[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 1.99.2.1

1.99.2.1! kent        1: /*     $NetBSD: subr_pool.c,v 1.100 2005/04/01 11:59:37 yamt 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>
1.99.2.1! kent       41: __KERNEL_RCSID(0, "$NetBSD: subr_pool.c,v 1.100 2005/04/01 11:59:37 yamt 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 */
1.5       thorpej    73: TAILQ_HEAD(,pool) pool_head = TAILQ_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 {
                    115:                        TAILQ_HEAD(, pool_item)
                    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 */
                    142:        TAILQ_ENTRY(pool_item)  pi_list;
                    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:
1.43      thorpej   177: static void    pool_cache_reclaim(struct pool_cache *);
1.3       pk        178:
1.42      thorpej   179: static int     pool_catchup(struct pool *);
1.55      thorpej   180: static void    pool_prime_page(struct pool *, caddr_t,
                    181:                    struct pool_item_header *);
1.88      chs       182: static void    pool_update_curpage(struct pool *);
1.66      thorpej   183:
                    184: void           *pool_allocator_alloc(struct pool *, int);
                    185: void           pool_allocator_free(struct pool *, void *);
1.3       pk        186:
1.97      yamt      187: static void pool_print_pagelist(struct pool *, struct pool_pagelist *,
1.88      chs       188:        void (*)(const char *, ...));
1.42      thorpej   189: static void pool_print1(struct pool *, const char *,
                    190:        void (*)(const char *, ...));
1.3       pk        191:
1.88      chs       192: static int pool_chk_page(struct pool *, const char *,
                    193:                         struct pool_item_header *);
                    194:
1.3       pk        195: /*
1.52      thorpej   196:  * Pool log entry. An array of these is allocated in pool_init().
1.3       pk        197:  */
                    198: struct pool_log {
                    199:        const char      *pl_file;
                    200:        long            pl_line;
                    201:        int             pl_action;
1.25      thorpej   202: #define        PRLOG_GET       1
                    203: #define        PRLOG_PUT       2
1.3       pk        204:        void            *pl_addr;
1.1       pk        205: };
                    206:
1.86      matt      207: #ifdef POOL_DIAGNOSTIC
1.3       pk        208: /* Number of entries in pool log buffers */
1.17      thorpej   209: #ifndef POOL_LOGSIZE
                    210: #define        POOL_LOGSIZE    10
                    211: #endif
                    212:
                    213: int pool_logsize = POOL_LOGSIZE;
1.1       pk        214:
1.42      thorpej   215: static __inline void
                    216: pr_log(struct pool *pp, void *v, int action, const char *file, long line)
1.3       pk        217: {
                    218:        int n = pp->pr_curlogentry;
                    219:        struct pool_log *pl;
                    220:
1.20      thorpej   221:        if ((pp->pr_roflags & PR_LOGGING) == 0)
1.3       pk        222:                return;
                    223:
                    224:        /*
                    225:         * Fill in the current entry. Wrap around and overwrite
                    226:         * the oldest entry if necessary.
                    227:         */
                    228:        pl = &pp->pr_log[n];
                    229:        pl->pl_file = file;
                    230:        pl->pl_line = line;
                    231:        pl->pl_action = action;
                    232:        pl->pl_addr = v;
                    233:        if (++n >= pp->pr_logsize)
                    234:                n = 0;
                    235:        pp->pr_curlogentry = n;
                    236: }
                    237:
                    238: static void
1.42      thorpej   239: pr_printlog(struct pool *pp, struct pool_item *pi,
                    240:     void (*pr)(const char *, ...))
1.3       pk        241: {
                    242:        int i = pp->pr_logsize;
                    243:        int n = pp->pr_curlogentry;
                    244:
1.20      thorpej   245:        if ((pp->pr_roflags & PR_LOGGING) == 0)
1.3       pk        246:                return;
                    247:
                    248:        /*
                    249:         * Print all entries in this pool's log.
                    250:         */
                    251:        while (i-- > 0) {
                    252:                struct pool_log *pl = &pp->pr_log[n];
                    253:                if (pl->pl_action != 0) {
1.25      thorpej   254:                        if (pi == NULL || pi == pl->pl_addr) {
                    255:                                (*pr)("\tlog entry %d:\n", i);
                    256:                                (*pr)("\t\taction = %s, addr = %p\n",
                    257:                                    pl->pl_action == PRLOG_GET ? "get" : "put",
                    258:                                    pl->pl_addr);
                    259:                                (*pr)("\t\tfile: %s at line %lu\n",
                    260:                                    pl->pl_file, pl->pl_line);
                    261:                        }
1.3       pk        262:                }
                    263:                if (++n >= pp->pr_logsize)
                    264:                        n = 0;
                    265:        }
                    266: }
1.25      thorpej   267:
1.42      thorpej   268: static __inline void
                    269: pr_enter(struct pool *pp, const char *file, long line)
1.25      thorpej   270: {
                    271:
1.34      thorpej   272:        if (__predict_false(pp->pr_entered_file != NULL)) {
1.25      thorpej   273:                printf("pool %s: reentrancy at file %s line %ld\n",
                    274:                    pp->pr_wchan, file, line);
                    275:                printf("         previous entry at file %s line %ld\n",
                    276:                    pp->pr_entered_file, pp->pr_entered_line);
                    277:                panic("pr_enter");
                    278:        }
                    279:
                    280:        pp->pr_entered_file = file;
                    281:        pp->pr_entered_line = line;
                    282: }
                    283:
1.42      thorpej   284: static __inline void
                    285: pr_leave(struct pool *pp)
1.25      thorpej   286: {
                    287:
1.34      thorpej   288:        if (__predict_false(pp->pr_entered_file == NULL)) {
1.25      thorpej   289:                printf("pool %s not entered?\n", pp->pr_wchan);
                    290:                panic("pr_leave");
                    291:        }
                    292:
                    293:        pp->pr_entered_file = NULL;
                    294:        pp->pr_entered_line = 0;
                    295: }
                    296:
1.42      thorpej   297: static __inline void
                    298: pr_enter_check(struct pool *pp, void (*pr)(const char *, ...))
1.25      thorpej   299: {
                    300:
                    301:        if (pp->pr_entered_file != NULL)
                    302:                (*pr)("\n\tcurrently entered from file %s line %ld\n",
                    303:                    pp->pr_entered_file, pp->pr_entered_line);
                    304: }
1.3       pk        305: #else
1.25      thorpej   306: #define        pr_log(pp, v, action, file, line)
                    307: #define        pr_printlog(pp, pi, pr)
                    308: #define        pr_enter(pp, file, line)
                    309: #define        pr_leave(pp)
                    310: #define        pr_enter_check(pp, pr)
1.59      thorpej   311: #endif /* POOL_DIAGNOSTIC */
1.3       pk        312:
1.88      chs       313: static __inline int
1.97      yamt      314: pr_item_notouch_index(const struct pool *pp, const struct pool_item_header *ph,
                    315:     const void *v)
                    316: {
                    317:        const char *cp = v;
                    318:        int idx;
                    319:
                    320:        KASSERT(pp->pr_roflags & PR_NOTOUCH);
                    321:        idx = (cp - ph->ph_page - ph->ph_off) / pp->pr_size;
                    322:        KASSERT(idx < pp->pr_itemsperpage);
                    323:        return idx;
                    324: }
                    325:
1.99      yamt      326: #define        PR_FREELIST_ALIGN(p) \
                    327:        roundup((uintptr_t)(p), sizeof(pool_item_freelist_t))
                    328: #define        PR_FREELIST(ph) ((pool_item_freelist_t *)PR_FREELIST_ALIGN((ph) + 1))
                    329: #define        PR_INDEX_USED   ((pool_item_freelist_t)-1)
                    330: #define        PR_INDEX_EOL    ((pool_item_freelist_t)-2)
1.97      yamt      331:
                    332: static __inline void
                    333: pr_item_notouch_put(const struct pool *pp, struct pool_item_header *ph,
                    334:     void *obj)
                    335: {
                    336:        int idx = pr_item_notouch_index(pp, ph, obj);
1.99      yamt      337:        pool_item_freelist_t *freelist = PR_FREELIST(ph);
1.97      yamt      338:
                    339:        KASSERT(freelist[idx] == PR_INDEX_USED);
                    340:        freelist[idx] = ph->ph_firstfree;
                    341:        ph->ph_firstfree = idx;
                    342: }
                    343:
                    344: static __inline void *
                    345: pr_item_notouch_get(const struct pool *pp, struct pool_item_header *ph)
                    346: {
                    347:        int idx = ph->ph_firstfree;
1.99      yamt      348:        pool_item_freelist_t *freelist = PR_FREELIST(ph);
1.97      yamt      349:
                    350:        KASSERT(freelist[idx] != PR_INDEX_USED);
                    351:        ph->ph_firstfree = freelist[idx];
                    352:        freelist[idx] = PR_INDEX_USED;
                    353:
                    354:        return ph->ph_page + ph->ph_off + idx * pp->pr_size;
                    355: }
                    356:
                    357: static __inline int
1.88      chs       358: phtree_compare(struct pool_item_header *a, struct pool_item_header *b)
                    359: {
                    360:        if (a->ph_page < b->ph_page)
                    361:                return (-1);
                    362:        else if (a->ph_page > b->ph_page)
                    363:                return (1);
                    364:        else
                    365:                return (0);
                    366: }
                    367:
                    368: SPLAY_PROTOTYPE(phtree, pool_item_header, ph_node, phtree_compare);
                    369: SPLAY_GENERATE(phtree, pool_item_header, ph_node, phtree_compare);
                    370:
1.3       pk        371: /*
                    372:  * Return the pool page header based on page address.
                    373:  */
1.42      thorpej   374: static __inline struct pool_item_header *
                    375: pr_find_pagehead(struct pool *pp, caddr_t page)
1.3       pk        376: {
1.88      chs       377:        struct pool_item_header *ph, tmp;
1.3       pk        378:
1.20      thorpej   379:        if ((pp->pr_roflags & PR_PHINPAGE) != 0)
1.3       pk        380:                return ((struct pool_item_header *)(page + pp->pr_phoffset));
                    381:
1.88      chs       382:        tmp.ph_page = page;
                    383:        ph = SPLAY_FIND(phtree, &pp->pr_phtree, &tmp);
                    384:        return ph;
1.3       pk        385: }
                    386:
                    387: /*
                    388:  * Remove a page from the pool.
                    389:  */
1.42      thorpej   390: static __inline void
1.61      chs       391: pr_rmpage(struct pool *pp, struct pool_item_header *ph,
                    392:      struct pool_pagelist *pq)
1.3       pk        393: {
1.61      chs       394:        int s;
1.3       pk        395:
1.91      yamt      396:        LOCK_ASSERT(!simple_lock_held(&pp->pr_slock) || pq != NULL);
                    397:
1.3       pk        398:        /*
1.7       thorpej   399:         * If the page was idle, decrement the idle page count.
1.3       pk        400:         */
1.6       thorpej   401:        if (ph->ph_nmissing == 0) {
                    402: #ifdef DIAGNOSTIC
                    403:                if (pp->pr_nidle == 0)
                    404:                        panic("pr_rmpage: nidle inconsistent");
1.20      thorpej   405:                if (pp->pr_nitems < pp->pr_itemsperpage)
                    406:                        panic("pr_rmpage: nitems inconsistent");
1.6       thorpej   407: #endif
                    408:                pp->pr_nidle--;
                    409:        }
1.7       thorpej   410:
1.20      thorpej   411:        pp->pr_nitems -= pp->pr_itemsperpage;
                    412:
1.7       thorpej   413:        /*
1.61      chs       414:         * Unlink a page from the pool and release it (or queue it for release).
1.7       thorpej   415:         */
1.88      chs       416:        LIST_REMOVE(ph, ph_pagelist);
1.91      yamt      417:        if ((pp->pr_roflags & PR_PHINPAGE) == 0)
                    418:                SPLAY_REMOVE(phtree, &pp->pr_phtree, ph);
1.61      chs       419:        if (pq) {
1.88      chs       420:                LIST_INSERT_HEAD(pq, ph, ph_pagelist);
1.61      chs       421:        } else {
1.66      thorpej   422:                pool_allocator_free(pp, ph->ph_page);
1.61      chs       423:                if ((pp->pr_roflags & PR_PHINPAGE) == 0) {
1.85      pk        424:                        s = splvm();
1.97      yamt      425:                        pool_put(pp->pr_phpool, ph);
1.61      chs       426:                        splx(s);
                    427:                }
                    428:        }
1.7       thorpej   429:        pp->pr_npages--;
                    430:        pp->pr_npagefree++;
1.6       thorpej   431:
1.88      chs       432:        pool_update_curpage(pp);
1.3       pk        433: }
                    434:
                    435: /*
1.94      simonb    436:  * Initialize all the pools listed in the "pools" link set.
                    437:  */
                    438: void
                    439: link_pool_init(void)
                    440: {
                    441:        __link_set_decl(pools, struct link_pool_init);
                    442:        struct link_pool_init * const *pi;
                    443:
                    444:        __link_set_foreach(pi, pools)
                    445:                pool_init((*pi)->pp, (*pi)->size, (*pi)->align,
                    446:                    (*pi)->align_offset, (*pi)->flags, (*pi)->wchan,
                    447:                    (*pi)->palloc);
                    448: }
                    449:
                    450: /*
1.3       pk        451:  * Initialize the given pool resource structure.
