[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 1.189

1.189   ! pooka       1: /*     $NetBSD: subr_pool.c,v 1.188 2011/01/17 07:36:58 uebayasi Exp $ */
1.1       pk          2:
                      3: /*-
1.183     ad          4:  * Copyright (c) 1997, 1999, 2000, 2002, 2007, 2008, 2010
                      5:  *     The NetBSD Foundation, Inc.
1.1       pk          6:  * All rights reserved.
                      7:  *
                      8:  * This code is derived from software contributed to The NetBSD Foundation
1.20      thorpej     9:  * by Paul Kranenburg; by Jason R. Thorpe of the Numerical Aerospace
1.134     ad         10:  * Simulation Facility, NASA Ames Research Center, and by Andrew Doran.
1.1       pk         11:  *
                     12:  * Redistribution and use in source and binary forms, with or without
                     13:  * modification, are permitted provided that the following conditions
                     14:  * are met:
                     15:  * 1. Redistributions of source code must retain the above copyright
                     16:  *    notice, this list of conditions and the following disclaimer.
                     17:  * 2. Redistributions in binary form must reproduce the above copyright
                     18:  *    notice, this list of conditions and the following disclaimer in the
                     19:  *    documentation and/or other materials provided with the distribution.
                     20:  *
                     21:  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
                     22:  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
                     23:  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
                     24:  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
                     25:  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
                     26:  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
                     27:  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
                     28:  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
                     29:  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
                     30:  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
                     31:  * POSSIBILITY OF SUCH DAMAGE.
                     32:  */
1.64      lukem      33:
                     34: #include <sys/cdefs.h>
1.189   ! pooka      35: __KERNEL_RCSID(0, "$NetBSD: subr_pool.c,v 1.188 2011/01/17 07:36:58 uebayasi Exp $");
1.24      scottr     36:
1.141     yamt       37: #include "opt_ddb.h"
1.25      thorpej    38: #include "opt_pool.h"
1.24      scottr     39: #include "opt_poollog.h"
1.28      thorpej    40: #include "opt_lockdebug.h"
1.1       pk         41:
                     42: #include <sys/param.h>
                     43: #include <sys/systm.h>
1.135     yamt       44: #include <sys/bitops.h>
1.1       pk         45: #include <sys/proc.h>
                     46: #include <sys/errno.h>
                     47: #include <sys/kernel.h>
                     48: #include <sys/malloc.h>
                     49: #include <sys/pool.h>
1.20      thorpej    50: #include <sys/syslog.h>
1.125     ad         51: #include <sys/debug.h>
1.134     ad         52: #include <sys/lockdebug.h>
                     53: #include <sys/xcall.h>
                     54: #include <sys/cpu.h>
1.145     ad         55: #include <sys/atomic.h>
1.3       pk         56:
1.187     uebayasi   57: #include <uvm/uvm_extern.h>
1.188     uebayasi   58: #ifdef DIAGNOSTIC
                     59: #include <uvm/uvm_km.h>        /* uvm_km_va_drain */
                     60: #endif
1.3       pk         61:
1.1       pk         62: /*
                     63:  * Pool resource management utility.
1.3       pk         64:  *
1.88      chs        65:  * Memory is allocated in pages which are split into pieces according to
                     66:  * the pool item size. Each page is kept on one of three lists in the
                     67:  * pool structure: `pr_emptypages', `pr_fullpages' and `pr_partpages',
                     68:  * for empty, full and partially-full pages respectively. The individual
                     69:  * pool items are on a linked list headed by `ph_itemlist' in each page
                     70:  * header. The memory for building the page list is either taken from
                     71:  * the allocated pages themselves (for small pool items) or taken from
                     72:  * an internal pool of page headers (`phpool').
1.1       pk         73:  */
                     74:
1.3       pk         75: /* List of all pools */
1.173     rmind      76: static TAILQ_HEAD(, pool) pool_head = TAILQ_HEAD_INITIALIZER(pool_head);
1.134     ad         77:
1.3       pk         78: /* Private pool for page header structures */
1.97      yamt       79: #define        PHPOOL_MAX      8
                     80: static struct pool phpool[PHPOOL_MAX];
1.135     yamt       81: #define        PHPOOL_FREELIST_NELEM(idx) \
                     82:        (((idx) == 0) ? 0 : BITMAP_SIZE * (1 << (idx)))
1.3       pk         83:
1.62      bjh21      84: #ifdef POOL_SUBPAGE
                     85: /* Pool of subpages for use by normal pools. */
                     86: static struct pool psppool;
                     87: #endif
                     88:
1.117     yamt       89: static SLIST_HEAD(, pool_allocator) pa_deferinitq =
                     90:     SLIST_HEAD_INITIALIZER(pa_deferinitq);
                     91:
1.98      yamt       92: static void *pool_page_alloc_meta(struct pool *, int);
                     93: static void pool_page_free_meta(struct pool *, void *);
                     94:
                     95: /* allocator for pool metadata */
1.134     ad         96: struct pool_allocator pool_allocator_meta = {
1.117     yamt       97:        pool_page_alloc_meta, pool_page_free_meta,
                     98:        .pa_backingmapptr = &kmem_map,
1.98      yamt       99: };
                    100:
1.3       pk        101: /* # of seconds to retain page after last use */
                    102: int pool_inactive_time = 10;
                    103:
                    104: /* Next candidate for drainage (see pool_drain()) */
1.23      thorpej   105: static struct pool     *drainpp;
                    106:
1.134     ad        107: /* This lock protects both pool_head and drainpp. */
                    108: static kmutex_t pool_head_lock;
                    109: static kcondvar_t pool_busy;
1.3       pk        110:
1.178     elad      111: /* This lock protects initialization of a potentially shared pool allocator */
                    112: static kmutex_t pool_allocator_lock;
                    113:
1.135     yamt      114: typedef uint32_t pool_item_bitmap_t;
                    115: #define        BITMAP_SIZE     (CHAR_BIT * sizeof(pool_item_bitmap_t))
                    116: #define        BITMAP_MASK     (BITMAP_SIZE - 1)
1.99      yamt      117:
1.3       pk        118: struct pool_item_header {
                    119:        /* Page headers */
1.88      chs       120:        LIST_ENTRY(pool_item_header)
1.3       pk        121:                                ph_pagelist;    /* pool page list */
1.88      chs       122:        SPLAY_ENTRY(pool_item_header)
                    123:                                ph_node;        /* Off-page page headers */
1.128     christos  124:        void *                  ph_page;        /* this page's address */
1.151     yamt      125:        uint32_t                ph_time;        /* last referenced */
1.135     yamt      126:        uint16_t                ph_nmissing;    /* # of chunks in use */
1.141     yamt      127:        uint16_t                ph_off;         /* start offset in page */
1.97      yamt      128:        union {
                    129:                /* !PR_NOTOUCH */
                    130:                struct {
1.102     chs       131:                        LIST_HEAD(, pool_item)
1.97      yamt      132:                                phu_itemlist;   /* chunk list for this page */
                    133:                } phu_normal;
                    134:                /* PR_NOTOUCH */
                    135:                struct {
1.141     yamt      136:                        pool_item_bitmap_t phu_bitmap[1];
1.97      yamt      137:                } phu_notouch;
                    138:        } ph_u;
1.3       pk        139: };
1.97      yamt      140: #define        ph_itemlist     ph_u.phu_normal.phu_itemlist
1.135     yamt      141: #define        ph_bitmap       ph_u.phu_notouch.phu_bitmap
1.3       pk        142:
1.1       pk        143: struct pool_item {
1.3       pk        144: #ifdef DIAGNOSTIC
1.82      thorpej   145:        u_int pi_magic;
1.33      chs       146: #endif
1.134     ad        147: #define        PI_MAGIC 0xdeaddeadU
1.3       pk        148:        /* Other entries use only this list entry */
1.102     chs       149:        LIST_ENTRY(pool_item)   pi_list;
1.3       pk        150: };
                    151:
1.53      thorpej   152: #define        POOL_NEEDS_CATCHUP(pp)                                          \
                    153:        ((pp)->pr_nitems < (pp)->pr_minitems)
                    154:
1.43      thorpej   155: /*
                    156:  * Pool cache management.
                    157:  *
                    158:  * Pool caches provide a way for constructed objects to be cached by the
                    159:  * pool subsystem.  This can lead to performance improvements by avoiding
                    160:  * needless object construction/destruction; it is deferred until absolutely
                    161:  * necessary.
                    162:  *
1.134     ad        163:  * Caches are grouped into cache groups.  Each cache group references up
                    164:  * to PCG_NUMOBJECTS constructed objects.  When a cache allocates an
                    165:  * object from the pool, it calls the object's constructor and places it
                    166:  * into a cache group.  When a cache group frees an object back to the
                    167:  * pool, it first calls the object's destructor.  This allows the object
                    168:  * to persist in constructed form while freed to the cache.
                    169:  *
                    170:  * The pool references each cache, so that when a pool is drained by the
                    171:  * pagedaemon, it can drain each individual cache as well.  Each time a
                    172:  * cache is drained, the most idle cache group is freed to the pool in
                    173:  * its entirety.
1.43      thorpej   174:  *
                    175:  * Pool caches are layed on top of pools.  By layering them, we can avoid
                    176:  * the complexity of cache management for pools which would not benefit
                    177:  * from it.
                    178:  */
                    179:
1.142     ad        180: static struct pool pcg_normal_pool;
                    181: static struct pool pcg_large_pool;
1.134     ad        182: static struct pool cache_pool;
                    183: static struct pool cache_cpu_pool;
1.3       pk        184:
1.189   ! pooka     185: pool_cache_t pnbuf_cache;      /* pathname buffer cache */
        !           186:
1.145     ad        187: /* List of all caches. */
                    188: TAILQ_HEAD(,pool_cache) pool_cache_head =
                    189:     TAILQ_HEAD_INITIALIZER(pool_cache_head);
                    190:
1.162     ad        191: int pool_cache_disable;                /* global disable for caching */
1.169     yamt      192: static const pcg_t pcg_dummy;  /* zero sized: always empty, yet always full */
1.145     ad        193:
1.162     ad        194: static bool    pool_cache_put_slow(pool_cache_cpu_t *, int,
                    195:                                    void *);
                    196: static bool    pool_cache_get_slow(pool_cache_cpu_t *, int,
                    197:                                    void **, paddr_t *, int);
1.134     ad        198: static void    pool_cache_cpu_init1(struct cpu_info *, pool_cache_t);
                    199: static void    pool_cache_invalidate_groups(pool_cache_t, pcg_t *);
1.175     jym       200: static void    pool_cache_invalidate_cpu(pool_cache_t, u_int);
1.134     ad        201: static void    pool_cache_xcall(pool_cache_t);
1.3       pk        202:
1.42      thorpej   203: static int     pool_catchup(struct pool *);
1.128     christos  204: static void    pool_prime_page(struct pool *, void *,
1.55      thorpej   205:                    struct pool_item_header *);
1.88      chs       206: static void    pool_update_curpage(struct pool *);
1.66      thorpej   207:
1.113     yamt      208: static int     pool_grow(struct pool *, int);
1.117     yamt      209: static void    *pool_allocator_alloc(struct pool *, int);
                    210: static void    pool_allocator_free(struct pool *, void *);
1.3       pk        211:
1.97      yamt      212: static void pool_print_pagelist(struct pool *, struct pool_pagelist *,
1.88      chs       213:        void (*)(const char *, ...));
1.42      thorpej   214: static void pool_print1(struct pool *, const char *,
                    215:        void (*)(const char *, ...));
1.3       pk        216:
1.88      chs       217: static int pool_chk_page(struct pool *, const char *,
                    218:                         struct pool_item_header *);
                    219:
1.3       pk        220: /*
1.52      thorpej   221:  * Pool log entry. An array of these is allocated in pool_init().
1.3       pk        222:  */
                    223: struct pool_log {
                    224:        const char      *pl_file;
                    225:        long            pl_line;
                    226:        int             pl_action;
1.25      thorpej   227: #define        PRLOG_GET       1
                    228: #define        PRLOG_PUT       2
1.3       pk        229:        void            *pl_addr;
1.1       pk        230: };
                    231:
1.86      matt      232: #ifdef POOL_DIAGNOSTIC
1.3       pk        233: /* Number of entries in pool log buffers */
1.17      thorpej   234: #ifndef POOL_LOGSIZE
                    235: #define        POOL_LOGSIZE    10
                    236: #endif
                    237:
                    238: int pool_logsize = POOL_LOGSIZE;
1.1       pk        239:
1.110     perry     240: static inline void
1.42      thorpej   241: pr_log(struct pool *pp, void *v, int action, const char *file, long line)
1.3       pk        242: {
1.179     mlelstv   243:        int n;
1.3       pk        244:        struct pool_log *pl;
                    245:
1.20      thorpej   246:        if ((pp->pr_roflags & PR_LOGGING) == 0)
1.3       pk        247:                return;
                    248:
1.179     mlelstv   249:        if (pp->pr_log == NULL) {
                    250:                if (kmem_map != NULL)
                    251:                        pp->pr_log = malloc(
                    252:                                pool_logsize * sizeof(struct pool_log),
                    253:                                M_TEMP, M_NOWAIT | M_ZERO);
                    254:                if (pp->pr_log == NULL)
                    255:                        return;
                    256:                pp->pr_curlogentry = 0;
                    257:                pp->pr_logsize = pool_logsize;
                    258:        }
                    259:
1.3       pk        260:        /*
                    261:         * Fill in the current entry. Wrap around and overwrite
                    262:         * the oldest entry if necessary.
                    263:         */
1.179     mlelstv   264:        n = pp->pr_curlogentry;
1.3       pk        265:        pl = &pp->pr_log[n];
                    266:        pl->pl_file = file;
                    267:        pl->pl_line = line;
                    268:        pl->pl_action = action;
                    269:        pl->pl_addr = v;
                    270:        if (++n >= pp->pr_logsize)
                    271:                n = 0;
                    272:        pp->pr_curlogentry = n;
                    273: }
                    274:
                    275: static void
1.42      thorpej   276: pr_printlog(struct pool *pp, struct pool_item *pi,
                    277:     void (*pr)(const char *, ...))
1.3       pk        278: {
                    279:        int i = pp->pr_logsize;
                    280:        int n = pp->pr_curlogentry;
                    281:
1.179     mlelstv   282:        if (pp->pr_log == NULL)
1.3       pk        283:                return;
                    284:
                    285:        /*
                    286:         * Print all entries in this pool's log.
                    287:         */
                    288:        while (i-- > 0) {
                    289:                struct pool_log *pl = &pp->pr_log[n];
                    290:                if (pl->pl_action != 0) {
1.25      thorpej   291:                        if (pi == NULL || pi == pl->pl_addr) {
                    292:                                (*pr)("\tlog entry %d:\n", i);
                    293:                                (*pr)("\t\taction = %s, addr = %p\n",
                    294:                                    pl->pl_action == PRLOG_GET ? "get" : "put",
                    295:                                    pl->pl_addr);
                    296:                                (*pr)("\t\tfile: %s at line %lu\n",
                    297:                                    pl->pl_file, pl->pl_line);
                    298:                        }
1.3       pk        299:                }
                    300:                if (++n >= pp->pr_logsize)
                    301:                        n = 0;
                    302:        }
                    303: }
1.25      thorpej   304:
1.110     perry     305: static inline void
1.42      thorpej   306: pr_enter(struct pool *pp, const char *file, long line)
1.25      thorpej   307: {
                    308:
1.34      thorpej   309:        if (__predict_false(pp->pr_entered_file != NULL)) {
1.25      thorpej   310:                printf("pool %s: reentrancy at file %s line %ld\n",
                    311:                    pp->pr_wchan, file, line);
                    312:                printf("         previous entry at file %s line %ld\n",
                    313:                    pp->pr_entered_file, pp->pr_entered_line);
                    314:                panic("pr_enter");
                    315:        }
                    316:
                    317:        pp->pr_entered_file = file;
                    318:        pp->pr_entered_line = line;
                    319: }
                    320:
1.110     perry     321: static inline void
1.42      thorpej   322: pr_leave(struct pool *pp)
1.25      thorpej   323: {
                    324:
1.34      thorpej   325:        if (__predict_false(pp->pr_entered_file == NULL)) {
1.25      thorpej   326:                printf("pool %s not entered?\n", pp->pr_wchan);
                    327:                panic("pr_leave");
                    328:        }
                    329:
                    330:        pp->pr_entered_file = NULL;
                    331:        pp->pr_entered_line = 0;
                    332: }
                    333:
1.110     perry     334: static inline void
1.42      thorpej   335: pr_enter_check(struct pool *pp, void (*pr)(const char *, ...))
1.25      thorpej   336: {
                    337:
                    338:        if (pp->pr_entered_file != NULL)
                    339:                (*pr)("\n\tcurrently entered from file %s line %ld\n",
                    340:                    pp->pr_entered_file, pp->pr_entered_line);
                    341: }
1.3       pk        342: #else
1.25      thorpej   343: #define        pr_log(pp, v, action, file, line)
                    344: #define        pr_printlog(pp, pi, pr)
                    345: #define        pr_enter(pp, file, line)
                    346: #define        pr_leave(pp)
                    347: #define        pr_enter_check(pp, pr)
1.59      thorpej   348: #endif /* POOL_DIAGNOSTIC */
1.3       pk        349:
1.135     yamt      350: static inline unsigned int
1.97      yamt      351: pr_item_notouch_index(const struct pool *pp, const struct pool_item_header *ph,
                    352:     const void *v)
                    353: {
                    354:        const char *cp = v;
1.135     yamt      355:        unsigned int idx;
1.97      yamt      356:
                    357:        KASSERT(pp->pr_roflags & PR_NOTOUCH);
1.128     christos  358:        idx = (cp - (char *)ph->ph_page - ph->ph_off) / pp->pr_size;
1.97      yamt      359:        KASSERT(idx < pp->pr_itemsperpage);
                    360:        return idx;
                    361: }
                    362:
1.110     perry     363: static inline void
1.97      yamt      364: pr_item_notouch_put(const struct pool *pp, struct pool_item_header *ph,
                    365:     void *obj)
                    366: {
1.135     yamt      367:        unsigned int idx = pr_item_notouch_index(pp, ph, obj);
                    368:        pool_item_bitmap_t *bitmap = ph->ph_bitmap + (idx / BITMAP_SIZE);
                    369:        pool_item_bitmap_t mask = 1 << (idx & BITMAP_MASK);
1.97      yamt      370:
1.135     yamt      371:        KASSERT((*bitmap & mask) == 0);
                    372:        *bitmap |= mask;
1.97      yamt      373: }
                    374:
1.110     perry     375: static inline void *
1.97      yamt      376: pr_item_notouch_get(const struct pool *pp, struct pool_item_header *ph)
                    377: {
1.135     yamt      378:        pool_item_bitmap_t *bitmap = ph->ph_bitmap;
                    379:        unsigned int idx;
                    380:        int i;
1.97      yamt      381:
1.135     yamt      382:        for (i = 0; ; i++) {
                    383:                int bit;
1.97      yamt      384:
1.135     yamt      385:                KASSERT((i * BITMAP_SIZE) < pp->pr_itemsperpage);
                    386:                bit = ffs32(bitmap[i]);
                    387:                if (bit) {
                    388:                        pool_item_bitmap_t mask;
                    389:
                    390:                        bit--;
                    391:                        idx = (i * BITMAP_SIZE) + bit;
                    392:                        mask = 1 << bit;
                    393:                        KASSERT((bitmap[i] & mask) != 0);
                    394:                        bitmap[i] &= ~mask;
                    395:                        break;
                    396:                }
                    397:        }
                    398:        KASSERT(idx < pp->pr_itemsperpage);
1.128     christos  399:        return (char *)ph->ph_page + ph->ph_off + idx * pp->pr_size;
1.97      yamt      400: }
                    401:
1.135     yamt      402: static inline void
1.141     yamt      403: pr_item_notouch_init(const struct pool *pp, struct pool_item_header *ph)
1.135     yamt      404: {
                    405:        pool_item_bitmap_t *bitmap = ph->ph_bitmap;
                    406:        const int n = howmany(pp->pr_itemsperpage, BITMAP_SIZE);
                    407:        int i;
                    408:
                    409:        for (i = 0; i < n; i++) {
                    410:                bitmap[i] = (pool_item_bitmap_t)-1;
                    411:        }
                    412: }
                    413:
1.110     perry     414: static inline int
1.88      chs       415: phtree_compare(struct pool_item_header *a, struct pool_item_header *b)
                    416: {
1.121     yamt      417:
                    418:        /*
                    419:         * we consider pool_item_header with smaller ph_page bigger.
