[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 1.116

1.116   ! simonb      1: /*     $NetBSD: subr_pool.c,v 1.115 2006/04/15 04:47:11 christos Exp $ */
1.1       pk          2:
                      3: /*-
1.43      thorpej     4:  * Copyright (c) 1997, 1999, 2000 The NetBSD Foundation, Inc.
1.1       pk          5:  * All rights reserved.
                      6:  *
                      7:  * This code is derived from software contributed to The NetBSD Foundation
1.20      thorpej     8:  * by Paul Kranenburg; by Jason R. Thorpe of the Numerical Aerospace
                      9:  * Simulation Facility, NASA Ames Research Center.
1.1       pk         10:  *
                     11:  * Redistribution and use in source and binary forms, with or without
                     12:  * modification, are permitted provided that the following conditions
                     13:  * are met:
                     14:  * 1. Redistributions of source code must retain the above copyright
                     15:  *    notice, this list of conditions and the following disclaimer.
                     16:  * 2. Redistributions in binary form must reproduce the above copyright
                     17:  *    notice, this list of conditions and the following disclaimer in the
                     18:  *    documentation and/or other materials provided with the distribution.
                     19:  * 3. All advertising materials mentioning features or use of this software
                     20:  *    must display the following acknowledgement:
1.13      christos   21:  *     This product includes software developed by the NetBSD
                     22:  *     Foundation, Inc. and its contributors.
1.1       pk         23:  * 4. Neither the name of The NetBSD Foundation nor the names of its
                     24:  *    contributors may be used to endorse or promote products derived
                     25:  *    from this software without specific prior written permission.
                     26:  *
                     27:  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
                     28:  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
                     29:  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
                     30:  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
                     31:  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
                     32:  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
                     33:  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
                     34:  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
                     35:  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
                     36:  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
                     37:  * POSSIBILITY OF SUCH DAMAGE.
                     38:  */
1.64      lukem      39:
                     40: #include <sys/cdefs.h>
1.116   ! simonb     41: __KERNEL_RCSID(0, "$NetBSD: subr_pool.c,v 1.115 2006/04/15 04:47:11 christos Exp $");
1.24      scottr     42:
1.25      thorpej    43: #include "opt_pool.h"
1.24      scottr     44: #include "opt_poollog.h"
1.28      thorpej    45: #include "opt_lockdebug.h"
1.1       pk         46:
                     47: #include <sys/param.h>
                     48: #include <sys/systm.h>
                     49: #include <sys/proc.h>
                     50: #include <sys/errno.h>
                     51: #include <sys/kernel.h>
                     52: #include <sys/malloc.h>
                     53: #include <sys/lock.h>
                     54: #include <sys/pool.h>
1.20      thorpej    55: #include <sys/syslog.h>
1.3       pk         56:
                     57: #include <uvm/uvm.h>
                     58:
1.1       pk         59: /*
                     60:  * Pool resource management utility.
1.3       pk         61:  *
1.88      chs        62:  * Memory is allocated in pages which are split into pieces according to
                     63:  * the pool item size. Each page is kept on one of three lists in the
                     64:  * pool structure: `pr_emptypages', `pr_fullpages' and `pr_partpages',
                     65:  * for empty, full and partially-full pages respectively. The individual
                     66:  * pool items are on a linked list headed by `ph_itemlist' in each page
                     67:  * header. The memory for building the page list is either taken from
                     68:  * the allocated pages themselves (for small pool items) or taken from
                     69:  * an internal pool of page headers (`phpool').
1.1       pk         70:  */
                     71:
1.3       pk         72: /* List of all pools */
1.102     chs        73: LIST_HEAD(,pool) pool_head = LIST_HEAD_INITIALIZER(pool_head);
1.3       pk         74:
                     75: /* Private pool for page header structures */
1.97      yamt       76: #define        PHPOOL_MAX      8
                     77: static struct pool phpool[PHPOOL_MAX];
                     78: #define        PHPOOL_FREELIST_NELEM(idx)      (((idx) == 0) ? 0 : (1 << (idx)))
1.3       pk         79:
1.62      bjh21      80: #ifdef POOL_SUBPAGE
                     81: /* Pool of subpages for use by normal pools. */
                     82: static struct pool psppool;
                     83: #endif
                     84:
1.98      yamt       85: static void *pool_page_alloc_meta(struct pool *, int);
                     86: static void pool_page_free_meta(struct pool *, void *);
                     87:
                     88: /* allocator for pool metadata */
                     89: static struct pool_allocator pool_allocator_meta = {
                     90:        pool_page_alloc_meta, pool_page_free_meta
                     91: };
                     92:
1.3       pk         93: /* # of seconds to retain page after last use */
                     94: int pool_inactive_time = 10;
                     95:
                     96: /* Next candidate for drainage (see pool_drain()) */
1.23      thorpej    97: static struct pool     *drainpp;
                     98:
                     99: /* This spin lock protects both pool_head and drainpp. */
                    100: struct simplelock pool_head_slock = SIMPLELOCK_INITIALIZER;
1.3       pk        101:
1.99      yamt      102: typedef uint8_t pool_item_freelist_t;
                    103:
1.3       pk        104: struct pool_item_header {
                    105:        /* Page headers */
1.88      chs       106:        LIST_ENTRY(pool_item_header)
1.3       pk        107:                                ph_pagelist;    /* pool page list */
1.88      chs       108:        SPLAY_ENTRY(pool_item_header)
                    109:                                ph_node;        /* Off-page page headers */
1.3       pk        110:        caddr_t                 ph_page;        /* this page's address */
                    111:        struct timeval          ph_time;        /* last referenced */
1.97      yamt      112:        union {
                    113:                /* !PR_NOTOUCH */
                    114:                struct {
1.102     chs       115:                        LIST_HEAD(, pool_item)
1.97      yamt      116:                                phu_itemlist;   /* chunk list for this page */
                    117:                } phu_normal;
                    118:                /* PR_NOTOUCH */
                    119:                struct {
                    120:                        uint16_t
                    121:                                phu_off;        /* start offset in page */
1.99      yamt      122:                        pool_item_freelist_t
1.97      yamt      123:                                phu_firstfree;  /* first free item */
1.99      yamt      124:                        /*
                    125:                         * XXX it might be better to use
                    126:                         * a simple bitmap and ffs(3)
                    127:                         */
1.97      yamt      128:                } phu_notouch;
                    129:        } ph_u;
                    130:        uint16_t                ph_nmissing;    /* # of chunks in use */
1.3       pk        131: };
1.97      yamt      132: #define        ph_itemlist     ph_u.phu_normal.phu_itemlist
                    133: #define        ph_off          ph_u.phu_notouch.phu_off
                    134: #define        ph_firstfree    ph_u.phu_notouch.phu_firstfree
1.3       pk        135:
1.1       pk        136: struct pool_item {
1.3       pk        137: #ifdef DIAGNOSTIC
1.82      thorpej   138:        u_int pi_magic;
1.33      chs       139: #endif
1.82      thorpej   140: #define        PI_MAGIC 0xdeadbeefU
1.3       pk        141:        /* Other entries use only this list entry */
1.102     chs       142:        LIST_ENTRY(pool_item)   pi_list;
1.3       pk        143: };
                    144:
1.53      thorpej   145: #define        POOL_NEEDS_CATCHUP(pp)                                          \
                    146:        ((pp)->pr_nitems < (pp)->pr_minitems)
                    147:
1.43      thorpej   148: /*
                    149:  * Pool cache management.
                    150:  *
                    151:  * Pool caches provide a way for constructed objects to be cached by the
                    152:  * pool subsystem.  This can lead to performance improvements by avoiding
                    153:  * needless object construction/destruction; it is deferred until absolutely
                    154:  * necessary.
                    155:  *
                    156:  * Caches are grouped into cache groups.  Each cache group references
                    157:  * up to 16 constructed objects.  When a cache allocates an object
                    158:  * from the pool, it calls the object's constructor and places it into
                    159:  * a cache group.  When a cache group frees an object back to the pool,
                    160:  * it first calls the object's destructor.  This allows the object to
                    161:  * persist in constructed form while freed to the cache.
                    162:  *
                    163:  * Multiple caches may exist for each pool.  This allows a single
                    164:  * object type to have multiple constructed forms.  The pool references
                    165:  * each cache, so that when a pool is drained by the pagedaemon, it can
                    166:  * drain each individual cache as well.  Each time a cache is drained,
                    167:  * the most idle cache group is freed to the pool in its entirety.
                    168:  *
                    169:  * Pool caches are layed on top of pools.  By layering them, we can avoid
                    170:  * the complexity of cache management for pools which would not benefit
                    171:  * from it.
                    172:  */
                    173:
                    174: /* The cache group pool. */
                    175: static struct pool pcgpool;
1.3       pk        176:
1.102     chs       177: static void    pool_cache_reclaim(struct pool_cache *, struct pool_pagelist *,
                    178:                                   struct pool_cache_grouplist *);
                    179: static void    pcg_grouplist_free(struct pool_cache_grouplist *);
1.3       pk        180:
1.42      thorpej   181: static int     pool_catchup(struct pool *);
1.55      thorpej   182: static void    pool_prime_page(struct pool *, caddr_t,
                    183:                    struct pool_item_header *);
1.88      chs       184: static void    pool_update_curpage(struct pool *);
1.66      thorpej   185:
1.113     yamt      186: static int     pool_grow(struct pool *, int);
1.66      thorpej   187: void           *pool_allocator_alloc(struct pool *, int);
                    188: void           pool_allocator_free(struct pool *, void *);
1.3       pk        189:
1.97      yamt      190: static void pool_print_pagelist(struct pool *, struct pool_pagelist *,
1.88      chs       191:        void (*)(const char *, ...));
1.42      thorpej   192: static void pool_print1(struct pool *, const char *,
                    193:        void (*)(const char *, ...));
1.3       pk        194:
1.88      chs       195: static int pool_chk_page(struct pool *, const char *,
                    196:                         struct pool_item_header *);
                    197:
1.3       pk        198: /*
1.52      thorpej   199:  * Pool log entry. An array of these is allocated in pool_init().
1.3       pk        200:  */
                    201: struct pool_log {
                    202:        const char      *pl_file;
                    203:        long            pl_line;
                    204:        int             pl_action;
1.25      thorpej   205: #define        PRLOG_GET       1
                    206: #define        PRLOG_PUT       2
1.3       pk        207:        void            *pl_addr;
1.1       pk        208: };
                    209:
1.86      matt      210: #ifdef POOL_DIAGNOSTIC
1.3       pk        211: /* Number of entries in pool log buffers */
1.17      thorpej   212: #ifndef POOL_LOGSIZE
                    213: #define        POOL_LOGSIZE    10
                    214: #endif
                    215:
                    216: int pool_logsize = POOL_LOGSIZE;
1.1       pk        217:
1.110     perry     218: static inline void
1.42      thorpej   219: pr_log(struct pool *pp, void *v, int action, const char *file, long line)
1.3       pk        220: {
                    221:        int n = pp->pr_curlogentry;
                    222:        struct pool_log *pl;
                    223:
1.20      thorpej   224:        if ((pp->pr_roflags & PR_LOGGING) == 0)
1.3       pk        225:                return;
                    226:
                    227:        /*
                    228:         * Fill in the current entry. Wrap around and overwrite
                    229:         * the oldest entry if necessary.
                    230:         */
                    231:        pl = &pp->pr_log[n];
                    232:        pl->pl_file = file;
                    233:        pl->pl_line = line;
                    234:        pl->pl_action = action;
                    235:        pl->pl_addr = v;
                    236:        if (++n >= pp->pr_logsize)
                    237:                n = 0;
                    238:        pp->pr_curlogentry = n;
                    239: }
                    240:
                    241: static void
1.42      thorpej   242: pr_printlog(struct pool *pp, struct pool_item *pi,
                    243:     void (*pr)(const char *, ...))
1.3       pk        244: {
                    245:        int i = pp->pr_logsize;
                    246:        int n = pp->pr_curlogentry;
                    247:
1.20      thorpej   248:        if ((pp->pr_roflags & PR_LOGGING) == 0)
1.3       pk        249:                return;
                    250:
                    251:        /*
                    252:         * Print all entries in this pool's log.
                    253:         */
                    254:        while (i-- > 0) {
                    255:                struct pool_log *pl = &pp->pr_log[n];
                    256:                if (pl->pl_action != 0) {
1.25      thorpej   257:                        if (pi == NULL || pi == pl->pl_addr) {
                    258:                                (*pr)("\tlog entry %d:\n", i);
                    259:                                (*pr)("\t\taction = %s, addr = %p\n",
                    260:                                    pl->pl_action == PRLOG_GET ? "get" : "put",
                    261:                                    pl->pl_addr);
                    262:                                (*pr)("\t\tfile: %s at line %lu\n",
                    263:                                    pl->pl_file, pl->pl_line);
                    264:                        }
1.3       pk        265:                }
                    266:                if (++n >= pp->pr_logsize)
                    267:                        n = 0;
                    268:        }
                    269: }
1.25      thorpej   270:
1.110     perry     271: static inline void
1.42      thorpej   272: pr_enter(struct pool *pp, const char *file, long line)
1.25      thorpej   273: {
                    274:
1.34      thorpej   275:        if (__predict_false(pp->pr_entered_file != NULL)) {
1.25      thorpej   276:                printf("pool %s: reentrancy at file %s line %ld\n",
                    277:                    pp->pr_wchan, file, line);
                    278:                printf("         previous entry at file %s line %ld\n",
                    279:                    pp->pr_entered_file, pp->pr_entered_line);
                    280:                panic("pr_enter");
                    281:        }
                    282:
                    283:        pp->pr_entered_file = file;
                    284:        pp->pr_entered_line = line;
                    285: }
                    286:
1.110     perry     287: static inline void
1.42      thorpej   288: pr_leave(struct pool *pp)
1.25      thorpej   289: {
                    290:
1.34      thorpej   291:        if (__predict_false(pp->pr_entered_file == NULL)) {
1.25      thorpej   292:                printf("pool %s not entered?\n", pp->pr_wchan);
                    293:                panic("pr_leave");
                    294:        }
                    295:
                    296:        pp->pr_entered_file = NULL;
                    297:        pp->pr_entered_line = 0;
                    298: }
                    299:
1.110     perry     300: static inline void
1.42      thorpej   301: pr_enter_check(struct pool *pp, void (*pr)(const char *, ...))
