[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 1.131.2.2

1.131.2.2! matt        1: /*     $NetBSD$        */
1.1       pk          2:
                      3: /*-
1.131.2.2! matt        4:  * Copyright (c) 1997, 1999, 2000, 2002, 2007 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
1.131.2.2! matt        9:  * Simulation Facility, NASA Ames Research Center, and by Andrew Doran.
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>
1.131.2.2! matt       41: __KERNEL_RCSID(0, "$NetBSD$");
1.24      scottr     42:
1.25      thorpej    43: #include "opt_pool.h"
1.24      scottr     44: #include "opt_poollog.h"
1.28      thorpej    45: #include "opt_lockdebug.h"
1.1       pk         46:
                     47: #include <sys/param.h>
                     48: #include <sys/systm.h>
                     49: #include <sys/proc.h>
                     50: #include <sys/errno.h>
                     51: #include <sys/kernel.h>
                     52: #include <sys/malloc.h>
                     53: #include <sys/lock.h>
                     54: #include <sys/pool.h>
1.20      thorpej    55: #include <sys/syslog.h>
1.125     ad         56: #include <sys/debug.h>
1.131.2.2! matt       57: #include <sys/lockdebug.h>
        !            58: #include <sys/xcall.h>
        !            59: #include <sys/cpu.h>
1.3       pk         60:
                     61: #include <uvm/uvm.h>
                     62:
1.1       pk         63: /*
                     64:  * Pool resource management utility.
1.3       pk         65:  *
1.88      chs        66:  * Memory is allocated in pages which are split into pieces according to
                     67:  * the pool item size. Each page is kept on one of three lists in the
                     68:  * pool structure: `pr_emptypages', `pr_fullpages' and `pr_partpages',
                     69:  * for empty, full and partially-full pages respectively. The individual
                     70:  * pool items are on a linked list headed by `ph_itemlist' in each page
                     71:  * header. The memory for building the page list is either taken from
                     72:  * the allocated pages themselves (for small pool items) or taken from
                     73:  * an internal pool of page headers (`phpool').
1.1       pk         74:  */
                     75:
1.3       pk         76: /* List of all pools */
1.102     chs        77: LIST_HEAD(,pool) pool_head = LIST_HEAD_INITIALIZER(pool_head);
1.3       pk         78:
1.131.2.2! matt       79: /* List of all caches. */
        !            80: LIST_HEAD(,pool_cache) pool_cache_head =
        !            81:     LIST_HEAD_INITIALIZER(pool_cache_head);
        !            82:
1.3       pk         83: /* Private pool for page header structures */
1.97      yamt       84: #define        PHPOOL_MAX      8
                     85: static struct pool phpool[PHPOOL_MAX];
                     86: #define        PHPOOL_FREELIST_NELEM(idx)      (((idx) == 0) ? 0 : (1 << (idx)))
1.3       pk         87:
1.62      bjh21      88: #ifdef POOL_SUBPAGE
                     89: /* Pool of subpages for use by normal pools. */
                     90: static struct pool psppool;
                     91: #endif
                     92:
1.117     yamt       93: static SLIST_HEAD(, pool_allocator) pa_deferinitq =
                     94:     SLIST_HEAD_INITIALIZER(pa_deferinitq);
                     95:
1.98      yamt       96: static void *pool_page_alloc_meta(struct pool *, int);
                     97: static void pool_page_free_meta(struct pool *, void *);
                     98:
                     99: /* allocator for pool metadata */
1.131.2.2! matt      100: struct pool_allocator pool_allocator_meta = {
1.117     yamt      101:        pool_page_alloc_meta, pool_page_free_meta,
                    102:        .pa_backingmapptr = &kmem_map,
1.98      yamt      103: };
                    104:
1.3       pk        105: /* # of seconds to retain page after last use */
                    106: int pool_inactive_time = 10;
                    107:
                    108: /* Next candidate for drainage (see pool_drain()) */
1.23      thorpej   109: static struct pool     *drainpp;
                    110:
1.131.2.2! matt      111: /* This lock protects both pool_head and drainpp. */
        !           112: static kmutex_t pool_head_lock;
        !           113: static kcondvar_t pool_busy;
1.3       pk        114:
1.99      yamt      115: typedef uint8_t pool_item_freelist_t;
                    116:
1.3       pk        117: struct pool_item_header {
                    118:        /* Page headers */
1.88      chs       119:        LIST_ENTRY(pool_item_header)
1.3       pk        120:                                ph_pagelist;    /* pool page list */
1.88      chs       121:        SPLAY_ENTRY(pool_item_header)
                    122:                                ph_node;        /* Off-page page headers */
1.128     christos  123:        void *                  ph_page;        /* this page's address */
1.3       pk        124:        struct timeval          ph_time;        /* last referenced */
1.97      yamt      125:        union {
                    126:                /* !PR_NOTOUCH */
                    127:                struct {
1.102     chs       128:                        LIST_HEAD(, pool_item)
1.97      yamt      129:                                phu_itemlist;   /* chunk list for this page */
                    130:                } phu_normal;
                    131:                /* PR_NOTOUCH */
                    132:                struct {
                    133:                        uint16_t
                    134:                                phu_off;        /* start offset in page */
1.99      yamt      135:                        pool_item_freelist_t
1.97      yamt      136:                                phu_firstfree;  /* first free item */
1.99      yamt      137:                        /*
                    138:                         * XXX it might be better to use
                    139:                         * a simple bitmap and ffs(3)
                    140:                         */
1.97      yamt      141:                } phu_notouch;
                    142:        } ph_u;
                    143:        uint16_t                ph_nmissing;    /* # of chunks in use */
1.3       pk        144: };
1.97      yamt      145: #define        ph_itemlist     ph_u.phu_normal.phu_itemlist
                    146: #define        ph_off          ph_u.phu_notouch.phu_off
                    147: #define        ph_firstfree    ph_u.phu_notouch.phu_firstfree
1.3       pk        148:
1.1       pk        149: struct pool_item {
1.3       pk        150: #ifdef DIAGNOSTIC
1.82      thorpej   151:        u_int pi_magic;
1.33      chs       152: #endif
1.131.2.2! matt      153: #define        PI_MAGIC 0xdeaddeadU
1.3       pk        154:        /* Other entries use only this list entry */
1.102     chs       155:        LIST_ENTRY(pool_item)   pi_list;
1.3       pk        156: };
                    157:
1.53      thorpej   158: #define        POOL_NEEDS_CATCHUP(pp)                                          \
                    159:        ((pp)->pr_nitems < (pp)->pr_minitems)
                    160:
1.43      thorpej   161: /*
                    162:  * Pool cache management.
                    163:  *
                    164:  * Pool caches provide a way for constructed objects to be cached by the
                    165:  * pool subsystem.  This can lead to performance improvements by avoiding
                    166:  * needless object construction/destruction; it is deferred until absolutely
                    167:  * necessary.
                    168:  *
1.131.2.2! matt      169:  * Caches are grouped into cache groups.  Each cache group references up
        !           170:  * to PCG_NUMOBJECTS constructed objects.  When a cache allocates an
        !           171:  * object from the pool, it calls the object's constructor and places it
        !           172:  * into a cache group.  When a cache group frees an object back to the
        !           173:  * pool, it first calls the object's destructor.  This allows the object
        !           174:  * to persist in constructed form while freed to the cache.
        !           175:  *
        !           176:  * The pool references each cache, so that when a pool is drained by the
        !           177:  * pagedaemon, it can drain each individual cache as well.  Each time a
        !           178:  * cache is drained, the most idle cache group is freed to the pool in
        !           179:  * its entirety.
1.43      thorpej   180:  *
                    181:  * Pool caches are layed on top of pools.  By layering them, we can avoid
                    182:  * the complexity of cache management for pools which would not benefit
                    183:  * from it.
                    184:  */
                    185:
                    186: static struct pool pcgpool;
1.131.2.2! matt      187: static struct pool cache_pool;
        !           188: static struct pool cache_cpu_pool;
1.3       pk        189:
1.131.2.2! matt      190: static pool_cache_cpu_t *pool_cache_put_slow(pool_cache_cpu_t *, int *,
        !           191:                                             void *, paddr_t);
        !           192: static pool_cache_cpu_t *pool_cache_get_slow(pool_cache_cpu_t *, int *,
        !           193:                                             void **, paddr_t *, int);
        !           194: static void    pool_cache_cpu_init1(struct cpu_info *, pool_cache_t);
        !           195: static void    pool_cache_invalidate_groups(pool_cache_t, pcg_t *);
        !           196: static void    pool_cache_xcall(pool_cache_t);
1.3       pk        197:
1.42      thorpej   198: static int     pool_catchup(struct pool *);
1.128     christos  199: static void    pool_prime_page(struct pool *, void *,
1.55      thorpej   200:                    struct pool_item_header *);
1.88      chs       201: static void    pool_update_curpage(struct pool *);
1.66      thorpej   202:
1.113     yamt      203: static int     pool_grow(struct pool *, int);
1.117     yamt      204: static void    *pool_allocator_alloc(struct pool *, int);
                    205: static void    pool_allocator_free(struct pool *, void *);
1.3       pk        206:
1.97      yamt      207: static void pool_print_pagelist(struct pool *, struct pool_pagelist *,
1.88      chs       208:        void (*)(const char *, ...));
1.42      thorpej   209: static void pool_print1(struct pool *, const char *,
                    210:        void (*)(const char *, ...));
1.3       pk        211:
1.88      chs       212: static int pool_chk_page(struct pool *, const char *,
                    213:                         struct pool_item_header *);
                    214:
1.3       pk        215: /*
1.52      thorpej   216:  * Pool log entry. An array of these is allocated in pool_init().
1.3       pk        217:  */
                    218: struct pool_log {
                    219:        const char      *pl_file;
                    220:        long            pl_line;
                    221:        int             pl_action;
1.25      thorpej   222: #define        PRLOG_GET       1
                    223: #define        PRLOG_PUT       2
1.3       pk        224:        void            *pl_addr;
1.1       pk        225: };
                    226:
1.86      matt      227: #ifdef POOL_DIAGNOSTIC
1.3       pk        228: /* Number of entries in pool log buffers */
1.17      thorpej   229: #ifndef POOL_LOGSIZE
                    230: #define        POOL_LOGSIZE    10
                    231: #endif
                    232:
                    233: int pool_logsize = POOL_LOGSIZE;
1.1       pk        234:
1.110     perry     235: static inline void
1.42      thorpej   236: pr_log(struct pool *pp, void *v, int action, const char *file, long line)
1.3       pk        237: {
                    238:        int n = pp->pr_curlogentry;
                    239:        struct pool_log *pl;
                    240:
1.20      thorpej   241:        if ((pp->pr_roflags & PR_LOGGING) == 0)
1.3       pk        242:                return;
                    243:
                    244:        /*
                    245:         * Fill in the current entry. Wrap around and overwrite
                    246:         * the oldest entry if necessary.
                    247:         */
                    248:        pl = &pp->pr_log[n];
                    249:        pl->pl_file = file;
                    250:        pl->pl_line = line;
                    251:        pl->pl_action = action;
                    252:        pl->pl_addr = v;
                    253:        if (++n >= pp->pr_logsize)
                    254:                n = 0;
                    255:        pp->pr_curlogentry = n;
                    256: }
                    257:
                    258: static void
1.42      thorpej   259: pr_printlog(struct pool *pp, struct pool_item *pi,
                    260:     void (*pr)(const char *, ...))
1.3       pk        261: {
                    262:        int i = pp->pr_logsize;
                    263:        int n = pp->pr_curlogentry;
                    264:
1.20      thorpej   265:        if ((pp->pr_roflags & PR_LOGGING) == 0)
1.3       pk        266:                return;
                    267:
                    268:        /*
                    269:         * Print all entries in this pool's log.
                    270:         */
                    271:        while (i-- > 0) {
                    272:                struct pool_log *pl = &pp->pr_log[n];
                    273:                if (pl->pl_action != 0) {
1.25      thorpej   274:                        if (pi == NULL || pi == pl->pl_addr) {
                    275:                                (*pr)("\tlog entry %d:\n", i);
                    276:                                (*pr)("\t\taction = %s, addr = %p\n",
                    277:                                    pl->pl_action == PRLOG_GET ? "get" : "put",
                    278:                                    pl->pl_addr);
                    279:                                (*pr)("\t\tfile: %s at line %lu\n",
                    280:                                    pl->pl_file, pl->pl_line);
                    281:                        }
1.3       pk        282:                }
                    283:                if (++n >= pp->pr_logsize)
                    284:                        n = 0;
                    285:        }
                    286: }
1.25      thorpej   287:
1.110     perry     288: static inline void
1.42      thorpej   289: pr_enter(struct pool *pp, const char *file, long line)
1.25      thorpej   290: {
                    291:
1.34      thorpej   292:        if (__predict_false(pp->pr_entered_file != NULL)) {
1.25      thorpej   293:                printf("pool %s: reentrancy at file %s line %ld\n",
                    294:                    pp->pr_wchan, file, line);
                    295:                printf("         previous entry at file %s line %ld\n",
                    296:                    pp->pr_entered_file, pp->pr_entered_line);
                    297:                panic("pr_enter");
                    298:        }
                    299:
                    300:        pp->pr_entered_file = file;
                    301:        pp->pr_entered_line = line;
                    302: }
                    303:
1.110     perry     304: static inline void
1.42      thorpej   305: pr_leave(struct pool *pp)
1.25      thorpej   306: {
                    307:
1.34      thorpej   308:        if (__predict_false(pp->pr_entered_file == NULL)) {
1.25      thorpej   309:                printf("pool %s not entered?\n", pp->pr_wchan);
                    310:                panic("pr_leave");
                    311:        }
                    312:
                    313:        pp->pr_entered_file = NULL;
                    314:        pp->pr_entered_line = 0;
                    315: }
                    316:
1.110     perry     317: static inline void
1.42      thorpej   318: pr_enter_check(struct pool *pp, void (*pr)(const char *, ...))
1.25      thorpej   319: {
                    320:
                    321:        if (pp->pr_entered_file != NULL)
                    322:                (*pr)("\n\tcurrently entered from file %s line %ld\n",
                    323:                    pp->pr_entered_file, pp->pr_entered_line);
                    324: }
1.3       pk        325: #else
1.25      thorpej   326: #define        pr_log(pp, v, action, file, line)
                    327: #define        pr_printlog(pp, pi, pr)
                    328: #define        pr_enter(pp, file, line)
                    329: #define        pr_leave(pp)
                    330: #define        pr_enter_check(pp, pr)
1.59      thorpej   331: #endif /* POOL_DIAGNOSTIC */
1.3       pk        332:
1.110     perry     333: static inline int
1.97      yamt      334: pr_item_notouch_index(const struct pool *pp, const struct pool_item_header *ph,
                    335:     const void *v)
                    336: {
                    337:        const char *cp = v;
                    338:        int idx;
                    339:
                    340:        KASSERT(pp->pr_roflags & PR_NOTOUCH);
1.128     christos  341:        idx = (cp - (char *)ph->ph_page - ph->ph_off) / pp->pr_size;
1.97      yamt      342:        KASSERT(idx < pp->pr_itemsperpage);
                    343:        return idx;
                    344: }
                    345:
1.99      yamt      346: #define        PR_FREELIST_ALIGN(p) \
                    347:        roundup((uintptr_t)(p), sizeof(pool_item_freelist_t))
                    348: #define        PR_FREELIST(ph) ((pool_item_freelist_t *)PR_FREELIST_ALIGN((ph) + 1))
                    349: #define        PR_INDEX_USED   ((pool_item_freelist_t)-1)
                    350: #define        PR_INDEX_EOL    ((pool_item_freelist_t)-2)
1.97      yamt      351:
1.110     perry     352: static inline void
1.97      yamt      353: pr_item_notouch_put(const struct pool *pp, struct pool_item_header *ph,
                    354:     void *obj)
                    355: {
                    356:        int idx = pr_item_notouch_index(pp, ph, obj);
1.99      yamt      357:        pool_item_freelist_t *freelist = PR_FREELIST(ph);
1.97      yamt      358:
                    359:        KASSERT(freelist[idx] == PR_INDEX_USED);
                    360:        freelist[idx] = ph->ph_firstfree;
                    361:        ph->ph_firstfree = idx;
                    362: }
                    363:
1.110     perry     364: static inline void *
1.97      yamt      365: pr_item_notouch_get(const struct pool *pp, struct pool_item_header *ph)
                    366: {
                    367:        int idx = ph->ph_firstfree;
1.99      yamt      368:        pool_item_freelist_t *freelist = PR_FREELIST(ph);
1.97      yamt      369:
                    370:        KASSERT(freelist[idx] != PR_INDEX_USED);
                    371:        ph->ph_firstfree = freelist[idx];
                    372:        freelist[idx] = PR_INDEX_USED;
                    373:
1.128     christos  374:        return (char *)ph->ph_page + ph->ph_off + idx * pp->pr_size;
1.97      yamt      375: }
                    376:
1.110     perry     377: static inline int
1.88      chs       378: phtree_compare(struct pool_item_header *a, struct pool_item_header *b)
                    379: {
1.121     yamt      380:
                    381:        /*
                    382:         * we consider pool_item_header with smaller ph_page bigger.
                    383:         * (this unnatural ordering is for the benefit of pr_find_pagehead.)
                    384:         */
                    385:
1.88      chs       386:        if (a->ph_page < b->ph_page)
1.121     yamt      387:                return (1);
                    388:        else if (a->ph_page > b->ph_page)
1.88      chs       389:                return (-1);
                    390:        else
                    391:                return (0);
                    392: }
                    393:
                    394: SPLAY_PROTOTYPE(phtree, pool_item_header, ph_node, phtree_compare);
                    395: SPLAY_GENERATE(phtree, pool_item_header, ph_node, phtree_compare);
                    396:
1.3       pk        397: /*
1.121     yamt      398:  * Return the pool page header based on item address.