                    420:         * (this unnatural ordering is for the benefit of pr_find_pagehead.)
                    421:         */
                    422:
1.88      chs       423:        if (a->ph_page < b->ph_page)
1.121     yamt      424:                return (1);
                    425:        else if (a->ph_page > b->ph_page)
1.88      chs       426:                return (-1);
                    427:        else
                    428:                return (0);
                    429: }
                    430:
                    431: SPLAY_PROTOTYPE(phtree, pool_item_header, ph_node, phtree_compare);
                    432: SPLAY_GENERATE(phtree, pool_item_header, ph_node, phtree_compare);
                    433:
1.141     yamt      434: static inline struct pool_item_header *
                    435: pr_find_pagehead_noalign(struct pool *pp, void *v)
                    436: {
                    437:        struct pool_item_header *ph, tmp;
                    438:
                    439:        tmp.ph_page = (void *)(uintptr_t)v;
                    440:        ph = SPLAY_FIND(phtree, &pp->pr_phtree, &tmp);
                    441:        if (ph == NULL) {
                    442:                ph = SPLAY_ROOT(&pp->pr_phtree);
                    443:                if (ph != NULL && phtree_compare(&tmp, ph) >= 0) {
                    444:                        ph = SPLAY_NEXT(phtree, &pp->pr_phtree, ph);
                    445:                }
                    446:                KASSERT(ph == NULL || phtree_compare(&tmp, ph) < 0);
                    447:        }
                    448:
                    449:        return ph;
                    450: }
                    451:
1.3       pk        452: /*
1.121     yamt      453:  * Return the pool page header based on item address.
1.3       pk        454:  */
1.110     perry     455: static inline struct pool_item_header *
1.121     yamt      456: pr_find_pagehead(struct pool *pp, void *v)
1.3       pk        457: {
1.88      chs       458:        struct pool_item_header *ph, tmp;
1.3       pk        459:
1.121     yamt      460:        if ((pp->pr_roflags & PR_NOALIGN) != 0) {
1.141     yamt      461:                ph = pr_find_pagehead_noalign(pp, v);
1.121     yamt      462:        } else {
1.128     christos  463:                void *page =
                    464:                    (void *)((uintptr_t)v & pp->pr_alloc->pa_pagemask);
1.121     yamt      465:
                    466:                if ((pp->pr_roflags & PR_PHINPAGE) != 0) {
1.128     christos  467:                        ph = (struct pool_item_header *)((char *)page + pp->pr_phoffset);
1.121     yamt      468:                } else {
                    469:                        tmp.ph_page = page;
                    470:                        ph = SPLAY_FIND(phtree, &pp->pr_phtree, &tmp);
                    471:                }
                    472:        }
1.3       pk        473:
1.121     yamt      474:        KASSERT(ph == NULL || ((pp->pr_roflags & PR_PHINPAGE) != 0) ||
1.128     christos  475:            ((char *)ph->ph_page <= (char *)v &&
                    476:            (char *)v < (char *)ph->ph_page + pp->pr_alloc->pa_pagesz));
1.88      chs       477:        return ph;
1.3       pk        478: }
                    479:
1.101     thorpej   480: static void
                    481: pr_pagelist_free(struct pool *pp, struct pool_pagelist *pq)
                    482: {
                    483:        struct pool_item_header *ph;
                    484:
                    485:        while ((ph = LIST_FIRST(pq)) != NULL) {
                    486:                LIST_REMOVE(ph, ph_pagelist);
                    487:                pool_allocator_free(pp, ph->ph_page);
1.134     ad        488:                if ((pp->pr_roflags & PR_PHINPAGE) == 0)
1.101     thorpej   489:                        pool_put(pp->pr_phpool, ph);
                    490:        }
                    491: }
                    492:
1.3       pk        493: /*
                    494:  * Remove a page from the pool.
                    495:  */
1.110     perry     496: static inline void
1.61      chs       497: pr_rmpage(struct pool *pp, struct pool_item_header *ph,
                    498:      struct pool_pagelist *pq)
1.3       pk        499: {
                    500:
1.134     ad        501:        KASSERT(mutex_owned(&pp->pr_lock));
1.91      yamt      502:
1.3       pk        503:        /*
1.7       thorpej   504:         * If the page was idle, decrement the idle page count.
1.3       pk        505:         */
1.6       thorpej   506:        if (ph->ph_nmissing == 0) {
                    507: #ifdef DIAGNOSTIC
                    508:                if (pp->pr_nidle == 0)
                    509:                        panic("pr_rmpage: nidle inconsistent");
1.20      thorpej   510:                if (pp->pr_nitems < pp->pr_itemsperpage)
                    511:                        panic("pr_rmpage: nitems inconsistent");
1.6       thorpej   512: #endif
                    513:                pp->pr_nidle--;
                    514:        }
1.7       thorpej   515:
1.20      thorpej   516:        pp->pr_nitems -= pp->pr_itemsperpage;
                    517:
1.7       thorpej   518:        /*
1.101     thorpej   519:         * Unlink the page from the pool and queue it for release.
1.7       thorpej   520:         */
1.88      chs       521:        LIST_REMOVE(ph, ph_pagelist);
1.91      yamt      522:        if ((pp->pr_roflags & PR_PHINPAGE) == 0)
                    523:                SPLAY_REMOVE(phtree, &pp->pr_phtree, ph);
1.101     thorpej   524:        LIST_INSERT_HEAD(pq, ph, ph_pagelist);
                    525:
1.7       thorpej   526:        pp->pr_npages--;
                    527:        pp->pr_npagefree++;
1.6       thorpej   528:
1.88      chs       529:        pool_update_curpage(pp);
1.3       pk        530: }
                    531:
1.126     thorpej   532: static bool
1.117     yamt      533: pa_starved_p(struct pool_allocator *pa)
                    534: {
                    535:
                    536:        if (pa->pa_backingmap != NULL) {
                    537:                return vm_map_starved_p(pa->pa_backingmap);
                    538:        }
1.127     thorpej   539:        return false;
1.117     yamt      540: }
                    541:
                    542: static int
1.124     yamt      543: pool_reclaim_callback(struct callback_entry *ce, void *obj, void *arg)
1.117     yamt      544: {
                    545:        struct pool *pp = obj;
                    546:        struct pool_allocator *pa = pp->pr_alloc;
                    547:
                    548:        KASSERT(&pp->pr_reclaimerentry == ce);
                    549:        pool_reclaim(pp);
                    550:        if (!pa_starved_p(pa)) {
                    551:                return CALLBACK_CHAIN_ABORT;
                    552:        }
                    553:        return CALLBACK_CHAIN_CONTINUE;
                    554: }
                    555:
                    556: static void
                    557: pool_reclaim_register(struct pool *pp)
                    558: {
                    559:        struct vm_map *map = pp->pr_alloc->pa_backingmap;
                    560:        int s;
                    561:
                    562:        if (map == NULL) {
                    563:                return;
                    564:        }
                    565:
                    566:        s = splvm(); /* not necessary for INTRSAFE maps, but don't care. */
                    567:        callback_register(&vm_map_to_kernel(map)->vmk_reclaim_callback,
                    568:            &pp->pr_reclaimerentry, pp, pool_reclaim_callback);
                    569:        splx(s);
1.184     rmind     570:
                    571: #ifdef DIAGNOSTIC
                    572:        /* Diagnostic drain attempt. */
                    573:        uvm_km_va_drain(map, 0);
                    574: #endif
1.117     yamt      575: }
                    576:
                    577: static void
                    578: pool_reclaim_unregister(struct pool *pp)
                    579: {
                    580:        struct vm_map *map = pp->pr_alloc->pa_backingmap;
                    581:        int s;
                    582:
                    583:        if (map == NULL) {
                    584:                return;
                    585:        }
                    586:
                    587:        s = splvm(); /* not necessary for INTRSAFE maps, but don't care. */
                    588:        callback_unregister(&vm_map_to_kernel(map)->vmk_reclaim_callback,
                    589:            &pp->pr_reclaimerentry);
                    590:        splx(s);
                    591: }
                    592:
                    593: static void
                    594: pa_reclaim_register(struct pool_allocator *pa)
                    595: {
                    596:        struct vm_map *map = *pa->pa_backingmapptr;
                    597:        struct pool *pp;
                    598:
                    599:        KASSERT(pa->pa_backingmap == NULL);
                    600:        if (map == NULL) {
                    601:                SLIST_INSERT_HEAD(&pa_deferinitq, pa, pa_q);
                    602:                return;
                    603:        }
                    604:        pa->pa_backingmap = map;
                    605:        TAILQ_FOREACH(pp, &pa->pa_list, pr_alloc_list) {
                    606:                pool_reclaim_register(pp);
                    607:        }
                    608: }
                    609:
1.3       pk        610: /*
1.94      simonb    611:  * Initialize all the pools listed in the "pools" link set.
                    612:  */
                    613: void
1.117     yamt      614: pool_subsystem_init(void)
1.94      simonb    615: {
1.117     yamt      616:        struct pool_allocator *pa;
1.94      simonb    617:
1.134     ad        618:        mutex_init(&pool_head_lock, MUTEX_DEFAULT, IPL_NONE);
1.179     mlelstv   619:        mutex_init(&pool_allocator_lock, MUTEX_DEFAULT, IPL_NONE);
1.134     ad        620:        cv_init(&pool_busy, "poolbusy");
                    621:
1.117     yamt      622:        while ((pa = SLIST_FIRST(&pa_deferinitq)) != NULL) {
                    623:                KASSERT(pa->pa_backingmapptr != NULL);
                    624:                KASSERT(*pa->pa_backingmapptr != NULL);
                    625:                SLIST_REMOVE_HEAD(&pa_deferinitq, pa_q);
                    626:                pa_reclaim_register(pa);
                    627:        }
1.134     ad        628:
1.156     ad        629:        pool_init(&cache_pool, sizeof(struct pool_cache), coherency_unit,
1.134     ad        630:            0, 0, "pcache", &pool_allocator_nointr, IPL_NONE);
                    631:
1.156     ad        632:        pool_init(&cache_cpu_pool, sizeof(pool_cache_cpu_t), coherency_unit,
1.134     ad        633:            0, 0, "pcachecpu", &pool_allocator_nointr, IPL_NONE);
1.94      simonb    634: }
                    635:
                    636: /*
1.3       pk        637:  * Initialize the given pool resource structure.
                    638:  *
                    639:  * We export this routine to allow other kernel parts to declare
                    640:  * static pools that must be initialized before malloc() is available.
                    641:  */
                    642: void
1.42      thorpej   643: pool_init(struct pool *pp, size_t size, u_int align, u_int ioff, int flags,
1.129     ad        644:     const char *wchan, struct pool_allocator *palloc, int ipl)
1.3       pk        645: {
1.116     simonb    646:        struct pool *pp1;
1.92      enami     647:        size_t trysize, phsize;
1.134     ad        648:        int off, slack;
1.3       pk        649:
1.116     simonb    650: #ifdef DEBUG
                    651:        /*
                    652:         * Check that the pool hasn't already been initialised and
                    653:         * added to the list of all pools.
                    654:         */
1.145     ad        655:        TAILQ_FOREACH(pp1, &pool_head, pr_poollist) {
1.116     simonb    656:                if (pp == pp1)
                    657:                        panic("pool_init: pool %s already initialised",
                    658:                            wchan);
                    659:        }
                    660: #endif
                    661:
1.25      thorpej   662: #ifdef POOL_DIAGNOSTIC
                    663:        /*
                    664:         * Always log if POOL_DIAGNOSTIC is defined.
                    665:         */
                    666:        if (pool_logsize != 0)
                    667:                flags |= PR_LOGGING;
                    668: #endif
                    669:
1.66      thorpej   670:        if (palloc == NULL)
                    671:                palloc = &pool_allocator_kmem;
1.112     bjh21     672: #ifdef POOL_SUBPAGE
                    673:        if (size > palloc->pa_pagesz) {
                    674:                if (palloc == &pool_allocator_kmem)
                    675:                        palloc = &pool_allocator_kmem_fullpage;
                    676:                else if (palloc == &pool_allocator_nointr)
                    677:                        palloc = &pool_allocator_nointr_fullpage;
                    678:        }
1.66      thorpej   679: #endif /* POOL_SUBPAGE */
1.180     mlelstv   680:        if (!cold)
                    681:                mutex_enter(&pool_allocator_lock);
1.178     elad      682:        if (palloc->pa_refcnt++ == 0) {
1.112     bjh21     683:                if (palloc->pa_pagesz == 0)
1.66      thorpej   684:                        palloc->pa_pagesz = PAGE_SIZE;
                    685:
                    686:                TAILQ_INIT(&palloc->pa_list);
                    687:
1.134     ad        688:                mutex_init(&palloc->pa_lock, MUTEX_DEFAULT, IPL_VM);
1.66      thorpej   689:                palloc->pa_pagemask = ~(palloc->pa_pagesz - 1);
                    690:                palloc->pa_pageshift = ffs(palloc->pa_pagesz) - 1;
1.117     yamt      691:
                    692:                if (palloc->pa_backingmapptr != NULL) {
                    693:                        pa_reclaim_register(palloc);
                    694:                }
1.4       thorpej   695:        }
1.180     mlelstv   696:        if (!cold)
                    697:                mutex_exit(&pool_allocator_lock);
1.3       pk        698:
                    699:        if (align == 0)
                    700:                align = ALIGN(1);
1.14      thorpej   701:
1.120     yamt      702:        if ((flags & PR_NOTOUCH) == 0 && size < sizeof(struct pool_item))
1.14      thorpej   703:                size = sizeof(struct pool_item);
1.3       pk        704:
1.78      thorpej   705:        size = roundup(size, align);
1.66      thorpej   706: #ifdef DIAGNOSTIC
                    707:        if (size > palloc->pa_pagesz)
1.121     yamt      708:                panic("pool_init: pool item size (%zu) too large", size);
1.66      thorpej   709: #endif
1.35      pk        710:
1.3       pk        711:        /*
                    712:         * Initialize the pool structure.
                    713:         */
1.88      chs       714:        LIST_INIT(&pp->pr_emptypages);
                    715:        LIST_INIT(&pp->pr_fullpages);
                    716:        LIST_INIT(&pp->pr_partpages);
1.134     ad        717:        pp->pr_cache = NULL;
1.3       pk        718:        pp->pr_curpage = NULL;
                    719:        pp->pr_npages = 0;
                    720:        pp->pr_minitems = 0;
                    721:        pp->pr_minpages = 0;
                    722:        pp->pr_maxpages = UINT_MAX;
1.20      thorpej   723:        pp->pr_roflags = flags;
                    724:        pp->pr_flags = 0;
1.35      pk        725:        pp->pr_size = size;
1.3       pk        726:        pp->pr_align = align;
                    727:        pp->pr_wchan = wchan;
1.66      thorpej   728:        pp->pr_alloc = palloc;
1.20      thorpej   729:        pp->pr_nitems = 0;
                    730:        pp->pr_nout = 0;
                    731:        pp->pr_hardlimit = UINT_MAX;
                    732:        pp->pr_hardlimit_warning = NULL;
1.31      thorpej   733:        pp->pr_hardlimit_ratecap.tv_sec = 0;
                    734:        pp->pr_hardlimit_ratecap.tv_usec = 0;
                    735:        pp->pr_hardlimit_warning_last.tv_sec = 0;
                    736:        pp->pr_hardlimit_warning_last.tv_usec = 0;
1.68      thorpej   737:        pp->pr_drain_hook = NULL;
                    738:        pp->pr_drain_hook_arg = NULL;
1.125     ad        739:        pp->pr_freecheck = NULL;
1.3       pk        740:
                    741:        /*
                    742:         * Decide whether to put the page header off page to avoid
1.92      enami     743:         * wasting too large a part of the page or too big item.
                    744:         * Off-page page headers go on a hash table, so we can match
                    745:         * a returned item with its header based on the page address.
                    746:         * We use 1/16 of the page size and about 8 times of the item
                    747:         * size as the threshold (XXX: tune)
                    748:         *
                    749:         * However, we'll put the header into the page if we can put
                    750:         * it without wasting any items.
                    751:         *
                    752:         * Silently enforce `0 <= ioff < align'.
1.3       pk        753:         */
1.92      enami     754:        pp->pr_itemoffset = ioff %= align;
                    755:        /* See the comment below about reserved bytes. */
                    756:        trysize = palloc->pa_pagesz - ((align - ioff) % align);
                    757:        phsize = ALIGN(sizeof(struct pool_item_header));
1.121     yamt      758:        if ((pp->pr_roflags & (PR_NOTOUCH | PR_NOALIGN)) == 0 &&
1.97      yamt      759:            (pp->pr_size < MIN(palloc->pa_pagesz / 16, phsize << 3) ||
                    760:            trysize / pp->pr_size == (trysize - phsize) / pp->pr_size)) {
1.3       pk        761:                /* Use the end of the page for the page header */
1.20      thorpej   762:                pp->pr_roflags |= PR_PHINPAGE;
1.92      enami     763:                pp->pr_phoffset = off = palloc->pa_pagesz - phsize;
1.2       pk        764:        } else {
1.3       pk        765:                /* The page header will be taken from our page header pool */
                    766:                pp->pr_phoffset = 0;
1.66      thorpej   767:                off = palloc->pa_pagesz;
1.88      chs       768:                SPLAY_INIT(&pp->pr_phtree);
1.2       pk        769:        }
1.1       pk        770:
1.3       pk        771:        /*
                    772:         * Alignment is to take place at `ioff' within the item. This means
                    773:         * we must reserve up to `align - 1' bytes on the page to allow
                    774:         * appropriate positioning of each item.
