[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 1.112

1.112   ! bjh21       1: /*     $NetBSD: subr_pool.c,v 1.111 2006/01/26 15:07:25 christos Exp $ */
1.1       pk          2:
                      3: /*-
1.43      thorpej     4:  * Copyright (c) 1997, 1999, 2000 The NetBSD Foundation, Inc.
1.1       pk          5:  * All rights reserved.
                      6:  *
                      7:  * This code is derived from software contributed to The NetBSD Foundation
1.20      thorpej     8:  * by Paul Kranenburg; by Jason R. Thorpe of the Numerical Aerospace
                      9:  * Simulation Facility, NASA Ames Research Center.
1.1       pk         10:  *
                     11:  * Redistribution and use in source and binary forms, with or without
                     12:  * modification, are permitted provided that the following conditions
                     13:  * are met:
                     14:  * 1. Redistributions of source code must retain the above copyright
                     15:  *    notice, this list of conditions and the following disclaimer.
                     16:  * 2. Redistributions in binary form must reproduce the above copyright
                     17:  *    notice, this list of conditions and the following disclaimer in the
                     18:  *    documentation and/or other materials provided with the distribution.
                     19:  * 3. All advertising materials mentioning features or use of this software
                     20:  *    must display the following acknowledgement:
1.13      christos   21:  *     This product includes software developed by the NetBSD
                     22:  *     Foundation, Inc. and its contributors.
1.1       pk         23:  * 4. Neither the name of The NetBSD Foundation nor the names of its
                     24:  *    contributors may be used to endorse or promote products derived
                     25:  *    from this software without specific prior written permission.
                     26:  *
                     27:  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
                     28:  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
                     29:  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
                     30:  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
                     31:  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
                     32:  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
                     33:  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
                     34:  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
                     35:  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
                     36:  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
                     37:  * POSSIBILITY OF SUCH DAMAGE.
                     38:  */
1.64      lukem      39:
                     40: #include <sys/cdefs.h>
1.112   ! bjh21      41: __KERNEL_RCSID(0, "$NetBSD: subr_pool.c,v 1.111 2006/01/26 15:07:25 christos Exp $");
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:
1.98      yamt       85: static void *pool_page_alloc_meta(struct pool *, int);
                     86: static void pool_page_free_meta(struct pool *, void *);
                     87:
                     88: /* allocator for pool metadata */
                     89: static struct pool_allocator pool_allocator_meta = {
                     90:        pool_page_alloc_meta, pool_page_free_meta
                     91: };
                     92:
1.3       pk         93: /* # of seconds to retain page after last use */
                     94: int pool_inactive_time = 10;
                     95:
                     96: /* Next candidate for drainage (see pool_drain()) */
1.23      thorpej    97: static struct pool     *drainpp;
                     98:
                     99: /* This spin lock protects both pool_head and drainpp. */
                    100: struct simplelock pool_head_slock = SIMPLELOCK_INITIALIZER;
1.3       pk        101:
1.99      yamt      102: typedef uint8_t pool_item_freelist_t;
                    103:
1.3       pk        104: struct pool_item_header {
                    105:        /* Page headers */
1.88      chs       106:        LIST_ENTRY(pool_item_header)
1.3       pk        107:                                ph_pagelist;    /* pool page list */
1.88      chs       108:        SPLAY_ENTRY(pool_item_header)
                    109:                                ph_node;        /* Off-page page headers */
1.3       pk        110:        caddr_t                 ph_page;        /* this page's address */
                    111:        struct timeval          ph_time;        /* last referenced */
1.97      yamt      112:        union {
                    113:                /* !PR_NOTOUCH */
                    114:                struct {
1.102     chs       115:                        LIST_HEAD(, pool_item)
1.97      yamt      116:                                phu_itemlist;   /* chunk list for this page */
                    117:                } phu_normal;
                    118:                /* PR_NOTOUCH */
                    119:                struct {
                    120:                        uint16_t
                    121:                                phu_off;        /* start offset in page */
1.99      yamt      122:                        pool_item_freelist_t
1.97      yamt      123:                                phu_firstfree;  /* first free item */
1.99      yamt      124:                        /*
                    125:                         * XXX it might be better to use
                    126:                         * a simple bitmap and ffs(3)
                    127:                         */
1.97      yamt      128:                } phu_notouch;
                    129:        } ph_u;
                    130:        uint16_t                ph_nmissing;    /* # of chunks in use */
1.3       pk        131: };
1.97      yamt      132: #define        ph_itemlist     ph_u.phu_normal.phu_itemlist
                    133: #define        ph_off          ph_u.phu_notouch.phu_off
                    134: #define        ph_firstfree    ph_u.phu_notouch.phu_firstfree
1.3       pk        135:
1.1       pk        136: struct pool_item {
1.3       pk        137: #ifdef DIAGNOSTIC
1.82      thorpej   138:        u_int pi_magic;
1.33      chs       139: #endif
1.82      thorpej   140: #define        PI_MAGIC 0xdeadbeefU
1.3       pk        141:        /* Other entries use only this list entry */
1.102     chs       142:        LIST_ENTRY(pool_item)   pi_list;
1.3       pk        143: };
                    144:
1.53      thorpej   145: #define        POOL_NEEDS_CATCHUP(pp)                                          \
                    146:        ((pp)->pr_nitems < (pp)->pr_minitems)
                    147:
1.43      thorpej   148: /*
                    149:  * Pool cache management.
                    150:  *
                    151:  * Pool caches provide a way for constructed objects to be cached by the
                    152:  * pool subsystem.  This can lead to performance improvements by avoiding
                    153:  * needless object construction/destruction; it is deferred until absolutely
                    154:  * necessary.
                    155:  *
                    156:  * Caches are grouped into cache groups.  Each cache group references
                    157:  * up to 16 constructed objects.  When a cache allocates an object
                    158:  * from the pool, it calls the object's constructor and places it into
                    159:  * a cache group.  When a cache group frees an object back to the pool,
                    160:  * it first calls the object's destructor.  This allows the object to
                    161:  * persist in constructed form while freed to the cache.
                    162:  *
                    163:  * Multiple caches may exist for each pool.  This allows a single
                    164:  * object type to have multiple constructed forms.  The pool references
                    165:  * each cache, so that when a pool is drained by the pagedaemon, it can
                    166:  * drain each individual cache as well.  Each time a cache is drained,
                    167:  * the most idle cache group is freed to the pool in its entirety.
                    168:  *
                    169:  * Pool caches are layed on top of pools.  By layering them, we can avoid
                    170:  * the complexity of cache management for pools which would not benefit
                    171:  * from it.
                    172:  */
                    173:
                    174: /* The cache group pool. */
                    175: static struct pool pcgpool;
1.3       pk        176:
1.102     chs       177: static void    pool_cache_reclaim(struct pool_cache *, struct pool_pagelist *,
                    178:                                   struct pool_cache_grouplist *);
                    179: static void    pcg_grouplist_free(struct pool_cache_grouplist *);
1.3       pk        180:
1.42      thorpej   181: static int     pool_catchup(struct pool *);
1.55      thorpej   182: static void    pool_prime_page(struct pool *, caddr_t,
                    183:                    struct pool_item_header *);
1.88      chs       184: static void    pool_update_curpage(struct pool *);
1.66      thorpej   185:
                    186: void           *pool_allocator_alloc(struct pool *, int);
                    187: void           pool_allocator_free(struct pool *, void *);
1.3       pk        188:
1.97      yamt      189: static void pool_print_pagelist(struct pool *, struct pool_pagelist *,
1.88      chs       190:        void (*)(const char *, ...));
1.42      thorpej   191: static void pool_print1(struct pool *, const char *,
                    192:        void (*)(const char *, ...));
1.3       pk        193:
1.88      chs       194: static int pool_chk_page(struct pool *, const char *,
                    195:                         struct pool_item_header *);
                    196:
1.3       pk        197: /*
1.52      thorpej   198:  * Pool log entry. An array of these is allocated in pool_init().
1.3       pk        199:  */
                    200: struct pool_log {
                    201:        const char      *pl_file;
                    202:        long            pl_line;
                    203:        int             pl_action;
1.25      thorpej   204: #define        PRLOG_GET       1
                    205: #define        PRLOG_PUT       2
1.3       pk        206:        void            *pl_addr;
1.1       pk        207: };
                    208:
1.86      matt      209: #ifdef POOL_DIAGNOSTIC
1.3       pk        210: /* Number of entries in pool log buffers */
1.17      thorpej   211: #ifndef POOL_LOGSIZE
                    212: #define        POOL_LOGSIZE    10
                    213: #endif
                    214:
                    215: int pool_logsize = POOL_LOGSIZE;
1.1       pk        216:
1.110     perry     217: static inline void
1.42      thorpej   218: pr_log(struct pool *pp, void *v, int action, const char *file, long line)
1.3       pk        219: {
                    220:        int n = pp->pr_curlogentry;
                    221:        struct pool_log *pl;
                    222:
1.20      thorpej   223:        if ((pp->pr_roflags & PR_LOGGING) == 0)
1.3       pk        224:                return;
                    225:
                    226:        /*
                    227:         * Fill in the current entry. Wrap around and overwrite
                    228:         * the oldest entry if necessary.
                    229:         */
                    230:        pl = &pp->pr_log[n];
                    231:        pl->pl_file = file;
                    232:        pl->pl_line = line;
                    233:        pl->pl_action = action;
                    234:        pl->pl_addr = v;
                    235:        if (++n >= pp->pr_logsize)
                    236:                n = 0;
                    237:        pp->pr_curlogentry = n;
                    238: }
                    239:
                    240: static void
1.42      thorpej   241: pr_printlog(struct pool *pp, struct pool_item *pi,
                    242:     void (*pr)(const char *, ...))
1.3       pk        243: {
                    244:        int i = pp->pr_logsize;
                    245:        int n = pp->pr_curlogentry;
                    246:
1.20      thorpej   247:        if ((pp->pr_roflags & PR_LOGGING) == 0)
1.3       pk        248:                return;
                    249:
                    250:        /*
                    251:         * Print all entries in this pool's log.
                    252:         */
                    253:        while (i-- > 0) {
                    254:                struct pool_log *pl = &pp->pr_log[n];
                    255:                if (pl->pl_action != 0) {
1.25      thorpej   256:                        if (pi == NULL || pi == pl->pl_addr) {
                    257:                                (*pr)("\tlog entry %d:\n", i);
                    258:                                (*pr)("\t\taction = %s, addr = %p\n",
                    259:                                    pl->pl_action == PRLOG_GET ? "get" : "put",
                    260:                                    pl->pl_addr);
                    261:                                (*pr)("\t\tfile: %s at line %lu\n",
                    262:                                    pl->pl_file, pl->pl_line);
                    263:                        }
1.3       pk        264:                }
                    265:                if (++n >= pp->pr_logsize)
                    266:                        n = 0;
                    267:        }
                    268: }
1.25      thorpej   269:
1.110     perry     270: static inline void
1.42      thorpej   271: pr_enter(struct pool *pp, const char *file, long line)
1.25      thorpej   272: {
                    273:
1.34      thorpej   274:        if (__predict_false(pp->pr_entered_file != NULL)) {
1.25      thorpej   275:                printf("pool %s: reentrancy at file %s line %ld\n",
                    276:                    pp->pr_wchan, file, line);
                    277:                printf("         previous entry at file %s line %ld\n",
                    278:                    pp->pr_entered_file, pp->pr_entered_line);
                    279:                panic("pr_enter");
                    280:        }
                    281:
                    282:        pp->pr_entered_file = file;
                    283:        pp->pr_entered_line = line;
                    284: }
                    285:
1.110     perry     286: static inline void
1.42      thorpej   287: pr_leave(struct pool *pp)
1.25      thorpej   288: {
                    289:
1.34      thorpej   290:        if (__predict_false(pp->pr_entered_file == NULL)) {
1.25      thorpej   291:                printf("pool %s not entered?\n", pp->pr_wchan);
                    292:                panic("pr_leave");
                    293:        }
                    294:
                    295:        pp->pr_entered_file = NULL;
                    296:        pp->pr_entered_line = 0;
                    297: }
                    298:
1.110     perry     299: static inline void
1.42      thorpej   300: pr_enter_check(struct pool *pp, void (*pr)(const char *, ...))
1.25      thorpej   301: {
                    302:
                    303:        if (pp->pr_entered_file != NULL)
                    304:                (*pr)("\n\tcurrently entered from file %s line %ld\n",
                    305:                    pp->pr_entered_file, pp->pr_entered_line);
                    306: }
1.3       pk        307: #else
1.25      thorpej   308: #define        pr_log(pp, v, action, file, line)
                    309: #define        pr_printlog(pp, pi, pr)
                    310: #define        pr_enter(pp, file, line)
                    311: #define        pr_leave(pp)
                    312: #define        pr_enter_check(pp, pr)
1.59      thorpej   313: #endif /* POOL_DIAGNOSTIC */
1.3       pk        314:
1.110     perry     315: static inline int
1.97      yamt      316: pr_item_notouch_index(const struct pool *pp, const struct pool_item_header *ph,
                    317:     const void *v)
                    318: {
                    319:        const char *cp = v;
                    320:        int idx;
                    321:
                    322:        KASSERT(pp->pr_roflags & PR_NOTOUCH);
                    323:        idx = (cp - ph->ph_page - ph->ph_off) / pp->pr_size;
                    324:        KASSERT(idx < pp->pr_itemsperpage);
                    325:        return idx;
                    326: }
                    327:
1.99      yamt      328: #define        PR_FREELIST_ALIGN(p) \
                    329:        roundup((uintptr_t)(p), sizeof(pool_item_freelist_t))
                    330: #define        PR_FREELIST(ph) ((pool_item_freelist_t *)PR_FREELIST_ALIGN((ph) + 1))
                    331: #define        PR_INDEX_USED   ((pool_item_freelist_t)-1)
                    332: #define        PR_INDEX_EOL    ((pool_item_freelist_t)-2)
1.97      yamt      333:
1.110     perry     334: static inline void
1.97      yamt      335: pr_item_notouch_put(const struct pool *pp, struct pool_item_header *ph,
                    336:     void *obj)
                    337: {
                    338:        int idx = pr_item_notouch_index(pp, ph, obj);
1.99      yamt      339:        pool_item_freelist_t *freelist = PR_FREELIST(ph);
1.97      yamt      340:
                    341:        KASSERT(freelist[idx] == PR_INDEX_USED);
                    342:        freelist[idx] = ph->ph_firstfree;
                    343:        ph->ph_firstfree = idx;
                    344: }
                    345:
1.110     perry     346: static inline void *
1.97      yamt      347: pr_item_notouch_get(const struct pool *pp, struct pool_item_header *ph)
                    348: {
                    349:        int idx = ph->ph_firstfree;
1.99      yamt      350:        pool_item_freelist_t *freelist = PR_FREELIST(ph);
1.97      yamt      351:
                    352:        KASSERT(freelist[idx] != PR_INDEX_USED);
                    353:        ph->ph_firstfree = freelist[idx];
                    354:        freelist[idx] = PR_INDEX_USED;
                    355:
                    356:        return ph->ph_page + ph->ph_off + idx * pp->pr_size;
                    357: }
                    358:
1.110     perry     359: static inline int
1.88      chs       360: phtree_compare(struct pool_item_header *a, struct pool_item_header *b)
                    361: {
                    362:        if (a->ph_page < b->ph_page)
                    363:                return (-1);
                    364:        else if (a->ph_page > b->ph_page)
                    365:                return (1);
                    366:        else
                    367:                return (0);
                    368: }
                    369:
                    370: SPLAY_PROTOTYPE(phtree, pool_item_header, ph_node, phtree_compare);
                    371: SPLAY_GENERATE(phtree, pool_item_header, ph_node, phtree_compare);
                    372:
1.3       pk        373: /*
                    374:  * Return the pool page header based on page address.
