[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 1.72

1.72    ! thorpej     1: /*     $NetBSD: subr_pool.c,v 1.71 2002/03/09 01:37:19 thorpej 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>
1.72    ! thorpej    41: __KERNEL_RCSID(0, "$NetBSD: subr_pool.c,v 1.71 2002/03/09 01:37:19 thorpej 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:  *
                     62:  * Memory is allocated in pages which are split into pieces according
                     63:  * to the pool item size. Each page is kept on a list headed by `pr_pagelist'
                     64:  * in the pool structure and the individual pool items are on a linked list
                     65:  * headed by `ph_itemlist' in each page header. The memory for building
                     66:  * the page list is either taken from the allocated pages themselves (for
                     67:  * small pool items) or taken from an internal pool of page headers (`phpool').
1.1       pk         68:  */
                     69:
1.3       pk         70: /* List of all pools */
1.5       thorpej    71: TAILQ_HEAD(,pool) pool_head = TAILQ_HEAD_INITIALIZER(pool_head);
1.3       pk         72:
                     73: /* Private pool for page header structures */
                     74: static struct pool phpool;
                     75:
1.62      bjh21      76: #ifdef POOL_SUBPAGE
                     77: /* Pool of subpages for use by normal pools. */
                     78: static struct pool psppool;
                     79: #endif
                     80:
1.3       pk         81: /* # of seconds to retain page after last use */
                     82: int pool_inactive_time = 10;
                     83:
                     84: /* Next candidate for drainage (see pool_drain()) */
1.23      thorpej    85: static struct pool     *drainpp;
                     86:
                     87: /* This spin lock protects both pool_head and drainpp. */
                     88: struct simplelock pool_head_slock = SIMPLELOCK_INITIALIZER;
1.3       pk         89:
                     90: struct pool_item_header {
                     91:        /* Page headers */
                     92:        TAILQ_ENTRY(pool_item_header)
                     93:                                ph_pagelist;    /* pool page list */
                     94:        TAILQ_HEAD(,pool_item)  ph_itemlist;    /* chunk list for this page */
                     95:        LIST_ENTRY(pool_item_header)
                     96:                                ph_hashlist;    /* Off-page page headers */
                     97:        int                     ph_nmissing;    /* # of chunks in use */
                     98:        caddr_t                 ph_page;        /* this page's address */
                     99:        struct timeval          ph_time;        /* last referenced */
                    100: };
1.61      chs       101: TAILQ_HEAD(pool_pagelist,pool_item_header);
1.3       pk        102:
1.1       pk        103: struct pool_item {
1.3       pk        104: #ifdef DIAGNOSTIC
                    105:        int pi_magic;
1.33      chs       106: #endif
1.25      thorpej   107: #define        PI_MAGIC 0xdeadbeef
1.3       pk        108:        /* Other entries use only this list entry */
                    109:        TAILQ_ENTRY(pool_item)  pi_list;
                    110: };
                    111:
1.25      thorpej   112: #define        PR_HASH_INDEX(pp,addr) \
1.66      thorpej   113:        (((u_long)(addr) >> (pp)->pr_alloc->pa_pageshift) & \
                    114:         (PR_HASHTABSIZE - 1))
1.3       pk        115:
1.53      thorpej   116: #define        POOL_NEEDS_CATCHUP(pp)                                          \
                    117:        ((pp)->pr_nitems < (pp)->pr_minitems)
                    118:
1.43      thorpej   119: /*
                    120:  * Pool cache management.
                    121:  *
                    122:  * Pool caches provide a way for constructed objects to be cached by the
                    123:  * pool subsystem.  This can lead to performance improvements by avoiding
                    124:  * needless object construction/destruction; it is deferred until absolutely
                    125:  * necessary.
                    126:  *
                    127:  * Caches are grouped into cache groups.  Each cache group references
                    128:  * up to 16 constructed objects.  When a cache allocates an object
                    129:  * from the pool, it calls the object's constructor and places it into
                    130:  * a cache group.  When a cache group frees an object back to the pool,
                    131:  * it first calls the object's destructor.  This allows the object to
                    132:  * persist in constructed form while freed to the cache.
                    133:  *
                    134:  * Multiple caches may exist for each pool.  This allows a single
                    135:  * object type to have multiple constructed forms.  The pool references
                    136:  * each cache, so that when a pool is drained by the pagedaemon, it can
                    137:  * drain each individual cache as well.  Each time a cache is drained,
                    138:  * the most idle cache group is freed to the pool in its entirety.
                    139:  *
                    140:  * Pool caches are layed on top of pools.  By layering them, we can avoid
                    141:  * the complexity of cache management for pools which would not benefit
                    142:  * from it.
                    143:  */
                    144:
                    145: /* The cache group pool. */
                    146: static struct pool pcgpool;
                    147:
                    148: /* The pool cache group. */
                    149: #define        PCG_NOBJECTS            16
                    150: struct pool_cache_group {
                    151:        TAILQ_ENTRY(pool_cache_group)
                    152:                pcg_list;       /* link in the pool cache's group list */
                    153:        u_int   pcg_avail;      /* # available objects */
                    154:                                /* pointers to the objects */
                    155:        void    *pcg_objects[PCG_NOBJECTS];
                    156: };
1.3       pk        157:
1.43      thorpej   158: static void    pool_cache_reclaim(struct pool_cache *);
1.3       pk        159:
1.42      thorpej   160: static int     pool_catchup(struct pool *);
1.55      thorpej   161: static void    pool_prime_page(struct pool *, caddr_t,
                    162:                    struct pool_item_header *);
1.66      thorpej   163:
                    164: void           *pool_allocator_alloc(struct pool *, int);
                    165: void           pool_allocator_free(struct pool *, void *);
1.3       pk        166:
1.42      thorpej   167: static void pool_print1(struct pool *, const char *,
                    168:        void (*)(const char *, ...));
1.3       pk        169:
                    170: /*
1.52      thorpej   171:  * Pool log entry. An array of these is allocated in pool_init().
1.3       pk        172:  */
                    173: struct pool_log {
                    174:        const char      *pl_file;
                    175:        long            pl_line;
                    176:        int             pl_action;
1.25      thorpej   177: #define        PRLOG_GET       1
                    178: #define        PRLOG_PUT       2
1.3       pk        179:        void            *pl_addr;
1.1       pk        180: };
                    181:
1.3       pk        182: /* Number of entries in pool log buffers */
1.17      thorpej   183: #ifndef POOL_LOGSIZE
                    184: #define        POOL_LOGSIZE    10
                    185: #endif
                    186:
                    187: int pool_logsize = POOL_LOGSIZE;
1.1       pk        188:
1.59      thorpej   189: #ifdef POOL_DIAGNOSTIC
1.42      thorpej   190: static __inline void
                    191: pr_log(struct pool *pp, void *v, int action, const char *file, long line)
1.3       pk        192: {
                    193:        int n = pp->pr_curlogentry;
                    194:        struct pool_log *pl;
                    195:
1.20      thorpej   196:        if ((pp->pr_roflags & PR_LOGGING) == 0)
1.3       pk        197:                return;
                    198:
                    199:        /*
                    200:         * Fill in the current entry. Wrap around and overwrite
                    201:         * the oldest entry if necessary.
                    202:         */
                    203:        pl = &pp->pr_log[n];
                    204:        pl->pl_file = file;
                    205:        pl->pl_line = line;
                    206:        pl->pl_action = action;
                    207:        pl->pl_addr = v;
                    208:        if (++n >= pp->pr_logsize)
                    209:                n = 0;
                    210:        pp->pr_curlogentry = n;
                    211: }
                    212:
                    213: static void
1.42      thorpej   214: pr_printlog(struct pool *pp, struct pool_item *pi,
                    215:     void (*pr)(const char *, ...))
1.3       pk        216: {
                    217:        int i = pp->pr_logsize;
                    218:        int n = pp->pr_curlogentry;
                    219:
1.20      thorpej   220:        if ((pp->pr_roflags & PR_LOGGING) == 0)
1.3       pk        221:                return;
                    222:
                    223:        /*
                    224:         * Print all entries in this pool's log.
                    225:         */
                    226:        while (i-- > 0) {
                    227:                struct pool_log *pl = &pp->pr_log[n];
                    228:                if (pl->pl_action != 0) {
1.25      thorpej   229:                        if (pi == NULL || pi == pl->pl_addr) {
                    230:                                (*pr)("\tlog entry %d:\n", i);
                    231:                                (*pr)("\t\taction = %s, addr = %p\n",
                    232:                                    pl->pl_action == PRLOG_GET ? "get" : "put",
                    233:                                    pl->pl_addr);
                    234:                                (*pr)("\t\tfile: %s at line %lu\n",
                    235:                                    pl->pl_file, pl->pl_line);
                    236:                        }
1.3       pk        237:                }
                    238:                if (++n >= pp->pr_logsize)
                    239:                        n = 0;
                    240:        }
                    241: }
1.25      thorpej   242:
1.42      thorpej   243: static __inline void
                    244: pr_enter(struct pool *pp, const char *file, long line)
1.25      thorpej   245: {
                    246:
1.34      thorpej   247:        if (__predict_false(pp->pr_entered_file != NULL)) {
1.25      thorpej   248:                printf("pool %s: reentrancy at file %s line %ld\n",
                    249:                    pp->pr_wchan, file, line);
                    250:                printf("         previous entry at file %s line %ld\n",
                    251:                    pp->pr_entered_file, pp->pr_entered_line);
                    252:                panic("pr_enter");
                    253:        }
                    254:
                    255:        pp->pr_entered_file = file;
                    256:        pp->pr_entered_line = line;
                    257: }
                    258:
1.42      thorpej   259: static __inline void
                    260: pr_leave(struct pool *pp)
1.25      thorpej   261: {
                    262:
1.34      thorpej   263:        if (__predict_false(pp->pr_entered_file == NULL)) {
1.25      thorpej   264:                printf("pool %s not entered?\n", pp->pr_wchan);
                    265:                panic("pr_leave");
                    266:        }
                    267:
                    268:        pp->pr_entered_file = NULL;
                    269:        pp->pr_entered_line = 0;
                    270: }
                    271:
1.42      thorpej   272: static __inline void
                    273: pr_enter_check(struct pool *pp, void (*pr)(const char *, ...))
1.25      thorpej   274: {
                    275:
                    276:        if (pp->pr_entered_file != NULL)
                    277:                (*pr)("\n\tcurrently entered from file %s line %ld\n",
                    278:                    pp->pr_entered_file, pp->pr_entered_line);
                    279: }
1.3       pk        280: #else
1.25      thorpej   281: #define        pr_log(pp, v, action, file, line)
                    282: #define        pr_printlog(pp, pi, pr)
                    283: #define        pr_enter(pp, file, line)
                    284: #define        pr_leave(pp)
                    285: #define        pr_enter_check(pp, pr)
1.59      thorpej   286: #endif /* POOL_DIAGNOSTIC */
1.3       pk        287:
                    288: /*
                    289:  * Return the pool page header based on page address.
                    290:  */
1.42      thorpej   291: static __inline struct pool_item_header *
                    292: pr_find_pagehead(struct pool *pp, caddr_t page)
1.3       pk        293: {
                    294:        struct pool_item_header *ph;
                    295:
1.20      thorpej   296:        if ((pp->pr_roflags & PR_PHINPAGE) != 0)
1.3       pk        297:                return ((struct pool_item_header *)(page + pp->pr_phoffset));
                    298:
                    299:        for (ph = LIST_FIRST(&pp->pr_hashtab[PR_HASH_INDEX(pp, page)]);
                    300:             ph != NULL;
                    301:             ph = LIST_NEXT(ph, ph_hashlist)) {
                    302:                if (ph->ph_page == page)
                    303:                        return (ph);
                    304:        }
                    305:        return (NULL);
                    306: }
                    307:
                    308: /*
                    309:  * Remove a page from the pool.
