[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 1.129

1.129   ! ad          1: /*     $NetBSD: subr_pool.c,v 1.128 2007/03/04 06:03:07 christos Exp $ */
1.1       pk          2:
                      3: /*-
1.125     ad          4:  * Copyright (c) 1997, 1999, 2000, 2002 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>
1.129   ! ad         41: __KERNEL_RCSID(0, "$NetBSD: subr_pool.c,v 1.128 2007/03/04 06:03:07 christos Exp $");
1.24      scottr     42:
1.25      thorpej    43: #include "opt_pool.h"
1.24      scottr     44: #include "opt_poollog.h"
1.28      thorpej    45: #include "opt_lockdebug.h"
1.1       pk         46:
                     47: #include <sys/param.h>
                     48: #include <sys/systm.h>
                     49: #include <sys/proc.h>
                     50: #include <sys/errno.h>
                     51: #include <sys/kernel.h>
                     52: #include <sys/malloc.h>
                     53: #include <sys/lock.h>
                     54: #include <sys/pool.h>
1.20      thorpej    55: #include <sys/syslog.h>
1.125     ad         56: #include <sys/debug.h>
1.3       pk         57:
                     58: #include <uvm/uvm.h>
                     59:
1.1       pk         60: /*
                     61:  * Pool resource management utility.
1.3       pk         62:  *
1.88      chs        63:  * Memory is allocated in pages which are split into pieces according to
                     64:  * the pool item size. Each page is kept on one of three lists in the
                     65:  * pool structure: `pr_emptypages', `pr_fullpages' and `pr_partpages',
                     66:  * for empty, full and partially-full pages respectively. The individual
                     67:  * pool items are on a linked list headed by `ph_itemlist' in each page
                     68:  * header. The memory for building the page list is either taken from
                     69:  * the allocated pages themselves (for small pool items) or taken from
                     70:  * an internal pool of page headers (`phpool').
1.1       pk         71:  */
                     72:
1.3       pk         73: /* List of all pools */
1.102     chs        74: LIST_HEAD(,pool) pool_head = LIST_HEAD_INITIALIZER(pool_head);
1.3       pk         75:
                     76: /* Private pool for page header structures */
1.97      yamt       77: #define        PHPOOL_MAX      8
                     78: static struct pool phpool[PHPOOL_MAX];
                     79: #define        PHPOOL_FREELIST_NELEM(idx)      (((idx) == 0) ? 0 : (1 << (idx)))
1.3       pk         80:
1.62      bjh21      81: #ifdef POOL_SUBPAGE
                     82: /* Pool of subpages for use by normal pools. */
                     83: static struct pool psppool;
                     84: #endif
                     85:
1.117     yamt       86: static SLIST_HEAD(, pool_allocator) pa_deferinitq =
                     87:     SLIST_HEAD_INITIALIZER(pa_deferinitq);
                     88:
1.98      yamt       89: static void *pool_page_alloc_meta(struct pool *, int);
                     90: static void pool_page_free_meta(struct pool *, void *);
                     91:
                     92: /* allocator for pool metadata */
                     93: static struct pool_allocator pool_allocator_meta = {
1.117     yamt       94:        pool_page_alloc_meta, pool_page_free_meta,
                     95:        .pa_backingmapptr = &kmem_map,
1.98      yamt       96: };
                     97:
1.3       pk         98: /* # of seconds to retain page after last use */
                     99: int pool_inactive_time = 10;
                    100:
                    101: /* Next candidate for drainage (see pool_drain()) */
1.23      thorpej   102: static struct pool     *drainpp;
                    103:
                    104: /* This spin lock protects both pool_head and drainpp. */
                    105: struct simplelock pool_head_slock = SIMPLELOCK_INITIALIZER;
1.3       pk        106:
1.99      yamt      107: typedef uint8_t pool_item_freelist_t;
                    108:
1.3       pk        109: struct pool_item_header {
                    110:        /* Page headers */
1.88      chs       111:        LIST_ENTRY(pool_item_header)
1.3       pk        112:                                ph_pagelist;    /* pool page list */
1.88      chs       113:        SPLAY_ENTRY(pool_item_header)
                    114:                                ph_node;        /* Off-page page headers */
1.128     christos  115:        void *                  ph_page;        /* this page's address */
1.3       pk        116:        struct timeval          ph_time;        /* last referenced */
1.97      yamt      117:        union {
                    118:                /* !PR_NOTOUCH */
                    119:                struct {
1.102     chs       120:                        LIST_HEAD(, pool_item)
1.97      yamt      121:                                phu_itemlist;   /* chunk list for this page */
                    122:                } phu_normal;
                    123:                /* PR_NOTOUCH */
                    124:                struct {
                    125:                        uint16_t
                    126:                                phu_off;        /* start offset in page */
1.99      yamt      127:                        pool_item_freelist_t
1.97      yamt      128:                                phu_firstfree;  /* first free item */
1.99      yamt      129:                        /*
                    130:                         * XXX it might be better to use
                    131:                         * a simple bitmap and ffs(3)
                    132:                         */
1.97      yamt      133:                } phu_notouch;
                    134:        } ph_u;
                    135:        uint16_t                ph_nmissing;    /* # of chunks in use */
1.3       pk        136: };
1.97      yamt      137: #define        ph_itemlist     ph_u.phu_normal.phu_itemlist
                    138: #define        ph_off          ph_u.phu_notouch.phu_off
                    139: #define        ph_firstfree    ph_u.phu_notouch.phu_firstfree
1.3       pk        140:
1.1       pk        141: struct pool_item {
1.3       pk        142: #ifdef DIAGNOSTIC
1.82      thorpej   143:        u_int pi_magic;
1.33      chs       144: #endif
1.82      thorpej   145: #define        PI_MAGIC 0xdeadbeefU
1.3       pk        146:        /* Other entries use only this list entry */
1.102     chs       147:        LIST_ENTRY(pool_item)   pi_list;
1.3       pk        148: };
                    149:
1.53      thorpej   150: #define        POOL_NEEDS_CATCHUP(pp)                                          \
                    151:        ((pp)->pr_nitems < (pp)->pr_minitems)
                    152:
1.43      thorpej   153: /*
                    154:  * Pool cache management.
                    155:  *
                    156:  * Pool caches provide a way for constructed objects to be cached by the
                    157:  * pool subsystem.  This can lead to performance improvements by avoiding
                    158:  * needless object construction/destruction; it is deferred until absolutely
                    159:  * necessary.
                    160:  *
                    161:  * Caches are grouped into cache groups.  Each cache group references
                    162:  * up to 16 constructed objects.  When a cache allocates an object
                    163:  * from the pool, it calls the object's constructor and places it into
                    164:  * a cache group.  When a cache group frees an object back to the pool,
                    165:  * it first calls the object's destructor.  This allows the object to
                    166:  * persist in constructed form while freed to the cache.
                    167:  *
                    168:  * Multiple caches may exist for each pool.  This allows a single
                    169:  * object type to have multiple constructed forms.  The pool references
                    170:  * each cache, so that when a pool is drained by the pagedaemon, it can
                    171:  * drain each individual cache as well.  Each time a cache is drained,
                    172:  * the most idle cache group is freed to the pool in its entirety.
                    173:  *
                    174:  * Pool caches are layed on top of pools.  By layering them, we can avoid
                    175:  * the complexity of cache management for pools which would not benefit
                    176:  * from it.
                    177:  */
                    178:
                    179: /* The cache group pool. */
                    180: static struct pool pcgpool;
1.3       pk        181:
1.102     chs       182: static void    pool_cache_reclaim(struct pool_cache *, struct pool_pagelist *,
                    183:                                   struct pool_cache_grouplist *);
                    184: static void    pcg_grouplist_free(struct pool_cache_grouplist *);
1.3       pk        185:
1.42      thorpej   186: static int     pool_catchup(struct pool *);
1.128     christos  187: static void    pool_prime_page(struct pool *, void *,
1.55      thorpej   188:                    struct pool_item_header *);
1.88      chs       189: static void    pool_update_curpage(struct pool *);
1.66      thorpej   190:
1.113     yamt      191: static int     pool_grow(struct pool *, int);
1.117     yamt      192: static void    *pool_allocator_alloc(struct pool *, int);
                    193: static void    pool_allocator_free(struct pool *, void *);
1.3       pk        194:
1.97      yamt      195: static void pool_print_pagelist(struct pool *, struct pool_pagelist *,
1.88      chs       196:        void (*)(const char *, ...));
1.42      thorpej   197: static void pool_print1(struct pool *, const char *,
                    198:        void (*)(const char *, ...));
1.3       pk        199:
1.88      chs       200: static int pool_chk_page(struct pool *, const char *,
                    201:                         struct pool_item_header *);
                    202:
1.3       pk        203: /*
1.52      thorpej   204:  * Pool log entry. An array of these is allocated in pool_init().
1.3       pk        205:  */
                    206: struct pool_log {
                    207:        const char      *pl_file;
                    208:        long            pl_line;
                    209:        int             pl_action;
1.25      thorpej   210: #define        PRLOG_GET       1
                    211: #define        PRLOG_PUT       2
1.3       pk        212:        void            *pl_addr;
1.1       pk        213: };
                    214:
1.86      matt      215: #ifdef POOL_DIAGNOSTIC
1.3       pk        216: /* Number of entries in pool log buffers */
1.17      thorpej   217: #ifndef POOL_LOGSIZE
                    218: #define        POOL_LOGSIZE    10
                    219: #endif
                    220:
                    221: int pool_logsize = POOL_LOGSIZE;
1.1       pk        222:
1.110     perry     223: static inline void
1.42      thorpej   224: pr_log(struct pool *pp, void *v, int action, const char *file, long line)
1.3       pk        225: {
                    226:        int n = pp->pr_curlogentry;
                    227:        struct pool_log *pl;
                    228:
1.20      thorpej   229:        if ((pp->pr_roflags & PR_LOGGING) == 0)
1.3       pk        230:                return;
                    231:
                    232:        /*
                    233:         * Fill in the current entry. Wrap around and overwrite
                    234:         * the oldest entry if necessary.
                    235:         */
                    236:        pl = &pp->pr_log[n];
                    237:        pl->pl_file = file;
                    238:        pl->pl_line = line;
                    239:        pl->pl_action = action;
                    240:        pl->pl_addr = v;
                    241:        if (++n >= pp->pr_logsize)
                    242:                n = 0;
                    243:        pp->pr_curlogentry = n;
                    244: }
                    245:
                    246: static void
1.42      thorpej   247: pr_printlog(struct pool *pp, struct pool_item *pi,
                    248:     void (*pr)(const char *, ...))
1.3       pk        249: {
                    250:        int i = pp->pr_logsize;
                    251:        int n = pp->pr_curlogentry;
                    252:
1.20      thorpej   253:        if ((pp->pr_roflags & PR_LOGGING) == 0)
1.3       pk        254:                return;
                    255:
                    256:        /*
                    257:         * Print all entries in this pool's log.
                    258:         */
                    259:        while (i-- > 0) {
                    260:                struct pool_log *pl = &pp->pr_log[n];
                    261:                if (pl->pl_action != 0) {
1.25      thorpej   262:                        if (pi == NULL || pi == pl->pl_addr) {
                    263:                                (*pr)("\tlog entry %d:\n", i);
                    264:                                (*pr)("\t\taction = %s, addr = %p\n",
                    265:                                    pl->pl_action == PRLOG_GET ? "get" : "put",
                    266:                                    pl->pl_addr);
                    267:                                (*pr)("\t\tfile: %s at line %lu\n",
                    268:                                    pl->pl_file, pl->pl_line);
                    269:                        }
1.3       pk        270:                }
                    271:                if (++n >= pp->pr_logsize)
                    272:                        n = 0;
                    273:        }
                    274: }
1.25      thorpej   275:
1.110     perry     276: static inline void
1.42      thorpej   277: pr_enter(struct pool *pp, const char *file, long line)
1.25      thorpej   278: {
                    279:
1.34      thorpej   280:        if (__predict_false(pp->pr_entered_file != NULL)) {
1.25      thorpej   281:                printf("pool %s: reentrancy at file %s line %ld\n",
                    282:                    pp->pr_wchan, file, line);
                    283:                printf("         previous entry at file %s line %ld\n",
                    284:                    pp->pr_entered_file, pp->pr_entered_line);
                    285:                panic("pr_enter");
                    286:        }
                    287:
                    288:        pp->pr_entered_file = file;
                    289:        pp->pr_entered_line = line;
                    290: }
                    291:
1.110     perry     292: static inline void
1.42      thorpej   293: pr_leave(struct pool *pp)
1.25      thorpej   294: {
                    295:
1.34      thorpej   296:        if (__predict_false(pp->pr_entered_file == NULL)) {
1.25      thorpej   297:                printf("pool %s not entered?\n", pp->pr_wchan);
                    298:                panic("pr_leave");
                    299:        }
                    300:
                    301:        pp->pr_entered_file = NULL;
                    302:        pp->pr_entered_line = 0;
                    303: }
                    304:
1.110     perry     305: static inline void
1.42      thorpej   306: pr_enter_check(struct pool *pp, void (*pr)(const char *, ...))
1.25      thorpej   307: {
                    308:
                    309:        if (pp->pr_entered_file != NULL)
                    310:                (*pr)("\n\tcurrently entered from file %s line %ld\n",
                    311:                    pp->pr_entered_file, pp->pr_entered_line);
                    312: }
1.3       pk        313: #else
1.25      thorpej   314: #define        pr_log(pp, v, action, file, line)
                    315: #define        pr_printlog(pp, pi, pr)
                    316: #define        pr_enter(pp, file, line)
                    317: #define        pr_leave(pp)
                    318: #define        pr_enter_check(pp, pr)
1.59      thorpej   319: #endif /* POOL_DIAGNOSTIC */
1.3       pk        320:
1.110     perry     321: static inline int
1.97      yamt      322: pr_item_notouch_index(const struct pool *pp, const struct pool_item_header *ph,
                    323:     const void *v)
                    324: {
                    325:        const char *cp = v;
                    326:        int idx;
                    327:
                    328:        KASSERT(pp->pr_roflags & PR_NOTOUCH);
1.128     christos  329:        idx = (cp - (char *)ph->ph_page - ph->ph_off) / pp->pr_size;
1.97      yamt      330:        KASSERT(idx < pp->pr_itemsperpage);
                    331:        return idx;
                    332: }
                    333:
1.99      yamt      334: #define        PR_FREELIST_ALIGN(p) \
                    335:        roundup((uintptr_t)(p), sizeof(pool_item_freelist_t))
                    336: #define        PR_FREELIST(ph) ((pool_item_freelist_t *)PR_FREELIST_ALIGN((ph) + 1))
                    337: #define        PR_INDEX_USED   ((pool_item_freelist_t)-1)
                    338: #define        PR_INDEX_EOL    ((pool_item_freelist_t)-2)
1.97      yamt      339:
1.110     perry     340: static inline void
1.97      yamt      341: pr_item_notouch_put(const struct pool *pp, struct pool_item_header *ph,
                    342:     void *obj)
                    343: {
                    344:        int idx = pr_item_notouch_index(pp, ph, obj);
1.99      yamt      345:        pool_item_freelist_t *freelist = PR_FREELIST(ph);
1.97      yamt      346:
                    347:        KASSERT(freelist[idx] == PR_INDEX_USED);
                    348:        freelist[idx] = ph->ph_firstfree;
                    349:        ph->ph_firstfree = idx;
                    350: }
                    351:
1.110     perry     352: static inline void *
1.97      yamt      353: pr_item_notouch_get(const struct pool *pp, struct pool_item_header *ph)
                    354: {
                    355:        int idx = ph->ph_firstfree;
1.99      yamt      356:        pool_item_freelist_t *freelist = PR_FREELIST(ph);
1.97      yamt      357:
                    358:        KASSERT(freelist[idx] != PR_INDEX_USED);
                    359:        ph->ph_firstfree = freelist[idx];
                    360:        freelist[idx] = PR_INDEX_USED;
                    361:
1.128     christos  362:        return (char *)ph->ph_page + ph->ph_off + idx * pp->pr_size;
1.97      yamt      363: }
                    364:
1.110     perry     365: static inline int
1.88      chs       366: phtree_compare(struct pool_item_header *a, struct pool_item_header *b)
                    367: {
1.121     yamt      368:
                    369:        /*
                    370:         * we consider pool_item_header with smaller ph_page bigger.