                    375:  */
1.110     perry     376: static inline struct pool_item_header *
1.42      thorpej   377: pr_find_pagehead(struct pool *pp, caddr_t page)
1.3       pk        378: {
1.88      chs       379:        struct pool_item_header *ph, tmp;
1.3       pk        380:
1.20      thorpej   381:        if ((pp->pr_roflags & PR_PHINPAGE) != 0)
1.3       pk        382:                return ((struct pool_item_header *)(page + pp->pr_phoffset));
                    383:
1.88      chs       384:        tmp.ph_page = page;
                    385:        ph = SPLAY_FIND(phtree, &pp->pr_phtree, &tmp);
                    386:        return ph;
1.3       pk        387: }
                    388:
1.101     thorpej   389: static void
                    390: pr_pagelist_free(struct pool *pp, struct pool_pagelist *pq)
                    391: {
                    392:        struct pool_item_header *ph;
                    393:        int s;
                    394:
                    395:        while ((ph = LIST_FIRST(pq)) != NULL) {
                    396:                LIST_REMOVE(ph, ph_pagelist);
                    397:                pool_allocator_free(pp, ph->ph_page);
                    398:                if ((pp->pr_roflags & PR_PHINPAGE) == 0) {
                    399:                        s = splvm();
                    400:                        pool_put(pp->pr_phpool, ph);
                    401:                        splx(s);
                    402:                }
                    403:        }
                    404: }
                    405:
1.3       pk        406: /*
                    407:  * Remove a page from the pool.
                    408:  */
1.110     perry     409: static inline void
1.61      chs       410: pr_rmpage(struct pool *pp, struct pool_item_header *ph,
                    411:      struct pool_pagelist *pq)
1.3       pk        412: {
                    413:
1.101     thorpej   414:        LOCK_ASSERT(simple_lock_held(&pp->pr_slock));
1.91      yamt      415:
1.3       pk        416:        /*
1.7       thorpej   417:         * If the page was idle, decrement the idle page count.
1.3       pk        418:         */
1.6       thorpej   419:        if (ph->ph_nmissing == 0) {
                    420: #ifdef DIAGNOSTIC
                    421:                if (pp->pr_nidle == 0)
                    422:                        panic("pr_rmpage: nidle inconsistent");
1.20      thorpej   423:                if (pp->pr_nitems < pp->pr_itemsperpage)
                    424:                        panic("pr_rmpage: nitems inconsistent");
1.6       thorpej   425: #endif
                    426:                pp->pr_nidle--;
                    427:        }
1.7       thorpej   428:
1.20      thorpej   429:        pp->pr_nitems -= pp->pr_itemsperpage;
                    430:
1.7       thorpej   431:        /*
1.101     thorpej   432:         * Unlink the page from the pool and queue it for release.
1.7       thorpej   433:         */
1.88      chs       434:        LIST_REMOVE(ph, ph_pagelist);
1.91      yamt      435:        if ((pp->pr_roflags & PR_PHINPAGE) == 0)
                    436:                SPLAY_REMOVE(phtree, &pp->pr_phtree, ph);
1.101     thorpej   437:        LIST_INSERT_HEAD(pq, ph, ph_pagelist);
                    438:
1.7       thorpej   439:        pp->pr_npages--;
                    440:        pp->pr_npagefree++;
1.6       thorpej   441:
1.88      chs       442:        pool_update_curpage(pp);
1.3       pk        443: }
                    444:
                    445: /*
1.94      simonb    446:  * Initialize all the pools listed in the "pools" link set.
                    447:  */
                    448: void
                    449: link_pool_init(void)
                    450: {
                    451:        __link_set_decl(pools, struct link_pool_init);
                    452:        struct link_pool_init * const *pi;
                    453:
                    454:        __link_set_foreach(pi, pools)
                    455:                pool_init((*pi)->pp, (*pi)->size, (*pi)->align,
                    456:                    (*pi)->align_offset, (*pi)->flags, (*pi)->wchan,
                    457:                    (*pi)->palloc);
                    458: }
                    459:
                    460: /*
1.3       pk        461:  * Initialize the given pool resource structure.
                    462:  *
                    463:  * We export this routine to allow other kernel parts to declare
                    464:  * static pools that must be initialized before malloc() is available.
                    465:  */
                    466: void
1.42      thorpej   467: pool_init(struct pool *pp, size_t size, u_int align, u_int ioff, int flags,
1.66      thorpej   468:     const char *wchan, struct pool_allocator *palloc)
1.3       pk        469: {
1.88      chs       470:        int off, slack;
1.92      enami     471:        size_t trysize, phsize;
1.93      dbj       472:        int s;
1.3       pk        473:
1.99      yamt      474:        KASSERT((1UL << (CHAR_BIT * sizeof(pool_item_freelist_t))) - 2 >=
                    475:            PHPOOL_FREELIST_NELEM(PHPOOL_MAX - 1));
                    476:
1.25      thorpej   477: #ifdef POOL_DIAGNOSTIC
                    478:        /*
                    479:         * Always log if POOL_DIAGNOSTIC is defined.
                    480:         */
                    481:        if (pool_logsize != 0)
                    482:                flags |= PR_LOGGING;
                    483: #endif
                    484:
1.66      thorpej   485:        if (palloc == NULL)
                    486:                palloc = &pool_allocator_kmem;
1.112   ! bjh21     487: #ifdef POOL_SUBPAGE
        !           488:        if (size > palloc->pa_pagesz) {
        !           489:                if (palloc == &pool_allocator_kmem)
        !           490:                        palloc = &pool_allocator_kmem_fullpage;
        !           491:                else if (palloc == &pool_allocator_nointr)
        !           492:                        palloc = &pool_allocator_nointr_fullpage;
        !           493:        }
1.66      thorpej   494: #endif /* POOL_SUBPAGE */
                    495:        if ((palloc->pa_flags & PA_INITIALIZED) == 0) {
1.112   ! bjh21     496:                if (palloc->pa_pagesz == 0)
1.66      thorpej   497:                        palloc->pa_pagesz = PAGE_SIZE;
                    498:
                    499:                TAILQ_INIT(&palloc->pa_list);
                    500:
                    501:                simple_lock_init(&palloc->pa_slock);
                    502:                palloc->pa_pagemask = ~(palloc->pa_pagesz - 1);
                    503:                palloc->pa_pageshift = ffs(palloc->pa_pagesz) - 1;
                    504:                palloc->pa_flags |= PA_INITIALIZED;
1.4       thorpej   505:        }
1.3       pk        506:
                    507:        if (align == 0)
                    508:                align = ALIGN(1);
1.14      thorpej   509:
                    510:        if (size < sizeof(struct pool_item))
                    511:                size = sizeof(struct pool_item);
1.3       pk        512:
1.78      thorpej   513:        size = roundup(size, align);
1.66      thorpej   514: #ifdef DIAGNOSTIC
                    515:        if (size > palloc->pa_pagesz)
1.35      pk        516:                panic("pool_init: pool item size (%lu) too large",
                    517:                      (u_long)size);
1.66      thorpej   518: #endif
1.35      pk        519:
1.3       pk        520:        /*
                    521:         * Initialize the pool structure.
                    522:         */
1.88      chs       523:        LIST_INIT(&pp->pr_emptypages);
                    524:        LIST_INIT(&pp->pr_fullpages);
                    525:        LIST_INIT(&pp->pr_partpages);
1.102     chs       526:        LIST_INIT(&pp->pr_cachelist);
1.3       pk        527:        pp->pr_curpage = NULL;
                    528:        pp->pr_npages = 0;
                    529:        pp->pr_minitems = 0;
                    530:        pp->pr_minpages = 0;
                    531:        pp->pr_maxpages = UINT_MAX;
1.20      thorpej   532:        pp->pr_roflags = flags;
                    533:        pp->pr_flags = 0;
1.35      pk        534:        pp->pr_size = size;
1.3       pk        535:        pp->pr_align = align;
                    536:        pp->pr_wchan = wchan;
1.66      thorpej   537:        pp->pr_alloc = palloc;
1.20      thorpej   538:        pp->pr_nitems = 0;
                    539:        pp->pr_nout = 0;
                    540:        pp->pr_hardlimit = UINT_MAX;
                    541:        pp->pr_hardlimit_warning = NULL;
1.31      thorpej   542:        pp->pr_hardlimit_ratecap.tv_sec = 0;
                    543:        pp->pr_hardlimit_ratecap.tv_usec = 0;
                    544:        pp->pr_hardlimit_warning_last.tv_sec = 0;
                    545:        pp->pr_hardlimit_warning_last.tv_usec = 0;
1.68      thorpej   546:        pp->pr_drain_hook = NULL;
                    547:        pp->pr_drain_hook_arg = NULL;
1.3       pk        548:
                    549:        /*
                    550:         * Decide whether to put the page header off page to avoid
1.92      enami     551:         * wasting too large a part of the page or too big item.
                    552:         * Off-page page headers go on a hash table, so we can match
                    553:         * a returned item with its header based on the page address.
                    554:         * We use 1/16 of the page size and about 8 times of the item
                    555:         * size as the threshold (XXX: tune)
                    556:         *
                    557:         * However, we'll put the header into the page if we can put
                    558:         * it without wasting any items.
                    559:         *
                    560:         * Silently enforce `0 <= ioff < align'.
1.3       pk        561:         */
1.92      enami     562:        pp->pr_itemoffset = ioff %= align;
                    563:        /* See the comment below about reserved bytes. */
                    564:        trysize = palloc->pa_pagesz - ((align - ioff) % align);
                    565:        phsize = ALIGN(sizeof(struct pool_item_header));
1.97      yamt      566:        if ((pp->pr_roflags & PR_NOTOUCH) == 0 &&
                    567:            (pp->pr_size < MIN(palloc->pa_pagesz / 16, phsize << 3) ||
                    568:            trysize / pp->pr_size == (trysize - phsize) / pp->pr_size)) {
1.3       pk        569:                /* Use the end of the page for the page header */
1.20      thorpej   570:                pp->pr_roflags |= PR_PHINPAGE;
1.92      enami     571:                pp->pr_phoffset = off = palloc->pa_pagesz - phsize;
1.2       pk        572:        } else {
1.3       pk        573:                /* The page header will be taken from our page header pool */
                    574:                pp->pr_phoffset = 0;
1.66      thorpej   575:                off = palloc->pa_pagesz;
1.88      chs       576:                SPLAY_INIT(&pp->pr_phtree);
1.2       pk        577:        }
1.1       pk        578:
1.3       pk        579:        /*
                    580:         * Alignment is to take place at `ioff' within the item. This means
                    581:         * we must reserve up to `align - 1' bytes on the page to allow
                    582:         * appropriate positioning of each item.
                    583:         */
                    584:        pp->pr_itemsperpage = (off - ((align - ioff) % align)) / pp->pr_size;
1.43      thorpej   585:        KASSERT(pp->pr_itemsperpage != 0);
1.97      yamt      586:        if ((pp->pr_roflags & PR_NOTOUCH)) {
                    587:                int idx;
                    588:
                    589:                for (idx = 0; pp->pr_itemsperpage > PHPOOL_FREELIST_NELEM(idx);
                    590:                    idx++) {
                    591:                        /* nothing */
                    592:                }
                    593:                if (idx >= PHPOOL_MAX) {
                    594:                        /*
                    595:                         * if you see this panic, consider to tweak
                    596:                         * PHPOOL_MAX and PHPOOL_FREELIST_NELEM.
                    597:                         */
                    598:                        panic("%s: too large itemsperpage(%d) for PR_NOTOUCH",
                    599:                            pp->pr_wchan, pp->pr_itemsperpage);
                    600:                }
                    601:                pp->pr_phpool = &phpool[idx];
                    602:        } else if ((pp->pr_roflags & PR_PHINPAGE) == 0) {
                    603:                pp->pr_phpool = &phpool[0];
                    604:        }
                    605: #if defined(DIAGNOSTIC)
                    606:        else {
                    607:                pp->pr_phpool = NULL;
                    608:        }
                    609: #endif
1.3       pk        610:
                    611:        /*
                    612:         * Use the slack between the chunks and the page header
                    613:         * for "cache coloring".
                    614:         */
                    615:        slack = off - pp->pr_itemsperpage * pp->pr_size;
                    616:        pp->pr_maxcolor = (slack / align) * align;
                    617:        pp->pr_curcolor = 0;
                    618:
                    619:        pp->pr_nget = 0;
                    620:        pp->pr_nfail = 0;
                    621:        pp->pr_nput = 0;
                    622:        pp->pr_npagealloc = 0;
                    623:        pp->pr_npagefree = 0;
1.1       pk        624:        pp->pr_hiwat = 0;
1.8       thorpej   625:        pp->pr_nidle = 0;
1.3       pk        626:
1.59      thorpej   627: #ifdef POOL_DIAGNOSTIC
1.25      thorpej   628:        if (flags & PR_LOGGING) {
                    629:                if (kmem_map == NULL ||
                    630:                    (pp->pr_log = malloc(pool_logsize * sizeof(struct pool_log),
                    631:                     M_TEMP, M_NOWAIT)) == NULL)
1.20      thorpej   632:                        pp->pr_roflags &= ~PR_LOGGING;
1.3       pk        633:                pp->pr_curlogentry = 0;
                    634:                pp->pr_logsize = pool_logsize;
                    635:        }
1.59      thorpej   636: #endif
1.25      thorpej   637:
                    638:        pp->pr_entered_file = NULL;
                    639:        pp->pr_entered_line = 0;
1.3       pk        640:
1.21      thorpej   641:        simple_lock_init(&pp->pr_slock);
1.1       pk        642:
1.3       pk        643:        /*
1.43      thorpej   644:         * Initialize private page header pool and cache magazine pool if we
                    645:         * haven't done so yet.