1.3       pk        399:  */
1.110     perry     400: static inline struct pool_item_header *
1.121     yamt      401: pr_find_pagehead(struct pool *pp, void *v)
1.3       pk        402: {
1.88      chs       403:        struct pool_item_header *ph, tmp;
1.3       pk        404:
1.121     yamt      405:        if ((pp->pr_roflags & PR_NOALIGN) != 0) {
1.128     christos  406:                tmp.ph_page = (void *)(uintptr_t)v;
1.121     yamt      407:                ph = SPLAY_FIND(phtree, &pp->pr_phtree, &tmp);
                    408:                if (ph == NULL) {
                    409:                        ph = SPLAY_ROOT(&pp->pr_phtree);
                    410:                        if (ph != NULL && phtree_compare(&tmp, ph) >= 0) {
                    411:                                ph = SPLAY_NEXT(phtree, &pp->pr_phtree, ph);
                    412:                        }
                    413:                        KASSERT(ph == NULL || phtree_compare(&tmp, ph) < 0);
                    414:                }
                    415:        } else {
1.128     christos  416:                void *page =
                    417:                    (void *)((uintptr_t)v & pp->pr_alloc->pa_pagemask);
1.121     yamt      418:
                    419:                if ((pp->pr_roflags & PR_PHINPAGE) != 0) {
1.128     christos  420:                        ph = (struct pool_item_header *)((char *)page + pp->pr_phoffset);
1.121     yamt      421:                } else {
                    422:                        tmp.ph_page = page;
                    423:                        ph = SPLAY_FIND(phtree, &pp->pr_phtree, &tmp);
                    424:                }
                    425:        }
1.3       pk        426:
1.121     yamt      427:        KASSERT(ph == NULL || ((pp->pr_roflags & PR_PHINPAGE) != 0) ||
1.128     christos  428:            ((char *)ph->ph_page <= (char *)v &&
                    429:            (char *)v < (char *)ph->ph_page + pp->pr_alloc->pa_pagesz));
1.88      chs       430:        return ph;
1.3       pk        431: }
                    432:
1.101     thorpej   433: static void
                    434: pr_pagelist_free(struct pool *pp, struct pool_pagelist *pq)
                    435: {
                    436:        struct pool_item_header *ph;
                    437:
                    438:        while ((ph = LIST_FIRST(pq)) != NULL) {
                    439:                LIST_REMOVE(ph, ph_pagelist);
                    440:                pool_allocator_free(pp, ph->ph_page);
1.131.2.2! matt      441:                if ((pp->pr_roflags & PR_PHINPAGE) == 0)
1.101     thorpej   442:                        pool_put(pp->pr_phpool, ph);
                    443:        }
                    444: }
                    445:
1.3       pk        446: /*
                    447:  * Remove a page from the pool.
                    448:  */
1.110     perry     449: static inline void
1.61      chs       450: pr_rmpage(struct pool *pp, struct pool_item_header *ph,
                    451:      struct pool_pagelist *pq)
1.3       pk        452: {
                    453:
1.131.2.2! matt      454:        KASSERT(mutex_owned(&pp->pr_lock));
1.91      yamt      455:
1.3       pk        456:        /*
1.7       thorpej   457:         * If the page was idle, decrement the idle page count.
1.3       pk        458:         */
1.6       thorpej   459:        if (ph->ph_nmissing == 0) {
                    460: #ifdef DIAGNOSTIC
                    461:                if (pp->pr_nidle == 0)
                    462:                        panic("pr_rmpage: nidle inconsistent");
1.20      thorpej   463:                if (pp->pr_nitems < pp->pr_itemsperpage)
                    464:                        panic("pr_rmpage: nitems inconsistent");
1.6       thorpej   465: #endif
                    466:                pp->pr_nidle--;
                    467:        }
1.7       thorpej   468:
1.20      thorpej   469:        pp->pr_nitems -= pp->pr_itemsperpage;
                    470:
1.7       thorpej   471:        /*
1.101     thorpej   472:         * Unlink the page from the pool and queue it for release.
1.7       thorpej   473:         */
1.88      chs       474:        LIST_REMOVE(ph, ph_pagelist);
1.91      yamt      475:        if ((pp->pr_roflags & PR_PHINPAGE) == 0)
                    476:                SPLAY_REMOVE(phtree, &pp->pr_phtree, ph);
1.101     thorpej   477:        LIST_INSERT_HEAD(pq, ph, ph_pagelist);
                    478:
1.7       thorpej   479:        pp->pr_npages--;
                    480:        pp->pr_npagefree++;
1.6       thorpej   481:
1.88      chs       482:        pool_update_curpage(pp);
1.3       pk        483: }
                    484:
1.126     thorpej   485: static bool
1.117     yamt      486: pa_starved_p(struct pool_allocator *pa)
                    487: {
                    488:
                    489:        if (pa->pa_backingmap != NULL) {
                    490:                return vm_map_starved_p(pa->pa_backingmap);
                    491:        }
1.127     thorpej   492:        return false;
1.117     yamt      493: }
                    494:
                    495: static int
1.124     yamt      496: pool_reclaim_callback(struct callback_entry *ce, void *obj, void *arg)
1.117     yamt      497: {
                    498:        struct pool *pp = obj;
                    499:        struct pool_allocator *pa = pp->pr_alloc;
                    500:
                    501:        KASSERT(&pp->pr_reclaimerentry == ce);
                    502:        pool_reclaim(pp);
                    503:        if (!pa_starved_p(pa)) {
                    504:                return CALLBACK_CHAIN_ABORT;
                    505:        }
                    506:        return CALLBACK_CHAIN_CONTINUE;
                    507: }
                    508:
                    509: static void
                    510: pool_reclaim_register(struct pool *pp)
                    511: {
                    512:        struct vm_map *map = pp->pr_alloc->pa_backingmap;
                    513:        int s;
                    514:
                    515:        if (map == NULL) {
                    516:                return;
                    517:        }
                    518:
                    519:        s = splvm(); /* not necessary for INTRSAFE maps, but don't care. */
                    520:        callback_register(&vm_map_to_kernel(map)->vmk_reclaim_callback,
                    521:            &pp->pr_reclaimerentry, pp, pool_reclaim_callback);
                    522:        splx(s);
                    523: }
                    524:
                    525: static void
                    526: pool_reclaim_unregister(struct pool *pp)
                    527: {
                    528:        struct vm_map *map = pp->pr_alloc->pa_backingmap;
                    529:        int s;
                    530:
                    531:        if (map == NULL) {
                    532:                return;
                    533:        }
                    534:
                    535:        s = splvm(); /* not necessary for INTRSAFE maps, but don't care. */
                    536:        callback_unregister(&vm_map_to_kernel(map)->vmk_reclaim_callback,
                    537:            &pp->pr_reclaimerentry);
                    538:        splx(s);
                    539: }
                    540:
                    541: static void
                    542: pa_reclaim_register(struct pool_allocator *pa)
                    543: {
                    544:        struct vm_map *map = *pa->pa_backingmapptr;
                    545:        struct pool *pp;
                    546:
                    547:        KASSERT(pa->pa_backingmap == NULL);
                    548:        if (map == NULL) {
                    549:                SLIST_INSERT_HEAD(&pa_deferinitq, pa, pa_q);
                    550:                return;
                    551:        }
                    552:        pa->pa_backingmap = map;
                    553:        TAILQ_FOREACH(pp, &pa->pa_list, pr_alloc_list) {
                    554:                pool_reclaim_register(pp);
                    555:        }
                    556: }
                    557:
1.3       pk        558: /*
1.94      simonb    559:  * Initialize all the pools listed in the "pools" link set.
                    560:  */
                    561: void
1.117     yamt      562: pool_subsystem_init(void)
1.94      simonb    563: {
1.117     yamt      564:        struct pool_allocator *pa;
1.94      simonb    565:        __link_set_decl(pools, struct link_pool_init);
                    566:        struct link_pool_init * const *pi;
                    567:
1.131.2.2! matt      568:        mutex_init(&pool_head_lock, MUTEX_DEFAULT, IPL_NONE);
        !           569:        cv_init(&pool_busy, "poolbusy");
        !           570:
1.94      simonb    571:        __link_set_foreach(pi, pools)
                    572:                pool_init((*pi)->pp, (*pi)->size, (*pi)->align,
                    573:                    (*pi)->align_offset, (*pi)->flags, (*pi)->wchan,
1.129     ad        574:                    (*pi)->palloc, (*pi)->ipl);
1.117     yamt      575:
                    576:        while ((pa = SLIST_FIRST(&pa_deferinitq)) != NULL) {
                    577:                KASSERT(pa->pa_backingmapptr != NULL);
                    578:                KASSERT(*pa->pa_backingmapptr != NULL);
                    579:                SLIST_REMOVE_HEAD(&pa_deferinitq, pa_q);
                    580:                pa_reclaim_register(pa);
                    581:        }
1.131.2.2! matt      582:
        !           583:        pool_init(&cache_pool, sizeof(struct pool_cache), CACHE_LINE_SIZE,
        !           584:            0, 0, "pcache", &pool_allocator_nointr, IPL_NONE);
        !           585:
        !           586:        pool_init(&cache_cpu_pool, sizeof(pool_cache_cpu_t), CACHE_LINE_SIZE,
        !           587:            0, 0, "pcachecpu", &pool_allocator_nointr, IPL_NONE);
1.94      simonb    588: }
                    589:
                    590: /*
1.3       pk        591:  * Initialize the given pool resource structure.
                    592:  *
                    593:  * We export this routine to allow other kernel parts to declare
                    594:  * static pools that must be initialized before malloc() is available.
                    595:  */
                    596: void
1.42      thorpej   597: pool_init(struct pool *pp, size_t size, u_int align, u_int ioff, int flags,
1.129     ad        598:     const char *wchan, struct pool_allocator *palloc, int ipl)
1.3       pk        599: {
1.116     simonb    600: #ifdef DEBUG
                    601:        struct pool *pp1;
                    602: #endif
1.92      enami     603:        size_t trysize, phsize;
1.131.2.2! matt      604:        int off, slack;
1.3       pk        605:
1.99      yamt      606:        KASSERT((1UL << (CHAR_BIT * sizeof(pool_item_freelist_t))) - 2 >=
                    607:            PHPOOL_FREELIST_NELEM(PHPOOL_MAX - 1));
                    608:
1.116     simonb    609: #ifdef DEBUG
                    610:        /*
                    611:         * Check that the pool hasn't already been initialised and
                    612:         * added to the list of all pools.
                    613:         */
                    614:        LIST_FOREACH(pp1, &pool_head, pr_poollist) {
                    615:                if (pp == pp1)
                    616:                        panic("pool_init: pool %s already initialised",
                    617:                            wchan);
                    618:        }
                    619: #endif
                    620:
1.25      thorpej   621: #ifdef POOL_DIAGNOSTIC
                    622:        /*
                    623:         * Always log if POOL_DIAGNOSTIC is defined.
                    624:         */
                    625:        if (pool_logsize != 0)
                    626:                flags |= PR_LOGGING;
                    627: #endif
                    628:
1.66      thorpej   629:        if (palloc == NULL)
                    630:                palloc = &pool_allocator_kmem;
1.112     bjh21     631: #ifdef POOL_SUBPAGE
                    632:        if (size > palloc->pa_pagesz) {
                    633:                if (palloc == &pool_allocator_kmem)
                    634:                        palloc = &pool_allocator_kmem_fullpage;
                    635:                else if (palloc == &pool_allocator_nointr)
                    636:                        palloc = &pool_allocator_nointr_fullpage;
                    637:        }
1.66      thorpej   638: #endif /* POOL_SUBPAGE */
                    639:        if ((palloc->pa_flags & PA_INITIALIZED) == 0) {
1.112     bjh21     640:                if (palloc->pa_pagesz == 0)
1.66      thorpej   641:                        palloc->pa_pagesz = PAGE_SIZE;
                    642:
                    643:                TAILQ_INIT(&palloc->pa_list);
                    644:
1.131.2.2! matt      645:                mutex_init(&palloc->pa_lock, MUTEX_DEFAULT, IPL_VM);
1.66      thorpej   646:                palloc->pa_pagemask = ~(palloc->pa_pagesz - 1);
                    647:                palloc->pa_pageshift = ffs(palloc->pa_pagesz) - 1;
1.117     yamt      648:
                    649:                if (palloc->pa_backingmapptr != NULL) {
                    650:                        pa_reclaim_register(palloc);
                    651:                }
1.66      thorpej   652:                palloc->pa_flags |= PA_INITIALIZED;
1.4       thorpej   653:        }
1.3       pk        654:
                    655:        if (align == 0)
                    656:                align = ALIGN(1);
1.14      thorpej   657:
1.120     yamt      658:        if ((flags & PR_NOTOUCH) == 0 && size < sizeof(struct pool_item))
1.14      thorpej   659:                size = sizeof(struct pool_item);
1.3       pk        660:
1.78      thorpej   661:        size = roundup(size, align);
1.66      thorpej   662: #ifdef DIAGNOSTIC
                    663:        if (size > palloc->pa_pagesz)
1.121     yamt      664:                panic("pool_init: pool item size (%zu) too large", size);
1.66      thorpej   665: #endif
1.35      pk        666:
1.3       pk        667:        /*
                    668:         * Initialize the pool structure.
                    669:         */
1.88      chs       670:        LIST_INIT(&pp->pr_emptypages);
                    671:        LIST_INIT(&pp->pr_fullpages);
                    672:        LIST_INIT(&pp->pr_partpages);
1.131.2.2! matt      673:        pp->pr_cache = NULL;
1.3       pk        674:        pp->pr_curpage = NULL;
                    675:        pp->pr_npages = 0;
                    676:        pp->pr_minitems = 0;
                    677:        pp->pr_minpages = 0;
                    678:        pp->pr_maxpages = UINT_MAX;
1.20      thorpej   679:        pp->pr_roflags = flags;
                    680:        pp->pr_flags = 0;
1.35      pk        681:        pp->pr_size = size;
1.3       pk        682:        pp->pr_align = align;
                    683:        pp->pr_wchan = wchan;
1.66      thorpej   684:        pp->pr_alloc = palloc;
1.20      thorpej   685:        pp->pr_nitems = 0;
                    686:        pp->pr_nout = 0;
                    687:        pp->pr_hardlimit = UINT_MAX;
                    688:        pp->pr_hardlimit_warning = NULL;
1.31      thorpej   689:        pp->pr_hardlimit_ratecap.tv_sec = 0;
                    690:        pp->pr_hardlimit_ratecap.tv_usec = 0;
                    691:        pp->pr_hardlimit_warning_last.tv_sec = 0;
                    692:        pp->pr_hardlimit_warning_last.tv_usec = 0;
1.68      thorpej   693:        pp->pr_drain_hook = NULL;
                    694:        pp->pr_drain_hook_arg = NULL;
1.125     ad        695:        pp->pr_freecheck = NULL;
1.3       pk        696:
                    697:        /*
                    698:         * Decide whether to put the page header off page to avoid
1.92      enami     699:         * wasting too large a part of the page or too big item.
                    700:         * Off-page page headers go on a hash table, so we can match
                    701:         * a returned item with its header based on the page address.
                    702:         * We use 1/16 of the page size and about 8 times of the item
                    703:         * size as the threshold (XXX: tune)
                    704:         *
                    705:         * However, we'll put the header into the page if we can put
                    706:         * it without wasting any items.
                    707:         *
                    708:         * Silently enforce `0 <= ioff < align'.
1.3       pk        709:         */
1.92      enami     710:        pp->pr_itemoffset = ioff %= align;
                    711:        /* See the comment below about reserved bytes. */
                    712:        trysize = palloc->pa_pagesz - ((align - ioff) % align);
                    713:        phsize = ALIGN(sizeof(struct pool_item_header));
1.121     yamt      714:        if ((pp->pr_roflags & (PR_NOTOUCH | PR_NOALIGN)) == 0 &&
1.97      yamt      715:            (pp->pr_size < MIN(palloc->pa_pagesz / 16, phsize << 3) ||
                    716:            trysize / pp->pr_size == (trysize - phsize) / pp->pr_size)) {
1.3       pk        717:                /* Use the end of the page for the page header */
1.20      thorpej   718:                pp->pr_roflags |= PR_PHINPAGE;
1.92      enami     719:                pp->pr_phoffset = off = palloc->pa_pagesz - phsize;
1.2       pk        720:        } else {
1.3       pk        721:                /* The page header will be taken from our page header pool */
                    722:                pp->pr_phoffset = 0;
1.66      thorpej   723:                off = palloc->pa_pagesz;
1.88      chs       724:                SPLAY_INIT(&pp->pr_phtree);
1.2       pk        725:        }
1.1       pk        726:
1.3       pk        727:        /*
                    728:         * Alignment is to take place at `ioff' within the item. This means
                    729:         * we must reserve up to `align - 1' bytes on the page to allow
                    730:         * appropriate positioning of each item.
                    731:         */
                    732:        pp->pr_itemsperpage = (off - ((align - ioff) % align)) / pp->pr_size;
1.43      thorpej   733:        KASSERT(pp->pr_itemsperpage != 0);
1.97      yamt      734:        if ((pp->pr_roflags & PR_NOTOUCH)) {
                    735:                int idx;
                    736:
                    737:                for (idx = 0; pp->pr_itemsperpage > PHPOOL_FREELIST_NELEM(idx);
                    738:                    idx++) {
                    739:                        /* nothing */
                    740:                }
                    741:                if (idx >= PHPOOL_MAX) {
                    742:                        /*
                    743:                         * if you see this panic, consider to tweak
                    744:                         * PHPOOL_MAX and PHPOOL_FREELIST_NELEM.