                    310:  */
1.42      thorpej   311: static __inline void
1.61      chs       312: pr_rmpage(struct pool *pp, struct pool_item_header *ph,
                    313:      struct pool_pagelist *pq)
1.3       pk        314: {
1.61      chs       315:        int s;
1.3       pk        316:
                    317:        /*
1.7       thorpej   318:         * If the page was idle, decrement the idle page count.
1.3       pk        319:         */
1.6       thorpej   320:        if (ph->ph_nmissing == 0) {
                    321: #ifdef DIAGNOSTIC
                    322:                if (pp->pr_nidle == 0)
                    323:                        panic("pr_rmpage: nidle inconsistent");
1.20      thorpej   324:                if (pp->pr_nitems < pp->pr_itemsperpage)
                    325:                        panic("pr_rmpage: nitems inconsistent");
1.6       thorpej   326: #endif
                    327:                pp->pr_nidle--;
                    328:        }
1.7       thorpej   329:
1.20      thorpej   330:        pp->pr_nitems -= pp->pr_itemsperpage;
                    331:
1.7       thorpej   332:        /*
1.61      chs       333:         * Unlink a page from the pool and release it (or queue it for release).
1.7       thorpej   334:         */
                    335:        TAILQ_REMOVE(&pp->pr_pagelist, ph, ph_pagelist);
1.61      chs       336:        if (pq) {
                    337:                TAILQ_INSERT_HEAD(pq, ph, ph_pagelist);
                    338:        } else {
1.66      thorpej   339:                pool_allocator_free(pp, ph->ph_page);
1.61      chs       340:                if ((pp->pr_roflags & PR_PHINPAGE) == 0) {
                    341:                        LIST_REMOVE(ph, ph_hashlist);
                    342:                        s = splhigh();
                    343:                        pool_put(&phpool, ph);
                    344:                        splx(s);
                    345:                }
                    346:        }
1.7       thorpej   347:        pp->pr_npages--;
                    348:        pp->pr_npagefree++;
1.6       thorpej   349:
1.3       pk        350:        if (pp->pr_curpage == ph) {
                    351:                /*
                    352:                 * Find a new non-empty page header, if any.
                    353:                 * Start search from the page head, to increase the
                    354:                 * chance for "high water" pages to be freed.
                    355:                 */
1.61      chs       356:                TAILQ_FOREACH(ph, &pp->pr_pagelist, ph_pagelist)
1.3       pk        357:                        if (TAILQ_FIRST(&ph->ph_itemlist) != NULL)
                    358:                                break;
                    359:
                    360:                pp->pr_curpage = ph;
1.21      thorpej   361:        }
1.3       pk        362: }
                    363:
                    364: /*
                    365:  * Initialize the given pool resource structure.
                    366:  *
                    367:  * We export this routine to allow other kernel parts to declare
                    368:  * static pools that must be initialized before malloc() is available.
                    369:  */
                    370: void
1.42      thorpej   371: pool_init(struct pool *pp, size_t size, u_int align, u_int ioff, int flags,
1.66      thorpej   372:     const char *wchan, struct pool_allocator *palloc)
1.3       pk        373: {
1.16      briggs    374:        int off, slack, i;
1.3       pk        375:
1.25      thorpej   376: #ifdef POOL_DIAGNOSTIC
                    377:        /*
                    378:         * Always log if POOL_DIAGNOSTIC is defined.
                    379:         */
                    380:        if (pool_logsize != 0)
                    381:                flags |= PR_LOGGING;
                    382: #endif
                    383:
1.66      thorpej   384: #ifdef POOL_SUBPAGE
                    385:        /*
                    386:         * XXX We don't provide a real `nointr' back-end
                    387:         * yet; all sub-pages come from a kmem back-end.
                    388:         * maybe some day...
                    389:         */
                    390:        if (palloc == NULL) {
                    391:                extern struct pool_allocator pool_allocator_kmem_subpage;
                    392:                palloc = &pool_allocator_kmem_subpage;
                    393:        }
1.3       pk        394:        /*
1.66      thorpej   395:         * We'll assume any user-specified back-end allocator
                    396:         * will deal with sub-pages, or simply don't care.
1.3       pk        397:         */
1.66      thorpej   398: #else
                    399:        if (palloc == NULL)
                    400:                palloc = &pool_allocator_kmem;
                    401: #endif /* POOL_SUBPAGE */
                    402:        if ((palloc->pa_flags & PA_INITIALIZED) == 0) {
                    403:                if (palloc->pa_pagesz == 0) {
1.62      bjh21     404: #ifdef POOL_SUBPAGE
1.66      thorpej   405:                        if (palloc == &pool_allocator_kmem)
                    406:                                palloc->pa_pagesz = PAGE_SIZE;
                    407:                        else
                    408:                                palloc->pa_pagesz = POOL_SUBPAGE;
1.62      bjh21     409: #else
1.66      thorpej   410:                        palloc->pa_pagesz = PAGE_SIZE;
                    411: #endif /* POOL_SUBPAGE */
                    412:                }
                    413:
                    414:                TAILQ_INIT(&palloc->pa_list);
                    415:
                    416:                simple_lock_init(&palloc->pa_slock);
                    417:                palloc->pa_pagemask = ~(palloc->pa_pagesz - 1);
                    418:                palloc->pa_pageshift = ffs(palloc->pa_pagesz) - 1;
                    419:                palloc->pa_flags |= PA_INITIALIZED;
1.4       thorpej   420:        }
1.3       pk        421:
                    422:        if (align == 0)
                    423:                align = ALIGN(1);
1.14      thorpej   424:
                    425:        if (size < sizeof(struct pool_item))
                    426:                size = sizeof(struct pool_item);
1.3       pk        427:
1.35      pk        428:        size = ALIGN(size);
1.66      thorpej   429: #ifdef DIAGNOSTIC
                    430:        if (size > palloc->pa_pagesz)
1.35      pk        431:                panic("pool_init: pool item size (%lu) too large",
                    432:                      (u_long)size);
1.66      thorpej   433: #endif
1.35      pk        434:
1.3       pk        435:        /*
                    436:         * Initialize the pool structure.
                    437:         */
                    438:        TAILQ_INIT(&pp->pr_pagelist);
1.43      thorpej   439:        TAILQ_INIT(&pp->pr_cachelist);
1.3       pk        440:        pp->pr_curpage = NULL;
                    441:        pp->pr_npages = 0;
                    442:        pp->pr_minitems = 0;
                    443:        pp->pr_minpages = 0;
                    444:        pp->pr_maxpages = UINT_MAX;
1.20      thorpej   445:        pp->pr_roflags = flags;
                    446:        pp->pr_flags = 0;
1.35      pk        447:        pp->pr_size = size;
1.3       pk        448:        pp->pr_align = align;
                    449:        pp->pr_wchan = wchan;
1.66      thorpej   450:        pp->pr_alloc = palloc;
1.20      thorpej   451:        pp->pr_nitems = 0;
                    452:        pp->pr_nout = 0;
                    453:        pp->pr_hardlimit = UINT_MAX;
                    454:        pp->pr_hardlimit_warning = NULL;
1.31      thorpej   455:        pp->pr_hardlimit_ratecap.tv_sec = 0;
                    456:        pp->pr_hardlimit_ratecap.tv_usec = 0;
                    457:        pp->pr_hardlimit_warning_last.tv_sec = 0;
                    458:        pp->pr_hardlimit_warning_last.tv_usec = 0;
1.68      thorpej   459:        pp->pr_drain_hook = NULL;
                    460:        pp->pr_drain_hook_arg = NULL;
1.3       pk        461:
                    462:        /*
                    463:         * Decide whether to put the page header off page to avoid
                    464:         * wasting too large a part of the page. Off-page page headers
                    465:         * go on a hash table, so we can match a returned item
                    466:         * with its header based on the page address.
                    467:         * We use 1/16 of the page size as the threshold (XXX: tune)
                    468:         */
1.66      thorpej   469:        if (pp->pr_size < palloc->pa_pagesz/16) {
1.3       pk        470:                /* Use the end of the page for the page header */
1.20      thorpej   471:                pp->pr_roflags |= PR_PHINPAGE;
1.66      thorpej   472:                pp->pr_phoffset = off = palloc->pa_pagesz -
                    473:                    ALIGN(sizeof(struct pool_item_header));
1.2       pk        474:        } else {
1.3       pk        475:                /* The page header will be taken from our page header pool */
                    476:                pp->pr_phoffset = 0;
1.66      thorpej   477:                off = palloc->pa_pagesz;
1.16      briggs    478:                for (i = 0; i < PR_HASHTABSIZE; i++) {
                    479:                        LIST_INIT(&pp->pr_hashtab[i]);
                    480:                }
1.2       pk        481:        }
1.1       pk        482:
1.3       pk        483:        /*
                    484:         * Alignment is to take place at `ioff' within the item. This means
                    485:         * we must reserve up to `align - 1' bytes on the page to allow
                    486:         * appropriate positioning of each item.
                    487:         *
                    488:         * Silently enforce `0 <= ioff < align'.
                    489:         */
                    490:        pp->pr_itemoffset = ioff = ioff % align;
                    491:        pp->pr_itemsperpage = (off - ((align - ioff) % align)) / pp->pr_size;
1.43      thorpej   492:        KASSERT(pp->pr_itemsperpage != 0);
1.3       pk        493:
                    494:        /*
                    495:         * Use the slack between the chunks and the page header
                    496:         * for "cache coloring".
                    497:         */
                    498:        slack = off - pp->pr_itemsperpage * pp->pr_size;
                    499:        pp->pr_maxcolor = (slack / align) * align;
                    500:        pp->pr_curcolor = 0;
                    501:
                    502:        pp->pr_nget = 0;
                    503:        pp->pr_nfail = 0;
                    504:        pp->pr_nput = 0;
                    505:        pp->pr_npagealloc = 0;
                    506:        pp->pr_npagefree = 0;
1.1       pk        507:        pp->pr_hiwat = 0;
1.8       thorpej   508:        pp->pr_nidle = 0;
1.3       pk        509:
1.59      thorpej   510: #ifdef POOL_DIAGNOSTIC
1.25      thorpej   511:        if (flags & PR_LOGGING) {
                    512:                if (kmem_map == NULL ||
                    513:                    (pp->pr_log = malloc(pool_logsize * sizeof(struct pool_log),
                    514:                     M_TEMP, M_NOWAIT)) == NULL)
1.20      thorpej   515:                        pp->pr_roflags &= ~PR_LOGGING;
1.3       pk        516:                pp->pr_curlogentry = 0;
                    517:                pp->pr_logsize = pool_logsize;
                    518:        }
1.59      thorpej   519: #endif
1.25      thorpej   520:
                    521:        pp->pr_entered_file = NULL;
                    522:        pp->pr_entered_line = 0;
1.3       pk        523:
1.21      thorpej   524:        simple_lock_init(&pp->pr_slock);
1.1       pk        525:
1.3       pk        526:        /*
1.43      thorpej   527:         * Initialize private page header pool and cache magazine pool if we
                    528:         * haven't done so yet.
1.23      thorpej   529:         * XXX LOCKING.