1.23      thorpej   646:         * XXX LOCKING.
1.3       pk        647:         */
1.97      yamt      648:        if (phpool[0].pr_size == 0) {
                    649:                int idx;
                    650:                for (idx = 0; idx < PHPOOL_MAX; idx++) {
                    651:                        static char phpool_names[PHPOOL_MAX][6+1+6+1];
                    652:                        int nelem;
                    653:                        size_t sz;
                    654:
                    655:                        nelem = PHPOOL_FREELIST_NELEM(idx);
                    656:                        snprintf(phpool_names[idx], sizeof(phpool_names[idx]),
                    657:                            "phpool-%d", nelem);
                    658:                        sz = sizeof(struct pool_item_header);
                    659:                        if (nelem) {
                    660:                                sz = PR_FREELIST_ALIGN(sz)
1.99      yamt      661:                                    + nelem * sizeof(pool_item_freelist_t);
1.97      yamt      662:                        }
                    663:                        pool_init(&phpool[idx], sz, 0, 0, 0,
1.98      yamt      664:                            phpool_names[idx], &pool_allocator_meta);
1.97      yamt      665:                }
1.62      bjh21     666: #ifdef POOL_SUBPAGE
                    667:                pool_init(&psppool, POOL_SUBPAGE, POOL_SUBPAGE, 0,
1.98      yamt      668:                    PR_RECURSIVE, "psppool", &pool_allocator_meta);
1.62      bjh21     669: #endif
1.43      thorpej   670:                pool_init(&pcgpool, sizeof(struct pool_cache_group), 0, 0,
1.98      yamt      671:                    0, "pcgpool", &pool_allocator_meta);
1.1       pk        672:        }
                    673:
1.23      thorpej   674:        /* Insert into the list of all pools. */
                    675:        simple_lock(&pool_head_slock);
1.102     chs       676:        LIST_INSERT_HEAD(&pool_head, pp, pr_poollist);
1.23      thorpej   677:        simple_unlock(&pool_head_slock);
1.66      thorpej   678:
                    679:        /* Insert this into the list of pools using this allocator. */
1.93      dbj       680:        s = splvm();
1.66      thorpej   681:        simple_lock(&palloc->pa_slock);
                    682:        TAILQ_INSERT_TAIL(&palloc->pa_list, pp, pr_alloc_list);
                    683:        simple_unlock(&palloc->pa_slock);
1.93      dbj       684:        splx(s);
1.1       pk        685: }
                    686:
                    687: /*
                    688:  * De-commision a pool resource.
                    689:  */
                    690: void
1.42      thorpej   691: pool_destroy(struct pool *pp)
1.1       pk        692: {
1.101     thorpej   693:        struct pool_pagelist pq;
1.3       pk        694:        struct pool_item_header *ph;
1.93      dbj       695:        int s;
1.43      thorpej   696:
1.101     thorpej   697:        /* Remove from global pool list */
                    698:        simple_lock(&pool_head_slock);
1.102     chs       699:        LIST_REMOVE(pp, pr_poollist);
1.101     thorpej   700:        if (drainpp == pp)
                    701:                drainpp = NULL;
                    702:        simple_unlock(&pool_head_slock);
                    703:
                    704:        /* Remove this pool from its allocator's list of pools. */
1.93      dbj       705:        s = splvm();
1.66      thorpej   706:        simple_lock(&pp->pr_alloc->pa_slock);
                    707:        TAILQ_REMOVE(&pp->pr_alloc->pa_list, pp, pr_alloc_list);
                    708:        simple_unlock(&pp->pr_alloc->pa_slock);
1.93      dbj       709:        splx(s);
1.66      thorpej   710:
1.101     thorpej   711:        s = splvm();
                    712:        simple_lock(&pp->pr_slock);
                    713:
1.102     chs       714:        KASSERT(LIST_EMPTY(&pp->pr_cachelist));
1.3       pk        715:
                    716: #ifdef DIAGNOSTIC
1.20      thorpej   717:        if (pp->pr_nout != 0) {
1.25      thorpej   718:                pr_printlog(pp, NULL, printf);
1.80      provos    719:                panic("pool_destroy: pool busy: still out: %u",
1.20      thorpej   720:                    pp->pr_nout);
1.3       pk        721:        }
                    722: #endif
1.1       pk        723:
1.101     thorpej   724:        KASSERT(LIST_EMPTY(&pp->pr_fullpages));
                    725:        KASSERT(LIST_EMPTY(&pp->pr_partpages));
                    726:
1.3       pk        727:        /* Remove all pages */
1.101     thorpej   728:        LIST_INIT(&pq);
1.88      chs       729:        while ((ph = LIST_FIRST(&pp->pr_emptypages)) != NULL)
1.101     thorpej   730:                pr_rmpage(pp, ph, &pq);
                    731:
                    732:        simple_unlock(&pp->pr_slock);
                    733:        splx(s);
1.3       pk        734:
1.101     thorpej   735:        pr_pagelist_free(pp, &pq);
1.3       pk        736:
1.59      thorpej   737: #ifdef POOL_DIAGNOSTIC
1.20      thorpej   738:        if ((pp->pr_roflags & PR_LOGGING) != 0)
1.3       pk        739:                free(pp->pr_log, M_TEMP);
1.59      thorpej   740: #endif
1.1       pk        741: }
                    742:
1.68      thorpej   743: void
                    744: pool_set_drain_hook(struct pool *pp, void (*fn)(void *, int), void *arg)
                    745: {
                    746:
                    747:        /* XXX no locking -- must be used just after pool_init() */
                    748: #ifdef DIAGNOSTIC
                    749:        if (pp->pr_drain_hook != NULL)
                    750:                panic("pool_set_drain_hook(%s): already set", pp->pr_wchan);
                    751: #endif
                    752:        pp->pr_drain_hook = fn;
                    753:        pp->pr_drain_hook_arg = arg;
                    754: }
                    755:
1.88      chs       756: static struct pool_item_header *
1.55      thorpej   757: pool_alloc_item_header(struct pool *pp, caddr_t storage, int flags)
                    758: {
                    759:        struct pool_item_header *ph;
                    760:        int s;
                    761:
                    762:        LOCK_ASSERT(simple_lock_held(&pp->pr_slock) == 0);
                    763:
                    764:        if ((pp->pr_roflags & PR_PHINPAGE) != 0)
                    765:                ph = (struct pool_item_header *) (storage + pp->pr_phoffset);
                    766:        else {
1.85      pk        767:                s = splvm();
1.97      yamt      768:                ph = pool_get(pp->pr_phpool, flags);
1.55      thorpej   769:                splx(s);
                    770:        }
                    771:
                    772:        return (ph);
                    773: }
1.1       pk        774:
                    775: /*
1.3       pk        776:  * Grab an item from the pool; must be called at appropriate spl level
1.1       pk        777:  */
1.3       pk        778: void *
1.59      thorpej   779: #ifdef POOL_DIAGNOSTIC
1.42      thorpej   780: _pool_get(struct pool *pp, int flags, const char *file, long line)
1.56      sommerfe  781: #else
                    782: pool_get(struct pool *pp, int flags)
                    783: #endif
1.1       pk        784: {
                    785:        struct pool_item *pi;
1.3       pk        786:        struct pool_item_header *ph;
1.55      thorpej   787:        void *v;
1.1       pk        788:
1.2       pk        789: #ifdef DIAGNOSTIC
1.95      atatat    790:        if (__predict_false(pp->pr_itemsperpage == 0))
                    791:                panic("pool_get: pool %p: pr_itemsperpage is zero, "
                    792:                    "pool not initialized?", pp);
1.84      thorpej   793:        if (__predict_false(curlwp == NULL && doing_shutdown == 0 &&
1.37      sommerfe  794:                            (flags & PR_WAITOK) != 0))
1.77      matt      795:                panic("pool_get: %s: must have NOWAIT", pp->pr_wchan);
1.58      thorpej   796:
1.102     chs       797: #endif /* DIAGNOSTIC */
1.58      thorpej   798: #ifdef LOCKDEBUG
                    799:        if (flags & PR_WAITOK)
                    800:                simple_lock_only_held(NULL, "pool_get(PR_WAITOK)");
1.102     chs       801:        SCHED_ASSERT_UNLOCKED();
1.56      sommerfe  802: #endif
1.1       pk        803:
1.21      thorpej   804:        simple_lock(&pp->pr_slock);
1.25      thorpej   805:        pr_enter(pp, file, line);
1.20      thorpej   806:
                    807:  startover:
                    808:        /*
                    809:         * Check to see if we've reached the hard limit.  If we have,
                    810:         * and we can wait, then wait until an item has been returned to
                    811:         * the pool.
                    812:         */
                    813: #ifdef DIAGNOSTIC
1.34      thorpej   814:        if (__predict_false(pp->pr_nout > pp->pr_hardlimit)) {
1.25      thorpej   815:                pr_leave(pp);
1.21      thorpej   816:                simple_unlock(&pp->pr_slock);
1.20      thorpej   817:                panic("pool_get: %s: crossed hard limit", pp->pr_wchan);
                    818:        }
                    819: #endif
1.34      thorpej   820:        if (__predict_false(pp->pr_nout == pp->pr_hardlimit)) {
1.68      thorpej   821:                if (pp->pr_drain_hook != NULL) {
                    822:                        /*
                    823:                         * Since the drain hook is going to free things
                    824:                         * back to the pool, unlock, call the hook, re-lock,
                    825:                         * and check the hardlimit condition again.
                    826:                         */
                    827:                        pr_leave(pp);
                    828:                        simple_unlock(&pp->pr_slock);
                    829:                        (*pp->pr_drain_hook)(pp->pr_drain_hook_arg, flags);
                    830:                        simple_lock(&pp->pr_slock);
                    831:                        pr_enter(pp, file, line);
                    832:                        if (pp->pr_nout < pp->pr_hardlimit)
                    833:                                goto startover;
                    834:                }
                    835:
1.29      sommerfe  836:                if ((flags & PR_WAITOK) && !(flags & PR_LIMITFAIL)) {
1.20      thorpej   837:                        /*
                    838:                         * XXX: A warning isn't logged in this case.  Should
                    839:                         * it be?
                    840:                         */
                    841:                        pp->pr_flags |= PR_WANTED;
1.25      thorpej   842:                        pr_leave(pp);
1.40      sommerfe  843:                        ltsleep(pp, PSWP, pp->pr_wchan, 0, &pp->pr_slock);
1.25      thorpej   844:                        pr_enter(pp, file, line);
1.20      thorpej   845:                        goto startover;
                    846:                }
1.31      thorpej   847:
                    848:                /*
                    849:                 * Log a message that the hard limit has been hit.
                    850:                 */
                    851:                if (pp->pr_hardlimit_warning != NULL &&
                    852:                    ratecheck(&pp->pr_hardlimit_warning_last,
                    853:                              &pp->pr_hardlimit_ratecap))
                    854:                        log(LOG_ERR, "%s\n", pp->pr_hardlimit_warning);
1.21      thorpej   855:
                    856:                pp->pr_nfail++;
                    857:
1.25      thorpej   858:                pr_leave(pp);
1.21      thorpej   859:                simple_unlock(&pp->pr_slock);
1.20      thorpej   860:                return (NULL);
                    861:        }
                    862:
1.3       pk        863:        /*
                    864:         * The convention we use is that if `curpage' is not NULL, then
                    865:         * it points at a non-empty bucket. In particular, `curpage'
                    866:         * never points at a page header which has PR_PHINPAGE set and
                    867:         * has no items in its bucket.
                    868:         */
1.20      thorpej   869:        if ((ph = pp->pr_curpage) == NULL) {
                    870: #ifdef DIAGNOSTIC
                    871:                if (pp->pr_nitems != 0) {
1.21      thorpej   872:                        simple_unlock(&pp->pr_slock);
1.20      thorpej   873:                        printf("pool_get: %s: curpage NULL, nitems %u\n",
                    874:                            pp->pr_wchan, pp->pr_nitems);
1.80      provos    875:                        panic("pool_get: nitems inconsistent");
1.20      thorpej   876:                }
                    877: #endif
                    878:
1.21      thorpej   879:                /*
                    880:                 * Call the back-end page allocator for more memory.
                    881:                 * Release the pool lock, as the back-end page allocator
                    882:                 * may block.
                    883:                 */
1.25      thorpej   884:                pr_leave(pp);
1.21      thorpej   885:                simple_unlock(&pp->pr_slock);
1.66      thorpej   886:                v = pool_allocator_alloc(pp, flags);
1.55      thorpej   887:                if (__predict_true(v != NULL))
                    888:                        ph = pool_alloc_item_header(pp, v, flags);
1.15      pk        889:
1.55      thorpej   890:                if (__predict_false(v == NULL || ph == NULL)) {
                    891:                        if (v != NULL)
1.66      thorpej   892:                                pool_allocator_free(pp, v);
1.55      thorpej   893:
1.91      yamt      894:                        simple_lock(&pp->pr_slock);
                    895:                        pr_enter(pp, file, line);
                    896:
1.21      thorpej   897:                        /*
1.55      thorpej   898:                         * We were unable to allocate a page or item
                    899:                         * header, but we released the lock during
                    900:                         * allocation, so perhaps items were freed
                    901:                         * back to the pool.  Check for this case.
1.21      thorpej   902:                         */
                    903:                        if (pp->pr_curpage != NULL)
                    904:                                goto startover;
1.15      pk        905:
1.3       pk        906:                        if ((flags & PR_WAITOK) == 0) {
                    907:                                pp->pr_nfail++;
1.25      thorpej   908:                                pr_leave(pp);
1.21      thorpej   909:                                simple_unlock(&pp->pr_slock);
1.1       pk        910:                                return (NULL);
1.3       pk        911:                        }
                    912:
1.15      pk        913:                        /*
                    914:                         * Wait for items to be returned to this pool.
1.21      thorpej   915:                         *
1.109     christos  916:                         * wake up once a second and try again,
                    917:                         * as the check in pool_cache_put_paddr() is racy.
