[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 1.141

1.141   ! yamt        1: /*     $NetBSD: subr_pool.c,v 1.140 2007/12/13 01:22:50 yamt Exp $     */
1.1       pk          2:
                      3: /*-
1.134     ad          4:  * Copyright (c) 1997, 1999, 2000, 2002, 2007 The NetBSD Foundation, Inc.
1.1       pk          5:  * All rights reserved.
                      6:  *
                      7:  * This code is derived from software contributed to The NetBSD Foundation
1.20      thorpej     8:  * by Paul Kranenburg; by Jason R. Thorpe of the Numerical Aerospace
1.134     ad          9:  * Simulation Facility, NASA Ames Research Center, and by Andrew Doran.
1.1       pk         10:  *
                     11:  * Redistribution and use in source and binary forms, with or without
                     12:  * modification, are permitted provided that the following conditions
                     13:  * are met:
                     14:  * 1. Redistributions of source code must retain the above copyright
                     15:  *    notice, this list of conditions and the following disclaimer.
                     16:  * 2. Redistributions in binary form must reproduce the above copyright
                     17:  *    notice, this list of conditions and the following disclaimer in the
                     18:  *    documentation and/or other materials provided with the distribution.
                     19:  * 3. All advertising materials mentioning features or use of this software
                     20:  *    must display the following acknowledgement:
1.13      christos   21:  *     This product includes software developed by the NetBSD
                     22:  *     Foundation, Inc. and its contributors.
1.1       pk         23:  * 4. Neither the name of The NetBSD Foundation nor the names of its
                     24:  *    contributors may be used to endorse or promote products derived
                     25:  *    from this software without specific prior written permission.
                     26:  *
                     27:  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
                     28:  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
                     29:  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
                     30:  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
                     31:  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
                     32:  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
                     33:  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
                     34:  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
                     35:  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
                     36:  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
                     37:  * POSSIBILITY OF SUCH DAMAGE.
                     38:  */
1.64      lukem      39:
                     40: #include <sys/cdefs.h>
1.141   ! yamt       41: __KERNEL_RCSID(0, "$NetBSD: subr_pool.c,v 1.140 2007/12/13 01:22:50 yamt Exp $");
1.24      scottr     42:
1.141   ! yamt       43: #include "opt_ddb.h"
1.25      thorpej    44: #include "opt_pool.h"
1.24      scottr     45: #include "opt_poollog.h"
1.28      thorpej    46: #include "opt_lockdebug.h"
1.1       pk         47:
                     48: #include <sys/param.h>
                     49: #include <sys/systm.h>
1.135     yamt       50: #include <sys/bitops.h>
1.1       pk         51: #include <sys/proc.h>
                     52: #include <sys/errno.h>
                     53: #include <sys/kernel.h>
                     54: #include <sys/malloc.h>
                     55: #include <sys/lock.h>
                     56: #include <sys/pool.h>
1.20      thorpej    57: #include <sys/syslog.h>
1.125     ad         58: #include <sys/debug.h>
1.134     ad         59: #include <sys/lockdebug.h>
                     60: #include <sys/xcall.h>
                     61: #include <sys/cpu.h>
1.3       pk         62:
                     63: #include <uvm/uvm.h>
                     64:
1.1       pk         65: /*
                     66:  * Pool resource management utility.
1.3       pk         67:  *
1.88      chs        68:  * Memory is allocated in pages which are split into pieces according to
                     69:  * the pool item size. Each page is kept on one of three lists in the
                     70:  * pool structure: `pr_emptypages', `pr_fullpages' and `pr_partpages',
                     71:  * for empty, full and partially-full pages respectively. The individual
                     72:  * pool items are on a linked list headed by `ph_itemlist' in each page
                     73:  * header. The memory for building the page list is either taken from
                     74:  * the allocated pages themselves (for small pool items) or taken from
                     75:  * an internal pool of page headers (`phpool').
1.1       pk         76:  */
                     77:
1.3       pk         78: /* List of all pools */
1.102     chs        79: LIST_HEAD(,pool) pool_head = LIST_HEAD_INITIALIZER(pool_head);
1.3       pk         80:
1.134     ad         81: /* List of all caches. */
                     82: LIST_HEAD(,pool_cache) pool_cache_head =
                     83:     LIST_HEAD_INITIALIZER(pool_cache_head);
                     84:
1.3       pk         85: /* Private pool for page header structures */
1.97      yamt       86: #define        PHPOOL_MAX      8
                     87: static struct pool phpool[PHPOOL_MAX];
1.135     yamt       88: #define        PHPOOL_FREELIST_NELEM(idx) \
                     89:        (((idx) == 0) ? 0 : BITMAP_SIZE * (1 << (idx)))
1.3       pk         90:
1.62      bjh21      91: #ifdef POOL_SUBPAGE
                     92: /* Pool of subpages for use by normal pools. */
                     93: static struct pool psppool;
                     94: #endif
                     95:
1.117     yamt       96: static SLIST_HEAD(, pool_allocator) pa_deferinitq =
                     97:     SLIST_HEAD_INITIALIZER(pa_deferinitq);
                     98:
1.98      yamt       99: static void *pool_page_alloc_meta(struct pool *, int);
                    100: static void pool_page_free_meta(struct pool *, void *);
                    101:
                    102: /* allocator for pool metadata */
1.134     ad        103: struct pool_allocator pool_allocator_meta = {
1.117     yamt      104:        pool_page_alloc_meta, pool_page_free_meta,
                    105:        .pa_backingmapptr = &kmem_map,
1.98      yamt      106: };
                    107:
1.3       pk        108: /* # of seconds to retain page after last use */
                    109: int pool_inactive_time = 10;
                    110:
                    111: /* Next candidate for drainage (see pool_drain()) */
1.23      thorpej   112: static struct pool     *drainpp;
                    113:
1.134     ad        114: /* This lock protects both pool_head and drainpp. */
                    115: static kmutex_t pool_head_lock;
                    116: static kcondvar_t pool_busy;
1.3       pk        117:
1.135     yamt      118: typedef uint32_t pool_item_bitmap_t;
                    119: #define        BITMAP_SIZE     (CHAR_BIT * sizeof(pool_item_bitmap_t))
                    120: #define        BITMAP_MASK     (BITMAP_SIZE - 1)
1.99      yamt      121:
1.3       pk        122: struct pool_item_header {
                    123:        /* Page headers */
1.88      chs       124:        LIST_ENTRY(pool_item_header)
1.3       pk        125:                                ph_pagelist;    /* pool page list */
1.88      chs       126:        SPLAY_ENTRY(pool_item_header)
                    127:                                ph_node;        /* Off-page page headers */
1.128     christos  128:        void *                  ph_page;        /* this page's address */
1.3       pk        129:        struct timeval          ph_time;        /* last referenced */
1.135     yamt      130:        uint16_t                ph_nmissing;    /* # of chunks in use */
1.141   ! yamt      131:        uint16_t                ph_off;         /* start offset in page */
1.97      yamt      132:        union {
                    133:                /* !PR_NOTOUCH */
                    134:                struct {
1.102     chs       135:                        LIST_HEAD(, pool_item)
1.97      yamt      136:                                phu_itemlist;   /* chunk list for this page */
                    137:                } phu_normal;
                    138:                /* PR_NOTOUCH */
                    139:                struct {
1.141   ! yamt      140:                        pool_item_bitmap_t phu_bitmap[1];
1.97      yamt      141:                } phu_notouch;
                    142:        } ph_u;
1.3       pk        143: };
1.97      yamt      144: #define        ph_itemlist     ph_u.phu_normal.phu_itemlist
1.135     yamt      145: #define        ph_bitmap       ph_u.phu_notouch.phu_bitmap
1.3       pk        146:
1.1       pk        147: struct pool_item {
1.3       pk        148: #ifdef DIAGNOSTIC
1.82      thorpej   149:        u_int pi_magic;
1.33      chs       150: #endif
1.134     ad        151: #define        PI_MAGIC 0xdeaddeadU
1.3       pk        152:        /* Other entries use only this list entry */
1.102     chs       153:        LIST_ENTRY(pool_item)   pi_list;
1.3       pk        154: };
                    155:
1.53      thorpej   156: #define        POOL_NEEDS_CATCHUP(pp)                                          \
                    157:        ((pp)->pr_nitems < (pp)->pr_minitems)
                    158:
1.43      thorpej   159: /*
                    160:  * Pool cache management.