                    745:                         */
                    746:                        panic("%s: too large itemsperpage(%d) for PR_NOTOUCH",
                    747:                            pp->pr_wchan, pp->pr_itemsperpage);
                    748:                }
                    749:                pp->pr_phpool = &phpool[idx];
                    750:        } else if ((pp->pr_roflags & PR_PHINPAGE) == 0) {
                    751:                pp->pr_phpool = &phpool[0];
                    752:        }
                    753: #if defined(DIAGNOSTIC)
                    754:        else {
                    755:                pp->pr_phpool = NULL;
                    756:        }
                    757: #endif
1.3       pk        758:
                    759:        /*
                    760:         * Use the slack between the chunks and the page header
                    761:         * for "cache coloring".
                    762:         */
                    763:        slack = off - pp->pr_itemsperpage * pp->pr_size;
                    764:        pp->pr_maxcolor = (slack / align) * align;
                    765:        pp->pr_curcolor = 0;
                    766:
                    767:        pp->pr_nget = 0;
                    768:        pp->pr_nfail = 0;
                    769:        pp->pr_nput = 0;
                    770:        pp->pr_npagealloc = 0;
                    771:        pp->pr_npagefree = 0;
1.1       pk        772:        pp->pr_hiwat = 0;
1.8       thorpej   773:        pp->pr_nidle = 0;
1.131.2.2! matt      774:        pp->pr_refcnt = 0;
1.3       pk        775:
1.59      thorpej   776: #ifdef POOL_DIAGNOSTIC
1.25      thorpej   777:        if (flags & PR_LOGGING) {
                    778:                if (kmem_map == NULL ||
                    779:                    (pp->pr_log = malloc(pool_logsize * sizeof(struct pool_log),
                    780:                     M_TEMP, M_NOWAIT)) == NULL)
1.20      thorpej   781:                        pp->pr_roflags &= ~PR_LOGGING;
1.3       pk        782:                pp->pr_curlogentry = 0;
                    783:                pp->pr_logsize = pool_logsize;
                    784:        }
1.59      thorpej   785: #endif
1.25      thorpej   786:
                    787:        pp->pr_entered_file = NULL;
                    788:        pp->pr_entered_line = 0;
1.3       pk        789:
1.131.2.2! matt      790:        mutex_init(&pp->pr_lock, MUTEX_DEFAULT, ipl);
        !           791:        cv_init(&pp->pr_cv, wchan);
        !           792:        pp->pr_ipl = ipl;
1.1       pk        793:
1.3       pk        794:        /*
1.43      thorpej   795:         * Initialize private page header pool and cache magazine pool if we
                    796:         * haven't done so yet.
1.23      thorpej   797:         * XXX LOCKING.
1.3       pk        798:         */
1.97      yamt      799:        if (phpool[0].pr_size == 0) {
                    800:                int idx;
                    801:                for (idx = 0; idx < PHPOOL_MAX; idx++) {
                    802:                        static char phpool_names[PHPOOL_MAX][6+1+6+1];
                    803:                        int nelem;
                    804:                        size_t sz;
                    805:
                    806:                        nelem = PHPOOL_FREELIST_NELEM(idx);
                    807:                        snprintf(phpool_names[idx], sizeof(phpool_names[idx]),
                    808:                            "phpool-%d", nelem);
                    809:                        sz = sizeof(struct pool_item_header);
                    810:                        if (nelem) {
                    811:                                sz = PR_FREELIST_ALIGN(sz)
1.99      yamt      812:                                    + nelem * sizeof(pool_item_freelist_t);
1.97      yamt      813:                        }
                    814:                        pool_init(&phpool[idx], sz, 0, 0, 0,
1.129     ad        815:                            phpool_names[idx], &pool_allocator_meta, IPL_VM);
1.97      yamt      816:                }
1.62      bjh21     817: #ifdef POOL_SUBPAGE
                    818:                pool_init(&psppool, POOL_SUBPAGE, POOL_SUBPAGE, 0,
1.129     ad        819:                    PR_RECURSIVE, "psppool", &pool_allocator_meta, IPL_VM);
1.62      bjh21     820: #endif
1.131.2.2! matt      821:                pool_init(&pcgpool, sizeof(pcg_t), CACHE_LINE_SIZE, 0, 0,
        !           822:                    "cachegrp", &pool_allocator_meta, IPL_VM);
1.1       pk        823:        }
                    824:
1.131.2.2! matt      825:        if (__predict_true(!cold)) {
        !           826:                /* Insert into the list of all pools. */
        !           827:                mutex_enter(&pool_head_lock);
        !           828:                LIST_INSERT_HEAD(&pool_head, pp, pr_poollist);
        !           829:                mutex_exit(&pool_head_lock);
        !           830:
        !           831:                /* Insert this into the list of pools using this allocator. */
        !           832:                mutex_enter(&palloc->pa_lock);
        !           833:                TAILQ_INSERT_TAIL(&palloc->pa_list, pp, pr_alloc_list);
        !           834:                mutex_exit(&palloc->pa_lock);
        !           835:        } else {
        !           836:                LIST_INSERT_HEAD(&pool_head, pp, pr_poollist);
        !           837:                TAILQ_INSERT_TAIL(&palloc->pa_list, pp, pr_alloc_list);
        !           838:        }
1.66      thorpej   839:
1.117     yamt      840:        pool_reclaim_register(pp);
1.1       pk        841: }
                    842:
                    843: /*
                    844:  * De-commision a pool resource.
                    845:  */
                    846: void
1.42      thorpej   847: pool_destroy(struct pool *pp)
1.1       pk        848: {
1.101     thorpej   849:        struct pool_pagelist pq;
1.3       pk        850:        struct pool_item_header *ph;
1.43      thorpej   851:
1.101     thorpej   852:        /* Remove from global pool list */
1.131.2.2! matt      853:        mutex_enter(&pool_head_lock);
        !           854:        while (pp->pr_refcnt != 0)
        !           855:                cv_wait(&pool_busy, &pool_head_lock);
1.102     chs       856:        LIST_REMOVE(pp, pr_poollist);
1.101     thorpej   857:        if (drainpp == pp)
                    858:                drainpp = NULL;
1.131.2.2! matt      859:        mutex_exit(&pool_head_lock);
1.101     thorpej   860:
                    861:        /* Remove this pool from its allocator's list of pools. */
1.117     yamt      862:        pool_reclaim_unregister(pp);
1.131.2.2! matt      863:        mutex_enter(&pp->pr_alloc->pa_lock);
1.66      thorpej   864:        TAILQ_REMOVE(&pp->pr_alloc->pa_list, pp, pr_alloc_list);
1.131.2.2! matt      865:        mutex_exit(&pp->pr_alloc->pa_lock);
1.66      thorpej   866:
1.131.2.2! matt      867:        mutex_enter(&pp->pr_lock);
1.101     thorpej   868:
1.131.2.2! matt      869:        KASSERT(pp->pr_cache == NULL);
1.3       pk        870:
                    871: #ifdef DIAGNOSTIC
1.20      thorpej   872:        if (pp->pr_nout != 0) {
1.25      thorpej   873:                pr_printlog(pp, NULL, printf);
1.80      provos    874:                panic("pool_destroy: pool busy: still out: %u",
1.20      thorpej   875:                    pp->pr_nout);
1.3       pk        876:        }
                    877: #endif
1.1       pk        878:
1.101     thorpej   879:        KASSERT(LIST_EMPTY(&pp->pr_fullpages));
                    880:        KASSERT(LIST_EMPTY(&pp->pr_partpages));
                    881:
1.3       pk        882:        /* Remove all pages */
1.101     thorpej   883:        LIST_INIT(&pq);
1.88      chs       884:        while ((ph = LIST_FIRST(&pp->pr_emptypages)) != NULL)
1.101     thorpej   885:                pr_rmpage(pp, ph, &pq);
                    886:
1.131.2.2! matt      887:        mutex_exit(&pp->pr_lock);
1.3       pk        888:
1.101     thorpej   889:        pr_pagelist_free(pp, &pq);
1.3       pk        890:
1.59      thorpej   891: #ifdef POOL_DIAGNOSTIC
1.20      thorpej   892:        if ((pp->pr_roflags & PR_LOGGING) != 0)
1.3       pk        893:                free(pp->pr_log, M_TEMP);
1.59      thorpej   894: #endif
1.131.2.2! matt      895:
        !           896:        cv_destroy(&pp->pr_cv);
        !           897:        mutex_destroy(&pp->pr_lock);
1.1       pk        898: }
                    899:
1.68      thorpej   900: void
                    901: pool_set_drain_hook(struct pool *pp, void (*fn)(void *, int), void *arg)
                    902: {
                    903:
                    904:        /* XXX no locking -- must be used just after pool_init() */
                    905: #ifdef DIAGNOSTIC
                    906:        if (pp->pr_drain_hook != NULL)
                    907:                panic("pool_set_drain_hook(%s): already set", pp->pr_wchan);
                    908: #endif
                    909:        pp->pr_drain_hook = fn;
                    910:        pp->pr_drain_hook_arg = arg;
                    911: }
                    912:
1.88      chs       913: static struct pool_item_header *
1.128     christos  914: pool_alloc_item_header(struct pool *pp, void *storage, int flags)
1.55      thorpej   915: {
                    916:        struct pool_item_header *ph;
                    917:
                    918:        if ((pp->pr_roflags & PR_PHINPAGE) != 0)
1.128     christos  919:                ph = (struct pool_item_header *) ((char *)storage + pp->pr_phoffset);
1.131.2.2! matt      920:        else
1.97      yamt      921:                ph = pool_get(pp->pr_phpool, flags);
1.55      thorpej   922:
                    923:        return (ph);
                    924: }
1.1       pk        925:
                    926: /*
1.131.2.2! matt      927:  * Grab an item from the pool.
1.1       pk        928:  */
1.3       pk        929: void *
1.59      thorpej   930: #ifdef POOL_DIAGNOSTIC
1.42      thorpej   931: _pool_get(struct pool *pp, int flags, const char *file, long line)
1.56      sommerfe  932: #else
                    933: pool_get(struct pool *pp, int flags)
                    934: #endif
1.1       pk        935: {
                    936:        struct pool_item *pi;
1.3       pk        937:        struct pool_item_header *ph;
1.55      thorpej   938:        void *v;
1.1       pk        939:
1.2       pk        940: #ifdef DIAGNOSTIC
1.95      atatat    941:        if (__predict_false(pp->pr_itemsperpage == 0))
                    942:                panic("pool_get: pool %p: pr_itemsperpage is zero, "
                    943:                    "pool not initialized?", pp);
1.84      thorpej   944:        if (__predict_false(curlwp == NULL && doing_shutdown == 0 &&
1.37      sommerfe  945:                            (flags & PR_WAITOK) != 0))
1.77      matt      946:                panic("pool_get: %s: must have NOWAIT", pp->pr_wchan);
1.58      thorpej   947:
1.102     chs       948: #endif /* DIAGNOSTIC */
1.58      thorpej   949: #ifdef LOCKDEBUG
                    950:        if (flags & PR_WAITOK)
1.119     yamt      951:                ASSERT_SLEEPABLE(NULL, "pool_get(PR_WAITOK)");
1.56      sommerfe  952: #endif
1.1       pk        953:
1.131.2.2! matt      954:        mutex_enter(&pp->pr_lock);
1.25      thorpej   955:        pr_enter(pp, file, line);
1.20      thorpej   956:
                    957:  startover:
                    958:        /*
                    959:         * Check to see if we've reached the hard limit.  If we have,
                    960:         * and we can wait, then wait until an item has been returned to
                    961:         * the pool.
                    962:         */
                    963: #ifdef DIAGNOSTIC
1.34      thorpej   964:        if (__predict_false(pp->pr_nout > pp->pr_hardlimit)) {
1.25      thorpej   965:                pr_leave(pp);
1.131.2.2! matt      966:                mutex_exit(&pp->pr_lock);
1.20      thorpej   967:                panic("pool_get: %s: crossed hard limit", pp->pr_wchan);
                    968:        }
                    969: #endif
1.34      thorpej   970:        if (__predict_false(pp->pr_nout == pp->pr_hardlimit)) {
1.68      thorpej   971:                if (pp->pr_drain_hook != NULL) {
                    972:                        /*
                    973:                         * Since the drain hook is going to free things
                    974:                         * back to the pool, unlock, call the hook, re-lock,
                    975:                         * and check the hardlimit condition again.
                    976:                         */
                    977:                        pr_leave(pp);
1.131.2.2! matt      978:                        mutex_exit(&pp->pr_lock);
1.68      thorpej   979:                        (*pp->pr_drain_hook)(pp->pr_drain_hook_arg, flags);
1.131.2.2! matt      980:                        mutex_enter(&pp->pr_lock);
1.68      thorpej   981:                        pr_enter(pp, file, line);
                    982:                        if (pp->pr_nout < pp->pr_hardlimit)
                    983:                                goto startover;
                    984:                }
                    985:
1.29      sommerfe  986:                if ((flags & PR_WAITOK) && !(flags & PR_LIMITFAIL)) {
1.20      thorpej   987:                        /*
                    988:                         * XXX: A warning isn't logged in this case.  Should
                    989:                         * it be?
                    990:                         */
                    991:                        pp->pr_flags |= PR_WANTED;
1.25      thorpej   992:                        pr_leave(pp);
1.131.2.2! matt      993:                        cv_wait(&pp->pr_cv, &pp->pr_lock);
1.25      thorpej   994:                        pr_enter(pp, file, line);
1.20      thorpej   995:                        goto startover;
                    996:                }
1.31      thorpej   997:
                    998:                /*
                    999:                 * Log a message that the hard limit has been hit.
                   1000:                 */
                   1001:                if (pp->pr_hardlimit_warning != NULL &&
                   1002:                    ratecheck(&pp->pr_hardlimit_warning_last,
                   1003:                              &pp->pr_hardlimit_ratecap))
                   1004:                        log(LOG_ERR, "%s\n", pp->pr_hardlimit_warning);
1.21      thorpej  1005:
                   1006:                pp->pr_nfail++;
                   1007:
1.25      thorpej  1008:                pr_leave(pp);
1.131.2.2! matt     1009:                mutex_exit(&pp->pr_lock);
1.20      thorpej  1010:                return (NULL);
                   1011:        }
                   1012:
1.3       pk       1013:        /*
                   1014:         * The convention we use is that if `curpage' is not NULL, then
                   1015:         * it points at a non-empty bucket. In particular, `curpage'
                   1016:         * never points at a page header which has PR_PHINPAGE set and
                   1017:         * has no items in its bucket.
                   1018:         */
1.20      thorpej  1019:        if ((ph = pp->pr_curpage) == NULL) {
1.113     yamt     1020:                int error;
                   1021:
1.20      thorpej  1022: #ifdef DIAGNOSTIC
                   1023:                if (pp->pr_nitems != 0) {
1.131.2.2! matt     1024:                        mutex_exit(&pp->pr_lock);
1.20      thorpej  1025:                        printf("pool_get: %s: curpage NULL, nitems %u\n",
                   1026:                            pp->pr_wchan, pp->pr_nitems);
1.80      provos   1027:                        panic("pool_get: nitems inconsistent");
1.20      thorpej  1028:                }
                   1029: #endif
                   1030:
1.21      thorpej  1031:                /*
                   1032:                 * Call the back-end page allocator for more memory.
                   1033:                 * Release the pool lock, as the back-end page allocator
                   1034:                 * may block.
                   1035:                 */
1.25      thorpej  1036:                pr_leave(pp);
1.113     yamt     1037:                error = pool_grow(pp, flags);
                   1038:                pr_enter(pp, file, line);
                   1039:                if (error != 0) {
1.21      thorpej  1040:                        /*
1.55      thorpej  1041:                         * We were unable to allocate a page or item
                   1042:                         * header, but we released the lock during
                   1043:                         * allocation, so perhaps items were freed
                   1044:                         * back to the pool.  Check for this case.
1.21      thorpej  1045:                         */
                   1046:                        if (pp->pr_curpage != NULL)
                   1047:                                goto startover;
1.15      pk       1048:
1.117     yamt     1049:                        pp->pr_nfail++;
1.25      thorpej  1050:                        pr_leave(pp);
1.131.2.2! matt     1051:                        mutex_exit(&pp->pr_lock);
1.117     yamt     1052:                        return (NULL);
1.1       pk       1053:                }
1.3       pk       1054:
1.20      thorpej  1055:                /* Start the allocation process over. */
                   1056:                goto startover;
1.3       pk       1057:        }
1.97      yamt     1058:        if (pp->pr_roflags & PR_NOTOUCH) {
                   1059: #ifdef DIAGNOSTIC
                   1060:                if (__predict_false(ph->ph_nmissing == pp->pr_itemsperpage)) {
                   1061:                        pr_leave(pp);
1.131.2.2! matt     1062:                        mutex_exit(&pp->pr_lock);
1.97      yamt     1063:                        panic("pool_get: %s: page empty", pp->pr_wchan);
                   1064:                }
                   1065: #endif
                   1066:                v = pr_item_notouch_get(pp, ph);
                   1067: #ifdef POOL_DIAGNOSTIC
                   1068:                pr_log(pp, v, PRLOG_GET, file, line);
                   1069: #endif
                   1070:        } else {
1.102     chs      1071:                v = pi = LIST_FIRST(&ph->ph_itemlist);
1.97      yamt     1072:                if (__predict_false(v == NULL)) {
                   1073:                        pr_leave(pp);
1.131.2.2! matt     1074:                        mutex_exit(&pp->pr_lock);
1.97      yamt     1075:                        panic("pool_get: %s: page empty", pp->pr_wchan);
                   1076:                }
1.20      thorpej  1077: #ifdef DIAGNOSTIC
1.97      yamt     1078:                if (__predict_false(pp->pr_nitems == 0)) {
                   1079:                        pr_leave(pp);
1.131.2.2! matt     1080:                        mutex_exit(&pp->pr_lock);
1.97      yamt     1081:                        printf("pool_get: %s: items on itemlist, nitems %u\n",
                   1082:                            pp->pr_wchan, pp->pr_nitems);
                   1083:                        panic("pool_get: nitems inconsistent");
                   1084:                }
1.65      enami    1085: #endif
1.56      sommerfe 1086:
1.65      enami    1087: #ifdef POOL_DIAGNOSTIC
1.97      yamt     1088:                pr_log(pp, v, PRLOG_GET, file, line);
1.65      enami    1089: #endif
1.3       pk       1090:
1.65      enami    1091: #ifdef DIAGNOSTIC
1.97      yamt     1092:                if (__predict_false(pi->pi_magic != PI_MAGIC)) {
                   1093:                        pr_printlog(pp, pi, printf);
                   1094:                        panic("pool_get(%s): free list modified: "
                   1095:                            "magic=%x; page %p; item addr %p\n",
                   1096:                            pp->pr_wchan, pi->pi_magic, ph->ph_page, pi);
                   1097:                }
1.3       pk       1098: #endif
                   1099:
1.97      yamt     1100:                /*
                   1101:                 * Remove from item list.