1.3       pk        530:         */
                    531:        if (phpool.pr_size == 0) {
1.62      bjh21     532: #ifdef POOL_SUBPAGE
                    533:                pool_init(&phpool, sizeof(struct pool_item_header), 0, 0, 0,
1.66      thorpej   534:                    "phpool", &pool_allocator_kmem);
1.62      bjh21     535:                pool_init(&psppool, POOL_SUBPAGE, POOL_SUBPAGE, 0,
1.66      thorpej   536:                    PR_RECURSIVE, "psppool", &pool_allocator_kmem);
1.62      bjh21     537: #else
1.3       pk        538:                pool_init(&phpool, sizeof(struct pool_item_header), 0, 0,
1.66      thorpej   539:                    0, "phpool", NULL);
1.62      bjh21     540: #endif
1.43      thorpej   541:                pool_init(&pcgpool, sizeof(struct pool_cache_group), 0, 0,
1.66      thorpej   542:                    0, "pcgpool", NULL);
1.1       pk        543:        }
                    544:
1.23      thorpej   545:        /* Insert into the list of all pools. */
                    546:        simple_lock(&pool_head_slock);
                    547:        TAILQ_INSERT_TAIL(&pool_head, pp, pr_poollist);
                    548:        simple_unlock(&pool_head_slock);
1.66      thorpej   549:
                    550:        /* Insert this into the list of pools using this allocator. */
                    551:        simple_lock(&palloc->pa_slock);
                    552:        TAILQ_INSERT_TAIL(&palloc->pa_list, pp, pr_alloc_list);
                    553:        simple_unlock(&palloc->pa_slock);
1.1       pk        554: }
                    555:
                    556: /*
                    557:  * De-commision a pool resource.
                    558:  */
                    559: void
1.42      thorpej   560: pool_destroy(struct pool *pp)
1.1       pk        561: {
1.3       pk        562:        struct pool_item_header *ph;
1.43      thorpej   563:        struct pool_cache *pc;
                    564:
1.66      thorpej   565:        /* Locking order: pool_allocator -> pool */
                    566:        simple_lock(&pp->pr_alloc->pa_slock);
                    567:        TAILQ_REMOVE(&pp->pr_alloc->pa_list, pp, pr_alloc_list);
                    568:        simple_unlock(&pp->pr_alloc->pa_slock);
                    569:
1.43      thorpej   570:        /* Destroy all caches for this pool. */
                    571:        while ((pc = TAILQ_FIRST(&pp->pr_cachelist)) != NULL)
                    572:                pool_cache_destroy(pc);
1.3       pk        573:
                    574: #ifdef DIAGNOSTIC
1.20      thorpej   575:        if (pp->pr_nout != 0) {
1.25      thorpej   576:                pr_printlog(pp, NULL, printf);
1.20      thorpej   577:                panic("pool_destroy: pool busy: still out: %u\n",
                    578:                    pp->pr_nout);
1.3       pk        579:        }
                    580: #endif
1.1       pk        581:
1.3       pk        582:        /* Remove all pages */
1.70      thorpej   583:        while ((ph = TAILQ_FIRST(&pp->pr_pagelist)) != NULL)
                    584:                pr_rmpage(pp, ph, NULL);
1.3       pk        585:
                    586:        /* Remove from global pool list */
1.23      thorpej   587:        simple_lock(&pool_head_slock);
1.3       pk        588:        TAILQ_REMOVE(&pool_head, pp, pr_poollist);
1.61      chs       589:        if (drainpp == pp) {
                    590:                drainpp = NULL;
                    591:        }
1.23      thorpej   592:        simple_unlock(&pool_head_slock);
1.3       pk        593:
1.59      thorpej   594: #ifdef POOL_DIAGNOSTIC
1.20      thorpej   595:        if ((pp->pr_roflags & PR_LOGGING) != 0)
1.3       pk        596:                free(pp->pr_log, M_TEMP);
1.59      thorpej   597: #endif
1.1       pk        598: }
                    599:
1.68      thorpej   600: void
                    601: pool_set_drain_hook(struct pool *pp, void (*fn)(void *, int), void *arg)
                    602: {
                    603:
                    604:        /* XXX no locking -- must be used just after pool_init() */
                    605: #ifdef DIAGNOSTIC
                    606:        if (pp->pr_drain_hook != NULL)
                    607:                panic("pool_set_drain_hook(%s): already set", pp->pr_wchan);
                    608: #endif
                    609:        pp->pr_drain_hook = fn;
                    610:        pp->pr_drain_hook_arg = arg;
                    611: }
                    612:
1.55      thorpej   613: static __inline struct pool_item_header *
                    614: pool_alloc_item_header(struct pool *pp, caddr_t storage, int flags)
                    615: {
                    616:        struct pool_item_header *ph;
                    617:        int s;
                    618:
                    619:        LOCK_ASSERT(simple_lock_held(&pp->pr_slock) == 0);
                    620:
                    621:        if ((pp->pr_roflags & PR_PHINPAGE) != 0)
                    622:                ph = (struct pool_item_header *) (storage + pp->pr_phoffset);
                    623:        else {
                    624:                s = splhigh();
                    625:                ph = pool_get(&phpool, flags);
                    626:                splx(s);
                    627:        }
                    628:
                    629:        return (ph);
                    630: }
1.1       pk        631:
                    632: /*
1.3       pk        633:  * Grab an item from the pool; must be called at appropriate spl level
1.1       pk        634:  */
1.3       pk        635: void *
1.59      thorpej   636: #ifdef POOL_DIAGNOSTIC
1.42      thorpej   637: _pool_get(struct pool *pp, int flags, const char *file, long line)
1.56      sommerfe  638: #else
                    639: pool_get(struct pool *pp, int flags)
                    640: #endif
1.1       pk        641: {
                    642:        struct pool_item *pi;
1.3       pk        643:        struct pool_item_header *ph;
1.55      thorpej   644:        void *v;
1.1       pk        645:
1.2       pk        646: #ifdef DIAGNOSTIC
1.37      sommerfe  647:        if (__predict_false(curproc == NULL && doing_shutdown == 0 &&
                    648:                            (flags & PR_WAITOK) != 0))
1.3       pk        649:                panic("pool_get: must have NOWAIT");
1.58      thorpej   650:
                    651: #ifdef LOCKDEBUG
                    652:        if (flags & PR_WAITOK)
                    653:                simple_lock_only_held(NULL, "pool_get(PR_WAITOK)");
1.56      sommerfe  654: #endif
1.58      thorpej   655: #endif /* DIAGNOSTIC */
1.1       pk        656:
1.21      thorpej   657:        simple_lock(&pp->pr_slock);
1.25      thorpej   658:        pr_enter(pp, file, line);
1.20      thorpej   659:
                    660:  startover:
                    661:        /*
                    662:         * Check to see if we've reached the hard limit.  If we have,
                    663:         * and we can wait, then wait until an item has been returned to
                    664:         * the pool.
                    665:         */
                    666: #ifdef DIAGNOSTIC
1.34      thorpej   667:        if (__predict_false(pp->pr_nout > pp->pr_hardlimit)) {
1.25      thorpej   668:                pr_leave(pp);
1.21      thorpej   669:                simple_unlock(&pp->pr_slock);
1.20      thorpej   670:                panic("pool_get: %s: crossed hard limit", pp->pr_wchan);
                    671:        }
                    672: #endif
1.34      thorpej   673:        if (__predict_false(pp->pr_nout == pp->pr_hardlimit)) {
1.68      thorpej   674:                if (pp->pr_drain_hook != NULL) {
                    675:                        /*
                    676:                         * Since the drain hook is going to free things
                    677:                         * back to the pool, unlock, call the hook, re-lock,
                    678:                         * and check the hardlimit condition again.
                    679:                         */
                    680:                        pr_leave(pp);
                    681:                        simple_unlock(&pp->pr_slock);
                    682:                        (*pp->pr_drain_hook)(pp->pr_drain_hook_arg, flags);
                    683:                        simple_lock(&pp->pr_slock);
                    684:                        pr_enter(pp, file, line);
                    685:                        if (pp->pr_nout < pp->pr_hardlimit)
                    686:                                goto startover;
                    687:                }
                    688:
1.29      sommerfe  689:                if ((flags & PR_WAITOK) && !(flags & PR_LIMITFAIL)) {
1.20      thorpej   690:                        /*
                    691:                         * XXX: A warning isn't logged in this case.  Should
                    692:                         * it be?
                    693:                         */
                    694:                        pp->pr_flags |= PR_WANTED;
1.25      thorpej   695:                        pr_leave(pp);
1.40      sommerfe  696:                        ltsleep(pp, PSWP, pp->pr_wchan, 0, &pp->pr_slock);
1.25      thorpej   697:                        pr_enter(pp, file, line);
1.20      thorpej   698:                        goto startover;
                    699:                }
1.31      thorpej   700:
                    701:                /*
                    702:                 * Log a message that the hard limit has been hit.
                    703:                 */
                    704:                if (pp->pr_hardlimit_warning != NULL &&
                    705:                    ratecheck(&pp->pr_hardlimit_warning_last,
                    706:                              &pp->pr_hardlimit_ratecap))
                    707:                        log(LOG_ERR, "%s\n", pp->pr_hardlimit_warning);
1.21      thorpej   708:
                    709:                pp->pr_nfail++;
                    710:
1.25      thorpej   711:                pr_leave(pp);
1.21      thorpej   712:                simple_unlock(&pp->pr_slock);
1.20      thorpej   713:                return (NULL);
                    714:        }
                    715:
1.3       pk        716:        /*
                    717:         * The convention we use is that if `curpage' is not NULL, then
                    718:         * it points at a non-empty bucket. In particular, `curpage'
                    719:         * never points at a page header which has PR_PHINPAGE set and
                    720:         * has no items in its bucket.
                    721:         */
1.20      thorpej   722:        if ((ph = pp->pr_curpage) == NULL) {
                    723: #ifdef DIAGNOSTIC
                    724:                if (pp->pr_nitems != 0) {
1.21      thorpej   725:                        simple_unlock(&pp->pr_slock);
1.20      thorpej   726:                        printf("pool_get: %s: curpage NULL, nitems %u\n",
                    727:                            pp->pr_wchan, pp->pr_nitems);
                    728:                        panic("pool_get: nitems inconsistent\n");
                    729:                }
                    730: #endif
                    731:
1.21      thorpej   732:                /*
                    733:                 * Call the back-end page allocator for more memory.
                    734:                 * Release the pool lock, as the back-end page allocator
                    735:                 * may block.
                    736:                 */
1.25      thorpej   737:                pr_leave(pp);
1.21      thorpej   738:                simple_unlock(&pp->pr_slock);
1.66      thorpej   739:                v = pool_allocator_alloc(pp, flags);
1.55      thorpej   740:                if (__predict_true(v != NULL))
                    741:                        ph = pool_alloc_item_header(pp, v, flags);
1.21      thorpej   742:                simple_lock(&pp->pr_slock);
1.25      thorpej   743:                pr_enter(pp, file, line);
1.15      pk        744:
1.55      thorpej   745:                if (__predict_false(v == NULL || ph == NULL)) {
                    746:                        if (v != NULL)
1.66      thorpej   747:                                pool_allocator_free(pp, v);
1.55      thorpej   748:
1.21      thorpej   749:                        /*
1.55      thorpej   750:                         * We were unable to allocate a page or item
                    751:                         * header, but we released the lock during
                    752:                         * allocation, so perhaps items were freed
                    753:                         * back to the pool.  Check for this case.
1.21      thorpej   754:                         */
                    755:                        if (pp->pr_curpage != NULL)
                    756:                                goto startover;
1.15      pk        757:
1.3       pk        758:                        if ((flags & PR_WAITOK) == 0) {
                    759:                                pp->pr_nfail++;
1.25      thorpej   760:                                pr_leave(pp);
1.21      thorpej   761:                                simple_unlock(&pp->pr_slock);
1.1       pk        762:                                return (NULL);
1.3       pk        763:                        }
                    764:
1.15      pk        765:                        /*
                    766:                         * Wait for items to be returned to this pool.
1.21      thorpej   767:                         *
1.20      thorpej   768:                         * XXX: maybe we should wake up once a second and
                    769:                         * try again?