                    371:         * (this unnatural ordering is for the benefit of pr_find_pagehead.)
                    372:         */
                    373:
1.88      chs       374:        if (a->ph_page < b->ph_page)
1.121     yamt      375:                return (1);
                    376:        else if (a->ph_page > b->ph_page)
1.88      chs       377:                return (-1);
                    378:        else
                    379:                return (0);
                    380: }
                    381:
                    382: SPLAY_PROTOTYPE(phtree, pool_item_header, ph_node, phtree_compare);
                    383: SPLAY_GENERATE(phtree, pool_item_header, ph_node, phtree_compare);
                    384:
1.3       pk        385: /*
1.121     yamt      386:  * Return the pool page header based on item address.
1.3       pk        387:  */
1.110     perry     388: static inline struct pool_item_header *
1.121     yamt      389: pr_find_pagehead(struct pool *pp, void *v)
1.3       pk        390: {
1.88      chs       391:        struct pool_item_header *ph, tmp;
1.3       pk        392:
1.121     yamt      393:        if ((pp->pr_roflags & PR_NOALIGN) != 0) {
1.128     christos  394:                tmp.ph_page = (void *)(uintptr_t)v;
1.121     yamt      395:                ph = SPLAY_FIND(phtree, &pp->pr_phtree, &tmp);
                    396:                if (ph == NULL) {
                    397:                        ph = SPLAY_ROOT(&pp->pr_phtree);
                    398:                        if (ph != NULL && phtree_compare(&tmp, ph) >= 0) {
                    399:                                ph = SPLAY_NEXT(phtree, &pp->pr_phtree, ph);
                    400:                        }
                    401:                        KASSERT(ph == NULL || phtree_compare(&tmp, ph) < 0);
                    402:                }
                    403:        } else {
1.128     christos  404:                void *page =
                    405:                    (void *)((uintptr_t)v & pp->pr_alloc->pa_pagemask);
1.121     yamt      406:
                    407:                if ((pp->pr_roflags & PR_PHINPAGE) != 0) {
1.128     christos  408:                        ph = (struct pool_item_header *)((char *)page + pp->pr_phoffset);
1.121     yamt      409:                } else {
                    410:                        tmp.ph_page = page;
                    411:                        ph = SPLAY_FIND(phtree, &pp->pr_phtree, &tmp);
                    412:                }
                    413:        }
1.3       pk        414:
1.121     yamt      415:        KASSERT(ph == NULL || ((pp->pr_roflags & PR_PHINPAGE) != 0) ||
1.128     christos  416:            ((char *)ph->ph_page <= (char *)v &&
                    417:            (char *)v < (char *)ph->ph_page + pp->pr_alloc->pa_pagesz));
1.88      chs       418:        return ph;
1.3       pk        419: }
                    420:
1.101     thorpej   421: static void
                    422: pr_pagelist_free(struct pool *pp, struct pool_pagelist *pq)
                    423: {
                    424:        struct pool_item_header *ph;
                    425:        int s;
                    426:
                    427:        while ((ph = LIST_FIRST(pq)) != NULL) {
                    428:                LIST_REMOVE(ph, ph_pagelist);
                    429:                pool_allocator_free(pp, ph->ph_page);
                    430:                if ((pp->pr_roflags & PR_PHINPAGE) == 0) {
                    431:                        s = splvm();
                    432:                        pool_put(pp->pr_phpool, ph);
                    433:                        splx(s);
                    434:                }
                    435:        }
                    436: }
                    437:
1.3       pk        438: /*
                    439:  * Remove a page from the pool.
                    440:  */
1.110     perry     441: static inline void
1.61      chs       442: pr_rmpage(struct pool *pp, struct pool_item_header *ph,
                    443:      struct pool_pagelist *pq)
1.3       pk        444: {
                    445:
1.101     thorpej   446:        LOCK_ASSERT(simple_lock_held(&pp->pr_slock));
1.91      yamt      447:
1.3       pk        448:        /*
1.7       thorpej   449:         * If the page was idle, decrement the idle page count.
1.3       pk        450:         */
1.6       thorpej   451:        if (ph->ph_nmissing == 0) {
                    452: #ifdef DIAGNOSTIC
                    453:                if (pp->pr_nidle == 0)
                    454:                        panic("pr_rmpage: nidle inconsistent");
1.20      thorpej   455:                if (pp->pr_nitems < pp->pr_itemsperpage)
                    456:                        panic("pr_rmpage: nitems inconsistent");
1.6       thorpej   457: #endif
                    458:                pp->pr_nidle--;
                    459:        }
1.7       thorpej   460:
1.20      thorpej   461:        pp->pr_nitems -= pp->pr_itemsperpage;
                    462:
1.7       thorpej   463:        /*
1.101     thorpej   464:         * Unlink the page from the pool and queue it for release.
1.7       thorpej   465:         */
1.88      chs       466:        LIST_REMOVE(ph, ph_pagelist);
1.91      yamt      467:        if ((pp->pr_roflags & PR_PHINPAGE) == 0)
                    468:                SPLAY_REMOVE(phtree, &pp->pr_phtree, ph);
1.101     thorpej   469:        LIST_INSERT_HEAD(pq, ph, ph_pagelist);
                    470:
1.7       thorpej   471:        pp->pr_npages--;
                    472:        pp->pr_npagefree++;
1.6       thorpej   473:
1.88      chs       474:        pool_update_curpage(pp);
1.3       pk        475: }
                    476:
1.126     thorpej   477: static bool
1.117     yamt      478: pa_starved_p(struct pool_allocator *pa)
                    479: {
                    480:
                    481:        if (pa->pa_backingmap != NULL) {
                    482:                return vm_map_starved_p(pa->pa_backingmap);
                    483:        }
1.127     thorpej   484:        return false;
1.117     yamt      485: }
                    486:
                    487: static int
1.124     yamt      488: pool_reclaim_callback(struct callback_entry *ce, void *obj, void *arg)
1.117     yamt      489: {
                    490:        struct pool *pp = obj;
                    491:        struct pool_allocator *pa = pp->pr_alloc;
                    492:
                    493:        KASSERT(&pp->pr_reclaimerentry == ce);
                    494:        pool_reclaim(pp);
                    495:        if (!pa_starved_p(pa)) {
                    496:                return CALLBACK_CHAIN_ABORT;
                    497:        }
                    498:        return CALLBACK_CHAIN_CONTINUE;
                    499: }
                    500:
                    501: static void
                    502: pool_reclaim_register(struct pool *pp)
                    503: {
                    504:        struct vm_map *map = pp->pr_alloc->pa_backingmap;
                    505:        int s;
                    506:
                    507:        if (map == NULL) {
                    508:                return;
                    509:        }
                    510:
                    511:        s = splvm(); /* not necessary for INTRSAFE maps, but don't care. */
                    512:        callback_register(&vm_map_to_kernel(map)->vmk_reclaim_callback,
                    513:            &pp->pr_reclaimerentry, pp, pool_reclaim_callback);
                    514:        splx(s);
                    515: }
                    516:
                    517: static void
                    518: pool_reclaim_unregister(struct pool *pp)
                    519: {
                    520:        struct vm_map *map = pp->pr_alloc->pa_backingmap;
                    521:        int s;
                    522:
                    523:        if (map == NULL) {
                    524:                return;
                    525:        }
                    526:
                    527:        s = splvm(); /* not necessary for INTRSAFE maps, but don't care. */
                    528:        callback_unregister(&vm_map_to_kernel(map)->vmk_reclaim_callback,
                    529:            &pp->pr_reclaimerentry);
                    530:        splx(s);
                    531: }
                    532:
                    533: static void
                    534: pa_reclaim_register(struct pool_allocator *pa)
                    535: {
                    536:        struct vm_map *map = *pa->pa_backingmapptr;
                    537:        struct pool *pp;
                    538:
                    539:        KASSERT(pa->pa_backingmap == NULL);
                    540:        if (map == NULL) {
                    541:                SLIST_INSERT_HEAD(&pa_deferinitq, pa, pa_q);
                    542:                return;
                    543:        }
                    544:        pa->pa_backingmap = map;
                    545:        TAILQ_FOREACH(pp, &pa->pa_list, pr_alloc_list) {
                    546:                pool_reclaim_register(pp);
                    547:        }
                    548: }
                    549:
1.3       pk        550: /*
1.94      simonb    551:  * Initialize all the pools listed in the "pools" link set.
                    552:  */
                    553: void
1.117     yamt      554: pool_subsystem_init(void)
1.94      simonb    555: {
1.117     yamt      556:        struct pool_allocator *pa;
1.94      simonb    557:        __link_set_decl(pools, struct link_pool_init);
                    558:        struct link_pool_init * const *pi;
                    559:
                    560:        __link_set_foreach(pi, pools)
                    561:                pool_init((*pi)->pp, (*pi)->size, (*pi)->align,
                    562:                    (*pi)->align_offset, (*pi)->flags, (*pi)->wchan,
1.129   ! ad        563:                    (*pi)->palloc, (*pi)->ipl);
1.117     yamt      564:
                    565:        while ((pa = SLIST_FIRST(&pa_deferinitq)) != NULL) {
                    566:                KASSERT(pa->pa_backingmapptr != NULL);
                    567:                KASSERT(*pa->pa_backingmapptr != NULL);
                    568:                SLIST_REMOVE_HEAD(&pa_deferinitq, pa_q);
                    569:                pa_reclaim_register(pa);
                    570:        }
1.94      simonb    571: }
                    572:
                    573: /*
1.3       pk        574:  * Initialize the given pool resource structure.
                    575:  *
                    576:  * We export this routine to allow other kernel parts to declare
                    577:  * static pools that must be initialized before malloc() is available.
                    578:  */
                    579: void
1.42      thorpej   580: pool_init(struct pool *pp, size_t size, u_int align, u_int ioff, int flags,
1.129   ! ad        581:     const char *wchan, struct pool_allocator *palloc, int ipl)
1.3       pk        582: {
1.116     simonb    583: #ifdef DEBUG
                    584:        struct pool *pp1;
                    585: #endif
1.92      enami     586:        size_t trysize, phsize;
1.116     simonb    587:        int off, slack, s;
1.3       pk        588:
1.99      yamt      589:        KASSERT((1UL << (CHAR_BIT * sizeof(pool_item_freelist_t))) - 2 >=
                    590:            PHPOOL_FREELIST_NELEM(PHPOOL_MAX - 1));
                    591:
1.116     simonb    592: #ifdef DEBUG
                    593:        /*
                    594:         * Check that the pool hasn't already been initialised and
                    595:         * added to the list of all pools.
                    596:         */
                    597:        LIST_FOREACH(pp1, &pool_head, pr_poollist) {
                    598:                if (pp == pp1)
                    599:                        panic("pool_init: pool %s already initialised",
                    600:                            wchan);
                    601:        }
                    602: #endif
                    603:
1.25      thorpej   604: #ifdef POOL_DIAGNOSTIC
                    605:        /*
                    606:         * Always log if POOL_DIAGNOSTIC is defined.
                    607:         */
                    608:        if (pool_logsize != 0)
                    609:                flags |= PR_LOGGING;
                    610: #endif
                    611:
1.66      thorpej   612:        if (palloc == NULL)
                    613:                palloc = &pool_allocator_kmem;
1.112     bjh21     614: #ifdef POOL_SUBPAGE
                    615:        if (size > palloc->pa_pagesz) {
                    616:                if (palloc == &pool_allocator_kmem)
                    617:                        palloc = &pool_allocator_kmem_fullpage;
                    618:                else if (palloc == &pool_allocator_nointr)
                    619:                        palloc = &pool_allocator_nointr_fullpage;
                    620:        }
1.66      thorpej   621: #endif /* POOL_SUBPAGE */
                    622:        if ((palloc->pa_flags & PA_INITIALIZED) == 0) {
1.112     bjh21     623:                if (palloc->pa_pagesz == 0)
1.66      thorpej   624:                        palloc->pa_pagesz = PAGE_SIZE;
                    625:
                    626:                TAILQ_INIT(&palloc->pa_list);
                    627:
                    628:                simple_lock_init(&palloc->pa_slock);
                    629:                palloc->pa_pagemask = ~(palloc->pa_pagesz - 1);
                    630:                palloc->pa_pageshift = ffs(palloc->pa_pagesz) - 1;
1.117     yamt      631:
                    632:                if (palloc->pa_backingmapptr != NULL) {
                    633:                        pa_reclaim_register(palloc);
                    634:                }
1.66      thorpej   635:                palloc->pa_flags |= PA_INITIALIZED;
1.4       thorpej   636:        }
1.3       pk        637:
                    638:        if (align == 0)
                    639:                align = ALIGN(1);
1.14      thorpej   640:
1.120     yamt      641:        if ((flags & PR_NOTOUCH) == 0 && size < sizeof(struct pool_item))
1.14      thorpej   642:                size = sizeof(struct pool_item);
1.3       pk        643:
1.78      thorpej   644:        size = roundup(size, align);
1.66      thorpej   645: #ifdef DIAGNOSTIC
                    646:        if (size > palloc->pa_pagesz)
1.121     yamt      647:                panic("pool_init: pool item size (%zu) too large", size);
1.66      thorpej   648: #endif
1.35      pk        649:
1.3       pk        650:        /*
                    651:         * Initialize the pool structure.
