcgl)) != NULL) {
                   1994:                LIST_REMOVE(pcg, pcg_list);
                   1995:                pool_put(&pcgpool, pcg);
                   1996:        }
                   1997:        splx(s);
                   1998: }
                   1999:
1.43      thorpej  2000: /*
1.87      thorpej  2001:  * pool_cache_get{,_paddr}:
1.43      thorpej  2002:  *
1.87      thorpej  2003:  *     Get an object from a pool cache (optionally returning
                   2004:  *     the physical address of the object).
1.43      thorpej  2005:  */
                   2006: void *
1.87      thorpej  2007: pool_cache_get_paddr(struct pool_cache *pc, int flags, paddr_t *pap)
1.43      thorpej  2008: {
                   2009:        struct pool_cache_group *pcg;
                   2010:        void *object;
1.58      thorpej  2011:
                   2012: #ifdef LOCKDEBUG
                   2013:        if (flags & PR_WAITOK)
1.119     yamt     2014:                ASSERT_SLEEPABLE(NULL, "pool_cache_get(PR_WAITOK)");
1.58      thorpej  2015: #endif
1.43      thorpej  2016:
                   2017:        simple_lock(&pc->pc_slock);
                   2018:
1.102     chs      2019:        pcg = LIST_FIRST(&pc->pc_partgroups);
                   2020:        if (pcg == NULL) {
                   2021:                pcg = LIST_FIRST(&pc->pc_fullgroups);
                   2022:                if (pcg != NULL) {
                   2023:                        LIST_REMOVE(pcg, pcg_list);
                   2024:                        LIST_INSERT_HEAD(&pc->pc_partgroups, pcg, pcg_list);
1.43      thorpej  2025:                }
1.102     chs      2026:        }
                   2027:        if (pcg == NULL) {
1.43      thorpej  2028:
                   2029:                /*
                   2030:                 * No groups with any available objects.  Allocate
                   2031:                 * a new object, construct it, and return it to
                   2032:                 * the caller.  We will allocate a group, if necessary,
                   2033:                 * when the object is freed back to the cache.
                   2034:                 */
1.48      thorpej  2035:                pc->pc_misses++;
1.43      thorpej  2036:                simple_unlock(&pc->pc_slock);
                   2037:                object = pool_get(pc->pc_pool, flags);
                   2038:                if (object != NULL && pc->pc_ctor != NULL) {
                   2039:                        if ((*pc->pc_ctor)(pc->pc_arg, object, flags) != 0) {
                   2040:                                pool_put(pc->pc_pool, object);
                   2041:                                return (NULL);
                   2042:                        }
                   2043:                }
1.87      thorpej  2044:                if (object != NULL && pap != NULL) {
                   2045: #ifdef POOL_VTOPHYS
                   2046:                        *pap = POOL_VTOPHYS(object);
                   2047: #else
                   2048:                        *pap = POOL_PADDR_INVALID;
                   2049: #endif
                   2050:                }
1.43      thorpej  2051:                return (object);
                   2052:        }
                   2053:
1.48      thorpej  2054:        pc->pc_hits++;
                   2055:        pc->pc_nitems--;
1.87      thorpej  2056:        object = pcg_get(pcg, pap);
1.43      thorpej  2057:
1.102     chs      2058:        if (pcg->pcg_avail == 0) {
                   2059:                LIST_REMOVE(pcg, pcg_list);
                   2060:                LIST_INSERT_HEAD(&pc->pc_emptygroups, pcg, pcg_list);
                   2061:        }
1.43      thorpej  2062:        simple_unlock(&pc->pc_slock);
                   2063:
                   2064:        return (object);
                   2065: }
                   2066:
                   2067: /*
1.87      thorpej  2068:  * pool_cache_put{,_paddr}:
1.43      thorpej  2069:  *
1.87      thorpej  2070:  *     Put an object back to the pool cache (optionally caching the
                   2071:  *     physical address of the object).