1.15      pk        918:                         */
1.1       pk        919:                        pp->pr_flags |= PR_WANTED;
1.66      thorpej   920:                        /* PA_WANTED is already set on the allocator. */
1.25      thorpej   921:                        pr_leave(pp);
1.109     christos  922:                        ltsleep(pp, PSWP, pp->pr_wchan, hz, &pp->pr_slock);
1.25      thorpej   923:                        pr_enter(pp, file, line);
1.20      thorpej   924:                        goto startover;
1.1       pk        925:                }
1.3       pk        926:
1.15      pk        927:                /* We have more memory; add it to the pool */
1.91      yamt      928:                simple_lock(&pp->pr_slock);
                    929:                pr_enter(pp, file, line);
1.55      thorpej   930:                pool_prime_page(pp, v, ph);
1.15      pk        931:                pp->pr_npagealloc++;
                    932:
1.20      thorpej   933:                /* Start the allocation process over. */
                    934:                goto startover;
1.3       pk        935:        }
1.97      yamt      936:        if (pp->pr_roflags & PR_NOTOUCH) {
                    937: #ifdef DIAGNOSTIC
                    938:                if (__predict_false(ph->ph_nmissing == pp->pr_itemsperpage)) {
                    939:                        pr_leave(pp);
                    940:                        simple_unlock(&pp->pr_slock);
                    941:                        panic("pool_get: %s: page empty", pp->pr_wchan);
                    942:                }
                    943: #endif
                    944:                v = pr_item_notouch_get(pp, ph);
                    945: #ifdef POOL_DIAGNOSTIC
                    946:                pr_log(pp, v, PRLOG_GET, file, line);
                    947: #endif
                    948:        } else {
1.102     chs       949:                v = pi = LIST_FIRST(&ph->ph_itemlist);
1.97      yamt      950:                if (__predict_false(v == NULL)) {
                    951:                        pr_leave(pp);
                    952:                        simple_unlock(&pp->pr_slock);
                    953:                        panic("pool_get: %s: page empty", pp->pr_wchan);
                    954:                }
1.20      thorpej   955: #ifdef DIAGNOSTIC
1.97      yamt      956:                if (__predict_false(pp->pr_nitems == 0)) {
                    957:                        pr_leave(pp);
                    958:                        simple_unlock(&pp->pr_slock);
                    959:                        printf("pool_get: %s: items on itemlist, nitems %u\n",
                    960:                            pp->pr_wchan, pp->pr_nitems);
                    961:                        panic("pool_get: nitems inconsistent");
                    962:                }
1.65      enami     963: #endif
1.56      sommerfe  964:
1.65      enami     965: #ifdef POOL_DIAGNOSTIC
1.97      yamt      966:                pr_log(pp, v, PRLOG_GET, file, line);
1.65      enami     967: #endif
1.3       pk        968:
1.65      enami     969: #ifdef DIAGNOSTIC
1.97      yamt      970:                if (__predict_false(pi->pi_magic != PI_MAGIC)) {
                    971:                        pr_printlog(pp, pi, printf);
                    972:                        panic("pool_get(%s): free list modified: "
                    973:                            "magic=%x; page %p; item addr %p\n",
                    974:                            pp->pr_wchan, pi->pi_magic, ph->ph_page, pi);
                    975:                }
1.3       pk        976: #endif
                    977:
1.97      yamt      978:                /*
                    979:                 * Remove from item list.
                    980:                 */
1.102     chs       981:                LIST_REMOVE(pi, pi_list);
1.97      yamt      982:        }
1.20      thorpej   983:        pp->pr_nitems--;
                    984:        pp->pr_nout++;
1.6       thorpej   985:        if (ph->ph_nmissing == 0) {
                    986: #ifdef DIAGNOSTIC
1.34      thorpej   987:                if (__predict_false(pp->pr_nidle == 0))
1.6       thorpej   988:                        panic("pool_get: nidle inconsistent");
                    989: #endif
                    990:                pp->pr_nidle--;
1.88      chs       991:
                    992:                /*
                    993:                 * This page was previously empty.  Move it to the list of
                    994:                 * partially-full pages.  This page is already curpage.
                    995:                 */
                    996:                LIST_REMOVE(ph, ph_pagelist);
                    997:                LIST_INSERT_HEAD(&pp->pr_partpages, ph, ph_pagelist);
1.6       thorpej   998:        }
1.3       pk        999:        ph->ph_nmissing++;
1.97      yamt     1000:        if (ph->ph_nmissing == pp->pr_itemsperpage) {
1.21      thorpej  1001: #ifdef DIAGNOSTIC
1.97      yamt     1002:                if (__predict_false((pp->pr_roflags & PR_NOTOUCH) == 0 &&
1.102     chs      1003:                    !LIST_EMPTY(&ph->ph_itemlist))) {
1.25      thorpej  1004:                        pr_leave(pp);
1.21      thorpej  1005:                        simple_unlock(&pp->pr_slock);
                   1006:                        panic("pool_get: %s: nmissing inconsistent",
                   1007:                            pp->pr_wchan);
                   1008:                }
                   1009: #endif
1.3       pk       1010:                /*
1.88      chs      1011:                 * This page is now full.  Move it to the full list
                   1012:                 * and select a new current page.
1.3       pk       1013:                 */
1.88      chs      1014:                LIST_REMOVE(ph, ph_pagelist);
                   1015:                LIST_INSERT_HEAD(&pp->pr_fullpages, ph, ph_pagelist);
                   1016:                pool_update_curpage(pp);
1.1       pk       1017:        }
1.3       pk       1018:
                   1019:        pp->pr_nget++;
1.111     christos 1020:        pr_leave(pp);
1.20      thorpej  1021:
                   1022:        /*
                   1023:         * If we have a low water mark and we are now below that low
                   1024:         * water mark, add more items to the pool.
                   1025:         */
1.53      thorpej  1026:        if (POOL_NEEDS_CATCHUP(pp) && pool_catchup(pp) != 0) {
1.20      thorpej  1027:                /*
                   1028:                 * XXX: Should we log a warning?  Should we set up a timeout
                   1029:                 * to try again in a second or so?  The latter could break
                   1030:                 * a caller's assumptions about interrupt protection, etc.
                   1031:                 */
                   1032:        }
                   1033:
1.21      thorpej  1034:        simple_unlock(&pp->pr_slock);
1.1       pk       1035:        return (v);
                   1036: }
                   1037:
                   1038: /*
1.43      thorpej  1039:  * Internal version of pool_put().  Pool is already locked/entered.
1.1       pk       1040:  */
1.43      thorpej  1041: static void
1.101     thorpej  1042: pool_do_put(struct pool *pp, void *v, struct pool_pagelist *pq)
1.1       pk       1043: {
                   1044:        struct pool_item *pi = v;
1.3       pk       1045:        struct pool_item_header *ph;
                   1046:        caddr_t page;
1.21      thorpej  1047:        int s;
1.3       pk       1048:
1.61      chs      1049:        LOCK_ASSERT(simple_lock_held(&pp->pr_slock));
1.102     chs      1050:        SCHED_ASSERT_UNLOCKED();
1.61      chs      1051:
1.66      thorpej  1052:        page = (caddr_t)((u_long)v & pp->pr_alloc->pa_pagemask);
1.1       pk       1053:
1.30      thorpej  1054: #ifdef DIAGNOSTIC
1.34      thorpej  1055:        if (__predict_false(pp->pr_nout == 0)) {
1.30      thorpej  1056:                printf("pool %s: putting with none out\n",
                   1057:                    pp->pr_wchan);
                   1058:                panic("pool_put");
                   1059:        }
                   1060: #endif
1.3       pk       1061:
1.34      thorpej  1062:        if (__predict_false((ph = pr_find_pagehead(pp, page)) == NULL)) {
1.25      thorpej  1063:                pr_printlog(pp, NULL, printf);
1.3       pk       1064:                panic("pool_put: %s: page header missing", pp->pr_wchan);
                   1065:        }
1.28      thorpej  1066:
                   1067: #ifdef LOCKDEBUG
                   1068:        /*
                   1069:         * Check if we're freeing a locked simple lock.
                   1070:         */
                   1071:        simple_lock_freecheck((caddr_t)pi, ((caddr_t)pi) + pp->pr_size);
                   1072: #endif
1.3       pk       1073:
                   1074:        /*
                   1075:         * Return to item list.
                   1076:         */
1.97      yamt     1077:        if (pp->pr_roflags & PR_NOTOUCH) {
                   1078:                pr_item_notouch_put(pp, ph, v);
                   1079:        } else {
1.2       pk       1080: #ifdef DIAGNOSTIC
1.97      yamt     1081:                pi->pi_magic = PI_MAGIC;
1.3       pk       1082: #endif
1.32      chs      1083: #ifdef DEBUG
1.97      yamt     1084:                {
                   1085:                        int i, *ip = v;
1.32      chs      1086:
1.97      yamt     1087:                        for (i = 0; i < pp->pr_size / sizeof(int); i++) {
                   1088:                                *ip++ = PI_MAGIC;
                   1089:                        }
1.32      chs      1090:                }
                   1091: #endif
                   1092:
1.102     chs      1093:                LIST_INSERT_HEAD(&ph->ph_itemlist, pi, pi_list);
1.97      yamt     1094:        }
1.79      thorpej  1095:        KDASSERT(ph->ph_nmissing != 0);
1.3       pk       1096:        ph->ph_nmissing--;
                   1097:        pp->pr_nput++;
1.20      thorpej  1098:        pp->pr_nitems++;
                   1099:        pp->pr_nout--;
1.3       pk       1100:
                   1101:        /* Cancel "pool empty" condition if it exists */
                   1102:        if (pp->pr_curpage == NULL)
                   1103:                pp->pr_curpage = ph;
                   1104:
                   1105:        if (pp->pr_flags & PR_WANTED) {
                   1106:                pp->pr_flags &= ~PR_WANTED;
1.15      pk       1107:                if (ph->ph_nmissing == 0)
                   1108:                        pp->pr_nidle++;
1.3       pk       1109:                wakeup((caddr_t)pp);
                   1110:                return;
                   1111:        }
                   1112:
                   1113:        /*
1.88      chs      1114:         * If this page is now empty, do one of two things:
1.21      thorpej  1115:         *
1.88      chs      1116:         *      (1) If we have more pages than the page high water mark,
1.96      thorpej  1117:         *          free the page back to the system.  ONLY CONSIDER
1.90      thorpej  1118:         *          FREEING BACK A PAGE IF WE HAVE MORE THAN OUR MINIMUM PAGE
                   1119:         *          CLAIM.
1.21      thorpej  1120:         *
1.88      chs      1121:         *      (2) Otherwise, move the page to the empty page list.
                   1122:         *
                   1123:         * Either way, select a new current page (so we use a partially-full
                   1124:         * page if one is available).
1.3       pk       1125:         */
                   1126:        if (ph->ph_nmissing == 0) {
1.6       thorpej  1127:                pp->pr_nidle++;
1.90      thorpej  1128:                if (pp->pr_npages > pp->pr_minpages &&
                   1129:                    (pp->pr_npages > pp->pr_maxpages ||
                   1130:                     (pp->pr_alloc->pa_flags & PA_WANT) != 0)) {
1.101     thorpej  1131:                        pr_rmpage(pp, ph, pq);
1.3       pk       1132:                } else {
1.88      chs      1133:                        LIST_REMOVE(ph, ph_pagelist);
                   1134:                        LIST_INSERT_HEAD(&pp->pr_emptypages, ph, ph_pagelist);
1.3       pk       1135:
1.21      thorpej  1136:                        /*
                   1137:                         * Update the timestamp on the page.  A page must
                   1138:                         * be idle for some period of time before it can
                   1139:                         * be reclaimed by the pagedaemon.  This minimizes
                   1140:                         * ping-pong'ing for memory.
                   1141:                         */
                   1142:                        s = splclock();
                   1143:                        ph->ph_time = mono_time;
                   1144:                        splx(s);
1.1       pk       1145:                }
1.88      chs      1146:                pool_update_curpage(pp);
1.1       pk       1147:        }
1.88      chs      1148:
1.21      thorpej  1149:        /*
1.88      chs      1150:         * If the page was previously completely full, move it to the
                   1151:         * partially-full list and make it the current page.  The next
                   1152:         * allocation will get the item from this page, instead of
                   1153:         * further fragmenting the pool.
1.21      thorpej  1154:         */
                   1155:        else if (ph->ph_nmissing == (pp->pr_itemsperpage - 1)) {
1.88      chs      1156:                LIST_REMOVE(ph, ph_pagelist);
                   1157:                LIST_INSERT_HEAD(&pp->pr_partpages, ph, ph_pagelist);
1.21      thorpej  1158:                pp->pr_curpage = ph;
                   1159:        }
1.43      thorpej  1160: }
                   1161:
                   1162: /*
                   1163:  * Return resource to the pool; must be called at appropriate spl level
                   1164:  */
1.59      thorpej  1165: #ifdef POOL_DIAGNOSTIC
1.43      thorpej  1166: void
                   1167: _pool_put(struct pool *pp, void *v, const char *file, long line)
                   1168: {
1.101     thorpej  1169:        struct pool_pagelist pq;
                   1170:
                   1171:        LIST_INIT(&pq);
1.43      thorpej  1172:
                   1173:        simple_lock(&pp->pr_slock);
                   1174:        pr_enter(pp, file, line);
                   1175:
1.56      sommerfe 1176:        pr_log(pp, v, PRLOG_PUT, file, line);
                   1177:
1.101     thorpej  1178:        pool_do_put(pp, v, &pq);
1.21      thorpej  1179:
1.25      thorpej  1180:        pr_leave(pp);
1.21      thorpej  1181:        simple_unlock(&pp->pr_slock);
1.101     thorpej  1182:
1.102     chs      1183:        pr_pagelist_free(pp, &pq);
1.1       pk       1184: }
1.57      sommerfe 1185: #undef pool_put
1.59      thorpej  1186: #endif /* POOL_DIAGNOSTIC */
1.1       pk       1187:
1.56      sommerfe 1188: void
                   1189: pool_put(struct pool *pp, void *v)
                   1190: {
1.101     thorpej  1191:        struct pool_pagelist pq;
                   1192:
                   1193:        LIST_INIT(&pq);
1.56      sommerfe 1194:
                   1195:        simple_lock(&pp->pr_slock);
1.101     thorpej  1196:        pool_do_put(pp, v, &pq);
                   1197:        simple_unlock(&pp->pr_slock);
1.56      sommerfe 1198:
1.102     chs      1199:        pr_pagelist_free(pp, &pq);
1.56      sommerfe 1200: }
1.57      sommerfe 1201:
1.59      thorpej  1202: #ifdef POOL_DIAGNOSTIC
1.57      sommerfe 1203: #define                pool_put(h, v)  _pool_put((h), (v), __FILE__, __LINE__)
1.56      sommerfe 1204: #endif
1.74      thorpej  1205:
                   1206: /*
                   1207:  * Add N items to the pool.