                    452:  *
                    453:  * We export this routine to allow other kernel parts to declare
                    454:  * static pools that must be initialized before malloc() is available.
                    455:  */
                    456: void
1.42      thorpej   457: pool_init(struct pool *pp, size_t size, u_int align, u_int ioff, int flags,
1.66      thorpej   458:     const char *wchan, struct pool_allocator *palloc)
1.3       pk        459: {
1.88      chs       460:        int off, slack;
1.92      enami     461:        size_t trysize, phsize;
1.93      dbj       462:        int s;
1.3       pk        463:
1.99      yamt      464:        KASSERT((1UL << (CHAR_BIT * sizeof(pool_item_freelist_t))) - 2 >=
                    465:            PHPOOL_FREELIST_NELEM(PHPOOL_MAX - 1));
                    466:
1.25      thorpej   467: #ifdef POOL_DIAGNOSTIC
                    468:        /*
                    469:         * Always log if POOL_DIAGNOSTIC is defined.
                    470:         */
                    471:        if (pool_logsize != 0)
                    472:                flags |= PR_LOGGING;
                    473: #endif
                    474:
1.66      thorpej   475: #ifdef POOL_SUBPAGE
                    476:        /*
                    477:         * XXX We don't provide a real `nointr' back-end
                    478:         * yet; all sub-pages come from a kmem back-end.
                    479:         * maybe some day...
                    480:         */
                    481:        if (palloc == NULL) {
                    482:                extern struct pool_allocator pool_allocator_kmem_subpage;
                    483:                palloc = &pool_allocator_kmem_subpage;
                    484:        }
1.3       pk        485:        /*
1.66      thorpej   486:         * We'll assume any user-specified back-end allocator
                    487:         * will deal with sub-pages, or simply don't care.
1.3       pk        488:         */
1.66      thorpej   489: #else
                    490:        if (palloc == NULL)
                    491:                palloc = &pool_allocator_kmem;
                    492: #endif /* POOL_SUBPAGE */
                    493:        if ((palloc->pa_flags & PA_INITIALIZED) == 0) {
                    494:                if (palloc->pa_pagesz == 0) {
1.62      bjh21     495: #ifdef POOL_SUBPAGE
1.66      thorpej   496:                        if (palloc == &pool_allocator_kmem)
                    497:                                palloc->pa_pagesz = PAGE_SIZE;
                    498:                        else
                    499:                                palloc->pa_pagesz = POOL_SUBPAGE;
1.62      bjh21     500: #else
1.66      thorpej   501:                        palloc->pa_pagesz = PAGE_SIZE;
                    502: #endif /* POOL_SUBPAGE */
                    503:                }
                    504:
                    505:                TAILQ_INIT(&palloc->pa_list);
                    506:
                    507:                simple_lock_init(&palloc->pa_slock);
                    508:                palloc->pa_pagemask = ~(palloc->pa_pagesz - 1);
                    509:                palloc->pa_pageshift = ffs(palloc->pa_pagesz) - 1;
                    510:                palloc->pa_flags |= PA_INITIALIZED;
1.4       thorpej   511:        }
1.3       pk        512:
                    513:        if (align == 0)
                    514:                align = ALIGN(1);
1.14      thorpej   515:
                    516:        if (size < sizeof(struct pool_item))
                    517:                size = sizeof(struct pool_item);
1.3       pk        518:
1.78      thorpej   519:        size = roundup(size, align);
1.66      thorpej   520: #ifdef DIAGNOSTIC
                    521:        if (size > palloc->pa_pagesz)
1.35      pk        522:                panic("pool_init: pool item size (%lu) too large",
                    523:                      (u_long)size);
1.66      thorpej   524: #endif
1.35      pk        525:
1.3       pk        526:        /*
                    527:         * Initialize the pool structure.
                    528:         */
1.88      chs       529:        LIST_INIT(&pp->pr_emptypages);
                    530:        LIST_INIT(&pp->pr_fullpages);
                    531:        LIST_INIT(&pp->pr_partpages);
1.43      thorpej   532:        TAILQ_INIT(&pp->pr_cachelist);
1.3       pk        533:        pp->pr_curpage = NULL;
                    534:        pp->pr_npages = 0;
                    535:        pp->pr_minitems = 0;
                    536:        pp->pr_minpages = 0;
                    537:        pp->pr_maxpages = UINT_MAX;
1.20      thorpej   538:        pp->pr_roflags = flags;
                    539:        pp->pr_flags = 0;
1.35      pk        540:        pp->pr_size = size;
1.3       pk        541:        pp->pr_align = align;
                    542:        pp->pr_wchan = wchan;
1.66      thorpej   543:        pp->pr_alloc = palloc;
1.20      thorpej   544:        pp->pr_nitems = 0;
                    545:        pp->pr_nout = 0;
                    546:        pp->pr_hardlimit = UINT_MAX;
                    547:        pp->pr_hardlimit_warning = NULL;
1.31      thorpej   548:        pp->pr_hardlimit_ratecap.tv_sec = 0;
                    549:        pp->pr_hardlimit_ratecap.tv_usec = 0;
                    550:        pp->pr_hardlimit_warning_last.tv_sec = 0;
                    551:        pp->pr_hardlimit_warning_last.tv_usec = 0;
1.68      thorpej   552:        pp->pr_drain_hook = NULL;
                    553:        pp->pr_drain_hook_arg = NULL;
1.3       pk        554:
                    555:        /*
                    556:         * Decide whether to put the page header off page to avoid
1.92      enami     557:         * wasting too large a part of the page or too big item.
                    558:         * Off-page page headers go on a hash table, so we can match
                    559:         * a returned item with its header based on the page address.
                    560:         * We use 1/16 of the page size and about 8 times of the item
                    561:         * size as the threshold (XXX: tune)
                    562:         *
                    563:         * However, we'll put the header into the page if we can put
                    564:         * it without wasting any items.
                    565:         *
                    566:         * Silently enforce `0 <= ioff < align'.
1.3       pk        567:         */
1.92      enami     568:        pp->pr_itemoffset = ioff %= align;
                    569:        /* See the comment below about reserved bytes. */
                    570:        trysize = palloc->pa_pagesz - ((align - ioff) % align);
                    571:        phsize = ALIGN(sizeof(struct pool_item_header));
1.97      yamt      572:        if ((pp->pr_roflags & PR_NOTOUCH) == 0 &&
                    573:            (pp->pr_size < MIN(palloc->pa_pagesz / 16, phsize << 3) ||
                    574:            trysize / pp->pr_size == (trysize - phsize) / pp->pr_size)) {
1.3       pk        575:                /* Use the end of the page for the page header */
1.20      thorpej   576:                pp->pr_roflags |= PR_PHINPAGE;
1.92      enami     577:                pp->pr_phoffset = off = palloc->pa_pagesz - phsize;
1.2       pk        578:        } else {
1.3       pk        579:                /* The page header will be taken from our page header pool */
                    580:                pp->pr_phoffset = 0;
1.66      thorpej   581:                off = palloc->pa_pagesz;
1.88      chs       582:                SPLAY_INIT(&pp->pr_phtree);
1.2       pk        583:        }
1.1       pk        584:
1.3       pk        585:        /*
                    586:         * Alignment is to take place at `ioff' within the item. This means
                    587:         * we must reserve up to `align - 1' bytes on the page to allow
                    588:         * appropriate positioning of each item.
                    589:         */
                    590:        pp->pr_itemsperpage = (off - ((align - ioff) % align)) / pp->pr_size;
1.43      thorpej   591:        KASSERT(pp->pr_itemsperpage != 0);
1.97      yamt      592:        if ((pp->pr_roflags & PR_NOTOUCH)) {
                    593:                int idx;
                    594:
                    595:                for (idx = 0; pp->pr_itemsperpage > PHPOOL_FREELIST_NELEM(idx);
                    596:                    idx++) {
                    597:                        /* nothing */
                    598:                }
                    599:                if (idx >= PHPOOL_MAX) {
                    600:                        /*
                    601:                         * if you see this panic, consider to tweak
                    602:                         * PHPOOL_MAX and PHPOOL_FREELIST_NELEM.
                    603:                         */
                    604:                        panic("%s: too large itemsperpage(%d) for PR_NOTOUCH",
                    605:                            pp->pr_wchan, pp->pr_itemsperpage);
                    606:                }
                    607:                pp->pr_phpool = &phpool[idx];
                    608:        } else if ((pp->pr_roflags & PR_PHINPAGE) == 0) {
                    609:                pp->pr_phpool = &phpool[0];
                    610:        }
                    611: #if defined(DIAGNOSTIC)
                    612:        else {
                    613:                pp->pr_phpool = NULL;
                    614:        }
                    615: #endif
1.3       pk        616:
                    617:        /*
                    618:         * Use the slack between the chunks and the page header
                    619:         * for "cache coloring".
                    620:         */
                    621:        slack = off - pp->pr_itemsperpage * pp->pr_size;
                    622:        pp->pr_maxcolor = (slack / align) * align;
                    623:        pp->pr_curcolor = 0;
                    624:
                    625:        pp->pr_nget = 0;
                    626:        pp->pr_nfail = 0;
                    627:        pp->pr_nput = 0;
                    628:        pp->pr_npagealloc = 0;
                    629:        pp->pr_npagefree = 0;
1.1       pk        630:        pp->pr_hiwat = 0;
1.8       thorpej   631:        pp->pr_nidle = 0;
1.3       pk        632:
1.59      thorpej   633: #ifdef POOL_DIAGNOSTIC
1.25      thorpej   634:        if (flags & PR_LOGGING) {
                    635:                if (kmem_map == NULL ||
                    636:                    (pp->pr_log = malloc(pool_logsize * sizeof(struct pool_log),
                    637:                     M_TEMP, M_NOWAIT)) == NULL)
1.20      thorpej   638:                        pp->pr_roflags &= ~PR_LOGGING;
1.3       pk        639:                pp->pr_curlogentry = 0;
                    640:                pp->pr_logsize = pool_logsize;
                    641:        }
1.59      thorpej   642: #endif
1.25      thorpej   643:
                    644:        pp->pr_entered_file = NULL;
                    645:        pp->pr_entered_line = 0;
1.3       pk        646:
1.21      thorpej   647:        simple_lock_init(&pp->pr_slock);
1.1       pk        648:
1.3       pk        649:        /*
1.43      thorpej   650:         * Initialize private page header pool and cache magazine pool if we
                    651:         * haven't done so yet.
1.23      thorpej   652:         * XXX LOCKING.
1.3       pk        653:         */
1.97      yamt      654:        if (phpool[0].pr_size == 0) {
                    655:                int idx;
                    656:                for (idx = 0; idx < PHPOOL_MAX; idx++) {
                    657:                        static char phpool_names[PHPOOL_MAX][6+1+6+1];
                    658:                        int nelem;
                    659:                        size_t sz;
                    660:
                    661:                        nelem = PHPOOL_FREELIST_NELEM(idx);
                    662:                        snprintf(phpool_names[idx], sizeof(phpool_names[idx]),
                    663:                            "phpool-%d", nelem);
                    664:                        sz = sizeof(struct pool_item_header);
                    665:                        if (nelem) {
                    666:                                sz = PR_FREELIST_ALIGN(sz)
1.99      yamt      667:                                    + nelem * sizeof(pool_item_freelist_t);
1.97      yamt      668:                        }
                    669:                        pool_init(&phpool[idx], sz, 0, 0, 0,
1.98      yamt      670:                            phpool_names[idx], &pool_allocator_meta);
1.97      yamt      671:                }
1.62      bjh21     672: #ifdef POOL_SUBPAGE
                    673:                pool_init(&psppool, POOL_SUBPAGE, POOL_SUBPAGE, 0,
1.98      yamt      674:                    PR_RECURSIVE, "psppool", &pool_allocator_meta);
1.62      bjh21     675: #endif
1.43      thorpej   676:                pool_init(&pcgpool, sizeof(struct pool_cache_group), 0, 0,
1.98      yamt      677:                    0, "pcgpool", &pool_allocator_meta);
1.1       pk        678:        }
                    679:
1.23      thorpej   680:        /* Insert into the list of all pools. */
                    681:        simple_lock(&pool_head_slock);
                    682:        TAILQ_INSERT_TAIL(&pool_head, pp, pr_poollist);
                    683:        simple_unlock(&pool_head_slock);
1.66      thorpej   684:
                    685:        /* Insert this into the list of pools using this allocator. */
1.93      dbj       686:        s = splvm();
1.66      thorpej   687:        simple_lock(&palloc->pa_slock);
                    688:        TAILQ_INSERT_TAIL(&palloc->pa_list, pp, pr_alloc_list);
                    689:        simple_unlock(&palloc->pa_slock);
1.93      dbj       690:        splx(s);
1.1       pk        691: }
                    692:
                    693: /*
                    694:  * De-commision a pool resource.