1.15      pk        770:                         */
1.1       pk        771:                        pp->pr_flags |= PR_WANTED;
1.66      thorpej   772:                        /* PA_WANTED is already set on the allocator. */
1.25      thorpej   773:                        pr_leave(pp);
1.40      sommerfe  774:                        ltsleep(pp, PSWP, pp->pr_wchan, 0, &pp->pr_slock);
1.25      thorpej   775:                        pr_enter(pp, file, line);
1.20      thorpej   776:                        goto startover;
1.1       pk        777:                }
1.3       pk        778:
1.15      pk        779:                /* We have more memory; add it to the pool */
1.55      thorpej   780:                pool_prime_page(pp, v, ph);
1.15      pk        781:                pp->pr_npagealloc++;
                    782:
1.20      thorpej   783:                /* Start the allocation process over. */
                    784:                goto startover;
1.3       pk        785:        }
                    786:
1.34      thorpej   787:        if (__predict_false((v = pi = TAILQ_FIRST(&ph->ph_itemlist)) == NULL)) {
1.25      thorpej   788:                pr_leave(pp);
1.21      thorpej   789:                simple_unlock(&pp->pr_slock);
1.3       pk        790:                panic("pool_get: %s: page empty", pp->pr_wchan);
1.21      thorpej   791:        }
1.20      thorpej   792: #ifdef DIAGNOSTIC
1.34      thorpej   793:        if (__predict_false(pp->pr_nitems == 0)) {
1.25      thorpej   794:                pr_leave(pp);
1.21      thorpej   795:                simple_unlock(&pp->pr_slock);
1.20      thorpej   796:                printf("pool_get: %s: items on itemlist, nitems %u\n",
                    797:                    pp->pr_wchan, pp->pr_nitems);
                    798:                panic("pool_get: nitems inconsistent\n");
                    799:        }
1.65      enami     800: #endif
1.56      sommerfe  801:
1.65      enami     802: #ifdef POOL_DIAGNOSTIC
1.3       pk        803:        pr_log(pp, v, PRLOG_GET, file, line);
1.65      enami     804: #endif
1.3       pk        805:
1.65      enami     806: #ifdef DIAGNOSTIC
1.34      thorpej   807:        if (__predict_false(pi->pi_magic != PI_MAGIC)) {
1.25      thorpej   808:                pr_printlog(pp, pi, printf);
1.3       pk        809:                panic("pool_get(%s): free list modified: magic=%x; page %p;"
                    810:                       " item addr %p\n",
                    811:                        pp->pr_wchan, pi->pi_magic, ph->ph_page, pi);
                    812:        }
                    813: #endif
                    814:
                    815:        /*
                    816:         * Remove from item list.
                    817:         */
                    818:        TAILQ_REMOVE(&ph->ph_itemlist, pi, pi_list);
1.20      thorpej   819:        pp->pr_nitems--;
                    820:        pp->pr_nout++;
1.6       thorpej   821:        if (ph->ph_nmissing == 0) {
                    822: #ifdef DIAGNOSTIC
1.34      thorpej   823:                if (__predict_false(pp->pr_nidle == 0))
1.6       thorpej   824:                        panic("pool_get: nidle inconsistent");
                    825: #endif
                    826:                pp->pr_nidle--;
                    827:        }
1.3       pk        828:        ph->ph_nmissing++;
                    829:        if (TAILQ_FIRST(&ph->ph_itemlist) == NULL) {
1.21      thorpej   830: #ifdef DIAGNOSTIC
1.34      thorpej   831:                if (__predict_false(ph->ph_nmissing != pp->pr_itemsperpage)) {
1.25      thorpej   832:                        pr_leave(pp);
1.21      thorpej   833:                        simple_unlock(&pp->pr_slock);
                    834:                        panic("pool_get: %s: nmissing inconsistent",
                    835:                            pp->pr_wchan);
                    836:                }
                    837: #endif
1.3       pk        838:                /*
                    839:                 * Find a new non-empty page header, if any.
                    840:                 * Start search from the page head, to increase
                    841:                 * the chance for "high water" pages to be freed.
                    842:                 *
1.21      thorpej   843:                 * Migrate empty pages to the end of the list.  This
                    844:                 * will speed the update of curpage as pages become
                    845:                 * idle.  Empty pages intermingled with idle pages
                    846:                 * is no big deal.  As soon as a page becomes un-empty,
                    847:                 * it will move back to the head of the list.
1.3       pk        848:                 */
                    849:                TAILQ_REMOVE(&pp->pr_pagelist, ph, ph_pagelist);
1.21      thorpej   850:                TAILQ_INSERT_TAIL(&pp->pr_pagelist, ph, ph_pagelist);
1.61      chs       851:                TAILQ_FOREACH(ph, &pp->pr_pagelist, ph_pagelist)
1.3       pk        852:                        if (TAILQ_FIRST(&ph->ph_itemlist) != NULL)
                    853:                                break;
                    854:
                    855:                pp->pr_curpage = ph;
1.1       pk        856:        }
1.3       pk        857:
                    858:        pp->pr_nget++;
1.20      thorpej   859:
                    860:        /*
                    861:         * If we have a low water mark and we are now below that low
                    862:         * water mark, add more items to the pool.
                    863:         */
1.53      thorpej   864:        if (POOL_NEEDS_CATCHUP(pp) && pool_catchup(pp) != 0) {
1.20      thorpej   865:                /*
                    866:                 * XXX: Should we log a warning?  Should we set up a timeout
                    867:                 * to try again in a second or so?  The latter could break
                    868:                 * a caller's assumptions about interrupt protection, etc.
                    869:                 */
                    870:        }
                    871:
1.25      thorpej   872:        pr_leave(pp);
1.21      thorpej   873:        simple_unlock(&pp->pr_slock);
1.1       pk        874:        return (v);
                    875: }
                    876:
                    877: /*
1.43      thorpej   878:  * Internal version of pool_put().  Pool is already locked/entered.
1.1       pk        879:  */
1.43      thorpej   880: static void
1.56      sommerfe  881: pool_do_put(struct pool *pp, void *v)
1.1       pk        882: {
                    883:        struct pool_item *pi = v;
1.3       pk        884:        struct pool_item_header *ph;
                    885:        caddr_t page;
1.21      thorpej   886:        int s;
1.3       pk        887:
1.61      chs       888:        LOCK_ASSERT(simple_lock_held(&pp->pr_slock));
                    889:
1.66      thorpej   890:        page = (caddr_t)((u_long)v & pp->pr_alloc->pa_pagemask);
1.1       pk        891:
1.30      thorpej   892: #ifdef DIAGNOSTIC
1.34      thorpej   893:        if (__predict_false(pp->pr_nout == 0)) {
1.30      thorpej   894:                printf("pool %s: putting with none out\n",
                    895:                    pp->pr_wchan);
                    896:                panic("pool_put");
                    897:        }
                    898: #endif
1.3       pk        899:
1.34      thorpej   900:        if (__predict_false((ph = pr_find_pagehead(pp, page)) == NULL)) {
1.25      thorpej   901:                pr_printlog(pp, NULL, printf);
1.3       pk        902:                panic("pool_put: %s: page header missing", pp->pr_wchan);
                    903:        }
1.28      thorpej   904:
                    905: #ifdef LOCKDEBUG
                    906:        /*
                    907:         * Check if we're freeing a locked simple lock.
                    908:         */
                    909:        simple_lock_freecheck((caddr_t)pi, ((caddr_t)pi) + pp->pr_size);
                    910: #endif
1.3       pk        911:
                    912:        /*
                    913:         * Return to item list.
                    914:         */
1.2       pk        915: #ifdef DIAGNOSTIC
1.3       pk        916:        pi->pi_magic = PI_MAGIC;
                    917: #endif
1.32      chs       918: #ifdef DEBUG
                    919:        {
                    920:                int i, *ip = v;
                    921:
                    922:                for (i = 0; i < pp->pr_size / sizeof(int); i++) {
                    923:                        *ip++ = PI_MAGIC;
                    924:                }
                    925:        }
                    926: #endif
                    927:
1.3       pk        928:        TAILQ_INSERT_HEAD(&ph->ph_itemlist, pi, pi_list);
                    929:        ph->ph_nmissing--;
                    930:        pp->pr_nput++;
1.20      thorpej   931:        pp->pr_nitems++;
                    932:        pp->pr_nout--;
1.3       pk        933:
                    934:        /* Cancel "pool empty" condition if it exists */
                    935:        if (pp->pr_curpage == NULL)
                    936:                pp->pr_curpage = ph;
                    937:
                    938:        if (pp->pr_flags & PR_WANTED) {
                    939:                pp->pr_flags &= ~PR_WANTED;
1.15      pk        940:                if (ph->ph_nmissing == 0)
                    941:                        pp->pr_nidle++;
1.3       pk        942:                wakeup((caddr_t)pp);
                    943:                return;
                    944:        }
                    945:
                    946:        /*
1.21      thorpej   947:         * If this page is now complete, do one of two things:
                    948:         *
                    949:         *      (1) If we have more pages than the page high water
                    950:         *          mark, free the page back to the system.
                    951:         *
                    952:         *      (2) Move it to the end of the page list, so that
                    953:         *          we minimize our chances of fragmenting the
                    954:         *          pool.  Idle pages migrate to the end (along with
                    955:         *          completely empty pages, so that we find un-empty
                    956:         *          pages more quickly when we update curpage) of the
                    957:         *          list so they can be more easily swept up by
                    958:         *          the pagedaemon when pages are scarce.
1.3       pk        959:         */
                    960:        if (ph->ph_nmissing == 0) {
1.6       thorpej   961:                pp->pr_nidle++;
1.71      thorpej   962:                if (pp->pr_npages > pp->pr_maxpages ||
                    963:                    (pp->pr_alloc->pa_flags & PA_WANT) != 0) {
1.61      chs       964:                        pr_rmpage(pp, ph, NULL);
1.3       pk        965:                } else {
                    966:                        TAILQ_REMOVE(&pp->pr_pagelist, ph, ph_pagelist);
                    967:                        TAILQ_INSERT_TAIL(&pp->pr_pagelist, ph, ph_pagelist);
                    968:
1.21      thorpej   969:                        /*
                    970:                         * Update the timestamp on the page.  A page must
                    971:                         * be idle for some period of time before it can
                    972:                         * be reclaimed by the pagedaemon.  This minimizes
                    973:                         * ping-pong'ing for memory.
                    974:                         */
                    975:                        s = splclock();
                    976:                        ph->ph_time = mono_time;
                    977:                        splx(s);
                    978:
                    979:                        /*
                    980:                         * Update the current page pointer.  Just look for
                    981:                         * the first page with any free items.
                    982:                         *
                    983:                         * XXX: Maybe we want an option to look for the
                    984:                         * page with the fewest available items, to minimize
                    985:                         * fragmentation?
                    986:                         */
1.61      chs       987:                        TAILQ_FOREACH(ph, &pp->pr_pagelist, ph_pagelist)
1.3       pk        988:                                if (TAILQ_FIRST(&ph->ph_itemlist) != NULL)
                    989:                                        break;
1.1       pk        990:
1.3       pk        991:                        pp->pr_curpage = ph;
1.1       pk        992:                }
                    993:        }
1.21      thorpej   994:        /*
                    995:         * If the page has just become un-empty, move it to the head of
                    996:         * the list, and make it the current page.  The next allocation
                    997:         * will get the item from this page, instead of further fragmenting
                    998:         * the pool.