                    652:         */
1.88      chs       653:        LIST_INIT(&pp->pr_emptypages);
                    654:        LIST_INIT(&pp->pr_fullpages);
                    655:        LIST_INIT(&pp->pr_partpages);
1.102     chs       656:        LIST_INIT(&pp->pr_cachelist);
1.3       pk        657:        pp->pr_curpage = NULL;
                    658:        pp->pr_npages = 0;
                    659:        pp->pr_minitems = 0;
                    660:        pp->pr_minpages = 0;
                    661:        pp->pr_maxpages = UINT_MAX;
1.20      thorpej   662:        pp->pr_roflags = flags;
                    663:        pp->pr_flags = 0;
1.35      pk        664:        pp->pr_size = size;
1.3       pk        665:        pp->pr_align = align;
                    666:        pp->pr_wchan = wchan;
1.66      thorpej   667:        pp->pr_alloc = palloc;
1.20      thorpej   668:        pp->pr_nitems = 0;
                    669:        pp->pr_nout = 0;
                    670:        pp->pr_hardlimit = UINT_MAX;
                    671:        pp->pr_hardlimit_warning = NULL;
1.31      thorpej   672:        pp->pr_hardlimit_ratecap.tv_sec = 0;
                    673:        pp->pr_hardlimit_ratecap.tv_usec = 0;
                    674:        pp->pr_hardlimit_warning_last.tv_sec = 0;
                    675:        pp->pr_hardlimit_warning_last.tv_usec = 0;
1.68      thorpej   676:        pp->pr_drain_hook = NULL;
                    677:        pp->pr_drain_hook_arg = NULL;
1.125     ad        678:        pp->pr_freecheck = NULL;
1.3       pk        679:
                    680:        /*
                    681:         * Decide whether to put the page header off page to avoid
1.92      enami     682:         * wasting too large a part of the page or too big item.
                    683:         * Off-page page headers go on a hash table, so we can match
                    684:         * a returned item with its header based on the page address.
                    685:         * We use 1/16 of the page size and about 8 times of the item
                    686:         * size as the threshold (XXX: tune)
                    687:         *
                    688:         * However, we'll put the header into the page if we can put
                    689:         * it without wasting any items.
                    690:         *
                    691:         * Silently enforce `0 <= ioff < align'.
1.3       pk        692:         */
1.92      enami     693:        pp->pr_itemoffset = ioff %= align;
                    694:        /* See the comment below about reserved bytes. */
                    695:        trysize = palloc->pa_pagesz - ((align - ioff) % align);
                    696:        phsize = ALIGN(sizeof(struct pool_item_header));
1.121     yamt      697:        if ((pp->pr_roflags & (PR_NOTOUCH | PR_NOALIGN)) == 0 &&
1.97      yamt      698:            (pp->pr_size < MIN(palloc->pa_pagesz / 16, phsize << 3) ||
                    699:            trysize / pp->pr_size == (trysize - phsize) / pp->pr_size)) {
1.3       pk        700:                /* Use the end of the page for the page header */
1.20      thorpej   701:                pp->pr_roflags |= PR_PHINPAGE;
1.92      enami     702:                pp->pr_phoffset = off = palloc->pa_pagesz - phsize;
1.2       pk        703:        } else {
1.3       pk        704:                /* The page header will be taken from our page header pool */
                    705:                pp->pr_phoffset = 0;
1.66      thorpej   706:                off = palloc->pa_pagesz;
1.88      chs       707:                SPLAY_INIT(&pp->pr_phtree);
1.2       pk        708:        }
1.1       pk        709:
1.3       pk        710:        /*
                    711:         * Alignment is to take place at `ioff' within the item. This means
                    712:         * we must reserve up to `align - 1' bytes on the page to allow
                    713:         * appropriate positioning of each item.
                    714:         */
                    715:        pp->pr_itemsperpage = (off - ((align - ioff) % align)) / pp->pr_size;
1.43      thorpej   716:        KASSERT(pp->pr_itemsperpage != 0);
1.97      yamt      717:        if ((pp->pr_roflags & PR_NOTOUCH)) {
                    718:                int idx;
                    719:
                    720:                for (idx = 0; pp->pr_itemsperpage > PHPOOL_FREELIST_NELEM(idx);
                    721:                    idx++) {
                    722:                        /* nothing */
                    723:                }
                    724:                if (idx >= PHPOOL_MAX) {
                    725:                        /*
                    726:                         * if you see this panic, consider to tweak
                    727:                         * PHPOOL_MAX and PHPOOL_FREELIST_NELEM.
                    728:                         */
                    729:                        panic("%s: too large itemsperpage(%d) for PR_NOTOUCH",
                    730:                            pp->pr_wchan, pp->pr_itemsperpage);
                    731:                }
                    732:                pp->pr_phpool = &phpool[idx];
                    733:        } else if ((pp->pr_roflags & PR_PHINPAGE) == 0) {
                    734:                pp->pr_phpool = &phpool[0];
                    735:        }
                    736: #if defined(DIAGNOSTIC)
                    737:        else {
                    738:                pp->pr_phpool = NULL;
                    739:        }
                    740: #endif
1.3       pk        741:
                    742:        /*
                    743:         * Use the slack between the chunks and the page header
                    744:         * for "cache coloring".
                    745:         */
                    746:        slack = off - pp->pr_itemsperpage * pp->pr_size;
                    747:        pp->pr_maxcolor = (slack / align) * align;
                    748:        pp->pr_curcolor = 0;
                    749:
                    750:        pp->pr_nget = 0;
                    751:        pp->pr_nfail = 0;
                    752:        pp->pr_nput = 0;
                    753:        pp->pr_npagealloc = 0;
                    754:        pp->pr_npagefree = 0;
1.1       pk        755:        pp->pr_hiwat = 0;
1.8       thorpej   756:        pp->pr_nidle = 0;
1.3       pk        757:
1.59      thorpej   758: #ifdef POOL_DIAGNOSTIC
1.25      thorpej   759:        if (flags & PR_LOGGING) {
                    760:                if (kmem_map == NULL ||
                    761:                    (pp->pr_log = malloc(pool_logsize * sizeof(struct pool_log),
                    762:                     M_TEMP, M_NOWAIT)) == NULL)
1.20      thorpej   763:                        pp->pr_roflags &= ~PR_LOGGING;
1.3       pk        764:                pp->pr_curlogentry = 0;
                    765:                pp->pr_logsize = pool_logsize;
                    766:        }
1.59      thorpej   767: #endif
1.25      thorpej   768:
                    769:        pp->pr_entered_file = NULL;
                    770:        pp->pr_entered_line = 0;
1.3       pk        771:
1.21      thorpej   772:        simple_lock_init(&pp->pr_slock);
1.1       pk        773:
1.3       pk        774:        /*
1.43      thorpej   775:         * Initialize private page header pool and cache magazine pool if we
                    776:         * haven't done so yet.
1.23      thorpej   777:         * XXX LOCKING.
1.3       pk        778:         */
1.97      yamt      779:        if (phpool[0].pr_size == 0) {
                    780:                int idx;
                    781:                for (idx = 0; idx < PHPOOL_MAX; idx++) {
                    782:                        static char phpool_names[PHPOOL_MAX][6+1+6+1];
                    783:                        int nelem;
                    784:                        size_t sz;
                    785:
                    786:                        nelem = PHPOOL_FREELIST_NELEM(idx);
                    787:                        snprintf(phpool_names[idx], sizeof(phpool_names[idx]),
                    788:                            "phpool-%d", nelem);
                    789:                        sz = sizeof(struct pool_item_header);
                    790:                        if (nelem) {
                    791:                                sz = PR_FREELIST_ALIGN(sz)
1.99      yamt      792:                                    + nelem * sizeof(pool_item_freelist_t);
1.97      yamt      793:                        }
                    794:                        pool_init(&phpool[idx], sz, 0, 0, 0,
1.129   ! ad        795:                            phpool_names[idx], &pool_allocator_meta, IPL_VM);
1.97      yamt      796:                }
1.62      bjh21     797: #ifdef POOL_SUBPAGE
                    798:                pool_init(&psppool, POOL_SUBPAGE, POOL_SUBPAGE, 0,
1.129   ! ad        799:                    PR_RECURSIVE, "psppool", &pool_allocator_meta, IPL_VM);
1.62      bjh21     800: #endif
1.43      thorpej   801:                pool_init(&pcgpool, sizeof(struct pool_cache_group), 0, 0,
1.129   ! ad        802:                    0, "pcgpool", &pool_allocator_meta, IPL_VM);
1.1       pk        803:        }
                    804:
1.23      thorpej   805:        /* Insert into the list of all pools. */
                    806:        simple_lock(&pool_head_slock);
1.102     chs       807:        LIST_INSERT_HEAD(&pool_head, pp, pr_poollist);
1.23      thorpej   808:        simple_unlock(&pool_head_slock);
1.66      thorpej   809:
                    810:        /* Insert this into the list of pools using this allocator. */
1.93      dbj       811:        s = splvm();
1.66      thorpej   812:        simple_lock(&palloc->pa_slock);
                    813:        TAILQ_INSERT_TAIL(&palloc->pa_list, pp, pr_alloc_list);
                    814:        simple_unlock(&palloc->pa_slock);
1.93      dbj       815:        splx(s);
1.117     yamt      816:        pool_reclaim_register(pp);
1.1       pk        817: }
                    818:
                    819: /*
                    820:  * De-commision a pool resource.
                    821:  */
                    822: void
1.42      thorpej   823: pool_destroy(struct pool *pp)
1.1       pk        824: {
1.101     thorpej   825:        struct pool_pagelist pq;
1.3       pk        826:        struct pool_item_header *ph;
1.93      dbj       827:        int s;
1.43      thorpej   828:
1.101     thorpej   829:        /* Remove from global pool list */
                    830:        simple_lock(&pool_head_slock);
1.102     chs       831:        LIST_REMOVE(pp, pr_poollist);
1.101     thorpej   832:        if (drainpp == pp)
                    833:                drainpp = NULL;
                    834:        simple_unlock(&pool_head_slock);
                    835:
                    836:        /* Remove this pool from its allocator's list of pools. */
1.117     yamt      837:        pool_reclaim_unregister(pp);
1.93      dbj       838:        s = splvm();
1.66      thorpej   839:        simple_lock(&pp->pr_alloc->pa_slock);
                    840:        TAILQ_REMOVE(&pp->pr_alloc->pa_list, pp, pr_alloc_list);
                    841:        simple_unlock(&pp->pr_alloc->pa_slock);
1.93      dbj       842:        splx(s);
1.66      thorpej   843:
1.101     thorpej   844:        s = splvm();
                    845:        simple_lock(&pp->pr_slock);
                    846:
1.102     chs       847:        KASSERT(LIST_EMPTY(&pp->pr_cachelist));
1.3       pk        848:
                    849: #ifdef DIAGNOSTIC
1.20      thorpej   850:        if (pp->pr_nout != 0) {
1.25      thorpej   851:                pr_printlog(pp, NULL, printf);
1.80      provos    852:                panic("pool_destroy: pool busy: still out: %u",
1.20      thorpej   853:                    pp->pr_nout);
1.3       pk        854:        }
                    855: #endif
1.1       pk        856:
1.101     thorpej   857:        KASSERT(LIST_EMPTY(&pp->pr_fullpages));
                    858:        KASSERT(LIST_EMPTY(&pp->pr_partpages));
                    859:
1.3       pk        860:        /* Remove all pages */
1.101     thorpej   861:        LIST_INIT(&pq);
1.88      chs       862:        while ((ph = LIST_FIRST(&pp->pr_emptypages)) != NULL)
1.101     thorpej   863:                pr_rmpage(pp, ph, &pq);
                    864:
                    865:        simple_unlock(&pp->pr_slock);
                    866:        splx(s);
1.3       pk        867:
1.101     thorpej   868:        pr_pagelist_free(pp, &pq);
1.3       pk        869:
1.59      thorpej   870: #ifdef POOL_DIAGNOSTIC
1.20      thorpej   871:        if ((pp->pr_roflags & PR_LOGGING) != 0)
1.3       pk        872:                free(pp->pr_log, M_TEMP);
1.59      thorpej   873: #endif
1.1       pk        874: }
                    875:
1.68      thorpej   876: void
                    877: pool_set_drain_hook(struct pool *pp, void (*fn)(void *, int), void *arg)
                    878: {
                    879:
                    880:        /* XXX no locking -- must be used just after pool_init() */
                    881: #ifdef DIAGNOSTIC
                    882:        if (pp->pr_drain_hook != NULL)
                    883:                panic("pool_set_drain_hook(%s): already set", pp->pr_wchan);
                    884: #endif
                    885:        pp->pr_drain_hook = fn;
                    886:        pp->pr_drain_hook_arg = arg;
                    887: }
                    888:
1.88      chs       889: static struct pool_item_header *
1.128     christos  890: pool_alloc_item_header(struct pool *pp, void *storage, int flags)
1.55      thorpej   891: {
                    892:        struct pool_item_header *ph;
                    893:        int s;
                    894:
                    895:        LOCK_ASSERT(simple_lock_held(&pp->pr_slock) == 0);
                    896:
                    897:        if ((pp->pr_roflags & PR_PHINPAGE) != 0)
1.128     christos  898:                ph = (struct pool_item_header *) ((char *)storage + pp->pr_phoffset);
1.55      thorpej   899:        else {
1.85      pk        900:                s = splvm();
1.97      yamt      901:                ph = pool_get(pp->pr_phpool, flags);
1.55      thorpej   902:                splx(s);
                    903:        }
                    904:
                    905:        return (ph);
                    906: }
1.1       pk        907:
                    908: /*
1.3       pk        909:  * Grab an item from the pool; must be called at appropriate spl level
1.1       pk        910:  */
1.3       pk        911: void *
1.59      thorpej   912: #ifdef POOL_DIAGNOSTIC
1.42      thorpej   913: _pool_get(struct pool *pp, int flags, const char *file, long line)
1.56      sommerfe  914: #else
                    915: pool_get(struct pool *pp, int flags)
                    916: #endif
1.1       pk        917: {
                    918:        struct pool_item *pi;
1.3       pk        919:        struct pool_item_header *ph;
1.55      thorpej   920:        void *v;
1.1       pk        921:
1.2       pk        922: #ifdef DIAGNOSTIC
1.95      atatat    923:        if (__predict_false(pp->pr_itemsperpage == 0))
                    924:                panic("pool_get: pool %p: pr_itemsperpage is zero, "
                    925:                    "pool not initialized?", pp);
1.84      thorpej   926:        if (__predict_false(curlwp == NULL && doing_shutdown == 0 &&
1.37      sommerfe  927:                            (flags & PR_WAITOK) != 0))
1.77      matt      928:                panic("pool_get: %s: must have NOWAIT", pp->pr_wchan);
1.58      thorpej   929:
1.102     chs       930: #endif /* DIAGNOSTIC */
1.58      thorpej   931: #ifdef LOCKDEBUG
                    932:        if (flags & PR_WAITOK)
1.119     yamt      933:                ASSERT_SLEEPABLE(NULL, "pool_get(PR_WAITOK)");
1.56      sommerfe  934: #endif
1.1       pk        935:
1.21      thorpej   936:        simple_lock(&pp->pr_slock);
1.25      thorpej   937:        pr_enter(pp, file, line);
1.20      thorpej   938:
                    939:  startover:
                    940:        /*
                    941:         * Check to see if we've reached the hard limit.  If we have,
                    942:         * and we can wait, then wait until an item has been returned to
                    943:         * the pool.