                    161:  *
                    162:  * Pool caches provide a way for constructed objects to be cached by the
                    163:  * pool subsystem.  This can lead to performance improvements by avoiding
                    164:  * needless object construction/destruction; it is deferred until absolutely
                    165:  * necessary.
                    166:  *
1.134     ad        167:  * Caches are grouped into cache groups.  Each cache group references up
                    168:  * to PCG_NUMOBJECTS constructed objects.  When a cache allocates an
                    169:  * object from the pool, it calls the object's constructor and places it
                    170:  * into a cache group.  When a cache group frees an object back to the
                    171:  * pool, it first calls the object's destructor.  This allows the object
                    172:  * to persist in constructed form while freed to the cache.
                    173:  *
                    174:  * The pool references each cache, so that when a pool is drained by the
                    175:  * pagedaemon, it can drain each individual cache as well.  Each time a
                    176:  * cache is drained, the most idle cache group is freed to the pool in
                    177:  * its entirety.
1.43      thorpej   178:  *
                    179:  * Pool caches are layed on top of pools.  By layering them, we can avoid
                    180:  * the complexity of cache management for pools which would not benefit
                    181:  * from it.
                    182:  */
                    183:
                    184: static struct pool pcgpool;
1.134     ad        185: static struct pool cache_pool;
                    186: static struct pool cache_cpu_pool;
1.3       pk        187:
1.134     ad        188: static pool_cache_cpu_t *pool_cache_put_slow(pool_cache_cpu_t *, int *,
                    189:                                             void *, paddr_t);
                    190: static pool_cache_cpu_t *pool_cache_get_slow(pool_cache_cpu_t *, int *,
                    191:                                             void **, paddr_t *, int);
                    192: static void    pool_cache_cpu_init1(struct cpu_info *, pool_cache_t);
                    193: static void    pool_cache_invalidate_groups(pool_cache_t, pcg_t *);
                    194: static void    pool_cache_xcall(pool_cache_t);
1.3       pk        195:
1.42      thorpej   196: static int     pool_catchup(struct pool *);
1.128     christos  197: static void    pool_prime_page(struct pool *, void *,
1.55      thorpej   198:                    struct pool_item_header *);
1.88      chs       199: static void    pool_update_curpage(struct pool *);
1.66      thorpej   200:
1.113     yamt      201: static int     pool_grow(struct pool *, int);
1.117     yamt      202: static void    *pool_allocator_alloc(struct pool *, int);
                    203: static void    pool_allocator_free(struct pool *, void *);
1.3       pk        204:
1.97      yamt      205: static void pool_print_pagelist(struct pool *, struct pool_pagelist *,
1.88      chs       206:        void (*)(const char *, ...));
1.42      thorpej   207: static void pool_print1(struct pool *, const char *,
                    208:        void (*)(const char *, ...));
1.3       pk        209:
1.88      chs       210: static int pool_chk_page(struct pool *, const char *,
                    211:                         struct pool_item_header *);
                    212:
1.3       pk        213: /*
1.52      thorpej   214:  * Pool log entry. An array of these is allocated in pool_init().
1.3       pk        215:  */
                    216: struct pool_log {
                    217:        const char      *pl_file;
                    218:        long            pl_line;
                    219:        int             pl_action;
1.25      thorpej   220: #define        PRLOG_GET       1
                    221: #define        PRLOG_PUT       2
1.3       pk        222:        void            *pl_addr;
1.1       pk        223: };
                    224:
1.86      matt      225: #ifdef POOL_DIAGNOSTIC
1.3       pk        226: /* Number of entries in pool log buffers */
1.17      thorpej   227: #ifndef POOL_LOGSIZE
                    228: #define        POOL_LOGSIZE    10
                    229: #endif
                    230:
                    231: int pool_logsize = POOL_LOGSIZE;
1.1       pk        232:
1.110     perry     233: static inline void
1.42      thorpej   234: pr_log(struct pool *pp, void *v, int action, const char *file, long line)
1.3       pk        235: {
                    236:        int n = pp->pr_curlogentry;
                    237:        struct pool_log *pl;
                    238:
1.20      thorpej   239:        if ((pp->pr_roflags & PR_LOGGING) == 0)
1.3       pk        240:                return;
                    241:
                    242:        /*
                    243:         * Fill in the current entry. Wrap around and overwrite
                    244:         * the oldest entry if necessary.
                    245:         */
                    246:        pl = &pp->pr_log[n];
                    247:        pl->pl_file = file;
                    248:        pl->pl_line = line;
                    249:        pl->pl_action = action;
                    250:        pl->pl_addr = v;
                    251:        if (++n >= pp->pr_logsize)
                    252:                n = 0;
                    253:        pp->pr_curlogentry = n;
                    254: }
                    255:
                    256: static void
1.42      thorpej   257: pr_printlog(struct pool *pp, struct pool_item *pi,
                    258:     void (*pr)(const char *, ...))