1.43      thorpej  2072:  */
                   2073: void
1.87      thorpej  2074: pool_cache_put_paddr(struct pool_cache *pc, void *object, paddr_t pa)
1.43      thorpej  2075: {
                   2076:        struct pool_cache_group *pcg;
1.60      thorpej  2077:        int s;
1.43      thorpej  2078:
1.109     christos 2079:        if (__predict_false((pc->pc_pool->pr_flags & PR_WANTED) != 0)) {
                   2080:                goto destruct;
                   2081:        }
                   2082:
1.43      thorpej  2083:        simple_lock(&pc->pc_slock);
                   2084:
1.102     chs      2085:        pcg = LIST_FIRST(&pc->pc_partgroups);
                   2086:        if (pcg == NULL) {
                   2087:                pcg = LIST_FIRST(&pc->pc_emptygroups);
                   2088:                if (pcg != NULL) {
                   2089:                        LIST_REMOVE(pcg, pcg_list);
                   2090:                        LIST_INSERT_HEAD(&pc->pc_partgroups, pcg, pcg_list);
1.43      thorpej  2091:                }
1.102     chs      2092:        }
                   2093:        if (pcg == NULL) {
1.43      thorpej  2094:
                   2095:                /*
                   2096:                 * No empty groups to free the object to.  Attempt to
1.47      thorpej  2097:                 * allocate one.
1.43      thorpej  2098:                 */
1.47      thorpej  2099:                simple_unlock(&pc->pc_slock);
1.60      thorpej  2100:                s = splvm();
1.43      thorpej  2101:                pcg = pool_get(&pcgpool, PR_NOWAIT);
1.60      thorpej  2102:                splx(s);
1.102     chs      2103:                if (pcg == NULL) {
1.109     christos 2104: destruct:
1.102     chs      2105:
                   2106:                        /*
                   2107:                         * Unable to allocate a cache group; destruct the object
                   2108:                         * and free it back to the pool.
                   2109:                         */
                   2110:                        pool_cache_destruct_object(pc, object);
                   2111:                        return;
1.43      thorpej  2112:                }
1.102     chs      2113:                memset(pcg, 0, sizeof(*pcg));
                   2114:                simple_lock(&pc->pc_slock);
                   2115:                pc->pc_ngroups++;
                   2116:                LIST_INSERT_HEAD(&pc->pc_partgroups, pcg, pcg_list);
1.43      thorpej  2117:        }
                   2118:
1.48      thorpej  2119:        pc->pc_nitems++;
1.87      thorpej  2120:        pcg_put(pcg, object, pa);
1.43      thorpej  2121:
1.102     chs      2122:        if (pcg->pcg_avail == PCG_NOBJECTS) {
                   2123:                LIST_REMOVE(pcg, pcg_list);
                   2124:                LIST_INSERT_HEAD(&pc->pc_fullgroups, pcg, pcg_list);
                   2125:        }
1.43      thorpej  2126:        simple_unlock(&pc->pc_slock);
1.51      thorpej  2127: }
                   2128:
                   2129: /*
                   2130:  * pool_cache_destruct_object:
                   2131:  *
                   2132:  *     Force destruction of an object and its release back into
                   2133:  *     the pool.
                   2134:  */
                   2135: void
                   2136: pool_cache_destruct_object(struct pool_cache *pc, void *object)
                   2137: {
                   2138:
                   2139:        if (pc->pc_dtor != NULL)
                   2140:                (*pc->pc_dtor)(pc->pc_arg, object);
                   2141:        pool_put(pc->pc_pool, object);
1.43      thorpej  2142: }
                   2143:
1.102     chs      2144: static void
1.106     christos 2145: pool_do_cache_invalidate_grouplist(struct pool_cache_grouplist *pcgsl,
1.105     christos 2146:     struct pool_cache *pc, struct pool_pagelist *pq,
1.106     christos 2147:     struct pool_cache_grouplist *pcgdl)
1.102     chs      2148: {
1.106     christos 2149:        struct pool_cache_group *pcg, *npcg;
1.102     chs      2150:        void *object;
                   2151:
1.106     christos 2152:        for (pcg = LIST_FIRST(pcgsl); pcg != NULL; pcg = npcg) {
1.102     chs      2153:                npcg = LIST_NEXT(pcg, pcg_list);
                   2154:                while (pcg->pcg_avail != 0) {