                   1102:                 */
1.102     chs      1103:                LIST_REMOVE(pi, pi_list);
1.97      yamt     1104:        }
1.20      thorpej  1105:        pp->pr_nitems--;
                   1106:        pp->pr_nout++;
1.6       thorpej  1107:        if (ph->ph_nmissing == 0) {
                   1108: #ifdef DIAGNOSTIC
1.34      thorpej  1109:                if (__predict_false(pp->pr_nidle == 0))
1.6       thorpej  1110:                        panic("pool_get: nidle inconsistent");
                   1111: #endif
                   1112:                pp->pr_nidle--;
1.88      chs      1113:
                   1114:                /*
                   1115:                 * This page was previously empty.  Move it to the list of
                   1116:                 * partially-full pages.  This page is already curpage.
                   1117:                 */
                   1118:                LIST_REMOVE(ph, ph_pagelist);
                   1119:                LIST_INSERT_HEAD(&pp->pr_partpages, ph, ph_pagelist);
1.6       thorpej  1120:        }
1.3       pk       1121:        ph->ph_nmissing++;
1.97      yamt     1122:        if (ph->ph_nmissing == pp->pr_itemsperpage) {
1.21      thorpej  1123: #ifdef DIAGNOSTIC
1.97      yamt     1124:                if (__predict_false((pp->pr_roflags & PR_NOTOUCH) == 0 &&
1.102     chs      1125:                    !LIST_EMPTY(&ph->ph_itemlist))) {
1.25      thorpej  1126:                        pr_leave(pp);
1.131.2.2! matt     1127:                        mutex_exit(&pp->pr_lock);
1.21      thorpej  1128:                        panic("pool_get: %s: nmissing inconsistent",
                   1129:                            pp->pr_wchan);
                   1130:                }
                   1131: #endif
1.3       pk       1132:                /*
1.88      chs      1133:                 * This page is now full.  Move it to the full list
                   1134:                 * and select a new current page.
1.3       pk       1135:                 */
1.88      chs      1136:                LIST_REMOVE(ph, ph_pagelist);
                   1137:                LIST_INSERT_HEAD(&pp->pr_fullpages, ph, ph_pagelist);
                   1138:                pool_update_curpage(pp);
1.1       pk       1139:        }
1.3       pk       1140:
                   1141:        pp->pr_nget++;
1.111     christos 1142:        pr_leave(pp);
1.20      thorpej  1143:
                   1144:        /*
                   1145:         * If we have a low water mark and we are now below that low
                   1146:         * water mark, add more items to the pool.
                   1147:         */
1.53      thorpej  1148:        if (POOL_NEEDS_CATCHUP(pp) && pool_catchup(pp) != 0) {
1.20      thorpej  1149:                /*
                   1150:                 * XXX: Should we log a warning?  Should we set up a timeout
                   1151:                 * to try again in a second or so?  The latter could break
                   1152:                 * a caller's assumptions about interrupt protection, etc.
                   1153:                 */
                   1154:        }
                   1155:
1.131.2.2! matt     1156:        mutex_exit(&pp->pr_lock);
1.125     ad       1157:        KASSERT((((vaddr_t)v + pp->pr_itemoffset) & (pp->pr_align - 1)) == 0);
                   1158:        FREECHECK_OUT(&pp->pr_freecheck, v);
1.1       pk       1159:        return (v);
                   1160: }
                   1161:
                   1162: /*
1.43      thorpej  1163:  * Internal version of pool_put().  Pool is already locked/entered.
1.1       pk       1164:  */
1.43      thorpej  1165: static void
1.101     thorpej  1166: pool_do_put(struct pool *pp, void *v, struct pool_pagelist *pq)
1.1       pk       1167: {
                   1168:        struct pool_item *pi = v;
1.3       pk       1169:        struct pool_item_header *ph;
                   1170:
1.131.2.2! matt     1171:        KASSERT(mutex_owned(&pp->pr_lock));
1.125     ad       1172:        FREECHECK_IN(&pp->pr_freecheck, v);
1.131.2.2! matt     1173:        LOCKDEBUG_MEM_CHECK(v, pp->pr_size);
1.61      chs      1174:
1.30      thorpej  1175: #ifdef DIAGNOSTIC
1.34      thorpej  1176:        if (__predict_false(pp->pr_nout == 0)) {
1.30      thorpej  1177:                printf("pool %s: putting with none out\n",
                   1178:                    pp->pr_wchan);
                   1179:                panic("pool_put");
                   1180:        }
                   1181: #endif
1.3       pk       1182:
1.121     yamt     1183:        if (__predict_false((ph = pr_find_pagehead(pp, v)) == NULL)) {
1.25      thorpej  1184:                pr_printlog(pp, NULL, printf);
1.3       pk       1185:                panic("pool_put: %s: page header missing", pp->pr_wchan);
                   1186:        }
1.28      thorpej  1187:
1.3       pk       1188:        /*
                   1189:         * Return to item list.
                   1190:         */
1.97      yamt     1191:        if (pp->pr_roflags & PR_NOTOUCH) {
                   1192:                pr_item_notouch_put(pp, ph, v);
                   1193:        } else {
1.2       pk       1194: #ifdef DIAGNOSTIC
1.97      yamt     1195:                pi->pi_magic = PI_MAGIC;
1.3       pk       1196: #endif
1.32      chs      1197: #ifdef DEBUG
1.97      yamt     1198:                {
                   1199:                        int i, *ip = v;
1.32      chs      1200:
1.97      yamt     1201:                        for (i = 0; i < pp->pr_size / sizeof(int); i++) {
                   1202:                                *ip++ = PI_MAGIC;
                   1203:                        }
1.32      chs      1204:                }
                   1205: #endif
                   1206:
1.102     chs      1207:                LIST_INSERT_HEAD(&ph->ph_itemlist, pi, pi_list);
1.97      yamt     1208:        }
1.79      thorpej  1209:        KDASSERT(ph->ph_nmissing != 0);
1.3       pk       1210:        ph->ph_nmissing--;
                   1211:        pp->pr_nput++;
1.20      thorpej  1212:        pp->pr_nitems++;
                   1213:        pp->pr_nout--;
1.3       pk       1214:
                   1215:        /* Cancel "pool empty" condition if it exists */
                   1216:        if (pp->pr_curpage == NULL)
                   1217:                pp->pr_curpage = ph;
                   1218:
                   1219:        if (pp->pr_flags & PR_WANTED) {
                   1220:                pp->pr_flags &= ~PR_WANTED;
1.15      pk       1221:                if (ph->ph_nmissing == 0)
                   1222:                        pp->pr_nidle++;
1.131.2.2! matt     1223:                cv_broadcast(&pp->pr_cv);
1.3       pk       1224:                return;
                   1225:        }
                   1226:
                   1227:        /*
1.88      chs      1228:         * If this page is now empty, do one of two things:
1.21      thorpej  1229:         *
1.88      chs      1230:         *      (1) If we have more pages than the page high water mark,
1.96      thorpej  1231:         *          free the page back to the system.  ONLY CONSIDER
1.90      thorpej  1232:         *          FREEING BACK A PAGE IF WE HAVE MORE THAN OUR MINIMUM PAGE
                   1233:         *          CLAIM.
1.21      thorpej  1234:         *
1.88      chs      1235:         *      (2) Otherwise, move the page to the empty page list.
                   1236:         *
                   1237:         * Either way, select a new current page (so we use a partially-full
                   1238:         * page if one is available).
1.3       pk       1239:         */
                   1240:        if (ph->ph_nmissing == 0) {
1.6       thorpej  1241:                pp->pr_nidle++;
1.90      thorpej  1242:                if (pp->pr_npages > pp->pr_minpages &&
                   1243:                    (pp->pr_npages > pp->pr_maxpages ||
1.117     yamt     1244:                     pa_starved_p(pp->pr_alloc))) {
1.101     thorpej  1245:                        pr_rmpage(pp, ph, pq);
1.3       pk       1246:                } else {
1.88      chs      1247:                        LIST_REMOVE(ph, ph_pagelist);
                   1248:                        LIST_INSERT_HEAD(&pp->pr_emptypages, ph, ph_pagelist);
1.3       pk       1249:
1.21      thorpej  1250:                        /*
                   1251:                         * Update the timestamp on the page.  A page must
                   1252:                         * be idle for some period of time before it can
                   1253:                         * be reclaimed by the pagedaemon.  This minimizes
                   1254:                         * ping-pong'ing for memory.
                   1255:                         */
1.118     kardel   1256:                        getmicrotime(&ph->ph_time);
1.1       pk       1257:                }
1.88      chs      1258:                pool_update_curpage(pp);
1.1       pk       1259:        }
1.88      chs      1260:
1.21      thorpej  1261:        /*
1.88      chs      1262:         * If the page was previously completely full, move it to the
                   1263:         * partially-full list and make it the current page.  The next
                   1264:         * allocation will get the item from this page, instead of
                   1265:         * further fragmenting the pool.
1.21      thorpej  1266:         */
                   1267:        else if (ph->ph_nmissing == (pp->pr_itemsperpage - 1)) {
1.88      chs      1268:                LIST_REMOVE(ph, ph_pagelist);
                   1269:                LIST_INSERT_HEAD(&pp->pr_partpages, ph, ph_pagelist);
1.21      thorpej  1270:                pp->pr_curpage = ph;
                   1271:        }
1.43      thorpej  1272: }
                   1273:
                   1274: /*
1.131.2.2! matt     1275:  * Return resource to the pool.
1.43      thorpej  1276:  */
1.59      thorpej  1277: #ifdef POOL_DIAGNOSTIC
1.43      thorpej  1278: void
                   1279: _pool_put(struct pool *pp, void *v, const char *file, long line)
                   1280: {
1.101     thorpej  1281:        struct pool_pagelist pq;
                   1282:
                   1283:        LIST_INIT(&pq);
1.43      thorpej  1284:
1.131.2.2! matt     1285:        mutex_enter(&pp->pr_lock);
1.43      thorpej  1286:        pr_enter(pp, file, line);
                   1287:
1.56      sommerfe 1288:        pr_log(pp, v, PRLOG_PUT, file, line);
                   1289:
1.101     thorpej  1290:        pool_do_put(pp, v, &pq);
1.21      thorpej  1291:
1.25      thorpej  1292:        pr_leave(pp);
1.131.2.2! matt     1293:        mutex_exit(&pp->pr_lock);
1.101     thorpej  1294:
1.102     chs      1295:        pr_pagelist_free(pp, &pq);
1.1       pk       1296: }
1.57      sommerfe 1297: #undef pool_put
1.59      thorpej  1298: #endif /* POOL_DIAGNOSTIC */
1.1       pk       1299:
1.56      sommerfe 1300: void
                   1301: pool_put(struct pool *pp, void *v)
                   1302: {
1.101     thorpej  1303:        struct pool_pagelist pq;
                   1304:
                   1305:        LIST_INIT(&pq);
1.56      sommerfe 1306:
1.131.2.2! matt     1307:        mutex_enter(&pp->pr_lock);
1.101     thorpej  1308:        pool_do_put(pp, v, &pq);
1.131.2.2! matt     1309:        mutex_exit(&pp->pr_lock);
1.56      sommerfe 1310:
1.102     chs      1311:        pr_pagelist_free(pp, &pq);
1.56      sommerfe 1312: }
1.57      sommerfe 1313:
1.59      thorpej  1314: #ifdef POOL_DIAGNOSTIC
1.57      sommerfe 1315: #define                pool_put(h, v)  _pool_put((h), (v), __FILE__, __LINE__)
1.56      sommerfe 1316: #endif
1.74      thorpej  1317:
                   1318: /*
1.113     yamt     1319:  * pool_grow: grow a pool by a page.
                   1320:  *
                   1321:  * => called with pool locked.
                   1322:  * => unlock and relock the pool.
                   1323:  * => return with pool locked.
                   1324:  */
                   1325:
                   1326: static int
                   1327: pool_grow(struct pool *pp, int flags)
                   1328: {
                   1329:        struct pool_item_header *ph = NULL;
                   1330:        char *cp;
                   1331:
1.131.2.2! matt     1332:        mutex_exit(&pp->pr_lock);
1.113     yamt     1333:        cp = pool_allocator_alloc(pp, flags);
                   1334:        if (__predict_true(cp != NULL)) {
                   1335:                ph = pool_alloc_item_header(pp, cp, flags);
                   1336:        }
                   1337:        if (__predict_false(cp == NULL || ph == NULL)) {
                   1338:                if (cp != NULL) {
                   1339:                        pool_allocator_free(pp, cp);
                   1340:                }
1.131.2.2! matt     1341:                mutex_enter(&pp->pr_lock);
1.113     yamt     1342:                return ENOMEM;
                   1343:        }
                   1344:
1.131.2.2! matt     1345:        mutex_enter(&pp->pr_lock);
1.113     yamt     1346:        pool_prime_page(pp, cp, ph);
                   1347:        pp->pr_npagealloc++;
                   1348:        return 0;
                   1349: }
                   1350:
                   1351: /*
1.74      thorpej  1352:  * Add N items to the pool.
                   1353:  */
                   1354: int
                   1355: pool_prime(struct pool *pp, int n)
                   1356: {
1.75      simonb   1357:        int newpages;
1.113     yamt     1358:        int error = 0;
1.74      thorpej  1359:
1.131.2.2! matt     1360:        mutex_enter(&pp->pr_lock);
1.74      thorpej  1361:
                   1362:        newpages = roundup(n, pp->pr_itemsperpage) / pp->pr_itemsperpage;
                   1363:
                   1364:        while (newpages-- > 0) {
1.113     yamt     1365:                error = pool_grow(pp, PR_NOWAIT);
                   1366:                if (error) {
1.74      thorpej  1367:                        break;
                   1368:                }
                   1369:                pp->pr_minpages++;
                   1370:        }
                   1371:
                   1372:        if (pp->pr_minpages >= pp->pr_maxpages)
                   1373:                pp->pr_maxpages = pp->pr_minpages + 1;  /* XXX */
                   1374:
1.131.2.2! matt     1375:        mutex_exit(&pp->pr_lock);
1.113     yamt     1376:        return error;
1.74      thorpej  1377: }
1.55      thorpej  1378:
                   1379: /*
1.3       pk       1380:  * Add a page worth of items to the pool.
1.21      thorpej  1381:  *
                   1382:  * Note, we must be called with the pool descriptor LOCKED.
1.3       pk       1383:  */
1.55      thorpej  1384: static void
1.128     christos 1385: pool_prime_page(struct pool *pp, void *storage, struct pool_item_header *ph)
1.3       pk       1386: {
                   1387:        struct pool_item *pi;
1.128     christos 1388:        void *cp = storage;
1.125     ad       1389:        const unsigned int align = pp->pr_align;
                   1390:        const unsigned int ioff = pp->pr_itemoffset;
1.55      thorpej  1391:        int n;
1.36      pk       1392:
1.131.2.2! matt     1393:        KASSERT(mutex_owned(&pp->pr_lock));
1.91      yamt     1394:
1.66      thorpej  1395: #ifdef DIAGNOSTIC
1.121     yamt     1396:        if ((pp->pr_roflags & PR_NOALIGN) == 0 &&
                   1397:            ((uintptr_t)cp & (pp->pr_alloc->pa_pagesz - 1)) != 0)
1.36      pk       1398:                panic("pool_prime_page: %s: unaligned page", pp->pr_wchan);
1.66      thorpej  1399: #endif
1.3       pk       1400:
                   1401:        /*
                   1402:         * Insert page header.
                   1403:         */
1.88      chs      1404:        LIST_INSERT_HEAD(&pp->pr_emptypages, ph, ph_pagelist);
1.102     chs      1405:        LIST_INIT(&ph->ph_itemlist);
1.3       pk       1406:        ph->ph_page = storage;
                   1407:        ph->ph_nmissing = 0;
1.118     kardel   1408:        getmicrotime(&ph->ph_time);
1.88      chs      1409:        if ((pp->pr_roflags & PR_PHINPAGE) == 0)
                   1410:                SPLAY_INSERT(phtree, &pp->pr_phtree, ph);
1.3       pk       1411:
1.6       thorpej  1412:        pp->pr_nidle++;
                   1413:
1.3       pk       1414:        /*
                   1415:         * Color this page.
                   1416:         */
1.128     christos 1417:        cp = (char *)cp + pp->pr_curcolor;
1.3       pk       1418:        if ((pp->pr_curcolor += align) > pp->pr_maxcolor)
                   1419:                pp->pr_curcolor = 0;
                   1420:
                   1421:        /*
                   1422:         * Adjust storage to apply aligment to `pr_itemoffset' in each item.