1.3       pk        259: {
                    260:        int i = pp->pr_logsize;
                    261:        int n = pp->pr_curlogentry;
                    262:
1.20      thorpej   263:        if ((pp->pr_roflags & PR_LOGGING) == 0)
1.3       pk        264:                return;
                    265:
                    266:        /*
                    267:         * Print all entries in this pool's log.
                    268:         */
                    269:        while (i-- > 0) {
                    270:                struct pool_log *pl = &pp->pr_log[n];
                    271:                if (pl->pl_action != 0) {
1.25      thorpej   272:                        if (pi == NULL || pi == pl->pl_addr) {
                    273:                                (*pr)("\tlog entry %d:\n", i);
                    274:                                (*pr)("\t\taction = %s, addr = %p\n",
                    275:                                    pl->pl_action == PRLOG_GET ? "get" : "put",
                    276:                                    pl->pl_addr);
                    277:                                (*pr)("\t\tfile: %s at line %lu\n",
                    278:                                    pl->pl_file, pl->pl_line);
                    279:                        }
1.3       pk        280:                }
                    281:                if (++n >= pp->pr_logsize)
                    282:                        n = 0;
                    283:        }
                    284: }
1.25      thorpej   285:
1.110     perry     286: static inline void
1.42      thorpej   287: pr_enter(struct pool *pp, const char *file, long line)
1.25      thorpej   288: {
                    289:
1.34      thorpej   290:        if (__predict_false(pp->pr_entered_file != NULL)) {
1.25      thorpej   291:                printf("pool %s: reentrancy at file %s line %ld\n",
                    292:                    pp->pr_wchan, file, line);
                    293:                printf("         previous entry at file %s line %ld\n",
                    294:                    pp->pr_entered_file, pp->pr_entered_line);
                    295:                panic("pr_enter");
                    296:        }
                    297:
                    298:        pp->pr_entered_file = file;
                    299:        pp->pr_entered_line = line;
                    300: }
                    301:
1.110     perry     302: static inline void
1.42      thorpej   303: pr_leave(struct pool *pp)
1.25      thorpej   304: {
                    305:
1.34      thorpej   306:        if (__predict_false(pp->pr_entered_file == NULL)) {
1.25      thorpej   307:                printf("pool %s not entered?\n", pp->pr_wchan);
                    308:                panic("pr_leave");
                    309:        }
                    310:
                    311:        pp->pr_entered_file = NULL;
                    312:        pp->pr_entered_line = 0;
                    313: }
                    314:
1.110     perry     315: static inline void
1.42      thorpej   316: pr_enter_check(struct pool *pp, void (*pr)(const char *, ...))
1.25      thorpej   317: {
                    318:
                    319:        if (pp->pr_entered_file != NULL)
                    320:                (*pr)("\n\tcurrently entered from file %s line %ld\n",
                    321:                    pp->pr_entered_file, pp->pr_entered_line);
                    322: }
1.3       pk        323: #else
1.25      thorpej   324: #define        pr_log(pp, v, action, file, line)
                    325: #define        pr_printlog(pp, pi, pr)
                    326: #define        pr_enter(pp, file, line)
                    327: #define        pr_leave(pp)
                    328: #define        pr_enter_check(pp, pr)
1.59      thorpej   329: #endif /* POOL_DIAGNOSTIC */
1.3       pk        330:
1.135     yamt      331: static inline unsigned int
1.97      yamt      332: pr_item_notouch_index(const struct pool *pp, const struct pool_item_header *ph,
                    333:     const void *v)
                    334: {
                    335:        const char *cp = v;
1.135     yamt      336:        unsigned int idx;
1.97      yamt      337:
                    338:        KASSERT(pp->pr_roflags & PR_NOTOUCH);
1.128     christos  339:        idx = (cp - (char *)ph->ph_page - ph->ph_off) / pp->pr_size;
1.97      yamt      340:        KASSERT(idx < pp->pr_itemsperpage);
                    341:        return idx;
                    342: }
                    343:
1.110     perry     344: static inline void
1.97      yamt      345: pr_item_notouch_put(const struct pool *pp, struct pool_item_header *ph,
                    346:     void *obj)
                    347: {
1.135     yamt      348:        unsigned int idx = pr_item_notouch_index(pp, ph, obj);
                    349:        pool_item_bitmap_t *bitmap = ph->ph_bitmap + (idx / BITMAP_SIZE);
                    350:        pool_item_bitmap_t mask = 1 << (idx & BITMAP_MASK);
1.97      yamt      351:
1.135     yamt      352:        KASSERT((*bitmap & mask) == 0);
                    353:        *bitmap |= mask;
1.97      yamt      354: }
                    355:
1.110     perry     356: static inline void *
1.97      yamt      357: pr_item_notouch_get(const struct pool *pp, struct pool_item_header *ph)
                    358: {
1.135     yamt      359:        pool_item_bitmap_t *bitmap = ph->ph_bitmap;
                    360:        unsigned int idx;
                    361:        int i;
1.97      yamt      362:
1.135     yamt      363:        for (i = 0; ; i++) {
                    364:                int bit;
1.97      yamt      365:
1.135     yamt      366:                KASSERT((i * BITMAP_SIZE) < pp->pr_itemsperpage);
                    367:                bit = ffs32(bitmap[i]);
                    368:                if (bit) {
                    369:                        pool_item_bitmap_t mask;
                    370:
                    371:                        bit--;
                    372:                        idx = (i * BITMAP_SIZE) + bit;
                    373:                        mask = 1 << bit;
                    374:                        KASSERT((bitmap[i] & mask) != 0);
                    375:                        bitmap[i] &= ~mask;
                    376:                        break;
                    377:                }
                    378:        }
                    379:        KASSERT(idx < pp->pr_itemsperpage);
1.128     christos  380:        return (char *)ph->ph_page + ph->ph_off + idx * pp->pr_size;
1.97      yamt      381: }
                    382:
1.135     yamt      383: static inline void
1.141   ! yamt      384: pr_item_notouch_init(const struct pool *pp, struct pool_item_header *ph)
1.135     yamt      385: {
                    386:        pool_item_bitmap_t *bitmap = ph->ph_bitmap;
                    387:        const int n = howmany(pp->pr_itemsperpage, BITMAP_SIZE);
                    388:        int i;
                    389:
                    390:        for (i = 0; i < n; i++) {
                    391:                bitmap[i] = (pool_item_bitmap_t)-1;
                    392:        }
                    393: }
                    394:
1.110     perry     395: static inline int
1.88      chs       396: phtree_compare(struct pool_item_header *a, struct pool_item_header *b)
                    397: {
1.121     yamt      398:
                    399:        /*
                    400:         * we consider pool_item_header with smaller ph_page bigger.