                   2155:                        pc->pc_nitems--;
                   2156:                        object = pcg_get(pcg, NULL);
                   2157:                        if (pc->pc_dtor != NULL)
                   2158:                                (*pc->pc_dtor)(pc->pc_arg, object);
                   2159:                        pool_do_put(pc->pc_pool, object, pq);
                   2160:                }
1.103     chs      2161:                pc->pc_ngroups--;
1.102     chs      2162:                LIST_REMOVE(pcg, pcg_list);
1.106     christos 2163:                LIST_INSERT_HEAD(pcgdl, pcg, pcg_list);
1.102     chs      2164:        }
1.105     christos 2165: }
                   2166:
                   2167: static void
                   2168: pool_do_cache_invalidate(struct pool_cache *pc, struct pool_pagelist *pq,
                   2169:     struct pool_cache_grouplist *pcgl)
                   2170: {
                   2171:
                   2172:        LOCK_ASSERT(simple_lock_held(&pc->pc_slock));
                   2173:        LOCK_ASSERT(simple_lock_held(&pc->pc_pool->pr_slock));
                   2174:
1.106     christos 2175:        pool_do_cache_invalidate_grouplist(&pc->pc_fullgroups, pc, pq, pcgl);
                   2176:        pool_do_cache_invalidate_grouplist(&pc->pc_partgroups, pc, pq, pcgl);
1.103     chs      2177:
                   2178:        KASSERT(LIST_EMPTY(&pc->pc_partgroups));
                   2179:        KASSERT(LIST_EMPTY(&pc->pc_fullgroups));
                   2180:        KASSERT(pc->pc_nitems == 0);
1.102     chs      2181: }
                   2182:
1.43      thorpej  2183: /*
1.101     thorpej  2184:  * pool_cache_invalidate:
1.43      thorpej  2185:  *
1.101     thorpej  2186:  *     Invalidate a pool cache (destruct and release all of the
                   2187:  *     cached objects).
1.43      thorpej  2188:  */
1.101     thorpej  2189: void
                   2190: pool_cache_invalidate(struct pool_cache *pc)
1.43      thorpej  2191: {
1.101     thorpej  2192:        struct pool_pagelist pq;
1.102     chs      2193:        struct pool_cache_grouplist pcgl;
1.101     thorpej  2194:
                   2195:        LIST_INIT(&pq);
1.102     chs      2196:        LIST_INIT(&pcgl);
1.101     thorpej  2197:
                   2198:        simple_lock(&pc->pc_slock);
                   2199:        simple_lock(&pc->pc_pool->pr_slock);
1.43      thorpej  2200:
1.102     chs      2201:        pool_do_cache_invalidate(pc, &pq, &pcgl);
1.43      thorpej  2202:
1.101     thorpej  2203:        simple_unlock(&pc->pc_pool->pr_slock);
                   2204:        simple_unlock(&pc->pc_slock);
1.43      thorpej  2205:
1.102     chs      2206:        pr_pagelist_free(pc->pc_pool, &pq);
                   2207:        pcg_grouplist_free(&pcgl);
1.43      thorpej  2208: }
                   2209:
                   2210: /*
                   2211:  * pool_cache_reclaim:
                   2212:  *
                   2213:  *     Reclaim a pool cache for pool_reclaim().
                   2214:  */
                   2215: static void
1.102     chs      2216: pool_cache_reclaim(struct pool_cache *pc, struct pool_pagelist *pq,
                   2217:     struct pool_cache_grouplist *pcgl)
1.43      thorpej  2218: {
1.101     thorpej  2219:
                   2220:        /*
                   2221:         * We're locking in the wrong order (normally pool_cache -> pool,
                   2222:         * but the pool is already locked when we get here), so we have
                   2223:         * to use trylock.  If we can't lock the pool_cache, it's not really
                   2224:         * a big deal here.
                   2225:         */
                   2226:        if (simple_lock_try(&pc->pc_slock) == 0)
                   2227:                return;
                   2228:
1.102     chs      2229:        pool_do_cache_invalidate(pc, pq, pcgl);
1.43      thorpej  2230:
                   2231:        simple_unlock(&pc->pc_slock);
1.3       pk       2232: }
1.66      thorpej  2233:
                   2234: /*
                   2235:  * Pool backend allocators.
                   2236:  *
                   2237:  * Each pool has a backend allocator that handles allocation, deallocation,
                   2238:  * and any additional draining that might be needed.