                    944:         */
                    945: #ifdef DIAGNOSTIC
1.34      thorpej   946:        if (__predict_false(pp->pr_nout > pp->pr_hardlimit)) {
1.25      thorpej   947:                pr_leave(pp);
1.21      thorpej   948:                simple_unlock(&pp->pr_slock);
1.20      thorpej   949:                panic("pool_get: %s: crossed hard limit", pp->pr_wchan);
                    950:        }
                    951: #endif
1.34      thorpej   952:        if (__predict_false(pp->pr_nout == pp->pr_hardlimit)) {
1.68      thorpej   953:                if (pp->pr_drain_hook != NULL) {
                    954:                        /*
                    955:                         * Since the drain hook is going to free things
                    956:                         * back to the pool, unlock, call the hook, re-lock,
                    957:                         * and check the hardlimit condition again.
                    958:                         */
                    959:                        pr_leave(pp);
                    960:                        simple_unlock(&pp->pr_slock);
                    961:                        (*pp->pr_drain_hook)(pp->pr_drain_hook_arg, flags);
                    962:                        simple_lock(&pp->pr_slock);
                    963:                        pr_enter(pp, file, line);
                    964:                        if (pp->pr_nout < pp->pr_hardlimit)
                    965:                                goto startover;
                    966:                }
                    967:
1.29      sommerfe  968:                if ((flags & PR_WAITOK) && !(flags & PR_LIMITFAIL)) {
1.20      thorpej   969:                        /*
                    970:                         * XXX: A warning isn't logged in this case.  Should
                    971:                         * it be?
                    972:                         */
                    973:                        pp->pr_flags |= PR_WANTED;
1.25      thorpej   974:                        pr_leave(pp);
1.40      sommerfe  975:                        ltsleep(pp, PSWP, pp->pr_wchan, 0, &pp->pr_slock);
1.25      thorpej   976:                        pr_enter(pp, file, line);
1.20      thorpej   977:                        goto startover;
                    978:                }
1.31      thorpej   979:
                    980:                /*
                    981:                 * Log a message that the hard limit has been hit.
                    982:                 */
                    983:                if (pp->pr_hardlimit_warning != NULL &&
                    984:                    ratecheck(&pp->pr_hardlimit_warning_last,
                    985:                              &pp->pr_hardlimit_ratecap))
                    986:                        log(LOG_ERR, "%s\n", pp->pr_hardlimit_warning);
1.21      thorpej   987:
                    988:                pp->pr_nfail++;
                    989:
1.25      thorpej   990:                pr_leave(pp);
1.21      thorpej   991:                simple_unlock(&pp->pr_slock);
1.20      thorpej   992:                return (NULL);
                    993:        }
                    994:
1.3       pk        995:        /*
                    996:         * The convention we use is that if `curpage' is not NULL, then
                    997:         * it points at a non-empty bucket. In particular, `curpage'
                    998:         * never points at a page header which has PR_PHINPAGE set and
                    999:         * has no items in its bucket.
                   1000:         */
1.20      thorpej  1001:        if ((ph = pp->pr_curpage) == NULL) {
1.113     yamt     1002:                int error;
                   1003:
1.20      thorpej  1004: #ifdef DIAGNOSTIC
                   1005:                if (pp->pr_nitems != 0) {
1.21      thorpej  1006:                        simple_unlock(&pp->pr_slock);
1.20      thorpej  1007:                        printf("pool_get: %s: curpage NULL, nitems %u\n",
                   1008:                            pp->pr_wchan, pp->pr_nitems);
1.80      provos   1009:                        panic("pool_get: nitems inconsistent");
1.20      thorpej  1010:                }
                   1011: #endif
                   1012:
1.21      thorpej  1013:                /*
                   1014:                 * Call the back-end page allocator for more memory.
                   1015:                 * Release the pool lock, as the back-end page allocator
                   1016:                 * may block.
                   1017:                 */
1.25      thorpej  1018:                pr_leave(pp);
1.113     yamt     1019:                error = pool_grow(pp, flags);
                   1020:                pr_enter(pp, file, line);
                   1021:                if (error != 0) {
1.21      thorpej  1022:                        /*
1.55      thorpej  1023:                         * We were unable to allocate a page or item
                   1024:                         * header, but we released the lock during
                   1025:                         * allocation, so perhaps items were freed
                   1026:                         * back to the pool.  Check for this case.
1.21      thorpej  1027:                         */
                   1028:                        if (pp->pr_curpage != NULL)
                   1029:                                goto startover;
1.15      pk       1030:
1.117     yamt     1031:                        pp->pr_nfail++;
1.25      thorpej  1032:                        pr_leave(pp);
1.117     yamt     1033:                        simple_unlock(&pp->pr_slock);
                   1034:                        return (NULL);
1.1       pk       1035:                }
1.3       pk       1036:
1.20      thorpej  1037:                /* Start the allocation process over. */
                   1038:                goto startover;
1.3       pk       1039:        }
1.97      yamt     1040:        if (pp->pr_roflags & PR_NOTOUCH) {
                   1041: #ifdef DIAGNOSTIC
                   1042:                if (__predict_false(ph->ph_nmissing == pp->pr_itemsperpage)) {
                   1043:                        pr_leave(pp);
                   1044:                        simple_unlock(&pp->pr_slock);
                   1045:                        panic("pool_get: %s: page empty", pp->pr_wchan);
                   1046:                }
                   1047: #endif
                   1048:                v = pr_item_notouch_get(pp, ph);
                   1049: #ifdef POOL_DIAGNOSTIC
                   1050:                pr_log(pp, v, PRLOG_GET, file, line);
                   1051: #endif
                   1052:        } else {
1.102     chs      1053:                v = pi = LIST_FIRST(&ph->ph_itemlist);
1.97      yamt     1054:                if (__predict_false(v == NULL)) {
                   1055:                        pr_leave(pp);
                   1056:                        simple_unlock(&pp->pr_slock);
                   1057:                        panic("pool_get: %s: page empty", pp->pr_wchan);
                   1058:                }
1.20      thorpej  1059: #ifdef DIAGNOSTIC
1.97      yamt     1060:                if (__predict_false(pp->pr_nitems == 0)) {
                   1061:                        pr_leave(pp);
                   1062:                        simple_unlock(&pp->pr_slock);
                   1063:                        printf("pool_get: %s: items on itemlist, nitems %u\n",
                   1064:                            pp->pr_wchan, pp->pr_nitems);
                   1065:                        panic("pool_get: nitems inconsistent");
                   1066:                }
1.65      enami    1067: #endif
1.56      sommerfe 1068:
1.65      enami    1069: #ifdef POOL_DIAGNOSTIC
1.97      yamt     1070:                pr_log(pp, v, PRLOG_GET, file, line);
1.65      enami    1071: #endif
1.3       pk       1072:
1.65      enami    1073: #ifdef DIAGNOSTIC
1.97      yamt     1074:                if (__predict_false(pi->pi_magic != PI_MAGIC)) {
                   1075:                        pr_printlog(pp, pi, printf);
                   1076:                        panic("pool_get(%s): free list modified: "
                   1077:                            "magic=%x; page %p; item addr %p\n",
                   1078:                            pp->pr_wchan, pi->pi_magic, ph->ph_page, pi);
                   1079:                }
1.3       pk       1080: #endif
                   1081:
1.97      yamt     1082:                /*
                   1083:                 * Remove from item list.
                   1084:                 */
1.102     chs      1085:                LIST_REMOVE(pi, pi_list);
1.97      yamt     1086:        }
1.20      thorpej  1087:        pp->pr_nitems--;
                   1088:        pp->pr_nout++;
1.6       thorpej  1089:        if (ph->ph_nmissing == 0) {
                   1090: #ifdef DIAGNOSTIC
1.34      thorpej  1091:                if (__predict_false(pp->pr_nidle == 0))
1.6       thorpej  1092:                        panic("pool_get: nidle inconsistent");
                   1093: #endif
                   1094:                pp->pr_nidle--;
1.88      chs      1095:
                   1096:                /*
                   1097:                 * This page was previously empty.  Move it to the list of
                   1098:                 * partially-full pages.  This page is already curpage.
                   1099:                 */
                   1100:                LIST_REMOVE(ph, ph_pagelist);
                   1101:                LIST_INSERT_HEAD(&pp->pr_partpages, ph, ph_pagelist);
1.6       thorpej  1102:        }
1.3       pk       1103:        ph->ph_nmissing++;
1.97      yamt     1104:        if (ph->ph_nmissing == pp->pr_itemsperpage) {
1.21      thorpej  1105: #ifdef DIAGNOSTIC
1.97      yamt     1106:                if (__predict_false((pp->pr_roflags & PR_NOTOUCH) == 0 &&
1.102     chs      1107:                    !LIST_EMPTY(&ph->ph_itemlist))) {
1.25      thorpej  1108:                        pr_leave(pp);
1.21      thorpej  1109:                        simple_unlock(&pp->pr_slock);
                   1110:                        panic("pool_get: %s: nmissing inconsistent",
                   1111:                            pp->pr_wchan);
                   1112:                }
                   1113: #endif
1.3       pk       1114:                /*
1.88      chs      1115:                 * This page is now full.  Move it to the full list
                   1116:                 * and select a new current page.
1.3       pk       1117:                 */
1.88      chs      1118:                LIST_REMOVE(ph, ph_pagelist);
                   1119:                LIST_INSERT_HEAD(&pp->pr_fullpages, ph, ph_pagelist);
                   1120:                pool_update_curpage(pp);
1.1       pk       1121:        }
1.3       pk       1122:
                   1123:        pp->pr_nget++;
1.111     christos 1124:        pr_leave(pp);
1.20      thorpej  1125:
                   1126:        /*
                   1127:         * If we have a low water mark and we are now below that low
                   1128:         * water mark, add more items to the pool.
                   1129:         */
1.53      thorpej  1130:        if (POOL_NEEDS_CATCHUP(pp) && pool_catchup(pp) != 0) {
1.20      thorpej  1131:                /*
                   1132:                 * XXX: Should we log a warning?  Should we set up a timeout
                   1133:                 * to try again in a second or so?  The latter could break
                   1134:                 * a caller's assumptions about interrupt protection, etc.
                   1135:                 */
                   1136:        }
                   1137:
1.21      thorpej  1138:        simple_unlock(&pp->pr_slock);
1.125     ad       1139:        KASSERT((((vaddr_t)v + pp->pr_itemoffset) & (pp->pr_align - 1)) == 0);
                   1140:        FREECHECK_OUT(&pp->pr_freecheck, v);
1.1       pk       1141:        return (v);
                   1142: }
                   1143:
                   1144: /*
1.43      thorpej  1145:  * Internal version of pool_put().  Pool is already locked/entered.
1.1       pk       1146:  */
1.43      thorpej  1147: static void
1.101     thorpej  1148: pool_do_put(struct pool *pp, void *v, struct pool_pagelist *pq)
1.1       pk       1149: {
                   1150:        struct pool_item *pi = v;
1.3       pk       1151:        struct pool_item_header *ph;
                   1152:
1.61      chs      1153:        LOCK_ASSERT(simple_lock_held(&pp->pr_slock));
1.125     ad       1154:        FREECHECK_IN(&pp->pr_freecheck, v);
1.61      chs      1155:
1.30      thorpej  1156: #ifdef DIAGNOSTIC
1.34      thorpej  1157:        if (__predict_false(pp->pr_nout == 0)) {
1.30      thorpej  1158:                printf("pool %s: putting with none out\n",
                   1159:                    pp->pr_wchan);
                   1160:                panic("pool_put");
                   1161:        }
                   1162: #endif
1.3       pk       1163:
1.121     yamt     1164:        if (__predict_false((ph = pr_find_pagehead(pp, v)) == NULL)) {
1.25      thorpej  1165:                pr_printlog(pp, NULL, printf);
1.3       pk       1166:                panic("pool_put: %s: page header missing", pp->pr_wchan);
                   1167:        }
1.28      thorpej  1168:
                   1169: #ifdef LOCKDEBUG
                   1170:        /*
                   1171:         * Check if we're freeing a locked simple lock.
                   1172:         */
1.128     christos 1173:        simple_lock_freecheck(pi, (char *)pi + pp->pr_size);
1.28      thorpej  1174: #endif
1.3       pk       1175:
                   1176:        /*
                   1177:         * Return to item list.
                   1178:         */
1.97      yamt     1179:        if (pp->pr_roflags & PR_NOTOUCH) {
                   1180:                pr_item_notouch_put(pp, ph, v);
                   1181:        } else {
1.2       pk       1182: #ifdef DIAGNOSTIC
1.97      yamt     1183:                pi->pi_magic = PI_MAGIC;
1.3       pk       1184: #endif
1.32      chs      1185: #ifdef DEBUG
1.97      yamt     1186:                {
                   1187:                        int i, *ip = v;
1.32      chs      1188:
1.97      yamt     1189:                        for (i = 0; i < pp->pr_size / sizeof(int); i++) {
                   1190:                                *ip++ = PI_MAGIC;
                   1191:                        }
1.32      chs      1192:                }
                   1193: #endif
                   1194:
1.102     chs      1195:                LIST_INSERT_HEAD(&ph->ph_itemlist, pi, pi_list);
1.97      yamt     1196:        }
1.79      thorpej  1197:        KDASSERT(ph->ph_nmissing != 0);
1.3       pk       1198:        ph->ph_nmissing--;
                   1199:        pp->pr_nput++;
1.20      thorpej  1200:        pp->pr_nitems++;
                   1201:        pp->pr_nout--;
1.3       pk       1202:
                   1203:        /* Cancel "pool empty" condition if it exists */
                   1204:        if (pp->pr_curpage == NULL)
                   1205:                pp->pr_curpage = ph;
                   1206:
                   1207:        if (pp->pr_flags & PR_WANTED) {
                   1208:                pp->pr_flags &= ~PR_WANTED;
1.15      pk       1209:                if (ph->ph_nmissing == 0)
                   1210:                        pp->pr_nidle++;
1.128     christos 1211:                wakeup((void *)pp);
1.3       pk       1212:                return;
                   1213:        }
                   1214:
                   1215:        /*
1.88      chs      1216:         * If this page is now empty, do one of two things:
1.21      thorpej  1217:         *
1.88      chs      1218:         *      (1) If we have more pages than the page high water mark,
1.96      thorpej  1219:         *          free the page back to the system.  ONLY CONSIDER
1.90      thorpej  1220:         *          FREEING BACK A PAGE IF WE HAVE MORE THAN OUR MINIMUM PAGE
                   1221:         *          CLAIM.
1.21      thorpej  1222:         *
1.88      chs      1223:         *      (2) Otherwise, move the page to the empty page list.
                   1224:         *
                   1225:         * Either way, select a new current page (so we use a partially-full
                   1226:         * page if one is available).
1.3       pk       1227:         */
                   1228:        if (ph->ph_nmissing == 0) {
1.6       thorpej  1229:                pp->pr_nidle++;
1.90      thorpej  1230:                if (pp->pr_npages > pp->pr_minpages &&
                   1231:                    (pp->pr_npages > pp->pr_maxpages ||
1.117     yamt     1232:                     pa_starved_p(pp->pr_alloc))) {
1.101     thorpej  1233:                        pr_rmpage(pp, ph, pq);
1.3       pk       1234:                } else {
1.88      chs      1235:                        LIST_REMOVE(ph, ph_pagelist);
                   1236:                        LIST_INSERT_HEAD(&pp->pr_emptypages, ph, ph_pagelist);
1.3       pk       1237:
1.21      thorpej  1238:                        /*
                   1239:                         * Update the timestamp on the page.  A page must
                   1240:                         * be idle for some period of time before it can
                   1241:                         * be reclaimed by the pagedaemon.  This minimizes
                   1242:                         * ping-pong'ing for memory.