                    695:  */
                    696: void
1.42      thorpej   697: pool_destroy(struct pool *pp)
1.1       pk        698: {
1.3       pk        699:        struct pool_item_header *ph;
1.43      thorpej   700:        struct pool_cache *pc;
1.93      dbj       701:        int s;
1.43      thorpej   702:
1.66      thorpej   703:        /* Locking order: pool_allocator -> pool */
1.93      dbj       704:        s = splvm();
1.66      thorpej   705:        simple_lock(&pp->pr_alloc->pa_slock);
                    706:        TAILQ_REMOVE(&pp->pr_alloc->pa_list, pp, pr_alloc_list);
                    707:        simple_unlock(&pp->pr_alloc->pa_slock);
1.93      dbj       708:        splx(s);
1.66      thorpej   709:
1.43      thorpej   710:        /* Destroy all caches for this pool. */
                    711:        while ((pc = TAILQ_FIRST(&pp->pr_cachelist)) != NULL)
                    712:                pool_cache_destroy(pc);
1.3       pk        713:
                    714: #ifdef DIAGNOSTIC
1.20      thorpej   715:        if (pp->pr_nout != 0) {
1.25      thorpej   716:                pr_printlog(pp, NULL, printf);
1.80      provos    717:                panic("pool_destroy: pool busy: still out: %u",
1.20      thorpej   718:                    pp->pr_nout);
1.3       pk        719:        }
                    720: #endif
1.1       pk        721:
1.3       pk        722:        /* Remove all pages */
1.88      chs       723:        while ((ph = LIST_FIRST(&pp->pr_emptypages)) != NULL)
1.70      thorpej   724:                pr_rmpage(pp, ph, NULL);
1.88      chs       725:        KASSERT(LIST_EMPTY(&pp->pr_fullpages));
                    726:        KASSERT(LIST_EMPTY(&pp->pr_partpages));
1.3       pk        727:
                    728:        /* Remove from global pool list */
1.23      thorpej   729:        simple_lock(&pool_head_slock);
1.3       pk        730:        TAILQ_REMOVE(&pool_head, pp, pr_poollist);
1.61      chs       731:        if (drainpp == pp) {
                    732:                drainpp = NULL;
                    733:        }
1.23      thorpej   734:        simple_unlock(&pool_head_slock);
1.3       pk        735:
1.59      thorpej   736: #ifdef POOL_DIAGNOSTIC
1.20      thorpej   737:        if ((pp->pr_roflags & PR_LOGGING) != 0)
1.3       pk        738:                free(pp->pr_log, M_TEMP);
1.59      thorpej   739: #endif
1.1       pk        740: }
                    741:
1.68      thorpej   742: void
                    743: pool_set_drain_hook(struct pool *pp, void (*fn)(void *, int), void *arg)
                    744: {
                    745:
                    746:        /* XXX no locking -- must be used just after pool_init() */
                    747: #ifdef DIAGNOSTIC
                    748:        if (pp->pr_drain_hook != NULL)
                    749:                panic("pool_set_drain_hook(%s): already set", pp->pr_wchan);
                    750: #endif
                    751:        pp->pr_drain_hook = fn;
                    752:        pp->pr_drain_hook_arg = arg;
                    753: }
                    754:
1.88      chs       755: static struct pool_item_header *
1.55      thorpej   756: pool_alloc_item_header(struct pool *pp, caddr_t storage, int flags)
                    757: {
                    758:        struct pool_item_header *ph;
                    759:        int s;
                    760:
                    761:        LOCK_ASSERT(simple_lock_held(&pp->pr_slock) == 0);
                    762:
                    763:        if ((pp->pr_roflags & PR_PHINPAGE) != 0)
                    764:                ph = (struct pool_item_header *) (storage + pp->pr_phoffset);
                    765:        else {
1.85      pk        766:                s = splvm();
1.97      yamt      767:                ph = pool_get(pp->pr_phpool, flags);
1.55      thorpej   768:                splx(s);
                    769:        }
                    770:
                    771:        return (ph);
                    772: }
1.1       pk        773:
                    774: /*
1.3       pk        775:  * Grab an item from the pool; must be called at appropriate spl level
1.1       pk        776:  */
1.3       pk        777: void *
1.59      thorpej   778: #ifdef POOL_DIAGNOSTIC
1.42      thorpej   779: _pool_get(struct pool *pp, int flags, const char *file, long line)
1.56      sommerfe  780: #else
                    781: pool_get(struct pool *pp, int flags)
                    782: #endif
1.1       pk        783: {
                    784:        struct pool_item *pi;
1.3       pk        785:        struct pool_item_header *ph;
1.55      thorpej   786:        void *v;
1.1       pk        787:
1.2       pk        788: #ifdef DIAGNOSTIC
1.95      atatat    789:        if (__predict_false(pp->pr_itemsperpage == 0))
                    790:                panic("pool_get: pool %p: pr_itemsperpage is zero, "
                    791:                    "pool not initialized?", pp);
1.84      thorpej   792:        if (__predict_false(curlwp == NULL && doing_shutdown == 0 &&
1.37      sommerfe  793:                            (flags & PR_WAITOK) != 0))
1.77      matt      794:                panic("pool_get: %s: must have NOWAIT", pp->pr_wchan);
1.58      thorpej   795:
                    796: #ifdef LOCKDEBUG
                    797:        if (flags & PR_WAITOK)
                    798:                simple_lock_only_held(NULL, "pool_get(PR_WAITOK)");
1.56      sommerfe  799: #endif
1.58      thorpej   800: #endif /* DIAGNOSTIC */
1.1       pk        801:
1.21      thorpej   802:        simple_lock(&pp->pr_slock);
1.25      thorpej   803:        pr_enter(pp, file, line);
1.20      thorpej   804:
                    805:  startover:
                    806:        /*
                    807:         * Check to see if we've reached the hard limit.  If we have,
                    808:         * and we can wait, then wait until an item has been returned to
                    809:         * the pool.
                    810:         */
                    811: #ifdef DIAGNOSTIC
1.34      thorpej   812:        if (__predict_false(pp->pr_nout > pp->pr_hardlimit)) {
1.25      thorpej   813:                pr_leave(pp);
1.21      thorpej   814:                simple_unlock(&pp->pr_slock);
1.20      thorpej   815:                panic("pool_get: %s: crossed hard limit", pp->pr_wchan);
                    816:        }
                    817: #endif
1.34      thorpej   818:        if (__predict_false(pp->pr_nout == pp->pr_hardlimit)) {
1.68      thorpej   819:                if (pp->pr_drain_hook != NULL) {
                    820:                        /*
                    821:                         * Since the drain hook is going to free things
                    822:                         * back to the pool, unlock, call the hook, re-lock,
                    823:                         * and check the hardlimit condition again.
                    824:                         */
                    825:                        pr_leave(pp);
                    826:                        simple_unlock(&pp->pr_slock);
                    827:                        (*pp->pr_drain_hook)(pp->pr_drain_hook_arg, flags);
                    828:                        simple_lock(&pp->pr_slock);
                    829:                        pr_enter(pp, file, line);
                    830:                        if (pp->pr_nout < pp->pr_hardlimit)
                    831:                                goto startover;
                    832:                }
                    833:
1.29      sommerfe  834:                if ((flags & PR_WAITOK) && !(flags & PR_LIMITFAIL)) {
1.20      thorpej   835:                        /*
                    836:                         * XXX: A warning isn't logged in this case.  Should
                    837:                         * it be?
                    838:                         */
                    839:                        pp->pr_flags |= PR_WANTED;
1.25      thorpej   840:                        pr_leave(pp);
1.40      sommerfe  841:                        ltsleep(pp, PSWP, pp->pr_wchan, 0, &pp->pr_slock);
1.25      thorpej   842:                        pr_enter(pp, file, line);
1.20      thorpej   843:                        goto startover;
                    844:                }
1.31      thorpej   845:
                    846:                /*
                    847:                 * Log a message that the hard limit has been hit.
                    848:                 */
                    849:                if (pp->pr_hardlimit_warning != NULL &&
                    850:                    ratecheck(&pp->pr_hardlimit_warning_last,
                    851:                              &pp->pr_hardlimit_ratecap))
                    852:                        log(LOG_ERR, "%s\n", pp->pr_hardlimit_warning);
1.21      thorpej   853:
                    854:                pp->pr_nfail++;
                    855:
1.25      thorpej   856:                pr_leave(pp);
1.21      thorpej   857:                simple_unlock(&pp->pr_slock);
1.20      thorpej   858:                return (NULL);
                    859:        }
                    860:
1.3       pk        861:        /*
                    862:         * The convention we use is that if `curpage' is not NULL, then
                    863:         * it points at a non-empty bucket. In particular, `curpage'
                    864:         * never points at a page header which has PR_PHINPAGE set and
                    865:         * has no items in its bucket.
                    866:         */
1.20      thorpej   867:        if ((ph = pp->pr_curpage) == NULL) {
                    868: #ifdef DIAGNOSTIC
                    869:                if (pp->pr_nitems != 0) {
1.21      thorpej   870:                        simple_unlock(&pp->pr_slock);
1.20      thorpej   871:                        printf("pool_get: %s: curpage NULL, nitems %u\n",
                    872:                            pp->pr_wchan, pp->pr_nitems);
1.80      provos    873:                        panic("pool_get: nitems inconsistent");
1.20      thorpej   874:                }
                    875: #endif
                    876:
1.21      thorpej   877:                /*
                    878:                 * Call the back-end page allocator for more memory.
                    879:                 * Release the pool lock, as the back-end page allocator
                    880:                 * may block.
                    881:                 */
1.25      thorpej   882:                pr_leave(pp);
1.21      thorpej   883:                simple_unlock(&pp->pr_slock);
1.66      thorpej   884:                v = pool_allocator_alloc(pp, flags);
1.55      thorpej   885:                if (__predict_true(v != NULL))
                    886:                        ph = pool_alloc_item_header(pp, v, flags);
1.15      pk        887:
1.55      thorpej   888:                if (__predict_false(v == NULL || ph == NULL)) {
                    889:                        if (v != NULL)
1.66      thorpej   890:                                pool_allocator_free(pp, v);
1.55      thorpej   891:
1.91      yamt      892:                        simple_lock(&pp->pr_slock);
                    893:                        pr_enter(pp, file, line);
                    894:
1.21      thorpej   895:                        /*
1.55      thorpej   896:                         * We were unable to allocate a page or item
                    897:                         * header, but we released the lock during
                    898:                         * allocation, so perhaps items were freed
                    899:                         * back to the pool.  Check for this case.
1.21      thorpej   900:                         */
                    901:                        if (pp->pr_curpage != NULL)
                    902:                                goto startover;
1.15      pk        903:
1.3       pk        904:                        if ((flags & PR_WAITOK) == 0) {
                    905:                                pp->pr_nfail++;
1.25      thorpej   906:                                pr_leave(pp);
1.21      thorpej   907:                                simple_unlock(&pp->pr_slock);
1.1       pk        908:                                return (NULL);
1.3       pk        909:                        }
                    910:
1.15      pk        911:                        /*
                    912:                         * Wait for items to be returned to this pool.
1.21      thorpej   913:                         *
1.20      thorpej   914:                         * XXX: maybe we should wake up once a second and
                    915:                         * try again?
1.15      pk        916:                         */
1.1       pk        917:                        pp->pr_flags |= PR_WANTED;
1.66      thorpej   918:                        /* PA_WANTED is already set on the allocator. */
1.25      thorpej   919:                        pr_leave(pp);
1.40      sommerfe  920:                        ltsleep(pp, PSWP, pp->pr_wchan, 0, &pp->pr_slock);
1.25      thorpej   921:                        pr_enter(pp, file, line);
1.20      thorpej   922:                        goto startover;
1.1       pk        923:                }
1.3       pk        924:
1.15      pk        925:                /* We have more memory; add it to the pool */
1.91      yamt      926:                simple_lock(&pp->pr_slock);
                    927:                pr_enter(pp, file, line);
1.55      thorpej   928:                pool_prime_page(pp, v, ph);
1.15      pk        929:                pp->pr_npagealloc++;
                    930:
1.20      thorpej   931:                /* Start the allocation process over. */
                    932:                goto startover;
1.3       pk        933:        }
1.97      yamt      934:        if (pp->pr_roflags & PR_NOTOUCH) {
                    935: #ifdef DIAGNOSTIC
                    936:                if (__predict_false(ph->ph_nmissing == pp->pr_itemsperpage)) {
                    937:                        pr_leave(pp);
                    938:                        simple_unlock(&pp->pr_slock);
                    939:                        panic("pool_get: %s: page empty", pp->pr_wchan);
                    940:                }
                    941: #endif
                    942:                v = pr_item_notouch_get(pp, ph);
                    943: #ifdef POOL_DIAGNOSTIC
                    944:                pr_log(pp, v, PRLOG_GET, file, line);
                    945: #endif
                    946:        } else {
                    947:                v = pi = TAILQ_FIRST(&ph->ph_itemlist);
                    948:                if (__predict_false(v == NULL)) {
                    949:                        pr_leave(pp);
                    950:                        simple_unlock(&pp->pr_slock);
                    951:                        panic("pool_get: %s: page empty", pp->pr_wchan);
                    952:                }
1.20      thorpej   953: #ifdef DIAGNOSTIC
1.97      yamt      954:                if (__predict_false(pp->pr_nitems == 0)) {
                    955:                        pr_leave(pp);
                    956:                        simple_unlock(&pp->pr_slock);
                    957:                        printf("pool_get: %s: items on itemlist, nitems %u\n",
                    958:                            pp->pr_wchan, pp->pr_nitems);
                    959:                        panic("pool_get: nitems inconsistent");
                    960:                }
1.65      enami     961: #endif
1.56      sommerfe  962:
1.65      enami     963: #ifdef POOL_DIAGNOSTIC
1.97      yamt      964:                pr_log(pp, v, PRLOG_GET, file, line);
1.65      enami     965: #endif
1.3       pk        966:
1.65      enami     967: #ifdef DIAGNOSTIC
1.97      yamt      968:                if (__predict_false(pi->pi_magic != PI_MAGIC)) {
                    969:                        pr_printlog(pp, pi, printf);
                    970:                        panic("pool_get(%s): free list modified: "
                    971:                            "magic=%x; page %p; item addr %p\n",
                    972:                            pp->pr_wchan, pi->pi_magic, ph->ph_page, pi);
                    973:                }
1.3       pk        974: #endif
                    975:
1.97      yamt      976:                /*
                    977:                 * Remove from item list.