                   1423:         */
                   1424:        if (ioff != 0)
1.128     christos 1425:                cp = (char *)cp + align - ioff;
1.3       pk       1426:
1.125     ad       1427:        KASSERT((((vaddr_t)cp + ioff) & (align - 1)) == 0);
                   1428:
1.3       pk       1429:        /*
                   1430:         * Insert remaining chunks on the bucket list.
                   1431:         */
                   1432:        n = pp->pr_itemsperpage;
1.20      thorpej  1433:        pp->pr_nitems += n;
1.3       pk       1434:
1.97      yamt     1435:        if (pp->pr_roflags & PR_NOTOUCH) {
1.99      yamt     1436:                pool_item_freelist_t *freelist = PR_FREELIST(ph);
1.97      yamt     1437:                int i;
                   1438:
1.128     christos 1439:                ph->ph_off = (char *)cp - (char *)storage;
1.97      yamt     1440:                ph->ph_firstfree = 0;
                   1441:                for (i = 0; i < n - 1; i++)
                   1442:                        freelist[i] = i + 1;
                   1443:                freelist[n - 1] = PR_INDEX_EOL;
                   1444:        } else {
                   1445:                while (n--) {
                   1446:                        pi = (struct pool_item *)cp;
1.78      thorpej  1447:
1.97      yamt     1448:                        KASSERT(((((vaddr_t)pi) + ioff) & (align - 1)) == 0);
1.3       pk       1449:
1.97      yamt     1450:                        /* Insert on page list */
1.102     chs      1451:                        LIST_INSERT_HEAD(&ph->ph_itemlist, pi, pi_list);
1.3       pk       1452: #ifdef DIAGNOSTIC
1.97      yamt     1453:                        pi->pi_magic = PI_MAGIC;
1.3       pk       1454: #endif
1.128     christos 1455:                        cp = (char *)cp + pp->pr_size;
1.125     ad       1456:
                   1457:                        KASSERT((((vaddr_t)cp + ioff) & (align - 1)) == 0);
1.97      yamt     1458:                }
1.3       pk       1459:        }
                   1460:
                   1461:        /*
                   1462:         * If the pool was depleted, point at the new page.
                   1463:         */
                   1464:        if (pp->pr_curpage == NULL)
                   1465:                pp->pr_curpage = ph;
                   1466:
                   1467:        if (++pp->pr_npages > pp->pr_hiwat)
                   1468:                pp->pr_hiwat = pp->pr_npages;
                   1469: }
                   1470:
1.20      thorpej  1471: /*
1.52      thorpej  1472:  * Used by pool_get() when nitems drops below the low water mark.  This
1.88      chs      1473:  * is used to catch up pr_nitems with the low water mark.
1.20      thorpej  1474:  *
1.21      thorpej  1475:  * Note 1, we never wait for memory here, we let the caller decide what to do.
1.20      thorpej  1476:  *
1.73      thorpej  1477:  * Note 2, we must be called with the pool already locked, and we return
1.20      thorpej  1478:  * with it locked.
                   1479:  */
                   1480: static int
1.42      thorpej  1481: pool_catchup(struct pool *pp)
1.20      thorpej  1482: {
                   1483:        int error = 0;
                   1484:
1.54      thorpej  1485:        while (POOL_NEEDS_CATCHUP(pp)) {
1.113     yamt     1486:                error = pool_grow(pp, PR_NOWAIT);
                   1487:                if (error) {
1.20      thorpej  1488:                        break;
                   1489:                }
                   1490:        }
1.113     yamt     1491:        return error;
1.20      thorpej  1492: }
                   1493:
1.88      chs      1494: static void
                   1495: pool_update_curpage(struct pool *pp)
                   1496: {
                   1497:
                   1498:        pp->pr_curpage = LIST_FIRST(&pp->pr_partpages);
                   1499:        if (pp->pr_curpage == NULL) {
                   1500:                pp->pr_curpage = LIST_FIRST(&pp->pr_emptypages);
                   1501:        }
                   1502: }
                   1503:
1.3       pk       1504: void
1.42      thorpej  1505: pool_setlowat(struct pool *pp, int n)
1.3       pk       1506: {
1.15      pk       1507:
1.131.2.2! matt     1508:        mutex_enter(&pp->pr_lock);
1.21      thorpej  1509:
1.3       pk       1510:        pp->pr_minitems = n;
1.15      pk       1511:        pp->pr_minpages = (n == 0)
                   1512:                ? 0
1.18      thorpej  1513:                : roundup(n, pp->pr_itemsperpage) / pp->pr_itemsperpage;
1.20      thorpej  1514:
                   1515:        /* Make sure we're caught up with the newly-set low water mark. */
1.75      simonb   1516:        if (POOL_NEEDS_CATCHUP(pp) && pool_catchup(pp) != 0) {
1.20      thorpej  1517:                /*
                   1518:                 * XXX: Should we log a warning?  Should we set up a timeout
                   1519:                 * to try again in a second or so?  The latter could break
                   1520:                 * a caller's assumptions about interrupt protection, etc.
                   1521:                 */
                   1522:        }
1.21      thorpej  1523:
1.131.2.2! matt     1524:        mutex_exit(&pp->pr_lock);
1.3       pk       1525: }
                   1526:
                   1527: void
1.42      thorpej  1528: pool_sethiwat(struct pool *pp, int n)
1.3       pk       1529: {
1.15      pk       1530:
1.131.2.2! matt     1531:        mutex_enter(&pp->pr_lock);
1.21      thorpej  1532:
1.15      pk       1533:        pp->pr_maxpages = (n == 0)
                   1534:                ? 0
1.18      thorpej  1535:                : roundup(n, pp->pr_itemsperpage) / pp->pr_itemsperpage;
1.21      thorpej  1536:
1.131.2.2! matt     1537:        mutex_exit(&pp->pr_lock);
1.3       pk       1538: }
                   1539:
1.20      thorpej  1540: void
1.42      thorpej  1541: pool_sethardlimit(struct pool *pp, int n, const char *warnmess, int ratecap)
1.20      thorpej  1542: {
                   1543:
1.131.2.2! matt     1544:        mutex_enter(&pp->pr_lock);
1.20      thorpej  1545:
                   1546:        pp->pr_hardlimit = n;
                   1547:        pp->pr_hardlimit_warning = warnmess;
1.31      thorpej  1548:        pp->pr_hardlimit_ratecap.tv_sec = ratecap;
                   1549:        pp->pr_hardlimit_warning_last.tv_sec = 0;
                   1550:        pp->pr_hardlimit_warning_last.tv_usec = 0;
1.20      thorpej  1551:
                   1552:        /*
1.21      thorpej  1553:         * In-line version of pool_sethiwat(), because we don't want to
                   1554:         * release the lock.
1.20      thorpej  1555:         */
                   1556:        pp->pr_maxpages = (n == 0)
                   1557:                ? 0
                   1558:                : roundup(n, pp->pr_itemsperpage) / pp->pr_itemsperpage;
1.21      thorpej  1559:
1.131.2.2! matt     1560:        mutex_exit(&pp->pr_lock);
1.20      thorpej  1561: }
1.3       pk       1562:
                   1563: /*
                   1564:  * Release all complete pages that have not been used recently.
                   1565:  */
1.66      thorpej  1566: int
1.59      thorpej  1567: #ifdef POOL_DIAGNOSTIC
1.42      thorpej  1568: _pool_reclaim(struct pool *pp, const char *file, long line)
1.56      sommerfe 1569: #else
                   1570: pool_reclaim(struct pool *pp)
                   1571: #endif
1.3       pk       1572: {
                   1573:        struct pool_item_header *ph, *phnext;
1.61      chs      1574:        struct pool_pagelist pq;
1.102     chs      1575:        struct timeval curtime, diff;
1.131.2.2! matt     1576:        bool klock;
        !          1577:        int rv;
1.3       pk       1578:
1.68      thorpej  1579:        if (pp->pr_drain_hook != NULL) {
                   1580:                /*
                   1581:                 * The drain hook must be called with the pool unlocked.
                   1582:                 */
                   1583:                (*pp->pr_drain_hook)(pp->pr_drain_hook_arg, PR_NOWAIT);
                   1584:        }
                   1585:
1.131.2.2! matt     1586:        /*
        !          1587:         * XXXSMP Because mutexes at IPL_SOFTXXX are still spinlocks,
        !          1588:         * and we are called from the pagedaemon without kernel_lock.
        !          1589:         * Does not apply to IPL_SOFTBIO.
        !          1590:         */
        !          1591:        if (pp->pr_ipl == IPL_SOFTNET || pp->pr_ipl == IPL_SOFTCLOCK ||
        !          1592:            pp->pr_ipl == IPL_SOFTSERIAL) {
        !          1593:                KERNEL_LOCK(1, NULL);
        !          1594:                klock = true;
        !          1595:        } else
        !          1596:                klock = false;
        !          1597:
        !          1598:        /* Reclaim items from the pool's cache (if any). */
        !          1599:        if (pp->pr_cache != NULL)
        !          1600:                pool_cache_invalidate(pp->pr_cache);
        !          1601:
        !          1602:        if (mutex_tryenter(&pp->pr_lock) == 0) {
        !          1603:                if (klock) {
        !          1604:                        KERNEL_UNLOCK_ONE(NULL);
        !          1605:                }
1.66      thorpej  1606:                return (0);
1.131.2.2! matt     1607:        }
1.25      thorpej  1608:        pr_enter(pp, file, line);
1.68      thorpej  1609:
1.88      chs      1610:        LIST_INIT(&pq);
1.43      thorpej  1611:
1.118     kardel   1612:        getmicrotime(&curtime);
1.21      thorpej  1613:
1.88      chs      1614:        for (ph = LIST_FIRST(&pp->pr_emptypages); ph != NULL; ph = phnext) {
                   1615:                phnext = LIST_NEXT(ph, ph_pagelist);
1.3       pk       1616:
                   1617:                /* Check our minimum page claim */
                   1618:                if (pp->pr_npages <= pp->pr_minpages)
                   1619:                        break;
                   1620:
1.88      chs      1621:                KASSERT(ph->ph_nmissing == 0);
                   1622:                timersub(&curtime, &ph->ph_time, &diff);
1.117     yamt     1623:                if (diff.tv_sec < pool_inactive_time
                   1624:                    && !pa_starved_p(pp->pr_alloc))
1.88      chs      1625:                        continue;
1.21      thorpej  1626:
1.88      chs      1627:                /*
                   1628:                 * If freeing this page would put us below
                   1629:                 * the low water mark, stop now.
                   1630:                 */
                   1631:                if ((pp->pr_nitems - pp->pr_itemsperpage) <
                   1632:                    pp->pr_minitems)
                   1633:                        break;
1.21      thorpej  1634:
1.88      chs      1635:                pr_rmpage(pp, ph, &pq);
1.3       pk       1636:        }
                   1637:
1.25      thorpej  1638:        pr_leave(pp);
1.131.2.2! matt     1639:        mutex_exit(&pp->pr_lock);
1.66      thorpej  1640:
1.131.2.2! matt     1641:        if (LIST_EMPTY(&pq))
        !          1642:                rv = 0;
        !          1643:        else {
        !          1644:                pr_pagelist_free(pp, &pq);
        !          1645:                rv = 1;
        !          1646:        }
        !          1647:
        !          1648:        if (klock) {
        !          1649:                KERNEL_UNLOCK_ONE(NULL);
        !          1650:        }
        !          1651:
        !          1652:        return (rv);
1.3       pk       1653: }
                   1654:
                   1655: /*
1.131.2.2! matt     1656:  * Drain pools, one at a time.  This is a two stage process;
        !          1657:  * drain_start kicks off a cross call to drain CPU-level caches
        !          1658:  * if the pool has an associated pool_cache.  drain_end waits
        !          1659:  * for those cross calls to finish, and then drains the cache
        !          1660:  * (if any) and pool.
1.131     ad       1661:  *
1.131.2.2! matt     1662:  * Note, must never be called from interrupt context.
1.3       pk       1663:  */
                   1664: void
1.131.2.2! matt     1665: pool_drain_start(struct pool **ppp, uint64_t *wp)
1.3       pk       1666: {
                   1667:        struct pool *pp;
1.131.2.2! matt     1668:
        !          1669:        KASSERT(!LIST_EMPTY(&pool_head));
1.3       pk       1670:
1.61      chs      1671:        pp = NULL;
1.131.2.2! matt     1672:
        !          1673:        /* Find next pool to drain, and add a reference. */
        !          1674:        mutex_enter(&pool_head_lock);
        !          1675:        do {
        !          1676:                if (drainpp == NULL) {
        !          1677:                        drainpp = LIST_FIRST(&pool_head);
        !          1678:                }
        !          1679:                if (drainpp != NULL) {
        !          1680:                        pp = drainpp;
        !          1681:                        drainpp = LIST_NEXT(pp, pr_poollist);
        !          1682:                }
        !          1683:                /*
        !          1684:                 * Skip completely idle pools.  We depend on at least
        !          1685:                 * one pool in the system being active.
        !          1686:                 */
        !          1687:        } while (pp == NULL || pp->pr_npages == 0);
        !          1688:        pp->pr_refcnt++;
        !          1689:        mutex_exit(&pool_head_lock);
        !          1690:
        !          1691:        /* If there is a pool_cache, drain CPU level caches. */
        !          1692:        *ppp = pp;
        !          1693:        if (pp->pr_cache != NULL) {
        !          1694:                *wp = xc_broadcast(0, (xcfunc_t)pool_cache_xcall,
        !          1695:                    pp->pr_cache, NULL);
        !          1696:        }
        !          1697: }
        !          1698:
        !          1699: void
        !          1700: pool_drain_end(struct pool *pp, uint64_t where)
        !          1701: {
        !          1702:
        !          1703:        if (pp == NULL)
        !          1704:                return;
        !          1705:
        !          1706:        KASSERT(pp->pr_refcnt > 0);
        !          1707:
        !          1708:        /* Wait for remote draining to complete. */
        !          1709:        if (pp->pr_cache != NULL)
        !          1710:                xc_wait(where);
        !          1711:
        !          1712:        /* Drain the cache (if any) and pool.. */
        !          1713:        pool_reclaim(pp);
        !          1714:
        !          1715:        /* Finally, unlock the pool. */
        !          1716:        mutex_enter(&pool_head_lock);
        !          1717:        pp->pr_refcnt--;
        !          1718:        cv_broadcast(&pool_busy);
        !          1719:        mutex_exit(&pool_head_lock);
1.3       pk       1720: }
                   1721:
                   1722: /*
                   1723:  * Diagnostic helpers.
                   1724:  */
                   1725: void
1.42      thorpej  1726: pool_print(struct pool *pp, const char *modif)
1.21      thorpej  1727: {
                   1728:
1.25      thorpej  1729:        pool_print1(pp, modif, printf);
1.21      thorpej  1730: }
                   1731:
1.25      thorpej  1732: void
1.108     yamt     1733: pool_printall(const char *modif, void (*pr)(const char *, ...))
                   1734: {
                   1735:        struct pool *pp;
                   1736:
                   1737:        LIST_FOREACH(pp, &pool_head, pr_poollist) {
                   1738:                pool_printit(pp, modif, pr);
                   1739:        }
                   1740: }
                   1741:
                   1742: void
1.42      thorpej  1743: pool_printit(struct pool *pp, const char *modif, void (*pr)(const char *, ...))
1.25      thorpej  1744: {
                   1745:
                   1746:        if (pp == NULL) {
                   1747:                (*pr)("Must specify a pool to print.\n");
                   1748:                return;
                   1749:        }
                   1750:
                   1751:        pool_print1(pp, modif, pr);
                   1752: }
                   1753:
1.21      thorpej  1754: static void
1.124     yamt     1755: pool_print_pagelist(struct pool *pp, struct pool_pagelist *pl,
1.97      yamt     1756:     void (*pr)(const char *, ...))
1.88      chs      1757: {
                   1758:        struct pool_item_header *ph;
                   1759: #ifdef DIAGNOSTIC
                   1760:        struct pool_item *pi;
                   1761: #endif
                   1762:
                   1763:        LIST_FOREACH(ph, pl, ph_pagelist) {
                   1764:                (*pr)("\t\tpage %p, nmissing %d, time %lu,%lu\n",
                   1765:                    ph->ph_page, ph->ph_nmissing,
                   1766:                    (u_long)ph->ph_time.tv_sec,
                   1767:                    (u_long)ph->ph_time.tv_usec);
                   1768: #ifdef DIAGNOSTIC
1.97      yamt     1769:                if (!(pp->pr_roflags & PR_NOTOUCH)) {
1.102     chs      1770:                        LIST_FOREACH(pi, &ph->ph_itemlist, pi_list) {
1.97      yamt     1771:                                if (pi->pi_magic != PI_MAGIC) {
                   1772:                                        (*pr)("\t\t\titem %p, magic 0x%x\n",
                   1773:                                            pi, pi->pi_magic);
                   1774:                                }
1.88      chs      1775:                        }
                   1776:                }
                   1777: #endif
                   1778:        }
                   1779: }
                   1780:
                   1781: static void
1.42      thorpej  1782: pool_print1(struct pool *pp, const char *modif, void (*pr)(const char *, ...))