                   1243:                         */
1.118     kardel   1244:                        getmicrotime(&ph->ph_time);
1.1       pk       1245:                }
1.88      chs      1246:                pool_update_curpage(pp);
1.1       pk       1247:        }
1.88      chs      1248:
1.21      thorpej  1249:        /*
1.88      chs      1250:         * If the page was previously completely full, move it to the
                   1251:         * partially-full list and make it the current page.  The next
                   1252:         * allocation will get the item from this page, instead of
                   1253:         * further fragmenting the pool.
1.21      thorpej  1254:         */
                   1255:        else if (ph->ph_nmissing == (pp->pr_itemsperpage - 1)) {
1.88      chs      1256:                LIST_REMOVE(ph, ph_pagelist);
                   1257:                LIST_INSERT_HEAD(&pp->pr_partpages, ph, ph_pagelist);
1.21      thorpej  1258:                pp->pr_curpage = ph;
                   1259:        }
1.43      thorpej  1260: }
                   1261:
                   1262: /*
                   1263:  * Return resource to the pool; must be called at appropriate spl level
                   1264:  */
1.59      thorpej  1265: #ifdef POOL_DIAGNOSTIC
1.43      thorpej  1266: void
                   1267: _pool_put(struct pool *pp, void *v, const char *file, long line)
                   1268: {
1.101     thorpej  1269:        struct pool_pagelist pq;
                   1270:
                   1271:        LIST_INIT(&pq);
1.43      thorpej  1272:
                   1273:        simple_lock(&pp->pr_slock);
                   1274:        pr_enter(pp, file, line);
                   1275:
1.56      sommerfe 1276:        pr_log(pp, v, PRLOG_PUT, file, line);
                   1277:
1.101     thorpej  1278:        pool_do_put(pp, v, &pq);
1.21      thorpej  1279:
1.25      thorpej  1280:        pr_leave(pp);
1.21      thorpej  1281:        simple_unlock(&pp->pr_slock);
1.101     thorpej  1282:
1.102     chs      1283:        pr_pagelist_free(pp, &pq);
1.1       pk       1284: }
1.57      sommerfe 1285: #undef pool_put
1.59      thorpej  1286: #endif /* POOL_DIAGNOSTIC */
1.1       pk       1287:
1.56      sommerfe 1288: void
                   1289: pool_put(struct pool *pp, void *v)
                   1290: {
1.101     thorpej  1291:        struct pool_pagelist pq;
                   1292:
                   1293:        LIST_INIT(&pq);
1.56      sommerfe 1294:
                   1295:        simple_lock(&pp->pr_slock);
1.101     thorpej  1296:        pool_do_put(pp, v, &pq);
                   1297:        simple_unlock(&pp->pr_slock);
1.56      sommerfe 1298:
1.102     chs      1299:        pr_pagelist_free(pp, &pq);
1.56      sommerfe 1300: }
1.57      sommerfe 1301:
1.59      thorpej  1302: #ifdef POOL_DIAGNOSTIC
1.57      sommerfe 1303: #define                pool_put(h, v)  _pool_put((h), (v), __FILE__, __LINE__)
1.56      sommerfe 1304: #endif
1.74      thorpej  1305:
                   1306: /*
1.113     yamt     1307:  * pool_grow: grow a pool by a page.
                   1308:  *
                   1309:  * => called with pool locked.
                   1310:  * => unlock and relock the pool.
                   1311:  * => return with pool locked.
                   1312:  */
                   1313:
                   1314: static int
                   1315: pool_grow(struct pool *pp, int flags)
                   1316: {
                   1317:        struct pool_item_header *ph = NULL;
                   1318:        char *cp;
                   1319:
                   1320:        simple_unlock(&pp->pr_slock);
                   1321:        cp = pool_allocator_alloc(pp, flags);
                   1322:        if (__predict_true(cp != NULL)) {
                   1323:                ph = pool_alloc_item_header(pp, cp, flags);
                   1324:        }
                   1325:        if (__predict_false(cp == NULL || ph == NULL)) {
                   1326:                if (cp != NULL) {
                   1327:                        pool_allocator_free(pp, cp);
                   1328:                }
                   1329:                simple_lock(&pp->pr_slock);
                   1330:                return ENOMEM;
                   1331:        }
                   1332:
                   1333:        simple_lock(&pp->pr_slock);
                   1334:        pool_prime_page(pp, cp, ph);
                   1335:        pp->pr_npagealloc++;
                   1336:        return 0;
                   1337: }
                   1338:
                   1339: /*
1.74      thorpej  1340:  * Add N items to the pool.
                   1341:  */
                   1342: int
                   1343: pool_prime(struct pool *pp, int n)
                   1344: {
1.75      simonb   1345:        int newpages;
1.113     yamt     1346:        int error = 0;
1.74      thorpej  1347:
                   1348:        simple_lock(&pp->pr_slock);
                   1349:
                   1350:        newpages = roundup(n, pp->pr_itemsperpage) / pp->pr_itemsperpage;
                   1351:
                   1352:        while (newpages-- > 0) {
1.113     yamt     1353:                error = pool_grow(pp, PR_NOWAIT);
                   1354:                if (error) {
1.74      thorpej  1355:                        break;
                   1356:                }
                   1357:                pp->pr_minpages++;
                   1358:        }
                   1359:
                   1360:        if (pp->pr_minpages >= pp->pr_maxpages)
                   1361:                pp->pr_maxpages = pp->pr_minpages + 1;  /* XXX */
                   1362:
                   1363:        simple_unlock(&pp->pr_slock);
1.113     yamt     1364:        return error;
1.74      thorpej  1365: }
1.55      thorpej  1366:
                   1367: /*
1.3       pk       1368:  * Add a page worth of items to the pool.
1.21      thorpej  1369:  *
                   1370:  * Note, we must be called with the pool descriptor LOCKED.
1.3       pk       1371:  */
1.55      thorpej  1372: static void
1.128     christos 1373: pool_prime_page(struct pool *pp, void *storage, struct pool_item_header *ph)
1.3       pk       1374: {
                   1375:        struct pool_item *pi;
1.128     christos 1376:        void *cp = storage;
1.125     ad       1377:        const unsigned int align = pp->pr_align;
                   1378:        const unsigned int ioff = pp->pr_itemoffset;
1.55      thorpej  1379:        int n;
1.36      pk       1380:
1.91      yamt     1381:        LOCK_ASSERT(simple_lock_held(&pp->pr_slock));
                   1382:
1.66      thorpej  1383: #ifdef DIAGNOSTIC
1.121     yamt     1384:        if ((pp->pr_roflags & PR_NOALIGN) == 0 &&
                   1385:            ((uintptr_t)cp & (pp->pr_alloc->pa_pagesz - 1)) != 0)
1.36      pk       1386:                panic("pool_prime_page: %s: unaligned page", pp->pr_wchan);
1.66      thorpej  1387: #endif
1.3       pk       1388:
                   1389:        /*
                   1390:         * Insert page header.
                   1391:         */
1.88      chs      1392:        LIST_INSERT_HEAD(&pp->pr_emptypages, ph, ph_pagelist);
1.102     chs      1393:        LIST_INIT(&ph->ph_itemlist);
1.3       pk       1394:        ph->ph_page = storage;
                   1395:        ph->ph_nmissing = 0;
1.118     kardel   1396:        getmicrotime(&ph->ph_time);
1.88      chs      1397:        if ((pp->pr_roflags & PR_PHINPAGE) == 0)
                   1398:                SPLAY_INSERT(phtree, &pp->pr_phtree, ph);
1.3       pk       1399:
1.6       thorpej  1400:        pp->pr_nidle++;
                   1401:
1.3       pk       1402:        /*
                   1403:         * Color this page.
                   1404:         */
1.128     christos 1405:        cp = (char *)cp + pp->pr_curcolor;
1.3       pk       1406:        if ((pp->pr_curcolor += align) > pp->pr_maxcolor)
                   1407:                pp->pr_curcolor = 0;
                   1408:
                   1409:        /*
                   1410:         * Adjust storage to apply aligment to `pr_itemoffset' in each item.
                   1411:         */
                   1412:        if (ioff != 0)
1.128     christos 1413:                cp = (char *)cp + align - ioff;
1.3       pk       1414:
1.125     ad       1415:        KASSERT((((vaddr_t)cp + ioff) & (align - 1)) == 0);
                   1416:
1.3       pk       1417:        /*
                   1418:         * Insert remaining chunks on the bucket list.
                   1419:         */
                   1420:        n = pp->pr_itemsperpage;
1.20      thorpej  1421:        pp->pr_nitems += n;
1.3       pk       1422:
1.97      yamt     1423:        if (pp->pr_roflags & PR_NOTOUCH) {
1.99      yamt     1424:                pool_item_freelist_t *freelist = PR_FREELIST(ph);
1.97      yamt     1425:                int i;
                   1426:
1.128     christos 1427:                ph->ph_off = (char *)cp - (char *)storage;
1.97      yamt     1428:                ph->ph_firstfree = 0;
                   1429:                for (i = 0; i < n - 1; i++)
                   1430:                        freelist[i] = i + 1;
                   1431:                freelist[n - 1] = PR_INDEX_EOL;
                   1432:        } else {
                   1433:                while (n--) {
                   1434:                        pi = (struct pool_item *)cp;
1.78      thorpej  1435:
1.97      yamt     1436:                        KASSERT(((((vaddr_t)pi) + ioff) & (align - 1)) == 0);
1.3       pk       1437:
1.97      yamt     1438:                        /* Insert on page list */
1.102     chs      1439:                        LIST_INSERT_HEAD(&ph->ph_itemlist, pi, pi_list);
1.3       pk       1440: #ifdef DIAGNOSTIC
1.97      yamt     1441:                        pi->pi_magic = PI_MAGIC;
1.3       pk       1442: #endif
1.128     christos 1443:                        cp = (char *)cp + pp->pr_size;
1.125     ad       1444:
                   1445:                        KASSERT((((vaddr_t)cp + ioff) & (align - 1)) == 0);
1.97      yamt     1446:                }
1.3       pk       1447:        }
                   1448:
                   1449:        /*
                   1450:         * If the pool was depleted, point at the new page.
                   1451:         */
                   1452:        if (pp->pr_curpage == NULL)
                   1453:                pp->pr_curpage = ph;
                   1454:
                   1455:        if (++pp->pr_npages > pp->pr_hiwat)
                   1456:                pp->pr_hiwat = pp->pr_npages;
                   1457: }
                   1458:
1.20      thorpej  1459: /*
1.52      thorpej  1460:  * Used by pool_get() when nitems drops below the low water mark.  This
1.88      chs      1461:  * is used to catch up pr_nitems with the low water mark.
1.20      thorpej  1462:  *
1.21      thorpej  1463:  * Note 1, we never wait for memory here, we let the caller decide what to do.
1.20      thorpej  1464:  *
1.73      thorpej  1465:  * Note 2, we must be called with the pool already locked, and we return
1.20      thorpej  1466:  * with it locked.
                   1467:  */
                   1468: static int
1.42      thorpej  1469: pool_catchup(struct pool *pp)
1.20      thorpej  1470: {
                   1471:        int error = 0;
                   1472:
1.54      thorpej  1473:        while (POOL_NEEDS_CATCHUP(pp)) {
1.113     yamt     1474:                error = pool_grow(pp, PR_NOWAIT);
                   1475:                if (error) {
1.20      thorpej  1476:                        break;
                   1477:                }
                   1478:        }
1.113     yamt     1479:        return error;
1.20      thorpej  1480: }
                   1481:
1.88      chs      1482: static void
                   1483: pool_update_curpage(struct pool *pp)
                   1484: {
                   1485:
                   1486:        pp->pr_curpage = LIST_FIRST(&pp->pr_partpages);
                   1487:        if (pp->pr_curpage == NULL) {
                   1488:                pp->pr_curpage = LIST_FIRST(&pp->pr_emptypages);
                   1489:        }
                   1490: }
                   1491:
1.3       pk       1492: void
1.42      thorpej  1493: pool_setlowat(struct pool *pp, int n)
1.3       pk       1494: {
1.15      pk       1495:
1.21      thorpej  1496:        simple_lock(&pp->pr_slock);
                   1497:
1.3       pk       1498:        pp->pr_minitems = n;
1.15      pk       1499:        pp->pr_minpages = (n == 0)
                   1500:                ? 0
1.18      thorpej  1501:                : roundup(n, pp->pr_itemsperpage) / pp->pr_itemsperpage;
1.20      thorpej  1502:
                   1503:        /* Make sure we're caught up with the newly-set low water mark. */
1.75      simonb   1504:        if (POOL_NEEDS_CATCHUP(pp) && pool_catchup(pp) != 0) {
1.20      thorpej  1505:                /*
                   1506:                 * XXX: Should we log a warning?  Should we set up a timeout
                   1507:                 * to try again in a second or so?  The latter could break
                   1508:                 * a caller's assumptions about interrupt protection, etc.
                   1509:                 */
                   1510:        }
1.21      thorpej  1511:
                   1512:        simple_unlock(&pp->pr_slock);
1.3       pk       1513: }
                   1514:
                   1515: void
1.42      thorpej  1516: pool_sethiwat(struct pool *pp, int n)
1.3       pk       1517: {
1.15      pk       1518:
1.21      thorpej  1519:        simple_lock(&pp->pr_slock);
                   1520:
1.15      pk       1521:        pp->pr_maxpages = (n == 0)
                   1522:                ? 0
1.18      thorpej  1523:                : roundup(n, pp->pr_itemsperpage) / pp->pr_itemsperpage;
1.21      thorpej  1524:
                   1525:        simple_unlock(&pp->pr_slock);
1.3       pk       1526: }
                   1527:
1.20      thorpej  1528: void
1.42      thorpej  1529: pool_sethardlimit(struct pool *pp, int n, const char *warnmess, int ratecap)
1.20      thorpej  1530: {
                   1531:
1.21      thorpej  1532:        simple_lock(&pp->pr_slock);
1.20      thorpej  1533:
                   1534:        pp->pr_hardlimit = n;
                   1535:        pp->pr_hardlimit_warning = warnmess;
1.31      thorpej  1536:        pp->pr_hardlimit_ratecap.tv_sec = ratecap;
                   1537:        pp->pr_hardlimit_warning_last.tv_sec = 0;
                   1538:        pp->pr_hardlimit_warning_last.tv_usec = 0;
1.20      thorpej  1539:
                   1540:        /*
1.21      thorpej  1541:         * In-line version of pool_sethiwat(), because we don't want to
                   1542:         * release the lock.