                    978:                 */
                    979:                TAILQ_REMOVE(&ph->ph_itemlist, pi, pi_list);
                    980:        }
1.20      thorpej   981:        pp->pr_nitems--;
                    982:        pp->pr_nout++;
1.6       thorpej   983:        if (ph->ph_nmissing == 0) {
                    984: #ifdef DIAGNOSTIC
1.34      thorpej   985:                if (__predict_false(pp->pr_nidle == 0))
1.6       thorpej   986:                        panic("pool_get: nidle inconsistent");
                    987: #endif
                    988:                pp->pr_nidle--;
1.88      chs       989:
                    990:                /*
                    991:                 * This page was previously empty.  Move it to the list of
                    992:                 * partially-full pages.  This page is already curpage.
                    993:                 */
                    994:                LIST_REMOVE(ph, ph_pagelist);
                    995:                LIST_INSERT_HEAD(&pp->pr_partpages, ph, ph_pagelist);
1.6       thorpej   996:        }
1.3       pk        997:        ph->ph_nmissing++;
1.97      yamt      998:        if (ph->ph_nmissing == pp->pr_itemsperpage) {
1.21      thorpej   999: #ifdef DIAGNOSTIC
1.97      yamt     1000:                if (__predict_false((pp->pr_roflags & PR_NOTOUCH) == 0 &&
                   1001:                    !TAILQ_EMPTY(&ph->ph_itemlist))) {
1.25      thorpej  1002:                        pr_leave(pp);
1.21      thorpej  1003:                        simple_unlock(&pp->pr_slock);
                   1004:                        panic("pool_get: %s: nmissing inconsistent",
                   1005:                            pp->pr_wchan);
                   1006:                }
                   1007: #endif
1.3       pk       1008:                /*
1.88      chs      1009:                 * This page is now full.  Move it to the full list
                   1010:                 * and select a new current page.
1.3       pk       1011:                 */
1.88      chs      1012:                LIST_REMOVE(ph, ph_pagelist);
                   1013:                LIST_INSERT_HEAD(&pp->pr_fullpages, ph, ph_pagelist);
                   1014:                pool_update_curpage(pp);
1.1       pk       1015:        }
1.3       pk       1016:
                   1017:        pp->pr_nget++;
1.20      thorpej  1018:
                   1019:        /*
                   1020:         * If we have a low water mark and we are now below that low
                   1021:         * water mark, add more items to the pool.
                   1022:         */
1.53      thorpej  1023:        if (POOL_NEEDS_CATCHUP(pp) && pool_catchup(pp) != 0) {
1.20      thorpej  1024:                /*
                   1025:                 * XXX: Should we log a warning?  Should we set up a timeout
                   1026:                 * to try again in a second or so?  The latter could break
                   1027:                 * a caller's assumptions about interrupt protection, etc.
                   1028:                 */
                   1029:        }
                   1030:
1.25      thorpej  1031:        pr_leave(pp);
1.21      thorpej  1032:        simple_unlock(&pp->pr_slock);
1.1       pk       1033:        return (v);
                   1034: }
                   1035:
                   1036: /*
1.43      thorpej  1037:  * Internal version of pool_put().  Pool is already locked/entered.
1.1       pk       1038:  */
1.43      thorpej  1039: static void
1.56      sommerfe 1040: pool_do_put(struct pool *pp, void *v)
1.1       pk       1041: {
                   1042:        struct pool_item *pi = v;
1.3       pk       1043:        struct pool_item_header *ph;
                   1044:        caddr_t page;
1.21      thorpej  1045:        int s;
1.3       pk       1046:
1.61      chs      1047:        LOCK_ASSERT(simple_lock_held(&pp->pr_slock));
                   1048:
1.66      thorpej  1049:        page = (caddr_t)((u_long)v & pp->pr_alloc->pa_pagemask);
1.1       pk       1050:
1.30      thorpej  1051: #ifdef DIAGNOSTIC
1.34      thorpej  1052:        if (__predict_false(pp->pr_nout == 0)) {
1.30      thorpej  1053:                printf("pool %s: putting with none out\n",
                   1054:                    pp->pr_wchan);
                   1055:                panic("pool_put");
                   1056:        }
                   1057: #endif
1.3       pk       1058:
1.34      thorpej  1059:        if (__predict_false((ph = pr_find_pagehead(pp, page)) == NULL)) {
1.25      thorpej  1060:                pr_printlog(pp, NULL, printf);
1.3       pk       1061:                panic("pool_put: %s: page header missing", pp->pr_wchan);
                   1062:        }
1.28      thorpej  1063:
                   1064: #ifdef LOCKDEBUG
                   1065:        /*
                   1066:         * Check if we're freeing a locked simple lock.
                   1067:         */
                   1068:        simple_lock_freecheck((caddr_t)pi, ((caddr_t)pi) + pp->pr_size);
                   1069: #endif
1.3       pk       1070:
                   1071:        /*
                   1072:         * Return to item list.
                   1073:         */
1.97      yamt     1074:        if (pp->pr_roflags & PR_NOTOUCH) {
                   1075:                pr_item_notouch_put(pp, ph, v);
                   1076:        } else {
1.2       pk       1077: #ifdef DIAGNOSTIC
1.97      yamt     1078:                pi->pi_magic = PI_MAGIC;
1.3       pk       1079: #endif
1.32      chs      1080: #ifdef DEBUG
1.97      yamt     1081:                {
                   1082:                        int i, *ip = v;
1.32      chs      1083:
1.97      yamt     1084:                        for (i = 0; i < pp->pr_size / sizeof(int); i++) {
                   1085:                                *ip++ = PI_MAGIC;
                   1086:                        }
1.32      chs      1087:                }
                   1088: #endif
                   1089:
1.97      yamt     1090:                TAILQ_INSERT_HEAD(&ph->ph_itemlist, pi, pi_list);
                   1091:        }
1.79      thorpej  1092:        KDASSERT(ph->ph_nmissing != 0);
1.3       pk       1093:        ph->ph_nmissing--;
                   1094:        pp->pr_nput++;
1.20      thorpej  1095:        pp->pr_nitems++;
                   1096:        pp->pr_nout--;
1.3       pk       1097:
                   1098:        /* Cancel "pool empty" condition if it exists */
                   1099:        if (pp->pr_curpage == NULL)
                   1100:                pp->pr_curpage = ph;
                   1101:
                   1102:        if (pp->pr_flags & PR_WANTED) {
                   1103:                pp->pr_flags &= ~PR_WANTED;
1.15      pk       1104:                if (ph->ph_nmissing == 0)
                   1105:                        pp->pr_nidle++;
1.3       pk       1106:                wakeup((caddr_t)pp);
                   1107:                return;
                   1108:        }
                   1109:
                   1110:        /*
1.88      chs      1111:         * If this page is now empty, do one of two things:
1.21      thorpej  1112:         *
1.88      chs      1113:         *      (1) If we have more pages than the page high water mark,
1.96      thorpej  1114:         *          free the page back to the system.  ONLY CONSIDER
1.90      thorpej  1115:         *          FREEING BACK A PAGE IF WE HAVE MORE THAN OUR MINIMUM PAGE
                   1116:         *          CLAIM.
1.21      thorpej  1117:         *
1.88      chs      1118:         *      (2) Otherwise, move the page to the empty page list.
                   1119:         *
                   1120:         * Either way, select a new current page (so we use a partially-full
                   1121:         * page if one is available).
1.3       pk       1122:         */
                   1123:        if (ph->ph_nmissing == 0) {
1.6       thorpej  1124:                pp->pr_nidle++;
1.90      thorpej  1125:                if (pp->pr_npages > pp->pr_minpages &&
                   1126:                    (pp->pr_npages > pp->pr_maxpages ||
                   1127:                     (pp->pr_alloc->pa_flags & PA_WANT) != 0)) {
1.91      yamt     1128:                        simple_unlock(&pp->pr_slock);
1.61      chs      1129:                        pr_rmpage(pp, ph, NULL);
1.91      yamt     1130:                        simple_lock(&pp->pr_slock);
1.3       pk       1131:                } else {
1.88      chs      1132:                        LIST_REMOVE(ph, ph_pagelist);
                   1133:                        LIST_INSERT_HEAD(&pp->pr_emptypages, ph, ph_pagelist);
1.3       pk       1134:
1.21      thorpej  1135:                        /*
                   1136:                         * Update the timestamp on the page.  A page must
                   1137:                         * be idle for some period of time before it can
                   1138:                         * be reclaimed by the pagedaemon.  This minimizes
                   1139:                         * ping-pong'ing for memory.
                   1140:                         */
                   1141:                        s = splclock();
                   1142:                        ph->ph_time = mono_time;
                   1143:                        splx(s);
1.1       pk       1144:                }
1.88      chs      1145:                pool_update_curpage(pp);
1.1       pk       1146:        }
1.88      chs      1147:
1.21      thorpej  1148:        /*
1.88      chs      1149:         * If the page was previously completely full, move it to the
                   1150:         * partially-full list and make it the current page.  The next
                   1151:         * allocation will get the item from this page, instead of
                   1152:         * further fragmenting the pool.
1.21      thorpej  1153:         */
                   1154:        else if (ph->ph_nmissing == (pp->pr_itemsperpage - 1)) {
1.88      chs      1155:                LIST_REMOVE(ph, ph_pagelist);
                   1156:                LIST_INSERT_HEAD(&pp->pr_partpages, ph, ph_pagelist);
1.21      thorpej  1157:                pp->pr_curpage = ph;
                   1158:        }
1.43      thorpej  1159: }
                   1160:
                   1161: /*
                   1162:  * Return resource to the pool; must be called at appropriate spl level
                   1163:  */
1.59      thorpej  1164: #ifdef POOL_DIAGNOSTIC
1.43      thorpej  1165: void
                   1166: _pool_put(struct pool *pp, void *v, const char *file, long line)
                   1167: {
                   1168:
                   1169:        simple_lock(&pp->pr_slock);
                   1170:        pr_enter(pp, file, line);
                   1171:
1.56      sommerfe 1172:        pr_log(pp, v, PRLOG_PUT, file, line);
                   1173:
                   1174:        pool_do_put(pp, v);
1.21      thorpej  1175:
1.25      thorpej  1176:        pr_leave(pp);
1.21      thorpej  1177:        simple_unlock(&pp->pr_slock);
1.1       pk       1178: }
1.57      sommerfe 1179: #undef pool_put
1.59      thorpej  1180: #endif /* POOL_DIAGNOSTIC */
1.1       pk       1181:
1.56      sommerfe 1182: void
                   1183: pool_put(struct pool *pp, void *v)
                   1184: {
                   1185:
                   1186:        simple_lock(&pp->pr_slock);
                   1187:
                   1188:        pool_do_put(pp, v);
                   1189:
                   1190:        simple_unlock(&pp->pr_slock);
                   1191: }
1.57      sommerfe 1192:
1.59      thorpej  1193: #ifdef POOL_DIAGNOSTIC
1.57      sommerfe 1194: #define                pool_put(h, v)  _pool_put((h), (v), __FILE__, __LINE__)
1.56      sommerfe 1195: #endif
1.74      thorpej  1196:
                   1197: /*
                   1198:  * Add N items to the pool.
                   1199:  */
                   1200: int
                   1201: pool_prime(struct pool *pp, int n)
                   1202: {
1.83      scw      1203:        struct pool_item_header *ph = NULL;
1.74      thorpej  1204:        caddr_t cp;
1.75      simonb   1205:        int newpages;
1.74      thorpej  1206:
                   1207:        simple_lock(&pp->pr_slock);
                   1208:
                   1209:        newpages = roundup(n, pp->pr_itemsperpage) / pp->pr_itemsperpage;
                   1210:
                   1211:        while (newpages-- > 0) {
                   1212:                simple_unlock(&pp->pr_slock);
                   1213:                cp = pool_allocator_alloc(pp, PR_NOWAIT);
                   1214:                if (__predict_true(cp != NULL))
                   1215:                        ph = pool_alloc_item_header(pp, cp, PR_NOWAIT);
                   1216:
                   1217:                if (__predict_false(cp == NULL || ph == NULL)) {
                   1218:                        if (cp != NULL)
                   1219:                                pool_allocator_free(pp, cp);
1.91      yamt     1220:                        simple_lock(&pp->pr_slock);
1.74      thorpej  1221:                        break;
                   1222:                }
                   1223:
1.91      yamt     1224:                simple_lock(&pp->pr_slock);
1.74      thorpej  1225:                pool_prime_page(pp, cp, ph);
                   1226:                pp->pr_npagealloc++;
                   1227:                pp->pr_minpages++;
                   1228:        }
                   1229:
                   1230:        if (pp->pr_minpages >= pp->pr_maxpages)
                   1231:                pp->pr_maxpages = pp->pr_minpages + 1;  /* XXX */
                   1232:
                   1233:        simple_unlock(&pp->pr_slock);
                   1234:        return (0);
                   1235: }
1.55      thorpej  1236:
                   1237: /*
1.3       pk       1238:  * Add a page worth of items to the pool.