                    401:         * (this unnatural ordering is for the benefit of pr_find_pagehead.)
                    402:         */
                    403:
1.88      chs       404:        if (a->ph_page < b->ph_page)
1.121     yamt      405:                return (1);
                    406:        else if (a->ph_page > b->ph_page)
1.88      chs       407:                return (-1);
                    408:        else
                    409:                return (0);
                    410: }
                    411:
                    412: SPLAY_PROTOTYPE(phtree, pool_item_header, ph_node, phtree_compare);
                    413: SPLAY_GENERATE(phtree, pool_item_header, ph_node, phtree_compare);
                    414:
1.141   ! yamt      415: static inline struct pool_item_header *
        !           416: pr_find_pagehead_noalign(struct pool *pp, void *v)
        !           417: {
        !           418:        struct pool_item_header *ph, tmp;
        !           419:
        !           420:        tmp.ph_page = (void *)(uintptr_t)v;
        !           421:        ph = SPLAY_FIND(phtree, &pp->pr_phtree, &tmp);
        !           422:        if (ph == NULL) {
        !           423:                ph = SPLAY_ROOT(&pp->pr_phtree);
        !           424:                if (ph != NULL && phtree_compare(&tmp, ph) >= 0) {
        !           425:                        ph = SPLAY_NEXT(phtree, &pp->pr_phtree, ph);
        !           426:                }
        !           427:                KASSERT(ph == NULL || phtree_compare(&tmp, ph) < 0);
        !           428:        }
        !           429:
        !           430:        return ph;
        !           431: }
        !           432:
1.3       pk        433: /*
1.121     yamt      434:  * Return the pool page header based on item address.
1.3       pk        435:  */
1.110     perry     436: static inline struct pool_item_header *
1.121     yamt      437: pr_find_pagehead(struct pool *pp, void *v)
1.3       pk        438: {
1.88      chs       439:        struct pool_item_header *ph, tmp;
1.3       pk        440:
1.121     yamt      441:        if ((pp->pr_roflags & PR_NOALIGN) != 0) {
1.141   ! yamt      442:                ph = pr_find_pagehead_noalign(pp, v);
1.121     yamt      443:        } else {
1.128     christos  444:                void *page =
                    445:                    (void *)((uintptr_t)v & pp->pr_alloc->pa_pagemask);
1.121     yamt      446:
                    447:                if ((pp->pr_roflags & PR_PHINPAGE) != 0) {
1.128     christos  448:                        ph = (struct pool_item_header *)((char *)page + pp->pr_phoffset);
1.121     yamt      449:                } else {
                    450:                        tmp.ph_page = page;
                    451:                        ph = SPLAY_FIND(phtree, &pp->pr_phtree, &tmp);
                    452:                }
                    453:        }
1.3       pk        454:
1.121     yamt      455:        KASSERT(ph == NULL || ((pp->pr_roflags & PR_PHINPAGE) != 0) ||
1.128     christos  456:            ((char *)ph->ph_page <= (char *)v &&
                    457:            (char *)v < (char *)ph->ph_page + pp->pr_alloc->pa_pagesz));
1.88      chs       458:        return ph;
1.3       pk        459: }
                    460:
1.101     thorpej   461: static void
                    462: pr_pagelist_free(struct pool *pp, struct pool_pagelist *pq)
                    463: {
                    464:        struct pool_item_header *ph;
                    465:
                    466:        while ((ph = LIST_FIRST(pq)) != NULL) {
                    467:                LIST_REMOVE(ph, ph_pagelist);
                    468:                pool_allocator_free(pp, ph->ph_page);
1.134     ad        469:                if ((pp->pr_roflags & PR_PHINPAGE) == 0)
1.101     thorpej   470:                        pool_put(pp->pr_phpool, ph);
                    471:        }
                    472: }
                    473:
1.3       pk        474: /*
                    475:  * Remove a page from the pool.
                    476:  */
1.110     perry     477: static inline void
1.61      chs       478: pr_rmpage(struct pool *pp, struct pool_item_header *ph,
                    479:      struct pool_pagelist *pq)
1.3       pk        480: {
                    481:
1.134     ad        482:        KASSERT(mutex_owned(&pp->pr_lock));
1.91      yamt      483:
1.3       pk        484:        /*
1.7       thorpej   485:         * If the page was idle, decrement the idle page count.
1.3       pk        486:         */
1.6       thorpej   487:        if (ph->ph_nmissing == 0) {
                    488: #ifdef DIAGNOSTIC
                    489:                if (pp->pr_nidle == 0)
                    490:                        panic("pr_rmpage: nidle inconsistent");
1.20      thorpej   491:                if (pp->pr_nitems < pp->pr_itemsperpage)
                    492:                        panic("pr_rmpage: nitems inconsistent");
1.6       thorpej   493: #endif
                    494:                pp->pr_nidle--;
                    495:        }
1.7       thorpej   496:
1.20      thorpej   497:        pp->pr_nitems -= pp->pr_itemsperpage;
                    498:
1.7       thorpej   499:        /*
1.101     thorpej   500:         * Unlink the page from the pool and queue it for release.