                   2239:  *
                   2240:  * We provide two standard allocators:
                   2241:  *
                   2242:  *     pool_allocator_kmem - the default when no allocator is specified
                   2243:  *
                   2244:  *     pool_allocator_nointr - used for pools that will not be accessed
                   2245:  *     in interrupt context.
                   2246:  */
                   2247: void   *pool_page_alloc(struct pool *, int);
                   2248: void   pool_page_free(struct pool *, void *);
                   2249:
1.112     bjh21    2250: #ifdef POOL_SUBPAGE
                   2251: struct pool_allocator pool_allocator_kmem_fullpage = {
                   2252:        pool_page_alloc, pool_page_free, 0,
1.117     yamt     2253:        .pa_backingmapptr = &kmem_map,
1.112     bjh21    2254: };
                   2255: #else
1.66      thorpej  2256: struct pool_allocator pool_allocator_kmem = {
                   2257:        pool_page_alloc, pool_page_free, 0,
1.117     yamt     2258:        .pa_backingmapptr = &kmem_map,
1.66      thorpej  2259: };
1.112     bjh21    2260: #endif
1.66      thorpej  2261:
                   2262: void   *pool_page_alloc_nointr(struct pool *, int);
                   2263: void   pool_page_free_nointr(struct pool *, void *);
                   2264:
1.112     bjh21    2265: #ifdef POOL_SUBPAGE
                   2266: struct pool_allocator pool_allocator_nointr_fullpage = {
                   2267:        pool_page_alloc_nointr, pool_page_free_nointr, 0,
1.117     yamt     2268:        .pa_backingmapptr = &kernel_map,
1.112     bjh21    2269: };
                   2270: #else
1.66      thorpej  2271: struct pool_allocator pool_allocator_nointr = {
                   2272:        pool_page_alloc_nointr, pool_page_free_nointr, 0,
1.117     yamt     2273:        .pa_backingmapptr = &kernel_map,
1.66      thorpej  2274: };
1.112     bjh21    2275: #endif
1.66      thorpej  2276:
                   2277: #ifdef POOL_SUBPAGE
                   2278: void   *pool_subpage_alloc(struct pool *, int);
                   2279: void   pool_subpage_free(struct pool *, void *);
                   2280:
1.112     bjh21    2281: struct pool_allocator pool_allocator_kmem = {
                   2282:        pool_subpage_alloc, pool_subpage_free, POOL_SUBPAGE,
1.117     yamt     2283:        .pa_backingmapptr = &kmem_map,
1.112     bjh21    2284: };
                   2285:
                   2286: void   *pool_subpage_alloc_nointr(struct pool *, int);
                   2287: void   pool_subpage_free_nointr(struct pool *, void *);
                   2288:
                   2289: struct pool_allocator pool_allocator_nointr = {
                   2290:        pool_subpage_alloc, pool_subpage_free, POOL_SUBPAGE,
1.117     yamt     2291:        .pa_backingmapptr = &kmem_map,
1.66      thorpej  2292: };
                   2293: #endif /* POOL_SUBPAGE */
                   2294:
1.117     yamt     2295: static void *
                   2296: pool_allocator_alloc(struct pool *pp, int flags)
1.66      thorpej  2297: {
1.117     yamt     2298:        struct pool_allocator *pa = pp->pr_alloc;
1.66      thorpej  2299:        void *res;
                   2300:
1.117     yamt     2301:        LOCK_ASSERT(!simple_lock_held(&pp->pr_slock));
1.66      thorpej  2302:
1.117     yamt     2303:        res = (*pa->pa_alloc)(pp, flags);
                   2304:        if (res == NULL && (flags & PR_WAITOK) == 0) {
1.66      thorpej  2305:                /*
1.117     yamt     2306:                 * We only run the drain hook here if PR_NOWAIT.
                   2307:                 * In other cases, the hook will be run in
                   2308:                 * pool_reclaim().