1.21      thorpej  1239:  *
                   1240:  * Note, we must be called with the pool descriptor LOCKED.
1.3       pk       1241:  */
1.55      thorpej  1242: static void
                   1243: pool_prime_page(struct pool *pp, caddr_t storage, struct pool_item_header *ph)
1.3       pk       1244: {
                   1245:        struct pool_item *pi;
                   1246:        caddr_t cp = storage;
                   1247:        unsigned int align = pp->pr_align;
                   1248:        unsigned int ioff = pp->pr_itemoffset;
1.55      thorpej  1249:        int n;
1.89      yamt     1250:        int s;
1.36      pk       1251:
1.91      yamt     1252:        LOCK_ASSERT(simple_lock_held(&pp->pr_slock));
                   1253:
1.66      thorpej  1254: #ifdef DIAGNOSTIC
                   1255:        if (((u_long)cp & (pp->pr_alloc->pa_pagesz - 1)) != 0)
1.36      pk       1256:                panic("pool_prime_page: %s: unaligned page", pp->pr_wchan);
1.66      thorpej  1257: #endif
1.3       pk       1258:
                   1259:        /*
                   1260:         * Insert page header.
                   1261:         */
1.88      chs      1262:        LIST_INSERT_HEAD(&pp->pr_emptypages, ph, ph_pagelist);
1.3       pk       1263:        TAILQ_INIT(&ph->ph_itemlist);
                   1264:        ph->ph_page = storage;
                   1265:        ph->ph_nmissing = 0;
1.89      yamt     1266:        s = splclock();
                   1267:        ph->ph_time = mono_time;
                   1268:        splx(s);
1.88      chs      1269:        if ((pp->pr_roflags & PR_PHINPAGE) == 0)
                   1270:                SPLAY_INSERT(phtree, &pp->pr_phtree, ph);
1.3       pk       1271:
1.6       thorpej  1272:        pp->pr_nidle++;
                   1273:
1.3       pk       1274:        /*
                   1275:         * Color this page.
                   1276:         */
                   1277:        cp = (caddr_t)(cp + pp->pr_curcolor);
                   1278:        if ((pp->pr_curcolor += align) > pp->pr_maxcolor)
                   1279:                pp->pr_curcolor = 0;
                   1280:
                   1281:        /*
                   1282:         * Adjust storage to apply aligment to `pr_itemoffset' in each item.
                   1283:         */
                   1284:        if (ioff != 0)
                   1285:                cp = (caddr_t)(cp + (align - ioff));
                   1286:
                   1287:        /*
                   1288:         * Insert remaining chunks on the bucket list.
                   1289:         */
                   1290:        n = pp->pr_itemsperpage;
1.20      thorpej  1291:        pp->pr_nitems += n;
1.3       pk       1292:
1.97      yamt     1293:        if (pp->pr_roflags & PR_NOTOUCH) {
1.99      yamt     1294:                pool_item_freelist_t *freelist = PR_FREELIST(ph);
1.97      yamt     1295:                int i;
                   1296:
1.99      yamt     1297:                ph->ph_off = cp - storage;
1.97      yamt     1298:                ph->ph_firstfree = 0;
                   1299:                for (i = 0; i < n - 1; i++)
                   1300:                        freelist[i] = i + 1;
                   1301:                freelist[n - 1] = PR_INDEX_EOL;
                   1302:        } else {
                   1303:                while (n--) {
                   1304:                        pi = (struct pool_item *)cp;
1.78      thorpej  1305:
1.97      yamt     1306:                        KASSERT(((((vaddr_t)pi) + ioff) & (align - 1)) == 0);
1.3       pk       1307:
1.97      yamt     1308:                        /* Insert on page list */
                   1309:                        TAILQ_INSERT_TAIL(&ph->ph_itemlist, pi, pi_list);
1.3       pk       1310: #ifdef DIAGNOSTIC
1.97      yamt     1311:                        pi->pi_magic = PI_MAGIC;
1.3       pk       1312: #endif
1.97      yamt     1313:                        cp = (caddr_t)(cp + pp->pr_size);
                   1314:                }
1.3       pk       1315:        }
                   1316:
                   1317:        /*
                   1318:         * If the pool was depleted, point at the new page.
                   1319:         */
                   1320:        if (pp->pr_curpage == NULL)
                   1321:                pp->pr_curpage = ph;
                   1322:
                   1323:        if (++pp->pr_npages > pp->pr_hiwat)
                   1324:                pp->pr_hiwat = pp->pr_npages;
                   1325: }
                   1326:
1.20      thorpej  1327: /*
1.52      thorpej  1328:  * Used by pool_get() when nitems drops below the low water mark.  This
1.88      chs      1329:  * is used to catch up pr_nitems with the low water mark.
1.20      thorpej  1330:  *
1.21      thorpej  1331:  * Note 1, we never wait for memory here, we let the caller decide what to do.
1.20      thorpej  1332:  *
1.73      thorpej  1333:  * Note 2, we must be called with the pool already locked, and we return
1.20      thorpej  1334:  * with it locked.
                   1335:  */
                   1336: static int
1.42      thorpej  1337: pool_catchup(struct pool *pp)
1.20      thorpej  1338: {
1.83      scw      1339:        struct pool_item_header *ph = NULL;
1.20      thorpej  1340:        caddr_t cp;
                   1341:        int error = 0;
                   1342:
1.54      thorpej  1343:        while (POOL_NEEDS_CATCHUP(pp)) {
1.20      thorpej  1344:                /*
1.21      thorpej  1345:                 * Call the page back-end allocator for more memory.
                   1346:                 *
                   1347:                 * XXX: We never wait, so should we bother unlocking
                   1348:                 * the pool descriptor?
1.20      thorpej  1349:                 */
1.21      thorpej  1350:                simple_unlock(&pp->pr_slock);
1.66      thorpej  1351:                cp = pool_allocator_alloc(pp, PR_NOWAIT);
1.55      thorpej  1352:                if (__predict_true(cp != NULL))
                   1353:                        ph = pool_alloc_item_header(pp, cp, PR_NOWAIT);
                   1354:                if (__predict_false(cp == NULL || ph == NULL)) {
                   1355:                        if (cp != NULL)
1.66      thorpej  1356:                                pool_allocator_free(pp, cp);
1.20      thorpej  1357:                        error = ENOMEM;
1.91      yamt     1358:                        simple_lock(&pp->pr_slock);
1.20      thorpej  1359:                        break;
                   1360:                }
1.91      yamt     1361:                simple_lock(&pp->pr_slock);
1.55      thorpej  1362:                pool_prime_page(pp, cp, ph);
1.26      thorpej  1363:                pp->pr_npagealloc++;
1.20      thorpej  1364:        }
                   1365:
                   1366:        return (error);
                   1367: }
                   1368:
1.88      chs      1369: static void
                   1370: pool_update_curpage(struct pool *pp)
                   1371: {
                   1372:
                   1373:        pp->pr_curpage = LIST_FIRST(&pp->pr_partpages);
                   1374:        if (pp->pr_curpage == NULL) {
                   1375:                pp->pr_curpage = LIST_FIRST(&pp->pr_emptypages);
                   1376:        }
                   1377: }
                   1378:
1.3       pk       1379: void
1.42      thorpej  1380: pool_setlowat(struct pool *pp, int n)
1.3       pk       1381: {
1.15      pk       1382:
1.21      thorpej  1383:        simple_lock(&pp->pr_slock);
                   1384:
1.3       pk       1385:        pp->pr_minitems = n;
1.15      pk       1386:        pp->pr_minpages = (n == 0)
                   1387:                ? 0
1.18      thorpej  1388:                : roundup(n, pp->pr_itemsperpage) / pp->pr_itemsperpage;
1.20      thorpej  1389:
                   1390:        /* Make sure we're caught up with the newly-set low water mark. */
1.75      simonb   1391:        if (POOL_NEEDS_CATCHUP(pp) && pool_catchup(pp) != 0) {
1.20      thorpej  1392:                /*
                   1393:                 * XXX: Should we log a warning?  Should we set up a timeout
                   1394:                 * to try again in a second or so?  The latter could break
                   1395:                 * a caller's assumptions about interrupt protection, etc.
                   1396:                 */
                   1397:        }
1.21      thorpej  1398:
                   1399:        simple_unlock(&pp->pr_slock);
1.3       pk       1400: }
                   1401:
                   1402: void
1.42      thorpej  1403: pool_sethiwat(struct pool *pp, int n)
1.3       pk       1404: {
1.15      pk       1405:
1.21      thorpej  1406:        simple_lock(&pp->pr_slock);
                   1407:
1.15      pk       1408:        pp->pr_maxpages = (n == 0)
                   1409:                ? 0
1.18      thorpej  1410:                : roundup(n, pp->pr_itemsperpage) / pp->pr_itemsperpage;
1.21      thorpej  1411:
                   1412:        simple_unlock(&pp->pr_slock);
1.3       pk       1413: }
                   1414:
1.20      thorpej  1415: void
1.42      thorpej  1416: pool_sethardlimit(struct pool *pp, int n, const char *warnmess, int ratecap)
1.20      thorpej  1417: {
                   1418:
1.21      thorpej  1419:        simple_lock(&pp->pr_slock);
1.20      thorpej  1420:
                   1421:        pp->pr_hardlimit = n;
                   1422:        pp->pr_hardlimit_warning = warnmess;
1.31      thorpej  1423:        pp->pr_hardlimit_ratecap.tv_sec = ratecap;
                   1424:        pp->pr_hardlimit_warning_last.tv_sec = 0;
                   1425:        pp->pr_hardlimit_warning_last.tv_usec = 0;
1.20      thorpej  1426:
                   1427:        /*
1.21      thorpej  1428:         * In-line version of pool_sethiwat(), because we don't want to
                   1429:         * release the lock.
1.20      thorpej  1430:         */
                   1431:        pp->pr_maxpages = (n == 0)
                   1432:                ? 0
                   1433:                : roundup(n, pp->pr_itemsperpage) / pp->pr_itemsperpage;
1.21      thorpej  1434:
                   1435:        simple_unlock(&pp->pr_slock);
1.20      thorpej  1436: }
1.3       pk       1437:
                   1438: /*
                   1439:  * Release all complete pages that have not been used recently.
                   1440:  */
1.66      thorpej  1441: int
1.59      thorpej  1442: #ifdef POOL_DIAGNOSTIC
1.42      thorpej  1443: _pool_reclaim(struct pool *pp, const char *file, long line)
1.56      sommerfe 1444: #else
                   1445: pool_reclaim(struct pool *pp)
                   1446: #endif
1.3       pk       1447: {
                   1448:        struct pool_item_header *ph, *phnext;
1.43      thorpej  1449:        struct pool_cache *pc;
1.21      thorpej  1450:        struct timeval curtime;
1.61      chs      1451:        struct pool_pagelist pq;
1.88      chs      1452:        struct timeval diff;
1.21      thorpej  1453:        int s;
1.3       pk       1454:
1.68      thorpej  1455:        if (pp->pr_drain_hook != NULL) {
                   1456:                /*
                   1457:                 * The drain hook must be called with the pool unlocked.
                   1458:                 */
                   1459:                (*pp->pr_drain_hook)(pp->pr_drain_hook_arg, PR_NOWAIT);
                   1460:        }
                   1461:
1.21      thorpej  1462:        if (simple_lock_try(&pp->pr_slock) == 0)
1.66      thorpej  1463:                return (0);
1.25      thorpej  1464:        pr_enter(pp, file, line);
1.68      thorpej  1465:
1.88      chs      1466:        LIST_INIT(&pq);
1.3       pk       1467:
1.43      thorpej  1468:        /*
                   1469:         * Reclaim items from the pool's caches.
                   1470:         */
1.61      chs      1471:        TAILQ_FOREACH(pc, &pp->pr_cachelist, pc_poollist)
1.43      thorpej  1472:                pool_cache_reclaim(pc);
                   1473:
1.21      thorpej  1474:        s = splclock();
                   1475:        curtime = mono_time;
                   1476:        splx(s);
                   1477:
1.88      chs      1478:        for (ph = LIST_FIRST(&pp->pr_emptypages); ph != NULL; ph = phnext) {
                   1479:                phnext = LIST_NEXT(ph, ph_pagelist);
1.3       pk       1480:
                   1481:                /* Check our minimum page claim */
                   1482:                if (pp->pr_npages <= pp->pr_minpages)
                   1483:                        break;
                   1484:
1.88      chs      1485:                KASSERT(ph->ph_nmissing == 0);
                   1486:                timersub(&curtime, &ph->ph_time, &diff);
                   1487:                if (diff.tv_sec < pool_inactive_time)
                   1488:                        continue;
1.21      thorpej  1489:
1.88      chs      1490:                /*
                   1491:                 * If freeing this page would put us below
                   1492:                 * the low water mark, stop now.