1.20      thorpej  1543:         */
                   1544:        pp->pr_maxpages = (n == 0)
                   1545:                ? 0
                   1546:                : roundup(n, pp->pr_itemsperpage) / pp->pr_itemsperpage;
1.21      thorpej  1547:
                   1548:        simple_unlock(&pp->pr_slock);
1.20      thorpej  1549: }
1.3       pk       1550:
                   1551: /*
                   1552:  * Release all complete pages that have not been used recently.
                   1553:  */
1.66      thorpej  1554: int
1.59      thorpej  1555: #ifdef POOL_DIAGNOSTIC
1.42      thorpej  1556: _pool_reclaim(struct pool *pp, const char *file, long line)
1.56      sommerfe 1557: #else
                   1558: pool_reclaim(struct pool *pp)
                   1559: #endif
1.3       pk       1560: {
                   1561:        struct pool_item_header *ph, *phnext;
1.43      thorpej  1562:        struct pool_cache *pc;
1.61      chs      1563:        struct pool_pagelist pq;
1.102     chs      1564:        struct pool_cache_grouplist pcgl;
                   1565:        struct timeval curtime, diff;
1.3       pk       1566:
1.68      thorpej  1567:        if (pp->pr_drain_hook != NULL) {
                   1568:                /*
                   1569:                 * The drain hook must be called with the pool unlocked.
                   1570:                 */
                   1571:                (*pp->pr_drain_hook)(pp->pr_drain_hook_arg, PR_NOWAIT);
                   1572:        }
                   1573:
1.21      thorpej  1574:        if (simple_lock_try(&pp->pr_slock) == 0)
1.66      thorpej  1575:                return (0);
1.25      thorpej  1576:        pr_enter(pp, file, line);
1.68      thorpej  1577:
1.88      chs      1578:        LIST_INIT(&pq);
1.102     chs      1579:        LIST_INIT(&pcgl);
1.3       pk       1580:
1.43      thorpej  1581:        /*
                   1582:         * Reclaim items from the pool's caches.
                   1583:         */
1.102     chs      1584:        LIST_FOREACH(pc, &pp->pr_cachelist, pc_poollist)
                   1585:                pool_cache_reclaim(pc, &pq, &pcgl);
1.43      thorpej  1586:
1.118     kardel   1587:        getmicrotime(&curtime);
1.21      thorpej  1588:
1.88      chs      1589:        for (ph = LIST_FIRST(&pp->pr_emptypages); ph != NULL; ph = phnext) {
                   1590:                phnext = LIST_NEXT(ph, ph_pagelist);
1.3       pk       1591:
                   1592:                /* Check our minimum page claim */
                   1593:                if (pp->pr_npages <= pp->pr_minpages)
                   1594:                        break;
                   1595:
1.88      chs      1596:                KASSERT(ph->ph_nmissing == 0);
                   1597:                timersub(&curtime, &ph->ph_time, &diff);
1.117     yamt     1598:                if (diff.tv_sec < pool_inactive_time
                   1599:                    && !pa_starved_p(pp->pr_alloc))
1.88      chs      1600:                        continue;
1.21      thorpej  1601:
1.88      chs      1602:                /*
                   1603:                 * If freeing this page would put us below
                   1604:                 * the low water mark, stop now.
                   1605:                 */
                   1606:                if ((pp->pr_nitems - pp->pr_itemsperpage) <
                   1607:                    pp->pr_minitems)
                   1608:                        break;
1.21      thorpej  1609:
1.88      chs      1610:                pr_rmpage(pp, ph, &pq);
1.3       pk       1611:        }
                   1612:
1.25      thorpej  1613:        pr_leave(pp);
1.21      thorpej  1614:        simple_unlock(&pp->pr_slock);
1.102     chs      1615:        if (LIST_EMPTY(&pq) && LIST_EMPTY(&pcgl))
                   1616:                return 0;
1.66      thorpej  1617:
1.101     thorpej  1618:        pr_pagelist_free(pp, &pq);
1.102     chs      1619:        pcg_grouplist_free(&pcgl);
1.66      thorpej  1620:        return (1);
1.3       pk       1621: }
                   1622:
                   1623: /*
                   1624:  * Drain pools, one at a time.
1.21      thorpej  1625:  *
                   1626:  * Note, we must never be called from an interrupt context.
1.3       pk       1627:  */
                   1628: void
1.124     yamt     1629: pool_drain(void *arg)
1.3       pk       1630: {
                   1631:        struct pool *pp;
1.23      thorpej  1632:        int s;
1.3       pk       1633:
1.61      chs      1634:        pp = NULL;
1.49      thorpej  1635:        s = splvm();
1.23      thorpej  1636:        simple_lock(&pool_head_slock);
1.61      chs      1637:        if (drainpp == NULL) {
1.102     chs      1638:                drainpp = LIST_FIRST(&pool_head);
1.61      chs      1639:        }
                   1640:        if (drainpp) {
                   1641:                pp = drainpp;
1.102     chs      1642:                drainpp = LIST_NEXT(pp, pr_poollist);
1.61      chs      1643:        }
                   1644:        simple_unlock(&pool_head_slock);
1.115     christos 1645:        if (pp)
                   1646:                pool_reclaim(pp);
1.61      chs      1647:        splx(s);
1.3       pk       1648: }
                   1649:
                   1650: /*
                   1651:  * Diagnostic helpers.
                   1652:  */
                   1653: void
1.42      thorpej  1654: pool_print(struct pool *pp, const char *modif)
1.21      thorpej  1655: {
                   1656:        int s;
                   1657:
1.49      thorpej  1658:        s = splvm();
1.25      thorpej  1659:        if (simple_lock_try(&pp->pr_slock) == 0) {
                   1660:                printf("pool %s is locked; try again later\n",
                   1661:                    pp->pr_wchan);
                   1662:                splx(s);
                   1663:                return;
                   1664:        }
                   1665:        pool_print1(pp, modif, printf);
1.21      thorpej  1666:        simple_unlock(&pp->pr_slock);
                   1667:        splx(s);
                   1668: }
                   1669:
1.25      thorpej  1670: void
1.108     yamt     1671: pool_printall(const char *modif, void (*pr)(const char *, ...))
                   1672: {
                   1673:        struct pool *pp;
                   1674:
                   1675:        if (simple_lock_try(&pool_head_slock) == 0) {
                   1676:                (*pr)("WARNING: pool_head_slock is locked\n");
                   1677:        } else {
                   1678:                simple_unlock(&pool_head_slock);
                   1679:        }
                   1680:
                   1681:        LIST_FOREACH(pp, &pool_head, pr_poollist) {
                   1682:                pool_printit(pp, modif, pr);
                   1683:        }
                   1684: }
                   1685:
                   1686: void
1.42      thorpej  1687: pool_printit(struct pool *pp, const char *modif, void (*pr)(const char *, ...))
1.25      thorpej  1688: {
                   1689:
                   1690:        if (pp == NULL) {
                   1691:                (*pr)("Must specify a pool to print.\n");
                   1692:                return;
                   1693:        }
                   1694:
                   1695:        /*
                   1696:         * Called from DDB; interrupts should be blocked, and all
                   1697:         * other processors should be paused.  We can skip locking
                   1698:         * the pool in this case.
                   1699:         *
                   1700:         * We do a simple_lock_try() just to print the lock
                   1701:         * status, however.
                   1702:         */
                   1703:
                   1704:        if (simple_lock_try(&pp->pr_slock) == 0)
                   1705:                (*pr)("WARNING: pool %s is locked\n", pp->pr_wchan);
                   1706:        else
1.107     yamt     1707:                simple_unlock(&pp->pr_slock);
1.25      thorpej  1708:
                   1709:        pool_print1(pp, modif, pr);
                   1710: }
                   1711:
1.21      thorpej  1712: static void
1.124     yamt     1713: pool_print_pagelist(struct pool *pp, struct pool_pagelist *pl,
1.97      yamt     1714:     void (*pr)(const char *, ...))
1.88      chs      1715: {
                   1716:        struct pool_item_header *ph;
                   1717: #ifdef DIAGNOSTIC
                   1718:        struct pool_item *pi;
                   1719: #endif
                   1720:
                   1721:        LIST_FOREACH(ph, pl, ph_pagelist) {
                   1722:                (*pr)("\t\tpage %p, nmissing %d, time %lu,%lu\n",
                   1723:                    ph->ph_page, ph->ph_nmissing,
                   1724:                    (u_long)ph->ph_time.tv_sec,
                   1725:                    (u_long)ph->ph_time.tv_usec);
                   1726: #ifdef DIAGNOSTIC
1.97      yamt     1727:                if (!(pp->pr_roflags & PR_NOTOUCH)) {
1.102     chs      1728:                        LIST_FOREACH(pi, &ph->ph_itemlist, pi_list) {
1.97      yamt     1729:                                if (pi->pi_magic != PI_MAGIC) {
                   1730:                                        (*pr)("\t\t\titem %p, magic 0x%x\n",
                   1731:                                            pi, pi->pi_magic);
                   1732:                                }
1.88      chs      1733:                        }
                   1734:                }
                   1735: #endif
                   1736:        }
                   1737: }
                   1738:
                   1739: static void
1.42      thorpej  1740: pool_print1(struct pool *pp, const char *modif, void (*pr)(const char *, ...))
1.3       pk       1741: {
1.25      thorpej  1742:        struct pool_item_header *ph;
1.44      thorpej  1743:        struct pool_cache *pc;
                   1744:        struct pool_cache_group *pcg;
                   1745:        int i, print_log = 0, print_pagelist = 0, print_cache = 0;
1.25      thorpej  1746:        char c;
                   1747:
                   1748:        while ((c = *modif++) != '\0') {
                   1749:                if (c == 'l')
                   1750:                        print_log = 1;
                   1751:                if (c == 'p')
                   1752:                        print_pagelist = 1;
1.44      thorpej  1753:                if (c == 'c')
                   1754:                        print_cache = 1;
1.25      thorpej  1755:        }
                   1756:
                   1757:        (*pr)("POOL %s: size %u, align %u, ioff %u, roflags 0x%08x\n",
                   1758:            pp->pr_wchan, pp->pr_size, pp->pr_align, pp->pr_itemoffset,
                   1759:            pp->pr_roflags);
1.66      thorpej  1760:        (*pr)("\talloc %p\n", pp->pr_alloc);
1.25      thorpej  1761:        (*pr)("\tminitems %u, minpages %u, maxpages %u, npages %u\n",
                   1762:            pp->pr_minitems, pp->pr_minpages, pp->pr_maxpages, pp->pr_npages);
                   1763:        (*pr)("\titemsperpage %u, nitems %u, nout %u, hardlimit %u\n",
                   1764:            pp->pr_itemsperpage, pp->pr_nitems, pp->pr_nout, pp->pr_hardlimit);
                   1765:
                   1766:        (*pr)("\n\tnget %lu, nfail %lu, nput %lu\n",
                   1767:            pp->pr_nget, pp->pr_nfail, pp->pr_nput);
                   1768:        (*pr)("\tnpagealloc %lu, npagefree %lu, hiwat %u, nidle %lu\n",
                   1769:            pp->pr_npagealloc, pp->pr_npagefree, pp->pr_hiwat, pp->pr_nidle);
                   1770:
                   1771:        if (print_pagelist == 0)
                   1772:                goto skip_pagelist;
                   1773:
1.88      chs      1774:        if ((ph = LIST_FIRST(&pp->pr_emptypages)) != NULL)
                   1775:                (*pr)("\n\tempty page list:\n");
1.97      yamt     1776:        pool_print_pagelist(pp, &pp->pr_emptypages, pr);
1.88      chs      1777:        if ((ph = LIST_FIRST(&pp->pr_fullpages)) != NULL)
                   1778:                (*pr)("\n\tfull page list:\n");
1.97      yamt     1779:        pool_print_pagelist(pp, &pp->pr_fullpages, pr);
1.88      chs      1780:        if ((ph = LIST_FIRST(&pp->pr_partpages)) != NULL)
                   1781:                (*pr)("\n\tpartial-page list:\n");
1.97      yamt     1782:        pool_print_pagelist(pp, &pp->pr_partpages, pr);
1.88      chs      1783:
1.25      thorpej  1784:        if (pp->pr_curpage == NULL)
                   1785:                (*pr)("\tno current page\n");
                   1786:        else
                   1787:                (*pr)("\tcurpage %p\n", pp->pr_curpage->ph_page);
                   1788:
                   1789:  skip_pagelist:
                   1790:        if (print_log == 0)
                   1791:                goto skip_log;
                   1792:
                   1793:        (*pr)("\n");
                   1794:        if ((pp->pr_roflags & PR_LOGGING) == 0)
                   1795:                (*pr)("\tno log\n");
1.122     christos 1796:        else {
1.25      thorpej  1797:                pr_printlog(pp, NULL, pr);
1.122     christos 1798:        }
1.3       pk       1799:
1.25      thorpej  1800:  skip_log:
1.44      thorpej  1801:        if (print_cache == 0)
                   1802:                goto skip_cache;
                   1803:
1.102     chs      1804: #define PR_GROUPLIST(pcg)                                              \
                   1805:        (*pr)("\t\tgroup %p: avail %d\n", pcg, pcg->pcg_avail);         \
                   1806:        for (i = 0; i < PCG_NOBJECTS; i++) {                            \
                   1807:                if (pcg->pcg_objects[i].pcgo_pa !=                      \
                   1808:                    POOL_PADDR_INVALID) {                               \
                   1809:                        (*pr)("\t\t\t%p, 0x%llx\n",                     \
                   1810:                            pcg->pcg_objects[i].pcgo_va,                \
                   1811:                            (unsigned long long)                        \
                   1812:                            pcg->pcg_objects[i].pcgo_pa);               \
                   1813:                } else {                                                \
                   1814:                        (*pr)("\t\t\t%p\n",                             \
                   1815:                            pcg->pcg_objects[i].pcgo_va);               \
                   1816:                }                                                       \
                   1817:        }
                   1818:
                   1819:        LIST_FOREACH(pc, &pp->pr_cachelist, pc_poollist) {
1.103     chs      1820:                (*pr)("\tcache %p\n", pc);
1.48      thorpej  1821:                (*pr)("\t    hits %lu misses %lu ngroups %lu nitems %lu\n",
                   1822:                    pc->pc_hits, pc->pc_misses, pc->pc_ngroups, pc->pc_nitems);
1.102     chs      1823:                (*pr)("\t    full groups:\n");
1.103     chs      1824:                LIST_FOREACH(pcg, &pc->pc_fullgroups, pcg_list) {
1.102     chs      1825:                        PR_GROUPLIST(pcg);
1.103     chs      1826:                }
1.102     chs      1827:                (*pr)("\t    partial groups:\n");
1.103     chs      1828:                LIST_FOREACH(pcg, &pc->pc_partgroups, pcg_list) {
1.102     chs      1829:                        PR_GROUPLIST(pcg);
1.103     chs      1830:                }
1.102     chs      1831:                (*pr)("\t    empty groups:\n");
1.103     chs      1832:                LIST_FOREACH(pcg, &pc->pc_emptygroups, pcg_list) {
1.102     chs      1833:                        PR_GROUPLIST(pcg);
1.103     chs      1834:                }
1.44      thorpej  1835:        }
1.102     chs      1836: #undef PR_GROUPLIST
1.44      thorpej  1837:
                   1838:  skip_cache:
1.88      chs      1839:        pr_enter_check(pp, pr);