                    999:         */
                   1000:        else if (ph->ph_nmissing == (pp->pr_itemsperpage - 1)) {
                   1001:                TAILQ_REMOVE(&pp->pr_pagelist, ph, ph_pagelist);
                   1002:                TAILQ_INSERT_HEAD(&pp->pr_pagelist, ph, ph_pagelist);
                   1003:                pp->pr_curpage = ph;
                   1004:        }
1.43      thorpej  1005: }
                   1006:
                   1007: /*
                   1008:  * Return resource to the pool; must be called at appropriate spl level
                   1009:  */
1.59      thorpej  1010: #ifdef POOL_DIAGNOSTIC
1.43      thorpej  1011: void
                   1012: _pool_put(struct pool *pp, void *v, const char *file, long line)
                   1013: {
                   1014:
                   1015:        simple_lock(&pp->pr_slock);
                   1016:        pr_enter(pp, file, line);
                   1017:
1.56      sommerfe 1018:        pr_log(pp, v, PRLOG_PUT, file, line);
                   1019:
                   1020:        pool_do_put(pp, v);
1.21      thorpej  1021:
1.25      thorpej  1022:        pr_leave(pp);
1.21      thorpej  1023:        simple_unlock(&pp->pr_slock);
1.1       pk       1024: }
1.57      sommerfe 1025: #undef pool_put
1.59      thorpej  1026: #endif /* POOL_DIAGNOSTIC */
1.1       pk       1027:
1.56      sommerfe 1028: void
                   1029: pool_put(struct pool *pp, void *v)
                   1030: {
                   1031:
                   1032:        simple_lock(&pp->pr_slock);
                   1033:
                   1034:        pool_do_put(pp, v);
                   1035:
                   1036:        simple_unlock(&pp->pr_slock);
                   1037: }
1.57      sommerfe 1038:
1.59      thorpej  1039: #ifdef POOL_DIAGNOSTIC
1.57      sommerfe 1040: #define                pool_put(h, v)  _pool_put((h), (v), __FILE__, __LINE__)
1.56      sommerfe 1041: #endif
1.55      thorpej  1042:
                   1043: /*
1.3       pk       1044:  * Add a page worth of items to the pool.
1.21      thorpej  1045:  *
                   1046:  * Note, we must be called with the pool descriptor LOCKED.
1.3       pk       1047:  */
1.55      thorpej  1048: static void
                   1049: pool_prime_page(struct pool *pp, caddr_t storage, struct pool_item_header *ph)
1.3       pk       1050: {
                   1051:        struct pool_item *pi;
                   1052:        caddr_t cp = storage;
                   1053:        unsigned int align = pp->pr_align;
                   1054:        unsigned int ioff = pp->pr_itemoffset;
1.55      thorpej  1055:        int n;
1.36      pk       1056:
1.66      thorpej  1057: #ifdef DIAGNOSTIC
                   1058:        if (((u_long)cp & (pp->pr_alloc->pa_pagesz - 1)) != 0)
1.36      pk       1059:                panic("pool_prime_page: %s: unaligned page", pp->pr_wchan);
1.66      thorpej  1060: #endif
1.3       pk       1061:
1.55      thorpej  1062:        if ((pp->pr_roflags & PR_PHINPAGE) == 0)
1.3       pk       1063:                LIST_INSERT_HEAD(&pp->pr_hashtab[PR_HASH_INDEX(pp, cp)],
1.55      thorpej  1064:                    ph, ph_hashlist);
1.3       pk       1065:
                   1066:        /*
                   1067:         * Insert page header.
                   1068:         */
                   1069:        TAILQ_INSERT_HEAD(&pp->pr_pagelist, ph, ph_pagelist);
                   1070:        TAILQ_INIT(&ph->ph_itemlist);
                   1071:        ph->ph_page = storage;
                   1072:        ph->ph_nmissing = 0;
1.21      thorpej  1073:        memset(&ph->ph_time, 0, sizeof(ph->ph_time));
1.3       pk       1074:
1.6       thorpej  1075:        pp->pr_nidle++;
                   1076:
1.3       pk       1077:        /*
                   1078:         * Color this page.
                   1079:         */
                   1080:        cp = (caddr_t)(cp + pp->pr_curcolor);
                   1081:        if ((pp->pr_curcolor += align) > pp->pr_maxcolor)
                   1082:                pp->pr_curcolor = 0;
                   1083:
                   1084:        /*
                   1085:         * Adjust storage to apply aligment to `pr_itemoffset' in each item.
                   1086:         */
                   1087:        if (ioff != 0)
                   1088:                cp = (caddr_t)(cp + (align - ioff));
                   1089:
                   1090:        /*
                   1091:         * Insert remaining chunks on the bucket list.
                   1092:         */
                   1093:        n = pp->pr_itemsperpage;
1.20      thorpej  1094:        pp->pr_nitems += n;
1.3       pk       1095:
                   1096:        while (n--) {
                   1097:                pi = (struct pool_item *)cp;
                   1098:
                   1099:                /* Insert on page list */
                   1100:                TAILQ_INSERT_TAIL(&ph->ph_itemlist, pi, pi_list);
                   1101: #ifdef DIAGNOSTIC
                   1102:                pi->pi_magic = PI_MAGIC;
                   1103: #endif
                   1104:                cp = (caddr_t)(cp + pp->pr_size);
                   1105:        }
                   1106:
                   1107:        /*
                   1108:         * If the pool was depleted, point at the new page.
                   1109:         */
                   1110:        if (pp->pr_curpage == NULL)
                   1111:                pp->pr_curpage = ph;
                   1112:
                   1113:        if (++pp->pr_npages > pp->pr_hiwat)
                   1114:                pp->pr_hiwat = pp->pr_npages;
                   1115: }
                   1116:
1.20      thorpej  1117: /*
1.52      thorpej  1118:  * Used by pool_get() when nitems drops below the low water mark.  This
                   1119:  * is used to catch up nitmes with the low water mark.
1.20      thorpej  1120:  *
1.21      thorpej  1121:  * Note 1, we never wait for memory here, we let the caller decide what to do.
1.20      thorpej  1122:  *
                   1123:  * Note 2, this doesn't work with static pools.
                   1124:  *
                   1125:  * Note 3, we must be called with the pool already locked, and we return
                   1126:  * with it locked.
                   1127:  */
                   1128: static int
1.42      thorpej  1129: pool_catchup(struct pool *pp)
1.20      thorpej  1130: {
1.55      thorpej  1131:        struct pool_item_header *ph;
1.20      thorpej  1132:        caddr_t cp;
                   1133:        int error = 0;
                   1134:
1.54      thorpej  1135:        while (POOL_NEEDS_CATCHUP(pp)) {
1.20      thorpej  1136:                /*
1.21      thorpej  1137:                 * Call the page back-end allocator for more memory.
                   1138:                 *
                   1139:                 * XXX: We never wait, so should we bother unlocking
                   1140:                 * the pool descriptor?
1.20      thorpej  1141:                 */
1.21      thorpej  1142:                simple_unlock(&pp->pr_slock);
1.66      thorpej  1143:                cp = pool_allocator_alloc(pp, PR_NOWAIT);
1.55      thorpej  1144:                if (__predict_true(cp != NULL))
                   1145:                        ph = pool_alloc_item_header(pp, cp, PR_NOWAIT);
1.21      thorpej  1146:                simple_lock(&pp->pr_slock);
1.55      thorpej  1147:                if (__predict_false(cp == NULL || ph == NULL)) {
                   1148:                        if (cp != NULL)
1.66      thorpej  1149:                                pool_allocator_free(pp, cp);
1.20      thorpej  1150:                        error = ENOMEM;
                   1151:                        break;
                   1152:                }
1.55      thorpej  1153:                pool_prime_page(pp, cp, ph);
1.26      thorpej  1154:                pp->pr_npagealloc++;
1.20      thorpej  1155:        }
                   1156:
                   1157:        return (error);
                   1158: }
                   1159:
1.3       pk       1160: void
1.42      thorpej  1161: pool_setlowat(struct pool *pp, int n)
1.3       pk       1162: {
1.20      thorpej  1163:        int error;
1.15      pk       1164:
1.21      thorpej  1165:        simple_lock(&pp->pr_slock);
                   1166:
1.3       pk       1167:        pp->pr_minitems = n;
1.15      pk       1168:        pp->pr_minpages = (n == 0)
                   1169:                ? 0
1.18      thorpej  1170:                : roundup(n, pp->pr_itemsperpage) / pp->pr_itemsperpage;
1.20      thorpej  1171:
                   1172:        /* Make sure we're caught up with the newly-set low water mark. */
1.53      thorpej  1173:        if (POOL_NEEDS_CATCHUP(pp) && (error = pool_catchup(pp) != 0)) {
1.20      thorpej  1174:                /*
                   1175:                 * XXX: Should we log a warning?  Should we set up a timeout
                   1176:                 * to try again in a second or so?  The latter could break
                   1177:                 * a caller's assumptions about interrupt protection, etc.
                   1178:                 */
                   1179:        }
1.21      thorpej  1180:
                   1181:        simple_unlock(&pp->pr_slock);
1.3       pk       1182: }
                   1183:
                   1184: void
1.42      thorpej  1185: pool_sethiwat(struct pool *pp, int n)
1.3       pk       1186: {
1.15      pk       1187:
1.21      thorpej  1188:        simple_lock(&pp->pr_slock);
                   1189:
1.15      pk       1190:        pp->pr_maxpages = (n == 0)
                   1191:                ? 0
1.18      thorpej  1192:                : roundup(n, pp->pr_itemsperpage) / pp->pr_itemsperpage;
1.21      thorpej  1193:
                   1194:        simple_unlock(&pp->pr_slock);
1.3       pk       1195: }
                   1196:
1.20      thorpej  1197: void
1.42      thorpej  1198: pool_sethardlimit(struct pool *pp, int n, const char *warnmess, int ratecap)
1.20      thorpej  1199: {
                   1200:
1.21      thorpej  1201:        simple_lock(&pp->pr_slock);
1.20      thorpej  1202:
                   1203:        pp->pr_hardlimit = n;
                   1204:        pp->pr_hardlimit_warning = warnmess;
1.31      thorpej  1205:        pp->pr_hardlimit_ratecap.tv_sec = ratecap;
                   1206:        pp->pr_hardlimit_warning_last.tv_sec = 0;
                   1207:        pp->pr_hardlimit_warning_last.tv_usec = 0;
1.20      thorpej  1208:
                   1209:        /*
1.21      thorpej  1210:         * In-line version of pool_sethiwat(), because we don't want to
                   1211:         * release the lock.
1.20      thorpej  1212:         */
                   1213:        pp->pr_maxpages = (n == 0)
                   1214:                ? 0
                   1215:                : roundup(n, pp->pr_itemsperpage) / pp->pr_itemsperpage;
1.21      thorpej  1216:
                   1217:        simple_unlock(&pp->pr_slock);
1.20      thorpej  1218: }
1.3       pk       1219:
                   1220: /*
                   1221:  * Release all complete pages that have not been used recently.
                   1222:  */
1.66      thorpej  1223: int
1.59      thorpej  1224: #ifdef POOL_DIAGNOSTIC
1.42      thorpej  1225: _pool_reclaim(struct pool *pp, const char *file, long line)
1.56      sommerfe 1226: #else
                   1227: pool_reclaim(struct pool *pp)
                   1228: #endif
1.3       pk       1229: {
                   1230:        struct pool_item_header *ph, *phnext;
1.43      thorpej  1231:        struct pool_cache *pc;
1.21      thorpej  1232:        struct timeval curtime;
1.61      chs      1233:        struct pool_pagelist pq;
1.21      thorpej  1234:        int s;
1.3       pk       1235:
1.68      thorpej  1236:        if (pp->pr_drain_hook != NULL) {
                   1237:                /*
                   1238:                 * The drain hook must be called with the pool unlocked.
                   1239:                 */
                   1240:                (*pp->pr_drain_hook)(pp->pr_drain_hook_arg, PR_NOWAIT);
                   1241:        }
                   1242:
1.21      thorpej  1243:        if (simple_lock_try(&pp->pr_slock) == 0)
1.66      thorpej  1244:                return (0);
1.25      thorpej  1245:        pr_enter(pp, file, line);
1.68      thorpej  1246:
1.61      chs      1247:        TAILQ_INIT(&pq);
1.3       pk       1248:
1.43      thorpej  1249:        /*
                   1250:         * Reclaim items from the pool's caches.