1.7       thorpej   501:         */
1.88      chs       502:        LIST_REMOVE(ph, ph_pagelist);
1.91      yamt      503:        if ((pp->pr_roflags & PR_PHINPAGE) == 0)
                    504:                SPLAY_REMOVE(phtree, &pp->pr_phtree, ph);
1.101     thorpej   505:        LIST_INSERT_HEAD(pq, ph, ph_pagelist);
                    506:
1.7       thorpej   507:        pp->pr_npages--;
                    508:        pp->pr_npagefree++;
1.6       thorpej   509:
1.88      chs       510:        pool_update_curpage(pp);
1.3       pk        511: }
                    512:
1.126     thorpej   513: static bool
1.117     yamt      514: pa_starved_p(struct pool_allocator *pa)
                    515: {
                    516:
                    517:        if (pa->pa_backingmap != NULL) {
                    518:                return vm_map_starved_p(pa->pa_backingmap);
                    519:        }
1.127     thorpej   520:        return false;
1.117     yamt      521: }
                    522:
                    523: static int
1.124     yamt      524: pool_reclaim_callback(struct callback_entry *ce, void *obj, void *arg)
1.117     yamt      525: {
                    526:        struct pool *pp = obj;
                    527:        struct pool_allocator *pa = pp->pr_alloc;
                    528:
                    529:        KASSERT(&pp->pr_reclaimerentry == ce);
                    530:        pool_reclaim(pp);
                    531:        if (!pa_starved_p(pa)) {
                    532:                return CALLBACK_CHAIN_ABORT;
                    533:        }
                    534:        return CALLBACK_CHAIN_CONTINUE;
                    535: }
                    536:
                    537: static void
                    538: pool_reclaim_register(struct pool *pp)
                    539: {
                    540:        struct vm_map *map = pp->pr_alloc->pa_backingmap;
                    541:        int s;
                    542:
                    543:        if (map == NULL) {
                    544:                return;
                    545:        }
                    546:
                    547:        s = splvm(); /* not necessary for INTRSAFE maps, but don't care. */
                    548:        callback_register(&vm_map_to_kernel(map)->vmk_reclaim_callback,
                    549:            &pp->pr_reclaimerentry, pp, pool_reclaim_callback);
                    550:        splx(s);
                    551: }
                    552:
                    553: static void
                    554: pool_reclaim_unregister(struct pool *pp)
                    555: {
                    556:        struct vm_map *map = pp->pr_alloc->pa_backingmap;
                    557:        int s;
                    558:
                    559:        if (map == NULL) {
                    560:                return;
                    561:        }
                    562:
                    563:        s = splvm(); /* not necessary for INTRSAFE maps, but don't care. */
                    564:        callback_unregister(&vm_map_to_kernel(map)->vmk_reclaim_callback,
                    565:            &pp->pr_reclaimerentry);
                    566:        splx(s);
                    567: }
                    568:
                    569: static void
                    570: pa_reclaim_register(struct pool_allocator *pa)
                    571: {
                    572:        struct vm_map *map = *pa->pa_backingmapptr;
                    573:        struct pool *pp;
                    574:
                    575:        KASSERT(pa->pa_backingmap == NULL);
                    576:        if (map == NULL) {
                    577:                SLIST_INSERT_HEAD(&pa_deferinitq, pa, pa_q);
                    578:                return;
                    579:        }
                    580:        pa->pa_backingmap = map;
                    581:        TAILQ_FOREACH(pp, &pa->pa_list, pr_alloc_list) {
                    582:                pool_reclaim_register(pp);
                    583:        }
                    584: }
                    585:
1.3       pk        586: /*
1.94      simonb    587:  * Initialize all the pools listed in the "pools" link set.
                    588:  */
                    589: void
1.117     yamt      590: pool_subsystem_init(void)
1.94      simonb    591: {
1.117     yamt      592:        struct pool_allocator *pa;
1.94      simonb    593:        __link_set_decl(pools, struct link_pool_init);
                    594:        struct link_pool_init * const *pi;
                    595:
1.134     ad        596:        mutex_init(&pool_head_lock, MUTEX_DEFAULT, IPL_NONE);
                    597:        cv_init(&pool_busy, "poolbusy");
                    598:
1.94      simonb    599:        __link_set_foreach(pi, pools)
                    600:                pool_init((*pi)->pp, (*pi)->size, (*pi)->align,
                    601:                    (*pi)->align_offset, (*pi)->flags, (*pi)->wchan,
1.129     ad        602:                    (*pi)->palloc, (*pi)->ipl);
1.117     yamt      603:
                    604:        while ((pa = SLIST_FIRST(&pa_deferinitq)) != NULL) {
                    605:                KASSERT(pa->pa_backingmapptr != NULL);
                    606:                KASSERT(*pa->pa_backingmapptr != NULL);
                    607:                SLIST_REMOVE_HEAD(&pa_deferinitq, pa_q);
                    608:                pa_reclaim_register(pa);
                    609:        }
1.134     ad        610:
                    611:        pool_init(&cache_pool, sizeof(struct pool_cache), CACHE_LINE_SIZE,
                    612:            0, 0, "pcache", &pool_allocator_nointr, IPL_NONE);
                    613:
                    614:        pool_init(&cache_cpu_pool, sizeof(pool_cache_cpu_t), CACHE_LINE_SIZE,
                    615:            0, 0, "pcachecpu", &pool_allocator_nointr, IPL_NONE);
1.94      simonb    616: }
                    617:
                    618: /*
1.3       pk        619:  * Initialize the given pool resource structure.
                    620:  *
                    621:  * We export this routine to allow other kernel parts to declare
                    622:  * static pools that must be initialized before malloc() is available.
                    623:  */
                    624: void
1.42      thorpej   625: pool_init(struct pool *pp, size_t size, u_int align, u_int ioff, int flags,
1.129     ad        626:     const char *wchan, struct pool_allocator *palloc, int ipl)
1.3       pk        627: {
1.116     simonb    628: #ifdef DEBUG
                    629:        struct pool *pp1;
                    630: #endif
1.92      enami     631:        size_t trysize, phsize;
1.134     ad        632:        int off, slack;
1.3       pk        633:
1.116     simonb    634: #ifdef DEBUG
                    635:        /*
                    636:         * Check that the pool hasn't already been initialised and
                    637:         * added to the list of all pools.
                    638:         */
                    639:        LIST_FOREACH(pp1, &pool_head, pr_poollist) {
                    640:                if (pp == pp1)
                    641:                        panic("pool_init: pool %s already initialised",
                    642:                            wchan);
                    643:        }
                    644: #endif
                    645:
1.25      thorpej   646: #ifdef POOL_DIAGNOSTIC
                    647:        /*
                    648:         * Always log if POOL_DIAGNOSTIC is defined.