1.25      thorpej   302: {
                    303:
                    304:        if (pp->pr_entered_file != NULL)
                    305:                (*pr)("\n\tcurrently entered from file %s line %ld\n",
                    306:                    pp->pr_entered_file, pp->pr_entered_line);
                    307: }
1.3       pk        308: #else
1.25      thorpej   309: #define        pr_log(pp, v, action, file, line)
                    310: #define        pr_printlog(pp, pi, pr)
                    311: #define        pr_enter(pp, file, line)
                    312: #define        pr_leave(pp)
                    313: #define        pr_enter_check(pp, pr)
1.59      thorpej   314: #endif /* POOL_DIAGNOSTIC */
1.3       pk        315:
1.110     perry     316: static inline int
1.97      yamt      317: pr_item_notouch_index(const struct pool *pp, const struct pool_item_header *ph,
                    318:     const void *v)
                    319: {
                    320:        const char *cp = v;
                    321:        int idx;
                    322:
                    323:        KASSERT(pp->pr_roflags & PR_NOTOUCH);
                    324:        idx = (cp - ph->ph_page - ph->ph_off) / pp->pr_size;
                    325:        KASSERT(idx < pp->pr_itemsperpage);
                    326:        return idx;
                    327: }
                    328:
1.99      yamt      329: #define        PR_FREELIST_ALIGN(p) \
                    330:        roundup((uintptr_t)(p), sizeof(pool_item_freelist_t))
                    331: #define        PR_FREELIST(ph) ((pool_item_freelist_t *)PR_FREELIST_ALIGN((ph) + 1))
                    332: #define        PR_INDEX_USED   ((pool_item_freelist_t)-1)
                    333: #define        PR_INDEX_EOL    ((pool_item_freelist_t)-2)
1.97      yamt      334:
1.110     perry     335: static inline void
1.97      yamt      336: pr_item_notouch_put(const struct pool *pp, struct pool_item_header *ph,
                    337:     void *obj)
                    338: {
                    339:        int idx = pr_item_notouch_index(pp, ph, obj);
1.99      yamt      340:        pool_item_freelist_t *freelist = PR_FREELIST(ph);
1.97      yamt      341:
                    342:        KASSERT(freelist[idx] == PR_INDEX_USED);
                    343:        freelist[idx] = ph->ph_firstfree;
                    344:        ph->ph_firstfree = idx;
                    345: }
                    346:
1.110     perry     347: static inline void *
1.97      yamt      348: pr_item_notouch_get(const struct pool *pp, struct pool_item_header *ph)
                    349: {
                    350:        int idx = ph->ph_firstfree;
1.99      yamt      351:        pool_item_freelist_t *freelist = PR_FREELIST(ph);
1.97      yamt      352:
                    353:        KASSERT(freelist[idx] != PR_INDEX_USED);
                    354:        ph->ph_firstfree = freelist[idx];
                    355:        freelist[idx] = PR_INDEX_USED;
                    356:
                    357:        return ph->ph_page + ph->ph_off + idx * pp->pr_size;
                    358: }
                    359:
1.110     perry     360: static inline int
1.88      chs       361: phtree_compare(struct pool_item_header *a, struct pool_item_header *b)
                    362: {
                    363:        if (a->ph_page < b->ph_page)
                    364:                return (-1);
                    365:        else if (a->ph_page > b->ph_page)
                    366:                return (1);
                    367:        else
                    368:                return (0);
                    369: }
                    370:
                    371: SPLAY_PROTOTYPE(phtree, pool_item_header, ph_node, phtree_compare);
                    372: SPLAY_GENERATE(phtree, pool_item_header, ph_node, phtree_compare);
                    373:
1.3       pk        374: /*
                    375:  * Return the pool page header based on page address.
                    376:  */
1.110     perry     377: static inline struct pool_item_header *
1.42      thorpej   378: pr_find_pagehead(struct pool *pp, caddr_t page)
1.3       pk        379: {
1.88      chs       380:        struct pool_item_header *ph, tmp;
1.3       pk        381:
1.20      thorpej   382:        if ((pp->pr_roflags & PR_PHINPAGE) != 0)
1.3       pk        383:                return ((struct pool_item_header *)(page + pp->pr_phoffset));
                    384:
1.88      chs       385:        tmp.ph_page = page;
                    386:        ph = SPLAY_FIND(phtree, &pp->pr_phtree, &tmp);
                    387:        return ph;
1.3       pk        388: }
                    389:
1.101     thorpej   390: static void
                    391: pr_pagelist_free(struct pool *pp, struct pool_pagelist *pq)
                    392: {
                    393:        struct pool_item_header *ph;
                    394:        int s;
                    395:
                    396:        while ((ph = LIST_FIRST(pq)) != NULL) {
                    397:                LIST_REMOVE(ph, ph_pagelist);
                    398:                pool_allocator_free(pp, ph->ph_page);
                    399:                if ((pp->pr_roflags & PR_PHINPAGE) == 0) {
                    400:                        s = splvm();
                    401:                        pool_put(pp->pr_phpool, ph);
                    402:                        splx(s);
                    403:                }
                    404:        }
                    405: }
                    406:
1.3       pk        407: /*
                    408:  * Remove a page from the pool.
                    409:  */
1.110     perry     410: static inline void
1.61      chs       411: pr_rmpage(struct pool *pp, struct pool_item_header *ph,
                    412:      struct pool_pagelist *pq)
1.3       pk        413: {
                    414:
1.101     thorpej   415:        LOCK_ASSERT(simple_lock_held(&pp->pr_slock));
1.91      yamt      416:
1.3       pk        417:        /*
1.7       thorpej   418:         * If the page was idle, decrement the idle page count.
1.3       pk        419:         */
1.6       thorpej   420:        if (ph->ph_nmissing == 0) {
                    421: #ifdef DIAGNOSTIC
                    422:                if (pp->pr_nidle == 0)
                    423:                        panic("pr_rmpage: nidle inconsistent");
1.20      thorpej   424:                if (pp->pr_nitems < pp->pr_itemsperpage)
                    425:                        panic("pr_rmpage: nitems inconsistent");
1.6       thorpej   426: #endif
                    427:                pp->pr_nidle--;
                    428:        }
1.7       thorpej   429:
1.20      thorpej   430:        pp->pr_nitems -= pp->pr_itemsperpage;
                    431:
1.7       thorpej   432:        /*
1.101     thorpej   433:         * Unlink the page from the pool and queue it for release.
1.7       thorpej   434:         */
1.88      chs       435:        LIST_REMOVE(ph, ph_pagelist);
1.91      yamt      436:        if ((pp->pr_roflags & PR_PHINPAGE) == 0)
                    437:                SPLAY_REMOVE(phtree, &pp->pr_phtree, ph);
1.101     thorpej   438:        LIST_INSERT_HEAD(pq, ph, ph_pagelist);
                    439:
1.7       thorpej   440:        pp->pr_npages--;
                    441:        pp->pr_npagefree++;
1.6       thorpej   442:
1.88      chs       443:        pool_update_curpage(pp);
1.3       pk        444: }
                    445:
                    446: /*
1.94      simonb    447:  * Initialize all the pools listed in the "pools" link set.
                    448:  */
                    449: void
                    450: link_pool_init(void)
                    451: {
                    452:        __link_set_decl(pools, struct link_pool_init);
                    453:        struct link_pool_init * const *pi;
                    454:
                    455:        __link_set_foreach(pi, pools)
                    456:                pool_init((*pi)->pp, (*pi)->size, (*pi)->align,
                    457:                    (*pi)->align_offset, (*pi)->flags, (*pi)->wchan,
                    458:                    (*pi)->palloc);
                    459: }
                    460:
                    461: /*
1.3       pk        462:  * Initialize the given pool resource structure.
                    463:  *
                    464:  * We export this routine to allow other kernel parts to declare
                    465:  * static pools that must be initialized before malloc() is available.
                    466:  */
                    467: void
1.42      thorpej   468: pool_init(struct pool *pp, size_t size, u_int align, u_int ioff, int flags,
1.66      thorpej   469:     const char *wchan, struct pool_allocator *palloc)
1.3       pk        470: {
1.116   ! simonb    471: #ifdef DEBUG
        !           472:        struct pool *pp1;
        !           473: #endif
1.92      enami     474:        size_t trysize, phsize;
1.116   ! simonb    475:        int off, slack, s;
1.3       pk        476:
1.99      yamt      477:        KASSERT((1UL << (CHAR_BIT * sizeof(pool_item_freelist_t))) - 2 >=
                    478:            PHPOOL_FREELIST_NELEM(PHPOOL_MAX - 1));
                    479:
1.116   ! simonb    480: #ifdef DEBUG
        !           481:        /*
        !           482:         * Check that the pool hasn't already been initialised and
        !           483:         * added to the list of all pools.
        !           484:         */
        !           485:        LIST_FOREACH(pp1, &pool_head, pr_poollist) {
        !           486:                if (pp == pp1)
        !           487:                        panic("pool_init: pool %s already initialised",
        !           488:                            wchan);
        !           489:        }
        !           490: #endif
        !           491:
1.25      thorpej   492: #ifdef POOL_DIAGNOSTIC
                    493:        /*
                    494:         * Always log if POOL_DIAGNOSTIC is defined.
                    495:         */
                    496:        if (pool_logsize != 0)
                    497:                flags |= PR_LOGGING;
                    498: #endif
                    499:
1.66      thorpej   500:        if (palloc == NULL)
                    501:                palloc = &pool_allocator_kmem;
1.112     bjh21     502: #ifdef POOL_SUBPAGE
                    503:        if (size > palloc->pa_pagesz) {
                    504:                if (palloc == &pool_allocator_kmem)
                    505:                        palloc = &pool_allocator_kmem_fullpage;
                    506:                else if (palloc == &pool_allocator_nointr)
                    507:                        palloc = &pool_allocator_nointr_fullpage;
                    508:        }
1.66      thorpej   509: #endif /* POOL_SUBPAGE */
                    510:        if ((palloc->pa_flags & PA_INITIALIZED) == 0) {
1.112     bjh21     511:                if (palloc->pa_pagesz == 0)
1.66      thorpej   512:                        palloc->pa_pagesz = PAGE_SIZE;
                    513:
                    514:                TAILQ_INIT(&palloc->pa_list);
                    515:
                    516:                simple_lock_init(&palloc->pa_slock);
                    517:                palloc->pa_pagemask = ~(palloc->pa_pagesz - 1);
                    518:                palloc->pa_pageshift = ffs(palloc->pa_pagesz) - 1;
                    519:                palloc->pa_flags |= PA_INITIALIZED;
1.4       thorpej   520:        }
1.3       pk        521:
                    522:        if (align == 0)
                    523:                align = ALIGN(1);
1.14      thorpej   524:
                    525:        if (size < sizeof(struct pool_item))
                    526:                size = sizeof(struct pool_item);
1.3       pk        527:
1.78      thorpej   528:        size = roundup(size, align);
1.66      thorpej   529: #ifdef DIAGNOSTIC
                    530:        if (size > palloc->pa_pagesz)
1.35      pk        531:                panic("pool_init: pool item size (%lu) too large",
                    532:                      (u_long)size);
1.66      thorpej   533: #endif
1.35      pk        534:
1.3       pk        535:        /*
                    536:         * Initialize the pool structure.
                    537:         */
1.88      chs       538:        LIST_INIT(&pp->pr_emptypages);
                    539:        LIST_INIT(&pp->pr_fullpages);
                    540:        LIST_INIT(&pp->pr_partpages);
1.102     chs       541:        LIST_INIT(&pp->pr_cachelist);
1.3       pk        542:        pp->pr_curpage = NULL;
                    543:        pp->pr_npages = 0;
                    544:        pp->pr_minitems = 0;
                    545:        pp->pr_minpages = 0;
                    546:        pp->pr_maxpages = UINT_MAX;
1.20      thorpej   547:        pp->pr_roflags = flags;
                    548:        pp->pr_flags = 0;
1.35      pk        549:        pp->pr_size = size;
1.3       pk        550:        pp->pr_align = align;
                    551:        pp->pr_wchan = wchan;
1.66      thorpej   552:        pp->pr_alloc = palloc;
1.20      thorpej   553:        pp->pr_nitems = 0;
                    554:        pp->pr_nout = 0;
                    555:        pp->pr_hardlimit = UINT_MAX;
                    556:        pp->pr_hardlimit_warning = NULL;
1.31      thorpej   557:        pp->pr_hardlimit_ratecap.tv_sec = 0;
                    558:        pp->pr_hardlimit_ratecap.tv_usec = 0;
                    559:        pp->pr_hardlimit_warning_last.tv_sec = 0;
                    560:        pp->pr_hardlimit_warning_last.tv_usec = 0;
1.68      thorpej   561:        pp->pr_drain_hook = NULL;
                    562:        pp->pr_drain_hook_arg = NULL;
1.3       pk        563:
                    564:        /*
                    565:         * Decide whether to put the page header off page to avoid
1.92      enami     566:         * wasting too large a part of the page or too big item.
                    567:         * Off-page page headers go on a hash table, so we can match
                    568:         * a returned item with its header based on the page address.
                    569:         * We use 1/16 of the page size and about 8 times of the item
                    570:         * size as the threshold (XXX: tune)
                    571:         *
                    572:         * However, we'll put the header into the page if we can put
                    573:         * it without wasting any items.
                    574:         *
                    575:         * Silently enforce `0 <= ioff < align'.
1.3       pk        576:         */
1.92      enami     577:        pp->pr_itemoffset = ioff %= align;
                    578:        /* See the comment below about reserved bytes. */
                    579:        trysize = palloc->pa_pagesz - ((align - ioff) % align);
                    580:        phsize = ALIGN(sizeof(struct pool_item_header));
1.97      yamt      581:        if ((pp->pr_roflags & PR_NOTOUCH) == 0 &&
                    582:            (pp->pr_size < MIN(palloc->pa_pagesz / 16, phsize << 3) ||
                    583:            trysize / pp->pr_size == (trysize - phsize) / pp->pr_size)) {
1.3       pk        584:                /* Use the end of the page for the page header */
1.20      thorpej   585:                pp->pr_roflags |= PR_PHINPAGE;
1.92      enami     586:                pp->pr_phoffset = off = palloc->pa_pagesz - phsize;
1.2       pk        587:        } else {
1.3       pk        588:                /* The page header will be taken from our page header pool */
                    589:                pp->pr_phoffset = 0;
1.66      thorpej   590:                off = palloc->pa_pagesz;
1.88      chs       591:                SPLAY_INIT(&pp->pr_phtree);
1.2       pk        592:        }
1.1       pk        593:
1.3       pk        594:        /*
                    595:         * Alignment is to take place at `ioff' within the item. This means
                    596:         * we must reserve up to `align - 1' bytes on the page to allow
                    597:         * appropriate positioning of each item.
                    598:         */
                    599:        pp->pr_itemsperpage = (off - ((align - ioff) % align)) / pp->pr_size;
1.43      thorpej   600:        KASSERT(pp->pr_itemsperpage != 0);
1.97      yamt      601:        if ((pp->pr_roflags & PR_NOTOUCH)) {
                    602:                int idx;
                    603:
                    604:                for (idx = 0; pp->pr_itemsperpage > PHPOOL_FREELIST_NELEM(idx);
                    605:                    idx++) {
                    606:                        /* nothing */
                    607:                }
                    608:                if (idx >= PHPOOL_MAX) {
                    609:                        /*
                    610:                         * if you see this panic, consider to tweak
                    611:                         * PHPOOL_MAX and PHPOOL_FREELIST_NELEM.