                    775:         */
                    776:        pp->pr_itemsperpage = (off - ((align - ioff) % align)) / pp->pr_size;
1.43      thorpej   777:        KASSERT(pp->pr_itemsperpage != 0);
1.97      yamt      778:        if ((pp->pr_roflags & PR_NOTOUCH)) {
                    779:                int idx;
                    780:
                    781:                for (idx = 0; pp->pr_itemsperpage > PHPOOL_FREELIST_NELEM(idx);
                    782:                    idx++) {
                    783:                        /* nothing */
                    784:                }
                    785:                if (idx >= PHPOOL_MAX) {
                    786:                        /*
                    787:                         * if you see this panic, consider to tweak
                    788:                         * PHPOOL_MAX and PHPOOL_FREELIST_NELEM.
                    789:                         */
                    790:                        panic("%s: too large itemsperpage(%d) for PR_NOTOUCH",
                    791:                            pp->pr_wchan, pp->pr_itemsperpage);
                    792:                }
                    793:                pp->pr_phpool = &phpool[idx];
                    794:        } else if ((pp->pr_roflags & PR_PHINPAGE) == 0) {
                    795:                pp->pr_phpool = &phpool[0];
                    796:        }
                    797: #if defined(DIAGNOSTIC)
                    798:        else {
                    799:                pp->pr_phpool = NULL;
                    800:        }
                    801: #endif
1.3       pk        802:
                    803:        /*
                    804:         * Use the slack between the chunks and the page header
                    805:         * for "cache coloring".
                    806:         */
                    807:        slack = off - pp->pr_itemsperpage * pp->pr_size;
                    808:        pp->pr_maxcolor = (slack / align) * align;
                    809:        pp->pr_curcolor = 0;
                    810:
                    811:        pp->pr_nget = 0;
                    812:        pp->pr_nfail = 0;
                    813:        pp->pr_nput = 0;
                    814:        pp->pr_npagealloc = 0;
                    815:        pp->pr_npagefree = 0;
1.1       pk        816:        pp->pr_hiwat = 0;
1.8       thorpej   817:        pp->pr_nidle = 0;
1.134     ad        818:        pp->pr_refcnt = 0;
1.3       pk        819:
1.179     mlelstv   820:        pp->pr_log = NULL;
1.25      thorpej   821:
                    822:        pp->pr_entered_file = NULL;
                    823:        pp->pr_entered_line = 0;
1.3       pk        824:
1.157     ad        825:        mutex_init(&pp->pr_lock, MUTEX_DEFAULT, ipl);
1.134     ad        826:        cv_init(&pp->pr_cv, wchan);
                    827:        pp->pr_ipl = ipl;
1.1       pk        828:
1.3       pk        829:        /*
1.43      thorpej   830:         * Initialize private page header pool and cache magazine pool if we
                    831:         * haven't done so yet.
1.23      thorpej   832:         * XXX LOCKING.
1.3       pk        833:         */
1.97      yamt      834:        if (phpool[0].pr_size == 0) {
                    835:                int idx;
                    836:                for (idx = 0; idx < PHPOOL_MAX; idx++) {
                    837:                        static char phpool_names[PHPOOL_MAX][6+1+6+1];
                    838:                        int nelem;
                    839:                        size_t sz;
                    840:
                    841:                        nelem = PHPOOL_FREELIST_NELEM(idx);
                    842:                        snprintf(phpool_names[idx], sizeof(phpool_names[idx]),
                    843:                            "phpool-%d", nelem);
                    844:                        sz = sizeof(struct pool_item_header);
                    845:                        if (nelem) {
1.135     yamt      846:                                sz = offsetof(struct pool_item_header,
                    847:                                    ph_bitmap[howmany(nelem, BITMAP_SIZE)]);
1.97      yamt      848:                        }
                    849:                        pool_init(&phpool[idx], sz, 0, 0, 0,
1.129     ad        850:                            phpool_names[idx], &pool_allocator_meta, IPL_VM);
1.97      yamt      851:                }
1.62      bjh21     852: #ifdef POOL_SUBPAGE
                    853:                pool_init(&psppool, POOL_SUBPAGE, POOL_SUBPAGE, 0,
1.129     ad        854:                    PR_RECURSIVE, "psppool", &pool_allocator_meta, IPL_VM);
1.62      bjh21     855: #endif
1.142     ad        856:
                    857:                size = sizeof(pcg_t) +
                    858:                    (PCG_NOBJECTS_NORMAL - 1) * sizeof(pcgpair_t);
1.156     ad        859:                pool_init(&pcg_normal_pool, size, coherency_unit, 0, 0,
1.142     ad        860:                    "pcgnormal", &pool_allocator_meta, IPL_VM);
                    861:
                    862:                size = sizeof(pcg_t) +
                    863:                    (PCG_NOBJECTS_LARGE - 1) * sizeof(pcgpair_t);
1.156     ad        864:                pool_init(&pcg_large_pool, size, coherency_unit, 0, 0,
1.142     ad        865:                    "pcglarge", &pool_allocator_meta, IPL_VM);
1.1       pk        866:        }
                    867:
1.145     ad        868:        /* Insert into the list of all pools. */
1.181     mlelstv   869:        if (!cold)
1.134     ad        870:                mutex_enter(&pool_head_lock);
1.145     ad        871:        TAILQ_FOREACH(pp1, &pool_head, pr_poollist) {
                    872:                if (strcmp(pp1->pr_wchan, pp->pr_wchan) > 0)
                    873:                        break;
                    874:        }
                    875:        if (pp1 == NULL)
                    876:                TAILQ_INSERT_TAIL(&pool_head, pp, pr_poollist);
                    877:        else
                    878:                TAILQ_INSERT_BEFORE(pp1, pp, pr_poollist);
1.181     mlelstv   879:        if (!cold)
1.134     ad        880:                mutex_exit(&pool_head_lock);
                    881:
1.167     skrll     882:        /* Insert this into the list of pools using this allocator. */
1.181     mlelstv   883:        if (!cold)
1.134     ad        884:                mutex_enter(&palloc->pa_lock);
1.145     ad        885:        TAILQ_INSERT_TAIL(&palloc->pa_list, pp, pr_alloc_list);
1.181     mlelstv   886:        if (!cold)
1.134     ad        887:                mutex_exit(&palloc->pa_lock);
1.66      thorpej   888:
1.117     yamt      889:        pool_reclaim_register(pp);
1.1       pk        890: }
                    891:
                    892: /*
                    893:  * De-commision a pool resource.
                    894:  */
                    895: void
1.42      thorpej   896: pool_destroy(struct pool *pp)
1.1       pk        897: {
1.101     thorpej   898:        struct pool_pagelist pq;
1.3       pk        899:        struct pool_item_header *ph;
1.43      thorpej   900:
1.101     thorpej   901:        /* Remove from global pool list */
1.134     ad        902:        mutex_enter(&pool_head_lock);
                    903:        while (pp->pr_refcnt != 0)
                    904:                cv_wait(&pool_busy, &pool_head_lock);
1.145     ad        905:        TAILQ_REMOVE(&pool_head, pp, pr_poollist);
1.101     thorpej   906:        if (drainpp == pp)
                    907:                drainpp = NULL;
1.134     ad        908:        mutex_exit(&pool_head_lock);
1.101     thorpej   909:
                    910:        /* Remove this pool from its allocator's list of pools. */
1.117     yamt      911:        pool_reclaim_unregister(pp);
1.134     ad        912:        mutex_enter(&pp->pr_alloc->pa_lock);
1.66      thorpej   913:        TAILQ_REMOVE(&pp->pr_alloc->pa_list, pp, pr_alloc_list);
1.134     ad        914:        mutex_exit(&pp->pr_alloc->pa_lock);
1.66      thorpej   915:
1.178     elad      916:        mutex_enter(&pool_allocator_lock);
                    917:        if (--pp->pr_alloc->pa_refcnt == 0)
                    918:                mutex_destroy(&pp->pr_alloc->pa_lock);
                    919:        mutex_exit(&pool_allocator_lock);
                    920:
1.134     ad        921:        mutex_enter(&pp->pr_lock);
1.101     thorpej   922:
1.134     ad        923:        KASSERT(pp->pr_cache == NULL);
1.3       pk        924:
                    925: #ifdef DIAGNOSTIC
1.20      thorpej   926:        if (pp->pr_nout != 0) {
1.25      thorpej   927:                pr_printlog(pp, NULL, printf);
1.80      provos    928:                panic("pool_destroy: pool busy: still out: %u",
1.20      thorpej   929:                    pp->pr_nout);
1.3       pk        930:        }
                    931: #endif
1.1       pk        932:
1.101     thorpej   933:        KASSERT(LIST_EMPTY(&pp->pr_fullpages));
                    934:        KASSERT(LIST_EMPTY(&pp->pr_partpages));
                    935:
1.3       pk        936:        /* Remove all pages */
1.101     thorpej   937:        LIST_INIT(&pq);
1.88      chs       938:        while ((ph = LIST_FIRST(&pp->pr_emptypages)) != NULL)
1.101     thorpej   939:                pr_rmpage(pp, ph, &pq);
                    940:
1.134     ad        941:        mutex_exit(&pp->pr_lock);
1.3       pk        942:
1.101     thorpej   943:        pr_pagelist_free(pp, &pq);
1.3       pk        944:
1.59      thorpej   945: #ifdef POOL_DIAGNOSTIC
1.179     mlelstv   946:        if (pp->pr_log != NULL) {
1.3       pk        947:                free(pp->pr_log, M_TEMP);
1.179     mlelstv   948:                pp->pr_log = NULL;
                    949:        }
1.59      thorpej   950: #endif
1.134     ad        951:
                    952:        cv_destroy(&pp->pr_cv);
                    953:        mutex_destroy(&pp->pr_lock);
1.1       pk        954: }
                    955:
1.68      thorpej   956: void
                    957: pool_set_drain_hook(struct pool *pp, void (*fn)(void *, int), void *arg)
                    958: {
                    959:
                    960:        /* XXX no locking -- must be used just after pool_init() */
                    961: #ifdef DIAGNOSTIC
                    962:        if (pp->pr_drain_hook != NULL)
                    963:                panic("pool_set_drain_hook(%s): already set", pp->pr_wchan);
                    964: #endif
                    965:        pp->pr_drain_hook = fn;
                    966:        pp->pr_drain_hook_arg = arg;
                    967: }
                    968:
1.88      chs       969: static struct pool_item_header *
1.128     christos  970: pool_alloc_item_header(struct pool *pp, void *storage, int flags)
1.55      thorpej   971: {
                    972:        struct pool_item_header *ph;
                    973:
                    974:        if ((pp->pr_roflags & PR_PHINPAGE) != 0)
1.128     christos  975:                ph = (struct pool_item_header *) ((char *)storage + pp->pr_phoffset);
1.134     ad        976:        else
1.97      yamt      977:                ph = pool_get(pp->pr_phpool, flags);
1.55      thorpej   978:
                    979:        return (ph);
                    980: }
1.1       pk        981:
                    982: /*
1.134     ad        983:  * Grab an item from the pool.
1.1       pk        984:  */
1.3       pk        985: void *
1.59      thorpej   986: #ifdef POOL_DIAGNOSTIC
1.42      thorpej   987: _pool_get(struct pool *pp, int flags, const char *file, long line)
1.56      sommerfe  988: #else
                    989: pool_get(struct pool *pp, int flags)
                    990: #endif
1.1       pk        991: {
                    992:        struct pool_item *pi;
1.3       pk        993:        struct pool_item_header *ph;
1.55      thorpej   994:        void *v;
1.1       pk        995:
1.2       pk        996: #ifdef DIAGNOSTIC
1.184     rmind     997:        if (pp->pr_itemsperpage == 0)
                    998:                panic("pool_get: pool '%s': pr_itemsperpage is zero, "
                    999:                    "pool not initialized?", pp->pr_wchan);
1.185     rmind    1000:        if ((cpu_intr_p() || cpu_softintr_p()) && pp->pr_ipl == IPL_NONE &&
                   1001:            !cold && panicstr == NULL)
1.184     rmind    1002:                panic("pool '%s' is IPL_NONE, but called from "
                   1003:                    "interrupt context\n", pp->pr_wchan);
                   1004: #endif
1.155     ad       1005:        if (flags & PR_WAITOK) {
1.154     yamt     1006:                ASSERT_SLEEPABLE();
1.155     ad       1007:        }
1.1       pk       1008:
1.134     ad       1009:        mutex_enter(&pp->pr_lock);
1.25      thorpej  1010:        pr_enter(pp, file, line);
1.20      thorpej  1011:
                   1012:  startover:
                   1013:        /*
                   1014:         * Check to see if we've reached the hard limit.  If we have,
                   1015:         * and we can wait, then wait until an item has been returned to
                   1016:         * the pool.
                   1017:         */
                   1018: #ifdef DIAGNOSTIC
1.34      thorpej  1019:        if (__predict_false(pp->pr_nout > pp->pr_hardlimit)) {
1.25      thorpej  1020:                pr_leave(pp);
1.134     ad       1021:                mutex_exit(&pp->pr_lock);
1.20      thorpej  1022:                panic("pool_get: %s: crossed hard limit", pp->pr_wchan);
                   1023:        }
                   1024: #endif
1.34      thorpej  1025:        if (__predict_false(pp->pr_nout == pp->pr_hardlimit)) {
1.68      thorpej  1026:                if (pp->pr_drain_hook != NULL) {
                   1027:                        /*
                   1028:                         * Since the drain hook is going to free things
                   1029:                         * back to the pool, unlock, call the hook, re-lock,
                   1030:                         * and check the hardlimit condition again.
                   1031:                         */
                   1032:                        pr_leave(pp);
1.134     ad       1033:                        mutex_exit(&pp->pr_lock);
1.68      thorpej  1034:                        (*pp->pr_drain_hook)(pp->pr_drain_hook_arg, flags);
1.134     ad       1035:                        mutex_enter(&pp->pr_lock);
1.68      thorpej  1036:                        pr_enter(pp, file, line);
                   1037:                        if (pp->pr_nout < pp->pr_hardlimit)
                   1038:                                goto startover;
                   1039:                }
                   1040:
1.29      sommerfe 1041:                if ((flags & PR_WAITOK) && !(flags & PR_LIMITFAIL)) {
1.20      thorpej  1042:                        /*
                   1043:                         * XXX: A warning isn't logged in this case.  Should
                   1044:                         * it be?
                   1045:                         */
                   1046:                        pp->pr_flags |= PR_WANTED;
1.25      thorpej  1047:                        pr_leave(pp);
1.134     ad       1048:                        cv_wait(&pp->pr_cv, &pp->pr_lock);
1.25      thorpej  1049:                        pr_enter(pp, file, line);
1.20      thorpej  1050:                        goto startover;
                   1051:                }
1.31      thorpej  1052:
                   1053:                /*
                   1054:                 * Log a message that the hard limit has been hit.
                   1055:                 */
                   1056:                if (pp->pr_hardlimit_warning != NULL &&
                   1057:                    ratecheck(&pp->pr_hardlimit_warning_last,
                   1058:                              &pp->pr_hardlimit_ratecap))
                   1059:                        log(LOG_ERR, "%s\n", pp->pr_hardlimit_warning);
1.21      thorpej  1060:
                   1061:                pp->pr_nfail++;
                   1062:
1.25      thorpej  1063:                pr_leave(pp);
1.134     ad       1064:                mutex_exit(&pp->pr_lock);
1.20      thorpej  1065:                return (NULL);
                   1066:        }
                   1067:
1.3       pk       1068:        /*
                   1069:         * The convention we use is that if `curpage' is not NULL, then
                   1070:         * it points at a non-empty bucket. In particular, `curpage'
                   1071:         * never points at a page header which has PR_PHINPAGE set and
                   1072:         * has no items in its bucket.
                   1073:         */
1.20      thorpej  1074:        if ((ph = pp->pr_curpage) == NULL) {
1.113     yamt     1075:                int error;
                   1076:
1.20      thorpej  1077: #ifdef DIAGNOSTIC
                   1078:                if (pp->pr_nitems != 0) {
1.134     ad       1079:                        mutex_exit(&pp->pr_lock);
1.20      thorpej  1080:                        printf("pool_get: %s: curpage NULL, nitems %u\n",
                   1081:                            pp->pr_wchan, pp->pr_nitems);
1.80      provos   1082:                        panic("pool_get: nitems inconsistent");
1.20      thorpej  1083:                }
                   1084: #endif
                   1085:
1.21      thorpej  1086:                /*
                   1087:                 * Call the back-end page allocator for more memory.
                   1088:                 * Release the pool lock, as the back-end page allocator
                   1089:                 * may block.
                   1090:                 */
1.25      thorpej  1091:                pr_leave(pp);
1.113     yamt     1092:                error = pool_grow(pp, flags);
                   1093:                pr_enter(pp, file, line);
                   1094:                if (error != 0) {
1.21      thorpej  1095:                        /*
1.55      thorpej  1096:                         * We were unable to allocate a page or item
                   1097:                         * header, but we released the lock during
                   1098:                         * allocation, so perhaps items were freed
                   1099:                         * back to the pool.  Check for this case.
1.21      thorpej  1100:                         */
                   1101:                        if (pp->pr_curpage != NULL)
                   1102:                                goto startover;
1.15      pk       1103:
1.117     yamt     1104:                        pp->pr_nfail++;
1.25      thorpej  1105:                        pr_leave(pp);
1.134     ad       1106:                        mutex_exit(&pp->pr_lock);
1.117     yamt     1107:                        return (NULL);
1.1       pk       1108:                }
1.3       pk       1109:
1.20      thorpej  1110:                /* Start the allocation process over. */
                   1111:                goto startover;
1.3       pk       1112:        }
1.97      yamt     1113:        if (pp->pr_roflags & PR_NOTOUCH) {
                   1114: #ifdef DIAGNOSTIC
                   1115:                if (__predict_false(ph->ph_nmissing == pp->pr_itemsperpage)) {
                   1116:                        pr_leave(pp);
1.134     ad       1117:                        mutex_exit(&pp->pr_lock);
1.97      yamt     1118:                        panic("pool_get: %s: page empty", pp->pr_wchan);
                   1119:                }
                   1120: #endif
                   1121:                v = pr_item_notouch_get(pp, ph);
                   1122: #ifdef POOL_DIAGNOSTIC
                   1123:                pr_log(pp, v, PRLOG_GET, file, line);
                   1124: #endif
                   1125:        } else {
1.102     chs      1126:                v = pi = LIST_FIRST(&ph->ph_itemlist);
1.97      yamt     1127:                if (__predict_false(v == NULL)) {
                   1128:                        pr_leave(pp);
1.134     ad       1129:                        mutex_exit(&pp->pr_lock);
1.97      yamt     1130:                        panic("pool_get: %s: page empty", pp->pr_wchan);
                   1131:                }
1.20      thorpej  1132: #ifdef DIAGNOSTIC
1.97      yamt     1133:                if (__predict_false(pp->pr_nitems == 0)) {
                   1134:                        pr_leave(pp);
1.134     ad       1135:                        mutex_exit(&pp->pr_lock);
1.97      yamt     1136:                        printf("pool_get: %s: items on itemlist, nitems %u\n",
                   1137:                            pp->pr_wchan, pp->pr_nitems);
                   1138:                        panic("pool_get: nitems inconsistent");
                   1139:                }
1.65      enami    1140: #endif
1.56      sommerfe 1141:
1.65      enami    1142: #ifdef POOL_DIAGNOSTIC
1.97      yamt     1143:                pr_log(pp, v, PRLOG_GET, file, line);
1.65      enami    1144: #endif
1.3       pk       1145:
1.65      enami    1146: #ifdef DIAGNOSTIC
1.97      yamt     1147:                if (__predict_false(pi->pi_magic != PI_MAGIC)) {
                   1148:                        pr_printlog(pp, pi, printf);
                   1149:                        panic("pool_get(%s): free list modified: "
                   1150:                            "magic=%x; page %p; item addr %p\n",
                   1151:                            pp->pr_wchan, pi->pi_magic, ph->ph_page, pi);
                   1152:                }
1.3       pk       1153: #endif
                   1154:
1.97      yamt     1155:                /*
                   1156:                 * Remove from item list.