                    649:         */
                    650:        if (pool_logsize != 0)
                    651:                flags |= PR_LOGGING;
                    652: #endif
                    653:
1.66      thorpej   654:        if (palloc == NULL)
                    655:                palloc = &pool_allocator_kmem;
1.112     bjh21     656: #ifdef POOL_SUBPAGE
                    657:        if (size > palloc->pa_pagesz) {
                    658:                if (palloc == &pool_allocator_kmem)
                    659:                        palloc = &pool_allocator_kmem_fullpage;
                    660:                else if (palloc == &pool_allocator_nointr)
                    661:                        palloc = &pool_allocator_nointr_fullpage;
                    662:        }
1.66      thorpej   663: #endif /* POOL_SUBPAGE */
                    664:        if ((palloc->pa_flags & PA_INITIALIZED) == 0) {
1.112     bjh21     665:                if (palloc->pa_pagesz == 0)
1.66      thorpej   666:                        palloc->pa_pagesz = PAGE_SIZE;
                    667:
                    668:                TAILQ_INIT(&palloc->pa_list);
                    669:
1.134     ad        670:                mutex_init(&palloc->pa_lock, MUTEX_DEFAULT, IPL_VM);
1.66      thorpej   671:                palloc->pa_pagemask = ~(palloc->pa_pagesz - 1);
                    672:                palloc->pa_pageshift = ffs(palloc->pa_pagesz) - 1;
1.117     yamt      673:
                    674:                if (palloc->pa_backingmapptr != NULL) {
                    675:                        pa_reclaim_register(palloc);
                    676:                }
1.66      thorpej   677:                palloc->pa_flags |= PA_INITIALIZED;
1.4       thorpej   678:        }
1.3       pk        679:
                    680:        if (align == 0)
                    681:                align = ALIGN(1);
1.14      thorpej   682:
1.120     yamt      683:        if ((flags & PR_NOTOUCH) == 0 && size < sizeof(struct pool_item))
1.14      thorpej   684:                size = sizeof(struct pool_item);
1.3       pk        685:
1.78      thorpej   686:        size = roundup(size, align);
1.66      thorpej   687: #ifdef DIAGNOSTIC
                    688:        if (size > palloc->pa_pagesz)
1.121     yamt      689:                panic("pool_init: pool item size (%zu) too large", size);
1.66      thorpej   690: #endif
1.35      pk        691:
1.3       pk        692:        /*
                    693:         * Initialize the pool structure.
                    694:         */
1.88      chs       695:        LIST_INIT(&pp->pr_emptypages);
                    696:        LIST_INIT(&pp->pr_fullpages);
                    697:        LIST_INIT(&pp->pr_partpages);
1.134     ad        698:        pp->pr_cache = NULL;
1.3       pk        699:        pp->pr_curpage = NULL;
                    700:        pp->pr_npages = 0;
                    701:        pp->pr_minitems = 0;
                    702:        pp->pr_minpages = 0;
                    703:        pp->pr_maxpages = UINT_MAX;
1.20      thorpej   704:        pp->pr_roflags = flags;
                    705:        pp->pr_flags = 0;
1.35      pk        706:        pp->pr_size = size;
1.3       pk        707:        pp->pr_align = align;
                    708:        pp->pr_wchan = wchan;
1.66      thorpej   709:        pp->pr_alloc = palloc;
1.20      thorpej   710:        pp->pr_nitems = 0;
                    711:        pp->pr_nout = 0;
                    712:        pp->pr_hardlimit = UINT_MAX;
                    713:        pp->pr_hardlimit_warning = NULL;
1.31      thorpej   714:        pp->pr_hardlimit_ratecap.tv_sec = 0;
                    715:        pp->pr_hardlimit_ratecap.tv_usec = 0;
                    716:        pp->pr_hardlimit_warning_last.tv_sec = 0;
                    717:        pp->pr_hardlimit_warning_last.tv_usec = 0;
1.68      thorpej   718:        pp->pr_drain_hook = NULL;
                    719:        pp->pr_drain_hook_arg = NULL;
1.125     ad        720:        pp->pr_freecheck = NULL;
1.3       pk        721:
                    722:        /*
                    723:         * Decide whether to put the page header off page to avoid
1.92      enami     724:         * wasting too large a part of the page or too big item.
                    725:         * Off-page page headers go on a hash table, so we can match
                    726:         * a returned item with its header based on the page address.
                    727:         * We use 1/16 of the page size and about 8 times of the item
                    728:         * size as the threshold (XXX: tune)
                    729:         *
                    730:         * However, we'll put the header into the page if we can put
                    731:         * it without wasting any items.
                    732:         *
                    733:         * Silently enforce `0 <= ioff < align'.
1.3       pk        734:         */
1.92      enami     735:        pp->pr_itemoffset = ioff %= align;
                    736:        /* See the comment below about reserved bytes. */
                    737:        trysize = palloc->pa_pagesz - ((align - ioff) % align);
                    738:        phsize = ALIGN(sizeof(struct pool_item_header));
1.121     yamt      739:        if ((pp->pr_roflags & (PR_NOTOUCH | PR_NOALIGN)) == 0 &&
1.97      yamt      740:            (pp->pr_size < MIN(palloc->pa_pagesz / 16, phsize << 3) ||
                    741:            trysize / pp->pr_size == (trysize - phsize) / pp->pr_size)) {
1.3       pk        742:                /* Use the end of the page for the page header */
1.20      thorpej   743:                pp->pr_roflags |= PR_PHINPAGE;
1.92      enami     744:                pp->pr_phoffset = off = palloc->pa_pagesz - phsize;
1.2       pk        745:        } else {
1.3       pk        746:                /* The page header will be taken from our page header pool */
                    747:                pp->pr_phoffset = 0;
1.66      thorpej   748:                off = palloc->pa_pagesz;
1.88      chs       749:                SPLAY_INIT(&pp->pr_phtree);
1.2       pk        750:        }
1.1       pk        751:
1.3       pk        752:        /*
                    753:         * Alignment is to take place at `ioff' within the item. This means
                    754:         * we must reserve up to `align - 1' bytes on the page to allow
                    755:         * appropriate positioning of each item.
                    756:         */
                    757:        pp->pr_itemsperpage = (off - ((align - ioff) % align)) / pp->pr_size;
1.43      thorpej   758:        KASSERT(pp->pr_itemsperpage != 0);
1.97      yamt      759:        if ((pp->pr_roflags & PR_NOTOUCH)) {
                    760:                int idx;
                    761:
                    762:                for (idx = 0; pp->pr_itemsperpage > PHPOOL_FREELIST_NELEM(idx);
                    763:                    idx++) {
                    764:                        /* nothing */
                    765:                }
                    766:                if (idx >= PHPOOL_MAX) {
                    767:                        /*
                    768:                         * if you see this panic, consider to tweak
                    769:                         * PHPOOL_MAX and PHPOOL_FREELIST_NELEM.
                    770:                         */
                    771:                        panic("%s: too large itemsperpage(%d) for PR_NOTOUCH",
                    772:                            pp->pr_wchan, pp->pr_itemsperpage);
                    773:                }
                    774:                pp->pr_phpool = &phpool[idx];
                    775:        } else if ((pp->pr_roflags & PR_PHINPAGE) == 0) {
                    776:                pp->pr_phpool = &phpool[0];
                    777:        }
                    778: #if defined(DIAGNOSTIC)
                    779:        else {
                    780:                pp->pr_phpool = NULL;
                    781:        }
                    782: #endif
1.3       pk        783:
                    784:        /*
                    785:         * Use the slack between the chunks and the page header
                    786:         * for "cache coloring".