                    612:                         */
                    613:                        panic("%s: too large itemsperpage(%d) for PR_NOTOUCH",
                    614:                            pp->pr_wchan, pp->pr_itemsperpage);
                    615:                }
                    616:                pp->pr_phpool = &phpool[idx];
                    617:        } else if ((pp->pr_roflags & PR_PHINPAGE) == 0) {
                    618:                pp->pr_phpool = &phpool[0];
                    619:        }
                    620: #if defined(DIAGNOSTIC)
                    621:        else {
                    622:                pp->pr_phpool = NULL;
                    623:        }
                    624: #endif
1.3       pk        625:
                    626:        /*
                    627:         * Use the slack between the chunks and the page header
                    628:         * for "cache coloring".
                    629:         */
                    630:        slack = off - pp->pr_itemsperpage * pp->pr_size;
                    631:        pp->pr_maxcolor = (slack / align) * align;
                    632:        pp->pr_curcolor = 0;
                    633:
                    634:        pp->pr_nget = 0;
                    635:        pp->pr_nfail = 0;
                    636:        pp->pr_nput = 0;
                    637:        pp->pr_npagealloc = 0;
                    638:        pp->pr_npagefree = 0;
1.1       pk        639:        pp->pr_hiwat = 0;
1.8       thorpej   640:        pp->pr_nidle = 0;
1.3       pk        641:
1.59      thorpej   642: #ifdef POOL_DIAGNOSTIC
1.25      thorpej   643:        if (flags & PR_LOGGING) {
                    644:                if (kmem_map == NULL ||
                    645:                    (pp->pr_log = malloc(pool_logsize * sizeof(struct pool_log),
                    646:                     M_TEMP, M_NOWAIT)) == NULL)
1.20      thorpej   647:                        pp->pr_roflags &= ~PR_LOGGING;
1.3       pk        648:                pp->pr_curlogentry = 0;
                    649:                pp->pr_logsize = pool_logsize;
                    650:        }
1.59      thorpej   651: #endif
1.25      thorpej   652:
                    653:        pp->pr_entered_file = NULL;
                    654:        pp->pr_entered_line = 0;
1.3       pk        655:
1.21      thorpej   656:        simple_lock_init(&pp->pr_slock);
1.1       pk        657:
1.3       pk        658:        /*
1.43      thorpej   659:         * Initialize private page header pool and cache magazine pool if we
                    660:         * haven't done so yet.
1.23      thorpej   661:         * XXX LOCKING.
1.3       pk        662:         */
1.97      yamt      663:        if (phpool[0].pr_size == 0) {
                    664:                int idx;
                    665:                for (idx = 0; idx < PHPOOL_MAX; idx++) {
                    666:                        static char phpool_names[PHPOOL_MAX][6+1+6+1];
                    667:                        int nelem;
                    668:                        size_t sz;
                    669:
                    670:                        nelem = PHPOOL_FREELIST_NELEM(idx);
                    671:                        snprintf(phpool_names[idx], sizeof(phpool_names[idx]),
                    672:                            "phpool-%d", nelem);
                    673:                        sz = sizeof(struct pool_item_header);
                    674:                        if (nelem) {
                    675:                                sz = PR_FREELIST_ALIGN(sz)
1.99      yamt      676:                                    + nelem * sizeof(pool_item_freelist_t);
1.97      yamt      677:                        }
                    678:                        pool_init(&phpool[idx], sz, 0, 0, 0,
1.98      yamt      679:                            phpool_names[idx], &pool_allocator_meta);
1.97      yamt      680:                }
1.62      bjh21     681: #ifdef POOL_SUBPAGE
                    682:                pool_init(&psppool, POOL_SUBPAGE, POOL_SUBPAGE, 0,
1.98      yamt      683:                    PR_RECURSIVE, "psppool", &pool_allocator_meta);
1.62      bjh21     684: #endif
1.43      thorpej   685:                pool_init(&pcgpool, sizeof(struct pool_cache_group), 0, 0,
1.98      yamt      686:                    0, "pcgpool", &pool_allocator_meta);
1.1       pk        687:        }
                    688:
1.23      thorpej   689:        /* Insert into the list of all pools. */
                    690:        simple_lock(&pool_head_slock);
1.102     chs       691:        LIST_INSERT_HEAD(&pool_head, pp, pr_poollist);
1.23      thorpej   692:        simple_unlock(&pool_head_slock);
1.66      thorpej   693:
                    694:        /* Insert this into the list of pools using this allocator. */
1.93      dbj       695:        s = splvm();
1.66      thorpej   696:        simple_lock(&palloc->pa_slock);
                    697:        TAILQ_INSERT_TAIL(&palloc->pa_list, pp, pr_alloc_list);
                    698:        simple_unlock(&palloc->pa_slock);
1.93      dbj       699:        splx(s);
1.1       pk        700: }
                    701:
                    702: /*
                    703:  * De-commision a pool resource.
                    704:  */
                    705: void
1.42      thorpej   706: pool_destroy(struct pool *pp)
1.1       pk        707: {
1.101     thorpej   708:        struct pool_pagelist pq;
1.3       pk        709:        struct pool_item_header *ph;
1.93      dbj       710:        int s;
1.43      thorpej   711:
1.101     thorpej   712:        /* Remove from global pool list */
                    713:        simple_lock(&pool_head_slock);
1.102     chs       714:        LIST_REMOVE(pp, pr_poollist);
1.101     thorpej   715:        if (drainpp == pp)
                    716:                drainpp = NULL;
                    717:        simple_unlock(&pool_head_slock);
                    718:
                    719:        /* Remove this pool from its allocator's list of pools. */
1.93      dbj       720:        s = splvm();
1.66      thorpej   721:        simple_lock(&pp->pr_alloc->pa_slock);
                    722:        TAILQ_REMOVE(&pp->pr_alloc->pa_list, pp, pr_alloc_list);
                    723:        simple_unlock(&pp->pr_alloc->pa_slock);
1.93      dbj       724:        splx(s);
1.66      thorpej   725:
1.101     thorpej   726:        s = splvm();
                    727:        simple_lock(&pp->pr_slock);
                    728:
1.102     chs       729:        KASSERT(LIST_EMPTY(&pp->pr_cachelist));
1.3       pk        730:
                    731: #ifdef DIAGNOSTIC
1.20      thorpej   732:        if (pp->pr_nout != 0) {
1.25      thorpej   733:                pr_printlog(pp, NULL, printf);
1.80      provos    734:                panic("pool_destroy: pool busy: still out: %u",
1.20      thorpej   735:                    pp->pr_nout);
1.3       pk        736:        }
                    737: #endif
1.1       pk        738:
1.101     thorpej   739:        KASSERT(LIST_EMPTY(&pp->pr_fullpages));
                    740:        KASSERT(LIST_EMPTY(&pp->pr_partpages));
                    741:
1.3       pk        742:        /* Remove all pages */
1.101     thorpej   743:        LIST_INIT(&pq);
1.88      chs       744:        while ((ph = LIST_FIRST(&pp->pr_emptypages)) != NULL)
1.101     thorpej   745:                pr_rmpage(pp, ph, &pq);
                    746:
                    747:        simple_unlock(&pp->pr_slock);
                    748:        splx(s);
1.3       pk        749:
1.101     thorpej   750:        pr_pagelist_free(pp, &pq);
1.3       pk        751:
1.59      thorpej   752: #ifdef POOL_DIAGNOSTIC
1.20      thorpej   753:        if ((pp->pr_roflags & PR_LOGGING) != 0)
1.3       pk        754:                free(pp->pr_log, M_TEMP);
1.59      thorpej   755: #endif
1.1       pk        756: }
                    757:
1.68      thorpej   758: void
                    759: pool_set_drain_hook(struct pool *pp, void (*fn)(void *, int), void *arg)
                    760: {
                    761:
                    762:        /* XXX no locking -- must be used just after pool_init() */
                    763: #ifdef DIAGNOSTIC
                    764:        if (pp->pr_drain_hook != NULL)
                    765:                panic("pool_set_drain_hook(%s): already set", pp->pr_wchan);
                    766: #endif
                    767:        pp->pr_drain_hook = fn;
                    768:        pp->pr_drain_hook_arg = arg;
                    769: }
                    770:
1.88      chs       771: static struct pool_item_header *
1.55      thorpej   772: pool_alloc_item_header(struct pool *pp, caddr_t storage, int flags)
                    773: {
                    774:        struct pool_item_header *ph;
                    775:        int s;
                    776:
                    777:        LOCK_ASSERT(simple_lock_held(&pp->pr_slock) == 0);
                    778:
                    779:        if ((pp->pr_roflags & PR_PHINPAGE) != 0)
                    780:                ph = (struct pool_item_header *) (storage + pp->pr_phoffset);
                    781:        else {
1.85      pk        782:                s = splvm();
1.97      yamt      783:                ph = pool_get(pp->pr_phpool, flags);
1.55      thorpej   784:                splx(s);
                    785:        }
                    786:
                    787:        return (ph);
                    788: }
1.1       pk        789:
                    790: /*
1.3       pk        791:  * Grab an item from the pool; must be called at appropriate spl level
1.1       pk        792:  */
1.3       pk        793: void *
1.59      thorpej   794: #ifdef POOL_DIAGNOSTIC
1.42      thorpej   795: _pool_get(struct pool *pp, int flags, const char *file, long line)
1.56      sommerfe  796: #else
                    797: pool_get(struct pool *pp, int flags)
                    798: #endif
1.1       pk        799: {
                    800:        struct pool_item *pi;
1.3       pk        801:        struct pool_item_header *ph;
1.55      thorpej   802:        void *v;
1.1       pk        803:
1.2       pk        804: #ifdef DIAGNOSTIC
1.95      atatat    805:        if (__predict_false(pp->pr_itemsperpage == 0))
                    806:                panic("pool_get: pool %p: pr_itemsperpage is zero, "
                    807:                    "pool not initialized?", pp);
1.84      thorpej   808:        if (__predict_false(curlwp == NULL && doing_shutdown == 0 &&
1.37      sommerfe  809:                            (flags & PR_WAITOK) != 0))
1.77      matt      810:                panic("pool_get: %s: must have NOWAIT", pp->pr_wchan);
1.58      thorpej   811:
1.102     chs       812: #endif /* DIAGNOSTIC */
1.58      thorpej   813: #ifdef LOCKDEBUG
                    814:        if (flags & PR_WAITOK)
                    815:                simple_lock_only_held(NULL, "pool_get(PR_WAITOK)");
1.102     chs       816:        SCHED_ASSERT_UNLOCKED();
1.56      sommerfe  817: #endif
1.1       pk        818:
1.21      thorpej   819:        simple_lock(&pp->pr_slock);
1.25      thorpej   820:        pr_enter(pp, file, line);
1.20      thorpej   821:
                    822:  startover:
                    823:        /*
                    824:         * Check to see if we've reached the hard limit.  If we have,
                    825:         * and we can wait, then wait until an item has been returned to
                    826:         * the pool.
                    827:         */
                    828: #ifdef DIAGNOSTIC
1.34      thorpej   829:        if (__predict_false(pp->pr_nout > pp->pr_hardlimit)) {
1.25      thorpej   830:                pr_leave(pp);
1.21      thorpej   831:                simple_unlock(&pp->pr_slock);
1.20      thorpej   832:                panic("pool_get: %s: crossed hard limit", pp->pr_wchan);
                    833:        }
                    834: #endif
1.34      thorpej   835:        if (__predict_false(pp->pr_nout == pp->pr_hardlimit)) {
1.68      thorpej   836:                if (pp->pr_drain_hook != NULL) {
                    837:                        /*
                    838:                         * Since the drain hook is going to free things
                    839:                         * back to the pool, unlock, call the hook, re-lock,
                    840:                         * and check the hardlimit condition again.
                    841:                         */
                    842:                        pr_leave(pp);
                    843:                        simple_unlock(&pp->pr_slock);
                    844:                        (*pp->pr_drain_hook)(pp->pr_drain_hook_arg, flags);
                    845:                        simple_lock(&pp->pr_slock);
                    846:                        pr_enter(pp, file, line);
                    847:                        if (pp->pr_nout < pp->pr_hardlimit)
                    848:                                goto startover;
                    849:                }
                    850:
1.29      sommerfe  851:                if ((flags & PR_WAITOK) && !(flags & PR_LIMITFAIL)) {
1.20      thorpej   852:                        /*
                    853:                         * XXX: A warning isn't logged in this case.  Should
                    854:                         * it be?
                    855:                         */
                    856:                        pp->pr_flags |= PR_WANTED;
1.25      thorpej   857:                        pr_leave(pp);
1.40      sommerfe  858:                        ltsleep(pp, PSWP, pp->pr_wchan, 0, &pp->pr_slock);
1.25      thorpej   859:                        pr_enter(pp, file, line);
1.20      thorpej   860:                        goto startover;
                    861:                }
1.31      thorpej   862:
                    863:                /*
                    864:                 * Log a message that the hard limit has been hit.
                    865:                 */
                    866:                if (pp->pr_hardlimit_warning != NULL &&
                    867:                    ratecheck(&pp->pr_hardlimit_warning_last,
                    868:                              &pp->pr_hardlimit_ratecap))
                    869:                        log(LOG_ERR, "%s\n", pp->pr_hardlimit_warning);
1.21      thorpej   870:
                    871:                pp->pr_nfail++;
                    872:
1.25      thorpej   873:                pr_leave(pp);
1.21      thorpej   874:                simple_unlock(&pp->pr_slock);
1.20      thorpej   875:                return (NULL);
                    876:        }
                    877:
1.3       pk        878:        /*
                    879:         * The convention we use is that if `curpage' is not NULL, then
                    880:         * it points at a non-empty bucket. In particular, `curpage'
                    881:         * never points at a page header which has PR_PHINPAGE set and
                    882:         * has no items in its bucket.
                    883:         */
1.20      thorpej   884:        if ((ph = pp->pr_curpage) == NULL) {
1.113     yamt      885:                int error;
                    886:
1.20      thorpej   887: #ifdef DIAGNOSTIC
                    888:                if (pp->pr_nitems != 0) {
1.21      thorpej   889:                        simple_unlock(&pp->pr_slock);
1.20      thorpej   890:                        printf("pool_get: %s: curpage NULL, nitems %u\n",
                    891:                            pp->pr_wchan, pp->pr_nitems);
1.80      provos    892:                        panic("pool_get: nitems inconsistent");
1.20      thorpej   893:                }
                    894: #endif
                    895:
1.21      thorpej   896:                /*
                    897:                 * Call the back-end page allocator for more memory.
                    898:                 * Release the pool lock, as the back-end page allocator
                    899:                 * may block.