1.3       pk       1783: {
1.25      thorpej  1784:        struct pool_item_header *ph;
1.131.2.2! matt     1785:        pool_cache_t pc;
        !          1786:        pcg_t *pcg;
        !          1787:        pool_cache_cpu_t *cc;
        !          1788:        uint64_t cpuhit, cpumiss;
1.44      thorpej  1789:        int i, print_log = 0, print_pagelist = 0, print_cache = 0;
1.25      thorpej  1790:        char c;
                   1791:
                   1792:        while ((c = *modif++) != '\0') {
                   1793:                if (c == 'l')
                   1794:                        print_log = 1;
                   1795:                if (c == 'p')
                   1796:                        print_pagelist = 1;
1.44      thorpej  1797:                if (c == 'c')
                   1798:                        print_cache = 1;
1.25      thorpej  1799:        }
                   1800:
1.131.2.2! matt     1801:        if ((pc = pp->pr_cache) != NULL) {
        !          1802:                (*pr)("POOL CACHE");
        !          1803:        } else {
        !          1804:                (*pr)("POOL");
        !          1805:        }
        !          1806:
        !          1807:        (*pr)(" %s: size %u, align %u, ioff %u, roflags 0x%08x\n",
1.25      thorpej  1808:            pp->pr_wchan, pp->pr_size, pp->pr_align, pp->pr_itemoffset,
                   1809:            pp->pr_roflags);
1.66      thorpej  1810:        (*pr)("\talloc %p\n", pp->pr_alloc);
1.25      thorpej  1811:        (*pr)("\tminitems %u, minpages %u, maxpages %u, npages %u\n",
                   1812:            pp->pr_minitems, pp->pr_minpages, pp->pr_maxpages, pp->pr_npages);
                   1813:        (*pr)("\titemsperpage %u, nitems %u, nout %u, hardlimit %u\n",
                   1814:            pp->pr_itemsperpage, pp->pr_nitems, pp->pr_nout, pp->pr_hardlimit);
                   1815:
1.131.2.2! matt     1816:        (*pr)("\tnget %lu, nfail %lu, nput %lu\n",
1.25      thorpej  1817:            pp->pr_nget, pp->pr_nfail, pp->pr_nput);
                   1818:        (*pr)("\tnpagealloc %lu, npagefree %lu, hiwat %u, nidle %lu\n",
                   1819:            pp->pr_npagealloc, pp->pr_npagefree, pp->pr_hiwat, pp->pr_nidle);
                   1820:
                   1821:        if (print_pagelist == 0)
                   1822:                goto skip_pagelist;
                   1823:
1.88      chs      1824:        if ((ph = LIST_FIRST(&pp->pr_emptypages)) != NULL)
                   1825:                (*pr)("\n\tempty page list:\n");
1.97      yamt     1826:        pool_print_pagelist(pp, &pp->pr_emptypages, pr);
1.88      chs      1827:        if ((ph = LIST_FIRST(&pp->pr_fullpages)) != NULL)
                   1828:                (*pr)("\n\tfull page list:\n");
1.97      yamt     1829:        pool_print_pagelist(pp, &pp->pr_fullpages, pr);
1.88      chs      1830:        if ((ph = LIST_FIRST(&pp->pr_partpages)) != NULL)
                   1831:                (*pr)("\n\tpartial-page list:\n");
1.97      yamt     1832:        pool_print_pagelist(pp, &pp->pr_partpages, pr);
1.88      chs      1833:
1.25      thorpej  1834:        if (pp->pr_curpage == NULL)
                   1835:                (*pr)("\tno current page\n");
                   1836:        else
                   1837:                (*pr)("\tcurpage %p\n", pp->pr_curpage->ph_page);
                   1838:
                   1839:  skip_pagelist:
                   1840:        if (print_log == 0)
                   1841:                goto skip_log;
                   1842:
                   1843:        (*pr)("\n");
                   1844:        if ((pp->pr_roflags & PR_LOGGING) == 0)
                   1845:                (*pr)("\tno log\n");
1.122     christos 1846:        else {
1.25      thorpej  1847:                pr_printlog(pp, NULL, pr);
1.122     christos 1848:        }
1.3       pk       1849:
1.25      thorpej  1850:  skip_log:
1.44      thorpej  1851:
1.102     chs      1852: #define PR_GROUPLIST(pcg)                                              \
                   1853:        (*pr)("\t\tgroup %p: avail %d\n", pcg, pcg->pcg_avail);         \
                   1854:        for (i = 0; i < PCG_NOBJECTS; i++) {                            \
                   1855:                if (pcg->pcg_objects[i].pcgo_pa !=                      \
                   1856:                    POOL_PADDR_INVALID) {                               \
                   1857:                        (*pr)("\t\t\t%p, 0x%llx\n",                     \
                   1858:                            pcg->pcg_objects[i].pcgo_va,                \
                   1859:                            (unsigned long long)                        \
                   1860:                            pcg->pcg_objects[i].pcgo_pa);               \
                   1861:                } else {                                                \
                   1862:                        (*pr)("\t\t\t%p\n",                             \
                   1863:                            pcg->pcg_objects[i].pcgo_va);               \
                   1864:                }                                                       \
                   1865:        }
                   1866:
1.131.2.2! matt     1867:        if (pc != NULL) {
        !          1868:                cpuhit = 0;
        !          1869:                cpumiss = 0;
        !          1870:                for (i = 0; i < MAXCPUS; i++) {
        !          1871:                        if ((cc = pc->pc_cpus[i]) == NULL)
        !          1872:                                continue;
        !          1873:                        cpuhit += cc->cc_hits;
        !          1874:                        cpumiss += cc->cc_misses;
        !          1875:                }
        !          1876:                (*pr)("\tcpu layer hits %llu misses %llu\n", cpuhit, cpumiss);
        !          1877:                (*pr)("\tcache layer hits %llu misses %llu\n",
        !          1878:                    pc->pc_hits, pc->pc_misses);
        !          1879:                (*pr)("\tcache layer entry uncontended %llu contended %llu\n",
        !          1880:                    pc->pc_hits + pc->pc_misses - pc->pc_contended,
        !          1881:                    pc->pc_contended);
        !          1882:                (*pr)("\tcache layer empty groups %u full groups %u\n",
        !          1883:                    pc->pc_nempty, pc->pc_nfull);
        !          1884:                if (print_cache) {
        !          1885:                        (*pr)("\tfull cache groups:\n");
        !          1886:                        for (pcg = pc->pc_fullgroups; pcg != NULL;
        !          1887:                            pcg = pcg->pcg_next) {
        !          1888:                                PR_GROUPLIST(pcg);
        !          1889:                        }
        !          1890:                        (*pr)("\tempty cache groups:\n");
        !          1891:                        for (pcg = pc->pc_emptygroups; pcg != NULL;
        !          1892:                            pcg = pcg->pcg_next) {
        !          1893:                                PR_GROUPLIST(pcg);
        !          1894:                        }
1.103     chs      1895:                }
1.44      thorpej  1896:        }
1.102     chs      1897: #undef PR_GROUPLIST
1.44      thorpej  1898:
1.88      chs      1899:        pr_enter_check(pp, pr);
                   1900: }
                   1901:
                   1902: static int
                   1903: pool_chk_page(struct pool *pp, const char *label, struct pool_item_header *ph)
                   1904: {
                   1905:        struct pool_item *pi;
1.128     christos 1906:        void *page;
1.88      chs      1907:        int n;
                   1908:
1.121     yamt     1909:        if ((pp->pr_roflags & PR_NOALIGN) == 0) {
1.128     christos 1910:                page = (void *)((uintptr_t)ph & pp->pr_alloc->pa_pagemask);
1.121     yamt     1911:                if (page != ph->ph_page &&
                   1912:                    (pp->pr_roflags & PR_PHINPAGE) != 0) {
                   1913:                        if (label != NULL)
                   1914:                                printf("%s: ", label);
                   1915:                        printf("pool(%p:%s): page inconsistency: page %p;"
                   1916:                               " at page head addr %p (p %p)\n", pp,
                   1917:                                pp->pr_wchan, ph->ph_page,
                   1918:                                ph, page);
                   1919:                        return 1;
                   1920:                }
1.88      chs      1921:        }
1.3       pk       1922:
1.97      yamt     1923:        if ((pp->pr_roflags & PR_NOTOUCH) != 0)
                   1924:                return 0;
                   1925:
1.102     chs      1926:        for (pi = LIST_FIRST(&ph->ph_itemlist), n = 0;
1.88      chs      1927:             pi != NULL;
1.102     chs      1928:             pi = LIST_NEXT(pi,pi_list), n++) {
1.88      chs      1929:
                   1930: #ifdef DIAGNOSTIC
                   1931:                if (pi->pi_magic != PI_MAGIC) {
                   1932:                        if (label != NULL)
                   1933:                                printf("%s: ", label);
                   1934:                        printf("pool(%s): free list modified: magic=%x;"
1.121     yamt     1935:                               " page %p; item ordinal %d; addr %p\n",
1.88      chs      1936:                                pp->pr_wchan, pi->pi_magic, ph->ph_page,
1.121     yamt     1937:                                n, pi);
1.88      chs      1938:                        panic("pool");
                   1939:                }
                   1940: #endif
1.121     yamt     1941:                if ((pp->pr_roflags & PR_NOALIGN) != 0) {
                   1942:                        continue;
                   1943:                }
1.128     christos 1944:                page = (void *)((uintptr_t)pi & pp->pr_alloc->pa_pagemask);
1.88      chs      1945:                if (page == ph->ph_page)
                   1946:                        continue;
                   1947:
                   1948:                if (label != NULL)
                   1949:                        printf("%s: ", label);
                   1950:                printf("pool(%p:%s): page inconsistency: page %p;"
                   1951:                       " item ordinal %d; addr %p (p %p)\n", pp,
                   1952:                        pp->pr_wchan, ph->ph_page,
                   1953:                        n, pi, page);
                   1954:                return 1;
                   1955:        }
                   1956:        return 0;
1.3       pk       1957: }
                   1958:
1.88      chs      1959:
1.3       pk       1960: int
1.42      thorpej  1961: pool_chk(struct pool *pp, const char *label)
1.3       pk       1962: {
                   1963:        struct pool_item_header *ph;
                   1964:        int r = 0;
                   1965:
1.131.2.2! matt     1966:        mutex_enter(&pp->pr_lock);
1.88      chs      1967:        LIST_FOREACH(ph, &pp->pr_emptypages, ph_pagelist) {
                   1968:                r = pool_chk_page(pp, label, ph);
                   1969:                if (r) {
                   1970:                        goto out;
                   1971:                }
                   1972:        }
                   1973:        LIST_FOREACH(ph, &pp->pr_fullpages, ph_pagelist) {
                   1974:                r = pool_chk_page(pp, label, ph);
                   1975:                if (r) {
1.3       pk       1976:                        goto out;
                   1977:                }
1.88      chs      1978:        }
                   1979:        LIST_FOREACH(ph, &pp->pr_partpages, ph_pagelist) {
                   1980:                r = pool_chk_page(pp, label, ph);
                   1981:                if (r) {
1.3       pk       1982:                        goto out;
                   1983:                }
                   1984:        }
1.88      chs      1985:
1.3       pk       1986: out:
1.131.2.2! matt     1987:        mutex_exit(&pp->pr_lock);
1.3       pk       1988:        return (r);
1.43      thorpej  1989: }
                   1990:
                   1991: /*
                   1992:  * pool_cache_init:
                   1993:  *
                   1994:  *     Initialize a pool cache.
1.131.2.2! matt     1995:  */
        !          1996: pool_cache_t
        !          1997: pool_cache_init(size_t size, u_int align, u_int align_offset, u_int flags,
        !          1998:     const char *wchan, struct pool_allocator *palloc, int ipl,
        !          1999:     int (*ctor)(void *, void *, int), void (*dtor)(void *, void *), void *arg)
        !          2000: {
        !          2001:        pool_cache_t pc;
        !          2002:
        !          2003:        pc = pool_get(&cache_pool, PR_WAITOK);
        !          2004:        if (pc == NULL)
        !          2005:                return NULL;
        !          2006:
        !          2007:        pool_cache_bootstrap(pc, size, align, align_offset, flags, wchan,
        !          2008:           palloc, ipl, ctor, dtor, arg);
        !          2009:
        !          2010:        return pc;
        !          2011: }
        !          2012:
        !          2013: /*
        !          2014:  * pool_cache_bootstrap:
1.43      thorpej  2015:  *
1.131.2.2! matt     2016:  *     Kernel-private version of pool_cache_init().  The caller
        !          2017:  *     provides initial storage.
1.43      thorpej  2018:  */
                   2019: void
1.131.2.2! matt     2020: pool_cache_bootstrap(pool_cache_t pc, size_t size, u_int align,
        !          2021:     u_int align_offset, u_int flags, const char *wchan,
        !          2022:     struct pool_allocator *palloc, int ipl,
        !          2023:     int (*ctor)(void *, void *, int), void (*dtor)(void *, void *),
1.43      thorpej  2024:     void *arg)
                   2025: {
1.131.2.2! matt     2026:        CPU_INFO_ITERATOR cii;
        !          2027:        struct cpu_info *ci;
        !          2028:        struct pool *pp;
        !          2029:
        !          2030:        pp = &pc->pc_pool;
        !          2031:        if (palloc == NULL && ipl == IPL_NONE)
        !          2032:                palloc = &pool_allocator_nointr;
        !          2033:        pool_init(pp, size, align, align_offset, flags, wchan, palloc, ipl);
1.43      thorpej  2034:
1.131.2.2! matt     2035:        mutex_init(&pc->pc_lock, MUTEX_DEFAULT, pp->pr_ipl);
1.43      thorpej  2036:
1.131.2.2! matt     2037:        if (ctor == NULL) {
        !          2038:                ctor = (int (*)(void *, void *, int))nullop;
        !          2039:        }
        !          2040:        if (dtor == NULL) {
        !          2041:                dtor = (void (*)(void *, void *))nullop;
        !          2042:        }
1.43      thorpej  2043:
1.131.2.2! matt     2044:        pc->pc_emptygroups = NULL;
        !          2045:        pc->pc_fullgroups = NULL;
        !          2046:        pc->pc_partgroups = NULL;
1.43      thorpej  2047:        pc->pc_ctor = ctor;
                   2048:        pc->pc_dtor = dtor;
                   2049:        pc->pc_arg  = arg;
1.131.2.2! matt     2050:        pc->pc_hits  = 0;
1.48      thorpej  2051:        pc->pc_misses = 0;
1.131.2.2! matt     2052:        pc->pc_nempty = 0;
        !          2053:        pc->pc_npart = 0;
        !          2054:        pc->pc_nfull = 0;
        !          2055:        pc->pc_contended = 0;
        !          2056:        pc->pc_refcnt = 0;
        !          2057:
        !          2058:        /* Allocate per-CPU caches. */
        !          2059:        memset(pc->pc_cpus, 0, sizeof(pc->pc_cpus));
        !          2060:        pc->pc_ncpu = 0;
        !          2061:        for (CPU_INFO_FOREACH(cii, ci)) {
        !          2062:                pool_cache_cpu_init1(ci, pc);
        !          2063:        }
        !          2064:
        !          2065:        if (__predict_true(!cold)) {
        !          2066:                mutex_enter(&pp->pr_lock);
        !          2067:                pp->pr_cache = pc;
        !          2068:                mutex_exit(&pp->pr_lock);
        !          2069:                mutex_enter(&pool_head_lock);
        !          2070:                LIST_INSERT_HEAD(&pool_cache_head, pc, pc_cachelist);
        !          2071:                mutex_exit(&pool_head_lock);
        !          2072:        } else {
        !          2073:                pp->pr_cache = pc;
        !          2074:                LIST_INSERT_HEAD(&pool_cache_head, pc, pc_cachelist);
        !          2075:        }
1.43      thorpej  2076: }
                   2077:
                   2078: /*
                   2079:  * pool_cache_destroy:
                   2080:  *
                   2081:  *     Destroy a pool cache.
                   2082:  */
                   2083: void
1.131.2.2! matt     2084: pool_cache_destroy(pool_cache_t pc)
1.43      thorpej  2085: {
1.131.2.2! matt     2086:        struct pool *pp = &pc->pc_pool;
        !          2087:        pool_cache_cpu_t *cc;
        !          2088:        pcg_t *pcg;
        !          2089:        int i;
        !          2090:
        !          2091:        /* Remove it from the global list. */
        !          2092:        mutex_enter(&pool_head_lock);
        !          2093:        while (pc->pc_refcnt != 0)
        !          2094:                cv_wait(&pool_busy, &pool_head_lock);
        !          2095:        LIST_REMOVE(pc, pc_cachelist);
        !          2096:        mutex_exit(&pool_head_lock);
1.43      thorpej  2097:
                   2098:        /* First, invalidate the entire cache. */
                   2099:        pool_cache_invalidate(pc);
                   2100:
1.131.2.2! matt     2101:        /* Disassociate it from the pool. */
        !          2102:        mutex_enter(&pp->pr_lock);
        !          2103:        pp->pr_cache = NULL;
        !          2104:        mutex_exit(&pp->pr_lock);
        !          2105:
        !          2106:        /* Destroy per-CPU data */
        !          2107:        for (i = 0; i < MAXCPUS; i++) {
        !          2108:                if ((cc = pc->pc_cpus[i]) == NULL)
        !          2109:                        continue;
        !          2110:                if ((pcg = cc->cc_current) != NULL) {
        !          2111:                        pcg->pcg_next = NULL;
        !          2112:                        pool_cache_invalidate_groups(pc, pcg);
        !          2113:                }
        !          2114:                if ((pcg = cc->cc_previous) != NULL) {
        !          2115:                        pcg->pcg_next = NULL;
        !          2116:                        pool_cache_invalidate_groups(pc, pcg);
        !          2117:                }
        !          2118:                if (cc != &pc->pc_cpu0)
        !          2119:                        pool_put(&cache_cpu_pool, cc);
        !          2120:        }
        !          2121:
        !          2122:        /* Finally, destroy it. */
        !          2123:        mutex_destroy(&pc->pc_lock);
        !          2124:        pool_destroy(pp);
        !          2125:        pool_put(&cache_pool, pc);
1.43      thorpej  2126: }
                   2127:
1.131.2.2! matt     2128: /*
        !          2129:  * pool_cache_cpu_init1:
        !          2130:  *
        !          2131:  *     Called for each pool_cache whenever a new CPU is attached.