1.66      thorpej  2309:                 */
1.117     yamt     2310:                if (pp->pr_drain_hook != NULL) {
                   2311:                        (*pp->pr_drain_hook)(pp->pr_drain_hook_arg, flags);
                   2312:                        res = (*pa->pa_alloc)(pp, flags);
1.66      thorpej  2313:                }
1.117     yamt     2314:        }
                   2315:        return res;
1.66      thorpej  2316: }
                   2317:
1.117     yamt     2318: static void
1.66      thorpej  2319: pool_allocator_free(struct pool *pp, void *v)
                   2320: {
                   2321:        struct pool_allocator *pa = pp->pr_alloc;
                   2322:
1.91      yamt     2323:        LOCK_ASSERT(!simple_lock_held(&pp->pr_slock));
                   2324:
1.66      thorpej  2325:        (*pa->pa_free)(pp, v);
                   2326: }
                   2327:
                   2328: void *
                   2329: pool_page_alloc(struct pool *pp, int flags)
                   2330: {
                   2331:        boolean_t waitok = (flags & PR_WAITOK) ? TRUE : FALSE;
                   2332:
1.100     yamt     2333:        return ((void *) uvm_km_alloc_poolpage_cache(kmem_map, waitok));
1.66      thorpej  2334: }
                   2335:
                   2336: void
                   2337: pool_page_free(struct pool *pp, void *v)
                   2338: {
                   2339:
1.98      yamt     2340:        uvm_km_free_poolpage_cache(kmem_map, (vaddr_t) v);
                   2341: }
                   2342:
                   2343: static void *
                   2344: pool_page_alloc_meta(struct pool *pp, int flags)
                   2345: {
                   2346:        boolean_t waitok = (flags & PR_WAITOK) ? TRUE : FALSE;
                   2347:
1.100     yamt     2348:        return ((void *) uvm_km_alloc_poolpage(kmem_map, waitok));
1.98      yamt     2349: }
                   2350:
                   2351: static void
                   2352: pool_page_free_meta(struct pool *pp, void *v)
                   2353: {
                   2354:
1.100     yamt     2355:        uvm_km_free_poolpage(kmem_map, (vaddr_t) v);
1.66      thorpej  2356: }
                   2357:
                   2358: #ifdef POOL_SUBPAGE
                   2359: /* Sub-page allocator, for machines with large hardware pages. */
                   2360: void *
                   2361: pool_subpage_alloc(struct pool *pp, int flags)
                   2362: {
1.93      dbj      2363:        void *v;
                   2364:        int s;
                   2365:        s = splvm();
                   2366:        v = pool_get(&psppool, flags);
                   2367:        splx(s);
                   2368:        return v;
1.66      thorpej  2369: }
                   2370:
                   2371: void
                   2372: pool_subpage_free(struct pool *pp, void *v)
                   2373: {
1.93      dbj      2374:        int s;
                   2375:        s = splvm();
1.66      thorpej  2376:        pool_put(&psppool, v);
1.93      dbj      2377:        splx(s);
1.66      thorpej  2378: }
                   2379:
                   2380: /* We don't provide a real nointr allocator.  Maybe later. */
                   2381: void *
1.112     bjh21    2382: pool_subpage_alloc_nointr(struct pool *pp, int flags)
1.66      thorpej  2383: {
                   2384:
                   2385:        return (pool_subpage_alloc(pp, flags));
                   2386: }
                   2387:
                   2388: void
1.112     bjh21    2389: pool_subpage_free_nointr(struct pool *pp, void *v)
1.66      thorpej  2390: {
                   2391:
                   2392:        pool_subpage_free(pp, v);
                   2393: }
1.112     bjh21    2394: #endif /* POOL_SUBPAGE */
1.66      thorpej  2395: void *
                   2396: pool_page_alloc_nointr(struct pool *pp, int flags)
                   2397: {
                   2398:        boolean_t waitok = (flags & PR_WAITOK) ? TRUE : FALSE;
                   2399:
1.100     yamt     2400:        return ((void *) uvm_km_alloc_poolpage_cache(kernel_map, waitok));
1.66      thorpej  2401: }
                   2402:
                   2403: void
                   2404: pool_page_free_nointr(struct pool *pp, void *v)
                   2405: {
                   2406:
1.98      yamt     2407:        uvm_km_free_poolpage_cache(kernel_map, (vaddr_t) v);
1.66      thorpej  2408: }

CVSweb <webmaster@jp.NetBSD.org>