                    900:                 */
1.25      thorpej   901:                pr_leave(pp);
1.113     yamt      902:                error = pool_grow(pp, flags);
                    903:                pr_enter(pp, file, line);
                    904:                if (error != 0) {
1.21      thorpej   905:                        /*
1.55      thorpej   906:                         * We were unable to allocate a page or item
                    907:                         * header, but we released the lock during
                    908:                         * allocation, so perhaps items were freed
                    909:                         * back to the pool.  Check for this case.
1.21      thorpej   910:                         */
                    911:                        if (pp->pr_curpage != NULL)
                    912:                                goto startover;
1.15      pk        913:
1.3       pk        914:                        if ((flags & PR_WAITOK) == 0) {
                    915:                                pp->pr_nfail++;
1.25      thorpej   916:                                pr_leave(pp);
1.21      thorpej   917:                                simple_unlock(&pp->pr_slock);
1.1       pk        918:                                return (NULL);
1.3       pk        919:                        }
                    920:
1.15      pk        921:                        /*
                    922:                         * Wait for items to be returned to this pool.
1.21      thorpej   923:                         *
1.109     christos  924:                         * wake up once a second and try again,
                    925:                         * as the check in pool_cache_put_paddr() is racy.
1.15      pk        926:                         */
1.1       pk        927:                        pp->pr_flags |= PR_WANTED;
1.66      thorpej   928:                        /* PA_WANTED is already set on the allocator. */
1.25      thorpej   929:                        pr_leave(pp);
1.109     christos  930:                        ltsleep(pp, PSWP, pp->pr_wchan, hz, &pp->pr_slock);
1.25      thorpej   931:                        pr_enter(pp, file, line);
1.1       pk        932:                }
1.3       pk        933:
1.20      thorpej   934:                /* Start the allocation process over. */
                    935:                goto startover;
1.3       pk        936:        }
1.97      yamt      937:        if (pp->pr_roflags & PR_NOTOUCH) {
                    938: #ifdef DIAGNOSTIC
                    939:                if (__predict_false(ph->ph_nmissing == pp->pr_itemsperpage)) {
                    940:                        pr_leave(pp);
                    941:                        simple_unlock(&pp->pr_slock);
                    942:                        panic("pool_get: %s: page empty", pp->pr_wchan);
                    943:                }
                    944: #endif
                    945:                v = pr_item_notouch_get(pp, ph);
                    946: #ifdef POOL_DIAGNOSTIC
                    947:                pr_log(pp, v, PRLOG_GET, file, line);
                    948: #endif
                    949:        } else {
1.102     chs       950:                v = pi = LIST_FIRST(&ph->ph_itemlist);
1.97      yamt      951:                if (__predict_false(v == NULL)) {
                    952:                        pr_leave(pp);
                    953:                        simple_unlock(&pp->pr_slock);
                    954:                        panic("pool_get: %s: page empty", pp->pr_wchan);
                    955:                }
1.20      thorpej   956: #ifdef DIAGNOSTIC
1.97      yamt      957:                if (__predict_false(pp->pr_nitems == 0)) {
                    958:                        pr_leave(pp);
                    959:                        simple_unlock(&pp->pr_slock);
                    960:                        printf("pool_get: %s: items on itemlist, nitems %u\n",
                    961:                            pp->pr_wchan, pp->pr_nitems);
                    962:                        panic("pool_get: nitems inconsistent");
                    963:                }
1.65      enami     964: #endif
1.56      sommerfe  965:
1.65      enami     966: #ifdef POOL_DIAGNOSTIC
1.97      yamt      967:                pr_log(pp, v, PRLOG_GET, file, line);
1.65      enami     968: #endif
1.3       pk        969:
1.65      enami     970: #ifdef DIAGNOSTIC
1.97      yamt      971:                if (__predict_false(pi->pi_magic != PI_MAGIC)) {
                    972:                        pr_printlog(pp, pi, printf);
                    973:                        panic("pool_get(%s): free list modified: "
                    974:                            "magic=%x; page %p; item addr %p\n",
                    975:                            pp->pr_wchan, pi->pi_magic, ph->ph_page, pi);
                    976:                }
1.3       pk        977: #endif
                    978:
1.97      yamt      979:                /*
                    980:                 * Remove from item list.
                    981:                 */
1.102     chs       982:                LIST_REMOVE(pi, pi_list);
1.97      yamt      983:        }
1.20      thorpej   984:        pp->pr_nitems--;
                    985:        pp->pr_nout++;
1.6       thorpej   986:        if (ph->ph_nmissing == 0) {
                    987: #ifdef DIAGNOSTIC
1.34      thorpej   988:                if (__predict_false(pp->pr_nidle == 0))
1.6       thorpej   989:                        panic("pool_get: nidle inconsistent");
                    990: #endif
                    991:                pp->pr_nidle--;
1.88      chs       992:
                    993:                /*
                    994:                 * This page was previously empty.  Move it to the list of
                    995:                 * partially-full pages.  This page is already curpage.
                    996:                 */
                    997:                LIST_REMOVE(ph, ph_pagelist);
                    998:                LIST_INSERT_HEAD(&pp->pr_partpages, ph, ph_pagelist);
1.6       thorpej   999:        }
1.3       pk       1000:        ph->ph_nmissing++;
1.97      yamt     1001:        if (ph->ph_nmissing == pp->pr_itemsperpage) {
1.21      thorpej  1002: #ifdef DIAGNOSTIC
1.97      yamt     1003:                if (__predict_false((pp->pr_roflags & PR_NOTOUCH) == 0 &&
1.102     chs      1004:                    !LIST_EMPTY(&ph->ph_itemlist))) {
1.25      thorpej  1005:                        pr_leave(pp);
1.21      thorpej  1006:                        simple_unlock(&pp->pr_slock);
                   1007:                        panic("pool_get: %s: nmissing inconsistent",
                   1008:                            pp->pr_wchan);
                   1009:                }
                   1010: #endif
1.3       pk       1011:                /*
1.88      chs      1012:                 * This page is now full.  Move it to the full list
                   1013:                 * and select a new current page.
1.3       pk       1014:                 */
1.88      chs      1015:                LIST_REMOVE(ph, ph_pagelist);
                   1016:                LIST_INSERT_HEAD(&pp->pr_fullpages, ph, ph_pagelist);
                   1017:                pool_update_curpage(pp);
1.1       pk       1018:        }
1.3       pk       1019:
                   1020:        pp->pr_nget++;
1.111     christos 1021:        pr_leave(pp);
1.20      thorpej  1022:
                   1023:        /*
                   1024:         * If we have a low water mark and we are now below that low
                   1025:         * water mark, add more items to the pool.
                   1026:         */
1.53      thorpej  1027:        if (POOL_NEEDS_CATCHUP(pp) && pool_catchup(pp) != 0) {
1.20      thorpej  1028:                /*
                   1029:                 * XXX: Should we log a warning?  Should we set up a timeout
                   1030:                 * to try again in a second or so?  The latter could break
                   1031:                 * a caller's assumptions about interrupt protection, etc.
                   1032:                 */
                   1033:        }
                   1034:
1.21      thorpej  1035:        simple_unlock(&pp->pr_slock);
1.1       pk       1036:        return (v);
                   1037: }
                   1038:
                   1039: /*
1.43      thorpej  1040:  * Internal version of pool_put().  Pool is already locked/entered.
1.1       pk       1041:  */
1.43      thorpej  1042: static void
1.101     thorpej  1043: pool_do_put(struct pool *pp, void *v, struct pool_pagelist *pq)
1.1       pk       1044: {
                   1045:        struct pool_item *pi = v;
1.3       pk       1046:        struct pool_item_header *ph;
                   1047:        caddr_t page;
1.21      thorpej  1048:        int s;
1.3       pk       1049:
1.61      chs      1050:        LOCK_ASSERT(simple_lock_held(&pp->pr_slock));
1.102     chs      1051:        SCHED_ASSERT_UNLOCKED();
1.61      chs      1052:
1.66      thorpej  1053:        page = (caddr_t)((u_long)v & pp->pr_alloc->pa_pagemask);
1.1       pk       1054:
1.30      thorpej  1055: #ifdef DIAGNOSTIC
1.34      thorpej  1056:        if (__predict_false(pp->pr_nout == 0)) {
1.30      thorpej  1057:                printf("pool %s: putting with none out\n",
                   1058:                    pp->pr_wchan);
                   1059:                panic("pool_put");
                   1060:        }
                   1061: #endif
1.3       pk       1062:
1.34      thorpej  1063:        if (__predict_false((ph = pr_find_pagehead(pp, page)) == NULL)) {
1.25      thorpej  1064:                pr_printlog(pp, NULL, printf);
1.3       pk       1065:                panic("pool_put: %s: page header missing", pp->pr_wchan);
                   1066:        }
1.28      thorpej  1067:
                   1068: #ifdef LOCKDEBUG
                   1069:        /*
                   1070:         * Check if we're freeing a locked simple lock.
                   1071:         */
                   1072:        simple_lock_freecheck((caddr_t)pi, ((caddr_t)pi) + pp->pr_size);
                   1073: #endif
1.3       pk       1074:
                   1075:        /*
                   1076:         * Return to item list.
                   1077:         */
1.97      yamt     1078:        if (pp->pr_roflags & PR_NOTOUCH) {
                   1079:                pr_item_notouch_put(pp, ph, v);
                   1080:        } else {
1.2       pk       1081: #ifdef DIAGNOSTIC
1.97      yamt     1082:                pi->pi_magic = PI_MAGIC;
1.3       pk       1083: #endif
1.32      chs      1084: #ifdef DEBUG
1.97      yamt     1085:                {
                   1086:                        int i, *ip = v;
1.32      chs      1087:
1.97      yamt     1088:                        for (i = 0; i < pp->pr_size / sizeof(int); i++) {
                   1089:                                *ip++ = PI_MAGIC;
                   1090:                        }
1.32      chs      1091:                }
                   1092: #endif
                   1093:
1.102     chs      1094:                LIST_INSERT_HEAD(&ph->ph_itemlist, pi, pi_list);
1.97      yamt     1095:        }
1.79      thorpej  1096:        KDASSERT(ph->ph_nmissing != 0);
1.3       pk       1097:        ph->ph_nmissing--;
                   1098:        pp->pr_nput++;
1.20      thorpej  1099:        pp->pr_nitems++;
                   1100:        pp->pr_nout--;
1.3       pk       1101:
                   1102:        /* Cancel "pool empty" condition if it exists */
                   1103:        if (pp->pr_curpage == NULL)
                   1104:                pp->pr_curpage = ph;
                   1105:
                   1106:        if (pp->pr_flags & PR_WANTED) {
                   1107:                pp->pr_flags &= ~PR_WANTED;
1.15      pk       1108:                if (ph->ph_nmissing == 0)
                   1109:                        pp->pr_nidle++;
1.3       pk       1110:                wakeup((caddr_t)pp);
                   1111:                return;
                   1112:        }
                   1113:
                   1114:        /*
1.88      chs      1115:         * If this page is now empty, do one of two things:
1.21      thorpej  1116:         *
1.88      chs      1117:         *      (1) If we have more pages than the page high water mark,
1.96      thorpej  1118:         *          free the page back to the system.  ONLY CONSIDER
1.90      thorpej  1119:         *          FREEING BACK A PAGE IF WE HAVE MORE THAN OUR MINIMUM PAGE
                   1120:         *          CLAIM.
1.21      thorpej  1121:         *
1.88      chs      1122:         *      (2) Otherwise, move the page to the empty page list.
                   1123:         *
                   1124:         * Either way, select a new current page (so we use a partially-full
                   1125:         * page if one is available).
1.3       pk       1126:         */
                   1127:        if (ph->ph_nmissing == 0) {
1.6       thorpej  1128:                pp->pr_nidle++;
1.90      thorpej  1129:                if (pp->pr_npages > pp->pr_minpages &&
                   1130:                    (pp->pr_npages > pp->pr_maxpages ||
                   1131:                     (pp->pr_alloc->pa_flags & PA_WANT) != 0)) {
1.101     thorpej  1132:                        pr_rmpage(pp, ph, pq);
1.3       pk       1133:                } else {
1.88      chs      1134:                        LIST_REMOVE(ph, ph_pagelist);
                   1135:                        LIST_INSERT_HEAD(&pp->pr_emptypages, ph, ph_pagelist);
1.3       pk       1136:
1.21      thorpej  1137:                        /*
                   1138:                         * Update the timestamp on the page.  A page must
                   1139:                         * be idle for some period of time before it can
                   1140:                         * be reclaimed by the pagedaemon.  This minimizes
                   1141:                         * ping-pong'ing for memory.
                   1142:                         */
                   1143:                        s = splclock();
                   1144:                        ph->ph_time = mono_time;
                   1145:                        splx(s);
1.1       pk       1146:                }
1.88      chs      1147:                pool_update_curpage(pp);
1.1       pk       1148:        }
1.88      chs      1149:
1.21      thorpej  1150:        /*
1.88      chs      1151:         * If the page was previously completely full, move it to the
                   1152:         * partially-full list and make it the current page.  The next
                   1153:         * allocation will get the item from this page, instead of
                   1154:         * further fragmenting the pool.
1.21      thorpej  1155:         */
                   1156:        else if (ph->ph_nmissing == (pp->pr_itemsperpage - 1)) {
1.88      chs      1157:                LIST_REMOVE(ph, ph_pagelist);
                   1158:                LIST_INSERT_HEAD(&pp->pr_partpages, ph, ph_pagelist);
1.21      thorpej  1159:                pp->pr_curpage = ph;
                   1160:        }
1.43      thorpej  1161: }
                   1162:
                   1163: /*
                   1164:  * Return resource to the pool; must be called at appropriate spl level
                   1165:  */
1.59      thorpej  1166: #ifdef POOL_DIAGNOSTIC
1.43      thorpej  1167: void
                   1168: _pool_put(struct pool *pp, void *v, const char *file, long line)
                   1169: {
1.101     thorpej  1170:        struct pool_pagelist pq;
                   1171:
                   1172:        LIST_INIT(&pq);
1.43      thorpej  1173:
                   1174:        simple_lock(&pp->pr_slock);
                   1175:        pr_enter(pp, file, line);
                   1176:
1.56      sommerfe 1177:        pr_log(pp, v, PRLOG_PUT, file, line);
                   1178:
1.101     thorpej  1179:        pool_do_put(pp, v, &pq);
1.21      thorpej  1180:
1.25      thorpej  1181:        pr_leave(pp);
1.21      thorpej  1182:        simple_unlock(&pp->pr_slock);
1.101     thorpej  1183:
1.102     chs      1184:        pr_pagelist_free(pp, &pq);
1.1       pk       1185: }
1.57      sommerfe 1186: #undef pool_put
1.59      thorpej  1187: #endif /* POOL_DIAGNOSTIC */
1.1       pk       1188:
1.56      sommerfe 1189: void
                   1190: pool_put(struct pool *pp, void *v)
                   1191: {
1.101     thorpej  1192:        struct pool_pagelist pq;
                   1193:
                   1194:        LIST_INIT(&pq);
1.56      sommerfe 1195:
                   1196:        simple_lock(&pp->pr_slock);
1.101     thorpej  1197:        pool_do_put(pp, v, &pq);
                   1198:        simple_unlock(&pp->pr_slock);
1.56      sommerfe 1199:
1.102     chs      1200:        pr_pagelist_free(pp, &pq);
1.56      sommerfe 1201: }
1.57      sommerfe 1202:
1.59      thorpej  1203: #ifdef POOL_DIAGNOSTIC
1.57      sommerfe 1204: #define                pool_put(h, v)  _pool_put((h), (v), __FILE__, __LINE__)
1.56      sommerfe 1205: #endif
1.74      thorpej  1206:
                   1207: /*
1.113     yamt     1208:  * pool_grow: grow a pool by a page.