                   1493:                 */
                   1494:                if ((pp->pr_nitems - pp->pr_itemsperpage) <
                   1495:                    pp->pr_minitems)
                   1496:                        break;
1.21      thorpej  1497:
1.88      chs      1498:                pr_rmpage(pp, ph, &pq);
1.3       pk       1499:        }
                   1500:
1.25      thorpej  1501:        pr_leave(pp);
1.21      thorpej  1502:        simple_unlock(&pp->pr_slock);
1.88      chs      1503:        if (LIST_EMPTY(&pq))
1.66      thorpej  1504:                return (0);
                   1505:
1.88      chs      1506:        while ((ph = LIST_FIRST(&pq)) != NULL) {
                   1507:                LIST_REMOVE(ph, ph_pagelist);
1.66      thorpej  1508:                pool_allocator_free(pp, ph->ph_page);
1.61      chs      1509:                if (pp->pr_roflags & PR_PHINPAGE) {
                   1510:                        continue;
                   1511:                }
1.85      pk       1512:                s = splvm();
1.97      yamt     1513:                pool_put(pp->pr_phpool, ph);
1.61      chs      1514:                splx(s);
                   1515:        }
1.66      thorpej  1516:
                   1517:        return (1);
1.3       pk       1518: }
                   1519:
                   1520: /*
                   1521:  * Drain pools, one at a time.
1.21      thorpej  1522:  *
                   1523:  * Note, we must never be called from an interrupt context.
1.3       pk       1524:  */
                   1525: void
1.42      thorpej  1526: pool_drain(void *arg)
1.3       pk       1527: {
                   1528:        struct pool *pp;
1.23      thorpej  1529:        int s;
1.3       pk       1530:
1.61      chs      1531:        pp = NULL;
1.49      thorpej  1532:        s = splvm();
1.23      thorpej  1533:        simple_lock(&pool_head_slock);
1.61      chs      1534:        if (drainpp == NULL) {
                   1535:                drainpp = TAILQ_FIRST(&pool_head);
                   1536:        }
                   1537:        if (drainpp) {
                   1538:                pp = drainpp;
                   1539:                drainpp = TAILQ_NEXT(pp, pr_poollist);
                   1540:        }
                   1541:        simple_unlock(&pool_head_slock);
1.63      chs      1542:        pool_reclaim(pp);
1.61      chs      1543:        splx(s);
1.3       pk       1544: }
                   1545:
                   1546: /*
                   1547:  * Diagnostic helpers.
                   1548:  */
                   1549: void
1.42      thorpej  1550: pool_print(struct pool *pp, const char *modif)
1.21      thorpej  1551: {
                   1552:        int s;
                   1553:
1.49      thorpej  1554:        s = splvm();
1.25      thorpej  1555:        if (simple_lock_try(&pp->pr_slock) == 0) {
                   1556:                printf("pool %s is locked; try again later\n",
                   1557:                    pp->pr_wchan);
                   1558:                splx(s);
                   1559:                return;
                   1560:        }
                   1561:        pool_print1(pp, modif, printf);
1.21      thorpej  1562:        simple_unlock(&pp->pr_slock);
                   1563:        splx(s);
                   1564: }
                   1565:
1.25      thorpej  1566: void
1.42      thorpej  1567: pool_printit(struct pool *pp, const char *modif, void (*pr)(const char *, ...))
1.25      thorpej  1568: {
                   1569:        int didlock = 0;
                   1570:
                   1571:        if (pp == NULL) {
                   1572:                (*pr)("Must specify a pool to print.\n");
                   1573:                return;
                   1574:        }
                   1575:
                   1576:        /*
                   1577:         * Called from DDB; interrupts should be blocked, and all
                   1578:         * other processors should be paused.  We can skip locking
                   1579:         * the pool in this case.
                   1580:         *
                   1581:         * We do a simple_lock_try() just to print the lock
                   1582:         * status, however.
                   1583:         */
                   1584:
                   1585:        if (simple_lock_try(&pp->pr_slock) == 0)
                   1586:                (*pr)("WARNING: pool %s is locked\n", pp->pr_wchan);
                   1587:        else
                   1588:                didlock = 1;
                   1589:
                   1590:        pool_print1(pp, modif, pr);
                   1591:
                   1592:        if (didlock)
                   1593:                simple_unlock(&pp->pr_slock);
                   1594: }
                   1595:
1.21      thorpej  1596: static void
1.97      yamt     1597: pool_print_pagelist(struct pool *pp, struct pool_pagelist *pl,
                   1598:     void (*pr)(const char *, ...))
1.88      chs      1599: {
                   1600:        struct pool_item_header *ph;
                   1601: #ifdef DIAGNOSTIC
                   1602:        struct pool_item *pi;
                   1603: #endif
                   1604:
                   1605:        LIST_FOREACH(ph, pl, ph_pagelist) {
                   1606:                (*pr)("\t\tpage %p, nmissing %d, time %lu,%lu\n",
                   1607:                    ph->ph_page, ph->ph_nmissing,
                   1608:                    (u_long)ph->ph_time.tv_sec,
                   1609:                    (u_long)ph->ph_time.tv_usec);
                   1610: #ifdef DIAGNOSTIC
1.97      yamt     1611:                if (!(pp->pr_roflags & PR_NOTOUCH)) {
                   1612:                        TAILQ_FOREACH(pi, &ph->ph_itemlist, pi_list) {
                   1613:                                if (pi->pi_magic != PI_MAGIC) {
                   1614:                                        (*pr)("\t\t\titem %p, magic 0x%x\n",
                   1615:                                            pi, pi->pi_magic);
                   1616:                                }
1.88      chs      1617:                        }
                   1618:                }
                   1619: #endif
                   1620:        }
                   1621: }
                   1622:
                   1623: static void
1.42      thorpej  1624: pool_print1(struct pool *pp, const char *modif, void (*pr)(const char *, ...))
1.3       pk       1625: {
1.25      thorpej  1626:        struct pool_item_header *ph;
1.44      thorpej  1627:        struct pool_cache *pc;
                   1628:        struct pool_cache_group *pcg;
                   1629:        int i, print_log = 0, print_pagelist = 0, print_cache = 0;
1.25      thorpej  1630:        char c;
                   1631:
                   1632:        while ((c = *modif++) != '\0') {
                   1633:                if (c == 'l')
                   1634:                        print_log = 1;
                   1635:                if (c == 'p')
                   1636:                        print_pagelist = 1;
1.44      thorpej  1637:                if (c == 'c')
                   1638:                        print_cache = 1;
1.25      thorpej  1639:        }
                   1640:
                   1641:        (*pr)("POOL %s: size %u, align %u, ioff %u, roflags 0x%08x\n",
                   1642:            pp->pr_wchan, pp->pr_size, pp->pr_align, pp->pr_itemoffset,
                   1643:            pp->pr_roflags);
1.66      thorpej  1644:        (*pr)("\talloc %p\n", pp->pr_alloc);
1.25      thorpej  1645:        (*pr)("\tminitems %u, minpages %u, maxpages %u, npages %u\n",
                   1646:            pp->pr_minitems, pp->pr_minpages, pp->pr_maxpages, pp->pr_npages);
                   1647:        (*pr)("\titemsperpage %u, nitems %u, nout %u, hardlimit %u\n",
                   1648:            pp->pr_itemsperpage, pp->pr_nitems, pp->pr_nout, pp->pr_hardlimit);
                   1649:
                   1650:        (*pr)("\n\tnget %lu, nfail %lu, nput %lu\n",
                   1651:            pp->pr_nget, pp->pr_nfail, pp->pr_nput);
                   1652:        (*pr)("\tnpagealloc %lu, npagefree %lu, hiwat %u, nidle %lu\n",
                   1653:            pp->pr_npagealloc, pp->pr_npagefree, pp->pr_hiwat, pp->pr_nidle);
                   1654:
                   1655:        if (print_pagelist == 0)
                   1656:                goto skip_pagelist;
                   1657:
1.88      chs      1658:        if ((ph = LIST_FIRST(&pp->pr_emptypages)) != NULL)
                   1659:                (*pr)("\n\tempty page list:\n");
1.97      yamt     1660:        pool_print_pagelist(pp, &pp->pr_emptypages, pr);
1.88      chs      1661:        if ((ph = LIST_FIRST(&pp->pr_fullpages)) != NULL)
                   1662:                (*pr)("\n\tfull page list:\n");
1.97      yamt     1663:        pool_print_pagelist(pp, &pp->pr_fullpages, pr);
1.88      chs      1664:        if ((ph = LIST_FIRST(&pp->pr_partpages)) != NULL)
                   1665:                (*pr)("\n\tpartial-page list:\n");
1.97      yamt     1666:        pool_print_pagelist(pp, &pp->pr_partpages, pr);
1.88      chs      1667:
1.25      thorpej  1668:        if (pp->pr_curpage == NULL)
                   1669:                (*pr)("\tno current page\n");
                   1670:        else
                   1671:                (*pr)("\tcurpage %p\n", pp->pr_curpage->ph_page);
                   1672:
                   1673:  skip_pagelist:
                   1674:        if (print_log == 0)
                   1675:                goto skip_log;
                   1676:
                   1677:        (*pr)("\n");
                   1678:        if ((pp->pr_roflags & PR_LOGGING) == 0)
                   1679:                (*pr)("\tno log\n");
                   1680:        else
                   1681:                pr_printlog(pp, NULL, pr);
1.3       pk       1682:
1.25      thorpej  1683:  skip_log:
1.44      thorpej  1684:        if (print_cache == 0)
                   1685:                goto skip_cache;
                   1686:
1.61      chs      1687:        TAILQ_FOREACH(pc, &pp->pr_cachelist, pc_poollist) {
1.44      thorpej  1688:                (*pr)("\tcache %p: allocfrom %p freeto %p\n", pc,
                   1689:                    pc->pc_allocfrom, pc->pc_freeto);
1.48      thorpej  1690:                (*pr)("\t    hits %lu misses %lu ngroups %lu nitems %lu\n",
                   1691:                    pc->pc_hits, pc->pc_misses, pc->pc_ngroups, pc->pc_nitems);
1.61      chs      1692:                TAILQ_FOREACH(pcg, &pc->pc_grouplist, pcg_list) {
1.44      thorpej  1693:                        (*pr)("\t\tgroup %p: avail %d\n", pcg, pcg->pcg_avail);
1.87      thorpej  1694:                        for (i = 0; i < PCG_NOBJECTS; i++) {
                   1695:                                if (pcg->pcg_objects[i].pcgo_pa !=
                   1696:                                    POOL_PADDR_INVALID) {
                   1697:                                        (*pr)("\t\t\t%p, 0x%llx\n",
                   1698:                                            pcg->pcg_objects[i].pcgo_va,
                   1699:                                            (unsigned long long)
                   1700:                                            pcg->pcg_objects[i].pcgo_pa);
                   1701:                                } else {
                   1702:                                        (*pr)("\t\t\t%p\n",
                   1703:                                            pcg->pcg_objects[i].pcgo_va);
                   1704:                                }
                   1705:                        }
1.44      thorpej  1706:                }
                   1707:        }
                   1708:
                   1709:  skip_cache:
1.88      chs      1710:        pr_enter_check(pp, pr);
                   1711: }
                   1712:
                   1713: static int
                   1714: pool_chk_page(struct pool *pp, const char *label, struct pool_item_header *ph)
                   1715: {
                   1716:        struct pool_item *pi;
                   1717:        caddr_t page;
                   1718:        int n;
                   1719:
                   1720:        page = (caddr_t)((u_long)ph & pp->pr_alloc->pa_pagemask);
                   1721:        if (page != ph->ph_page &&
                   1722:            (pp->pr_roflags & PR_PHINPAGE) != 0) {
                   1723:                if (label != NULL)
                   1724:                        printf("%s: ", label);
                   1725:                printf("pool(%p:%s): page inconsistency: page %p;"
                   1726:                       " at page head addr %p (p %p)\n", pp,
                   1727:                        pp->pr_wchan, ph->ph_page,
                   1728:                        ph, page);
                   1729:                return 1;
                   1730:        }
1.3       pk       1731:
1.97      yamt     1732:        if ((pp->pr_roflags & PR_NOTOUCH) != 0)
                   1733:                return 0;
                   1734:
1.88      chs      1735:        for (pi = TAILQ_FIRST(&ph->ph_itemlist), n = 0;
                   1736:             pi != NULL;
                   1737:             pi = TAILQ_NEXT(pi,pi_list), n++) {
                   1738:
                   1739: #ifdef DIAGNOSTIC
                   1740:                if (pi->pi_magic != PI_MAGIC) {
                   1741:                        if (label != NULL)
                   1742:                                printf("%s: ", label);
                   1743:                        printf("pool(%s): free list modified: magic=%x;"
                   1744:                               " page %p; item ordinal %d;"
                   1745:                               " addr %p (p %p)\n",
                   1746:                                pp->pr_wchan, pi->pi_magic, ph->ph_page,
                   1747:                                n, pi, page);
                   1748:                        panic("pool");
                   1749:                }
                   1750: #endif
                   1751:                page =
                   1752:                    (caddr_t)((u_long)pi & pp->pr_alloc->pa_pagemask);
                   1753:                if (page == ph->ph_page)
                   1754:                        continue;
                   1755:
                   1756:                if (label != NULL)
                   1757:                        printf("%s: ", label);
                   1758:                printf("pool(%p:%s): page inconsistency: page %p;"
                   1759:                       " item ordinal %d; addr %p (p %p)\n", pp,
                   1760:                        pp->pr_wchan, ph->ph_page,
                   1761:                        n, pi, page);
                   1762:                return 1;
                   1763:        }
                   1764:        return 0;
1.3       pk       1765: }
                   1766:
1.88      chs      1767:
1.3       pk       1768: int
1.42      thorpej  1769: pool_chk(struct pool *pp, const char *label)
1.3       pk       1770: {
                   1771:        struct pool_item_header *ph;
                   1772:        int r = 0;
                   1773:
1.21      thorpej  1774:        simple_lock(&pp->pr_slock);
1.88      chs      1775:        LIST_FOREACH(ph, &pp->pr_emptypages, ph_pagelist) {
                   1776:                r = pool_chk_page(pp, label, ph);
                   1777:                if (r) {
                   1778:                        goto out;
                   1779:                }
                   1780:        }
                   1781:        LIST_FOREACH(ph, &pp->pr_fullpages, ph_pagelist) {
                   1782:                r = pool_chk_page(pp, label, ph);
                   1783:                if (r) {
1.3       pk       1784:                        goto out;
                   1785:                }
1.88      chs      1786:        }
                   1787:        LIST_FOREACH(ph, &pp->pr_partpages, ph_pagelist) {
                   1788:                r = pool_chk_page(pp, label, ph);
                   1789:                if (r) {
1.3       pk       1790:                        goto out;
                   1791:                }
                   1792:        }
1.88      chs      1793:
1.3       pk       1794: out:
1.21      thorpej  1795:        simple_unlock(&pp->pr_slock);
1.3       pk       1796:        return (r);
1.43      thorpej  1797: }
                   1798:
                   1799: /*
                   1800:  * pool_cache_init:
                   1801:  *
                   1802:  *     Initialize a pool cache.