                   1208:  */
                   1209: int
                   1210: pool_prime(struct pool *pp, int n)
                   1211: {
1.83      scw      1212:        struct pool_item_header *ph = NULL;
1.74      thorpej  1213:        caddr_t cp;
1.75      simonb   1214:        int newpages;
1.74      thorpej  1215:
                   1216:        simple_lock(&pp->pr_slock);
                   1217:
                   1218:        newpages = roundup(n, pp->pr_itemsperpage) / pp->pr_itemsperpage;
                   1219:
                   1220:        while (newpages-- > 0) {
                   1221:                simple_unlock(&pp->pr_slock);
                   1222:                cp = pool_allocator_alloc(pp, PR_NOWAIT);
                   1223:                if (__predict_true(cp != NULL))
                   1224:                        ph = pool_alloc_item_header(pp, cp, PR_NOWAIT);
                   1225:
                   1226:                if (__predict_false(cp == NULL || ph == NULL)) {
                   1227:                        if (cp != NULL)
                   1228:                                pool_allocator_free(pp, cp);
1.91      yamt     1229:                        simple_lock(&pp->pr_slock);
1.74      thorpej  1230:                        break;
                   1231:                }
                   1232:
1.91      yamt     1233:                simple_lock(&pp->pr_slock);
1.74      thorpej  1234:                pool_prime_page(pp, cp, ph);
                   1235:                pp->pr_npagealloc++;
                   1236:                pp->pr_minpages++;
                   1237:        }
                   1238:
                   1239:        if (pp->pr_minpages >= pp->pr_maxpages)
                   1240:                pp->pr_maxpages = pp->pr_minpages + 1;  /* XXX */
                   1241:
                   1242:        simple_unlock(&pp->pr_slock);
                   1243:        return (0);
                   1244: }
1.55      thorpej  1245:
                   1246: /*
1.3       pk       1247:  * Add a page worth of items to the pool.
1.21      thorpej  1248:  *
                   1249:  * Note, we must be called with the pool descriptor LOCKED.
1.3       pk       1250:  */
1.55      thorpej  1251: static void
                   1252: pool_prime_page(struct pool *pp, caddr_t storage, struct pool_item_header *ph)
1.3       pk       1253: {
                   1254:        struct pool_item *pi;
                   1255:        caddr_t cp = storage;
                   1256:        unsigned int align = pp->pr_align;
                   1257:        unsigned int ioff = pp->pr_itemoffset;
1.55      thorpej  1258:        int n;
1.89      yamt     1259:        int s;
1.36      pk       1260:
1.91      yamt     1261:        LOCK_ASSERT(simple_lock_held(&pp->pr_slock));
                   1262:
1.66      thorpej  1263: #ifdef DIAGNOSTIC
                   1264:        if (((u_long)cp & (pp->pr_alloc->pa_pagesz - 1)) != 0)
1.36      pk       1265:                panic("pool_prime_page: %s: unaligned page", pp->pr_wchan);
1.66      thorpej  1266: #endif
1.3       pk       1267:
                   1268:        /*
                   1269:         * Insert page header.
                   1270:         */
1.88      chs      1271:        LIST_INSERT_HEAD(&pp->pr_emptypages, ph, ph_pagelist);
1.102     chs      1272:        LIST_INIT(&ph->ph_itemlist);
1.3       pk       1273:        ph->ph_page = storage;
                   1274:        ph->ph_nmissing = 0;
1.89      yamt     1275:        s = splclock();
                   1276:        ph->ph_time = mono_time;
                   1277:        splx(s);
1.88      chs      1278:        if ((pp->pr_roflags & PR_PHINPAGE) == 0)
                   1279:                SPLAY_INSERT(phtree, &pp->pr_phtree, ph);
1.3       pk       1280:
1.6       thorpej  1281:        pp->pr_nidle++;
                   1282:
1.3       pk       1283:        /*
                   1284:         * Color this page.
                   1285:         */
                   1286:        cp = (caddr_t)(cp + pp->pr_curcolor);
                   1287:        if ((pp->pr_curcolor += align) > pp->pr_maxcolor)
                   1288:                pp->pr_curcolor = 0;
                   1289:
                   1290:        /*
                   1291:         * Adjust storage to apply aligment to `pr_itemoffset' in each item.
                   1292:         */
                   1293:        if (ioff != 0)
                   1294:                cp = (caddr_t)(cp + (align - ioff));
                   1295:
                   1296:        /*
                   1297:         * Insert remaining chunks on the bucket list.
                   1298:         */
                   1299:        n = pp->pr_itemsperpage;
1.20      thorpej  1300:        pp->pr_nitems += n;
1.3       pk       1301:
1.97      yamt     1302:        if (pp->pr_roflags & PR_NOTOUCH) {
1.99      yamt     1303:                pool_item_freelist_t *freelist = PR_FREELIST(ph);
1.97      yamt     1304:                int i;
                   1305:
1.99      yamt     1306:                ph->ph_off = cp - storage;
1.97      yamt     1307:                ph->ph_firstfree = 0;
                   1308:                for (i = 0; i < n - 1; i++)
                   1309:                        freelist[i] = i + 1;
                   1310:                freelist[n - 1] = PR_INDEX_EOL;
                   1311:        } else {
                   1312:                while (n--) {
                   1313:                        pi = (struct pool_item *)cp;
1.78      thorpej  1314:
1.97      yamt     1315:                        KASSERT(((((vaddr_t)pi) + ioff) & (align - 1)) == 0);
1.3       pk       1316:
1.97      yamt     1317:                        /* Insert on page list */
1.102     chs      1318:                        LIST_INSERT_HEAD(&ph->ph_itemlist, pi, pi_list);
1.3       pk       1319: #ifdef DIAGNOSTIC
1.97      yamt     1320:                        pi->pi_magic = PI_MAGIC;
1.3       pk       1321: #endif
1.97      yamt     1322:                        cp = (caddr_t)(cp + pp->pr_size);
                   1323:                }
1.3       pk       1324:        }
                   1325:
                   1326:        /*
                   1327:         * If the pool was depleted, point at the new page.
                   1328:         */
                   1329:        if (pp->pr_curpage == NULL)
                   1330:                pp->pr_curpage = ph;
                   1331:
                   1332:        if (++pp->pr_npages > pp->pr_hiwat)
                   1333:                pp->pr_hiwat = pp->pr_npages;
                   1334: }
                   1335:
1.20      thorpej  1336: /*
1.52      thorpej  1337:  * Used by pool_get() when nitems drops below the low water mark.  This
1.88      chs      1338:  * is used to catch up pr_nitems with the low water mark.
1.20      thorpej  1339:  *
1.21      thorpej  1340:  * Note 1, we never wait for memory here, we let the caller decide what to do.
1.20      thorpej  1341:  *
1.73      thorpej  1342:  * Note 2, we must be called with the pool already locked, and we return
1.20      thorpej  1343:  * with it locked.
                   1344:  */
                   1345: static int
1.42      thorpej  1346: pool_catchup(struct pool *pp)
1.20      thorpej  1347: {
1.83      scw      1348:        struct pool_item_header *ph = NULL;
1.20      thorpej  1349:        caddr_t cp;
                   1350:        int error = 0;
                   1351:
1.54      thorpej  1352:        while (POOL_NEEDS_CATCHUP(pp)) {
1.20      thorpej  1353:                /*
1.21      thorpej  1354:                 * Call the page back-end allocator for more memory.
                   1355:                 *
                   1356:                 * XXX: We never wait, so should we bother unlocking
                   1357:                 * the pool descriptor?
1.20      thorpej  1358:                 */
1.21      thorpej  1359:                simple_unlock(&pp->pr_slock);
1.66      thorpej  1360:                cp = pool_allocator_alloc(pp, PR_NOWAIT);
1.55      thorpej  1361:                if (__predict_true(cp != NULL))
                   1362:                        ph = pool_alloc_item_header(pp, cp, PR_NOWAIT);
                   1363:                if (__predict_false(cp == NULL || ph == NULL)) {
                   1364:                        if (cp != NULL)
1.66      thorpej  1365:                                pool_allocator_free(pp, cp);
1.20      thorpej  1366:                        error = ENOMEM;
1.91      yamt     1367:                        simple_lock(&pp->pr_slock);
1.20      thorpej  1368:                        break;
                   1369:                }
1.91      yamt     1370:                simple_lock(&pp->pr_slock);
1.55      thorpej  1371:                pool_prime_page(pp, cp, ph);
1.26      thorpej  1372:                pp->pr_npagealloc++;
1.20      thorpej  1373:        }
                   1374:
                   1375:        return (error);
                   1376: }
                   1377:
1.88      chs      1378: static void
                   1379: pool_update_curpage(struct pool *pp)
                   1380: {
                   1381:
                   1382:        pp->pr_curpage = LIST_FIRST(&pp->pr_partpages);
                   1383:        if (pp->pr_curpage == NULL) {
                   1384:                pp->pr_curpage = LIST_FIRST(&pp->pr_emptypages);
                   1385:        }
                   1386: }
                   1387:
1.3       pk       1388: void
1.42      thorpej  1389: pool_setlowat(struct pool *pp, int n)
1.3       pk       1390: {
1.15      pk       1391:
1.21      thorpej  1392:        simple_lock(&pp->pr_slock);
                   1393:
1.3       pk       1394:        pp->pr_minitems = n;
1.15      pk       1395:        pp->pr_minpages = (n == 0)
                   1396:                ? 0
1.18      thorpej  1397:                : roundup(n, pp->pr_itemsperpage) / pp->pr_itemsperpage;
1.20      thorpej  1398:
                   1399:        /* Make sure we're caught up with the newly-set low water mark. */
1.75      simonb   1400:        if (POOL_NEEDS_CATCHUP(pp) && pool_catchup(pp) != 0) {
1.20      thorpej  1401:                /*
                   1402:                 * XXX: Should we log a warning?  Should we set up a timeout
                   1403:                 * to try again in a second or so?  The latter could break
                   1404:                 * a caller's assumptions about interrupt protection, etc.
                   1405:                 */
                   1406:        }
1.21      thorpej  1407:
                   1408:        simple_unlock(&pp->pr_slock);
1.3       pk       1409: }
                   1410:
                   1411: void
1.42      thorpej  1412: pool_sethiwat(struct pool *pp, int n)
1.3       pk       1413: {
1.15      pk       1414:
1.21      thorpej  1415:        simple_lock(&pp->pr_slock);
                   1416:
1.15      pk       1417:        pp->pr_maxpages = (n == 0)
                   1418:                ? 0
1.18      thorpej  1419:                : roundup(n, pp->pr_itemsperpage) / pp->pr_itemsperpage;
1.21      thorpej  1420:
                   1421:        simple_unlock(&pp->pr_slock);
1.3       pk       1422: }
                   1423:
1.20      thorpej  1424: void
1.42      thorpej  1425: pool_sethardlimit(struct pool *pp, int n, const char *warnmess, int ratecap)
1.20      thorpej  1426: {
                   1427:
1.21      thorpej  1428:        simple_lock(&pp->pr_slock);
1.20      thorpej  1429:
                   1430:        pp->pr_hardlimit = n;
                   1431:        pp->pr_hardlimit_warning = warnmess;
1.31      thorpej  1432:        pp->pr_hardlimit_ratecap.tv_sec = ratecap;
                   1433:        pp->pr_hardlimit_warning_last.tv_sec = 0;
                   1434:        pp->pr_hardlimit_warning_last.tv_usec = 0;
1.20      thorpej  1435:
                   1436:        /*
1.21      thorpej  1437:         * In-line version of pool_sethiwat(), because we don't want to
                   1438:         * release the lock.
1.20      thorpej  1439:         */
                   1440:        pp->pr_maxpages = (n == 0)
                   1441:                ? 0
                   1442:                : roundup(n, pp->pr_itemsperpage) / pp->pr_itemsperpage;
1.21      thorpej  1443:
                   1444:        simple_unlock(&pp->pr_slock);
1.20      thorpej  1445: }
1.3       pk       1446:
                   1447: /*
                   1448:  * Release all complete pages that have not been used recently.
                   1449:  */
1.66      thorpej  1450: int
1.59      thorpej  1451: #ifdef POOL_DIAGNOSTIC
1.42      thorpej  1452: _pool_reclaim(struct pool *pp, const char *file, long line)
1.56      sommerfe 1453: #else
                   1454: pool_reclaim(struct pool *pp)
                   1455: #endif
1.3       pk       1456: {
                   1457:        struct pool_item_header *ph, *phnext;
1.43      thorpej  1458:        struct pool_cache *pc;
1.61      chs      1459:        struct pool_pagelist pq;
1.102     chs      1460:        struct pool_cache_grouplist pcgl;
                   1461:        struct timeval curtime, diff;
1.21      thorpej  1462:        int s;
1.3       pk       1463:
1.68      thorpej  1464:        if (pp->pr_drain_hook != NULL) {
                   1465:                /*
                   1466:                 * The drain hook must be called with the pool unlocked.
                   1467:                 */
                   1468:                (*pp->pr_drain_hook)(pp->pr_drain_hook_arg, PR_NOWAIT);
                   1469:        }
                   1470:
1.21      thorpej  1471:        if (simple_lock_try(&pp->pr_slock) == 0)
1.66      thorpej  1472:                return (0);
1.25      thorpej  1473:        pr_enter(pp, file, line);
1.68      thorpej  1474:
1.88      chs      1475:        LIST_INIT(&pq);
1.102     chs      1476:        LIST_INIT(&pcgl);
1.3       pk       1477:
1.43      thorpej  1478:        /*
                   1479:         * Reclaim items from the pool's caches.
                   1480:         */
1.102     chs      1481:        LIST_FOREACH(pc, &pp->pr_cachelist, pc_poollist)
                   1482:                pool_cache_reclaim(pc, &pq, &pcgl);
1.43      thorpej  1483:
1.21      thorpej  1484:        s = splclock();
                   1485:        curtime = mono_time;
                   1486:        splx(s);
                   1487:
1.88      chs      1488:        for (ph = LIST_FIRST(&pp->pr_emptypages); ph != NULL; ph = phnext) {
                   1489:                phnext = LIST_NEXT(ph, ph_pagelist);
1.3       pk       1490:
                   1491:                /* Check our minimum page claim */
                   1492:                if (pp->pr_npages <= pp->pr_minpages)
                   1493:                        break;
                   1494:
1.88      chs      1495:                KASSERT(ph->ph_nmissing == 0);
                   1496:                timersub(&curtime, &ph->ph_time, &diff);
                   1497:                if (diff.tv_sec < pool_inactive_time)
                   1498:                        continue;
1.21      thorpej  1499:
1.88      chs      1500:                /*
                   1501:                 * If freeing this page would put us below
                   1502:                 * the low water mark, stop now.