                   1157:                 */
1.102     chs      1158:                LIST_REMOVE(pi, pi_list);
1.97      yamt     1159:        }
1.20      thorpej  1160:        pp->pr_nitems--;
                   1161:        pp->pr_nout++;
1.6       thorpej  1162:        if (ph->ph_nmissing == 0) {
                   1163: #ifdef DIAGNOSTIC
1.34      thorpej  1164:                if (__predict_false(pp->pr_nidle == 0))
1.6       thorpej  1165:                        panic("pool_get: nidle inconsistent");
                   1166: #endif
                   1167:                pp->pr_nidle--;
1.88      chs      1168:
                   1169:                /*
                   1170:                 * This page was previously empty.  Move it to the list of
                   1171:                 * partially-full pages.  This page is already curpage.
                   1172:                 */
                   1173:                LIST_REMOVE(ph, ph_pagelist);
                   1174:                LIST_INSERT_HEAD(&pp->pr_partpages, ph, ph_pagelist);
1.6       thorpej  1175:        }
1.3       pk       1176:        ph->ph_nmissing++;
1.97      yamt     1177:        if (ph->ph_nmissing == pp->pr_itemsperpage) {
1.21      thorpej  1178: #ifdef DIAGNOSTIC
1.97      yamt     1179:                if (__predict_false((pp->pr_roflags & PR_NOTOUCH) == 0 &&
1.102     chs      1180:                    !LIST_EMPTY(&ph->ph_itemlist))) {
1.25      thorpej  1181:                        pr_leave(pp);
1.134     ad       1182:                        mutex_exit(&pp->pr_lock);
1.21      thorpej  1183:                        panic("pool_get: %s: nmissing inconsistent",
                   1184:                            pp->pr_wchan);
                   1185:                }
                   1186: #endif
1.3       pk       1187:                /*
1.88      chs      1188:                 * This page is now full.  Move it to the full list
                   1189:                 * and select a new current page.
1.3       pk       1190:                 */
1.88      chs      1191:                LIST_REMOVE(ph, ph_pagelist);
                   1192:                LIST_INSERT_HEAD(&pp->pr_fullpages, ph, ph_pagelist);
                   1193:                pool_update_curpage(pp);
1.1       pk       1194:        }
1.3       pk       1195:
                   1196:        pp->pr_nget++;
1.111     christos 1197:        pr_leave(pp);
1.20      thorpej  1198:
                   1199:        /*
                   1200:         * If we have a low water mark and we are now below that low
                   1201:         * water mark, add more items to the pool.
                   1202:         */
1.53      thorpej  1203:        if (POOL_NEEDS_CATCHUP(pp) && pool_catchup(pp) != 0) {
1.20      thorpej  1204:                /*
                   1205:                 * XXX: Should we log a warning?  Should we set up a timeout
                   1206:                 * to try again in a second or so?  The latter could break
                   1207:                 * a caller's assumptions about interrupt protection, etc.
                   1208:                 */
                   1209:        }
                   1210:
1.134     ad       1211:        mutex_exit(&pp->pr_lock);
1.125     ad       1212:        KASSERT((((vaddr_t)v + pp->pr_itemoffset) & (pp->pr_align - 1)) == 0);
                   1213:        FREECHECK_OUT(&pp->pr_freecheck, v);
1.1       pk       1214:        return (v);
                   1215: }
                   1216:
                   1217: /*
1.43      thorpej  1218:  * Internal version of pool_put().  Pool is already locked/entered.
1.1       pk       1219:  */
1.43      thorpej  1220: static void
1.101     thorpej  1221: pool_do_put(struct pool *pp, void *v, struct pool_pagelist *pq)
1.1       pk       1222: {
                   1223:        struct pool_item *pi = v;
1.3       pk       1224:        struct pool_item_header *ph;
                   1225:
1.134     ad       1226:        KASSERT(mutex_owned(&pp->pr_lock));
1.125     ad       1227:        FREECHECK_IN(&pp->pr_freecheck, v);
1.134     ad       1228:        LOCKDEBUG_MEM_CHECK(v, pp->pr_size);
1.61      chs      1229:
1.30      thorpej  1230: #ifdef DIAGNOSTIC
1.34      thorpej  1231:        if (__predict_false(pp->pr_nout == 0)) {
1.30      thorpej  1232:                printf("pool %s: putting with none out\n",
                   1233:                    pp->pr_wchan);
                   1234:                panic("pool_put");
                   1235:        }
                   1236: #endif
1.3       pk       1237:
1.121     yamt     1238:        if (__predict_false((ph = pr_find_pagehead(pp, v)) == NULL)) {
1.25      thorpej  1239:                pr_printlog(pp, NULL, printf);
1.3       pk       1240:                panic("pool_put: %s: page header missing", pp->pr_wchan);
                   1241:        }
1.28      thorpej  1242:
1.3       pk       1243:        /*
                   1244:         * Return to item list.
                   1245:         */
1.97      yamt     1246:        if (pp->pr_roflags & PR_NOTOUCH) {
                   1247:                pr_item_notouch_put(pp, ph, v);
                   1248:        } else {
1.2       pk       1249: #ifdef DIAGNOSTIC
1.97      yamt     1250:                pi->pi_magic = PI_MAGIC;
1.3       pk       1251: #endif
1.32      chs      1252: #ifdef DEBUG
1.97      yamt     1253:                {
                   1254:                        int i, *ip = v;
1.32      chs      1255:
1.97      yamt     1256:                        for (i = 0; i < pp->pr_size / sizeof(int); i++) {
                   1257:                                *ip++ = PI_MAGIC;
                   1258:                        }
1.32      chs      1259:                }
                   1260: #endif
                   1261:
1.102     chs      1262:                LIST_INSERT_HEAD(&ph->ph_itemlist, pi, pi_list);
1.97      yamt     1263:        }
1.79      thorpej  1264:        KDASSERT(ph->ph_nmissing != 0);
1.3       pk       1265:        ph->ph_nmissing--;
                   1266:        pp->pr_nput++;
1.20      thorpej  1267:        pp->pr_nitems++;
                   1268:        pp->pr_nout--;
1.3       pk       1269:
                   1270:        /* Cancel "pool empty" condition if it exists */
                   1271:        if (pp->pr_curpage == NULL)
                   1272:                pp->pr_curpage = ph;
                   1273:
                   1274:        if (pp->pr_flags & PR_WANTED) {
                   1275:                pp->pr_flags &= ~PR_WANTED;
1.134     ad       1276:                cv_broadcast(&pp->pr_cv);
1.3       pk       1277:        }
                   1278:
                   1279:        /*
1.88      chs      1280:         * If this page is now empty, do one of two things:
1.21      thorpej  1281:         *
1.88      chs      1282:         *      (1) If we have more pages than the page high water mark,
1.96      thorpej  1283:         *          free the page back to the system.  ONLY CONSIDER
1.90      thorpej  1284:         *          FREEING BACK A PAGE IF WE HAVE MORE THAN OUR MINIMUM PAGE
                   1285:         *          CLAIM.
1.21      thorpej  1286:         *
1.88      chs      1287:         *      (2) Otherwise, move the page to the empty page list.
                   1288:         *
                   1289:         * Either way, select a new current page (so we use a partially-full
                   1290:         * page if one is available).
1.3       pk       1291:         */
                   1292:        if (ph->ph_nmissing == 0) {
1.6       thorpej  1293:                pp->pr_nidle++;
1.90      thorpej  1294:                if (pp->pr_npages > pp->pr_minpages &&
1.152     yamt     1295:                    pp->pr_npages > pp->pr_maxpages) {
1.101     thorpej  1296:                        pr_rmpage(pp, ph, pq);
1.3       pk       1297:                } else {
1.88      chs      1298:                        LIST_REMOVE(ph, ph_pagelist);
                   1299:                        LIST_INSERT_HEAD(&pp->pr_emptypages, ph, ph_pagelist);
1.3       pk       1300:
1.21      thorpej  1301:                        /*
                   1302:                         * Update the timestamp on the page.  A page must
                   1303:                         * be idle for some period of time before it can
                   1304:                         * be reclaimed by the pagedaemon.  This minimizes
                   1305:                         * ping-pong'ing for memory.
1.151     yamt     1306:                         *
                   1307:                         * note for 64-bit time_t: truncating to 32-bit is not
                   1308:                         * a problem for our usage.
1.21      thorpej  1309:                         */
1.151     yamt     1310:                        ph->ph_time = time_uptime;
1.1       pk       1311:                }
1.88      chs      1312:                pool_update_curpage(pp);
1.1       pk       1313:        }
1.88      chs      1314:
1.21      thorpej  1315:        /*
1.88      chs      1316:         * If the page was previously completely full, move it to the
                   1317:         * partially-full list and make it the current page.  The next
                   1318:         * allocation will get the item from this page, instead of
                   1319:         * further fragmenting the pool.
1.21      thorpej  1320:         */
                   1321:        else if (ph->ph_nmissing == (pp->pr_itemsperpage - 1)) {
1.88      chs      1322:                LIST_REMOVE(ph, ph_pagelist);
                   1323:                LIST_INSERT_HEAD(&pp->pr_partpages, ph, ph_pagelist);
1.21      thorpej  1324:                pp->pr_curpage = ph;
                   1325:        }
1.43      thorpej  1326: }
                   1327:
                   1328: /*
1.134     ad       1329:  * Return resource to the pool.
1.43      thorpej  1330:  */
1.59      thorpej  1331: #ifdef POOL_DIAGNOSTIC
1.43      thorpej  1332: void
                   1333: _pool_put(struct pool *pp, void *v, const char *file, long line)
                   1334: {
1.101     thorpej  1335:        struct pool_pagelist pq;
                   1336:
                   1337:        LIST_INIT(&pq);
1.43      thorpej  1338:
1.134     ad       1339:        mutex_enter(&pp->pr_lock);
1.43      thorpej  1340:        pr_enter(pp, file, line);
                   1341:
1.56      sommerfe 1342:        pr_log(pp, v, PRLOG_PUT, file, line);
                   1343:
1.101     thorpej  1344:        pool_do_put(pp, v, &pq);
1.21      thorpej  1345:
1.25      thorpej  1346:        pr_leave(pp);
1.134     ad       1347:        mutex_exit(&pp->pr_lock);
1.101     thorpej  1348:
1.102     chs      1349:        pr_pagelist_free(pp, &pq);
1.1       pk       1350: }
1.57      sommerfe 1351: #undef pool_put
1.59      thorpej  1352: #endif /* POOL_DIAGNOSTIC */
1.1       pk       1353:
1.56      sommerfe 1354: void
                   1355: pool_put(struct pool *pp, void *v)
                   1356: {
1.101     thorpej  1357:        struct pool_pagelist pq;
                   1358:
                   1359:        LIST_INIT(&pq);
1.56      sommerfe 1360:
1.134     ad       1361:        mutex_enter(&pp->pr_lock);
1.101     thorpej  1362:        pool_do_put(pp, v, &pq);
1.134     ad       1363:        mutex_exit(&pp->pr_lock);
1.56      sommerfe 1364:
1.102     chs      1365:        pr_pagelist_free(pp, &pq);
1.56      sommerfe 1366: }
1.57      sommerfe 1367:
1.59      thorpej  1368: #ifdef POOL_DIAGNOSTIC
1.57      sommerfe 1369: #define                pool_put(h, v)  _pool_put((h), (v), __FILE__, __LINE__)
1.56      sommerfe 1370: #endif
1.74      thorpej  1371:
                   1372: /*
1.113     yamt     1373:  * pool_grow: grow a pool by a page.
                   1374:  *
                   1375:  * => called with pool locked.
                   1376:  * => unlock and relock the pool.
                   1377:  * => return with pool locked.
                   1378:  */
                   1379:
                   1380: static int
                   1381: pool_grow(struct pool *pp, int flags)
                   1382: {
                   1383:        struct pool_item_header *ph = NULL;
                   1384:        char *cp;
                   1385:
1.134     ad       1386:        mutex_exit(&pp->pr_lock);
1.113     yamt     1387:        cp = pool_allocator_alloc(pp, flags);
                   1388:        if (__predict_true(cp != NULL)) {
                   1389:                ph = pool_alloc_item_header(pp, cp, flags);
                   1390:        }
                   1391:        if (__predict_false(cp == NULL || ph == NULL)) {
                   1392:                if (cp != NULL) {
                   1393:                        pool_allocator_free(pp, cp);
                   1394:                }
1.134     ad       1395:                mutex_enter(&pp->pr_lock);
1.113     yamt     1396:                return ENOMEM;
                   1397:        }
                   1398:
1.134     ad       1399:        mutex_enter(&pp->pr_lock);
1.113     yamt     1400:        pool_prime_page(pp, cp, ph);
                   1401:        pp->pr_npagealloc++;
                   1402:        return 0;
                   1403: }
                   1404:
                   1405: /*
1.74      thorpej  1406:  * Add N items to the pool.
                   1407:  */
                   1408: int
                   1409: pool_prime(struct pool *pp, int n)
                   1410: {
1.75      simonb   1411:        int newpages;
1.113     yamt     1412:        int error = 0;
1.74      thorpej  1413:
1.134     ad       1414:        mutex_enter(&pp->pr_lock);
1.74      thorpej  1415:
                   1416:        newpages = roundup(n, pp->pr_itemsperpage) / pp->pr_itemsperpage;
                   1417:
                   1418:        while (newpages-- > 0) {
1.113     yamt     1419:                error = pool_grow(pp, PR_NOWAIT);
                   1420:                if (error) {
1.74      thorpej  1421:                        break;
                   1422:                }
                   1423:                pp->pr_minpages++;
                   1424:        }
                   1425:
                   1426:        if (pp->pr_minpages >= pp->pr_maxpages)
                   1427:                pp->pr_maxpages = pp->pr_minpages + 1;  /* XXX */
                   1428:
1.134     ad       1429:        mutex_exit(&pp->pr_lock);
1.113     yamt     1430:        return error;
1.74      thorpej  1431: }
1.55      thorpej  1432:
                   1433: /*
1.3       pk       1434:  * Add a page worth of items to the pool.
1.21      thorpej  1435:  *
                   1436:  * Note, we must be called with the pool descriptor LOCKED.
1.3       pk       1437:  */
1.55      thorpej  1438: static void
1.128     christos 1439: pool_prime_page(struct pool *pp, void *storage, struct pool_item_header *ph)
1.3       pk       1440: {
                   1441:        struct pool_item *pi;
1.128     christos 1442:        void *cp = storage;
1.125     ad       1443:        const unsigned int align = pp->pr_align;
                   1444:        const unsigned int ioff = pp->pr_itemoffset;
1.55      thorpej  1445:        int n;
1.36      pk       1446:
1.134     ad       1447:        KASSERT(mutex_owned(&pp->pr_lock));
1.91      yamt     1448:
1.66      thorpej  1449: #ifdef DIAGNOSTIC
1.121     yamt     1450:        if ((pp->pr_roflags & PR_NOALIGN) == 0 &&
1.150     skrll    1451:            ((uintptr_t)cp & (pp->pr_alloc->pa_pagesz - 1)) != 0)
1.36      pk       1452:                panic("pool_prime_page: %s: unaligned page", pp->pr_wchan);
1.66      thorpej  1453: #endif
1.3       pk       1454:
                   1455:        /*
                   1456:         * Insert page header.
                   1457:         */
1.88      chs      1458:        LIST_INSERT_HEAD(&pp->pr_emptypages, ph, ph_pagelist);
1.102     chs      1459:        LIST_INIT(&ph->ph_itemlist);
1.3       pk       1460:        ph->ph_page = storage;
                   1461:        ph->ph_nmissing = 0;
1.151     yamt     1462:        ph->ph_time = time_uptime;
1.88      chs      1463:        if ((pp->pr_roflags & PR_PHINPAGE) == 0)
                   1464:                SPLAY_INSERT(phtree, &pp->pr_phtree, ph);
1.3       pk       1465:
1.6       thorpej  1466:        pp->pr_nidle++;
                   1467:
1.3       pk       1468:        /*
                   1469:         * Color this page.
                   1470:         */
1.141     yamt     1471:        ph->ph_off = pp->pr_curcolor;
                   1472:        cp = (char *)cp + ph->ph_off;
1.3       pk       1473:        if ((pp->pr_curcolor += align) > pp->pr_maxcolor)
                   1474:                pp->pr_curcolor = 0;
                   1475:
                   1476:        /*
                   1477:         * Adjust storage to apply aligment to `pr_itemoffset' in each item.
                   1478:         */
                   1479:        if (ioff != 0)
1.128     christos 1480:                cp = (char *)cp + align - ioff;
1.3       pk       1481:
1.125     ad       1482:        KASSERT((((vaddr_t)cp + ioff) & (align - 1)) == 0);
                   1483:
1.3       pk       1484:        /*
                   1485:         * Insert remaining chunks on the bucket list.
                   1486:         */
                   1487:        n = pp->pr_itemsperpage;
1.20      thorpej  1488:        pp->pr_nitems += n;
1.3       pk       1489:
1.97      yamt     1490:        if (pp->pr_roflags & PR_NOTOUCH) {
1.141     yamt     1491:                pr_item_notouch_init(pp, ph);
1.97      yamt     1492:        } else {
                   1493:                while (n--) {
                   1494:                        pi = (struct pool_item *)cp;
1.78      thorpej  1495:
1.97      yamt     1496:                        KASSERT(((((vaddr_t)pi) + ioff) & (align - 1)) == 0);
1.3       pk       1497:
1.97      yamt     1498:                        /* Insert on page list */
1.102     chs      1499:                        LIST_INSERT_HEAD(&ph->ph_itemlist, pi, pi_list);
1.3       pk       1500: #ifdef DIAGNOSTIC
1.97      yamt     1501:                        pi->pi_magic = PI_MAGIC;
1.3       pk       1502: #endif
1.128     christos 1503:                        cp = (char *)cp + pp->pr_size;
1.125     ad       1504:
                   1505:                        KASSERT((((vaddr_t)cp + ioff) & (align - 1)) == 0);
1.97      yamt     1506:                }
1.3       pk       1507:        }
                   1508:
                   1509:        /*
                   1510:         * If the pool was depleted, point at the new page.
                   1511:         */
                   1512:        if (pp->pr_curpage == NULL)
                   1513:                pp->pr_curpage = ph;
                   1514:
                   1515:        if (++pp->pr_npages > pp->pr_hiwat)
                   1516:                pp->pr_hiwat = pp->pr_npages;
                   1517: }
                   1518:
1.20      thorpej  1519: /*
1.52      thorpej  1520:  * Used by pool_get() when nitems drops below the low water mark.  This
1.88      chs      1521:  * is used to catch up pr_nitems with the low water mark.