                   1803:  *
                   1804:  *     NOTE: If the pool must be protected from interrupts, we expect
                   1805:  *     to be called at the appropriate interrupt priority level.
                   1806:  */
                   1807: void
                   1808: pool_cache_init(struct pool_cache *pc, struct pool *pp,
                   1809:     int (*ctor)(void *, void *, int),
                   1810:     void (*dtor)(void *, void *),
                   1811:     void *arg)
                   1812: {
                   1813:
                   1814:        TAILQ_INIT(&pc->pc_grouplist);
                   1815:        simple_lock_init(&pc->pc_slock);
                   1816:
                   1817:        pc->pc_allocfrom = NULL;
                   1818:        pc->pc_freeto = NULL;
                   1819:        pc->pc_pool = pp;
                   1820:
                   1821:        pc->pc_ctor = ctor;
                   1822:        pc->pc_dtor = dtor;
                   1823:        pc->pc_arg  = arg;
                   1824:
1.48      thorpej  1825:        pc->pc_hits   = 0;
                   1826:        pc->pc_misses = 0;
                   1827:
                   1828:        pc->pc_ngroups = 0;
                   1829:
                   1830:        pc->pc_nitems = 0;
                   1831:
1.43      thorpej  1832:        simple_lock(&pp->pr_slock);
                   1833:        TAILQ_INSERT_TAIL(&pp->pr_cachelist, pc, pc_poollist);
                   1834:        simple_unlock(&pp->pr_slock);
                   1835: }
                   1836:
                   1837: /*
                   1838:  * pool_cache_destroy:
                   1839:  *
                   1840:  *     Destroy a pool cache.
                   1841:  */
                   1842: void
                   1843: pool_cache_destroy(struct pool_cache *pc)
                   1844: {
                   1845:        struct pool *pp = pc->pc_pool;
                   1846:
                   1847:        /* First, invalidate the entire cache. */
                   1848:        pool_cache_invalidate(pc);
                   1849:
                   1850:        /* ...and remove it from the pool's cache list. */
                   1851:        simple_lock(&pp->pr_slock);
                   1852:        TAILQ_REMOVE(&pp->pr_cachelist, pc, pc_poollist);
                   1853:        simple_unlock(&pp->pr_slock);
                   1854: }
                   1855:
                   1856: static __inline void *
1.87      thorpej  1857: pcg_get(struct pool_cache_group *pcg, paddr_t *pap)
1.43      thorpej  1858: {
                   1859:        void *object;
                   1860:        u_int idx;
                   1861:
                   1862:        KASSERT(pcg->pcg_avail <= PCG_NOBJECTS);
1.45      thorpej  1863:        KASSERT(pcg->pcg_avail != 0);
1.43      thorpej  1864:        idx = --pcg->pcg_avail;
                   1865:
1.87      thorpej  1866:        KASSERT(pcg->pcg_objects[idx].pcgo_va != NULL);
                   1867:        object = pcg->pcg_objects[idx].pcgo_va;
                   1868:        if (pap != NULL)
                   1869:                *pap = pcg->pcg_objects[idx].pcgo_pa;
                   1870:        pcg->pcg_objects[idx].pcgo_va = NULL;
1.43      thorpej  1871:
                   1872:        return (object);
                   1873: }
                   1874:
                   1875: static __inline void
1.87      thorpej  1876: pcg_put(struct pool_cache_group *pcg, void *object, paddr_t pa)
1.43      thorpej  1877: {
                   1878:        u_int idx;
                   1879:
                   1880:        KASSERT(pcg->pcg_avail < PCG_NOBJECTS);
                   1881:        idx = pcg->pcg_avail++;
                   1882:
1.87      thorpej  1883:        KASSERT(pcg->pcg_objects[idx].pcgo_va == NULL);
                   1884:        pcg->pcg_objects[idx].pcgo_va = object;
                   1885:        pcg->pcg_objects[idx].pcgo_pa = pa;
1.43      thorpej  1886: }
                   1887:
                   1888: /*
1.87      thorpej  1889:  * pool_cache_get{,_paddr}:
1.43      thorpej  1890:  *
1.87      thorpej  1891:  *     Get an object from a pool cache (optionally returning
                   1892:  *     the physical address of the object).
1.43      thorpej  1893:  */
                   1894: void *
1.87      thorpej  1895: pool_cache_get_paddr(struct pool_cache *pc, int flags, paddr_t *pap)
1.43      thorpej  1896: {
                   1897:        struct pool_cache_group *pcg;
                   1898:        void *object;
1.58      thorpej  1899:
                   1900: #ifdef LOCKDEBUG
                   1901:        if (flags & PR_WAITOK)
                   1902:                simple_lock_only_held(NULL, "pool_cache_get(PR_WAITOK)");
                   1903: #endif
1.43      thorpej  1904:
                   1905:        simple_lock(&pc->pc_slock);
                   1906:
                   1907:        if ((pcg = pc->pc_allocfrom) == NULL) {
1.61      chs      1908:                TAILQ_FOREACH(pcg, &pc->pc_grouplist, pcg_list) {
1.43      thorpej  1909:                        if (pcg->pcg_avail != 0) {
                   1910:                                pc->pc_allocfrom = pcg;
                   1911:                                goto have_group;
                   1912:                        }
                   1913:                }
                   1914:
                   1915:                /*
                   1916:                 * No groups with any available objects.  Allocate
                   1917:                 * a new object, construct it, and return it to
                   1918:                 * the caller.  We will allocate a group, if necessary,
                   1919:                 * when the object is freed back to the cache.
                   1920:                 */
1.48      thorpej  1921:                pc->pc_misses++;
1.43      thorpej  1922:                simple_unlock(&pc->pc_slock);
                   1923:                object = pool_get(pc->pc_pool, flags);
                   1924:                if (object != NULL && pc->pc_ctor != NULL) {
                   1925:                        if ((*pc->pc_ctor)(pc->pc_arg, object, flags) != 0) {
                   1926:                                pool_put(pc->pc_pool, object);
                   1927:                                return (NULL);
                   1928:                        }
                   1929:                }
1.87      thorpej  1930:                if (object != NULL && pap != NULL) {
                   1931: #ifdef POOL_VTOPHYS
                   1932:                        *pap = POOL_VTOPHYS(object);
                   1933: #else
                   1934:                        *pap = POOL_PADDR_INVALID;
                   1935: #endif
                   1936:                }
1.43      thorpej  1937:                return (object);
                   1938:        }
                   1939:
                   1940:  have_group:
1.48      thorpej  1941:        pc->pc_hits++;
                   1942:        pc->pc_nitems--;
1.87      thorpej  1943:        object = pcg_get(pcg, pap);
1.43      thorpej  1944:
                   1945:        if (pcg->pcg_avail == 0)
                   1946:                pc->pc_allocfrom = NULL;
1.45      thorpej  1947:
1.43      thorpej  1948:        simple_unlock(&pc->pc_slock);
                   1949:
                   1950:        return (object);
                   1951: }
                   1952:
                   1953: /*
1.87      thorpej  1954:  * pool_cache_put{,_paddr}:
1.43      thorpej  1955:  *
1.87      thorpej  1956:  *     Put an object back to the pool cache (optionally caching the
                   1957:  *     physical address of the object).
1.43      thorpej  1958:  */
                   1959: void
1.87      thorpej  1960: pool_cache_put_paddr(struct pool_cache *pc, void *object, paddr_t pa)
1.43      thorpej  1961: {
                   1962:        struct pool_cache_group *pcg;
1.60      thorpej  1963:        int s;
1.43      thorpej  1964:
                   1965:        simple_lock(&pc->pc_slock);
                   1966:
                   1967:        if ((pcg = pc->pc_freeto) == NULL) {
1.61      chs      1968:                TAILQ_FOREACH(pcg, &pc->pc_grouplist, pcg_list) {
1.43      thorpej  1969:                        if (pcg->pcg_avail != PCG_NOBJECTS) {
                   1970:                                pc->pc_freeto = pcg;
                   1971:                                goto have_group;
                   1972:                        }
                   1973:                }
                   1974:
                   1975:                /*
                   1976:                 * No empty groups to free the object to.  Attempt to
1.47      thorpej  1977:                 * allocate one.
1.43      thorpej  1978:                 */
1.47      thorpej  1979:                simple_unlock(&pc->pc_slock);
1.60      thorpej  1980:                s = splvm();
1.43      thorpej  1981:                pcg = pool_get(&pcgpool, PR_NOWAIT);
1.60      thorpej  1982:                splx(s);
1.43      thorpej  1983:                if (pcg != NULL) {
                   1984:                        memset(pcg, 0, sizeof(*pcg));
1.47      thorpej  1985:                        simple_lock(&pc->pc_slock);
1.48      thorpej  1986:                        pc->pc_ngroups++;
1.43      thorpej  1987:                        TAILQ_INSERT_TAIL(&pc->pc_grouplist, pcg, pcg_list);
1.47      thorpej  1988:                        if (pc->pc_freeto == NULL)
                   1989:                                pc->pc_freeto = pcg;
1.43      thorpej  1990:                        goto have_group;
                   1991:                }
                   1992:
                   1993:                /*
                   1994:                 * Unable to allocate a cache group; destruct the object
                   1995:                 * and free it back to the pool.
                   1996:                 */
1.51      thorpej  1997:                pool_cache_destruct_object(pc, object);
1.43      thorpej  1998:                return;
                   1999:        }
                   2000:
                   2001:  have_group:
1.48      thorpej  2002:        pc->pc_nitems++;
1.87      thorpej  2003:        pcg_put(pcg, object, pa);
1.43      thorpej  2004:
                   2005:        if (pcg->pcg_avail == PCG_NOBJECTS)
                   2006:                pc->pc_freeto = NULL;
                   2007:
                   2008:        simple_unlock(&pc->pc_slock);
1.51      thorpej  2009: }
                   2010:
                   2011: /*
                   2012:  * pool_cache_destruct_object:
                   2013:  *
                   2014:  *     Force destruction of an object and its release back into
                   2015:  *     the pool.
                   2016:  */
                   2017: void
                   2018: pool_cache_destruct_object(struct pool_cache *pc, void *object)
                   2019: {
                   2020:
                   2021:        if (pc->pc_dtor != NULL)
                   2022:                (*pc->pc_dtor)(pc->pc_arg, object);
                   2023:        pool_put(pc->pc_pool, object);
1.43      thorpej  2024: }
                   2025:
                   2026: /*
                   2027:  * pool_cache_do_invalidate:
                   2028:  *
                   2029:  *     This internal function implements pool_cache_invalidate() and
                   2030:  *     pool_cache_reclaim().