                   1503:                 */
                   1504:                if ((pp->pr_nitems - pp->pr_itemsperpage) <
                   1505:                    pp->pr_minitems)
                   1506:                        break;
1.21      thorpej  1507:
1.88      chs      1508:                pr_rmpage(pp, ph, &pq);
1.3       pk       1509:        }
                   1510:
1.25      thorpej  1511:        pr_leave(pp);
1.21      thorpej  1512:        simple_unlock(&pp->pr_slock);
1.102     chs      1513:        if (LIST_EMPTY(&pq) && LIST_EMPTY(&pcgl))
                   1514:                return 0;
1.66      thorpej  1515:
1.101     thorpej  1516:        pr_pagelist_free(pp, &pq);
1.102     chs      1517:        pcg_grouplist_free(&pcgl);
1.66      thorpej  1518:        return (1);
1.3       pk       1519: }
                   1520:
                   1521: /*
                   1522:  * Drain pools, one at a time.
1.21      thorpej  1523:  *
                   1524:  * Note, we must never be called from an interrupt context.
1.3       pk       1525:  */
                   1526: void
1.42      thorpej  1527: pool_drain(void *arg)
1.3       pk       1528: {
                   1529:        struct pool *pp;
1.23      thorpej  1530:        int s;
1.3       pk       1531:
1.61      chs      1532:        pp = NULL;
1.49      thorpej  1533:        s = splvm();
1.23      thorpej  1534:        simple_lock(&pool_head_slock);
1.61      chs      1535:        if (drainpp == NULL) {
1.102     chs      1536:                drainpp = LIST_FIRST(&pool_head);
1.61      chs      1537:        }
                   1538:        if (drainpp) {
                   1539:                pp = drainpp;
1.102     chs      1540:                drainpp = LIST_NEXT(pp, pr_poollist);
1.61      chs      1541:        }
                   1542:        simple_unlock(&pool_head_slock);
1.63      chs      1543:        pool_reclaim(pp);
1.61      chs      1544:        splx(s);
1.3       pk       1545: }
                   1546:
                   1547: /*
                   1548:  * Diagnostic helpers.
                   1549:  */
                   1550: void
1.42      thorpej  1551: pool_print(struct pool *pp, const char *modif)
1.21      thorpej  1552: {
                   1553:        int s;
                   1554:
1.49      thorpej  1555:        s = splvm();
1.25      thorpej  1556:        if (simple_lock_try(&pp->pr_slock) == 0) {
                   1557:                printf("pool %s is locked; try again later\n",
                   1558:                    pp->pr_wchan);
                   1559:                splx(s);
                   1560:                return;
                   1561:        }
                   1562:        pool_print1(pp, modif, printf);
1.21      thorpej  1563:        simple_unlock(&pp->pr_slock);
                   1564:        splx(s);
                   1565: }
                   1566:
1.25      thorpej  1567: void
1.108     yamt     1568: pool_printall(const char *modif, void (*pr)(const char *, ...))
                   1569: {
                   1570:        struct pool *pp;
                   1571:
                   1572:        if (simple_lock_try(&pool_head_slock) == 0) {
                   1573:                (*pr)("WARNING: pool_head_slock is locked\n");
                   1574:        } else {
                   1575:                simple_unlock(&pool_head_slock);
                   1576:        }
                   1577:
                   1578:        LIST_FOREACH(pp, &pool_head, pr_poollist) {
                   1579:                pool_printit(pp, modif, pr);
                   1580:        }
                   1581: }
                   1582:
                   1583: void
1.42      thorpej  1584: pool_printit(struct pool *pp, const char *modif, void (*pr)(const char *, ...))
1.25      thorpej  1585: {
                   1586:
                   1587:        if (pp == NULL) {
                   1588:                (*pr)("Must specify a pool to print.\n");
                   1589:                return;
                   1590:        }
                   1591:
                   1592:        /*
                   1593:         * Called from DDB; interrupts should be blocked, and all
                   1594:         * other processors should be paused.  We can skip locking
                   1595:         * the pool in this case.
                   1596:         *
                   1597:         * We do a simple_lock_try() just to print the lock
                   1598:         * status, however.
                   1599:         */
                   1600:
                   1601:        if (simple_lock_try(&pp->pr_slock) == 0)
                   1602:                (*pr)("WARNING: pool %s is locked\n", pp->pr_wchan);
                   1603:        else
1.107     yamt     1604:                simple_unlock(&pp->pr_slock);
1.25      thorpej  1605:
                   1606:        pool_print1(pp, modif, pr);
                   1607: }
                   1608:
1.21      thorpej  1609: static void
1.97      yamt     1610: pool_print_pagelist(struct pool *pp, struct pool_pagelist *pl,
                   1611:     void (*pr)(const char *, ...))
1.88      chs      1612: {
                   1613:        struct pool_item_header *ph;
                   1614: #ifdef DIAGNOSTIC
                   1615:        struct pool_item *pi;
                   1616: #endif
                   1617:
                   1618:        LIST_FOREACH(ph, pl, ph_pagelist) {
                   1619:                (*pr)("\t\tpage %p, nmissing %d, time %lu,%lu\n",
                   1620:                    ph->ph_page, ph->ph_nmissing,
                   1621:                    (u_long)ph->ph_time.tv_sec,
                   1622:                    (u_long)ph->ph_time.tv_usec);
                   1623: #ifdef DIAGNOSTIC
1.97      yamt     1624:                if (!(pp->pr_roflags & PR_NOTOUCH)) {
1.102     chs      1625:                        LIST_FOREACH(pi, &ph->ph_itemlist, pi_list) {
1.97      yamt     1626:                                if (pi->pi_magic != PI_MAGIC) {
                   1627:                                        (*pr)("\t\t\titem %p, magic 0x%x\n",
                   1628:                                            pi, pi->pi_magic);
                   1629:                                }
1.88      chs      1630:                        }
                   1631:                }
                   1632: #endif
                   1633:        }
                   1634: }
                   1635:
                   1636: static void
1.42      thorpej  1637: pool_print1(struct pool *pp, const char *modif, void (*pr)(const char *, ...))
1.3       pk       1638: {
1.25      thorpej  1639:        struct pool_item_header *ph;
1.44      thorpej  1640:        struct pool_cache *pc;
                   1641:        struct pool_cache_group *pcg;
                   1642:        int i, print_log = 0, print_pagelist = 0, print_cache = 0;
1.25      thorpej  1643:        char c;
                   1644:
                   1645:        while ((c = *modif++) != '\0') {
                   1646:                if (c == 'l')
                   1647:                        print_log = 1;
                   1648:                if (c == 'p')
                   1649:                        print_pagelist = 1;
1.44      thorpej  1650:                if (c == 'c')
                   1651:                        print_cache = 1;
1.25      thorpej  1652:        }
                   1653:
                   1654:        (*pr)("POOL %s: size %u, align %u, ioff %u, roflags 0x%08x\n",
                   1655:            pp->pr_wchan, pp->pr_size, pp->pr_align, pp->pr_itemoffset,
                   1656:            pp->pr_roflags);
1.66      thorpej  1657:        (*pr)("\talloc %p\n", pp->pr_alloc);
1.25      thorpej  1658:        (*pr)("\tminitems %u, minpages %u, maxpages %u, npages %u\n",
                   1659:            pp->pr_minitems, pp->pr_minpages, pp->pr_maxpages, pp->pr_npages);
                   1660:        (*pr)("\titemsperpage %u, nitems %u, nout %u, hardlimit %u\n",
                   1661:            pp->pr_itemsperpage, pp->pr_nitems, pp->pr_nout, pp->pr_hardlimit);
                   1662:
                   1663:        (*pr)("\n\tnget %lu, nfail %lu, nput %lu\n",
                   1664:            pp->pr_nget, pp->pr_nfail, pp->pr_nput);
                   1665:        (*pr)("\tnpagealloc %lu, npagefree %lu, hiwat %u, nidle %lu\n",
                   1666:            pp->pr_npagealloc, pp->pr_npagefree, pp->pr_hiwat, pp->pr_nidle);
                   1667:
                   1668:        if (print_pagelist == 0)
                   1669:                goto skip_pagelist;
                   1670:
1.88      chs      1671:        if ((ph = LIST_FIRST(&pp->pr_emptypages)) != NULL)
                   1672:                (*pr)("\n\tempty page list:\n");
1.97      yamt     1673:        pool_print_pagelist(pp, &pp->pr_emptypages, pr);
1.88      chs      1674:        if ((ph = LIST_FIRST(&pp->pr_fullpages)) != NULL)
                   1675:                (*pr)("\n\tfull page list:\n");
1.97      yamt     1676:        pool_print_pagelist(pp, &pp->pr_fullpages, pr);
1.88      chs      1677:        if ((ph = LIST_FIRST(&pp->pr_partpages)) != NULL)
                   1678:                (*pr)("\n\tpartial-page list:\n");
1.97      yamt     1679:        pool_print_pagelist(pp, &pp->pr_partpages, pr);
1.88      chs      1680:
1.25      thorpej  1681:        if (pp->pr_curpage == NULL)
                   1682:                (*pr)("\tno current page\n");
                   1683:        else
                   1684:                (*pr)("\tcurpage %p\n", pp->pr_curpage->ph_page);
                   1685:
                   1686:  skip_pagelist:
                   1687:        if (print_log == 0)
                   1688:                goto skip_log;
                   1689:
                   1690:        (*pr)("\n");
                   1691:        if ((pp->pr_roflags & PR_LOGGING) == 0)
                   1692:                (*pr)("\tno log\n");
                   1693:        else
                   1694:                pr_printlog(pp, NULL, pr);
1.3       pk       1695:
1.25      thorpej  1696:  skip_log:
1.44      thorpej  1697:        if (print_cache == 0)
                   1698:                goto skip_cache;
                   1699:
1.102     chs      1700: #define PR_GROUPLIST(pcg)                                              \
                   1701:        (*pr)("\t\tgroup %p: avail %d\n", pcg, pcg->pcg_avail);         \
                   1702:        for (i = 0; i < PCG_NOBJECTS; i++) {                            \
                   1703:                if (pcg->pcg_objects[i].pcgo_pa !=                      \
                   1704:                    POOL_PADDR_INVALID) {                               \
                   1705:                        (*pr)("\t\t\t%p, 0x%llx\n",                     \
                   1706:                            pcg->pcg_objects[i].pcgo_va,                \
                   1707:                            (unsigned long long)                        \
                   1708:                            pcg->pcg_objects[i].pcgo_pa);               \
                   1709:                } else {                                                \
                   1710:                        (*pr)("\t\t\t%p\n",                             \
                   1711:                            pcg->pcg_objects[i].pcgo_va);               \
                   1712:                }                                                       \
                   1713:        }
                   1714:
                   1715:        LIST_FOREACH(pc, &pp->pr_cachelist, pc_poollist) {
1.103     chs      1716:                (*pr)("\tcache %p\n", pc);
1.48      thorpej  1717:                (*pr)("\t    hits %lu misses %lu ngroups %lu nitems %lu\n",
                   1718:                    pc->pc_hits, pc->pc_misses, pc->pc_ngroups, pc->pc_nitems);
1.102     chs      1719:                (*pr)("\t    full groups:\n");
1.103     chs      1720:                LIST_FOREACH(pcg, &pc->pc_fullgroups, pcg_list) {
1.102     chs      1721:                        PR_GROUPLIST(pcg);
1.103     chs      1722:                }
1.102     chs      1723:                (*pr)("\t    partial groups:\n");
1.103     chs      1724:                LIST_FOREACH(pcg, &pc->pc_partgroups, pcg_list) {
1.102     chs      1725:                        PR_GROUPLIST(pcg);
1.103     chs      1726:                }
1.102     chs      1727:                (*pr)("\t    empty groups:\n");
1.103     chs      1728:                LIST_FOREACH(pcg, &pc->pc_emptygroups, pcg_list) {
1.102     chs      1729:                        PR_GROUPLIST(pcg);
1.103     chs      1730:                }
1.44      thorpej  1731:        }
1.102     chs      1732: #undef PR_GROUPLIST
1.44      thorpej  1733:
                   1734:  skip_cache:
1.88      chs      1735:        pr_enter_check(pp, pr);
                   1736: }
                   1737:
                   1738: static int
                   1739: pool_chk_page(struct pool *pp, const char *label, struct pool_item_header *ph)
                   1740: {
                   1741:        struct pool_item *pi;
                   1742:        caddr_t page;
                   1743:        int n;
                   1744:
                   1745:        page = (caddr_t)((u_long)ph & pp->pr_alloc->pa_pagemask);
                   1746:        if (page != ph->ph_page &&
                   1747:            (pp->pr_roflags & PR_PHINPAGE) != 0) {
                   1748:                if (label != NULL)
                   1749:                        printf("%s: ", label);
                   1750:                printf("pool(%p:%s): page inconsistency: page %p;"
                   1751:                       " at page head addr %p (p %p)\n", pp,
                   1752:                        pp->pr_wchan, ph->ph_page,
                   1753:                        ph, page);
                   1754:                return 1;
                   1755:        }
1.3       pk       1756:
1.97      yamt     1757:        if ((pp->pr_roflags & PR_NOTOUCH) != 0)
                   1758:                return 0;
                   1759:
1.102     chs      1760:        for (pi = LIST_FIRST(&ph->ph_itemlist), n = 0;
1.88      chs      1761:             pi != NULL;
1.102     chs      1762:             pi = LIST_NEXT(pi,pi_list), n++) {
1.88      chs      1763:
                   1764: #ifdef DIAGNOSTIC
                   1765:                if (pi->pi_magic != PI_MAGIC) {
                   1766:                        if (label != NULL)
                   1767:                                printf("%s: ", label);
                   1768:                        printf("pool(%s): free list modified: magic=%x;"
                   1769:                               " page %p; item ordinal %d;"
                   1770:                               " addr %p (p %p)\n",
                   1771:                                pp->pr_wchan, pi->pi_magic, ph->ph_page,
                   1772:                                n, pi, page);
                   1773:                        panic("pool");
                   1774:                }
                   1775: #endif
                   1776:                page =
                   1777:                    (caddr_t)((u_long)pi & pp->pr_alloc->pa_pagemask);
                   1778:                if (page == ph->ph_page)
                   1779:                        continue;
                   1780:
                   1781:                if (label != NULL)
                   1782:                        printf("%s: ", label);
                   1783:                printf("pool(%p:%s): page inconsistency: page %p;"
                   1784:                       " item ordinal %d; addr %p (p %p)\n", pp,
                   1785:                        pp->pr_wchan, ph->ph_page,
                   1786:                        n, pi, page);
                   1787:                return 1;
                   1788:        }
                   1789:        return 0;
1.3       pk       1790: }
                   1791:
1.88      chs      1792:
1.3       pk       1793: int
1.42      thorpej  1794: pool_chk(struct pool *pp, const char *label)
1.3       pk       1795: {
                   1796:        struct pool_item_header *ph;
                   1797:        int r = 0;
                   1798:
1.21      thorpej  1799:        simple_lock(&pp->pr_slock);
1.88      chs      1800:        LIST_FOREACH(ph, &pp->pr_emptypages, ph_pagelist) {
                   1801:                r = pool_chk_page(pp, label, ph);
                   1802:                if (r) {
                   1803:                        goto out;
                   1804:                }
                   1805:        }
                   1806:        LIST_FOREACH(ph, &pp->pr_fullpages, ph_pagelist) {
                   1807:                r = pool_chk_page(pp, label, ph);
                   1808:                if (r) {
1.3       pk       1809:                        goto out;
                   1810:                }
1.88      chs      1811:        }
                   1812:        LIST_FOREACH(ph, &pp->pr_partpages, ph_pagelist) {
                   1813:                r = pool_chk_page(pp, label, ph);
                   1814:                if (r) {
1.3       pk       1815:                        goto out;
                   1816:                }
                   1817:        }
1.88      chs      1818:
1.3       pk       1819: out:
1.21      thorpej  1820:        simple_unlock(&pp->pr_slock);
1.3       pk       1821:        return (r);
1.43      thorpej  1822: }
                   1823:
                   1824: /*
                   1825:  * pool_cache_init:
                   1826:  *
                   1827:  *     Initialize a pool cache.