1.20      thorpej  1522:  *
1.21      thorpej  1523:  * Note 1, we never wait for memory here, we let the caller decide what to do.
1.20      thorpej  1524:  *
1.73      thorpej  1525:  * Note 2, we must be called with the pool already locked, and we return
1.20      thorpej  1526:  * with it locked.
                   1527:  */
                   1528: static int
1.42      thorpej  1529: pool_catchup(struct pool *pp)
1.20      thorpej  1530: {
                   1531:        int error = 0;
                   1532:
1.54      thorpej  1533:        while (POOL_NEEDS_CATCHUP(pp)) {
1.113     yamt     1534:                error = pool_grow(pp, PR_NOWAIT);
                   1535:                if (error) {
1.20      thorpej  1536:                        break;
                   1537:                }
                   1538:        }
1.113     yamt     1539:        return error;
1.20      thorpej  1540: }
                   1541:
1.88      chs      1542: static void
                   1543: pool_update_curpage(struct pool *pp)
                   1544: {
                   1545:
                   1546:        pp->pr_curpage = LIST_FIRST(&pp->pr_partpages);
                   1547:        if (pp->pr_curpage == NULL) {
                   1548:                pp->pr_curpage = LIST_FIRST(&pp->pr_emptypages);
                   1549:        }
1.168     yamt     1550:        KASSERT((pp->pr_curpage == NULL && pp->pr_nitems == 0) ||
                   1551:            (pp->pr_curpage != NULL && pp->pr_nitems > 0));
1.88      chs      1552: }
                   1553:
1.3       pk       1554: void
1.42      thorpej  1555: pool_setlowat(struct pool *pp, int n)
1.3       pk       1556: {
1.15      pk       1557:
1.134     ad       1558:        mutex_enter(&pp->pr_lock);
1.21      thorpej  1559:
1.3       pk       1560:        pp->pr_minitems = n;
1.15      pk       1561:        pp->pr_minpages = (n == 0)
                   1562:                ? 0
1.18      thorpej  1563:                : roundup(n, pp->pr_itemsperpage) / pp->pr_itemsperpage;
1.20      thorpej  1564:
                   1565:        /* Make sure we're caught up with the newly-set low water mark. */
1.75      simonb   1566:        if (POOL_NEEDS_CATCHUP(pp) && pool_catchup(pp) != 0) {
1.20      thorpej  1567:                /*
                   1568:                 * XXX: Should we log a warning?  Should we set up a timeout
                   1569:                 * to try again in a second or so?  The latter could break
                   1570:                 * a caller's assumptions about interrupt protection, etc.
                   1571:                 */
                   1572:        }
1.21      thorpej  1573:
1.134     ad       1574:        mutex_exit(&pp->pr_lock);
1.3       pk       1575: }
                   1576:
                   1577: void
1.42      thorpej  1578: pool_sethiwat(struct pool *pp, int n)
1.3       pk       1579: {
1.15      pk       1580:
1.134     ad       1581:        mutex_enter(&pp->pr_lock);
1.21      thorpej  1582:
1.15      pk       1583:        pp->pr_maxpages = (n == 0)
                   1584:                ? 0
1.18      thorpej  1585:                : roundup(n, pp->pr_itemsperpage) / pp->pr_itemsperpage;
1.21      thorpej  1586:
1.134     ad       1587:        mutex_exit(&pp->pr_lock);
1.3       pk       1588: }
                   1589:
1.20      thorpej  1590: void
1.42      thorpej  1591: pool_sethardlimit(struct pool *pp, int n, const char *warnmess, int ratecap)
1.20      thorpej  1592: {
                   1593:
1.134     ad       1594:        mutex_enter(&pp->pr_lock);
1.20      thorpej  1595:
                   1596:        pp->pr_hardlimit = n;
                   1597:        pp->pr_hardlimit_warning = warnmess;
1.31      thorpej  1598:        pp->pr_hardlimit_ratecap.tv_sec = ratecap;
                   1599:        pp->pr_hardlimit_warning_last.tv_sec = 0;
                   1600:        pp->pr_hardlimit_warning_last.tv_usec = 0;
1.20      thorpej  1601:
                   1602:        /*
1.21      thorpej  1603:         * In-line version of pool_sethiwat(), because we don't want to
                   1604:         * release the lock.
1.20      thorpej  1605:         */
                   1606:        pp->pr_maxpages = (n == 0)
                   1607:                ? 0
                   1608:                : roundup(n, pp->pr_itemsperpage) / pp->pr_itemsperpage;
1.21      thorpej  1609:
1.134     ad       1610:        mutex_exit(&pp->pr_lock);
1.20      thorpej  1611: }
1.3       pk       1612:
                   1613: /*
                   1614:  * Release all complete pages that have not been used recently.
1.184     rmind    1615:  *
                   1616:  * Might be called from interrupt context.
1.3       pk       1617:  */
1.66      thorpej  1618: int
1.59      thorpej  1619: #ifdef POOL_DIAGNOSTIC
1.42      thorpej  1620: _pool_reclaim(struct pool *pp, const char *file, long line)
1.56      sommerfe 1621: #else
                   1622: pool_reclaim(struct pool *pp)
                   1623: #endif
1.3       pk       1624: {
                   1625:        struct pool_item_header *ph, *phnext;
1.61      chs      1626:        struct pool_pagelist pq;
1.151     yamt     1627:        uint32_t curtime;
1.134     ad       1628:        bool klock;
                   1629:        int rv;
1.3       pk       1630:
1.184     rmind    1631:        if (cpu_intr_p() || cpu_softintr_p()) {
                   1632:                KASSERT(pp->pr_ipl != IPL_NONE);
                   1633:        }
                   1634:
1.68      thorpej  1635:        if (pp->pr_drain_hook != NULL) {
                   1636:                /*
                   1637:                 * The drain hook must be called with the pool unlocked.
                   1638:                 */
                   1639:                (*pp->pr_drain_hook)(pp->pr_drain_hook_arg, PR_NOWAIT);
                   1640:        }
                   1641:
1.134     ad       1642:        /*
1.157     ad       1643:         * XXXSMP Because we do not want to cause non-MPSAFE code
                   1644:         * to block.
1.134     ad       1645:         */
                   1646:        if (pp->pr_ipl == IPL_SOFTNET || pp->pr_ipl == IPL_SOFTCLOCK ||
                   1647:            pp->pr_ipl == IPL_SOFTSERIAL) {
                   1648:                KERNEL_LOCK(1, NULL);
                   1649:                klock = true;
                   1650:        } else
                   1651:                klock = false;
                   1652:
                   1653:        /* Reclaim items from the pool's cache (if any). */
                   1654:        if (pp->pr_cache != NULL)
                   1655:                pool_cache_invalidate(pp->pr_cache);
                   1656:
                   1657:        if (mutex_tryenter(&pp->pr_lock) == 0) {
                   1658:                if (klock) {
                   1659:                        KERNEL_UNLOCK_ONE(NULL);
                   1660:                }
1.66      thorpej  1661:                return (0);
1.134     ad       1662:        }
1.25      thorpej  1663:        pr_enter(pp, file, line);
1.68      thorpej  1664:
1.88      chs      1665:        LIST_INIT(&pq);
1.43      thorpej  1666:
1.151     yamt     1667:        curtime = time_uptime;
1.21      thorpej  1668:
1.88      chs      1669:        for (ph = LIST_FIRST(&pp->pr_emptypages); ph != NULL; ph = phnext) {
                   1670:                phnext = LIST_NEXT(ph, ph_pagelist);
1.3       pk       1671:
                   1672:                /* Check our minimum page claim */
                   1673:                if (pp->pr_npages <= pp->pr_minpages)
                   1674:                        break;
                   1675:
1.88      chs      1676:                KASSERT(ph->ph_nmissing == 0);
1.151     yamt     1677:                if (curtime - ph->ph_time < pool_inactive_time
1.117     yamt     1678:                    && !pa_starved_p(pp->pr_alloc))
1.88      chs      1679:                        continue;
1.21      thorpej  1680:
1.88      chs      1681:                /*
                   1682:                 * If freeing this page would put us below
                   1683:                 * the low water mark, stop now.
                   1684:                 */
                   1685:                if ((pp->pr_nitems - pp->pr_itemsperpage) <
                   1686:                    pp->pr_minitems)
                   1687:                        break;
1.21      thorpej  1688:
1.88      chs      1689:                pr_rmpage(pp, ph, &pq);
1.3       pk       1690:        }
                   1691:
1.25      thorpej  1692:        pr_leave(pp);
1.134     ad       1693:        mutex_exit(&pp->pr_lock);
                   1694:
                   1695:        if (LIST_EMPTY(&pq))
                   1696:                rv = 0;
                   1697:        else {
                   1698:                pr_pagelist_free(pp, &pq);
                   1699:                rv = 1;
                   1700:        }
                   1701:
                   1702:        if (klock) {
                   1703:                KERNEL_UNLOCK_ONE(NULL);
                   1704:        }
1.66      thorpej  1705:
1.134     ad       1706:        return (rv);
1.3       pk       1707: }
                   1708:
                   1709: /*
1.134     ad       1710:  * Drain pools, one at a time.  This is a two stage process;
                   1711:  * drain_start kicks off a cross call to drain CPU-level caches
                   1712:  * if the pool has an associated pool_cache.  drain_end waits
                   1713:  * for those cross calls to finish, and then drains the cache
                   1714:  * (if any) and pool.
1.131     ad       1715:  *
1.134     ad       1716:  * Note, must never be called from interrupt context.
1.3       pk       1717:  */
                   1718: void
1.134     ad       1719: pool_drain_start(struct pool **ppp, uint64_t *wp)
1.3       pk       1720: {
                   1721:        struct pool *pp;
1.134     ad       1722:
1.145     ad       1723:        KASSERT(!TAILQ_EMPTY(&pool_head));
1.3       pk       1724:
1.61      chs      1725:        pp = NULL;
1.134     ad       1726:
                   1727:        /* Find next pool to drain, and add a reference. */
                   1728:        mutex_enter(&pool_head_lock);
                   1729:        do {
                   1730:                if (drainpp == NULL) {
1.145     ad       1731:                        drainpp = TAILQ_FIRST(&pool_head);
1.134     ad       1732:                }
                   1733:                if (drainpp != NULL) {
                   1734:                        pp = drainpp;
1.145     ad       1735:                        drainpp = TAILQ_NEXT(pp, pr_poollist);
1.134     ad       1736:                }
                   1737:                /*
                   1738:                 * Skip completely idle pools.  We depend on at least
                   1739:                 * one pool in the system being active.
                   1740:                 */
                   1741:        } while (pp == NULL || pp->pr_npages == 0);
                   1742:        pp->pr_refcnt++;
                   1743:        mutex_exit(&pool_head_lock);
                   1744:
                   1745:        /* If there is a pool_cache, drain CPU level caches. */
                   1746:        *ppp = pp;
                   1747:        if (pp->pr_cache != NULL) {
                   1748:                *wp = xc_broadcast(0, (xcfunc_t)pool_cache_xcall,
                   1749:                    pp->pr_cache, NULL);
                   1750:        }
                   1751: }
                   1752:
1.186     pooka    1753: bool
1.134     ad       1754: pool_drain_end(struct pool *pp, uint64_t where)
                   1755: {
1.186     pooka    1756:        bool reclaimed;
1.134     ad       1757:
                   1758:        if (pp == NULL)
1.186     pooka    1759:                return false;
1.134     ad       1760:
                   1761:        KASSERT(pp->pr_refcnt > 0);
                   1762:
                   1763:        /* Wait for remote draining to complete. */
                   1764:        if (pp->pr_cache != NULL)
                   1765:                xc_wait(where);
                   1766:
                   1767:        /* Drain the cache (if any) and pool.. */
1.186     pooka    1768:        reclaimed = pool_reclaim(pp);
1.134     ad       1769:
                   1770:        /* Finally, unlock the pool. */
                   1771:        mutex_enter(&pool_head_lock);
                   1772:        pp->pr_refcnt--;
                   1773:        cv_broadcast(&pool_busy);
                   1774:        mutex_exit(&pool_head_lock);
1.186     pooka    1775:
                   1776:        return reclaimed;
1.3       pk       1777: }
                   1778:
                   1779: /*
                   1780:  * Diagnostic helpers.
                   1781:  */
                   1782: void
1.42      thorpej  1783: pool_print(struct pool *pp, const char *modif)
1.21      thorpej  1784: {
                   1785:
1.25      thorpej  1786:        pool_print1(pp, modif, printf);
1.21      thorpej  1787: }
                   1788:
1.25      thorpej  1789: void
1.108     yamt     1790: pool_printall(const char *modif, void (*pr)(const char *, ...))
                   1791: {
                   1792:        struct pool *pp;
                   1793:
1.145     ad       1794:        TAILQ_FOREACH(pp, &pool_head, pr_poollist) {
1.108     yamt     1795:                pool_printit(pp, modif, pr);
                   1796:        }
                   1797: }
                   1798:
                   1799: void
1.42      thorpej  1800: pool_printit(struct pool *pp, const char *modif, void (*pr)(const char *, ...))
1.25      thorpej  1801: {
                   1802:
                   1803:        if (pp == NULL) {
                   1804:                (*pr)("Must specify a pool to print.\n");
                   1805:                return;
                   1806:        }
                   1807:
                   1808:        pool_print1(pp, modif, pr);
                   1809: }
                   1810:
1.21      thorpej  1811: static void
1.124     yamt     1812: pool_print_pagelist(struct pool *pp, struct pool_pagelist *pl,
1.97      yamt     1813:     void (*pr)(const char *, ...))
1.88      chs      1814: {
                   1815:        struct pool_item_header *ph;
                   1816: #ifdef DIAGNOSTIC
                   1817:        struct pool_item *pi;
                   1818: #endif
                   1819:
                   1820:        LIST_FOREACH(ph, pl, ph_pagelist) {
1.151     yamt     1821:                (*pr)("\t\tpage %p, nmissing %d, time %" PRIu32 "\n",
                   1822:                    ph->ph_page, ph->ph_nmissing, ph->ph_time);
1.88      chs      1823: #ifdef DIAGNOSTIC
1.97      yamt     1824:                if (!(pp->pr_roflags & PR_NOTOUCH)) {
1.102     chs      1825:                        LIST_FOREACH(pi, &ph->ph_itemlist, pi_list) {
1.97      yamt     1826:                                if (pi->pi_magic != PI_MAGIC) {
                   1827:                                        (*pr)("\t\t\titem %p, magic 0x%x\n",
                   1828:                                            pi, pi->pi_magic);
                   1829:                                }
1.88      chs      1830:                        }
                   1831:                }
                   1832: #endif
                   1833:        }
                   1834: }
                   1835:
                   1836: static void
1.42      thorpej  1837: pool_print1(struct pool *pp, const char *modif, void (*pr)(const char *, ...))