                   1209:  *
                   1210:  * => called with pool locked.
                   1211:  * => unlock and relock the pool.
                   1212:  * => return with pool locked.
                   1213:  */
                   1214:
                   1215: static int
                   1216: pool_grow(struct pool *pp, int flags)
                   1217: {
                   1218:        struct pool_item_header *ph = NULL;
                   1219:        char *cp;
                   1220:
                   1221:        simple_unlock(&pp->pr_slock);
                   1222:        cp = pool_allocator_alloc(pp, flags);
                   1223:        if (__predict_true(cp != NULL)) {
                   1224:                ph = pool_alloc_item_header(pp, cp, flags);
                   1225:        }
                   1226:        if (__predict_false(cp == NULL || ph == NULL)) {
                   1227:                if (cp != NULL) {
                   1228:                        pool_allocator_free(pp, cp);
                   1229:                }
                   1230:                simple_lock(&pp->pr_slock);
                   1231:                return ENOMEM;
                   1232:        }
                   1233:
                   1234:        simple_lock(&pp->pr_slock);
                   1235:        pool_prime_page(pp, cp, ph);
                   1236:        pp->pr_npagealloc++;
                   1237:        return 0;
                   1238: }
                   1239:
                   1240: /*
1.74      thorpej  1241:  * Add N items to the pool.
                   1242:  */
                   1243: int
                   1244: pool_prime(struct pool *pp, int n)
                   1245: {
1.75      simonb   1246:        int newpages;
1.113     yamt     1247:        int error = 0;
1.74      thorpej  1248:
                   1249:        simple_lock(&pp->pr_slock);
                   1250:
                   1251:        newpages = roundup(n, pp->pr_itemsperpage) / pp->pr_itemsperpage;
                   1252:
                   1253:        while (newpages-- > 0) {
1.113     yamt     1254:                error = pool_grow(pp, PR_NOWAIT);
                   1255:                if (error) {
1.74      thorpej  1256:                        break;
                   1257:                }
                   1258:                pp->pr_minpages++;
                   1259:        }
                   1260:
                   1261:        if (pp->pr_minpages >= pp->pr_maxpages)
                   1262:                pp->pr_maxpages = pp->pr_minpages + 1;  /* XXX */
                   1263:
                   1264:        simple_unlock(&pp->pr_slock);
1.113     yamt     1265:        return error;
1.74      thorpej  1266: }
1.55      thorpej  1267:
                   1268: /*
1.3       pk       1269:  * Add a page worth of items to the pool.
1.21      thorpej  1270:  *
                   1271:  * Note, we must be called with the pool descriptor LOCKED.
1.3       pk       1272:  */
1.55      thorpej  1273: static void
                   1274: pool_prime_page(struct pool *pp, caddr_t storage, struct pool_item_header *ph)
1.3       pk       1275: {
                   1276:        struct pool_item *pi;
                   1277:        caddr_t cp = storage;
                   1278:        unsigned int align = pp->pr_align;
                   1279:        unsigned int ioff = pp->pr_itemoffset;
1.55      thorpej  1280:        int n;
1.89      yamt     1281:        int s;
1.36      pk       1282:
1.91      yamt     1283:        LOCK_ASSERT(simple_lock_held(&pp->pr_slock));
                   1284:
1.66      thorpej  1285: #ifdef DIAGNOSTIC
                   1286:        if (((u_long)cp & (pp->pr_alloc->pa_pagesz - 1)) != 0)
1.36      pk       1287:                panic("pool_prime_page: %s: unaligned page", pp->pr_wchan);
1.66      thorpej  1288: #endif
1.3       pk       1289:
                   1290:        /*
                   1291:         * Insert page header.
                   1292:         */
1.88      chs      1293:        LIST_INSERT_HEAD(&pp->pr_emptypages, ph, ph_pagelist);
1.102     chs      1294:        LIST_INIT(&ph->ph_itemlist);
1.3       pk       1295:        ph->ph_page = storage;
                   1296:        ph->ph_nmissing = 0;
1.89      yamt     1297:        s = splclock();
                   1298:        ph->ph_time = mono_time;
                   1299:        splx(s);
1.88      chs      1300:        if ((pp->pr_roflags & PR_PHINPAGE) == 0)
                   1301:                SPLAY_INSERT(phtree, &pp->pr_phtree, ph);
1.3       pk       1302:
1.6       thorpej  1303:        pp->pr_nidle++;
                   1304:
1.3       pk       1305:        /*
                   1306:         * Color this page.
                   1307:         */
                   1308:        cp = (caddr_t)(cp + pp->pr_curcolor);
                   1309:        if ((pp->pr_curcolor += align) > pp->pr_maxcolor)
                   1310:                pp->pr_curcolor = 0;
                   1311:
                   1312:        /*
                   1313:         * Adjust storage to apply aligment to `pr_itemoffset' in each item.
                   1314:         */
                   1315:        if (ioff != 0)
                   1316:                cp = (caddr_t)(cp + (align - ioff));
                   1317:
                   1318:        /*
                   1319:         * Insert remaining chunks on the bucket list.
                   1320:         */
                   1321:        n = pp->pr_itemsperpage;
1.20      thorpej  1322:        pp->pr_nitems += n;
1.3       pk       1323:
1.97      yamt     1324:        if (pp->pr_roflags & PR_NOTOUCH) {
1.99      yamt     1325:                pool_item_freelist_t *freelist = PR_FREELIST(ph);
1.97      yamt     1326:                int i;
                   1327:
1.99      yamt     1328:                ph->ph_off = cp - storage;
1.97      yamt     1329:                ph->ph_firstfree = 0;
                   1330:                for (i = 0; i < n - 1; i++)
                   1331:                        freelist[i] = i + 1;
                   1332:                freelist[n - 1] = PR_INDEX_EOL;
                   1333:        } else {
                   1334:                while (n--) {
                   1335:                        pi = (struct pool_item *)cp;
1.78      thorpej  1336:
1.97      yamt     1337:                        KASSERT(((((vaddr_t)pi) + ioff) & (align - 1)) == 0);
1.3       pk       1338:
1.97      yamt     1339:                        /* Insert on page list */
1.102     chs      1340:                        LIST_INSERT_HEAD(&ph->ph_itemlist, pi, pi_list);
1.3       pk       1341: #ifdef DIAGNOSTIC
1.97      yamt     1342:                        pi->pi_magic = PI_MAGIC;
1.3       pk       1343: #endif
1.97      yamt     1344:                        cp = (caddr_t)(cp + pp->pr_size);
                   1345:                }
1.3       pk       1346:        }
                   1347:
                   1348:        /*
                   1349:         * If the pool was depleted, point at the new page.
                   1350:         */
                   1351:        if (pp->pr_curpage == NULL)
                   1352:                pp->pr_curpage = ph;
                   1353:
                   1354:        if (++pp->pr_npages > pp->pr_hiwat)
                   1355:                pp->pr_hiwat = pp->pr_npages;
                   1356: }
                   1357:
1.20      thorpej  1358: /*
1.52      thorpej  1359:  * Used by pool_get() when nitems drops below the low water mark.  This
1.88      chs      1360:  * is used to catch up pr_nitems with the low water mark.
1.20      thorpej  1361:  *
1.21      thorpej  1362:  * Note 1, we never wait for memory here, we let the caller decide what to do.
1.20      thorpej  1363:  *
1.73      thorpej  1364:  * Note 2, we must be called with the pool already locked, and we return
1.20      thorpej  1365:  * with it locked.
                   1366:  */
                   1367: static int
1.42      thorpej  1368: pool_catchup(struct pool *pp)
1.20      thorpej  1369: {
                   1370:        int error = 0;
                   1371:
1.54      thorpej  1372:        while (POOL_NEEDS_CATCHUP(pp)) {
1.113     yamt     1373:                error = pool_grow(pp, PR_NOWAIT);
                   1374:                if (error) {
1.20      thorpej  1375:                        break;
                   1376:                }
                   1377:        }
1.113     yamt     1378:        return error;
1.20      thorpej  1379: }
                   1380:
1.88      chs      1381: static void
                   1382: pool_update_curpage(struct pool *pp)
                   1383: {
                   1384:
                   1385:        pp->pr_curpage = LIST_FIRST(&pp->pr_partpages);
                   1386:        if (pp->pr_curpage == NULL) {
                   1387:                pp->pr_curpage = LIST_FIRST(&pp->pr_emptypages);
                   1388:        }
                   1389: }
                   1390:
1.3       pk       1391: void
1.42      thorpej  1392: pool_setlowat(struct pool *pp, int n)
1.3       pk       1393: {
1.15      pk       1394:
1.21      thorpej  1395:        simple_lock(&pp->pr_slock);
                   1396:
1.3       pk       1397:        pp->pr_minitems = n;
1.15      pk       1398:        pp->pr_minpages = (n == 0)
                   1399:                ? 0
1.18      thorpej  1400:                : roundup(n, pp->pr_itemsperpage) / pp->pr_itemsperpage;
1.20      thorpej  1401:
                   1402:        /* Make sure we're caught up with the newly-set low water mark. */
1.75      simonb   1403:        if (POOL_NEEDS_CATCHUP(pp) && pool_catchup(pp) != 0) {
1.20      thorpej  1404:                /*
                   1405:                 * XXX: Should we log a warning?  Should we set up a timeout
                   1406:                 * to try again in a second or so?  The latter could break
                   1407:                 * a caller's assumptions about interrupt protection, etc.
                   1408:                 */
                   1409:        }
1.21      thorpej  1410:
                   1411:        simple_unlock(&pp->pr_slock);
1.3       pk       1412: }
                   1413:
                   1414: void
1.42      thorpej  1415: pool_sethiwat(struct pool *pp, int n)
1.3       pk       1416: {
1.15      pk       1417:
1.21      thorpej  1418:        simple_lock(&pp->pr_slock);
                   1419:
1.15      pk       1420:        pp->pr_maxpages = (n == 0)
                   1421:                ? 0
1.18      thorpej  1422:                : roundup(n, pp->pr_itemsperpage) / pp->pr_itemsperpage;
1.21      thorpej  1423:
                   1424:        simple_unlock(&pp->pr_slock);
1.3       pk       1425: }
                   1426:
1.20      thorpej  1427: void
1.42      thorpej  1428: pool_sethardlimit(struct pool *pp, int n, const char *warnmess, int ratecap)
1.20      thorpej  1429: {
                   1430:
1.21      thorpej  1431:        simple_lock(&pp->pr_slock);
1.20      thorpej  1432:
                   1433:        pp->pr_hardlimit = n;
                   1434:        pp->pr_hardlimit_warning = warnmess;
1.31      thorpej  1435:        pp->pr_hardlimit_ratecap.tv_sec = ratecap;
                   1436:        pp->pr_hardlimit_warning_last.tv_sec = 0;
                   1437:        pp->pr_hardlimit_warning_last.tv_usec = 0;
1.20      thorpej  1438:
                   1439:        /*
1.21      thorpej  1440:         * In-line version of pool_sethiwat(), because we don't want to
                   1441:         * release the lock.
1.20      thorpej  1442:         */
                   1443:        pp->pr_maxpages = (n == 0)
                   1444:                ? 0
                   1445:                : roundup(n, pp->pr_itemsperpage) / pp->pr_itemsperpage;
1.21      thorpej  1446:
                   1447:        simple_unlock(&pp->pr_slock);
1.20      thorpej  1448: }
1.3       pk       1449:
                   1450: /*
                   1451:  * Release all complete pages that have not been used recently.
                   1452:  */
1.66      thorpej  1453: int
1.59      thorpej  1454: #ifdef POOL_DIAGNOSTIC
1.42      thorpej  1455: _pool_reclaim(struct pool *pp, const char *file, long line)
1.56      sommerfe 1456: #else
                   1457: pool_reclaim(struct pool *pp)
                   1458: #endif
1.3       pk       1459: {
                   1460:        struct pool_item_header *ph, *phnext;
1.43      thorpej  1461:        struct pool_cache *pc;
1.61      chs      1462:        struct pool_pagelist pq;
1.102     chs      1463:        struct pool_cache_grouplist pcgl;
                   1464:        struct timeval curtime, diff;
1.21      thorpej  1465:        int s;
1.3       pk       1466:
1.68      thorpej  1467:        if (pp->pr_drain_hook != NULL) {
                   1468:                /*
                   1469:                 * The drain hook must be called with the pool unlocked.
                   1470:                 */
                   1471:                (*pp->pr_drain_hook)(pp->pr_drain_hook_arg, PR_NOWAIT);
                   1472:        }
                   1473:
1.21      thorpej  1474:        if (simple_lock_try(&pp->pr_slock) == 0)
1.66      thorpej  1475:                return (0);
1.25      thorpej  1476:        pr_enter(pp, file, line);
1.68      thorpej  1477:
1.88      chs      1478:        LIST_INIT(&pq);
1.102     chs      1479:        LIST_INIT(&pcgl);
1.3       pk       1480:
1.43      thorpej  1481:        /*
                   1482:         * Reclaim items from the pool's caches.
                   1483:         */
1.102     chs      1484:        LIST_FOREACH(pc, &pp->pr_cachelist, pc_poollist)
                   1485:                pool_cache_reclaim(pc, &pq, &pcgl);
1.43      thorpej  1486:
1.21      thorpej  1487:        s = splclock();
                   1488:        curtime = mono_time;
                   1489:        splx(s);
                   1490:
1.88      chs      1491:        for (ph = LIST_FIRST(&pp->pr_emptypages); ph != NULL; ph = phnext) {
                   1492:                phnext = LIST_NEXT(ph, ph_pagelist);
1.3       pk       1493:
                   1494:                /* Check our minimum page claim */
                   1495:                if (pp->pr_npages <= pp->pr_minpages)
                   1496:                        break;
                   1497:
1.88      chs      1498:                KASSERT(ph->ph_nmissing == 0);
                   1499:                timersub(&curtime, &ph->ph_time, &diff);
                   1500:                if (diff.tv_sec < pool_inactive_time)
                   1501:                        continue;
1.21      thorpej  1502:
1.88      chs      1503:                /*
                   1504:                 * If freeing this page would put us below
                   1505:                 * the low water mark, stop now.
