[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 1.101

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

CVSweb <webmaster@jp.NetBSD.org>