1.3       pk       1838: {
1.25      thorpej  1839:        struct pool_item_header *ph;
1.134     ad       1840:        pool_cache_t pc;
                   1841:        pcg_t *pcg;
                   1842:        pool_cache_cpu_t *cc;
                   1843:        uint64_t cpuhit, cpumiss;
1.44      thorpej  1844:        int i, print_log = 0, print_pagelist = 0, print_cache = 0;
1.25      thorpej  1845:        char c;
                   1846:
                   1847:        while ((c = *modif++) != '\0') {
                   1848:                if (c == 'l')
                   1849:                        print_log = 1;
                   1850:                if (c == 'p')
                   1851:                        print_pagelist = 1;
1.44      thorpej  1852:                if (c == 'c')
                   1853:                        print_cache = 1;
1.25      thorpej  1854:        }
                   1855:
1.134     ad       1856:        if ((pc = pp->pr_cache) != NULL) {
                   1857:                (*pr)("POOL CACHE");
                   1858:        } else {
                   1859:                (*pr)("POOL");
                   1860:        }
                   1861:
                   1862:        (*pr)(" %s: size %u, align %u, ioff %u, roflags 0x%08x\n",
1.25      thorpej  1863:            pp->pr_wchan, pp->pr_size, pp->pr_align, pp->pr_itemoffset,
                   1864:            pp->pr_roflags);
1.66      thorpej  1865:        (*pr)("\talloc %p\n", pp->pr_alloc);
1.25      thorpej  1866:        (*pr)("\tminitems %u, minpages %u, maxpages %u, npages %u\n",
                   1867:            pp->pr_minitems, pp->pr_minpages, pp->pr_maxpages, pp->pr_npages);
                   1868:        (*pr)("\titemsperpage %u, nitems %u, nout %u, hardlimit %u\n",
                   1869:            pp->pr_itemsperpage, pp->pr_nitems, pp->pr_nout, pp->pr_hardlimit);
                   1870:
1.134     ad       1871:        (*pr)("\tnget %lu, nfail %lu, nput %lu\n",
1.25      thorpej  1872:            pp->pr_nget, pp->pr_nfail, pp->pr_nput);
                   1873:        (*pr)("\tnpagealloc %lu, npagefree %lu, hiwat %u, nidle %lu\n",
                   1874:            pp->pr_npagealloc, pp->pr_npagefree, pp->pr_hiwat, pp->pr_nidle);
                   1875:
                   1876:        if (print_pagelist == 0)
                   1877:                goto skip_pagelist;
                   1878:
1.88      chs      1879:        if ((ph = LIST_FIRST(&pp->pr_emptypages)) != NULL)
                   1880:                (*pr)("\n\tempty page list:\n");
1.97      yamt     1881:        pool_print_pagelist(pp, &pp->pr_emptypages, pr);
1.88      chs      1882:        if ((ph = LIST_FIRST(&pp->pr_fullpages)) != NULL)
                   1883:                (*pr)("\n\tfull page list:\n");
1.97      yamt     1884:        pool_print_pagelist(pp, &pp->pr_fullpages, pr);
1.88      chs      1885:        if ((ph = LIST_FIRST(&pp->pr_partpages)) != NULL)
                   1886:                (*pr)("\n\tpartial-page list:\n");
1.97      yamt     1887:        pool_print_pagelist(pp, &pp->pr_partpages, pr);
1.88      chs      1888:
1.25      thorpej  1889:        if (pp->pr_curpage == NULL)
                   1890:                (*pr)("\tno current page\n");
                   1891:        else
                   1892:                (*pr)("\tcurpage %p\n", pp->pr_curpage->ph_page);
                   1893:
                   1894:  skip_pagelist:
                   1895:        if (print_log == 0)
                   1896:                goto skip_log;
                   1897:
                   1898:        (*pr)("\n");
                   1899:        if ((pp->pr_roflags & PR_LOGGING) == 0)
                   1900:                (*pr)("\tno log\n");
1.122     christos 1901:        else {
1.25      thorpej  1902:                pr_printlog(pp, NULL, pr);
1.122     christos 1903:        }
1.3       pk       1904:
1.25      thorpej  1905:  skip_log:
1.44      thorpej  1906:
1.102     chs      1907: #define PR_GROUPLIST(pcg)                                              \
                   1908:        (*pr)("\t\tgroup %p: avail %d\n", pcg, pcg->pcg_avail);         \
1.142     ad       1909:        for (i = 0; i < pcg->pcg_size; i++) {                           \
1.102     chs      1910:                if (pcg->pcg_objects[i].pcgo_pa !=                      \
                   1911:                    POOL_PADDR_INVALID) {                               \
                   1912:                        (*pr)("\t\t\t%p, 0x%llx\n",                     \
                   1913:                            pcg->pcg_objects[i].pcgo_va,                \
                   1914:                            (unsigned long long)                        \
                   1915:                            pcg->pcg_objects[i].pcgo_pa);               \
                   1916:                } else {                                                \
                   1917:                        (*pr)("\t\t\t%p\n",                             \
                   1918:                            pcg->pcg_objects[i].pcgo_va);               \
                   1919:                }                                                       \
                   1920:        }
                   1921:
1.134     ad       1922:        if (pc != NULL) {
                   1923:                cpuhit = 0;
                   1924:                cpumiss = 0;
1.183     ad       1925:                for (i = 0; i < __arraycount(pc->pc_cpus); i++) {
1.134     ad       1926:                        if ((cc = pc->pc_cpus[i]) == NULL)
                   1927:                                continue;
                   1928:                        cpuhit += cc->cc_hits;
                   1929:                        cpumiss += cc->cc_misses;
                   1930:                }
                   1931:                (*pr)("\tcpu layer hits %llu misses %llu\n", cpuhit, cpumiss);
                   1932:                (*pr)("\tcache layer hits %llu misses %llu\n",
                   1933:                    pc->pc_hits, pc->pc_misses);
                   1934:                (*pr)("\tcache layer entry uncontended %llu contended %llu\n",
                   1935:                    pc->pc_hits + pc->pc_misses - pc->pc_contended,
                   1936:                    pc->pc_contended);
                   1937:                (*pr)("\tcache layer empty groups %u full groups %u\n",
                   1938:                    pc->pc_nempty, pc->pc_nfull);
                   1939:                if (print_cache) {
                   1940:                        (*pr)("\tfull cache groups:\n");
                   1941:                        for (pcg = pc->pc_fullgroups; pcg != NULL;
                   1942:                            pcg = pcg->pcg_next) {
                   1943:                                PR_GROUPLIST(pcg);
                   1944:                        }
                   1945:                        (*pr)("\tempty cache groups:\n");
                   1946:                        for (pcg = pc->pc_emptygroups; pcg != NULL;
                   1947:                            pcg = pcg->pcg_next) {
                   1948:                                PR_GROUPLIST(pcg);
                   1949:                        }
1.103     chs      1950:                }
1.44      thorpej  1951:        }
1.102     chs      1952: #undef PR_GROUPLIST
1.44      thorpej  1953:
1.88      chs      1954:        pr_enter_check(pp, pr);
                   1955: }
                   1956:
                   1957: static int
                   1958: pool_chk_page(struct pool *pp, const char *label, struct pool_item_header *ph)
                   1959: {
                   1960:        struct pool_item *pi;
1.128     christos 1961:        void *page;
1.88      chs      1962:        int n;
                   1963:
1.121     yamt     1964:        if ((pp->pr_roflags & PR_NOALIGN) == 0) {
1.128     christos 1965:                page = (void *)((uintptr_t)ph & pp->pr_alloc->pa_pagemask);
1.121     yamt     1966:                if (page != ph->ph_page &&
                   1967:                    (pp->pr_roflags & PR_PHINPAGE) != 0) {
                   1968:                        if (label != NULL)
                   1969:                                printf("%s: ", label);
                   1970:                        printf("pool(%p:%s): page inconsistency: page %p;"
                   1971:                               " at page head addr %p (p %p)\n", pp,
                   1972:                                pp->pr_wchan, ph->ph_page,
                   1973:                                ph, page);
                   1974:                        return 1;
                   1975:                }
1.88      chs      1976:        }
1.3       pk       1977:
1.97      yamt     1978:        if ((pp->pr_roflags & PR_NOTOUCH) != 0)
                   1979:                return 0;
                   1980:
1.102     chs      1981:        for (pi = LIST_FIRST(&ph->ph_itemlist), n = 0;
1.88      chs      1982:             pi != NULL;
1.102     chs      1983:             pi = LIST_NEXT(pi,pi_list), n++) {
1.88      chs      1984:
                   1985: #ifdef DIAGNOSTIC
                   1986:                if (pi->pi_magic != PI_MAGIC) {
                   1987:                        if (label != NULL)
                   1988:                                printf("%s: ", label);
                   1989:                        printf("pool(%s): free list modified: magic=%x;"
1.121     yamt     1990:                               " page %p; item ordinal %d; addr %p\n",
1.88      chs      1991:                                pp->pr_wchan, pi->pi_magic, ph->ph_page,
1.121     yamt     1992:                                n, pi);
1.88      chs      1993:                        panic("pool");
                   1994:                }
                   1995: #endif
1.121     yamt     1996:                if ((pp->pr_roflags & PR_NOALIGN) != 0) {
                   1997:                        continue;
                   1998:                }
1.128     christos 1999:                page = (void *)((uintptr_t)pi & pp->pr_alloc->pa_pagemask);
1.88      chs      2000:                if (page == ph->ph_page)
                   2001:                        continue;
                   2002:
                   2003:                if (label != NULL)
                   2004:                        printf("%s: ", label);
                   2005:                printf("pool(%p:%s): page inconsistency: page %p;"
                   2006:                       " item ordinal %d; addr %p (p %p)\n", pp,
                   2007:                        pp->pr_wchan, ph->ph_page,
                   2008:                        n, pi, page);
                   2009:                return 1;
                   2010:        }
                   2011:        return 0;
1.3       pk       2012: }
                   2013:
1.88      chs      2014:
1.3       pk       2015: int
1.42      thorpej  2016: pool_chk(struct pool *pp, const char *label)
1.3       pk       2017: {
                   2018:        struct pool_item_header *ph;
                   2019:        int r = 0;
                   2020:
1.134     ad       2021:        mutex_enter(&pp->pr_lock);
1.88      chs      2022:        LIST_FOREACH(ph, &pp->pr_emptypages, ph_pagelist) {
                   2023:                r = pool_chk_page(pp, label, ph);
                   2024:                if (r) {
                   2025:                        goto out;
                   2026:                }
                   2027:        }
                   2028:        LIST_FOREACH(ph, &pp->pr_fullpages, ph_pagelist) {
                   2029:                r = pool_chk_page(pp, label, ph);
                   2030:                if (r) {
1.3       pk       2031:                        goto out;
                   2032:                }
1.88      chs      2033:        }
                   2034:        LIST_FOREACH(ph, &pp->pr_partpages, ph_pagelist) {
                   2035:                r = pool_chk_page(pp, label, ph);
                   2036:                if (r) {
1.3       pk       2037:                        goto out;
                   2038:                }
                   2039:        }
1.88      chs      2040:
1.3       pk       2041: out:
1.134     ad       2042:        mutex_exit(&pp->pr_lock);
1.3       pk       2043:        return (r);
1.43      thorpej  2044: }
                   2045:
                   2046: /*
                   2047:  * pool_cache_init:
                   2048:  *
                   2049:  *     Initialize a pool cache.
1.134     ad       2050:  */
                   2051: pool_cache_t
                   2052: pool_cache_init(size_t size, u_int align, u_int align_offset, u_int flags,
                   2053:     const char *wchan, struct pool_allocator *palloc, int ipl,
                   2054:     int (*ctor)(void *, void *, int), void (*dtor)(void *, void *), void *arg)
                   2055: {
                   2056:        pool_cache_t pc;
                   2057:
                   2058:        pc = pool_get(&cache_pool, PR_WAITOK);
                   2059:        if (pc == NULL)
                   2060:                return NULL;
                   2061:
                   2062:        pool_cache_bootstrap(pc, size, align, align_offset, flags, wchan,
                   2063:           palloc, ipl, ctor, dtor, arg);
                   2064:
                   2065:        return pc;
                   2066: }
                   2067:
                   2068: /*
                   2069:  * pool_cache_bootstrap:
1.43      thorpej  2070:  *
1.134     ad       2071:  *     Kernel-private version of pool_cache_init().  The caller
                   2072:  *     provides initial storage.
1.43      thorpej  2073:  */
                   2074: void
1.134     ad       2075: pool_cache_bootstrap(pool_cache_t pc, size_t size, u_int align,
                   2076:     u_int align_offset, u_int flags, const char *wchan,
                   2077:     struct pool_allocator *palloc, int ipl,
                   2078:     int (*ctor)(void *, void *, int), void (*dtor)(void *, void *),
1.43      thorpej  2079:     void *arg)
                   2080: {
1.134     ad       2081:        CPU_INFO_ITERATOR cii;
1.145     ad       2082:        pool_cache_t pc1;
1.134     ad       2083:        struct cpu_info *ci;
                   2084:        struct pool *pp;
                   2085:
                   2086:        pp = &pc->pc_pool;
                   2087:        if (palloc == NULL && ipl == IPL_NONE)
                   2088:                palloc = &pool_allocator_nointr;
                   2089:        pool_init(pp, size, align, align_offset, flags, wchan, palloc, ipl);
1.157     ad       2090:        mutex_init(&pc->pc_lock, MUTEX_DEFAULT, ipl);
1.43      thorpej  2091:
1.134     ad       2092:        if (ctor == NULL) {
                   2093:                ctor = (int (*)(void *, void *, int))nullop;
                   2094:        }
                   2095:        if (dtor == NULL) {
                   2096:                dtor = (void (*)(void *, void *))nullop;
                   2097:        }
1.43      thorpej  2098:
1.134     ad       2099:        pc->pc_emptygroups = NULL;
                   2100:        pc->pc_fullgroups = NULL;
                   2101:        pc->pc_partgroups = NULL;
1.43      thorpej  2102:        pc->pc_ctor = ctor;
                   2103:        pc->pc_dtor = dtor;
                   2104:        pc->pc_arg  = arg;
1.134     ad       2105:        pc->pc_hits  = 0;
1.48      thorpej  2106:        pc->pc_misses = 0;
1.134     ad       2107:        pc->pc_nempty = 0;
                   2108:        pc->pc_npart = 0;
                   2109:        pc->pc_nfull = 0;
                   2110:        pc->pc_contended = 0;
                   2111:        pc->pc_refcnt = 0;
1.136     yamt     2112:        pc->pc_freecheck = NULL;
1.134     ad       2113:
1.142     ad       2114:        if ((flags & PR_LARGECACHE) != 0) {
                   2115:                pc->pc_pcgsize = PCG_NOBJECTS_LARGE;
1.163     ad       2116:                pc->pc_pcgpool = &pcg_large_pool;
1.142     ad       2117:        } else {
                   2118:                pc->pc_pcgsize = PCG_NOBJECTS_NORMAL;
1.163     ad       2119:                pc->pc_pcgpool = &pcg_normal_pool;
1.142     ad       2120:        }
                   2121:
1.134     ad       2122:        /* Allocate per-CPU caches. */
                   2123:        memset(pc->pc_cpus, 0, sizeof(pc->pc_cpus));
                   2124:        pc->pc_ncpu = 0;
1.139     ad       2125:        if (ncpu < 2) {
1.137     ad       2126:                /* XXX For sparc: boot CPU is not attached yet. */
                   2127:                pool_cache_cpu_init1(curcpu(), pc);
                   2128:        } else {
                   2129:                for (CPU_INFO_FOREACH(cii, ci)) {
                   2130:                        pool_cache_cpu_init1(ci, pc);
                   2131:                }
1.134     ad       2132:        }
1.145     ad       2133:
                   2134:        /* Add to list of all pools. */
                   2135:        if (__predict_true(!cold))
1.134     ad       2136:                mutex_enter(&pool_head_lock);
1.145     ad       2137:        TAILQ_FOREACH(pc1, &pool_cache_head, pc_cachelist) {
                   2138:                if (strcmp(pc1->pc_pool.pr_wchan, pc->pc_pool.pr_wchan) > 0)
                   2139:                        break;
                   2140:        }
                   2141:        if (pc1 == NULL)
                   2142:                TAILQ_INSERT_TAIL(&pool_cache_head, pc, pc_cachelist);
                   2143:        else
                   2144:                TAILQ_INSERT_BEFORE(pc1, pc, pc_cachelist);
                   2145:        if (__predict_true(!cold))
1.134     ad       2146:                mutex_exit(&pool_head_lock);
1.145     ad       2147:
                   2148:        membar_sync();
                   2149:        pp->pr_cache = pc;
1.43      thorpej  2150: }
                   2151:
                   2152: /*
                   2153:  * pool_cache_destroy:
                   2154:  *
                   2155:  *     Destroy a pool cache.
                   2156:  */
                   2157: void
1.134     ad       2158: pool_cache_destroy(pool_cache_t pc)
1.43      thorpej  2159: {
1.134     ad       2160:        struct pool *pp = &pc->pc_pool;
1.175     jym      2161:        u_int i;
1.134     ad       2162:
                   2163:        /* Remove it from the global list. */
                   2164:        mutex_enter(&pool_head_lock);
                   2165:        while (pc->pc_refcnt != 0)
                   2166:                cv_wait(&pool_busy, &pool_head_lock);
1.145     ad       2167:        TAILQ_REMOVE(&pool_cache_head, pc, pc_cachelist);
1.134     ad       2168:        mutex_exit(&pool_head_lock);
1.43      thorpej  2169:
                   2170:        /* First, invalidate the entire cache. */
                   2171:        pool_cache_invalidate(pc);
                   2172:
1.134     ad       2173:        /* Disassociate it from the pool. */
                   2174:        mutex_enter(&pp->pr_lock);
                   2175:        pp->pr_cache = NULL;
                   2176:        mutex_exit(&pp->pr_lock);
                   2177:
                   2178:        /* Destroy per-CPU data */
1.183     ad       2179:        for (i = 0; i < __arraycount(pc->pc_cpus); i++)
1.175     jym      2180:                pool_cache_invalidate_cpu(pc, i);
1.134     ad       2181:
                   2182:        /* Finally, destroy it. */
                   2183:        mutex_destroy(&pc->pc_lock);
                   2184:        pool_destroy(pp);
                   2185:        pool_put(&cache_pool, pc);
                   2186: }
                   2187:
                   2188: /*
                   2189:  * pool_cache_cpu_init1:
                   2190:  *
                   2191:  *     Called for each pool_cache whenever a new CPU is attached.
                   2192:  */
                   2193: static void
                   2194: pool_cache_cpu_init1(struct cpu_info *ci, pool_cache_t pc)
                   2195: {
                   2196:        pool_cache_cpu_t *cc;
1.137     ad       2197:        int index;
1.134     ad       2198:
1.137     ad       2199:        index = ci->ci_index;
                   2200:
1.183     ad       2201:        KASSERT(index < __arraycount(pc->pc_cpus));
1.134     ad       2202:
1.137     ad       2203:        if ((cc = pc->pc_cpus[index]) != NULL) {
                   2204:                KASSERT(cc->cc_cpuindex == index);
1.134     ad       2205:                return;
                   2206:        }
                   2207:
                   2208:        /*
                   2209:         * The first CPU is 'free'.  This needs to be the case for
                   2210:         * bootstrap - we may not be able to allocate yet.
                   2211:         */
                   2212:        if (pc->pc_ncpu == 0) {
                   2213:                cc = &pc->pc_cpu0;
                   2214:                pc->pc_ncpu = 1;
                   2215:        } else {
                   2216:                mutex_enter(&pc->pc_lock);
                   2217:                pc->pc_ncpu++;
                   2218:                mutex_exit(&pc->pc_lock);
                   2219:                cc = pool_get(&cache_cpu_pool, PR_WAITOK);
                   2220:        }
                   2221:
                   2222:        cc->cc_ipl = pc->pc_pool.pr_ipl;
                   2223:        cc->cc_iplcookie = makeiplcookie(cc->cc_ipl);
                   2224:        cc->cc_cache = pc;
1.137     ad       2225:        cc->cc_cpuindex = index;
1.134     ad       2226:        cc->cc_hits = 0;
                   2227:        cc->cc_misses = 0;
1.169     yamt     2228:        cc->cc_current = __UNCONST(&pcg_dummy);
                   2229:        cc->cc_previous = __UNCONST(&pcg_dummy);
1.134     ad       2230:
1.137     ad       2231:        pc->pc_cpus[index] = cc;
1.43      thorpej  2232: }
                   2233:
1.134     ad       2234: /*
                   2235:  * pool_cache_cpu_init:
                   2236:  *
                   2237:  *     Called whenever a new CPU is attached.
                   2238:  */
                   2239: void
                   2240: pool_cache_cpu_init(struct cpu_info *ci)
1.43      thorpej  2241: {
1.134     ad       2242:        pool_cache_t pc;
                   2243:
                   2244:        mutex_enter(&pool_head_lock);
1.145     ad       2245:        TAILQ_FOREACH(pc, &pool_cache_head, pc_cachelist) {
1.134     ad       2246:                pc->pc_refcnt++;
                   2247:                mutex_exit(&pool_head_lock);
1.43      thorpej  2248:
1.134     ad       2249:                pool_cache_cpu_init1(ci, pc);
1.43      thorpej  2250:
1.134     ad       2251:                mutex_enter(&pool_head_lock);
                   2252:                pc->pc_refcnt--;
                   2253:                cv_broadcast(&pool_busy);
                   2254:        }
                   2255:        mutex_exit(&pool_head_lock);
1.43      thorpej  2256: }
                   2257:
1.134     ad       2258: /*
                   2259:  * pool_cache_reclaim:
                   2260:  *
                   2261:  *     Reclaim memory from a pool cache.
                   2262:  */
                   2263: bool
                   2264: pool_cache_reclaim(pool_cache_t pc)
1.43      thorpej  2265: {
                   2266:
1.134     ad       2267:        return pool_reclaim(&pc->pc_pool);
                   2268: }
1.43      thorpej  2269:
1.136     yamt     2270: static void
                   2271: pool_cache_destruct_object1(pool_cache_t pc, void *object)
                   2272: {
                   2273:
                   2274:        (*pc->pc_dtor)(pc->pc_arg, object);
                   2275:        pool_put(&pc->pc_pool, object);
                   2276: }
                   2277:
1.134     ad       2278: /*
                   2279:  * pool_cache_destruct_object:
                   2280:  *
                   2281:  *     Force destruction of an object and its release back into
                   2282:  *     the pool.
                   2283:  */
                   2284: void
                   2285: pool_cache_destruct_object(pool_cache_t pc, void *object)
                   2286: {
                   2287:
1.136     yamt     2288:        FREECHECK_IN(&pc->pc_freecheck, object);
                   2289:
                   2290:        pool_cache_destruct_object1(pc, object);
1.43      thorpej  2291: }
                   2292:
1.134     ad       2293: /*
                   2294:  * pool_cache_invalidate_groups:
                   2295:  *
                   2296:  *     Invalidate a chain of groups and destruct all objects.