                   1828:  *
                   1829:  *     NOTE: If the pool must be protected from interrupts, we expect
                   1830:  *     to be called at the appropriate interrupt priority level.
                   1831:  */
                   1832: void
                   1833: pool_cache_init(struct pool_cache *pc, struct pool *pp,
                   1834:     int (*ctor)(void *, void *, int),
                   1835:     void (*dtor)(void *, void *),
                   1836:     void *arg)
                   1837: {
                   1838:
1.102     chs      1839:        LIST_INIT(&pc->pc_emptygroups);
                   1840:        LIST_INIT(&pc->pc_fullgroups);
                   1841:        LIST_INIT(&pc->pc_partgroups);
1.43      thorpej  1842:        simple_lock_init(&pc->pc_slock);
                   1843:
                   1844:        pc->pc_pool = pp;
                   1845:
                   1846:        pc->pc_ctor = ctor;
                   1847:        pc->pc_dtor = dtor;
                   1848:        pc->pc_arg  = arg;
                   1849:
1.48      thorpej  1850:        pc->pc_hits   = 0;
                   1851:        pc->pc_misses = 0;
                   1852:
                   1853:        pc->pc_ngroups = 0;
                   1854:
                   1855:        pc->pc_nitems = 0;
                   1856:
1.43      thorpej  1857:        simple_lock(&pp->pr_slock);
1.102     chs      1858:        LIST_INSERT_HEAD(&pp->pr_cachelist, pc, pc_poollist);
1.43      thorpej  1859:        simple_unlock(&pp->pr_slock);
                   1860: }
                   1861:
                   1862: /*
                   1863:  * pool_cache_destroy:
                   1864:  *
                   1865:  *     Destroy a pool cache.
                   1866:  */
                   1867: void
                   1868: pool_cache_destroy(struct pool_cache *pc)
                   1869: {
                   1870:        struct pool *pp = pc->pc_pool;
                   1871:
                   1872:        /* First, invalidate the entire cache. */
                   1873:        pool_cache_invalidate(pc);
                   1874:
                   1875:        /* ...and remove it from the pool's cache list. */
                   1876:        simple_lock(&pp->pr_slock);
1.102     chs      1877:        LIST_REMOVE(pc, pc_poollist);
1.43      thorpej  1878:        simple_unlock(&pp->pr_slock);
                   1879: }
                   1880:
1.110     perry    1881: static inline void *
1.87      thorpej  1882: pcg_get(struct pool_cache_group *pcg, paddr_t *pap)
1.43      thorpej  1883: {
                   1884:        void *object;
                   1885:        u_int idx;
                   1886:
                   1887:        KASSERT(pcg->pcg_avail <= PCG_NOBJECTS);
1.45      thorpej  1888:        KASSERT(pcg->pcg_avail != 0);
1.43      thorpej  1889:        idx = --pcg->pcg_avail;
                   1890:
1.87      thorpej  1891:        KASSERT(pcg->pcg_objects[idx].pcgo_va != NULL);
                   1892:        object = pcg->pcg_objects[idx].pcgo_va;
                   1893:        if (pap != NULL)
                   1894:                *pap = pcg->pcg_objects[idx].pcgo_pa;
                   1895:        pcg->pcg_objects[idx].pcgo_va = NULL;
1.43      thorpej  1896:
                   1897:        return (object);
                   1898: }
                   1899:
1.110     perry    1900: static inline void
1.87      thorpej  1901: pcg_put(struct pool_cache_group *pcg, void *object, paddr_t pa)
1.43      thorpej  1902: {
                   1903:        u_int idx;
                   1904:
                   1905:        KASSERT(pcg->pcg_avail < PCG_NOBJECTS);
                   1906:        idx = pcg->pcg_avail++;
                   1907:
1.87      thorpej  1908:        KASSERT(pcg->pcg_objects[idx].pcgo_va == NULL);
                   1909:        pcg->pcg_objects[idx].pcgo_va = object;
                   1910:        pcg->pcg_objects[idx].pcgo_pa = pa;
1.43      thorpej  1911: }
                   1912:
1.102     chs      1913: static void
                   1914: pcg_grouplist_free(struct pool_cache_grouplist *pcgl)
                   1915: {
                   1916:        struct pool_cache_group *pcg;
                   1917:        int s;
                   1918:
                   1919:        s = splvm();
                   1920:        while ((pcg = LIST_FIRST(pcgl)) != NULL) {
                   1921:                LIST_REMOVE(pcg, pcg_list);
                   1922:                pool_put(&pcgpool, pcg);
                   1923:        }
                   1924:        splx(s);
                   1925: }
                   1926:
1.43      thorpej  1927: /*
1.87      thorpej  1928:  * pool_cache_get{,_paddr}:
1.43      thorpej  1929:  *
1.87      thorpej  1930:  *     Get an object from a pool cache (optionally returning
                   1931:  *     the physical address of the object).
1.43      thorpej  1932:  */
                   1933: void *
1.87      thorpej  1934: pool_cache_get_paddr(struct pool_cache *pc, int flags, paddr_t *pap)
1.43      thorpej  1935: {
                   1936:        struct pool_cache_group *pcg;
                   1937:        void *object;
1.58      thorpej  1938:
                   1939: #ifdef LOCKDEBUG
                   1940:        if (flags & PR_WAITOK)
                   1941:                simple_lock_only_held(NULL, "pool_cache_get(PR_WAITOK)");
                   1942: #endif
1.43      thorpej  1943:
                   1944:        simple_lock(&pc->pc_slock);
                   1945:
1.102     chs      1946:        pcg = LIST_FIRST(&pc->pc_partgroups);
                   1947:        if (pcg == NULL) {
                   1948:                pcg = LIST_FIRST(&pc->pc_fullgroups);
                   1949:                if (pcg != NULL) {
                   1950:                        LIST_REMOVE(pcg, pcg_list);
                   1951:                        LIST_INSERT_HEAD(&pc->pc_partgroups, pcg, pcg_list);
1.43      thorpej  1952:                }
1.102     chs      1953:        }
                   1954:        if (pcg == NULL) {
1.43      thorpej  1955:
                   1956:                /*
                   1957:                 * No groups with any available objects.  Allocate
                   1958:                 * a new object, construct it, and return it to
                   1959:                 * the caller.  We will allocate a group, if necessary,
                   1960:                 * when the object is freed back to the cache.
                   1961:                 */
1.48      thorpej  1962:                pc->pc_misses++;
1.43      thorpej  1963:                simple_unlock(&pc->pc_slock);
                   1964:                object = pool_get(pc->pc_pool, flags);
                   1965:                if (object != NULL && pc->pc_ctor != NULL) {
                   1966:                        if ((*pc->pc_ctor)(pc->pc_arg, object, flags) != 0) {
                   1967:                                pool_put(pc->pc_pool, object);
                   1968:                                return (NULL);
                   1969:                        }
                   1970:                }
1.87      thorpej  1971:                if (object != NULL && pap != NULL) {
                   1972: #ifdef POOL_VTOPHYS
                   1973:                        *pap = POOL_VTOPHYS(object);
                   1974: #else
                   1975:                        *pap = POOL_PADDR_INVALID;
                   1976: #endif
                   1977:                }
1.43      thorpej  1978:                return (object);
                   1979:        }
                   1980:
1.48      thorpej  1981:        pc->pc_hits++;
                   1982:        pc->pc_nitems--;
1.87      thorpej  1983:        object = pcg_get(pcg, pap);
1.43      thorpej  1984:
1.102     chs      1985:        if (pcg->pcg_avail == 0) {
                   1986:                LIST_REMOVE(pcg, pcg_list);
                   1987:                LIST_INSERT_HEAD(&pc->pc_emptygroups, pcg, pcg_list);
                   1988:        }
1.43      thorpej  1989:        simple_unlock(&pc->pc_slock);
                   1990:
                   1991:        return (object);
                   1992: }
                   1993:
                   1994: /*
1.87      thorpej  1995:  * pool_cache_put{,_paddr}:
1.43      thorpej  1996:  *
1.87      thorpej  1997:  *     Put an object back to the pool cache (optionally caching the
                   1998:  *     physical address of the object).
1.43      thorpej  1999:  */
                   2000: void
1.87      thorpej  2001: pool_cache_put_paddr(struct pool_cache *pc, void *object, paddr_t pa)
1.43      thorpej  2002: {
                   2003:        struct pool_cache_group *pcg;
1.60      thorpej  2004:        int s;
1.43      thorpej  2005:
1.109     christos 2006:        if (__predict_false((pc->pc_pool->pr_flags & PR_WANTED) != 0)) {
                   2007:                goto destruct;
                   2008:        }
                   2009:
1.43      thorpej  2010:        simple_lock(&pc->pc_slock);
                   2011:
1.102     chs      2012:        pcg = LIST_FIRST(&pc->pc_partgroups);
                   2013:        if (pcg == NULL) {
                   2014:                pcg = LIST_FIRST(&pc->pc_emptygroups);
                   2015:                if (pcg != NULL) {
                   2016:                        LIST_REMOVE(pcg, pcg_list);
                   2017:                        LIST_INSERT_HEAD(&pc->pc_partgroups, pcg, pcg_list);
1.43      thorpej  2018:                }
1.102     chs      2019:        }
                   2020:        if (pcg == NULL) {
1.43      thorpej  2021:
                   2022:                /*
                   2023:                 * No empty groups to free the object to.  Attempt to
1.47      thorpej  2024:                 * allocate one.
1.43      thorpej  2025:                 */
1.47      thorpej  2026:                simple_unlock(&pc->pc_slock);
1.60      thorpej  2027:                s = splvm();
1.43      thorpej  2028:                pcg = pool_get(&pcgpool, PR_NOWAIT);
1.60      thorpej  2029:                splx(s);
1.102     chs      2030:                if (pcg == NULL) {
1.109     christos 2031: destruct:
1.102     chs      2032:
                   2033:                        /*
                   2034:                         * Unable to allocate a cache group; destruct the object
                   2035:                         * and free it back to the pool.
                   2036:                         */
                   2037:                        pool_cache_destruct_object(pc, object);
                   2038:                        return;
1.43      thorpej  2039:                }
1.102     chs      2040:                memset(pcg, 0, sizeof(*pcg));
                   2041:                simple_lock(&pc->pc_slock);
                   2042:                pc->pc_ngroups++;
                   2043:                LIST_INSERT_HEAD(&pc->pc_partgroups, pcg, pcg_list);
1.43      thorpej  2044:        }
                   2045:
1.48      thorpej  2046:        pc->pc_nitems++;
1.87      thorpej  2047:        pcg_put(pcg, object, pa);
1.43      thorpej  2048:
1.102     chs      2049:        if (pcg->pcg_avail == PCG_NOBJECTS) {
                   2050:                LIST_REMOVE(pcg, pcg_list);
                   2051:                LIST_INSERT_HEAD(&pc->pc_fullgroups, pcg, pcg_list);
                   2052:        }
1.43      thorpej  2053:        simple_unlock(&pc->pc_slock);
1.51      thorpej  2054: }
                   2055:
                   2056: /*
                   2057:  * pool_cache_destruct_object:
                   2058:  *
                   2059:  *     Force destruction of an object and its release back into
                   2060:  *     the pool.
                   2061:  */
                   2062: void
                   2063: pool_cache_destruct_object(struct pool_cache *pc, void *object)
                   2064: {
                   2065:
                   2066:        if (pc->pc_dtor != NULL)
                   2067:                (*pc->pc_dtor)(pc->pc_arg, object);
                   2068:        pool_put(pc->pc_pool, object);
1.43      thorpej  2069: }
                   2070:
1.102     chs      2071: static void
1.106     christos 2072: pool_do_cache_invalidate_grouplist(struct pool_cache_grouplist *pcgsl,
1.105     christos 2073:     struct pool_cache *pc, struct pool_pagelist *pq,
1.106     christos 2074:     struct pool_cache_grouplist *pcgdl)
1.102     chs      2075: {
1.106     christos 2076:        struct pool_cache_group *pcg, *npcg;
1.102     chs      2077:        void *object;
                   2078:
1.106     christos 2079:        for (pcg = LIST_FIRST(pcgsl); pcg != NULL; pcg = npcg) {
1.102     chs      2080:                npcg = LIST_NEXT(pcg, pcg_list);
                   2081:                while (pcg->pcg_avail != 0) {
                   2082:                        pc->pc_nitems--;
                   2083:                        object = pcg_get(pcg, NULL);
                   2084:                        if (pc->pc_dtor != NULL)
                   2085:                                (*pc->pc_dtor)(pc->pc_arg, object);
                   2086:                        pool_do_put(pc->pc_pool, object, pq);
                   2087:                }
1.103     chs      2088:                pc->pc_ngroups--;
1.102     chs      2089:                LIST_REMOVE(pcg, pcg_list);
1.106     christos 2090:                LIST_INSERT_HEAD(pcgdl, pcg, pcg_list);
1.102     chs      2091:        }
1.105     christos 2092: }
                   2093:
                   2094: static void
                   2095: pool_do_cache_invalidate(struct pool_cache *pc, struct pool_pagelist *pq,
                   2096:     struct pool_cache_grouplist *pcgl)
                   2097: {
                   2098:
                   2099:        LOCK_ASSERT(simple_lock_held(&pc->pc_slock));
                   2100:        LOCK_ASSERT(simple_lock_held(&pc->pc_pool->pr_slock));
                   2101:
1.106     christos 2102:        pool_do_cache_invalidate_grouplist(&pc->pc_fullgroups, pc, pq, pcgl);
                   2103:        pool_do_cache_invalidate_grouplist(&pc->pc_partgroups, pc, pq, pcgl);
1.103     chs      2104:
                   2105:        KASSERT(LIST_EMPTY(&pc->pc_partgroups));
                   2106:        KASSERT(LIST_EMPTY(&pc->pc_fullgroups));
                   2107:        KASSERT(pc->pc_nitems == 0);
1.102     chs      2108: }
                   2109:
1.43      thorpej  2110: /*
1.101     thorpej  2111:  * pool_cache_invalidate:
1.43      thorpej  2112:  *
1.101     thorpej  2113:  *     Invalidate a pool cache (destruct and release all of the
                   2114:  *     cached objects).