                   1251:         */
1.61      chs      1252:        TAILQ_FOREACH(pc, &pp->pr_cachelist, pc_poollist)
1.43      thorpej  1253:                pool_cache_reclaim(pc);
                   1254:
1.21      thorpej  1255:        s = splclock();
                   1256:        curtime = mono_time;
                   1257:        splx(s);
                   1258:
1.3       pk       1259:        for (ph = TAILQ_FIRST(&pp->pr_pagelist); ph != NULL; ph = phnext) {
                   1260:                phnext = TAILQ_NEXT(ph, ph_pagelist);
                   1261:
                   1262:                /* Check our minimum page claim */
                   1263:                if (pp->pr_npages <= pp->pr_minpages)
                   1264:                        break;
                   1265:
                   1266:                if (ph->ph_nmissing == 0) {
                   1267:                        struct timeval diff;
                   1268:                        timersub(&curtime, &ph->ph_time, &diff);
                   1269:                        if (diff.tv_sec < pool_inactive_time)
                   1270:                                continue;
1.21      thorpej  1271:
                   1272:                        /*
                   1273:                         * If freeing this page would put us below
                   1274:                         * the low water mark, stop now.
                   1275:                         */
                   1276:                        if ((pp->pr_nitems - pp->pr_itemsperpage) <
                   1277:                            pp->pr_minitems)
                   1278:                                break;
                   1279:
1.61      chs      1280:                        pr_rmpage(pp, ph, &pq);
1.3       pk       1281:                }
                   1282:        }
                   1283:
1.25      thorpej  1284:        pr_leave(pp);
1.21      thorpej  1285:        simple_unlock(&pp->pr_slock);
1.66      thorpej  1286:        if (TAILQ_EMPTY(&pq))
                   1287:                return (0);
                   1288:
1.61      chs      1289:        while ((ph = TAILQ_FIRST(&pq)) != NULL) {
                   1290:                TAILQ_REMOVE(&pq, ph, ph_pagelist);
1.66      thorpej  1291:                pool_allocator_free(pp, ph->ph_page);
1.61      chs      1292:                if (pp->pr_roflags & PR_PHINPAGE) {
                   1293:                        continue;
                   1294:                }
                   1295:                LIST_REMOVE(ph, ph_hashlist);
                   1296:                s = splhigh();
                   1297:                pool_put(&phpool, ph);
                   1298:                splx(s);
                   1299:        }
1.66      thorpej  1300:
                   1301:        return (1);
1.3       pk       1302: }
                   1303:
                   1304: /*
                   1305:  * Drain pools, one at a time.
1.21      thorpej  1306:  *
                   1307:  * Note, we must never be called from an interrupt context.
1.3       pk       1308:  */
                   1309: void
1.42      thorpej  1310: pool_drain(void *arg)
1.3       pk       1311: {
                   1312:        struct pool *pp;
1.23      thorpej  1313:        int s;
1.3       pk       1314:
1.61      chs      1315:        pp = NULL;
1.49      thorpej  1316:        s = splvm();
1.23      thorpej  1317:        simple_lock(&pool_head_slock);
1.61      chs      1318:        if (drainpp == NULL) {
                   1319:                drainpp = TAILQ_FIRST(&pool_head);
                   1320:        }
                   1321:        if (drainpp) {
                   1322:                pp = drainpp;
                   1323:                drainpp = TAILQ_NEXT(pp, pr_poollist);
                   1324:        }
                   1325:        simple_unlock(&pool_head_slock);
1.63      chs      1326:        pool_reclaim(pp);
1.61      chs      1327:        splx(s);
1.3       pk       1328: }
                   1329:
                   1330: /*
                   1331:  * Diagnostic helpers.
                   1332:  */
                   1333: void
1.42      thorpej  1334: pool_print(struct pool *pp, const char *modif)
1.21      thorpej  1335: {
                   1336:        int s;
                   1337:
1.49      thorpej  1338:        s = splvm();
1.25      thorpej  1339:        if (simple_lock_try(&pp->pr_slock) == 0) {
                   1340:                printf("pool %s is locked; try again later\n",
                   1341:                    pp->pr_wchan);
                   1342:                splx(s);
                   1343:                return;
                   1344:        }
                   1345:        pool_print1(pp, modif, printf);
1.21      thorpej  1346:        simple_unlock(&pp->pr_slock);
                   1347:        splx(s);
                   1348: }
                   1349:
1.25      thorpej  1350: void
1.42      thorpej  1351: pool_printit(struct pool *pp, const char *modif, void (*pr)(const char *, ...))
1.25      thorpej  1352: {
                   1353:        int didlock = 0;
                   1354:
                   1355:        if (pp == NULL) {
                   1356:                (*pr)("Must specify a pool to print.\n");
                   1357:                return;
                   1358:        }
                   1359:
                   1360:        /*
                   1361:         * Called from DDB; interrupts should be blocked, and all
                   1362:         * other processors should be paused.  We can skip locking
                   1363:         * the pool in this case.
                   1364:         *
                   1365:         * We do a simple_lock_try() just to print the lock
                   1366:         * status, however.
                   1367:         */
                   1368:
                   1369:        if (simple_lock_try(&pp->pr_slock) == 0)
                   1370:                (*pr)("WARNING: pool %s is locked\n", pp->pr_wchan);
                   1371:        else
                   1372:                didlock = 1;
                   1373:
                   1374:        pool_print1(pp, modif, pr);
                   1375:
                   1376:        if (didlock)
                   1377:                simple_unlock(&pp->pr_slock);
                   1378: }
                   1379:
1.21      thorpej  1380: static void
1.42      thorpej  1381: pool_print1(struct pool *pp, const char *modif, void (*pr)(const char *, ...))
1.3       pk       1382: {
1.25      thorpej  1383:        struct pool_item_header *ph;
1.44      thorpej  1384:        struct pool_cache *pc;
                   1385:        struct pool_cache_group *pcg;
1.25      thorpej  1386: #ifdef DIAGNOSTIC
                   1387:        struct pool_item *pi;
                   1388: #endif
1.44      thorpej  1389:        int i, print_log = 0, print_pagelist = 0, print_cache = 0;
1.25      thorpej  1390:        char c;
                   1391:
                   1392:        while ((c = *modif++) != '\0') {
                   1393:                if (c == 'l')
                   1394:                        print_log = 1;
                   1395:                if (c == 'p')
                   1396:                        print_pagelist = 1;
1.44      thorpej  1397:                if (c == 'c')
                   1398:                        print_cache = 1;
1.25      thorpej  1399:                modif++;
                   1400:        }
                   1401:
                   1402:        (*pr)("POOL %s: size %u, align %u, ioff %u, roflags 0x%08x\n",
                   1403:            pp->pr_wchan, pp->pr_size, pp->pr_align, pp->pr_itemoffset,
                   1404:            pp->pr_roflags);
1.66      thorpej  1405:        (*pr)("\talloc %p\n", pp->pr_alloc);
1.25      thorpej  1406:        (*pr)("\tminitems %u, minpages %u, maxpages %u, npages %u\n",
                   1407:            pp->pr_minitems, pp->pr_minpages, pp->pr_maxpages, pp->pr_npages);
                   1408:        (*pr)("\titemsperpage %u, nitems %u, nout %u, hardlimit %u\n",
                   1409:            pp->pr_itemsperpage, pp->pr_nitems, pp->pr_nout, pp->pr_hardlimit);
                   1410:
                   1411:        (*pr)("\n\tnget %lu, nfail %lu, nput %lu\n",
                   1412:            pp->pr_nget, pp->pr_nfail, pp->pr_nput);
                   1413:        (*pr)("\tnpagealloc %lu, npagefree %lu, hiwat %u, nidle %lu\n",
                   1414:            pp->pr_npagealloc, pp->pr_npagefree, pp->pr_hiwat, pp->pr_nidle);
                   1415:
                   1416:        if (print_pagelist == 0)
                   1417:                goto skip_pagelist;
                   1418:
                   1419:        if ((ph = TAILQ_FIRST(&pp->pr_pagelist)) != NULL)
                   1420:                (*pr)("\n\tpage list:\n");
                   1421:        for (; ph != NULL; ph = TAILQ_NEXT(ph, ph_pagelist)) {
                   1422:                (*pr)("\t\tpage %p, nmissing %d, time %lu,%lu\n",
                   1423:                    ph->ph_page, ph->ph_nmissing,
                   1424:                    (u_long)ph->ph_time.tv_sec,
                   1425:                    (u_long)ph->ph_time.tv_usec);
                   1426: #ifdef DIAGNOSTIC
1.61      chs      1427:                TAILQ_FOREACH(pi, &ph->ph_itemlist, pi_list) {
1.25      thorpej  1428:                        if (pi->pi_magic != PI_MAGIC) {
                   1429:                                (*pr)("\t\t\titem %p, magic 0x%x\n",
                   1430:                                    pi, pi->pi_magic);
                   1431:                        }
                   1432:                }
                   1433: #endif
                   1434:        }
                   1435:        if (pp->pr_curpage == NULL)
                   1436:                (*pr)("\tno current page\n");
                   1437:        else
                   1438:                (*pr)("\tcurpage %p\n", pp->pr_curpage->ph_page);
                   1439:
                   1440:  skip_pagelist:
                   1441:
                   1442:        if (print_log == 0)
                   1443:                goto skip_log;
                   1444:
                   1445:        (*pr)("\n");
                   1446:        if ((pp->pr_roflags & PR_LOGGING) == 0)
                   1447:                (*pr)("\tno log\n");
                   1448:        else
                   1449:                pr_printlog(pp, NULL, pr);
1.3       pk       1450:
1.25      thorpej  1451:  skip_log:
1.44      thorpej  1452:
                   1453:        if (print_cache == 0)
                   1454:                goto skip_cache;
                   1455:
1.61      chs      1456:        TAILQ_FOREACH(pc, &pp->pr_cachelist, pc_poollist) {
1.44      thorpej  1457:                (*pr)("\tcache %p: allocfrom %p freeto %p\n", pc,
                   1458:                    pc->pc_allocfrom, pc->pc_freeto);
1.48      thorpej  1459:                (*pr)("\t    hits %lu misses %lu ngroups %lu nitems %lu\n",
                   1460:                    pc->pc_hits, pc->pc_misses, pc->pc_ngroups, pc->pc_nitems);
1.61      chs      1461:                TAILQ_FOREACH(pcg, &pc->pc_grouplist, pcg_list) {
1.44      thorpej  1462:                        (*pr)("\t\tgroup %p: avail %d\n", pcg, pcg->pcg_avail);
                   1463:                        for (i = 0; i < PCG_NOBJECTS; i++)
                   1464:                                (*pr)("\t\t\t%p\n", pcg->pcg_objects[i]);
                   1465:                }
                   1466:        }
                   1467:
                   1468:  skip_cache:
1.3       pk       1469:
1.25      thorpej  1470:        pr_enter_check(pp, pr);
1.3       pk       1471: }
                   1472:
                   1473: int
1.42      thorpej  1474: pool_chk(struct pool *pp, const char *label)
1.3       pk       1475: {
                   1476:        struct pool_item_header *ph;
                   1477:        int r = 0;
                   1478:
1.21      thorpej  1479:        simple_lock(&pp->pr_slock);
1.3       pk       1480:
1.61      chs      1481:        TAILQ_FOREACH(ph, &pp->pr_pagelist, ph_pagelist) {
1.3       pk       1482:                struct pool_item *pi;
                   1483:                int n;
                   1484:                caddr_t page;
                   1485:
1.66      thorpej  1486:                page = (caddr_t)((u_long)ph & pp->pr_alloc->pa_pagemask);
1.20      thorpej  1487:                if (page != ph->ph_page &&
                   1488:                    (pp->pr_roflags & PR_PHINPAGE) != 0) {
1.3       pk       1489:                        if (label != NULL)
                   1490:                                printf("%s: ", label);
1.16      briggs   1491:                        printf("pool(%p:%s): page inconsistency: page %p;"
                   1492:                               " at page head addr %p (p %p)\n", pp,
1.3       pk       1493:                                pp->pr_wchan, ph->ph_page,
                   1494:                                ph, page);
                   1495:                        r++;
                   1496:                        goto out;
                   1497:                }
                   1498:
                   1499:                for (pi = TAILQ_FIRST(&ph->ph_itemlist), n = 0;
                   1500:                     pi != NULL;
                   1501:                     pi = TAILQ_NEXT(pi,pi_list), n++) {
                   1502:
                   1503: #ifdef DIAGNOSTIC
                   1504:                        if (pi->pi_magic != PI_MAGIC) {
                   1505:                                if (label != NULL)
                   1506:                                        printf("%s: ", label);
                   1507:                                printf("pool(%s): free list modified: magic=%x;"
                   1508:                                       " page %p; item ordinal %d;"
                   1509:                                       " addr %p (p %p)\n",
                   1510:                                        pp->pr_wchan, pi->pi_magic, ph->ph_page,
                   1511:                                        n, pi, page);
                   1512:                                panic("pool");
                   1513:                        }
                   1514: #endif
1.66      thorpej  1515:                        page =
                   1516:                            (caddr_t)((u_long)pi & pp->pr_alloc->pa_pagemask);
1.3       pk       1517:                        if (page == ph->ph_page)
                   1518:                                continue;
                   1519:
                   1520:                        if (label != NULL)
                   1521:                                printf("%s: ", label);
1.16      briggs   1522:                        printf("pool(%p:%s): page inconsistency: page %p;"
                   1523:                               " item ordinal %d; addr %p (p %p)\n", pp,
1.3       pk       1524:                                pp->pr_wchan, ph->ph_page,
                   1525:                                n, pi, page);
                   1526:                        r++;
                   1527:                        goto out;
                   1528:                }
                   1529:        }
                   1530: out:
1.21      thorpej  1531:        simple_unlock(&pp->pr_slock);
1.3       pk       1532:        return (r);
1.43      thorpej  1533: }
                   1534:
                   1535: /*
                   1536:  * pool_cache_init:
                   1537:  *
                   1538:  *     Initialize a pool cache.