                   2297:  */
1.102     chs      2298: static void
1.134     ad       2299: pool_cache_invalidate_groups(pool_cache_t pc, pcg_t *pcg)
1.102     chs      2300: {
1.134     ad       2301:        void *object;
                   2302:        pcg_t *next;
                   2303:        int i;
                   2304:
                   2305:        for (; pcg != NULL; pcg = next) {
                   2306:                next = pcg->pcg_next;
                   2307:
                   2308:                for (i = 0; i < pcg->pcg_avail; i++) {
                   2309:                        object = pcg->pcg_objects[i].pcgo_va;
1.136     yamt     2310:                        pool_cache_destruct_object1(pc, object);
1.134     ad       2311:                }
1.102     chs      2312:
1.142     ad       2313:                if (pcg->pcg_size == PCG_NOBJECTS_LARGE) {
                   2314:                        pool_put(&pcg_large_pool, pcg);
                   2315:                } else {
                   2316:                        KASSERT(pcg->pcg_size == PCG_NOBJECTS_NORMAL);
                   2317:                        pool_put(&pcg_normal_pool, pcg);
                   2318:                }
1.102     chs      2319:        }
                   2320: }
                   2321:
1.43      thorpej  2322: /*
1.134     ad       2323:  * pool_cache_invalidate:
1.43      thorpej  2324:  *
1.134     ad       2325:  *     Invalidate a pool cache (destruct and release all of the
                   2326:  *     cached objects).  Does not reclaim objects from the pool.
1.176     thorpej  2327:  *
                   2328:  *     Note: For pool caches that provide constructed objects, there
                   2329:  *     is an assumption that another level of synchronization is occurring
                   2330:  *     between the input to the constructor and the cache invalidation.
1.43      thorpej  2331:  */
1.134     ad       2332: void
                   2333: pool_cache_invalidate(pool_cache_t pc)
                   2334: {
                   2335:        pcg_t *full, *empty, *part;
1.182     rmind    2336: #if 0
1.176     thorpej  2337:        uint64_t where;
                   2338:
1.177     jym      2339:        if (ncpu < 2 || !mp_online) {
1.176     thorpej  2340:                /*
                   2341:                 * We might be called early enough in the boot process
                   2342:                 * for the CPU data structures to not be fully initialized.
                   2343:                 * In this case, simply gather the local CPU's cache now
                   2344:                 * since it will be the only one running.
                   2345:                 */
                   2346:                pool_cache_xcall(pc);
                   2347:        } else {
                   2348:                /*
                   2349:                 * Gather all of the CPU-specific caches into the
                   2350:                 * global cache.
                   2351:                 */
                   2352:                where = xc_broadcast(0, (xcfunc_t)pool_cache_xcall, pc, NULL);
                   2353:                xc_wait(where);
                   2354:        }
1.182     rmind    2355: #endif
1.134     ad       2356:        mutex_enter(&pc->pc_lock);
                   2357:        full = pc->pc_fullgroups;
                   2358:        empty = pc->pc_emptygroups;
                   2359:        part = pc->pc_partgroups;
                   2360:        pc->pc_fullgroups = NULL;
                   2361:        pc->pc_emptygroups = NULL;
                   2362:        pc->pc_partgroups = NULL;
                   2363:        pc->pc_nfull = 0;
                   2364:        pc->pc_nempty = 0;
                   2365:        pc->pc_npart = 0;
                   2366:        mutex_exit(&pc->pc_lock);
                   2367:
                   2368:        pool_cache_invalidate_groups(pc, full);
                   2369:        pool_cache_invalidate_groups(pc, empty);
                   2370:        pool_cache_invalidate_groups(pc, part);
                   2371: }
                   2372:
1.175     jym      2373: /*
                   2374:  * pool_cache_invalidate_cpu:
                   2375:  *
                   2376:  *     Invalidate all CPU-bound cached objects in pool cache, the CPU being
                   2377:  *     identified by its associated index.
                   2378:  *     It is caller's responsibility to ensure that no operation is
                   2379:  *     taking place on this pool cache while doing this invalidation.
                   2380:  *     WARNING: as no inter-CPU locking is enforced, trying to invalidate
                   2381:  *     pool cached objects from a CPU different from the one currently running
                   2382:  *     may result in an undefined behaviour.
                   2383:  */
                   2384: static void
                   2385: pool_cache_invalidate_cpu(pool_cache_t pc, u_int index)
                   2386: {
                   2387:
                   2388:        pool_cache_cpu_t *cc;
                   2389:        pcg_t *pcg;
                   2390:
                   2391:        if ((cc = pc->pc_cpus[index]) == NULL)
                   2392:                return;
                   2393:
                   2394:        if ((pcg = cc->cc_current) != &pcg_dummy) {
                   2395:                pcg->pcg_next = NULL;
                   2396:                pool_cache_invalidate_groups(pc, pcg);
                   2397:        }
                   2398:        if ((pcg = cc->cc_previous) != &pcg_dummy) {
                   2399:                pcg->pcg_next = NULL;
                   2400:                pool_cache_invalidate_groups(pc, pcg);
                   2401:        }
                   2402:        if (cc != &pc->pc_cpu0)
                   2403:                pool_put(&cache_cpu_pool, cc);
                   2404:
                   2405: }
                   2406:
1.134     ad       2407: void
                   2408: pool_cache_set_drain_hook(pool_cache_t pc, void (*fn)(void *, int), void *arg)
                   2409: {
                   2410:
                   2411:        pool_set_drain_hook(&pc->pc_pool, fn, arg);
                   2412: }
                   2413:
                   2414: void
                   2415: pool_cache_setlowat(pool_cache_t pc, int n)
                   2416: {
                   2417:
                   2418:        pool_setlowat(&pc->pc_pool, n);
                   2419: }
                   2420:
                   2421: void
                   2422: pool_cache_sethiwat(pool_cache_t pc, int n)
                   2423: {
                   2424:
                   2425:        pool_sethiwat(&pc->pc_pool, n);
                   2426: }
                   2427:
                   2428: void
                   2429: pool_cache_sethardlimit(pool_cache_t pc, int n, const char *warnmess, int ratecap)
                   2430: {
                   2431:
                   2432:        pool_sethardlimit(&pc->pc_pool, n, warnmess, ratecap);
                   2433: }
                   2434:
1.162     ad       2435: static bool __noinline
                   2436: pool_cache_get_slow(pool_cache_cpu_t *cc, int s, void **objectp,
1.134     ad       2437:                    paddr_t *pap, int flags)
1.43      thorpej  2438: {
1.134     ad       2439:        pcg_t *pcg, *cur;
                   2440:        uint64_t ncsw;
                   2441:        pool_cache_t pc;
1.43      thorpej  2442:        void *object;
1.58      thorpej  2443:
1.168     yamt     2444:        KASSERT(cc->cc_current->pcg_avail == 0);
                   2445:        KASSERT(cc->cc_previous->pcg_avail == 0);
                   2446:
1.134     ad       2447:        pc = cc->cc_cache;
                   2448:        cc->cc_misses++;
1.43      thorpej  2449:
1.134     ad       2450:        /*
                   2451:         * Nothing was available locally.  Try and grab a group
                   2452:         * from the cache.
                   2453:         */
1.162     ad       2454:        if (__predict_false(!mutex_tryenter(&pc->pc_lock))) {
1.134     ad       2455:                ncsw = curlwp->l_ncsw;
                   2456:                mutex_enter(&pc->pc_lock);
                   2457:                pc->pc_contended++;
1.43      thorpej  2458:
1.134     ad       2459:                /*
                   2460:                 * If we context switched while locking, then
                   2461:                 * our view of the per-CPU data is invalid:
                   2462:                 * retry.
                   2463:                 */
                   2464:                if (curlwp->l_ncsw != ncsw) {
                   2465:                        mutex_exit(&pc->pc_lock);
1.162     ad       2466:                        return true;
1.43      thorpej  2467:                }
1.102     chs      2468:        }
1.43      thorpej  2469:
1.162     ad       2470:        if (__predict_true((pcg = pc->pc_fullgroups) != NULL)) {
1.43      thorpej  2471:                /*
1.134     ad       2472:                 * If there's a full group, release our empty
                   2473:                 * group back to the cache.  Install the full
                   2474:                 * group as cc_current and return.
1.43      thorpej  2475:                 */
1.162     ad       2476:                if (__predict_true((cur = cc->cc_current) != &pcg_dummy)) {
1.134     ad       2477:                        KASSERT(cur->pcg_avail == 0);
                   2478:                        cur->pcg_next = pc->pc_emptygroups;
                   2479:                        pc->pc_emptygroups = cur;
                   2480:                        pc->pc_nempty++;
1.87      thorpej  2481:                }
1.142     ad       2482:                KASSERT(pcg->pcg_avail == pcg->pcg_size);
1.134     ad       2483:                cc->cc_current = pcg;
                   2484:                pc->pc_fullgroups = pcg->pcg_next;
                   2485:                pc->pc_hits++;
                   2486:                pc->pc_nfull--;
                   2487:                mutex_exit(&pc->pc_lock);
1.162     ad       2488:                return true;
1.134     ad       2489:        }
                   2490:
                   2491:        /*
                   2492:         * Nothing available locally or in cache.  Take the slow
                   2493:         * path: fetch a new object from the pool and construct
                   2494:         * it.
                   2495:         */
                   2496:        pc->pc_misses++;
                   2497:        mutex_exit(&pc->pc_lock);
1.162     ad       2498:        splx(s);
1.134     ad       2499:
                   2500:        object = pool_get(&pc->pc_pool, flags);
                   2501:        *objectp = object;
1.162     ad       2502:        if (__predict_false(object == NULL))
                   2503:                return false;
1.125     ad       2504:
1.162     ad       2505:        if (__predict_false((*pc->pc_ctor)(pc->pc_arg, object, flags) != 0)) {
1.134     ad       2506:                pool_put(&pc->pc_pool, object);
                   2507:                *objectp = NULL;
1.162     ad       2508:                return false;
1.43      thorpej  2509:        }
                   2510:
1.134     ad       2511:        KASSERT((((vaddr_t)object + pc->pc_pool.pr_itemoffset) &
                   2512:            (pc->pc_pool.pr_align - 1)) == 0);
1.43      thorpej  2513:
1.134     ad       2514:        if (pap != NULL) {
                   2515: #ifdef POOL_VTOPHYS
                   2516:                *pap = POOL_VTOPHYS(object);
                   2517: #else
                   2518:                *pap = POOL_PADDR_INVALID;
                   2519: #endif
1.102     chs      2520:        }
1.43      thorpej  2521:
1.125     ad       2522:        FREECHECK_OUT(&pc->pc_freecheck, object);
1.162     ad       2523:        return false;
1.43      thorpej  2524: }
                   2525:
                   2526: /*
1.134     ad       2527:  * pool_cache_get{,_paddr}:
1.43      thorpej  2528:  *
1.134     ad       2529:  *     Get an object from a pool cache (optionally returning
                   2530:  *     the physical address of the object).
1.43      thorpej  2531:  */
1.134     ad       2532: void *
                   2533: pool_cache_get_paddr(pool_cache_t pc, int flags, paddr_t *pap)
1.43      thorpej  2534: {
1.134     ad       2535:        pool_cache_cpu_t *cc;
                   2536:        pcg_t *pcg;
                   2537:        void *object;
1.60      thorpej  2538:        int s;
1.43      thorpej  2539:
1.184     rmind    2540:        KASSERTMSG((!cpu_intr_p() && !cpu_softintr_p()) ||
1.185     rmind    2541:            (pc->pc_pool.pr_ipl != IPL_NONE || cold || panicstr != NULL),
1.184     rmind    2542:            ("pool '%s' is IPL_NONE, but called from interrupt context\n",
                   2543:            pc->pc_pool.pr_wchan));
                   2544:
1.155     ad       2545:        if (flags & PR_WAITOK) {
1.154     yamt     2546:                ASSERT_SLEEPABLE();
1.155     ad       2547:        }
1.125     ad       2548:
1.162     ad       2549:        /* Lock out interrupts and disable preemption. */
                   2550:        s = splvm();
1.165     yamt     2551:        while (/* CONSTCOND */ true) {
1.134     ad       2552:                /* Try and allocate an object from the current group. */
1.162     ad       2553:                cc = pc->pc_cpus[curcpu()->ci_index];
                   2554:                KASSERT(cc->cc_cache == pc);
1.134     ad       2555:                pcg = cc->cc_current;
1.162     ad       2556:                if (__predict_true(pcg->pcg_avail > 0)) {
1.134     ad       2557:                        object = pcg->pcg_objects[--pcg->pcg_avail].pcgo_va;
1.162     ad       2558:                        if (__predict_false(pap != NULL))
1.134     ad       2559:                                *pap = pcg->pcg_objects[pcg->pcg_avail].pcgo_pa;
1.148     yamt     2560: #if defined(DIAGNOSTIC)
1.134     ad       2561:                        pcg->pcg_objects[pcg->pcg_avail].pcgo_va = NULL;
1.163     ad       2562:                        KASSERT(pcg->pcg_avail < pcg->pcg_size);
1.134     ad       2563:                        KASSERT(object != NULL);
1.163     ad       2564: #endif
1.134     ad       2565:                        cc->cc_hits++;
1.162     ad       2566:                        splx(s);
1.134     ad       2567:                        FREECHECK_OUT(&pc->pc_freecheck, object);
                   2568:                        return object;
1.43      thorpej  2569:                }
                   2570:
                   2571:                /*
1.134     ad       2572:                 * That failed.  If the previous group isn't empty, swap
                   2573:                 * it with the current group and allocate from there.
1.43      thorpej  2574:                 */
1.134     ad       2575:                pcg = cc->cc_previous;
1.162     ad       2576:                if (__predict_true(pcg->pcg_avail > 0)) {
1.134     ad       2577:                        cc->cc_previous = cc->cc_current;
                   2578:                        cc->cc_current = pcg;
                   2579:                        continue;
1.43      thorpej  2580:                }
                   2581:
1.134     ad       2582:                /*
                   2583:                 * Can't allocate from either group: try the slow path.
                   2584:                 * If get_slow() allocated an object for us, or if
1.162     ad       2585:                 * no more objects are available, it will return false.
1.134     ad       2586:                 * Otherwise, we need to retry.
                   2587:                 */
1.165     yamt     2588:                if (!pool_cache_get_slow(cc, s, &object, pap, flags))
                   2589:                        break;
                   2590:        }
1.43      thorpej  2591:
1.134     ad       2592:        return object;
1.51      thorpej  2593: }
                   2594:
1.162     ad       2595: static bool __noinline
                   2596: pool_cache_put_slow(pool_cache_cpu_t *cc, int s, void *object)
1.51      thorpej  2597: {
1.163     ad       2598:        pcg_t *pcg, *cur;
1.134     ad       2599:        uint64_t ncsw;
                   2600:        pool_cache_t pc;
1.51      thorpej  2601:
1.168     yamt     2602:        KASSERT(cc->cc_current->pcg_avail == cc->cc_current->pcg_size);
                   2603:        KASSERT(cc->cc_previous->pcg_avail == cc->cc_previous->pcg_size);
                   2604:
1.134     ad       2605:        pc = cc->cc_cache;
1.171     ad       2606:        pcg = NULL;
1.134     ad       2607:        cc->cc_misses++;
1.43      thorpej  2608:
1.171     ad       2609:        /*
                   2610:         * If there are no empty groups in the cache then allocate one
                   2611:         * while still unlocked.
                   2612:         */
                   2613:        if (__predict_false(pc->pc_emptygroups == NULL)) {
                   2614:                if (__predict_true(!pool_cache_disable)) {
                   2615:                        pcg = pool_get(pc->pc_pcgpool, PR_NOWAIT);
                   2616:                }
                   2617:                if (__predict_true(pcg != NULL)) {
                   2618:                        pcg->pcg_avail = 0;
                   2619:                        pcg->pcg_size = pc->pc_pcgsize;
                   2620:                }
                   2621:        }
                   2622:
1.162     ad       2623:        /* Lock the cache. */
                   2624:        if (__predict_false(!mutex_tryenter(&pc->pc_lock))) {
1.164     ad       2625:                ncsw = curlwp->l_ncsw;
1.134     ad       2626:                mutex_enter(&pc->pc_lock);
                   2627:                pc->pc_contended++;
1.162     ad       2628:
1.163     ad       2629:                /*
                   2630:                 * If we context switched while locking, then our view of
                   2631:                 * the per-CPU data is invalid: retry.
                   2632:                 */
                   2633:                if (__predict_false(curlwp->l_ncsw != ncsw)) {
                   2634:                        mutex_exit(&pc->pc_lock);
1.171     ad       2635:                        if (pcg != NULL) {
                   2636:                                pool_put(pc->pc_pcgpool, pcg);
                   2637:                        }
1.163     ad       2638:                        return true;
                   2639:                }
1.162     ad       2640:        }
1.102     chs      2641:
1.163     ad       2642:        /* If there are no empty groups in the cache then allocate one. */
1.171     ad       2643:        if (pcg == NULL && pc->pc_emptygroups != NULL) {
                   2644:                pcg = pc->pc_emptygroups;
1.163     ad       2645:                pc->pc_emptygroups = pcg->pcg_next;
                   2646:                pc->pc_nempty--;
1.134     ad       2647:        }
1.130     ad       2648:
1.162     ad       2649:        /*
                   2650:         * If there's a empty group, release our full group back
                   2651:         * to the cache.  Install the empty group to the local CPU
                   2652:         * and return.
                   2653:         */
1.163     ad       2654:        if (pcg != NULL) {
1.134     ad       2655:                KASSERT(pcg->pcg_avail == 0);
1.162     ad       2656:                if (__predict_false(cc->cc_previous == &pcg_dummy)) {
1.146     ad       2657:                        cc->cc_previous = pcg;
                   2658:                } else {
1.162     ad       2659:                        cur = cc->cc_current;
                   2660:                        if (__predict_true(cur != &pcg_dummy)) {
1.163     ad       2661:                                KASSERT(cur->pcg_avail == cur->pcg_size);
1.146     ad       2662:                                cur->pcg_next = pc->pc_fullgroups;
                   2663:                                pc->pc_fullgroups = cur;
                   2664:                                pc->pc_nfull++;
                   2665:                        }
                   2666:                        cc->cc_current = pcg;
                   2667:                }
1.163     ad       2668:                pc->pc_hits++;
1.134     ad       2669:                mutex_exit(&pc->pc_lock);
1.162     ad       2670:                return true;
1.102     chs      2671:        }
1.105     christos 2672:
1.134     ad       2673:        /*
1.162     ad       2674:         * Nothing available locally or in cache, and we didn't
                   2675:         * allocate an empty group.  Take the slow path and destroy
                   2676:         * the object here and now.
1.134     ad       2677:         */
                   2678:        pc->pc_misses++;
                   2679:        mutex_exit(&pc->pc_lock);
1.162     ad       2680:        splx(s);
                   2681:        pool_cache_destruct_object(pc, object);
1.105     christos 2682:
1.162     ad       2683:        return false;
1.134     ad       2684: }
1.102     chs      2685:
1.43      thorpej  2686: /*
1.134     ad       2687:  * pool_cache_put{,_paddr}:
1.43      thorpej  2688:  *
1.134     ad       2689:  *     Put an object back to the pool cache (optionally caching the
                   2690:  *     physical address of the object).