                   1840: }
                   1841:
                   1842: static int
                   1843: pool_chk_page(struct pool *pp, const char *label, struct pool_item_header *ph)
                   1844: {
                   1845:        struct pool_item *pi;
1.128     christos 1846:        void *page;
1.88      chs      1847:        int n;
                   1848:
1.121     yamt     1849:        if ((pp->pr_roflags & PR_NOALIGN) == 0) {
1.128     christos 1850:                page = (void *)((uintptr_t)ph & pp->pr_alloc->pa_pagemask);
1.121     yamt     1851:                if (page != ph->ph_page &&
                   1852:                    (pp->pr_roflags & PR_PHINPAGE) != 0) {
                   1853:                        if (label != NULL)
                   1854:                                printf("%s: ", label);
                   1855:                        printf("pool(%p:%s): page inconsistency: page %p;"
                   1856:                               " at page head addr %p (p %p)\n", pp,
                   1857:                                pp->pr_wchan, ph->ph_page,
                   1858:                                ph, page);
                   1859:                        return 1;
                   1860:                }
1.88      chs      1861:        }
1.3       pk       1862:
1.97      yamt     1863:        if ((pp->pr_roflags & PR_NOTOUCH) != 0)
                   1864:                return 0;
                   1865:
1.102     chs      1866:        for (pi = LIST_FIRST(&ph->ph_itemlist), n = 0;
1.88      chs      1867:             pi != NULL;
1.102     chs      1868:             pi = LIST_NEXT(pi,pi_list), n++) {
1.88      chs      1869:
                   1870: #ifdef DIAGNOSTIC
                   1871:                if (pi->pi_magic != PI_MAGIC) {
                   1872:                        if (label != NULL)
                   1873:                                printf("%s: ", label);
                   1874:                        printf("pool(%s): free list modified: magic=%x;"
1.121     yamt     1875:                               " page %p; item ordinal %d; addr %p\n",
1.88      chs      1876:                                pp->pr_wchan, pi->pi_magic, ph->ph_page,
1.121     yamt     1877:                                n, pi);
1.88      chs      1878:                        panic("pool");
                   1879:                }
                   1880: #endif
1.121     yamt     1881:                if ((pp->pr_roflags & PR_NOALIGN) != 0) {
                   1882:                        continue;
                   1883:                }
1.128     christos 1884:                page = (void *)((uintptr_t)pi & pp->pr_alloc->pa_pagemask);
1.88      chs      1885:                if (page == ph->ph_page)
                   1886:                        continue;
                   1887:
                   1888:                if (label != NULL)
                   1889:                        printf("%s: ", label);
                   1890:                printf("pool(%p:%s): page inconsistency: page %p;"
                   1891:                       " item ordinal %d; addr %p (p %p)\n", pp,
                   1892:                        pp->pr_wchan, ph->ph_page,
                   1893:                        n, pi, page);
                   1894:                return 1;
                   1895:        }
                   1896:        return 0;
1.3       pk       1897: }
                   1898:
1.88      chs      1899:
1.3       pk       1900: int
1.42      thorpej  1901: pool_chk(struct pool *pp, const char *label)
1.3       pk       1902: {
                   1903:        struct pool_item_header *ph;
                   1904:        int r = 0;
                   1905:
1.21      thorpej  1906:        simple_lock(&pp->pr_slock);
1.88      chs      1907:        LIST_FOREACH(ph, &pp->pr_emptypages, ph_pagelist) {
                   1908:                r = pool_chk_page(pp, label, ph);
                   1909:                if (r) {
                   1910:                        goto out;
                   1911:                }
                   1912:        }
                   1913:        LIST_FOREACH(ph, &pp->pr_fullpages, ph_pagelist) {
                   1914:                r = pool_chk_page(pp, label, ph);
                   1915:                if (r) {
1.3       pk       1916:                        goto out;
                   1917:                }
1.88      chs      1918:        }
                   1919:        LIST_FOREACH(ph, &pp->pr_partpages, ph_pagelist) {
                   1920:                r = pool_chk_page(pp, label, ph);
                   1921:                if (r) {
1.3       pk       1922:                        goto out;
                   1923:                }
                   1924:        }
1.88      chs      1925:
1.3       pk       1926: out:
1.21      thorpej  1927:        simple_unlock(&pp->pr_slock);
1.3       pk       1928:        return (r);
1.43      thorpej  1929: }
                   1930:
                   1931: /*
                   1932:  * pool_cache_init:
                   1933:  *
                   1934:  *     Initialize a pool cache.
                   1935:  *
                   1936:  *     NOTE: If the pool must be protected from interrupts, we expect
                   1937:  *     to be called at the appropriate interrupt priority level.
                   1938:  */
                   1939: void
                   1940: pool_cache_init(struct pool_cache *pc, struct pool *pp,
                   1941:     int (*ctor)(void *, void *, int),
                   1942:     void (*dtor)(void *, void *),
                   1943:     void *arg)
                   1944: {
                   1945:
1.102     chs      1946:        LIST_INIT(&pc->pc_emptygroups);
                   1947:        LIST_INIT(&pc->pc_fullgroups);
                   1948:        LIST_INIT(&pc->pc_partgroups);
1.43      thorpej  1949:        simple_lock_init(&pc->pc_slock);
                   1950:
                   1951:        pc->pc_pool = pp;
                   1952:
                   1953:        pc->pc_ctor = ctor;
                   1954:        pc->pc_dtor = dtor;
                   1955:        pc->pc_arg  = arg;
                   1956:
1.48      thorpej  1957:        pc->pc_hits   = 0;
                   1958:        pc->pc_misses = 0;
                   1959:
                   1960:        pc->pc_ngroups = 0;
                   1961:
                   1962:        pc->pc_nitems = 0;
                   1963:
1.43      thorpej  1964:        simple_lock(&pp->pr_slock);
1.102     chs      1965:        LIST_INSERT_HEAD(&pp->pr_cachelist, pc, pc_poollist);
1.43      thorpej  1966:        simple_unlock(&pp->pr_slock);
                   1967: }
                   1968:
                   1969: /*
                   1970:  * pool_cache_destroy:
                   1971:  *
                   1972:  *     Destroy a pool cache.
                   1973:  */
                   1974: void
                   1975: pool_cache_destroy(struct pool_cache *pc)
                   1976: {
                   1977:        struct pool *pp = pc->pc_pool;
                   1978:
                   1979:        /* First, invalidate the entire cache. */
                   1980:        pool_cache_invalidate(pc);
                   1981:
                   1982:        /* ...and remove it from the pool's cache list. */
                   1983:        simple_lock(&pp->pr_slock);
1.102     chs      1984:        LIST_REMOVE(pc, pc_poollist);
1.43      thorpej  1985:        simple_unlock(&pp->pr_slock);
                   1986: }
                   1987:
1.110     perry    1988: static inline void *
1.87      thorpej  1989: pcg_get(struct pool_cache_group *pcg, paddr_t *pap)
1.43      thorpej  1990: {
                   1991:        void *object;
                   1992:        u_int idx;
                   1993:
                   1994:        KASSERT(pcg->pcg_avail <= PCG_NOBJECTS);
1.45      thorpej  1995:        KASSERT(pcg->pcg_avail != 0);
1.43      thorpej  1996:        idx = --pcg->pcg_avail;
                   1997:
1.87      thorpej  1998:        KASSERT(pcg->pcg_objects[idx].pcgo_va != NULL);
                   1999:        object = pcg->pcg_objects[idx].pcgo_va;
                   2000:        if (pap != NULL)
                   2001:                *pap = pcg->pcg_objects[idx].pcgo_pa;
                   2002:        pcg->pcg_objects[idx].pcgo_va = NULL;
1.43      thorpej  2003:
                   2004:        return (object);
                   2005: }
                   2006:
1.110     perry    2007: static inline void
1.87      thorpej  2008: pcg_put(struct pool_cache_group *pcg, void *object, paddr_t pa)
1.43      thorpej  2009: {
                   2010:        u_int idx;
                   2011:
                   2012:        KASSERT(pcg->pcg_avail < PCG_NOBJECTS);
                   2013:        idx = pcg->pcg_avail++;
                   2014:
1.87      thorpej  2015:        KASSERT(pcg->pcg_objects[idx].pcgo_va == NULL);
                   2016:        pcg->pcg_objects[idx].pcgo_va = object;
                   2017:        pcg->pcg_objects[idx].pcgo_pa = pa;
1.43      thorpej  2018: }
                   2019:
1.102     chs      2020: static void
                   2021: pcg_grouplist_free(struct pool_cache_grouplist *pcgl)
                   2022: {
                   2023:        struct pool_cache_group *pcg;
                   2024:        int s;
                   2025:
                   2026:        s = splvm();
                   2027:        while ((pcg = LIST_FIRST(pcgl)) != NULL) {
                   2028:                LIST_REMOVE(pcg, pcg_list);
                   2029:                pool_put(&pcgpool, pcg);
                   2030:        }
                   2031:        splx(s);
                   2032: }
                   2033:
1.43      thorpej  2034: /*
1.87      thorpej  2035:  * pool_cache_get{,_paddr}:
1.43      thorpej  2036:  *
1.87      thorpej  2037:  *     Get an object from a pool cache (optionally returning
                   2038:  *     the physical address of the object).
1.43      thorpej  2039:  */
                   2040: void *
1.87      thorpej  2041: pool_cache_get_paddr(struct pool_cache *pc, int flags, paddr_t *pap)
1.43      thorpej  2042: {
                   2043:        struct pool_cache_group *pcg;
                   2044:        void *object;
1.58      thorpej  2045:
                   2046: #ifdef LOCKDEBUG
                   2047:        if (flags & PR_WAITOK)
1.119     yamt     2048:                ASSERT_SLEEPABLE(NULL, "pool_cache_get(PR_WAITOK)");
1.58      thorpej  2049: #endif
1.43      thorpej  2050:
                   2051:        simple_lock(&pc->pc_slock);
                   2052:
1.102     chs      2053:        pcg = LIST_FIRST(&pc->pc_partgroups);
                   2054:        if (pcg == NULL) {
                   2055:                pcg = LIST_FIRST(&pc->pc_fullgroups);
                   2056:                if (pcg != NULL) {
                   2057:                        LIST_REMOVE(pcg, pcg_list);
                   2058:                        LIST_INSERT_HEAD(&pc->pc_partgroups, pcg, pcg_list);
1.43      thorpej  2059:                }
1.102     chs      2060:        }
                   2061:        if (pcg == NULL) {
1.43      thorpej  2062:
                   2063:                /*
                   2064:                 * No groups with any available objects.  Allocate
                   2065:                 * a new object, construct it, and return it to
                   2066:                 * the caller.  We will allocate a group, if necessary,
                   2067:                 * when the object is freed back to the cache.
                   2068:                 */
1.48      thorpej  2069:                pc->pc_misses++;
1.43      thorpej  2070:                simple_unlock(&pc->pc_slock);
                   2071:                object = pool_get(pc->pc_pool, flags);
                   2072:                if (object != NULL && pc->pc_ctor != NULL) {
                   2073:                        if ((*pc->pc_ctor)(pc->pc_arg, object, flags) != 0) {
                   2074:                                pool_put(pc->pc_pool, object);
                   2075:                                return (NULL);
                   2076:                        }
                   2077:                }
1.125     ad       2078:                KASSERT((((vaddr_t)object + pc->pc_pool->pr_itemoffset) &
                   2079:                    (pc->pc_pool->pr_align - 1)) == 0);
1.87      thorpej  2080:                if (object != NULL && pap != NULL) {
                   2081: #ifdef POOL_VTOPHYS
                   2082:                        *pap = POOL_VTOPHYS(object);
                   2083: #else
                   2084:                        *pap = POOL_PADDR_INVALID;
                   2085: #endif
                   2086:                }
1.125     ad       2087:
                   2088:                FREECHECK_OUT(&pc->pc_freecheck, object);
1.43      thorpej  2089:                return (object);
                   2090:        }
                   2091:
1.48      thorpej  2092:        pc->pc_hits++;
                   2093:        pc->pc_nitems--;
1.87      thorpej  2094:        object = pcg_get(pcg, pap);
1.43      thorpej  2095:
1.102     chs      2096:        if (pcg->pcg_avail == 0) {
                   2097:                LIST_REMOVE(pcg, pcg_list);
                   2098:                LIST_INSERT_HEAD(&pc->pc_emptygroups, pcg, pcg_list);
                   2099:        }
1.43      thorpej  2100:        simple_unlock(&pc->pc_slock);
                   2101:
1.125     ad       2102:        KASSERT((((vaddr_t)object + pc->pc_pool->pr_itemoffset) &
                   2103:            (pc->pc_pool->pr_align - 1)) == 0);
                   2104:        FREECHECK_OUT(&pc->pc_freecheck, object);
1.43      thorpej  2105:        return (object);
                   2106: }
                   2107:
                   2108: /*
1.87      thorpej  2109:  * pool_cache_put{,_paddr}:
1.43      thorpej  2110:  *
1.87      thorpej  2111:  *     Put an object back to the pool cache (optionally caching the
                   2112:  *     physical address of the object).
1.43      thorpej  2113:  */
                   2114: void
1.87      thorpej  2115: pool_cache_put_paddr(struct pool_cache *pc, void *object, paddr_t pa)
1.43      thorpej  2116: {
                   2117:        struct pool_cache_group *pcg;
1.60      thorpej  2118:        int s;
1.43      thorpej  2119:
1.125     ad       2120:        FREECHECK_IN(&pc->pc_freecheck, object);
                   2121:
1.109     christos 2122:        if (__predict_false((pc->pc_pool->pr_flags & PR_WANTED) != 0)) {
                   2123:                goto destruct;
                   2124:        }
                   2125:
1.43      thorpej  2126:        simple_lock(&pc->pc_slock);
                   2127:
1.102     chs      2128:        pcg = LIST_FIRST(&pc->pc_partgroups);
                   2129:        if (pcg == NULL) {
                   2130:                pcg = LIST_FIRST(&pc->pc_emptygroups);
                   2131:                if (pcg != NULL) {
                   2132:                        LIST_REMOVE(pcg, pcg_list);
                   2133:                        LIST_INSERT_HEAD(&pc->pc_partgroups, pcg, pcg_list);
1.43      thorpej  2134:                }
1.102     chs      2135:        }
                   2136:        if (pcg == NULL) {
1.43      thorpej  2137:
                   2138:                /*
                   2139:                 * No empty groups to free the object to.  Attempt to
1.47      thorpej  2140:                 * allocate one.
1.43      thorpej  2141:                 */
1.47      thorpej  2142:                simple_unlock(&pc->pc_slock);
1.60      thorpej  2143:                s = splvm();
1.43      thorpej  2144:                pcg = pool_get(&pcgpool, PR_NOWAIT);
1.60      thorpej  2145:                splx(s);
1.102     chs      2146:                if (pcg == NULL) {
1.109     christos 2147: destruct:
1.102     chs      2148:
                   2149:                        /*
                   2150:                         * Unable to allocate a cache group; destruct the object
                   2151:                         * and free it back to the pool.