        !          2132:  */
        !          2133: static void
        !          2134: pool_cache_cpu_init1(struct cpu_info *ci, pool_cache_t pc)
1.43      thorpej  2135: {
1.131.2.2! matt     2136:        pool_cache_cpu_t *cc;
1.43      thorpej  2137:
1.131.2.2! matt     2138:        KASSERT(((uintptr_t)pc->pc_cpus & (CACHE_LINE_SIZE - 1)) == 0);
1.43      thorpej  2139:
1.131.2.2! matt     2140:        if ((cc = pc->pc_cpus[ci->ci_index]) != NULL) {
        !          2141:                KASSERT(cc->cc_cpu = ci);
        !          2142:                return;
        !          2143:        }
        !          2144:
        !          2145:        /*
        !          2146:         * The first CPU is 'free'.  This needs to be the case for
        !          2147:         * bootstrap - we may not be able to allocate yet.
        !          2148:         */
        !          2149:        if (pc->pc_ncpu == 0) {
        !          2150:                cc = &pc->pc_cpu0;
        !          2151:                pc->pc_ncpu = 1;
        !          2152:        } else {
        !          2153:                mutex_enter(&pc->pc_lock);
        !          2154:                pc->pc_ncpu++;
        !          2155:                mutex_exit(&pc->pc_lock);
        !          2156:                cc = pool_get(&cache_cpu_pool, PR_WAITOK);
        !          2157:        }
        !          2158:
        !          2159:        cc->cc_ipl = pc->pc_pool.pr_ipl;
        !          2160:        cc->cc_iplcookie = makeiplcookie(cc->cc_ipl);
        !          2161:        cc->cc_cache = pc;
        !          2162:        cc->cc_cpu = ci;
        !          2163:        cc->cc_hits = 0;
        !          2164:        cc->cc_misses = 0;
        !          2165:        cc->cc_current = NULL;
        !          2166:        cc->cc_previous = NULL;
        !          2167:
        !          2168:        pc->pc_cpus[ci->ci_index] = cc;
1.43      thorpej  2169: }
                   2170:
1.131.2.2! matt     2171: /*
        !          2172:  * pool_cache_cpu_init:
        !          2173:  *
        !          2174:  *     Called whenever a new CPU is attached.
        !          2175:  */
        !          2176: void
        !          2177: pool_cache_cpu_init(struct cpu_info *ci)
        !          2178: {
        !          2179:        pool_cache_t pc;
        !          2180:
        !          2181:        mutex_enter(&pool_head_lock);
        !          2182:        LIST_FOREACH(pc, &pool_cache_head, pc_cachelist) {
        !          2183:                pc->pc_refcnt++;
        !          2184:                mutex_exit(&pool_head_lock);
        !          2185:
        !          2186:                pool_cache_cpu_init1(ci, pc);
        !          2187:
        !          2188:                mutex_enter(&pool_head_lock);
        !          2189:                pc->pc_refcnt--;
        !          2190:                cv_broadcast(&pool_busy);
        !          2191:        }
        !          2192:        mutex_exit(&pool_head_lock);
        !          2193: }
        !          2194:
        !          2195: /*
        !          2196:  * pool_cache_reclaim:
        !          2197:  *
        !          2198:  *     Reclaim memory from a pool cache.
        !          2199:  */
        !          2200: bool
        !          2201: pool_cache_reclaim(pool_cache_t pc)
1.43      thorpej  2202: {
                   2203:
1.131.2.2! matt     2204:        return pool_reclaim(&pc->pc_pool);
        !          2205: }
1.43      thorpej  2206:
1.131.2.2! matt     2207: /*
        !          2208:  * pool_cache_destruct_object:
        !          2209:  *
        !          2210:  *     Force destruction of an object and its release back into
        !          2211:  *     the pool.
        !          2212:  */
        !          2213: void
        !          2214: pool_cache_destruct_object(pool_cache_t pc, void *object)
        !          2215: {
        !          2216:
        !          2217:        (*pc->pc_dtor)(pc->pc_arg, object);
        !          2218:        pool_put(&pc->pc_pool, object);
1.43      thorpej  2219: }
                   2220:
1.131.2.2! matt     2221: /*
        !          2222:  * pool_cache_invalidate_groups:
        !          2223:  *
        !          2224:  *     Invalidate a chain of groups and destruct all objects.
        !          2225:  */
1.102     chs      2226: static void
1.131.2.2! matt     2227: pool_cache_invalidate_groups(pool_cache_t pc, pcg_t *pcg)
1.102     chs      2228: {
1.131.2.2! matt     2229:        void *object;
        !          2230:        pcg_t *next;
        !          2231:        int i;
        !          2232:
        !          2233:        for (; pcg != NULL; pcg = next) {
        !          2234:                next = pcg->pcg_next;
        !          2235:
        !          2236:                for (i = 0; i < pcg->pcg_avail; i++) {
        !          2237:                        object = pcg->pcg_objects[i].pcgo_va;
        !          2238:                        pool_cache_destruct_object(pc, object);
        !          2239:                }
1.102     chs      2240:
                   2241:                pool_put(&pcgpool, pcg);
                   2242:        }
                   2243: }
                   2244:
1.43      thorpej  2245: /*
1.131.2.2! matt     2246:  * pool_cache_invalidate:
1.43      thorpej  2247:  *
1.131.2.2! matt     2248:  *     Invalidate a pool cache (destruct and release all of the
        !          2249:  *     cached objects).  Does not reclaim objects from the pool.
1.43      thorpej  2250:  */
1.131.2.2! matt     2251: void
        !          2252: pool_cache_invalidate(pool_cache_t pc)
1.43      thorpej  2253: {
1.131.2.2! matt     2254:        pcg_t *full, *empty, *part;
1.58      thorpej  2255:
1.131.2.2! matt     2256:        mutex_enter(&pc->pc_lock);
        !          2257:        full = pc->pc_fullgroups;
        !          2258:        empty = pc->pc_emptygroups;
        !          2259:        part = pc->pc_partgroups;
        !          2260:        pc->pc_fullgroups = NULL;
        !          2261:        pc->pc_emptygroups = NULL;
        !          2262:        pc->pc_partgroups = NULL;
        !          2263:        pc->pc_nfull = 0;
        !          2264:        pc->pc_nempty = 0;
        !          2265:        pc->pc_npart = 0;
        !          2266:        mutex_exit(&pc->pc_lock);
1.43      thorpej  2267:
1.131.2.2! matt     2268:        pool_cache_invalidate_groups(pc, full);
        !          2269:        pool_cache_invalidate_groups(pc, empty);
        !          2270:        pool_cache_invalidate_groups(pc, part);
        !          2271: }
1.43      thorpej  2272:
1.131.2.2! matt     2273: void
        !          2274: pool_cache_set_drain_hook(pool_cache_t pc, void (*fn)(void *, int), void *arg)
        !          2275: {
1.43      thorpej  2276:
1.131.2.2! matt     2277:        pool_set_drain_hook(&pc->pc_pool, fn, arg);
        !          2278: }
1.125     ad       2279:
1.131.2.2! matt     2280: void
        !          2281: pool_cache_setlowat(pool_cache_t pc, int n)
        !          2282: {
1.43      thorpej  2283:
1.131.2.2! matt     2284:        pool_setlowat(&pc->pc_pool, n);
        !          2285: }
1.43      thorpej  2286:
1.131.2.2! matt     2287: void
        !          2288: pool_cache_sethiwat(pool_cache_t pc, int n)
        !          2289: {
1.43      thorpej  2290:
1.131.2.2! matt     2291:        pool_sethiwat(&pc->pc_pool, n);
1.43      thorpej  2292: }
                   2293:
                   2294: void
1.131.2.2! matt     2295: pool_cache_sethardlimit(pool_cache_t pc, int n, const char *warnmess, int ratecap)
1.43      thorpej  2296: {
                   2297:
1.131.2.2! matt     2298:        pool_sethardlimit(&pc->pc_pool, n, warnmess, ratecap);
        !          2299: }
        !          2300:
        !          2301: static inline pool_cache_cpu_t *
        !          2302: pool_cache_cpu_enter(pool_cache_t pc, int *s)
        !          2303: {
        !          2304:        pool_cache_cpu_t *cc;
        !          2305:        struct cpu_info *ci;
1.125     ad       2306:
1.131.2.2! matt     2307:        /*
        !          2308:         * Prevent other users of the cache from accessing our
        !          2309:         * CPU-local data.  To avoid touching shared state, we
        !          2310:         * pull the neccessary information from CPU local data.
        !          2311:         */
        !          2312:        ci = curcpu();
        !          2313:        KASSERT(ci->ci_data.cpu_index < MAXCPUS);
        !          2314:        cc = pc->pc_cpus[ci->ci_data.cpu_index];
        !          2315:        KASSERT(cc->cc_cache == pc);
        !          2316:        if (cc->cc_ipl == IPL_NONE) {
        !          2317:                crit_enter();
        !          2318:        } else {
        !          2319:                *s = splraiseipl(cc->cc_iplcookie);
1.109     christos 2320:        }
                   2321:
1.131.2.2! matt     2322:        /* Moved to another CPU before disabling preemption? */
        !          2323:        if (__predict_false(ci != curcpu())) {
        !          2324:                ci = curcpu();
        !          2325:                cc = pc->pc_cpus[ci->ci_data.cpu_index];
        !          2326:        }
1.43      thorpej  2327:
1.131.2.2! matt     2328: #ifdef DIAGNOSTIC
        !          2329:        KASSERT(cc->cc_cpu == ci);
        !          2330:        KASSERT(((uintptr_t)cc & (CACHE_LINE_SIZE - 1)) == 0);
        !          2331: #endif
        !          2332:
        !          2333:        return cc;
        !          2334: }
        !          2335:
        !          2336: static inline void
        !          2337: pool_cache_cpu_exit(pool_cache_cpu_t *cc, int *s)
        !          2338: {
        !          2339:
        !          2340:        /* No longer need exclusive access to the per-CPU data. */
        !          2341:        if (cc->cc_ipl == IPL_NONE) {
        !          2342:                crit_exit();
        !          2343:        } else {
        !          2344:                splx(*s);
1.102     chs      2345:        }
1.131.2.2! matt     2346: }
        !          2347:
        !          2348: #if __GNUC_PREREQ__(3, 0)
        !          2349: __attribute ((noinline))
        !          2350: #endif
        !          2351: pool_cache_cpu_t *
        !          2352: pool_cache_get_slow(pool_cache_cpu_t *cc, int *s, void **objectp,
        !          2353:                    paddr_t *pap, int flags)
        !          2354: {
        !          2355:        pcg_t *pcg, *cur;
        !          2356:        uint64_t ncsw;
        !          2357:        pool_cache_t pc;
        !          2358:        void *object;
        !          2359:
        !          2360:        pc = cc->cc_cache;
        !          2361:        cc->cc_misses++;
        !          2362:
        !          2363:        /*
        !          2364:         * Nothing was available locally.  Try and grab a group
        !          2365:         * from the cache.
        !          2366:         */
        !          2367:        if (!mutex_tryenter(&pc->pc_lock)) {
        !          2368:                ncsw = curlwp->l_ncsw;
        !          2369:                mutex_enter(&pc->pc_lock);
        !          2370:                pc->pc_contended++;
1.43      thorpej  2371:
                   2372:                /*
1.131.2.2! matt     2373:                 * If we context switched while locking, then
        !          2374:                 * our view of the per-CPU data is invalid:
        !          2375:                 * retry.
1.43      thorpej  2376:                 */
1.131.2.2! matt     2377:                if (curlwp->l_ncsw != ncsw) {
        !          2378:                        mutex_exit(&pc->pc_lock);
        !          2379:                        pool_cache_cpu_exit(cc, s);
        !          2380:                        return pool_cache_cpu_enter(pc, s);
        !          2381:                }
        !          2382:        }
1.102     chs      2383:
1.131.2.2! matt     2384:        if ((pcg = pc->pc_fullgroups) != NULL) {
        !          2385:                /*
        !          2386:                 * If there's a full group, release our empty
        !          2387:                 * group back to the cache.  Install the full
        !          2388:                 * group as cc_current and return.
        !          2389:                 */
        !          2390:                if ((cur = cc->cc_current) != NULL) {
        !          2391:                        KASSERT(cur->pcg_avail == 0);
        !          2392:                        cur->pcg_next = pc->pc_emptygroups;
        !          2393:                        pc->pc_emptygroups = cur;
        !          2394:                        pc->pc_nempty++;
1.43      thorpej  2395:                }
1.131.2.2! matt     2396:                KASSERT(pcg->pcg_avail == PCG_NOBJECTS);
        !          2397:                cc->cc_current = pcg;
        !          2398:                pc->pc_fullgroups = pcg->pcg_next;
        !          2399:                pc->pc_hits++;
        !          2400:                pc->pc_nfull--;
        !          2401:                mutex_exit(&pc->pc_lock);
        !          2402:                return cc;
1.43      thorpej  2403:        }
                   2404:
1.131.2.2! matt     2405:        /*
        !          2406:         * Nothing available locally or in cache.  Take the slow
        !          2407:         * path: fetch a new object from the pool and construct
        !          2408:         * it.
        !          2409:         */
        !          2410:        pc->pc_misses++;
        !          2411:        mutex_exit(&pc->pc_lock);
        !          2412:        pool_cache_cpu_exit(cc, s);
        !          2413:
        !          2414:        object = pool_get(&pc->pc_pool, flags);
        !          2415:        *objectp = object;
        !          2416:        if (object == NULL)
        !          2417:                return NULL;
1.43      thorpej  2418:
1.131.2.2! matt     2419:        if ((*pc->pc_ctor)(pc->pc_arg, object, flags) != 0) {
        !          2420:                pool_put(&pc->pc_pool, object);
        !          2421:                *objectp = NULL;
        !          2422:                return NULL;
1.102     chs      2423:        }
1.51      thorpej  2424:
1.131.2.2! matt     2425:        KASSERT((((vaddr_t)object + pc->pc_pool.pr_itemoffset) &
        !          2426:            (pc->pc_pool.pr_align - 1)) == 0);
        !          2427:
        !          2428:        if (pap != NULL) {
        !          2429: #ifdef POOL_VTOPHYS
        !          2430:                *pap = POOL_VTOPHYS(object);
        !          2431: #else
        !          2432:                *pap = POOL_PADDR_INVALID;
        !          2433: #endif
        !          2434:        }
1.51      thorpej  2435:
1.131.2.2! matt     2436:        FREECHECK_OUT(&pc->pc_freecheck, object);
        !          2437:        return NULL;
1.43      thorpej  2438: }
                   2439:
1.130     ad       2440: /*
1.131.2.2! matt     2441:  * pool_cache_get{,_paddr}:
1.130     ad       2442:  *
1.131.2.2! matt     2443:  *     Get an object from a pool cache (optionally returning
        !          2444:  *     the physical address of the object).
1.130     ad       2445:  */
1.131.2.2! matt     2446: void *
        !          2447: pool_cache_get_paddr(pool_cache_t pc, int flags, paddr_t *pap)
1.102     chs      2448: {
1.131.2.2! matt     2449:        pool_cache_cpu_t *cc;
        !          2450:        pcg_t *pcg;
1.102     chs      2451:        void *object;
1.131.2.2! matt     2452:        int s;
1.102     chs      2453:
1.131.2.2! matt     2454: #ifdef LOCKDEBUG
        !          2455:        if (flags & PR_WAITOK)
        !          2456:                ASSERT_SLEEPABLE(NULL, "pool_cache_get(PR_WAITOK)");
        !          2457: #endif
1.130     ad       2458:
1.131.2.2! matt     2459:        cc = pool_cache_cpu_enter(pc, &s);
        !          2460:        do {
        !          2461:                /* Try and allocate an object from the current group. */
        !          2462:                pcg = cc->cc_current;
        !          2463:                if (pcg != NULL && pcg->pcg_avail > 0) {
        !          2464:                        object = pcg->pcg_objects[--pcg->pcg_avail].pcgo_va;
        !          2465:                        if (pap != NULL)
        !          2466:                                *pap = pcg->pcg_objects[pcg->pcg_avail].pcgo_pa;
        !          2467:                        pcg->pcg_objects[pcg->pcg_avail].pcgo_va = NULL;
        !          2468:                        KASSERT(pcg->pcg_avail <= PCG_NOBJECTS);
        !          2469:                        KASSERT(object != NULL);
        !          2470:                        cc->cc_hits++;
        !          2471:                        pool_cache_cpu_exit(cc, &s);
        !          2472:                        FREECHECK_OUT(&pc->pc_freecheck, object);
        !          2473:                        return object;
1.102     chs      2474:                }
1.130     ad       2475:
1.131.2.2! matt     2476:                /*
        !          2477:                 * That failed.  If the previous group isn't empty, swap
        !          2478:                 * it with the current group and allocate from there.
        !          2479:                 */
        !          2480:                pcg = cc->cc_previous;
        !          2481:                if (pcg != NULL && pcg->pcg_avail > 0) {
        !          2482:                        cc->cc_previous = cc->cc_current;
        !          2483:                        cc->cc_current = pcg;
        !          2484:                        continue;
        !          2485:                }
        !          2486:
        !          2487:                /*
        !          2488:                 * Can't allocate from either group: try the slow path.
        !          2489:                 * If get_slow() allocated an object for us, or if
        !          2490:                 * no more objects are available, it will return NULL.
        !          2491:                 * Otherwise, we need to retry.