1.43      thorpej  2691:  */
1.101     thorpej  2692: void
1.134     ad       2693: pool_cache_put_paddr(pool_cache_t pc, void *object, paddr_t pa)
1.43      thorpej  2694: {
1.134     ad       2695:        pool_cache_cpu_t *cc;
                   2696:        pcg_t *pcg;
                   2697:        int s;
1.101     thorpej  2698:
1.172     yamt     2699:        KASSERT(object != NULL);
1.134     ad       2700:        FREECHECK_IN(&pc->pc_freecheck, object);
1.101     thorpej  2701:
1.162     ad       2702:        /* Lock out interrupts and disable preemption. */
                   2703:        s = splvm();
1.165     yamt     2704:        while (/* CONSTCOND */ true) {
1.134     ad       2705:                /* If the current group isn't full, release it there. */
1.162     ad       2706:                cc = pc->pc_cpus[curcpu()->ci_index];
                   2707:                KASSERT(cc->cc_cache == pc);
1.134     ad       2708:                pcg = cc->cc_current;
1.162     ad       2709:                if (__predict_true(pcg->pcg_avail < pcg->pcg_size)) {
1.134     ad       2710:                        pcg->pcg_objects[pcg->pcg_avail].pcgo_va = object;
                   2711:                        pcg->pcg_objects[pcg->pcg_avail].pcgo_pa = pa;
                   2712:                        pcg->pcg_avail++;
                   2713:                        cc->cc_hits++;
1.162     ad       2714:                        splx(s);
1.134     ad       2715:                        return;
                   2716:                }
1.43      thorpej  2717:
1.134     ad       2718:                /*
1.162     ad       2719:                 * That failed.  If the previous group isn't full, swap
1.134     ad       2720:                 * it with the current group and try again.
                   2721:                 */
                   2722:                pcg = cc->cc_previous;
1.162     ad       2723:                if (__predict_true(pcg->pcg_avail < pcg->pcg_size)) {
1.134     ad       2724:                        cc->cc_previous = cc->cc_current;
                   2725:                        cc->cc_current = pcg;
                   2726:                        continue;
                   2727:                }
1.43      thorpej  2728:
1.134     ad       2729:                /*
                   2730:                 * Can't free to either group: try the slow path.
                   2731:                 * If put_slow() releases the object for us, it
1.162     ad       2732:                 * will return false.  Otherwise we need to retry.
1.134     ad       2733:                 */
1.165     yamt     2734:                if (!pool_cache_put_slow(cc, s, object))
                   2735:                        break;
                   2736:        }
1.43      thorpej  2737: }
                   2738:
                   2739: /*
1.134     ad       2740:  * pool_cache_xcall:
1.43      thorpej  2741:  *
1.134     ad       2742:  *     Transfer objects from the per-CPU cache to the global cache.
                   2743:  *     Run within a cross-call thread.
1.43      thorpej  2744:  */
                   2745: static void
1.134     ad       2746: pool_cache_xcall(pool_cache_t pc)
1.43      thorpej  2747: {
1.134     ad       2748:        pool_cache_cpu_t *cc;
                   2749:        pcg_t *prev, *cur, **list;
1.162     ad       2750:        int s;
1.134     ad       2751:
1.162     ad       2752:        s = splvm();
                   2753:        mutex_enter(&pc->pc_lock);
                   2754:        cc = pc->pc_cpus[curcpu()->ci_index];
1.134     ad       2755:        cur = cc->cc_current;
1.169     yamt     2756:        cc->cc_current = __UNCONST(&pcg_dummy);
1.134     ad       2757:        prev = cc->cc_previous;
1.169     yamt     2758:        cc->cc_previous = __UNCONST(&pcg_dummy);
1.162     ad       2759:        if (cur != &pcg_dummy) {
1.142     ad       2760:                if (cur->pcg_avail == cur->pcg_size) {
1.134     ad       2761:                        list = &pc->pc_fullgroups;
                   2762:                        pc->pc_nfull++;
                   2763:                } else if (cur->pcg_avail == 0) {
                   2764:                        list = &pc->pc_emptygroups;
                   2765:                        pc->pc_nempty++;
                   2766:                } else {
                   2767:                        list = &pc->pc_partgroups;
                   2768:                        pc->pc_npart++;
                   2769:                }
                   2770:                cur->pcg_next = *list;
                   2771:                *list = cur;
                   2772:        }
1.162     ad       2773:        if (prev != &pcg_dummy) {
1.142     ad       2774:                if (prev->pcg_avail == prev->pcg_size) {
1.134     ad       2775:                        list = &pc->pc_fullgroups;
                   2776:                        pc->pc_nfull++;
                   2777:                } else if (prev->pcg_avail == 0) {
                   2778:                        list = &pc->pc_emptygroups;
                   2779:                        pc->pc_nempty++;
                   2780:                } else {
                   2781:                        list = &pc->pc_partgroups;
                   2782:                        pc->pc_npart++;
                   2783:                }
                   2784:                prev->pcg_next = *list;
                   2785:                *list = prev;
                   2786:        }
                   2787:        mutex_exit(&pc->pc_lock);
                   2788:        splx(s);
1.3       pk       2789: }
1.66      thorpej  2790:
                   2791: /*
                   2792:  * Pool backend allocators.
                   2793:  *
                   2794:  * Each pool has a backend allocator that handles allocation, deallocation,
                   2795:  * and any additional draining that might be needed.
                   2796:  *
                   2797:  * We provide two standard allocators:
                   2798:  *
                   2799:  *     pool_allocator_kmem - the default when no allocator is specified
                   2800:  *
                   2801:  *     pool_allocator_nointr - used for pools that will not be accessed
                   2802:  *     in interrupt context.
                   2803:  */
                   2804: void   *pool_page_alloc(struct pool *, int);
                   2805: void   pool_page_free(struct pool *, void *);
                   2806:
1.112     bjh21    2807: #ifdef POOL_SUBPAGE
                   2808: struct pool_allocator pool_allocator_kmem_fullpage = {
                   2809:        pool_page_alloc, pool_page_free, 0,
1.117     yamt     2810:        .pa_backingmapptr = &kmem_map,
1.112     bjh21    2811: };
                   2812: #else
1.66      thorpej  2813: struct pool_allocator pool_allocator_kmem = {
                   2814:        pool_page_alloc, pool_page_free, 0,
1.117     yamt     2815:        .pa_backingmapptr = &kmem_map,
1.66      thorpej  2816: };
1.112     bjh21    2817: #endif
1.66      thorpej  2818:
                   2819: void   *pool_page_alloc_nointr(struct pool *, int);
                   2820: void   pool_page_free_nointr(struct pool *, void *);
                   2821:
1.112     bjh21    2822: #ifdef POOL_SUBPAGE
                   2823: struct pool_allocator pool_allocator_nointr_fullpage = {
                   2824:        pool_page_alloc_nointr, pool_page_free_nointr, 0,
1.117     yamt     2825:        .pa_backingmapptr = &kernel_map,
1.112     bjh21    2826: };
                   2827: #else
1.66      thorpej  2828: struct pool_allocator pool_allocator_nointr = {
                   2829:        pool_page_alloc_nointr, pool_page_free_nointr, 0,
1.117     yamt     2830:        .pa_backingmapptr = &kernel_map,
1.66      thorpej  2831: };
1.112     bjh21    2832: #endif
1.66      thorpej  2833:
                   2834: #ifdef POOL_SUBPAGE
                   2835: void   *pool_subpage_alloc(struct pool *, int);
                   2836: void   pool_subpage_free(struct pool *, void *);
                   2837:
1.112     bjh21    2838: struct pool_allocator pool_allocator_kmem = {
                   2839:        pool_subpage_alloc, pool_subpage_free, POOL_SUBPAGE,
1.117     yamt     2840:        .pa_backingmapptr = &kmem_map,
1.112     bjh21    2841: };
                   2842:
                   2843: void   *pool_subpage_alloc_nointr(struct pool *, int);
                   2844: void   pool_subpage_free_nointr(struct pool *, void *);
                   2845:
                   2846: struct pool_allocator pool_allocator_nointr = {
                   2847:        pool_subpage_alloc, pool_subpage_free, POOL_SUBPAGE,
1.117     yamt     2848:        .pa_backingmapptr = &kmem_map,
1.66      thorpej  2849: };
                   2850: #endif /* POOL_SUBPAGE */
                   2851:
1.117     yamt     2852: static void *
                   2853: pool_allocator_alloc(struct pool *pp, int flags)
1.66      thorpej  2854: {
1.117     yamt     2855:        struct pool_allocator *pa = pp->pr_alloc;
1.66      thorpej  2856:        void *res;
                   2857:
1.117     yamt     2858:        res = (*pa->pa_alloc)(pp, flags);
                   2859:        if (res == NULL && (flags & PR_WAITOK) == 0) {
1.66      thorpej  2860:                /*
1.117     yamt     2861:                 * We only run the drain hook here if PR_NOWAIT.
                   2862:                 * In other cases, the hook will be run in
                   2863:                 * pool_reclaim().
1.66      thorpej  2864:                 */
1.117     yamt     2865:                if (pp->pr_drain_hook != NULL) {
                   2866:                        (*pp->pr_drain_hook)(pp->pr_drain_hook_arg, flags);
                   2867:                        res = (*pa->pa_alloc)(pp, flags);
1.66      thorpej  2868:                }
1.117     yamt     2869:        }
                   2870:        return res;
1.66      thorpej  2871: }
                   2872:
1.117     yamt     2873: static void
1.66      thorpej  2874: pool_allocator_free(struct pool *pp, void *v)
                   2875: {
                   2876:        struct pool_allocator *pa = pp->pr_alloc;
                   2877:
                   2878:        (*pa->pa_free)(pp, v);
                   2879: }
                   2880:
                   2881: void *
1.124     yamt     2882: pool_page_alloc(struct pool *pp, int flags)
1.66      thorpej  2883: {
1.127     thorpej  2884:        bool waitok = (flags & PR_WAITOK) ? true : false;
1.66      thorpej  2885:
1.100     yamt     2886:        return ((void *) uvm_km_alloc_poolpage_cache(kmem_map, waitok));
1.66      thorpej  2887: }
                   2888:
                   2889: void
1.124     yamt     2890: pool_page_free(struct pool *pp, void *v)
1.66      thorpej  2891: {
                   2892:
1.98      yamt     2893:        uvm_km_free_poolpage_cache(kmem_map, (vaddr_t) v);
                   2894: }
                   2895:
                   2896: static void *
1.124     yamt     2897: pool_page_alloc_meta(struct pool *pp, int flags)
1.98      yamt     2898: {
1.127     thorpej  2899:        bool waitok = (flags & PR_WAITOK) ? true : false;
1.98      yamt     2900:
1.100     yamt     2901:        return ((void *) uvm_km_alloc_poolpage(kmem_map, waitok));
1.98      yamt     2902: }
                   2903:
                   2904: static void
1.124     yamt     2905: pool_page_free_meta(struct pool *pp, void *v)
1.98      yamt     2906: {
                   2907:
1.100     yamt     2908:        uvm_km_free_poolpage(kmem_map, (vaddr_t) v);
1.66      thorpej  2909: }
                   2910:
                   2911: #ifdef POOL_SUBPAGE
                   2912: /* Sub-page allocator, for machines with large hardware pages. */
                   2913: void *
                   2914: pool_subpage_alloc(struct pool *pp, int flags)
                   2915: {
1.134     ad       2916:        return pool_get(&psppool, flags);
1.66      thorpej  2917: }
                   2918:
                   2919: void
                   2920: pool_subpage_free(struct pool *pp, void *v)
                   2921: {
                   2922:        pool_put(&psppool, v);
                   2923: }
                   2924:
                   2925: /* We don't provide a real nointr allocator.  Maybe later. */
                   2926: void *
1.112     bjh21    2927: pool_subpage_alloc_nointr(struct pool *pp, int flags)
1.66      thorpej  2928: {
                   2929:
                   2930:        return (pool_subpage_alloc(pp, flags));
                   2931: }
                   2932:
                   2933: void
1.112     bjh21    2934: pool_subpage_free_nointr(struct pool *pp, void *v)
1.66      thorpej  2935: {
                   2936:
                   2937:        pool_subpage_free(pp, v);
                   2938: }
1.112     bjh21    2939: #endif /* POOL_SUBPAGE */
1.66      thorpej  2940: void *
1.124     yamt     2941: pool_page_alloc_nointr(struct pool *pp, int flags)
1.66      thorpej  2942: {
1.127     thorpej  2943:        bool waitok = (flags & PR_WAITOK) ? true : false;
1.66      thorpej  2944:
1.100     yamt     2945:        return ((void *) uvm_km_alloc_poolpage_cache(kernel_map, waitok));
1.66      thorpej  2946: }
                   2947:
                   2948: void
1.124     yamt     2949: pool_page_free_nointr(struct pool *pp, void *v)
1.66      thorpej  2950: {
                   2951:
1.98      yamt     2952:        uvm_km_free_poolpage_cache(kernel_map, (vaddr_t) v);
1.66      thorpej  2953: }
1.141     yamt     2954:
                   2955: #if defined(DDB)
                   2956: static bool
                   2957: pool_in_page(struct pool *pp, struct pool_item_header *ph, uintptr_t addr)
                   2958: {
                   2959:
                   2960:        return (uintptr_t)ph->ph_page <= addr &&
                   2961:            addr < (uintptr_t)ph->ph_page + pp->pr_alloc->pa_pagesz;
                   2962: }
                   2963:
1.143     yamt     2964: static bool
                   2965: pool_in_item(struct pool *pp, void *item, uintptr_t addr)
                   2966: {
                   2967:
                   2968:        return (uintptr_t)item <= addr && addr < (uintptr_t)item + pp->pr_size;
                   2969: }
                   2970:
                   2971: static bool
                   2972: pool_in_cg(struct pool *pp, struct pool_cache_group *pcg, uintptr_t addr)
                   2973: {
                   2974:        int i;
                   2975:
                   2976:        if (pcg == NULL) {
                   2977:                return false;
                   2978:        }
1.144     yamt     2979:        for (i = 0; i < pcg->pcg_avail; i++) {
1.143     yamt     2980:                if (pool_in_item(pp, pcg->pcg_objects[i].pcgo_va, addr)) {
                   2981:                        return true;
                   2982:                }
                   2983:        }
                   2984:        return false;
                   2985: }
                   2986:
                   2987: static bool
                   2988: pool_allocated(struct pool *pp, struct pool_item_header *ph, uintptr_t addr)
                   2989: {
                   2990:
                   2991:        if ((pp->pr_roflags & PR_NOTOUCH) != 0) {
                   2992:                unsigned int idx = pr_item_notouch_index(pp, ph, (void *)addr);
                   2993:                pool_item_bitmap_t *bitmap =
                   2994:                    ph->ph_bitmap + (idx / BITMAP_SIZE);
                   2995:                pool_item_bitmap_t mask = 1 << (idx & BITMAP_MASK);
                   2996:
                   2997:                return (*bitmap & mask) == 0;
                   2998:        } else {
                   2999:                struct pool_item *pi;
                   3000:
                   3001:                LIST_FOREACH(pi, &ph->ph_itemlist, pi_list) {
                   3002:                        if (pool_in_item(pp, pi, addr)) {
                   3003:                                return false;
                   3004:                        }
                   3005:                }
                   3006:                return true;
                   3007:        }
                   3008: }
                   3009:
1.141     yamt     3010: void
                   3011: pool_whatis(uintptr_t addr, void (*pr)(const char *, ...))
                   3012: {
                   3013:        struct pool *pp;
                   3014:
1.145     ad       3015:        TAILQ_FOREACH(pp, &pool_head, pr_poollist) {
1.141     yamt     3016:                struct pool_item_header *ph;
                   3017:                uintptr_t item;
1.143     yamt     3018:                bool allocated = true;
                   3019:                bool incache = false;
                   3020:                bool incpucache = false;
                   3021:                char cpucachestr[32];
1.141     yamt     3022:
                   3023:                if ((pp->pr_roflags & PR_PHINPAGE) != 0) {
                   3024:                        LIST_FOREACH(ph, &pp->pr_fullpages, ph_pagelist) {
                   3025:                                if (pool_in_page(pp, ph, addr)) {
                   3026:                                        goto found;
                   3027:                                }
                   3028:                        }
                   3029:                        LIST_FOREACH(ph, &pp->pr_partpages, ph_pagelist) {
                   3030:                                if (pool_in_page(pp, ph, addr)) {
1.143     yamt     3031:                                        allocated =
                   3032:                                            pool_allocated(pp, ph, addr);
                   3033:                                        goto found;
                   3034:                                }
                   3035:                        }
                   3036:                        LIST_FOREACH(ph, &pp->pr_emptypages, ph_pagelist) {
                   3037:                                if (pool_in_page(pp, ph, addr)) {
                   3038:                                        allocated = false;
1.141     yamt     3039:                                        goto found;
                   3040:                                }
                   3041:                        }
                   3042:                        continue;
                   3043:                } else {
                   3044:                        ph = pr_find_pagehead_noalign(pp, (void *)addr);
                   3045:                        if (ph == NULL || !pool_in_page(pp, ph, addr)) {
                   3046:                                continue;
                   3047:                        }
1.143     yamt     3048:                        allocated = pool_allocated(pp, ph, addr);
1.141     yamt     3049:                }
                   3050: found:
1.143     yamt     3051:                if (allocated && pp->pr_cache) {
                   3052:                        pool_cache_t pc = pp->pr_cache;
                   3053:                        struct pool_cache_group *pcg;
                   3054:                        int i;
                   3055:
                   3056:                        for (pcg = pc->pc_fullgroups; pcg != NULL;
                   3057:                            pcg = pcg->pcg_next) {
                   3058:                                if (pool_in_cg(pp, pcg, addr)) {
                   3059:                                        incache = true;
                   3060:                                        goto print;
                   3061:                                }
                   3062:                        }
1.183     ad       3063:                        for (i = 0; i < __arraycount(pc->pc_cpus); i++) {
1.143     yamt     3064:                                pool_cache_cpu_t *cc;
                   3065:
                   3066:                                if ((cc = pc->pc_cpus[i]) == NULL) {
                   3067:                                        continue;
                   3068:                                }
                   3069:                                if (pool_in_cg(pp, cc->cc_current, addr) ||
                   3070:                                    pool_in_cg(pp, cc->cc_previous, addr)) {
                   3071:                                        struct cpu_info *ci =
1.170     ad       3072:                                            cpu_lookup(i);
1.143     yamt     3073:
                   3074:                                        incpucache = true;
                   3075:                                        snprintf(cpucachestr,
                   3076:                                            sizeof(cpucachestr),
                   3077:                                            "cached by CPU %u",
1.153     martin   3078:                                            ci->ci_index);
1.143     yamt     3079:                                        goto print;
                   3080:                                }
                   3081:                        }
                   3082:                }
                   3083: print:
1.141     yamt     3084:                item = (uintptr_t)ph->ph_page + ph->ph_off;
                   3085:                item = item + rounddown(addr - item, pp->pr_size);
1.143     yamt     3086:                (*pr)("%p is %p+%zu in POOL '%s' (%s)\n",
1.141     yamt     3087:                    (void *)addr, item, (size_t)(addr - item),
1.143     yamt     3088:                    pp->pr_wchan,
                   3089:                    incpucache ? cpucachestr :
                   3090:                    incache ? "cached" : allocated ? "allocated" : "free");
1.141     yamt     3091:        }
                   3092: }
                   3093: #endif /* defined(DDB) */

CVSweb <webmaster@jp.NetBSD.org>