                    787:         */
                    788:        slack = off - pp->pr_itemsperpage * pp->pr_size;
                    789:        pp->pr_maxcolor = (slack / align) * align;
                    790:        pp->pr_curcolor = 0;
                    791:
                    792:        pp->pr_nget = 0;
                    793:        pp->pr_nfail = 0;
                    794:        pp->pr_nput = 0;
                    795:        pp->pr_npagealloc = 0;
                    796:        pp->pr_npagefree = 0;
1.1       pk        797:        pp->pr_hiwat = 0;
1.8       thorpej   798:        pp->pr_nidle = 0;
1.134     ad        799:        pp->pr_refcnt = 0;
1.3       pk        800:
1.59      thorpej   801: #ifdef POOL_DIAGNOSTIC
1.25      thorpej   802:        if (flags & PR_LOGGING) {
                    803:                if (kmem_map == NULL ||
                    804:                    (pp->pr_log = malloc(pool_logsize * sizeof(struct pool_log),
                    805:                     M_TEMP, M_NOWAIT)) == NULL)
1.20      thorpej   806:                        pp->pr_roflags &= ~PR_LOGGING;
1.3       pk        807:                pp->pr_curlogentry = 0;
                    808:                pp->pr_logsize = pool_logsize;
                    809:        }
1.59      thorpej   810: #endif
1.25      thorpej   811:
                    812:        pp->pr_entered_file = NULL;
                    813:        pp->pr_entered_line = 0;
1.3       pk        814:
1.138     ad        815:        /*
                    816:         * XXXAD hack to prevent IP input processing from blocking.
                    817:         */
                    818:        if (ipl == IPL_SOFTNET) {
                    819:                mutex_init(&pp->pr_lock, MUTEX_DEFAULT, IPL_VM);
                    820:        } else {
                    821:                mutex_init(&pp->pr_lock, MUTEX_DEFAULT, ipl);
                    822:        }
1.134     ad        823:        cv_init(&pp->pr_cv, wchan);
                    824:        pp->pr_ipl = ipl;
1.1       pk        825:
1.3       pk        826:        /*
1.43      thorpej   827:         * Initialize private page header pool and cache magazine pool if we
                    828:         * haven't done so yet.
1.23      thorpej   829:         * XXX LOCKING.
1.3       pk        830:         */
1.97      yamt      831:        if (phpool[0].pr_size == 0) {
                    832:                int idx;
                    833:                for (idx = 0; idx < PHPOOL_MAX; idx++) {
                    834:                        static char phpool_names[PHPOOL_MAX][6+1+6+1];
                    835:                        int nelem;
                    836:                        size_t sz;
                    837:
                    838:                        nelem = PHPOOL_FREELIST_NELEM(idx);
                    839:                        snprintf(phpool_names[idx], sizeof(phpool_names[idx]),
                    840:                            "phpool-%d", nelem);
                    841:                        sz = sizeof(struct pool_item_header);
                    842:                        if (nelem) {
1.135     yamt      843:                                sz = offsetof(struct pool_item_header,
                    844:                                    ph_bitmap[howmany(nelem, BITMAP_SIZE)]);
1.97      yamt      845:                        }
                    846:                        pool_init(&phpool[idx], sz, 0, 0, 0,
1.129     ad        847:                            phpool_names[idx], &pool_allocator_meta, IPL_VM);
1.97      yamt      848:                }
1.62      bjh21     849: #ifdef POOL_SUBPAGE
                    850:                pool_init(&psppool, POOL_SUBPAGE, POOL_SUBPAGE, 0,
1.129     ad        851:                    PR_RECURSIVE, "psppool", &pool_allocator_meta, IPL_VM);
1.62      bjh21     852: #endif
1.134     ad        853:                pool_init(&pcgpool, sizeof(pcg_t), CACHE_LINE_SIZE, 0, 0,
                    854:                    "cachegrp", &pool_allocator_meta, IPL_VM);
1.1       pk        855:        }
                    856:
1.134     ad        857:        if (__predict_true(!cold)) {
                    858:                /* Insert into the list of all pools. */
                    859:                mutex_enter(&pool_head_lock);
                    860:                LIST_INSERT_HEAD(&pool_head, pp, pr_poollist);
                    861:                mutex_exit(&pool_head_lock);
                    862:
                    863:                /* Insert this into the list of pools using this allocator. */
                    864:                mutex_enter(&palloc->pa_lock);
                    865:                TAILQ_INSERT_TAIL(&palloc->pa_list, pp, pr_alloc_list);
                    866:                mutex_exit(&palloc->pa_lock);
                    867:        } else {
                    868:                LIST_INSERT_HEAD(&pool_head, pp, pr_poollist);
                    869:                TAILQ_INSERT_TAIL(&palloc->pa_list, pp, pr_alloc_list);
                    870:        }
1.66      thorpej   871:
1.117     yamt      872:        pool_reclaim_register(pp);
1.1       pk        873: }
                    874:
                    875: /*
                    876:  * De-commision a pool resource.