                   1506:                 */
                   1507:                if ((pp->pr_nitems - pp->pr_itemsperpage) <
                   1508:                    pp->pr_minitems)
                   1509:                        break;
1.21      thorpej  1510:
1.88      chs      1511:                pr_rmpage(pp, ph, &pq);
1.3       pk       1512:        }
                   1513:
1.25      thorpej  1514:        pr_leave(pp);
1.21      thorpej  1515:        simple_unlock(&pp->pr_slock);
1.102     chs      1516:        if (LIST_EMPTY(&pq) && LIST_EMPTY(&pcgl))
                   1517:                return 0;
1.66      thorpej  1518:
1.101     thorpej  1519:        pr_pagelist_free(pp, &pq);
1.102     chs      1520:        pcg_grouplist_free(&pcgl);
1.66      thorpej  1521:        return (1);
1.3       pk       1522: }
                   1523:
                   1524: /*
                   1525:  * Drain pools, one at a time.
1.21      thorpej  1526:  *
                   1527:  * Note, we must never be called from an interrupt context.
1.3       pk       1528:  */
                   1529: void
1.42      thorpej  1530: pool_drain(void *arg)
1.3       pk       1531: {
                   1532:        struct pool *pp;
1.23      thorpej  1533:        int s;
1.3       pk       1534:
1.61      chs      1535:        pp = NULL;
1.49      thorpej  1536:        s = splvm();
1.23      thorpej  1537:        simple_lock(&pool_head_slock);
1.61      chs      1538:        if (drainpp == NULL) {
1.102     chs      1539:                drainpp = LIST_FIRST(&pool_head);
1.61      chs      1540:        }
                   1541:        if (drainpp) {
                   1542:                pp = drainpp;
1.102     chs      1543:                drainpp = LIST_NEXT(pp, pr_poollist);
1.61      chs      1544:        }
                   1545:        simple_unlock(&pool_head_slock);
1.115     christos 1546:        if (pp)
                   1547:                pool_reclaim(pp);
1.61      chs      1548:        splx(s);
1.3       pk       1549: }
                   1550:
                   1551: /*
                   1552:  * Diagnostic helpers.
                   1553:  */
                   1554: void
1.42      thorpej  1555: pool_print(struct pool *pp, const char *modif)
1.21      thorpej  1556: {
                   1557:        int s;
                   1558:
1.49      thorpej  1559:        s = splvm();
1.25      thorpej  1560:        if (simple_lock_try(&pp->pr_slock) == 0) {
                   1561:                printf("pool %s is locked; try again later\n",
                   1562:                    pp->pr_wchan);
                   1563:                splx(s);
                   1564:                return;
                   1565:        }
                   1566:        pool_print1(pp, modif, printf);
1.21      thorpej  1567:        simple_unlock(&pp->pr_slock);
                   1568:        splx(s);
                   1569: }
                   1570:
1.25      thorpej  1571: void
1.108     yamt     1572: pool_printall(const char *modif, void (*pr)(const char *, ...))
                   1573: {
                   1574:        struct pool *pp;
                   1575:
                   1576:        if (simple_lock_try(&pool_head_slock) == 0) {
                   1577:                (*pr)("WARNING: pool_head_slock is locked\n");
                   1578:        } else {
                   1579:                simple_unlock(&pool_head_slock);
                   1580:        }
                   1581:
                   1582:        LIST_FOREACH(pp, &pool_head, pr_poollist) {
                   1583:                pool_printit(pp, modif, pr);
                   1584:        }
                   1585: }
                   1586:
                   1587: void
1.42      thorpej  1588: pool_printit(struct pool *pp, const char *modif, void (*pr)(const char *, ...))
1.25      thorpej  1589: {
                   1590:
                   1591:        if (pp == NULL) {
                   1592:                (*pr)("Must specify a pool to print.\n");
                   1593:                return;
                   1594:        }
                   1595:
                   1596:        /*
                   1597:         * Called from DDB; interrupts should be blocked, and all
                   1598:         * other processors should be paused.  We can skip locking
                   1599:         * the pool in this case.
                   1600:         *
                   1601:         * We do a simple_lock_try() just to print the lock
                   1602:         * status, however.
                   1603:         */
                   1604:
                   1605:        if (simple_lock_try(&pp->pr_slock) == 0)
                   1606:                (*pr)("WARNING: pool %s is locked\n", pp->pr_wchan);
                   1607:        else
1.107     yamt     1608:                simple_unlock(&pp->pr_slock);
1.25      thorpej  1609:
                   1610:        pool_print1(pp, modif, pr);
                   1611: }
                   1612:
1.21      thorpej  1613: static void
1.97      yamt     1614: pool_print_pagelist(struct pool *pp, struct pool_pagelist *pl,
                   1615:     void (*pr)(const char *, ...))
1.88      chs      1616: {
                   1617:        struct pool_item_header *ph;
                   1618: #ifdef DIAGNOSTIC
                   1619:        struct pool_item *pi;
                   1620: #endif
                   1621:
                   1622:        LIST_FOREACH(ph, pl, ph_pagelist) {
                   1623:                (*pr)("\t\tpage %p, nmissing %d, time %lu,%lu\n",
                   1624:                    ph->ph_page, ph->ph_nmissing,
                   1625:                    (u_long)ph->ph_time.tv_sec,
                   1626:                    (u_long)ph->ph_time.tv_usec);
                   1627: #ifdef DIAGNOSTIC
1.97      yamt     1628:                if (!(pp->pr_roflags & PR_NOTOUCH)) {
1.102     chs      1629:                        LIST_FOREACH(pi, &ph->ph_itemlist, pi_list) {
1.97      yamt     1630:                                if (pi->pi_magic != PI_MAGIC) {
                   1631:                                        (*pr)("\t\t\titem %p, magic 0x%x\n",
                   1632:                                            pi, pi->pi_magic);
                   1633:                                }
1.88      chs      1634:                        }
                   1635:                }
                   1636: #endif
                   1637:        }
                   1638: }
                   1639:
                   1640: static void
1.42      thorpej  1641: pool_print1(struct pool *pp, const char *modif, void (*pr)(const char *, ...))
1.3       pk       1642: {
1.25      thorpej  1643:        struct pool_item_header *ph;
1.44      thorpej  1644:        struct pool_cache *pc;
                   1645:        struct pool_cache_group *pcg;
                   1646:        int i, print_log = 0, print_pagelist = 0, print_cache = 0;
1.25      thorpej  1647:        char c;
                   1648:
                   1649:        while ((c = *modif++) != '\0') {
                   1650:                if (c == 'l')
                   1651:                        print_log = 1;
                   1652:                if (c == 'p')
                   1653:                        print_pagelist = 1;
1.44      thorpej  1654:                if (c == 'c')
                   1655:                        print_cache = 1;
1.25      thorpej  1656:        }
                   1657:
                   1658:        (*pr)("POOL %s: size %u, align %u, ioff %u, roflags 0x%08x\n",
                   1659:            pp->pr_wchan, pp->pr_size, pp->pr_align, pp->pr_itemoffset,
                   1660:            pp->pr_roflags);
1.66      thorpej  1661:        (*pr)("\talloc %p\n", pp->pr_alloc);
1.25      thorpej  1662:        (*pr)("\tminitems %u, minpages %u, maxpages %u, npages %u\n",
                   1663:            pp->pr_minitems, pp->pr_minpages, pp->pr_maxpages, pp->pr_npages);
                   1664:        (*pr)("\titemsperpage %u, nitems %u, nout %u, hardlimit %u\n",
                   1665:            pp->pr_itemsperpage, pp->pr_nitems, pp->pr_nout, pp->pr_hardlimit);
                   1666:
                   1667:        (*pr)("\n\tnget %lu, nfail %lu, nput %lu\n",
                   1668:            pp->pr_nget, pp->pr_nfail, pp->pr_nput);
                   1669:        (*pr)("\tnpagealloc %lu, npagefree %lu, hiwat %u, nidle %lu\n",
                   1670:            pp->pr_npagealloc, pp->pr_npagefree, pp->pr_hiwat, pp->pr_nidle);
                   1671:
                   1672:        if (print_pagelist == 0)
                   1673:                goto skip_pagelist;
                   1674:
1.88      chs      1675:        if ((ph = LIST_FIRST(&pp->pr_emptypages)) != NULL)
                   1676:                (*pr)("\n\tempty page list:\n");
1.97      yamt     1677:        pool_print_pagelist(pp, &pp->pr_emptypages, pr);
1.88      chs      1678:        if ((ph = LIST_FIRST(&pp->pr_fullpages)) != NULL)
                   1679:                (*pr)("\n\tfull page list:\n");
1.97      yamt     1680:        pool_print_pagelist(pp, &pp->pr_fullpages, pr);
1.88      chs      1681:        if ((ph = LIST_FIRST(&pp->pr_partpages)) != NULL)
                   1682:                (*pr)("\n\tpartial-page list:\n");
1.97      yamt     1683:        pool_print_pagelist(pp, &pp->pr_partpages, pr);
1.88      chs      1684:
1.25      thorpej  1685:        if (pp->pr_curpage == NULL)
                   1686:                (*pr)("\tno current page\n");
                   1687:        else
                   1688:                (*pr)("\tcurpage %p\n", pp->pr_curpage->ph_page);
                   1689:
                   1690:  skip_pagelist:
                   1691:        if (print_log == 0)
                   1692:                goto skip_log;
                   1693:
                   1694:        (*pr)("\n");
                   1695:        if ((pp->pr_roflags & PR_LOGGING) == 0)
                   1696:                (*pr)("\tno log\n");
                   1697:        else
                   1698:                pr_printlog(pp, NULL, pr);
1.3       pk       1699:
1.25      thorpej  1700:  skip_log:
1.44      thorpej  1701:        if (print_cache == 0)
                   1702:                goto skip_cache;
                   1703:
1.102     chs      1704: #define PR_GROUPLIST(pcg)                                              \
                   1705:        (*pr)("\t\tgroup %p: avail %d\n", pcg, pcg->pcg_avail);         \
                   1706:        for (i = 0; i < PCG_NOBJECTS; i++) {                            \
                   1707:                if (pcg->pcg_objects[i].pcgo_pa !=                      \
                   1708:                    POOL_PADDR_INVALID) {                               \
                   1709:                        (*pr)("\t\t\t%p, 0x%llx\n",                     \
                   1710:                            pcg->pcg_objects[i].pcgo_va,                \
                   1711:                            (unsigned long long)                        \
                   1712:                            pcg->pcg_objects[i].pcgo_pa);               \
                   1713:                } else {                                                \
                   1714:                        (*pr)("\t\t\t%p\n",                             \
                   1715:                            pcg->pcg_objects[i].pcgo_va);               \
                   1716:                }                                                       \
                   1717:        }
                   1718:
                   1719:        LIST_FOREACH(pc, &pp->pr_cachelist, pc_poollist) {
1.103     chs      1720:                (*pr)("\tcache %p\n", pc);
1.48      thorpej  1721:                (*pr)("\t    hits %lu misses %lu ngroups %lu nitems %lu\n",
                   1722:                    pc->pc_hits, pc->pc_misses, pc->pc_ngroups, pc->pc_nitems);
1.102     chs      1723:                (*pr)("\t    full groups:\n");
1.103     chs      1724:                LIST_FOREACH(pcg, &pc->pc_fullgroups, pcg_list) {
1.102     chs      1725:                        PR_GROUPLIST(pcg);
1.103     chs      1726:                }
1.102     chs      1727:                (*pr)("\t    partial groups:\n");
1.103     chs      1728:                LIST_FOREACH(pcg, &pc->pc_partgroups, pcg_list) {
1.102     chs      1729:                        PR_GROUPLIST(pcg);
1.103     chs      1730:                }
1.102     chs      1731:                (*pr)("\t    empty groups:\n");
1.103     chs      1732:                LIST_FOREACH(pcg, &pc->pc_emptygroups, pcg_list) {
1.102     chs      1733:                        PR_GROUPLIST(pcg);
1.103     chs      1734:                }
1.44      thorpej  1735:        }
1.102     chs      1736: #undef PR_GROUPLIST
1.44      thorpej  1737:
                   1738:  skip_cache:
1.88      chs      1739:        pr_enter_check(pp, pr);
                   1740: }
                   1741:
                   1742: static int
                   1743: pool_chk_page(struct pool *pp, const char *label, struct pool_item_header *ph)
                   1744: {
                   1745:        struct pool_item *pi;
                   1746:        caddr_t page;
                   1747:        int n;
                   1748:
                   1749:        page = (caddr_t)((u_long)ph & pp->pr_alloc->pa_pagemask);
                   1750:        if (page != ph->ph_page &&
                   1751:            (pp->pr_roflags & PR_PHINPAGE) != 0) {
                   1752:                if (label != NULL)
                   1753:                        printf("%s: ", label);
                   1754:                printf("pool(%p:%s): page inconsistency: page %p;"
                   1755:                       " at page head addr %p (p %p)\n", pp,
                   1756:                        pp->pr_wchan, ph->ph_page,
                   1757:                        ph, page);
                   1758:                return 1;
                   1759:        }
1.3       pk       1760:
1.97      yamt     1761:        if ((pp->pr_roflags & PR_NOTOUCH) != 0)
                   1762:                return 0;
                   1763:
1.102     chs      1764:        for (pi = LIST_FIRST(&ph->ph_itemlist), n = 0;
1.88      chs      1765:             pi != NULL;
1.102     chs      1766:             pi = LIST_NEXT(pi,pi_list), n++) {
1.88      chs      1767:
                   1768: #ifdef DIAGNOSTIC
                   1769:                if (pi->pi_magic != PI_MAGIC) {
                   1770:                        if (label != NULL)
                   1771:                                printf("%s: ", label);
                   1772:                        printf("pool(%s): free list modified: magic=%x;"
                   1773:                               " page %p; item ordinal %d;"
                   1774:                               " addr %p (p %p)\n",
                   1775:                                pp->pr_wchan, pi->pi_magic, ph->ph_page,
                   1776:                                n, pi, page);
                   1777:                        panic("pool");
                   1778:                }
                   1779: #endif
                   1780:                page =
                   1781:                    (caddr_t)((u_long)pi & pp->pr_alloc->pa_pagemask);
                   1782:                if (page == ph->ph_page)
                   1783:                        continue;
                   1784:
                   1785:                if (label != NULL)
                   1786:                        printf("%s: ", label);
                   1787:                printf("pool(%p:%s): page inconsistency: page %p;"
                   1788:                       " item ordinal %d; addr %p (p %p)\n", pp,
                   1789:                        pp->pr_wchan, ph->ph_page,
                   1790:                        n, pi, page);
                   1791:                return 1;
                   1792:        }
                   1793:        return 0;
1.3       pk       1794: }
                   1795:
1.88      chs      1796:
1.3       pk       1797: int
1.42      thorpej  1798: pool_chk(struct pool *pp, const char *label)
1.3       pk       1799: {
                   1800:        struct pool_item_header *ph;
                   1801:        int r = 0;
                   1802:
1.21      thorpej  1803:        simple_lock(&pp->pr_slock);
1.88      chs      1804:        LIST_FOREACH(ph, &pp->pr_emptypages, ph_pagelist) {
                   1805:                r = pool_chk_page(pp, label, ph);
                   1806:                if (r) {
                   1807:                        goto out;
                   1808:                }
                   1809:        }
                   1810:        LIST_FOREACH(ph, &pp->pr_fullpages, ph_pagelist) {
                   1811:                r = pool_chk_page(pp, label, ph);
                   1812:                if (r) {
1.3       pk       1813:                        goto out;
                   1814:                }
1.88      chs      1815:        }
                   1816:        LIST_FOREACH(ph, &pp->pr_partpages, ph_pagelist) {
                   1817:                r = pool_chk_page(pp, label, ph);
                   1818:                if (r) {
1.3       pk       1819:                        goto out;
                   1820:                }
                   1821:        }
1.88      chs      1822:
1.3       pk       1823: out:
1.21      thorpej  1824:        simple_unlock(&pp->pr_slock);
1.3       pk       1825:        return (r);
1.43      thorpej  1826: }
                   1827:
                   1828: /*
                   1829:  * pool_cache_init:
                   1830:  *
                   1831:  *     Initialize a pool cache.