                   2152:                         */
                   2153:                        pool_cache_destruct_object(pc, object);
                   2154:                        return;
1.43      thorpej  2155:                }
1.102     chs      2156:                memset(pcg, 0, sizeof(*pcg));
                   2157:                simple_lock(&pc->pc_slock);
                   2158:                pc->pc_ngroups++;
                   2159:                LIST_INSERT_HEAD(&pc->pc_partgroups, pcg, pcg_list);
1.43      thorpej  2160:        }
                   2161:
1.48      thorpej  2162:        pc->pc_nitems++;
1.87      thorpej  2163:        pcg_put(pcg, object, pa);
1.43      thorpej  2164:
1.102     chs      2165:        if (pcg->pcg_avail == PCG_NOBJECTS) {
                   2166:                LIST_REMOVE(pcg, pcg_list);
                   2167:                LIST_INSERT_HEAD(&pc->pc_fullgroups, pcg, pcg_list);
                   2168:        }
1.43      thorpej  2169:        simple_unlock(&pc->pc_slock);
1.51      thorpej  2170: }
                   2171:
                   2172: /*
                   2173:  * pool_cache_destruct_object:
                   2174:  *
                   2175:  *     Force destruction of an object and its release back into
                   2176:  *     the pool.
                   2177:  */
                   2178: void
                   2179: pool_cache_destruct_object(struct pool_cache *pc, void *object)
                   2180: {
                   2181:
                   2182:        if (pc->pc_dtor != NULL)
                   2183:                (*pc->pc_dtor)(pc->pc_arg, object);
                   2184:        pool_put(pc->pc_pool, object);
1.43      thorpej  2185: }
                   2186:
1.102     chs      2187: static void
1.106     christos 2188: pool_do_cache_invalidate_grouplist(struct pool_cache_grouplist *pcgsl,
1.105     christos 2189:     struct pool_cache *pc, struct pool_pagelist *pq,
1.106     christos 2190:     struct pool_cache_grouplist *pcgdl)
1.102     chs      2191: {
1.106     christos 2192:        struct pool_cache_group *pcg, *npcg;
1.102     chs      2193:        void *object;
                   2194:
1.106     christos 2195:        for (pcg = LIST_FIRST(pcgsl); pcg != NULL; pcg = npcg) {
1.102     chs      2196:                npcg = LIST_NEXT(pcg, pcg_list);
                   2197:                while (pcg->pcg_avail != 0) {
                   2198:                        pc->pc_nitems--;
                   2199:                        object = pcg_get(pcg, NULL);
                   2200:                        if (pc->pc_dtor != NULL)
                   2201:                                (*pc->pc_dtor)(pc->pc_arg, object);
                   2202:                        pool_do_put(pc->pc_pool, object, pq);
                   2203:                }
1.103     chs      2204:                pc->pc_ngroups--;
1.102     chs      2205:                LIST_REMOVE(pcg, pcg_list);
1.106     christos 2206:                LIST_INSERT_HEAD(pcgdl, pcg, pcg_list);
1.102     chs      2207:        }
1.105     christos 2208: }
                   2209:
                   2210: static void
                   2211: pool_do_cache_invalidate(struct pool_cache *pc, struct pool_pagelist *pq,
                   2212:     struct pool_cache_grouplist *pcgl)
                   2213: {
                   2214:
                   2215:        LOCK_ASSERT(simple_lock_held(&pc->pc_slock));
                   2216:        LOCK_ASSERT(simple_lock_held(&pc->pc_pool->pr_slock));
                   2217:
1.106     christos 2218:        pool_do_cache_invalidate_grouplist(&pc->pc_fullgroups, pc, pq, pcgl);
                   2219:        pool_do_cache_invalidate_grouplist(&pc->pc_partgroups, pc, pq, pcgl);
1.103     chs      2220:
                   2221:        KASSERT(LIST_EMPTY(&pc->pc_partgroups));
                   2222:        KASSERT(LIST_EMPTY(&pc->pc_fullgroups));
                   2223:        KASSERT(pc->pc_nitems == 0);
1.102     chs      2224: }
                   2225:
1.43      thorpej  2226: /*
1.101     thorpej  2227:  * pool_cache_invalidate:
1.43      thorpej  2228:  *
1.101     thorpej  2229:  *     Invalidate a pool cache (destruct and release all of the
                   2230:  *     cached objects).
1.43      thorpej  2231:  */
1.101     thorpej  2232: void
                   2233: pool_cache_invalidate(struct pool_cache *pc)
1.43      thorpej  2234: {
1.101     thorpej  2235:        struct pool_pagelist pq;
1.102     chs      2236:        struct pool_cache_grouplist pcgl;
1.101     thorpej  2237:
                   2238:        LIST_INIT(&pq);
1.102     chs      2239:        LIST_INIT(&pcgl);
1.101     thorpej  2240:
                   2241:        simple_lock(&pc->pc_slock);
                   2242:        simple_lock(&pc->pc_pool->pr_slock);
1.43      thorpej  2243:
1.102     chs      2244:        pool_do_cache_invalidate(pc, &pq, &pcgl);
1.43      thorpej  2245:
1.101     thorpej  2246:        simple_unlock(&pc->pc_pool->pr_slock);
                   2247:        simple_unlock(&pc->pc_slock);
1.43      thorpej  2248:
1.102     chs      2249:        pr_pagelist_free(pc->pc_pool, &pq);
                   2250:        pcg_grouplist_free(&pcgl);
1.43      thorpej  2251: }
                   2252:
                   2253: /*
                   2254:  * pool_cache_reclaim:
                   2255:  *
                   2256:  *     Reclaim a pool cache for pool_reclaim().
                   2257:  */
                   2258: static void
1.102     chs      2259: pool_cache_reclaim(struct pool_cache *pc, struct pool_pagelist *pq,
                   2260:     struct pool_cache_grouplist *pcgl)
1.43      thorpej  2261: {
1.101     thorpej  2262:
                   2263:        /*
                   2264:         * We're locking in the wrong order (normally pool_cache -> pool,
                   2265:         * but the pool is already locked when we get here), so we have
                   2266:         * to use trylock.  If we can't lock the pool_cache, it's not really
                   2267:         * a big deal here.
                   2268:         */
                   2269:        if (simple_lock_try(&pc->pc_slock) == 0)
                   2270:                return;
                   2271:
1.102     chs      2272:        pool_do_cache_invalidate(pc, pq, pcgl);
1.43      thorpej  2273:
                   2274:        simple_unlock(&pc->pc_slock);
1.3       pk       2275: }
1.66      thorpej  2276:
                   2277: /*
                   2278:  * Pool backend allocators.
                   2279:  *
                   2280:  * Each pool has a backend allocator that handles allocation, deallocation,
                   2281:  * and any additional draining that might be needed.
                   2282:  *
                   2283:  * We provide two standard allocators:
                   2284:  *
                   2285:  *     pool_allocator_kmem - the default when no allocator is specified
                   2286:  *
                   2287:  *     pool_allocator_nointr - used for pools that will not be accessed
                   2288:  *     in interrupt context.
                   2289:  */
                   2290: void   *pool_page_alloc(struct pool *, int);
                   2291: void   pool_page_free(struct pool *, void *);
                   2292:
1.112     bjh21    2293: #ifdef POOL_SUBPAGE
                   2294: struct pool_allocator pool_allocator_kmem_fullpage = {
                   2295:        pool_page_alloc, pool_page_free, 0,
1.117     yamt     2296:        .pa_backingmapptr = &kmem_map,
1.112     bjh21    2297: };
                   2298: #else
1.66      thorpej  2299: struct pool_allocator pool_allocator_kmem = {
                   2300:        pool_page_alloc, pool_page_free, 0,
1.117     yamt     2301:        .pa_backingmapptr = &kmem_map,
1.66      thorpej  2302: };
1.112     bjh21    2303: #endif
1.66      thorpej  2304:
                   2305: void   *pool_page_alloc_nointr(struct pool *, int);
                   2306: void   pool_page_free_nointr(struct pool *, void *);
                   2307:
1.112     bjh21    2308: #ifdef POOL_SUBPAGE
                   2309: struct pool_allocator pool_allocator_nointr_fullpage = {
                   2310:        pool_page_alloc_nointr, pool_page_free_nointr, 0,
1.117     yamt     2311:        .pa_backingmapptr = &kernel_map,
1.112     bjh21    2312: };
                   2313: #else
1.66      thorpej  2314: struct pool_allocator pool_allocator_nointr = {
                   2315:        pool_page_alloc_nointr, pool_page_free_nointr, 0,
1.117     yamt     2316:        .pa_backingmapptr = &kernel_map,
1.66      thorpej  2317: };
1.112     bjh21    2318: #endif
1.66      thorpej  2319:
                   2320: #ifdef POOL_SUBPAGE
                   2321: void   *pool_subpage_alloc(struct pool *, int);
                   2322: void   pool_subpage_free(struct pool *, void *);
                   2323:
1.112     bjh21    2324: struct pool_allocator pool_allocator_kmem = {
                   2325:        pool_subpage_alloc, pool_subpage_free, POOL_SUBPAGE,
1.117     yamt     2326:        .pa_backingmapptr = &kmem_map,
1.112     bjh21    2327: };
                   2328:
                   2329: void   *pool_subpage_alloc_nointr(struct pool *, int);
                   2330: void   pool_subpage_free_nointr(struct pool *, void *);
                   2331:
                   2332: struct pool_allocator pool_allocator_nointr = {
                   2333:        pool_subpage_alloc, pool_subpage_free, POOL_SUBPAGE,
1.117     yamt     2334:        .pa_backingmapptr = &kmem_map,
1.66      thorpej  2335: };
                   2336: #endif /* POOL_SUBPAGE */
                   2337:
1.117     yamt     2338: static void *
                   2339: pool_allocator_alloc(struct pool *pp, int flags)
1.66      thorpej  2340: {
1.117     yamt     2341:        struct pool_allocator *pa = pp->pr_alloc;
1.66      thorpej  2342:        void *res;
                   2343:
1.117     yamt     2344:        LOCK_ASSERT(!simple_lock_held(&pp->pr_slock));
1.66      thorpej  2345:
1.117     yamt     2346:        res = (*pa->pa_alloc)(pp, flags);
                   2347:        if (res == NULL && (flags & PR_WAITOK) == 0) {
1.66      thorpej  2348:                /*
1.117     yamt     2349:                 * We only run the drain hook here if PR_NOWAIT.
                   2350:                 * In other cases, the hook will be run in
                   2351:                 * pool_reclaim().
1.66      thorpej  2352:                 */
1.117     yamt     2353:                if (pp->pr_drain_hook != NULL) {
                   2354:                        (*pp->pr_drain_hook)(pp->pr_drain_hook_arg, flags);
                   2355:                        res = (*pa->pa_alloc)(pp, flags);
1.66      thorpej  2356:                }
1.117     yamt     2357:        }
                   2358:        return res;
1.66      thorpej  2359: }
                   2360:
1.117     yamt     2361: static void
1.66      thorpej  2362: pool_allocator_free(struct pool *pp, void *v)
                   2363: {
                   2364:        struct pool_allocator *pa = pp->pr_alloc;
                   2365:
1.91      yamt     2366:        LOCK_ASSERT(!simple_lock_held(&pp->pr_slock));
                   2367:
1.66      thorpej  2368:        (*pa->pa_free)(pp, v);
                   2369: }
                   2370:
                   2371: void *
1.124     yamt     2372: pool_page_alloc(struct pool *pp, int flags)
1.66      thorpej  2373: {
1.127     thorpej  2374:        bool waitok = (flags & PR_WAITOK) ? true : false;
1.66      thorpej  2375:
1.100     yamt     2376:        return ((void *) uvm_km_alloc_poolpage_cache(kmem_map, waitok));
1.66      thorpej  2377: }
                   2378:
                   2379: void
1.124     yamt     2380: pool_page_free(struct pool *pp, void *v)
1.66      thorpej  2381: {
                   2382:
1.98      yamt     2383:        uvm_km_free_poolpage_cache(kmem_map, (vaddr_t) v);
                   2384: }
                   2385:
                   2386: static void *
1.124     yamt     2387: pool_page_alloc_meta(struct pool *pp, int flags)
1.98      yamt     2388: {
1.127     thorpej  2389:        bool waitok = (flags & PR_WAITOK) ? true : false;
1.98      yamt     2390:
1.100     yamt     2391:        return ((void *) uvm_km_alloc_poolpage(kmem_map, waitok));
1.98      yamt     2392: }
                   2393:
                   2394: static void
1.124     yamt     2395: pool_page_free_meta(struct pool *pp, void *v)
1.98      yamt     2396: {
                   2397:
1.100     yamt     2398:        uvm_km_free_poolpage(kmem_map, (vaddr_t) v);
1.66      thorpej  2399: }
                   2400:
                   2401: #ifdef POOL_SUBPAGE
                   2402: /* Sub-page allocator, for machines with large hardware pages. */
                   2403: void *
                   2404: pool_subpage_alloc(struct pool *pp, int flags)
                   2405: {
1.93      dbj      2406:        void *v;
                   2407:        int s;
                   2408:        s = splvm();
                   2409:        v = pool_get(&psppool, flags);
                   2410:        splx(s);
                   2411:        return v;
1.66      thorpej  2412: }
                   2413:
                   2414: void
                   2415: pool_subpage_free(struct pool *pp, void *v)
                   2416: {
1.93      dbj      2417:        int s;
                   2418:        s = splvm();
1.66      thorpej  2419:        pool_put(&psppool, v);
1.93      dbj      2420:        splx(s);
1.66      thorpej  2421: }
                   2422:
                   2423: /* We don't provide a real nointr allocator.  Maybe later. */
                   2424: void *
1.112     bjh21    2425: pool_subpage_alloc_nointr(struct pool *pp, int flags)
1.66      thorpej  2426: {
                   2427:
                   2428:        return (pool_subpage_alloc(pp, flags));
                   2429: }
                   2430:
                   2431: void
1.112     bjh21    2432: pool_subpage_free_nointr(struct pool *pp, void *v)
1.66      thorpej  2433: {
                   2434:
                   2435:        pool_subpage_free(pp, v);
                   2436: }
1.112     bjh21    2437: #endif /* POOL_SUBPAGE */
1.66      thorpej  2438: void *
1.124     yamt     2439: pool_page_alloc_nointr(struct pool *pp, int flags)
1.66      thorpej  2440: {
1.127     thorpej  2441:        bool waitok = (flags & PR_WAITOK) ? true : false;
1.66      thorpej  2442:
1.100     yamt     2443:        return ((void *) uvm_km_alloc_poolpage_cache(kernel_map, waitok));
1.66      thorpej  2444: }
                   2445:
                   2446: void
1.124     yamt     2447: pool_page_free_nointr(struct pool *pp, void *v)
1.66      thorpej  2448: {
                   2449:
1.98      yamt     2450:        uvm_km_free_poolpage_cache(kernel_map, (vaddr_t) v);
1.66      thorpej  2451: }

CVSweb <webmaster@jp.NetBSD.org>