                   1539:  *
                   1540:  *     NOTE: If the pool must be protected from interrupts, we expect
                   1541:  *     to be called at the appropriate interrupt priority level.
                   1542:  */
                   1543: void
                   1544: pool_cache_init(struct pool_cache *pc, struct pool *pp,
                   1545:     int (*ctor)(void *, void *, int),
                   1546:     void (*dtor)(void *, void *),
                   1547:     void *arg)
                   1548: {
                   1549:
                   1550:        TAILQ_INIT(&pc->pc_grouplist);
                   1551:        simple_lock_init(&pc->pc_slock);
                   1552:
                   1553:        pc->pc_allocfrom = NULL;
                   1554:        pc->pc_freeto = NULL;
                   1555:        pc->pc_pool = pp;
                   1556:
                   1557:        pc->pc_ctor = ctor;
                   1558:        pc->pc_dtor = dtor;
                   1559:        pc->pc_arg  = arg;
                   1560:
1.48      thorpej  1561:        pc->pc_hits   = 0;
                   1562:        pc->pc_misses = 0;
                   1563:
                   1564:        pc->pc_ngroups = 0;
                   1565:
                   1566:        pc->pc_nitems = 0;
                   1567:
1.43      thorpej  1568:        simple_lock(&pp->pr_slock);
                   1569:        TAILQ_INSERT_TAIL(&pp->pr_cachelist, pc, pc_poollist);
                   1570:        simple_unlock(&pp->pr_slock);
                   1571: }
                   1572:
                   1573: /*
                   1574:  * pool_cache_destroy:
                   1575:  *
                   1576:  *     Destroy a pool cache.
                   1577:  */
                   1578: void
                   1579: pool_cache_destroy(struct pool_cache *pc)
                   1580: {
                   1581:        struct pool *pp = pc->pc_pool;
                   1582:
                   1583:        /* First, invalidate the entire cache. */
                   1584:        pool_cache_invalidate(pc);
                   1585:
                   1586:        /* ...and remove it from the pool's cache list. */
                   1587:        simple_lock(&pp->pr_slock);
                   1588:        TAILQ_REMOVE(&pp->pr_cachelist, pc, pc_poollist);
                   1589:        simple_unlock(&pp->pr_slock);
                   1590: }
                   1591:
                   1592: static __inline void *
                   1593: pcg_get(struct pool_cache_group *pcg)
                   1594: {
                   1595:        void *object;
                   1596:        u_int idx;
                   1597:
                   1598:        KASSERT(pcg->pcg_avail <= PCG_NOBJECTS);
1.45      thorpej  1599:        KASSERT(pcg->pcg_avail != 0);
1.43      thorpej  1600:        idx = --pcg->pcg_avail;
                   1601:
                   1602:        KASSERT(pcg->pcg_objects[idx] != NULL);
                   1603:        object = pcg->pcg_objects[idx];
                   1604:        pcg->pcg_objects[idx] = NULL;
                   1605:
                   1606:        return (object);
                   1607: }
                   1608:
                   1609: static __inline void
                   1610: pcg_put(struct pool_cache_group *pcg, void *object)
                   1611: {
                   1612:        u_int idx;
                   1613:
                   1614:        KASSERT(pcg->pcg_avail < PCG_NOBJECTS);
                   1615:        idx = pcg->pcg_avail++;
                   1616:
                   1617:        KASSERT(pcg->pcg_objects[idx] == NULL);
                   1618:        pcg->pcg_objects[idx] = object;
                   1619: }
                   1620:
                   1621: /*
                   1622:  * pool_cache_get:
                   1623:  *
                   1624:  *     Get an object from a pool cache.
                   1625:  */
                   1626: void *
                   1627: pool_cache_get(struct pool_cache *pc, int flags)
                   1628: {
                   1629:        struct pool_cache_group *pcg;
                   1630:        void *object;
1.58      thorpej  1631:
                   1632: #ifdef LOCKDEBUG
                   1633:        if (flags & PR_WAITOK)
                   1634:                simple_lock_only_held(NULL, "pool_cache_get(PR_WAITOK)");
                   1635: #endif
1.43      thorpej  1636:
                   1637:        simple_lock(&pc->pc_slock);
                   1638:
                   1639:        if ((pcg = pc->pc_allocfrom) == NULL) {
1.61      chs      1640:                TAILQ_FOREACH(pcg, &pc->pc_grouplist, pcg_list) {
1.43      thorpej  1641:                        if (pcg->pcg_avail != 0) {
                   1642:                                pc->pc_allocfrom = pcg;
                   1643:                                goto have_group;
                   1644:                        }
                   1645:                }
                   1646:
                   1647:                /*
                   1648:                 * No groups with any available objects.  Allocate
                   1649:                 * a new object, construct it, and return it to
                   1650:                 * the caller.  We will allocate a group, if necessary,
                   1651:                 * when the object is freed back to the cache.
                   1652:                 */
1.48      thorpej  1653:                pc->pc_misses++;
1.43      thorpej  1654:                simple_unlock(&pc->pc_slock);
                   1655:                object = pool_get(pc->pc_pool, flags);
                   1656:                if (object != NULL && pc->pc_ctor != NULL) {
                   1657:                        if ((*pc->pc_ctor)(pc->pc_arg, object, flags) != 0) {
                   1658:                                pool_put(pc->pc_pool, object);
                   1659:                                return (NULL);
                   1660:                        }
                   1661:                }
                   1662:                return (object);
                   1663:        }
                   1664:
                   1665:  have_group:
1.48      thorpej  1666:        pc->pc_hits++;
                   1667:        pc->pc_nitems--;
1.43      thorpej  1668:        object = pcg_get(pcg);
                   1669:
                   1670:        if (pcg->pcg_avail == 0)
                   1671:                pc->pc_allocfrom = NULL;
1.45      thorpej  1672:
1.43      thorpej  1673:        simple_unlock(&pc->pc_slock);
                   1674:
                   1675:        return (object);
                   1676: }
                   1677:
                   1678: /*
                   1679:  * pool_cache_put:
                   1680:  *
                   1681:  *     Put an object back to the pool cache.
                   1682:  */
                   1683: void
                   1684: pool_cache_put(struct pool_cache *pc, void *object)
                   1685: {
                   1686:        struct pool_cache_group *pcg;
1.60      thorpej  1687:        int s;
1.43      thorpej  1688:
                   1689:        simple_lock(&pc->pc_slock);
                   1690:
                   1691:        if ((pcg = pc->pc_freeto) == NULL) {
1.61      chs      1692:                TAILQ_FOREACH(pcg, &pc->pc_grouplist, pcg_list) {
1.43      thorpej  1693:                        if (pcg->pcg_avail != PCG_NOBJECTS) {
                   1694:                                pc->pc_freeto = pcg;
                   1695:                                goto have_group;
                   1696:                        }
                   1697:                }
                   1698:
                   1699:                /*
                   1700:                 * No empty groups to free the object to.  Attempt to
1.47      thorpej  1701:                 * allocate one.
1.43      thorpej  1702:                 */
1.47      thorpej  1703:                simple_unlock(&pc->pc_slock);
1.60      thorpej  1704:                s = splvm();
1.43      thorpej  1705:                pcg = pool_get(&pcgpool, PR_NOWAIT);
1.60      thorpej  1706:                splx(s);
1.43      thorpej  1707:                if (pcg != NULL) {
                   1708:                        memset(pcg, 0, sizeof(*pcg));
1.47      thorpej  1709:                        simple_lock(&pc->pc_slock);
1.48      thorpej  1710:                        pc->pc_ngroups++;
1.43      thorpej  1711:                        TAILQ_INSERT_TAIL(&pc->pc_grouplist, pcg, pcg_list);
1.47      thorpej  1712:                        if (pc->pc_freeto == NULL)
                   1713:                                pc->pc_freeto = pcg;
1.43      thorpej  1714:                        goto have_group;
                   1715:                }
                   1716:
                   1717:                /*
                   1718:                 * Unable to allocate a cache group; destruct the object
                   1719:                 * and free it back to the pool.
                   1720:                 */
1.51      thorpej  1721:                pool_cache_destruct_object(pc, object);
1.43      thorpej  1722:                return;
                   1723:        }
                   1724:
                   1725:  have_group:
1.48      thorpej  1726:        pc->pc_nitems++;
1.43      thorpej  1727:        pcg_put(pcg, object);
                   1728:
                   1729:        if (pcg->pcg_avail == PCG_NOBJECTS)
                   1730:                pc->pc_freeto = NULL;
                   1731:
                   1732:        simple_unlock(&pc->pc_slock);
1.51      thorpej  1733: }
                   1734:
                   1735: /*
                   1736:  * pool_cache_destruct_object:
                   1737:  *
                   1738:  *     Force destruction of an object and its release back into
                   1739:  *     the pool.
                   1740:  */
                   1741: void
                   1742: pool_cache_destruct_object(struct pool_cache *pc, void *object)
                   1743: {
                   1744:
                   1745:        if (pc->pc_dtor != NULL)
                   1746:                (*pc->pc_dtor)(pc->pc_arg, object);
                   1747:        pool_put(pc->pc_pool, object);
1.43      thorpej  1748: }
                   1749:
                   1750: /*
                   1751:  * pool_cache_do_invalidate:
                   1752:  *
                   1753:  *     This internal function implements pool_cache_invalidate() and
                   1754:  *     pool_cache_reclaim().