                   1832:  *
                   1833:  *     NOTE: If the pool must be protected from interrupts, we expect
                   1834:  *     to be called at the appropriate interrupt priority level.
                   1835:  */
                   1836: void
                   1837: pool_cache_init(struct pool_cache *pc, struct pool *pp,
                   1838:     int (*ctor)(void *, void *, int),
                   1839:     void (*dtor)(void *, void *),
                   1840:     void *arg)
                   1841: {
                   1842:
1.102     chs      1843:        LIST_INIT(&pc->pc_emptygroups);
                   1844:        LIST_INIT(&pc->pc_fullgroups);
                   1845:        LIST_INIT(&pc->pc_partgroups);
1.43      thorpej  1846:        simple_lock_init(&pc->pc_slock);
                   1847:
                   1848:        pc->pc_pool = pp;
                   1849:
                   1850:        pc->pc_ctor = ctor;
                   1851:        pc->pc_dtor = dtor;
                   1852:        pc->pc_arg  = arg;
                   1853:
1.48      thorpej  1854:        pc->pc_hits   = 0;
                   1855:        pc->pc_misses = 0;
                   1856:
                   1857:        pc->pc_ngroups = 0;
                   1858:
                   1859:        pc->pc_nitems = 0;
                   1860:
1.43      thorpej  1861:        simple_lock(&pp->pr_slock);
1.102     chs      1862:        LIST_INSERT_HEAD(&pp->pr_cachelist, pc, pc_poollist);
1.43      thorpej  1863:        simple_unlock(&pp->pr_slock);
                   1864: }
                   1865:
                   1866: /*
                   1867:  * pool_cache_destroy:
                   1868:  *
                   1869:  *     Destroy a pool cache.
                   1870:  */
                   1871: void
                   1872: pool_cache_destroy(struct pool_cache *pc)
                   1873: {
                   1874:        struct pool *pp = pc->pc_pool;
                   1875:
                   1876:        /* First, invalidate the entire cache. */
                   1877:        pool_cache_invalidate(pc);
                   1878:
                   1879:        /* ...and remove it from the pool's cache list. */
                   1880:        simple_lock(&pp->pr_slock);
1.102     chs      1881:        LIST_REMOVE(pc, pc_poollist);
1.43      thorpej  1882:        simple_unlock(&pp->pr_slock);
                   1883: }
                   1884:
1.110     perry    1885: static inline void *
1.87      thorpej  1886: pcg_get(struct pool_cache_group *pcg, paddr_t *pap)
1.43      thorpej  1887: {
                   1888:        void *object;
                   1889:        u_int idx;
                   1890:
                   1891:        KASSERT(pcg->pcg_avail <= PCG_NOBJECTS);
1.45      thorpej  1892:        KASSERT(pcg->pcg_avail != 0);
1.43      thorpej  1893:        idx = --pcg->pcg_avail;
                   1894:
1.87      thorpej  1895:        KASSERT(pcg->pcg_objects[idx].pcgo_va != NULL);
                   1896:        object = pcg->pcg_objects[idx].pcgo_va;
                   1897:        if (pap != NULL)
                   1898:                *pap = pcg->pcg_objects[idx].pcgo_pa;
                   1899:        pcg->pcg_objects[idx].pcgo_va = NULL;
1.43      thorpej  1900:
                   1901:        return (object);
                   1902: }
                   1903:
1.110     perry    1904: static inline void
1.87      thorpej  1905: pcg_put(struct pool_cache_group *pcg, void *object, paddr_t pa)
1.43      thorpej  1906: {
                   1907:        u_int idx;
                   1908:
                   1909:        KASSERT(pcg->pcg_avail < PCG_NOBJECTS);
                   1910:        idx = pcg->pcg_avail++;
                   1911:
1.87      thorpej  1912:        KASSERT(pcg->pcg_objects[idx].pcgo_va == NULL);
                   1913:        pcg->pcg_objects[idx].pcgo_va = object;
                   1914:        pcg->pcg_objects[idx].pcgo_pa = pa;
1.43      thorpej  1915: }
                   1916:
1.102     chs      1917: static void
                   1918: pcg_grouplist_free(struct pool_cache_grouplist *pcgl)
                   1919: {
                   1920:        struct pool_cache_group *pcg;
                   1921:        int s;
                   1922:
                   1923:        s = splvm();
                   1924:        while ((pcg = LIST_FIRST(pcgl)) != NULL) {
                   1925:                LIST_REMOVE(pcg, pcg_list);
                   1926:                pool_put(&pcgpool, pcg);
                   1927:        }
                   1928:        splx(s);
                   1929: }
                   1930:
1.43      thorpej  1931: /*
1.87      thorpej  1932:  * pool_cache_get{,_paddr}:
1.43      thorpej  1933:  *
1.87      thorpej  1934:  *     Get an object from a pool cache (optionally returning
                   1935:  *     the physical address of the object).
1.43      thorpej  1936:  */
                   1937: void *
1.87      thorpej  1938: pool_cache_get_paddr(struct pool_cache *pc, int flags, paddr_t *pap)
1.43      thorpej  1939: {
                   1940:        struct pool_cache_group *pcg;
                   1941:        void *object;
1.58      thorpej  1942:
                   1943: #ifdef LOCKDEBUG
                   1944:        if (flags & PR_WAITOK)
                   1945:                simple_lock_only_held(NULL, "pool_cache_get(PR_WAITOK)");
                   1946: #endif
1.43      thorpej  1947:
                   1948:        simple_lock(&pc->pc_slock);
                   1949:
1.102     chs      1950:        pcg = LIST_FIRST(&pc->pc_partgroups);
                   1951:        if (pcg == NULL) {
                   1952:                pcg = LIST_FIRST(&pc->pc_fullgroups);
                   1953:                if (pcg != NULL) {
                   1954:                        LIST_REMOVE(pcg, pcg_list);
                   1955:                        LIST_INSERT_HEAD(&pc->pc_partgroups, pcg, pcg_list);
1.43      thorpej  1956:                }
1.102     chs      1957:        }
                   1958:        if (pcg == NULL) {
1.43      thorpej  1959:
                   1960:                /*
                   1961:                 * No groups with any available objects.  Allocate
                   1962:                 * a new object, construct it, and return it to
                   1963:                 * the caller.  We will allocate a group, if necessary,
                   1964:                 * when the object is freed back to the cache.
                   1965:                 */
1.48      thorpej  1966:                pc->pc_misses++;
1.43      thorpej  1967:                simple_unlock(&pc->pc_slock);
                   1968:                object = pool_get(pc->pc_pool, flags);
                   1969:                if (object != NULL && pc->pc_ctor != NULL) {
                   1970:                        if ((*pc->pc_ctor)(pc->pc_arg, object, flags) != 0) {
                   1971:                                pool_put(pc->pc_pool, object);
                   1972:                                return (NULL);
                   1973:                        }
                   1974:                }
1.87      thorpej  1975:                if (object != NULL && pap != NULL) {
                   1976: #ifdef POOL_VTOPHYS
                   1977:                        *pap = POOL_VTOPHYS(object);
                   1978: #else
                   1979:                        *pap = POOL_PADDR_INVALID;
                   1980: #endif
                   1981:                }
1.43      thorpej  1982:                return (object);
                   1983:        }
                   1984:
1.48      thorpej  1985:        pc->pc_hits++;
                   1986:        pc->pc_nitems--;
1.87      thorpej  1987:        object = pcg_get(pcg, pap);
1.43      thorpej  1988:
1.102     chs      1989:        if (pcg->pcg_avail == 0) {
                   1990:                LIST_REMOVE(pcg, pcg_list);
                   1991:                LIST_INSERT_HEAD(&pc->pc_emptygroups, pcg, pcg_list);
                   1992:        }
1.43      thorpej  1993:        simple_unlock(&pc->pc_slock);
                   1994:
                   1995:        return (object);
                   1996: }
                   1997:
                   1998: /*
1.87      thorpej  1999:  * pool_cache_put{,_paddr}:
1.43      thorpej  2000:  *
1.87      thorpej  2001:  *     Put an object back to the pool cache (optionally caching the
                   2002:  *     physical address of the object).
1.43      thorpej  2003:  */
                   2004: void
1.87      thorpej  2005: pool_cache_put_paddr(struct pool_cache *pc, void *object, paddr_t pa)
1.43      thorpej  2006: {
                   2007:        struct pool_cache_group *pcg;
1.60      thorpej  2008:        int s;
1.43      thorpej  2009:
1.109     christos 2010:        if (__predict_false((pc->pc_pool->pr_flags & PR_WANTED) != 0)) {
                   2011:                goto destruct;
                   2012:        }
                   2013:
1.43      thorpej  2014:        simple_lock(&pc->pc_slock);
                   2015:
1.102     chs      2016:        pcg = LIST_FIRST(&pc->pc_partgroups);
                   2017:        if (pcg == NULL) {
                   2018:                pcg = LIST_FIRST(&pc->pc_emptygroups);
                   2019:                if (pcg != NULL) {
                   2020:                        LIST_REMOVE(pcg, pcg_list);
                   2021:                        LIST_INSERT_HEAD(&pc->pc_partgroups, pcg, pcg_list);
1.43      thorpej  2022:                }
1.102     chs      2023:        }
                   2024:        if (pcg == NULL) {
1.43      thorpej  2025:
                   2026:                /*
                   2027:                 * No empty groups to free the object to.  Attempt to
1.47      thorpej  2028:                 * allocate one.
1.43      thorpej  2029:                 */
1.47      thorpej  2030:                simple_unlock(&pc->pc_slock);
1.60      thorpej  2031:                s = splvm();
1.43      thorpej  2032:                pcg = pool_get(&pcgpool, PR_NOWAIT);
1.60      thorpej  2033:                splx(s);
1.102     chs      2034:                if (pcg == NULL) {
1.109     christos 2035: destruct:
1.102     chs      2036:
                   2037:                        /*
                   2038:                         * Unable to allocate a cache group; destruct the object
                   2039:                         * and free it back to the pool.
                   2040:                         */
                   2041:                        pool_cache_destruct_object(pc, object);
                   2042:                        return;
1.43      thorpej  2043:                }
1.102     chs      2044:                memset(pcg, 0, sizeof(*pcg));
                   2045:                simple_lock(&pc->pc_slock);
                   2046:                pc->pc_ngroups++;
                   2047:                LIST_INSERT_HEAD(&pc->pc_partgroups, pcg, pcg_list);
1.43      thorpej  2048:        }
                   2049:
1.48      thorpej  2050:        pc->pc_nitems++;
1.87      thorpej  2051:        pcg_put(pcg, object, pa);
1.43      thorpej  2052:
1.102     chs      2053:        if (pcg->pcg_avail == PCG_NOBJECTS) {
                   2054:                LIST_REMOVE(pcg, pcg_list);
                   2055:                LIST_INSERT_HEAD(&pc->pc_fullgroups, pcg, pcg_list);
                   2056:        }
1.43      thorpej  2057:        simple_unlock(&pc->pc_slock);
1.51      thorpej  2058: }
                   2059:
                   2060: /*
                   2061:  * pool_cache_destruct_object:
                   2062:  *
                   2063:  *     Force destruction of an object and its release back into
                   2064:  *     the pool.
                   2065:  */
                   2066: void
                   2067: pool_cache_destruct_object(struct pool_cache *pc, void *object)
                   2068: {
                   2069:
                   2070:        if (pc->pc_dtor != NULL)
                   2071:                (*pc->pc_dtor)(pc->pc_arg, object);
                   2072:        pool_put(pc->pc_pool, object);
1.43      thorpej  2073: }
                   2074:
1.102     chs      2075: static void
1.106     christos 2076: pool_do_cache_invalidate_grouplist(struct pool_cache_grouplist *pcgsl,
1.105     christos 2077:     struct pool_cache *pc, struct pool_pagelist *pq,
1.106     christos 2078:     struct pool_cache_grouplist *pcgdl)
1.102     chs      2079: {
1.106     christos 2080:        struct pool_cache_group *pcg, *npcg;
1.102     chs      2081:        void *object;
                   2082:
1.106     christos 2083:        for (pcg = LIST_FIRST(pcgsl); pcg != NULL; pcg = npcg) {
1.102     chs      2084:                npcg = LIST_NEXT(pcg, pcg_list);
                   2085:                while (pcg->pcg_avail != 0) {
                   2086:                        pc->pc_nitems--;
                   2087:                        object = pcg_get(pcg, NULL);
                   2088:                        if (pc->pc_dtor != NULL)
                   2089:                                (*pc->pc_dtor)(pc->pc_arg, object);
                   2090:                        pool_do_put(pc->pc_pool, object, pq);
                   2091:                }
1.103     chs      2092:                pc->pc_ngroups--;
1.102     chs      2093:                LIST_REMOVE(pcg, pcg_list);
1.106     christos 2094:                LIST_INSERT_HEAD(pcgdl, pcg, pcg_list);
1.102     chs      2095:        }
1.105     christos 2096: }
                   2097:
                   2098: static void
                   2099: pool_do_cache_invalidate(struct pool_cache *pc, struct pool_pagelist *pq,
                   2100:     struct pool_cache_grouplist *pcgl)
                   2101: {
                   2102:
                   2103:        LOCK_ASSERT(simple_lock_held(&pc->pc_slock));
                   2104:        LOCK_ASSERT(simple_lock_held(&pc->pc_pool->pr_slock));
                   2105:
1.106     christos 2106:        pool_do_cache_invalidate_grouplist(&pc->pc_fullgroups, pc, pq, pcgl);
                   2107:        pool_do_cache_invalidate_grouplist(&pc->pc_partgroups, pc, pq, pcgl);
1.103     chs      2108:
                   2109:        KASSERT(LIST_EMPTY(&pc->pc_partgroups));
                   2110:        KASSERT(LIST_EMPTY(&pc->pc_fullgroups));
                   2111:        KASSERT(pc->pc_nitems == 0);
1.102     chs      2112: }
                   2113:
1.43      thorpej  2114: /*
1.101     thorpej  2115:  * pool_cache_invalidate:
1.43      thorpej  2116:  *
1.101     thorpej  2117:  *     Invalidate a pool cache (destruct and release all of the
                   2118:  *     cached objects).