                   2031:  */
                   2032: static void
                   2033: pool_cache_do_invalidate(struct pool_cache *pc, int free_groups,
1.56      sommerfe 2034:     void (*putit)(struct pool *, void *))
1.43      thorpej  2035: {
                   2036:        struct pool_cache_group *pcg, *npcg;
                   2037:        void *object;
1.60      thorpej  2038:        int s;
1.43      thorpej  2039:
                   2040:        for (pcg = TAILQ_FIRST(&pc->pc_grouplist); pcg != NULL;
                   2041:             pcg = npcg) {
                   2042:                npcg = TAILQ_NEXT(pcg, pcg_list);
                   2043:                while (pcg->pcg_avail != 0) {
1.48      thorpej  2044:                        pc->pc_nitems--;
1.87      thorpej  2045:                        object = pcg_get(pcg, NULL);
1.45      thorpej  2046:                        if (pcg->pcg_avail == 0 && pc->pc_allocfrom == pcg)
                   2047:                                pc->pc_allocfrom = NULL;
1.43      thorpej  2048:                        if (pc->pc_dtor != NULL)
                   2049:                                (*pc->pc_dtor)(pc->pc_arg, object);
1.56      sommerfe 2050:                        (*putit)(pc->pc_pool, object);
1.43      thorpej  2051:                }
                   2052:                if (free_groups) {
1.48      thorpej  2053:                        pc->pc_ngroups--;
1.43      thorpej  2054:                        TAILQ_REMOVE(&pc->pc_grouplist, pcg, pcg_list);
1.46      thorpej  2055:                        if (pc->pc_freeto == pcg)
                   2056:                                pc->pc_freeto = NULL;
1.60      thorpej  2057:                        s = splvm();
1.43      thorpej  2058:                        pool_put(&pcgpool, pcg);
1.60      thorpej  2059:                        splx(s);
1.43      thorpej  2060:                }
                   2061:        }
                   2062: }
                   2063:
                   2064: /*
                   2065:  * pool_cache_invalidate:
                   2066:  *
                   2067:  *     Invalidate a pool cache (destruct and release all of the
                   2068:  *     cached objects).
                   2069:  */
                   2070: void
                   2071: pool_cache_invalidate(struct pool_cache *pc)
                   2072: {
                   2073:
                   2074:        simple_lock(&pc->pc_slock);
1.56      sommerfe 2075:        pool_cache_do_invalidate(pc, 0, pool_put);
1.43      thorpej  2076:        simple_unlock(&pc->pc_slock);
                   2077: }
                   2078:
                   2079: /*
                   2080:  * pool_cache_reclaim:
                   2081:  *
                   2082:  *     Reclaim a pool cache for pool_reclaim().
                   2083:  */
                   2084: static void
                   2085: pool_cache_reclaim(struct pool_cache *pc)
                   2086: {
                   2087:
1.47      thorpej  2088:        simple_lock(&pc->pc_slock);
1.43      thorpej  2089:        pool_cache_do_invalidate(pc, 1, pool_do_put);
                   2090:        simple_unlock(&pc->pc_slock);
1.3       pk       2091: }
1.66      thorpej  2092:
                   2093: /*
                   2094:  * Pool backend allocators.
                   2095:  *
                   2096:  * Each pool has a backend allocator that handles allocation, deallocation,
                   2097:  * and any additional draining that might be needed.
                   2098:  *
                   2099:  * We provide two standard allocators:
                   2100:  *
                   2101:  *     pool_allocator_kmem - the default when no allocator is specified
                   2102:  *
                   2103:  *     pool_allocator_nointr - used for pools that will not be accessed
                   2104:  *     in interrupt context.
                   2105:  */
                   2106: void   *pool_page_alloc(struct pool *, int);
                   2107: void   pool_page_free(struct pool *, void *);
                   2108:
                   2109: struct pool_allocator pool_allocator_kmem = {
                   2110:        pool_page_alloc, pool_page_free, 0,
                   2111: };
                   2112:
                   2113: void   *pool_page_alloc_nointr(struct pool *, int);
                   2114: void   pool_page_free_nointr(struct pool *, void *);
                   2115:
                   2116: struct pool_allocator pool_allocator_nointr = {
                   2117:        pool_page_alloc_nointr, pool_page_free_nointr, 0,
                   2118: };
                   2119:
                   2120: #ifdef POOL_SUBPAGE
                   2121: void   *pool_subpage_alloc(struct pool *, int);
                   2122: void   pool_subpage_free(struct pool *, void *);
                   2123:
                   2124: struct pool_allocator pool_allocator_kmem_subpage = {
                   2125:        pool_subpage_alloc, pool_subpage_free, 0,
                   2126: };
                   2127: #endif /* POOL_SUBPAGE */
                   2128:
                   2129: /*
                   2130:  * We have at least three different resources for the same allocation and
                   2131:  * each resource can be depleted.  First, we have the ready elements in the
                   2132:  * pool.  Then we have the resource (typically a vm_map) for this allocator.
                   2133:  * Finally, we have physical memory.  Waiting for any of these can be
                   2134:  * unnecessary when any other is freed, but the kernel doesn't support
                   2135:  * sleeping on multiple wait channels, so we have to employ another strategy.
                   2136:  *
                   2137:  * The caller sleeps on the pool (so that it can be awakened when an item
                   2138:  * is returned to the pool), but we set PA_WANT on the allocator.  When a
                   2139:  * page is returned to the allocator and PA_WANT is set, pool_allocator_free
                   2140:  * will wake up all sleeping pools belonging to this allocator.
                   2141:  *
                   2142:  * XXX Thundering herd.
                   2143:  */
                   2144: void *
                   2145: pool_allocator_alloc(struct pool *org, int flags)
                   2146: {
                   2147:        struct pool_allocator *pa = org->pr_alloc;
                   2148:        struct pool *pp, *start;
                   2149:        int s, freed;
                   2150:        void *res;
                   2151:
1.91      yamt     2152:        LOCK_ASSERT(!simple_lock_held(&org->pr_slock));
                   2153:
1.66      thorpej  2154:        do {
                   2155:                if ((res = (*pa->pa_alloc)(org, flags)) != NULL)
                   2156:                        return (res);
1.68      thorpej  2157:                if ((flags & PR_WAITOK) == 0) {
                   2158:                        /*
                   2159:                         * We only run the drain hookhere if PR_NOWAIT.
                   2160:                         * In other cases, the hook will be run in
                   2161:                         * pool_reclaim().
                   2162:                         */
                   2163:                        if (org->pr_drain_hook != NULL) {
                   2164:                                (*org->pr_drain_hook)(org->pr_drain_hook_arg,
                   2165:                                    flags);
                   2166:                                if ((res = (*pa->pa_alloc)(org, flags)) != NULL)
                   2167:                                        return (res);
                   2168:                        }
1.66      thorpej  2169:                        break;
1.68      thorpej  2170:                }
1.66      thorpej  2171:
                   2172:                /*
                   2173:                 * Drain all pools, except "org", that use this
                   2174:                 * allocator.  We do this to reclaim VA space.
                   2175:                 * pa_alloc is responsible for waiting for
                   2176:                 * physical memory.
                   2177:                 *
                   2178:                 * XXX We risk looping forever if start if someone
                   2179:                 * calls pool_destroy on "start".  But there is no
                   2180:                 * other way to have potentially sleeping pool_reclaim,
                   2181:                 * non-sleeping locks on pool_allocator, and some
                   2182:                 * stirring of drained pools in the allocator.
1.68      thorpej  2183:                 *
                   2184:                 * XXX Maybe we should use pool_head_slock for locking
                   2185:                 * the allocators?
1.66      thorpej  2186:                 */
                   2187:                freed = 0;
                   2188:
                   2189:                s = splvm();
                   2190:                simple_lock(&pa->pa_slock);
                   2191:                pp = start = TAILQ_FIRST(&pa->pa_list);
                   2192:                do {
                   2193:                        TAILQ_REMOVE(&pa->pa_list, pp, pr_alloc_list);
                   2194:                        TAILQ_INSERT_TAIL(&pa->pa_list, pp, pr_alloc_list);
                   2195:                        if (pp == org)
                   2196:                                continue;
1.73      thorpej  2197:                        simple_unlock(&pa->pa_slock);
1.66      thorpej  2198:                        freed = pool_reclaim(pp);
1.73      thorpej  2199:                        simple_lock(&pa->pa_slock);
1.66      thorpej  2200:                } while ((pp = TAILQ_FIRST(&pa->pa_list)) != start &&
                   2201:                         freed == 0);
                   2202:
                   2203:                if (freed == 0) {
                   2204:                        /*
                   2205:                         * We set PA_WANT here, the caller will most likely
                   2206:                         * sleep waiting for pages (if not, this won't hurt
                   2207:                         * that much), and there is no way to set this in
                   2208:                         * the caller without violating locking order.
                   2209:                         */
                   2210:                        pa->pa_flags |= PA_WANT;
                   2211:                }
                   2212:                simple_unlock(&pa->pa_slock);
                   2213:                splx(s);
                   2214:        } while (freed);
                   2215:        return (NULL);
                   2216: }
                   2217:
                   2218: void
                   2219: pool_allocator_free(struct pool *pp, void *v)
                   2220: {
                   2221:        struct pool_allocator *pa = pp->pr_alloc;
                   2222:        int s;
                   2223:
1.91      yamt     2224:        LOCK_ASSERT(!simple_lock_held(&pp->pr_slock));
                   2225:
1.66      thorpej  2226:        (*pa->pa_free)(pp, v);
                   2227:
                   2228:        s = splvm();
                   2229:        simple_lock(&pa->pa_slock);
                   2230:        if ((pa->pa_flags & PA_WANT) == 0) {
                   2231:                simple_unlock(&pa->pa_slock);
                   2232:                splx(s);
                   2233:                return;
                   2234:        }
                   2235:
                   2236:        TAILQ_FOREACH(pp, &pa->pa_list, pr_alloc_list) {
                   2237:                simple_lock(&pp->pr_slock);
                   2238:                if ((pp->pr_flags & PR_WANTED) != 0) {
                   2239:                        pp->pr_flags &= ~PR_WANTED;
                   2240:                        wakeup(pp);
                   2241:                }
1.69      thorpej  2242:                simple_unlock(&pp->pr_slock);
1.66      thorpej  2243:        }
                   2244:        pa->pa_flags &= ~PA_WANT;
                   2245:        simple_unlock(&pa->pa_slock);
                   2246:        splx(s);
                   2247: }
                   2248:
                   2249: void *
                   2250: pool_page_alloc(struct pool *pp, int flags)
                   2251: {
                   2252:        boolean_t waitok = (flags & PR_WAITOK) ? TRUE : FALSE;
                   2253:
1.99.2.1! kent     2254:        return ((void *) uvm_km_alloc_poolpage_cache(kmem_map, waitok));
1.66      thorpej  2255: }
                   2256:
                   2257: void
                   2258: pool_page_free(struct pool *pp, void *v)
                   2259: {
                   2260:
1.98      yamt     2261:        uvm_km_free_poolpage_cache(kmem_map, (vaddr_t) v);
                   2262: }
                   2263:
                   2264: static void *
                   2265: pool_page_alloc_meta(struct pool *pp, int flags)
                   2266: {
                   2267:        boolean_t waitok = (flags & PR_WAITOK) ? TRUE : FALSE;
                   2268:
1.99.2.1! kent     2269:        return ((void *) uvm_km_alloc_poolpage(kmem_map, waitok));
1.98      yamt     2270: }
                   2271:
                   2272: static void
                   2273: pool_page_free_meta(struct pool *pp, void *v)
                   2274: {
                   2275:
1.99.2.1! kent     2276:        uvm_km_free_poolpage(kmem_map, (vaddr_t) v);
1.66      thorpej  2277: }
                   2278:
                   2279: #ifdef POOL_SUBPAGE
                   2280: /* Sub-page allocator, for machines with large hardware pages. */
                   2281: void *
                   2282: pool_subpage_alloc(struct pool *pp, int flags)
                   2283: {
1.93      dbj      2284:        void *v;
                   2285:        int s;
                   2286:        s = splvm();
                   2287:        v = pool_get(&psppool, flags);
                   2288:        splx(s);
                   2289:        return v;
1.66      thorpej  2290: }
                   2291:
                   2292: void
                   2293: pool_subpage_free(struct pool *pp, void *v)
                   2294: {
1.93      dbj      2295:        int s;
                   2296:        s = splvm();
1.66      thorpej  2297:        pool_put(&psppool, v);
1.93      dbj      2298:        splx(s);
1.66      thorpej  2299: }
                   2300:
                   2301: /* We don't provide a real nointr allocator.  Maybe later. */
                   2302: void *
                   2303: pool_page_alloc_nointr(struct pool *pp, int flags)
                   2304: {
                   2305:
                   2306:        return (pool_subpage_alloc(pp, flags));
                   2307: }
                   2308:
                   2309: void
                   2310: pool_page_free_nointr(struct pool *pp, void *v)
                   2311: {
                   2312:
                   2313:        pool_subpage_free(pp, v);
                   2314: }
                   2315: #else
                   2316: void *
                   2317: pool_page_alloc_nointr(struct pool *pp, int flags)
                   2318: {
                   2319:        boolean_t waitok = (flags & PR_WAITOK) ? TRUE : FALSE;
                   2320:
1.99.2.1! kent     2321:        return ((void *) uvm_km_alloc_poolpage_cache(kernel_map, waitok));
1.66      thorpej  2322: }
                   2323:
                   2324: void
                   2325: pool_page_free_nointr(struct pool *pp, void *v)
                   2326: {
                   2327:
1.98      yamt     2328:        uvm_km_free_poolpage_cache(kernel_map, (vaddr_t) v);
1.66      thorpej  2329: }
                   2330: #endif /* POOL_SUBPAGE */

CVSweb <webmaster@jp.NetBSD.org>