                   1755:  */
                   1756: static void
                   1757: pool_cache_do_invalidate(struct pool_cache *pc, int free_groups,
1.56      sommerfe 1758:     void (*putit)(struct pool *, void *))
1.43      thorpej  1759: {
                   1760:        struct pool_cache_group *pcg, *npcg;
                   1761:        void *object;
1.60      thorpej  1762:        int s;
1.43      thorpej  1763:
                   1764:        for (pcg = TAILQ_FIRST(&pc->pc_grouplist); pcg != NULL;
                   1765:             pcg = npcg) {
                   1766:                npcg = TAILQ_NEXT(pcg, pcg_list);
                   1767:                while (pcg->pcg_avail != 0) {
1.48      thorpej  1768:                        pc->pc_nitems--;
1.43      thorpej  1769:                        object = pcg_get(pcg);
1.45      thorpej  1770:                        if (pcg->pcg_avail == 0 && pc->pc_allocfrom == pcg)
                   1771:                                pc->pc_allocfrom = NULL;
1.43      thorpej  1772:                        if (pc->pc_dtor != NULL)
                   1773:                                (*pc->pc_dtor)(pc->pc_arg, object);
1.56      sommerfe 1774:                        (*putit)(pc->pc_pool, object);
1.43      thorpej  1775:                }
                   1776:                if (free_groups) {
1.48      thorpej  1777:                        pc->pc_ngroups--;
1.43      thorpej  1778:                        TAILQ_REMOVE(&pc->pc_grouplist, pcg, pcg_list);
1.46      thorpej  1779:                        if (pc->pc_freeto == pcg)
                   1780:                                pc->pc_freeto = NULL;
1.60      thorpej  1781:                        s = splvm();
1.43      thorpej  1782:                        pool_put(&pcgpool, pcg);
1.60      thorpej  1783:                        splx(s);
1.43      thorpej  1784:                }
                   1785:        }
                   1786: }
                   1787:
                   1788: /*
                   1789:  * pool_cache_invalidate:
                   1790:  *
                   1791:  *     Invalidate a pool cache (destruct and release all of the
                   1792:  *     cached objects).
                   1793:  */
                   1794: void
                   1795: pool_cache_invalidate(struct pool_cache *pc)
                   1796: {
                   1797:
                   1798:        simple_lock(&pc->pc_slock);
1.56      sommerfe 1799:        pool_cache_do_invalidate(pc, 0, pool_put);
1.43      thorpej  1800:        simple_unlock(&pc->pc_slock);
                   1801: }
                   1802:
                   1803: /*
                   1804:  * pool_cache_reclaim:
                   1805:  *
                   1806:  *     Reclaim a pool cache for pool_reclaim().
                   1807:  */
                   1808: static void
                   1809: pool_cache_reclaim(struct pool_cache *pc)
                   1810: {
                   1811:
1.47      thorpej  1812:        simple_lock(&pc->pc_slock);
1.43      thorpej  1813:        pool_cache_do_invalidate(pc, 1, pool_do_put);
                   1814:        simple_unlock(&pc->pc_slock);
1.3       pk       1815: }
1.66      thorpej  1816:
                   1817: /*
                   1818:  * Pool backend allocators.
                   1819:  *
                   1820:  * Each pool has a backend allocator that handles allocation, deallocation,
                   1821:  * and any additional draining that might be needed.
                   1822:  *
                   1823:  * We provide two standard allocators:
                   1824:  *
                   1825:  *     pool_allocator_kmem - the default when no allocator is specified
                   1826:  *
                   1827:  *     pool_allocator_nointr - used for pools that will not be accessed
                   1828:  *     in interrupt context.
                   1829:  */
                   1830: void   *pool_page_alloc(struct pool *, int);
                   1831: void   pool_page_free(struct pool *, void *);
                   1832:
                   1833: struct pool_allocator pool_allocator_kmem = {
                   1834:        pool_page_alloc, pool_page_free, 0,
                   1835: };
                   1836:
                   1837: void   *pool_page_alloc_nointr(struct pool *, int);
                   1838: void   pool_page_free_nointr(struct pool *, void *);
                   1839:
                   1840: struct pool_allocator pool_allocator_nointr = {
                   1841:        pool_page_alloc_nointr, pool_page_free_nointr, 0,
                   1842: };
                   1843:
                   1844: #ifdef POOL_SUBPAGE
                   1845: void   *pool_subpage_alloc(struct pool *, int);
                   1846: void   pool_subpage_free(struct pool *, void *);
                   1847:
                   1848: struct pool_allocator pool_allocator_kmem_subpage = {
                   1849:        pool_subpage_alloc, pool_subpage_free, 0,
                   1850: };
                   1851: #endif /* POOL_SUBPAGE */
                   1852:
                   1853: /*
                   1854:  * We have at least three different resources for the same allocation and
                   1855:  * each resource can be depleted.  First, we have the ready elements in the
                   1856:  * pool.  Then we have the resource (typically a vm_map) for this allocator.
                   1857:  * Finally, we have physical memory.  Waiting for any of these can be
                   1858:  * unnecessary when any other is freed, but the kernel doesn't support
                   1859:  * sleeping on multiple wait channels, so we have to employ another strategy.
                   1860:  *
                   1861:  * The caller sleeps on the pool (so that it can be awakened when an item
                   1862:  * is returned to the pool), but we set PA_WANT on the allocator.  When a
                   1863:  * page is returned to the allocator and PA_WANT is set, pool_allocator_free
                   1864:  * will wake up all sleeping pools belonging to this allocator.
                   1865:  *
                   1866:  * XXX Thundering herd.
                   1867:  */
                   1868: void *
                   1869: pool_allocator_alloc(struct pool *org, int flags)
                   1870: {
                   1871:        struct pool_allocator *pa = org->pr_alloc;
                   1872:        struct pool *pp, *start;
                   1873:        int s, freed;
                   1874:        void *res;
                   1875:
                   1876:        do {
                   1877:                if ((res = (*pa->pa_alloc)(org, flags)) != NULL)
                   1878:                        return (res);
1.68      thorpej  1879:                if ((flags & PR_WAITOK) == 0) {
                   1880:                        /*
                   1881:                         * We only run the drain hookhere if PR_NOWAIT.
                   1882:                         * In other cases, the hook will be run in
                   1883:                         * pool_reclaim().
                   1884:                         */
                   1885:                        if (org->pr_drain_hook != NULL) {
                   1886:                                (*org->pr_drain_hook)(org->pr_drain_hook_arg,
                   1887:                                    flags);
                   1888:                                if ((res = (*pa->pa_alloc)(org, flags)) != NULL)
                   1889:                                        return (res);
                   1890:                        }
1.66      thorpej  1891:                        break;
1.68      thorpej  1892:                }
1.66      thorpej  1893:
                   1894:                /*
                   1895:                 * Drain all pools, except "org", that use this
                   1896:                 * allocator.  We do this to reclaim VA space.
                   1897:                 * pa_alloc is responsible for waiting for
                   1898:                 * physical memory.
                   1899:                 *
                   1900:                 * XXX We risk looping forever if start if someone
                   1901:                 * calls pool_destroy on "start".  But there is no
                   1902:                 * other way to have potentially sleeping pool_reclaim,
                   1903:                 * non-sleeping locks on pool_allocator, and some
                   1904:                 * stirring of drained pools in the allocator.
1.68      thorpej  1905:                 *
                   1906:                 * XXX Maybe we should use pool_head_slock for locking
                   1907:                 * the allocators?
1.66      thorpej  1908:                 */
                   1909:                freed = 0;
                   1910:
                   1911:                s = splvm();
                   1912:                simple_lock(&pa->pa_slock);
                   1913:                pp = start = TAILQ_FIRST(&pa->pa_list);
                   1914:                do {
                   1915:                        TAILQ_REMOVE(&pa->pa_list, pp, pr_alloc_list);
                   1916:                        TAILQ_INSERT_TAIL(&pa->pa_list, pp, pr_alloc_list);
                   1917:                        if (pp == org)
                   1918:                                continue;
                   1919:                        simple_unlock(&pa->pa_list);
                   1920:                        freed = pool_reclaim(pp);
                   1921:                        simple_lock(&pa->pa_list);
                   1922:                } while ((pp = TAILQ_FIRST(&pa->pa_list)) != start &&
                   1923:                         freed == 0);
                   1924:
                   1925:                if (freed == 0) {
                   1926:                        /*
                   1927:                         * We set PA_WANT here, the caller will most likely
                   1928:                         * sleep waiting for pages (if not, this won't hurt
                   1929:                         * that much), and there is no way to set this in
                   1930:                         * the caller without violating locking order.
                   1931:                         */
                   1932:                        pa->pa_flags |= PA_WANT;
                   1933:                }
                   1934:                simple_unlock(&pa->pa_slock);
                   1935:                splx(s);
                   1936:        } while (freed);
                   1937:        return (NULL);
                   1938: }
                   1939:
                   1940: void
                   1941: pool_allocator_free(struct pool *pp, void *v)
                   1942: {
                   1943:        struct pool_allocator *pa = pp->pr_alloc;
                   1944:        int s;
                   1945:
                   1946:        (*pa->pa_free)(pp, v);
                   1947:
                   1948:        s = splvm();
                   1949:        simple_lock(&pa->pa_slock);
                   1950:        if ((pa->pa_flags & PA_WANT) == 0) {
                   1951:                simple_unlock(&pa->pa_slock);
                   1952:                splx(s);
                   1953:                return;
                   1954:        }
                   1955:
                   1956:        TAILQ_FOREACH(pp, &pa->pa_list, pr_alloc_list) {
                   1957:                simple_lock(&pp->pr_slock);
                   1958:                if ((pp->pr_flags & PR_WANTED) != 0) {
                   1959:                        pp->pr_flags &= ~PR_WANTED;
                   1960:                        wakeup(pp);
                   1961:                }
1.69      thorpej  1962:                simple_unlock(&pp->pr_slock);
1.66      thorpej  1963:        }
                   1964:        pa->pa_flags &= ~PA_WANT;
                   1965:        simple_unlock(&pa->pa_slock);
                   1966:        splx(s);
                   1967: }
                   1968:
                   1969: void *
                   1970: pool_page_alloc(struct pool *pp, int flags)
                   1971: {
                   1972:        boolean_t waitok = (flags & PR_WAITOK) ? TRUE : FALSE;
                   1973:
                   1974:        return ((void *) uvm_km_alloc_poolpage(waitok));
                   1975: }
                   1976:
                   1977: void
                   1978: pool_page_free(struct pool *pp, void *v)
                   1979: {
                   1980:
                   1981:        uvm_km_free_poolpage((vaddr_t) v);
                   1982: }
                   1983:
                   1984: #ifdef POOL_SUBPAGE
                   1985: /* Sub-page allocator, for machines with large hardware pages. */
                   1986: void *
                   1987: pool_subpage_alloc(struct pool *pp, int flags)
                   1988: {
                   1989:
                   1990:        return (pool_get(&psppool, flags));
                   1991: }
                   1992:
                   1993: void
                   1994: pool_subpage_free(struct pool *pp, void *v)
                   1995: {
                   1996:
                   1997:        pool_put(&psppool, v);
                   1998: }
                   1999:
                   2000: /* We don't provide a real nointr allocator.  Maybe later. */
                   2001: void *
                   2002: pool_page_alloc_nointr(struct pool *pp, int flags)
                   2003: {
                   2004:
                   2005:        return (pool_subpage_alloc(pp, flags));
                   2006: }
                   2007:
                   2008: void
                   2009: pool_page_free_nointr(struct pool *pp, void *v)
                   2010: {
                   2011:
                   2012:        pool_subpage_free(pp, v);
                   2013: }
                   2014: #else
                   2015: void *
                   2016: pool_page_alloc_nointr(struct pool *pp, int flags)
                   2017: {
                   2018:        boolean_t waitok = (flags & PR_WAITOK) ? TRUE : FALSE;
                   2019:
                   2020:        return ((void *) uvm_km_alloc_poolpage1(kernel_map,
                   2021:            uvm.kernel_object, waitok));
                   2022: }
                   2023:
                   2024: void
                   2025: pool_page_free_nointr(struct pool *pp, void *v)
                   2026: {
                   2027:
                   2028:        uvm_km_free_poolpage1(kernel_map, (vaddr_t) v);
                   2029: }
                   2030: #endif /* POOL_SUBPAGE */

CVSweb <webmaster@jp.NetBSD.org>