1.43      thorpej  2119:  */
1.101     thorpej  2120: void
                   2121: pool_cache_invalidate(struct pool_cache *pc)
1.43      thorpej  2122: {
1.101     thorpej  2123:        struct pool_pagelist pq;
1.102     chs      2124:        struct pool_cache_grouplist pcgl;
1.101     thorpej  2125:
                   2126:        LIST_INIT(&pq);
1.102     chs      2127:        LIST_INIT(&pcgl);
1.101     thorpej  2128:
                   2129:        simple_lock(&pc->pc_slock);
                   2130:        simple_lock(&pc->pc_pool->pr_slock);
1.43      thorpej  2131:
1.102     chs      2132:        pool_do_cache_invalidate(pc, &pq, &pcgl);
1.43      thorpej  2133:
1.101     thorpej  2134:        simple_unlock(&pc->pc_pool->pr_slock);
                   2135:        simple_unlock(&pc->pc_slock);
1.43      thorpej  2136:
1.102     chs      2137:        pr_pagelist_free(pc->pc_pool, &pq);
                   2138:        pcg_grouplist_free(&pcgl);
1.43      thorpej  2139: }
                   2140:
                   2141: /*
                   2142:  * pool_cache_reclaim:
                   2143:  *
                   2144:  *     Reclaim a pool cache for pool_reclaim().
                   2145:  */
                   2146: static void
1.102     chs      2147: pool_cache_reclaim(struct pool_cache *pc, struct pool_pagelist *pq,
                   2148:     struct pool_cache_grouplist *pcgl)
1.43      thorpej  2149: {
1.101     thorpej  2150:
                   2151:        /*
                   2152:         * We're locking in the wrong order (normally pool_cache -> pool,
                   2153:         * but the pool is already locked when we get here), so we have
                   2154:         * to use trylock.  If we can't lock the pool_cache, it's not really
                   2155:         * a big deal here.
                   2156:         */
                   2157:        if (simple_lock_try(&pc->pc_slock) == 0)
                   2158:                return;
                   2159:
1.102     chs      2160:        pool_do_cache_invalidate(pc, pq, pcgl);
1.43      thorpej  2161:
                   2162:        simple_unlock(&pc->pc_slock);
1.3       pk       2163: }
1.66      thorpej  2164:
                   2165: /*
                   2166:  * Pool backend allocators.
                   2167:  *
                   2168:  * Each pool has a backend allocator that handles allocation, deallocation,
                   2169:  * and any additional draining that might be needed.
                   2170:  *
                   2171:  * We provide two standard allocators:
                   2172:  *
                   2173:  *     pool_allocator_kmem - the default when no allocator is specified
                   2174:  *
                   2175:  *     pool_allocator_nointr - used for pools that will not be accessed
                   2176:  *     in interrupt context.
                   2177:  */
                   2178: void   *pool_page_alloc(struct pool *, int);
                   2179: void   pool_page_free(struct pool *, void *);
                   2180:
1.112     bjh21    2181: #ifdef POOL_SUBPAGE
                   2182: struct pool_allocator pool_allocator_kmem_fullpage = {
                   2183:        pool_page_alloc, pool_page_free, 0,
                   2184: };
                   2185: #else
1.66      thorpej  2186: struct pool_allocator pool_allocator_kmem = {
                   2187:        pool_page_alloc, pool_page_free, 0,
                   2188: };
1.112     bjh21    2189: #endif
1.66      thorpej  2190:
                   2191: void   *pool_page_alloc_nointr(struct pool *, int);
                   2192: void   pool_page_free_nointr(struct pool *, void *);
                   2193:
1.112     bjh21    2194: #ifdef POOL_SUBPAGE
                   2195: struct pool_allocator pool_allocator_nointr_fullpage = {
                   2196:        pool_page_alloc_nointr, pool_page_free_nointr, 0,
                   2197: };
                   2198: #else
1.66      thorpej  2199: struct pool_allocator pool_allocator_nointr = {
                   2200:        pool_page_alloc_nointr, pool_page_free_nointr, 0,
                   2201: };
1.112     bjh21    2202: #endif
1.66      thorpej  2203:
                   2204: #ifdef POOL_SUBPAGE
                   2205: void   *pool_subpage_alloc(struct pool *, int);
                   2206: void   pool_subpage_free(struct pool *, void *);
                   2207:
1.112     bjh21    2208: struct pool_allocator pool_allocator_kmem = {
                   2209:        pool_subpage_alloc, pool_subpage_free, POOL_SUBPAGE,
                   2210: };
                   2211:
                   2212: void   *pool_subpage_alloc_nointr(struct pool *, int);
                   2213: void   pool_subpage_free_nointr(struct pool *, void *);
                   2214:
                   2215: struct pool_allocator pool_allocator_nointr = {
                   2216:        pool_subpage_alloc, pool_subpage_free, POOL_SUBPAGE,
1.66      thorpej  2217: };
                   2218: #endif /* POOL_SUBPAGE */
                   2219:
                   2220: /*
                   2221:  * We have at least three different resources for the same allocation and
                   2222:  * each resource can be depleted.  First, we have the ready elements in the
                   2223:  * pool.  Then we have the resource (typically a vm_map) for this allocator.
                   2224:  * Finally, we have physical memory.  Waiting for any of these can be
                   2225:  * unnecessary when any other is freed, but the kernel doesn't support
                   2226:  * sleeping on multiple wait channels, so we have to employ another strategy.
                   2227:  *
                   2228:  * The caller sleeps on the pool (so that it can be awakened when an item
                   2229:  * is returned to the pool), but we set PA_WANT on the allocator.  When a
                   2230:  * page is returned to the allocator and PA_WANT is set, pool_allocator_free
                   2231:  * will wake up all sleeping pools belonging to this allocator.
                   2232:  *
                   2233:  * XXX Thundering herd.
                   2234:  */
                   2235: void *
                   2236: pool_allocator_alloc(struct pool *org, int flags)
                   2237: {
                   2238:        struct pool_allocator *pa = org->pr_alloc;
                   2239:        struct pool *pp, *start;
                   2240:        int s, freed;
                   2241:        void *res;
                   2242:
1.91      yamt     2243:        LOCK_ASSERT(!simple_lock_held(&org->pr_slock));
                   2244:
1.66      thorpej  2245:        do {
                   2246:                if ((res = (*pa->pa_alloc)(org, flags)) != NULL)
                   2247:                        return (res);
1.68      thorpej  2248:                if ((flags & PR_WAITOK) == 0) {
                   2249:                        /*
                   2250:                         * We only run the drain hookhere if PR_NOWAIT.
                   2251:                         * In other cases, the hook will be run in
                   2252:                         * pool_reclaim().
                   2253:                         */
                   2254:                        if (org->pr_drain_hook != NULL) {
                   2255:                                (*org->pr_drain_hook)(org->pr_drain_hook_arg,
                   2256:                                    flags);
                   2257:                                if ((res = (*pa->pa_alloc)(org, flags)) != NULL)
                   2258:                                        return (res);
                   2259:                        }
1.66      thorpej  2260:                        break;
1.68      thorpej  2261:                }
1.66      thorpej  2262:
                   2263:                /*
1.109     christos 2264:                 * Drain all pools, that use this allocator.
                   2265:                 * We do this to reclaim VA space.
1.66      thorpej  2266:                 * pa_alloc is responsible for waiting for
                   2267:                 * physical memory.
                   2268:                 *
                   2269:                 * XXX We risk looping forever if start if someone
                   2270:                 * calls pool_destroy on "start".  But there is no
                   2271:                 * other way to have potentially sleeping pool_reclaim,
                   2272:                 * non-sleeping locks on pool_allocator, and some
                   2273:                 * stirring of drained pools in the allocator.
1.68      thorpej  2274:                 *
                   2275:                 * XXX Maybe we should use pool_head_slock for locking
                   2276:                 * the allocators?
1.66      thorpej  2277:                 */
                   2278:                freed = 0;
                   2279:
                   2280:                s = splvm();
                   2281:                simple_lock(&pa->pa_slock);
                   2282:                pp = start = TAILQ_FIRST(&pa->pa_list);
                   2283:                do {
                   2284:                        TAILQ_REMOVE(&pa->pa_list, pp, pr_alloc_list);
                   2285:                        TAILQ_INSERT_TAIL(&pa->pa_list, pp, pr_alloc_list);
1.73      thorpej  2286:                        simple_unlock(&pa->pa_slock);
1.66      thorpej  2287:                        freed = pool_reclaim(pp);
1.73      thorpej  2288:                        simple_lock(&pa->pa_slock);
1.66      thorpej  2289:                } while ((pp = TAILQ_FIRST(&pa->pa_list)) != start &&
                   2290:                         freed == 0);
                   2291:
                   2292:                if (freed == 0) {
                   2293:                        /*
                   2294:                         * We set PA_WANT here, the caller will most likely
                   2295:                         * sleep waiting for pages (if not, this won't hurt
                   2296:                         * that much), and there is no way to set this in
                   2297:                         * the caller without violating locking order.
                   2298:                         */
                   2299:                        pa->pa_flags |= PA_WANT;
                   2300:                }
                   2301:                simple_unlock(&pa->pa_slock);
                   2302:                splx(s);
                   2303:        } while (freed);
                   2304:        return (NULL);
                   2305: }
                   2306:
                   2307: void
                   2308: pool_allocator_free(struct pool *pp, void *v)
                   2309: {
                   2310:        struct pool_allocator *pa = pp->pr_alloc;
                   2311:        int s;
                   2312:
1.91      yamt     2313:        LOCK_ASSERT(!simple_lock_held(&pp->pr_slock));
                   2314:
1.66      thorpej  2315:        (*pa->pa_free)(pp, v);
                   2316:
                   2317:        s = splvm();
                   2318:        simple_lock(&pa->pa_slock);
                   2319:        if ((pa->pa_flags & PA_WANT) == 0) {
                   2320:                simple_unlock(&pa->pa_slock);
                   2321:                splx(s);
                   2322:                return;
                   2323:        }
                   2324:
                   2325:        TAILQ_FOREACH(pp, &pa->pa_list, pr_alloc_list) {
                   2326:                simple_lock(&pp->pr_slock);
                   2327:                if ((pp->pr_flags & PR_WANTED) != 0) {
                   2328:                        pp->pr_flags &= ~PR_WANTED;
                   2329:                        wakeup(pp);
                   2330:                }
1.69      thorpej  2331:                simple_unlock(&pp->pr_slock);
1.66      thorpej  2332:        }
                   2333:        pa->pa_flags &= ~PA_WANT;
                   2334:        simple_unlock(&pa->pa_slock);
                   2335:        splx(s);
                   2336: }
                   2337:
                   2338: void *
                   2339: pool_page_alloc(struct pool *pp, int flags)
                   2340: {
                   2341:        boolean_t waitok = (flags & PR_WAITOK) ? TRUE : FALSE;
                   2342:
1.100     yamt     2343:        return ((void *) uvm_km_alloc_poolpage_cache(kmem_map, waitok));
1.66      thorpej  2344: }
                   2345:
                   2346: void
                   2347: pool_page_free(struct pool *pp, void *v)
                   2348: {
                   2349:
1.98      yamt     2350:        uvm_km_free_poolpage_cache(kmem_map, (vaddr_t) v);
                   2351: }
                   2352:
                   2353: static void *
                   2354: pool_page_alloc_meta(struct pool *pp, int flags)
                   2355: {
                   2356:        boolean_t waitok = (flags & PR_WAITOK) ? TRUE : FALSE;
                   2357:
1.100     yamt     2358:        return ((void *) uvm_km_alloc_poolpage(kmem_map, waitok));
1.98      yamt     2359: }
                   2360:
                   2361: static void
                   2362: pool_page_free_meta(struct pool *pp, void *v)
                   2363: {
                   2364:
1.100     yamt     2365:        uvm_km_free_poolpage(kmem_map, (vaddr_t) v);
1.66      thorpej  2366: }
                   2367:
                   2368: #ifdef POOL_SUBPAGE
                   2369: /* Sub-page allocator, for machines with large hardware pages. */
                   2370: void *
                   2371: pool_subpage_alloc(struct pool *pp, int flags)
                   2372: {
1.93      dbj      2373:        void *v;
                   2374:        int s;
                   2375:        s = splvm();
                   2376:        v = pool_get(&psppool, flags);
                   2377:        splx(s);
                   2378:        return v;
1.66      thorpej  2379: }
                   2380:
                   2381: void
                   2382: pool_subpage_free(struct pool *pp, void *v)
                   2383: {
1.93      dbj      2384:        int s;
                   2385:        s = splvm();
1.66      thorpej  2386:        pool_put(&psppool, v);
1.93      dbj      2387:        splx(s);
1.66      thorpej  2388: }
                   2389:
                   2390: /* We don't provide a real nointr allocator.  Maybe later. */
                   2391: void *
1.112     bjh21    2392: pool_subpage_alloc_nointr(struct pool *pp, int flags)
1.66      thorpej  2393: {
                   2394:
                   2395:        return (pool_subpage_alloc(pp, flags));
                   2396: }
                   2397:
                   2398: void
1.112     bjh21    2399: pool_subpage_free_nointr(struct pool *pp, void *v)
1.66      thorpej  2400: {
                   2401:
                   2402:        pool_subpage_free(pp, v);
                   2403: }
1.112     bjh21    2404: #endif /* POOL_SUBPAGE */
1.66      thorpej  2405: void *
                   2406: pool_page_alloc_nointr(struct pool *pp, int flags)
                   2407: {
                   2408:        boolean_t waitok = (flags & PR_WAITOK) ? TRUE : FALSE;
                   2409:
1.100     yamt     2410:        return ((void *) uvm_km_alloc_poolpage_cache(kernel_map, waitok));
1.66      thorpej  2411: }
                   2412:
                   2413: void
                   2414: pool_page_free_nointr(struct pool *pp, void *v)
                   2415: {
                   2416:
1.98      yamt     2417:        uvm_km_free_poolpage_cache(kernel_map, (vaddr_t) v);
1.66      thorpej  2418: }

CVSweb <webmaster@jp.NetBSD.org>