1.43      thorpej  2115:  */
1.101     thorpej  2116: void
                   2117: pool_cache_invalidate(struct pool_cache *pc)
1.43      thorpej  2118: {
1.101     thorpej  2119:        struct pool_pagelist pq;
1.102     chs      2120:        struct pool_cache_grouplist pcgl;
1.101     thorpej  2121:
                   2122:        LIST_INIT(&pq);
1.102     chs      2123:        LIST_INIT(&pcgl);
1.101     thorpej  2124:
                   2125:        simple_lock(&pc->pc_slock);
                   2126:        simple_lock(&pc->pc_pool->pr_slock);
1.43      thorpej  2127:
1.102     chs      2128:        pool_do_cache_invalidate(pc, &pq, &pcgl);
1.43      thorpej  2129:
1.101     thorpej  2130:        simple_unlock(&pc->pc_pool->pr_slock);
                   2131:        simple_unlock(&pc->pc_slock);
1.43      thorpej  2132:
1.102     chs      2133:        pr_pagelist_free(pc->pc_pool, &pq);
                   2134:        pcg_grouplist_free(&pcgl);
1.43      thorpej  2135: }
                   2136:
                   2137: /*
                   2138:  * pool_cache_reclaim:
                   2139:  *
                   2140:  *     Reclaim a pool cache for pool_reclaim().
                   2141:  */
                   2142: static void
1.102     chs      2143: pool_cache_reclaim(struct pool_cache *pc, struct pool_pagelist *pq,
                   2144:     struct pool_cache_grouplist *pcgl)
1.43      thorpej  2145: {
1.101     thorpej  2146:
                   2147:        /*
                   2148:         * We're locking in the wrong order (normally pool_cache -> pool,
                   2149:         * but the pool is already locked when we get here), so we have
                   2150:         * to use trylock.  If we can't lock the pool_cache, it's not really
                   2151:         * a big deal here.
                   2152:         */
                   2153:        if (simple_lock_try(&pc->pc_slock) == 0)
                   2154:                return;
                   2155:
1.102     chs      2156:        pool_do_cache_invalidate(pc, pq, pcgl);
1.43      thorpej  2157:
                   2158:        simple_unlock(&pc->pc_slock);
1.3       pk       2159: }
1.66      thorpej  2160:
                   2161: /*
                   2162:  * Pool backend allocators.
                   2163:  *
                   2164:  * Each pool has a backend allocator that handles allocation, deallocation,
                   2165:  * and any additional draining that might be needed.
                   2166:  *
                   2167:  * We provide two standard allocators:
                   2168:  *
                   2169:  *     pool_allocator_kmem - the default when no allocator is specified
                   2170:  *
                   2171:  *     pool_allocator_nointr - used for pools that will not be accessed
                   2172:  *     in interrupt context.
                   2173:  */
                   2174: void   *pool_page_alloc(struct pool *, int);
                   2175: void   pool_page_free(struct pool *, void *);
                   2176:
1.112   ! bjh21    2177: #ifdef POOL_SUBPAGE
        !          2178: struct pool_allocator pool_allocator_kmem_fullpage = {
        !          2179:        pool_page_alloc, pool_page_free, 0,
        !          2180: };
        !          2181: #else
1.66      thorpej  2182: struct pool_allocator pool_allocator_kmem = {
                   2183:        pool_page_alloc, pool_page_free, 0,
                   2184: };
1.112   ! bjh21    2185: #endif
1.66      thorpej  2186:
                   2187: void   *pool_page_alloc_nointr(struct pool *, int);
                   2188: void   pool_page_free_nointr(struct pool *, void *);
                   2189:
1.112   ! bjh21    2190: #ifdef POOL_SUBPAGE
        !          2191: struct pool_allocator pool_allocator_nointr_fullpage = {
        !          2192:        pool_page_alloc_nointr, pool_page_free_nointr, 0,
        !          2193: };
        !          2194: #else
1.66      thorpej  2195: struct pool_allocator pool_allocator_nointr = {
                   2196:        pool_page_alloc_nointr, pool_page_free_nointr, 0,
                   2197: };
1.112   ! bjh21    2198: #endif
1.66      thorpej  2199:
                   2200: #ifdef POOL_SUBPAGE
                   2201: void   *pool_subpage_alloc(struct pool *, int);
                   2202: void   pool_subpage_free(struct pool *, void *);
                   2203:
1.112   ! bjh21    2204: struct pool_allocator pool_allocator_kmem = {
        !          2205:        pool_subpage_alloc, pool_subpage_free, POOL_SUBPAGE,
        !          2206: };
        !          2207:
        !          2208: void   *pool_subpage_alloc_nointr(struct pool *, int);
        !          2209: void   pool_subpage_free_nointr(struct pool *, void *);
        !          2210:
        !          2211: struct pool_allocator pool_allocator_nointr = {
        !          2212:        pool_subpage_alloc, pool_subpage_free, POOL_SUBPAGE,
1.66      thorpej  2213: };
                   2214: #endif /* POOL_SUBPAGE */
                   2215:
                   2216: /*
                   2217:  * We have at least three different resources for the same allocation and
                   2218:  * each resource can be depleted.  First, we have the ready elements in the
                   2219:  * pool.  Then we have the resource (typically a vm_map) for this allocator.
                   2220:  * Finally, we have physical memory.  Waiting for any of these can be
                   2221:  * unnecessary when any other is freed, but the kernel doesn't support
                   2222:  * sleeping on multiple wait channels, so we have to employ another strategy.
                   2223:  *
                   2224:  * The caller sleeps on the pool (so that it can be awakened when an item
                   2225:  * is returned to the pool), but we set PA_WANT on the allocator.  When a
                   2226:  * page is returned to the allocator and PA_WANT is set, pool_allocator_free
                   2227:  * will wake up all sleeping pools belonging to this allocator.
                   2228:  *
                   2229:  * XXX Thundering herd.
                   2230:  */
                   2231: void *
                   2232: pool_allocator_alloc(struct pool *org, int flags)
                   2233: {
                   2234:        struct pool_allocator *pa = org->pr_alloc;
                   2235:        struct pool *pp, *start;
                   2236:        int s, freed;
                   2237:        void *res;
                   2238:
1.91      yamt     2239:        LOCK_ASSERT(!simple_lock_held(&org->pr_slock));
                   2240:
1.66      thorpej  2241:        do {
                   2242:                if ((res = (*pa->pa_alloc)(org, flags)) != NULL)
                   2243:                        return (res);
1.68      thorpej  2244:                if ((flags & PR_WAITOK) == 0) {
                   2245:                        /*
                   2246:                         * We only run the drain hookhere if PR_NOWAIT.
                   2247:                         * In other cases, the hook will be run in
                   2248:                         * pool_reclaim().
                   2249:                         */
                   2250:                        if (org->pr_drain_hook != NULL) {
                   2251:                                (*org->pr_drain_hook)(org->pr_drain_hook_arg,
                   2252:                                    flags);
                   2253:                                if ((res = (*pa->pa_alloc)(org, flags)) != NULL)
                   2254:                                        return (res);
                   2255:                        }
1.66      thorpej  2256:                        break;
1.68      thorpej  2257:                }
1.66      thorpej  2258:
                   2259:                /*
1.109     christos 2260:                 * Drain all pools, that use this allocator.
                   2261:                 * We do this to reclaim VA space.
1.66      thorpej  2262:                 * pa_alloc is responsible for waiting for
                   2263:                 * physical memory.
                   2264:                 *
                   2265:                 * XXX We risk looping forever if start if someone
                   2266:                 * calls pool_destroy on "start".  But there is no
                   2267:                 * other way to have potentially sleeping pool_reclaim,
                   2268:                 * non-sleeping locks on pool_allocator, and some
                   2269:                 * stirring of drained pools in the allocator.
1.68      thorpej  2270:                 *
                   2271:                 * XXX Maybe we should use pool_head_slock for locking
                   2272:                 * the allocators?
1.66      thorpej  2273:                 */
                   2274:                freed = 0;
                   2275:
                   2276:                s = splvm();
                   2277:                simple_lock(&pa->pa_slock);
                   2278:                pp = start = TAILQ_FIRST(&pa->pa_list);
                   2279:                do {
                   2280:                        TAILQ_REMOVE(&pa->pa_list, pp, pr_alloc_list);
                   2281:                        TAILQ_INSERT_TAIL(&pa->pa_list, pp, pr_alloc_list);
1.73      thorpej  2282:                        simple_unlock(&pa->pa_slock);
1.66      thorpej  2283:                        freed = pool_reclaim(pp);
1.73      thorpej  2284:                        simple_lock(&pa->pa_slock);
1.66      thorpej  2285:                } while ((pp = TAILQ_FIRST(&pa->pa_list)) != start &&
                   2286:                         freed == 0);
                   2287:
                   2288:                if (freed == 0) {
                   2289:                        /*
                   2290:                         * We set PA_WANT here, the caller will most likely
                   2291:                         * sleep waiting for pages (if not, this won't hurt
                   2292:                         * that much), and there is no way to set this in
                   2293:                         * the caller without violating locking order.
                   2294:                         */
                   2295:                        pa->pa_flags |= PA_WANT;
                   2296:                }
                   2297:                simple_unlock(&pa->pa_slock);
                   2298:                splx(s);
                   2299:        } while (freed);
                   2300:        return (NULL);
                   2301: }
                   2302:
                   2303: void
                   2304: pool_allocator_free(struct pool *pp, void *v)
                   2305: {
                   2306:        struct pool_allocator *pa = pp->pr_alloc;
                   2307:        int s;
                   2308:
1.91      yamt     2309:        LOCK_ASSERT(!simple_lock_held(&pp->pr_slock));
                   2310:
1.66      thorpej  2311:        (*pa->pa_free)(pp, v);
                   2312:
                   2313:        s = splvm();
                   2314:        simple_lock(&pa->pa_slock);
                   2315:        if ((pa->pa_flags & PA_WANT) == 0) {
                   2316:                simple_unlock(&pa->pa_slock);
                   2317:                splx(s);
                   2318:                return;
                   2319:        }
                   2320:
                   2321:        TAILQ_FOREACH(pp, &pa->pa_list, pr_alloc_list) {
                   2322:                simple_lock(&pp->pr_slock);
                   2323:                if ((pp->pr_flags & PR_WANTED) != 0) {
                   2324:                        pp->pr_flags &= ~PR_WANTED;
                   2325:                        wakeup(pp);
                   2326:                }
1.69      thorpej  2327:                simple_unlock(&pp->pr_slock);
1.66      thorpej  2328:        }
                   2329:        pa->pa_flags &= ~PA_WANT;
                   2330:        simple_unlock(&pa->pa_slock);
                   2331:        splx(s);
                   2332: }
                   2333:
                   2334: void *
                   2335: pool_page_alloc(struct pool *pp, int flags)
                   2336: {
                   2337:        boolean_t waitok = (flags & PR_WAITOK) ? TRUE : FALSE;
                   2338:
1.100     yamt     2339:        return ((void *) uvm_km_alloc_poolpage_cache(kmem_map, waitok));
1.66      thorpej  2340: }
                   2341:
                   2342: void
                   2343: pool_page_free(struct pool *pp, void *v)
                   2344: {
                   2345:
1.98      yamt     2346:        uvm_km_free_poolpage_cache(kmem_map, (vaddr_t) v);
                   2347: }
                   2348:
                   2349: static void *
                   2350: pool_page_alloc_meta(struct pool *pp, int flags)
                   2351: {
                   2352:        boolean_t waitok = (flags & PR_WAITOK) ? TRUE : FALSE;
                   2353:
1.100     yamt     2354:        return ((void *) uvm_km_alloc_poolpage(kmem_map, waitok));
1.98      yamt     2355: }
                   2356:
                   2357: static void
                   2358: pool_page_free_meta(struct pool *pp, void *v)
                   2359: {
                   2360:
1.100     yamt     2361:        uvm_km_free_poolpage(kmem_map, (vaddr_t) v);
1.66      thorpej  2362: }
                   2363:
                   2364: #ifdef POOL_SUBPAGE
                   2365: /* Sub-page allocator, for machines with large hardware pages. */
                   2366: void *
                   2367: pool_subpage_alloc(struct pool *pp, int flags)
                   2368: {
1.93      dbj      2369:        void *v;
                   2370:        int s;
                   2371:        s = splvm();
                   2372:        v = pool_get(&psppool, flags);
                   2373:        splx(s);
                   2374:        return v;
1.66      thorpej  2375: }
                   2376:
                   2377: void
                   2378: pool_subpage_free(struct pool *pp, void *v)
                   2379: {
1.93      dbj      2380:        int s;
                   2381:        s = splvm();
1.66      thorpej  2382:        pool_put(&psppool, v);
1.93      dbj      2383:        splx(s);
1.66      thorpej  2384: }
                   2385:
                   2386: /* We don't provide a real nointr allocator.  Maybe later. */
                   2387: void *
1.112   ! bjh21    2388: pool_subpage_alloc_nointr(struct pool *pp, int flags)
1.66      thorpej  2389: {
                   2390:
                   2391:        return (pool_subpage_alloc(pp, flags));
                   2392: }
                   2393:
                   2394: void
1.112   ! bjh21    2395: pool_subpage_free_nointr(struct pool *pp, void *v)
1.66      thorpej  2396: {
                   2397:
                   2398:        pool_subpage_free(pp, v);
                   2399: }
1.112   ! bjh21    2400: #endif /* POOL_SUBPAGE */
1.66      thorpej  2401: void *
                   2402: pool_page_alloc_nointr(struct pool *pp, int flags)
                   2403: {
                   2404:        boolean_t waitok = (flags & PR_WAITOK) ? TRUE : FALSE;
                   2405:
1.100     yamt     2406:        return ((void *) uvm_km_alloc_poolpage_cache(kernel_map, waitok));
1.66      thorpej  2407: }
                   2408:
                   2409: void
                   2410: pool_page_free_nointr(struct pool *pp, void *v)
                   2411: {
                   2412:
1.98      yamt     2413:        uvm_km_free_poolpage_cache(kernel_map, (vaddr_t) v);
1.66      thorpej  2414: }

CVSweb <webmaster@jp.NetBSD.org>