        !          2492:                 */
        !          2493:                cc = pool_cache_get_slow(cc, &s, &object, pap, flags);
        !          2494:        } while (cc != NULL);
        !          2495:
        !          2496:        return object;
1.105     christos 2497: }
                   2498:
1.131.2.2! matt     2499: #if __GNUC_PREREQ__(3, 0)
        !          2500: __attribute ((noinline))
        !          2501: #endif
        !          2502: pool_cache_cpu_t *
        !          2503: pool_cache_put_slow(pool_cache_cpu_t *cc, int *s, void *object, paddr_t pa)
1.105     christos 2504: {
1.131.2.2! matt     2505:        pcg_t *pcg, *cur;
        !          2506:        uint64_t ncsw;
        !          2507:        pool_cache_t pc;
1.105     christos 2508:
1.131.2.2! matt     2509:        pc = cc->cc_cache;
        !          2510:        cc->cc_misses++;
1.105     christos 2511:
1.131.2.2! matt     2512:        /*
        !          2513:         * No free slots locally.  Try to grab an empty, unused
        !          2514:         * group from the cache.
        !          2515:         */
        !          2516:        if (!mutex_tryenter(&pc->pc_lock)) {
        !          2517:                ncsw = curlwp->l_ncsw;
        !          2518:                mutex_enter(&pc->pc_lock);
        !          2519:                pc->pc_contended++;
1.103     chs      2520:
1.131.2.2! matt     2521:                /*
        !          2522:                 * If we context switched while locking, then
        !          2523:                 * our view of the per-CPU data is invalid:
        !          2524:                 * retry.
        !          2525:                 */
        !          2526:                if (curlwp->l_ncsw != ncsw) {
        !          2527:                        mutex_exit(&pc->pc_lock);
        !          2528:                        pool_cache_cpu_exit(cc, s);
        !          2529:                        return pool_cache_cpu_enter(pc, s);
        !          2530:                }
        !          2531:        }
        !          2532:
        !          2533:        if ((pcg = pc->pc_emptygroups) != NULL) {
        !          2534:                /*
        !          2535:                 * If there's a empty group, release our full
        !          2536:                 * group back to the cache.  Install the empty
        !          2537:                 * group as cc_current and return.
        !          2538:                 */
        !          2539:                if ((cur = cc->cc_current) != NULL) {
        !          2540:                        KASSERT(cur->pcg_avail == PCG_NOBJECTS);
        !          2541:                        cur->pcg_next = pc->pc_fullgroups;
        !          2542:                        pc->pc_fullgroups = cur;
        !          2543:                        pc->pc_nfull++;
        !          2544:                }
        !          2545:                KASSERT(pcg->pcg_avail == 0);
        !          2546:                cc->cc_current = pcg;
        !          2547:                pc->pc_emptygroups = pcg->pcg_next;
        !          2548:                pc->pc_hits++;
        !          2549:                pc->pc_nempty--;
        !          2550:                mutex_exit(&pc->pc_lock);
        !          2551:                return cc;
        !          2552:        }
        !          2553:
        !          2554:        /*
        !          2555:         * Nothing available locally or in cache.  Take the
        !          2556:         * slow path and try to allocate a new group that we
        !          2557:         * can release to.
        !          2558:         */
        !          2559:        pc->pc_misses++;
        !          2560:        mutex_exit(&pc->pc_lock);
        !          2561:        pool_cache_cpu_exit(cc, s);
        !          2562:
        !          2563:        /*
        !          2564:         * If we can't allocate a new group, just throw the
        !          2565:         * object away.
        !          2566:         */
        !          2567:        pcg = pool_get(&pcgpool, PR_NOWAIT);
        !          2568:        if (pcg == NULL) {
        !          2569:                pool_cache_destruct_object(pc, object);
        !          2570:                return NULL;
        !          2571:        }
        !          2572: #ifdef DIAGNOSTIC
        !          2573:        memset(pcg, 0, sizeof(*pcg));
        !          2574: #else
        !          2575:        pcg->pcg_avail = 0;
        !          2576: #endif
        !          2577:
        !          2578:        /*
        !          2579:         * Add the empty group to the cache and try again.
        !          2580:         */
        !          2581:        mutex_enter(&pc->pc_lock);
        !          2582:        pcg->pcg_next = pc->pc_emptygroups;
        !          2583:        pc->pc_emptygroups = pcg;
        !          2584:        pc->pc_nempty++;
        !          2585:        mutex_exit(&pc->pc_lock);
        !          2586:
        !          2587:        return pool_cache_cpu_enter(pc, s);
        !          2588: }
1.102     chs      2589:
1.43      thorpej  2590: /*
1.131.2.2! matt     2591:  * pool_cache_put{,_paddr}:
1.43      thorpej  2592:  *
1.131.2.2! matt     2593:  *     Put an object back to the pool cache (optionally caching the
        !          2594:  *     physical address of the object).
1.43      thorpej  2595:  */
1.101     thorpej  2596: void
1.131.2.2! matt     2597: pool_cache_put_paddr(pool_cache_t pc, void *object, paddr_t pa)
1.43      thorpej  2598: {
1.131.2.2! matt     2599:        pool_cache_cpu_t *cc;
        !          2600:        pcg_t *pcg;
        !          2601:        int s;
1.101     thorpej  2602:
1.131.2.2! matt     2603:        FREECHECK_IN(&pc->pc_freecheck, object);
1.43      thorpej  2604:
1.131.2.2! matt     2605:        cc = pool_cache_cpu_enter(pc, &s);
        !          2606:        do {
        !          2607:                /* If the current group isn't full, release it there. */
        !          2608:                pcg = cc->cc_current;
        !          2609:                if (pcg != NULL && pcg->pcg_avail < PCG_NOBJECTS) {
        !          2610:                        KASSERT(pcg->pcg_objects[pcg->pcg_avail].pcgo_va
        !          2611:                            == NULL);
        !          2612:                        pcg->pcg_objects[pcg->pcg_avail].pcgo_va = object;
        !          2613:                        pcg->pcg_objects[pcg->pcg_avail].pcgo_pa = pa;
        !          2614:                        pcg->pcg_avail++;
        !          2615:                        cc->cc_hits++;
        !          2616:                        pool_cache_cpu_exit(cc, &s);
        !          2617:                        return;
        !          2618:                }
1.43      thorpej  2619:
1.131.2.2! matt     2620:                /*
        !          2621:                 * That failed.  If the previous group is empty, swap
        !          2622:                 * it with the current group and try again.
        !          2623:                 */
        !          2624:                pcg = cc->cc_previous;
        !          2625:                if (pcg != NULL && pcg->pcg_avail == 0) {
        !          2626:                        cc->cc_previous = cc->cc_current;
        !          2627:                        cc->cc_current = pcg;
        !          2628:                        continue;
        !          2629:                }
1.43      thorpej  2630:
1.131.2.2! matt     2631:                /*
        !          2632:                 * Can't free to either group: try the slow path.
        !          2633:                 * If put_slow() releases the object for us, it
        !          2634:                 * will return NULL.  Otherwise we need to retry.
        !          2635:                 */
        !          2636:                cc = pool_cache_put_slow(cc, &s, object, pa);
        !          2637:        } while (cc != NULL);
1.43      thorpej  2638: }
                   2639:
                   2640: /*
1.131.2.2! matt     2641:  * pool_cache_xcall:
1.43      thorpej  2642:  *
1.131.2.2! matt     2643:  *     Transfer objects from the per-CPU cache to the global cache.
        !          2644:  *     Run within a cross-call thread.
1.43      thorpej  2645:  */
                   2646: static void
1.131.2.2! matt     2647: pool_cache_xcall(pool_cache_t pc)
1.43      thorpej  2648: {
1.131.2.2! matt     2649:        pool_cache_cpu_t *cc;
        !          2650:        pcg_t *prev, *cur, **list;
        !          2651:        int s = 0; /* XXXgcc */
        !          2652:
        !          2653:        cc = pool_cache_cpu_enter(pc, &s);
        !          2654:        cur = cc->cc_current;
        !          2655:        cc->cc_current = NULL;
        !          2656:        prev = cc->cc_previous;
        !          2657:        cc->cc_previous = NULL;
        !          2658:        pool_cache_cpu_exit(cc, &s);
        !          2659:
        !          2660:        /*
        !          2661:         * XXXSMP Go to splvm to prevent kernel_lock from being taken,
        !          2662:         * because locks at IPL_SOFTXXX are still spinlocks.  Does not
        !          2663:         * apply to IPL_SOFTBIO.  Cross-call threads do not take the
        !          2664:         * kernel_lock.
1.101     thorpej  2665:         */
1.131.2.2! matt     2666:        s = splvm();
        !          2667:        mutex_enter(&pc->pc_lock);
        !          2668:        if (cur != NULL) {
        !          2669:                if (cur->pcg_avail == PCG_NOBJECTS) {
        !          2670:                        list = &pc->pc_fullgroups;
        !          2671:                        pc->pc_nfull++;
        !          2672:                } else if (cur->pcg_avail == 0) {
        !          2673:                        list = &pc->pc_emptygroups;
        !          2674:                        pc->pc_nempty++;
        !          2675:                } else {
        !          2676:                        list = &pc->pc_partgroups;
        !          2677:                        pc->pc_npart++;
        !          2678:                }
        !          2679:                cur->pcg_next = *list;
        !          2680:                *list = cur;
        !          2681:        }
        !          2682:        if (prev != NULL) {
        !          2683:                if (prev->pcg_avail == PCG_NOBJECTS) {
        !          2684:                        list = &pc->pc_fullgroups;
        !          2685:                        pc->pc_nfull++;
        !          2686:                } else if (prev->pcg_avail == 0) {
        !          2687:                        list = &pc->pc_emptygroups;
        !          2688:                        pc->pc_nempty++;
        !          2689:                } else {
        !          2690:                        list = &pc->pc_partgroups;
        !          2691:                        pc->pc_npart++;
        !          2692:                }
        !          2693:                prev->pcg_next = *list;
        !          2694:                *list = prev;
        !          2695:        }
        !          2696:        mutex_exit(&pc->pc_lock);
        !          2697:        splx(s);
1.3       pk       2698: }
1.66      thorpej  2699:
                   2700: /*
                   2701:  * Pool backend allocators.
                   2702:  *
                   2703:  * Each pool has a backend allocator that handles allocation, deallocation,
                   2704:  * and any additional draining that might be needed.
                   2705:  *
                   2706:  * We provide two standard allocators:
                   2707:  *
                   2708:  *     pool_allocator_kmem - the default when no allocator is specified
                   2709:  *
                   2710:  *     pool_allocator_nointr - used for pools that will not be accessed
                   2711:  *     in interrupt context.
                   2712:  */
                   2713: void   *pool_page_alloc(struct pool *, int);
                   2714: void   pool_page_free(struct pool *, void *);
                   2715:
1.112     bjh21    2716: #ifdef POOL_SUBPAGE
                   2717: struct pool_allocator pool_allocator_kmem_fullpage = {
                   2718:        pool_page_alloc, pool_page_free, 0,
1.117     yamt     2719:        .pa_backingmapptr = &kmem_map,
1.112     bjh21    2720: };
                   2721: #else
1.66      thorpej  2722: struct pool_allocator pool_allocator_kmem = {
                   2723:        pool_page_alloc, pool_page_free, 0,
1.117     yamt     2724:        .pa_backingmapptr = &kmem_map,
1.66      thorpej  2725: };
1.112     bjh21    2726: #endif
1.66      thorpej  2727:
                   2728: void   *pool_page_alloc_nointr(struct pool *, int);
                   2729: void   pool_page_free_nointr(struct pool *, void *);
                   2730:
1.112     bjh21    2731: #ifdef POOL_SUBPAGE
                   2732: struct pool_allocator pool_allocator_nointr_fullpage = {
                   2733:        pool_page_alloc_nointr, pool_page_free_nointr, 0,
1.117     yamt     2734:        .pa_backingmapptr = &kernel_map,
1.112     bjh21    2735: };
                   2736: #else
1.66      thorpej  2737: struct pool_allocator pool_allocator_nointr = {
                   2738:        pool_page_alloc_nointr, pool_page_free_nointr, 0,
1.117     yamt     2739:        .pa_backingmapptr = &kernel_map,
1.66      thorpej  2740: };
1.112     bjh21    2741: #endif
1.66      thorpej  2742:
                   2743: #ifdef POOL_SUBPAGE
                   2744: void   *pool_subpage_alloc(struct pool *, int);
                   2745: void   pool_subpage_free(struct pool *, void *);
                   2746:
1.112     bjh21    2747: struct pool_allocator pool_allocator_kmem = {
                   2748:        pool_subpage_alloc, pool_subpage_free, POOL_SUBPAGE,
1.117     yamt     2749:        .pa_backingmapptr = &kmem_map,
1.112     bjh21    2750: };
                   2751:
                   2752: void   *pool_subpage_alloc_nointr(struct pool *, int);
                   2753: void   pool_subpage_free_nointr(struct pool *, void *);
                   2754:
                   2755: struct pool_allocator pool_allocator_nointr = {
                   2756:        pool_subpage_alloc, pool_subpage_free, POOL_SUBPAGE,
1.117     yamt     2757:        .pa_backingmapptr = &kmem_map,
1.66      thorpej  2758: };
                   2759: #endif /* POOL_SUBPAGE */
                   2760:
1.117     yamt     2761: static void *
                   2762: pool_allocator_alloc(struct pool *pp, int flags)
1.66      thorpej  2763: {
1.117     yamt     2764:        struct pool_allocator *pa = pp->pr_alloc;
1.66      thorpej  2765:        void *res;
                   2766:
1.117     yamt     2767:        res = (*pa->pa_alloc)(pp, flags);
                   2768:        if (res == NULL && (flags & PR_WAITOK) == 0) {
1.66      thorpej  2769:                /*
1.117     yamt     2770:                 * We only run the drain hook here if PR_NOWAIT.
                   2771:                 * In other cases, the hook will be run in
                   2772:                 * pool_reclaim().
1.66      thorpej  2773:                 */
1.117     yamt     2774:                if (pp->pr_drain_hook != NULL) {
                   2775:                        (*pp->pr_drain_hook)(pp->pr_drain_hook_arg, flags);
                   2776:                        res = (*pa->pa_alloc)(pp, flags);
1.66      thorpej  2777:                }
1.117     yamt     2778:        }
                   2779:        return res;
1.66      thorpej  2780: }
                   2781:
1.117     yamt     2782: static void
1.66      thorpej  2783: pool_allocator_free(struct pool *pp, void *v)
                   2784: {
                   2785:        struct pool_allocator *pa = pp->pr_alloc;
                   2786:
                   2787:        (*pa->pa_free)(pp, v);
                   2788: }
                   2789:
                   2790: void *
1.124     yamt     2791: pool_page_alloc(struct pool *pp, int flags)
1.66      thorpej  2792: {
1.127     thorpej  2793:        bool waitok = (flags & PR_WAITOK) ? true : false;
1.66      thorpej  2794:
1.100     yamt     2795:        return ((void *) uvm_km_alloc_poolpage_cache(kmem_map, waitok));
1.66      thorpej  2796: }
                   2797:
                   2798: void
1.124     yamt     2799: pool_page_free(struct pool *pp, void *v)
1.66      thorpej  2800: {
                   2801:
1.98      yamt     2802:        uvm_km_free_poolpage_cache(kmem_map, (vaddr_t) v);
                   2803: }
                   2804:
                   2805: static void *
1.124     yamt     2806: pool_page_alloc_meta(struct pool *pp, int flags)
1.98      yamt     2807: {
1.127     thorpej  2808:        bool waitok = (flags & PR_WAITOK) ? true : false;
1.98      yamt     2809:
1.100     yamt     2810:        return ((void *) uvm_km_alloc_poolpage(kmem_map, waitok));
1.98      yamt     2811: }
                   2812:
                   2813: static void
1.124     yamt     2814: pool_page_free_meta(struct pool *pp, void *v)
1.98      yamt     2815: {
                   2816:
1.100     yamt     2817:        uvm_km_free_poolpage(kmem_map, (vaddr_t) v);
1.66      thorpej  2818: }
                   2819:
                   2820: #ifdef POOL_SUBPAGE
                   2821: /* Sub-page allocator, for machines with large hardware pages. */
                   2822: void *
                   2823: pool_subpage_alloc(struct pool *pp, int flags)
                   2824: {
1.131.2.2! matt     2825:        return pool_get(&psppool, flags);
1.66      thorpej  2826: }
                   2827:
                   2828: void
                   2829: pool_subpage_free(struct pool *pp, void *v)
                   2830: {
                   2831:        pool_put(&psppool, v);
                   2832: }
                   2833:
                   2834: /* We don't provide a real nointr allocator.  Maybe later. */
                   2835: void *
1.112     bjh21    2836: pool_subpage_alloc_nointr(struct pool *pp, int flags)
1.66      thorpej  2837: {
                   2838:
                   2839:        return (pool_subpage_alloc(pp, flags));
                   2840: }
                   2841:
                   2842: void
1.112     bjh21    2843: pool_subpage_free_nointr(struct pool *pp, void *v)
1.66      thorpej  2844: {
                   2845:
                   2846:        pool_subpage_free(pp, v);
                   2847: }
1.112     bjh21    2848: #endif /* POOL_SUBPAGE */
1.66      thorpej  2849: void *
1.124     yamt     2850: pool_page_alloc_nointr(struct pool *pp, int flags)
1.66      thorpej  2851: {
1.127     thorpej  2852:        bool waitok = (flags & PR_WAITOK) ? true : false;
1.66      thorpej  2853:
1.100     yamt     2854:        return ((void *) uvm_km_alloc_poolpage_cache(kernel_map, waitok));
1.66      thorpej  2855: }
                   2856:
                   2857: void
1.124     yamt     2858: pool_page_free_nointr(struct pool *pp, void *v)
1.66      thorpej  2859: {
                   2860:
1.98      yamt     2861:        uvm_km_free_poolpage_cache(kernel_map, (vaddr_t) v);
1.66      thorpej  2862: }

CVSweb <webmaster@jp.NetBSD.org>