                    877:  */
                    878: void
1.42      thorpej   879: pool_destroy(struct pool *pp)
1.1       pk        880: {
1.101     thorpej   881:        struct pool_pagelist pq;
1.3       pk        882:        struct pool_item_header *ph;
1.43      thorpej   883:
1.101     thorpej   884:        /* Remove from global pool list */
1.134     ad        885:        mutex_enter(&pool_head_lock);
                    886:        while (pp->pr_refcnt != 0)
                    887:                cv_wait(&pool_busy, &pool_head_lock);
1.102     chs       888:        LIST_REMOVE(pp, pr_poollist);
1.101     thorpej   889:        if (drainpp == pp)
                    890:                drainpp = NULL;
1.134     ad        891:        mutex_exit(&pool_head_lock);
1.101     thorpej   892:
                    893:        /* Remove this pool from its allocator's list of pools. */
1.117     yamt      894:        pool_reclaim_unregister(pp);
1.134     ad        895:        mutex_enter(&pp->pr_alloc->pa_lock);
1.66      thorpej   896:        TAILQ_REMOVE(&pp->pr_alloc->pa_list, pp, pr_alloc_list);
1.134     ad        897:        mutex_exit(&pp->pr_alloc->pa_lock);
1.66      thorpej   898:
1.134     ad        899:        mutex_enter(&pp->pr_lock);
1.101     thorpej   900:
1.134     ad        901:        KASSERT(pp->pr_cache == NULL);
1.3       pk        902:
                    903: #ifdef DIAGNOSTIC
1.20      thorpej   904:        if (pp->pr_nout != 0) {
1.25      thorpej   905:                pr_printlog(pp, NULL, printf);
1.80      provos    906:                panic("pool_destroy: pool busy: still out: %u",
1.20      thorpej   907:                    pp->pr_nout);
1.3       pk        908:        }
                    909: #endif
1.1       pk        910:
1.101     thorpej   911:        KASSERT(LIST_EMPTY(&pp->pr_fullpages));
                    912:        KASSERT(LIST_EMPTY(&pp->pr_partpages));
                    913:
1.3       pk        914:        /* Remove all pages */
1.101     thorpej   915:        LIST_INIT(&pq);
1.88      chs       916:        while ((ph = LIST_FIRST(&pp->pr_emptypages)) != NULL)
1.101     thorpej   917:                pr_rmpage(pp, ph, &pq);
                    918:
1.134     ad        919:        mutex_exit(&pp->pr_lock);
1.3       pk        920:
1.101     thorpej   921:        pr_pagelist_free(pp, &pq);
1.3       pk        922:
1.59      thorpej   923: #ifdef POOL_DIAGNOSTIC
1.20      thorpej   924:        if ((pp->pr_roflags & PR_LOGGING) != 0)
1.3       pk        925:                free(pp->pr_log, M_TEMP);
1.59      thorpej   926: #endif
1.134     ad        927:
                    928:        cv_destroy(&pp->pr_cv);
                    929:        mutex_destroy(&pp->pr_lock);
1.1       pk        930: }
                    931:
1.68      thorpej   932: void
                    933: pool_set_drain_hook(struct pool *pp, void (*fn)(void *, int), void *arg)
                    934: {
                    935:
                    936:        /* XXX no locking -- must be used just after pool_init() */
                    937: #ifdef DIAGNOSTIC
                    938:        if (pp->pr_drain_hook != NULL)
                    939:                panic("pool_set_drain_hook(%s): already set", pp->pr_wchan);
                    940: #endif
                    941:        pp->pr_drain_hook = fn;
                    942:        pp->pr_drain_hook_arg = arg;
                    943: }
                    944:
1.88      chs       945: static struct pool_item_header *
1.128     christos  946: pool_alloc_item_header(struct pool *pp, void *storage, int flags)
1.55      thorpej   947: {
                    948:        struct pool_item_header *ph;
                    949:
                    950:        if ((pp->pr_roflags & PR_PHINPAGE) != 0)
1.128     christos  951:                ph = (struct pool_item_header *) ((char *)storage + pp->pr_phoffset);
1.134     ad        952:        else
1.97      yamt      953:                ph = pool_get(pp->pr_phpool, flags);
1.55      thorpej   954:
                    955:        return (ph);
                    956: }
1.1       pk        957:
                    958: /*
1.134     ad        959:  * Grab an item from the pool.
1.1       pk        960:  */
1.3       pk        961: void *
1.59      thorpej   962: #ifdef POOL_DIAGNOSTIC
1.42      thorpej   963: _pool_get(struct pool *pp, int flags, const char *file, long line)
1.56      sommerfe  964: #else
                    965: pool_get(struct pool *pp, int flags)
                    966: #endif
1.1       pk        967: {
                    968:        struct pool_item *pi;
1.3       pk        969:        struct pool_item_header *ph;
1.55      thorpej   970:        void *v;
1.1       pk        971:
1.2       pk        972: #ifdef DIAGNOSTIC
1.95      atatat    973:        if (__predict_false(pp->pr_itemsperpage == 0))
                    974:                panic("pool_get: pool %p: pr_itemsperpage is zero, "
                    975:                    "pool not initialized?", pp);
1.84      thorpej   976:        if (__predict_false(curlwp == NULL && doing_shutdown == 0 &&
1.37      sommerfe  977:                            (flags & PR_WAITOK) != 0))
1.77      matt      978:                panic("pool_get: %s: must have NOWAIT", pp->pr_wchan);
1.58      thorpej   979:
1.102     chs       980: #endif /* DIAGNOSTIC */
1.58      thorpej   981: #ifdef LOCKDEBUG
                    982:        if (flags & PR_WAITOK)
1.119     yamt      983:                ASSERT_SLEEPABLE(NULL, "pool_get(PR_WAITOK)");
1.56      sommerfe  984: #endif
1.1       pk        985:
1.134     ad        986:        mutex_enter(&pp->pr_lock);
1.25      thorpej   987:        pr_enter(pp, file, line);
1.20      thorpej   988:
                    989:  startover:
                    990:        /*
                    991:         * Check to see if we've reached the hard limit.  If we have,
                    992:         * and we can wait, then wait until an item has been returned to
                    993:         * the pool.
                    994:         */
                    995: #ifdef DIAGNOSTIC
1.34      thorpej   996:        if (__predict_false(pp->pr_nout > pp->pr_hardlimit)) {
1.25      thorpej   997:                pr_leave(pp);
1.134     ad        998:                mutex_exit(&pp->pr_lock);
1.20      thorpej   999:                panic("pool_get: %s: crossed hard limit", pp->pr_wchan);
                   1000:        }
                   1001: #endif
1.34      thorpej  1002:        if (__predict_false(pp->pr_nout == pp->pr_hardlimit)) {
1.68      thorpej  1003:                if (pp->pr_drain_hook != NULL) {
                   1004:                        /*
                   1005:                         * Since the drain hook is going to free things
                   1006:                         * back to the pool, unlock, call the hook, re-lock,
                   1007:                         * and check the hardlimit condition again.
                   1008:                         */
                   1009:                        pr_leave(pp);
1.134     ad       1010:                        mutex_exit(&pp->pr_lock);
1.68      thorpej  1011:                        (*pp->pr_drain_hook)(pp->pr_drain_hook_arg, flags);
1.134     ad       1012:                        mutex_enter(&pp->pr_lock);
1.68      thorpej  1013:                        pr_enter(pp, file, line);
                   1014:                        if (pp->pr_nout < pp->pr_hardlimit)
                   1015:                                goto startover;
                   1016:                }
                   1017:
1.29      sommerfe 1018:                if ((flags & PR_WAITOK) && !(flags & PR_LIMITFAIL)) {
1.20      thorpej  1019:                        /*
                   1020:                         * XXX: A warning isn't logged in this case.  Should
                   1021:                         * it be?
                   1022:                         */
                   1023:                        pp->pr_flags |= PR_WANTED;
1.25      thorpej  1024:                        pr_leave(pp);
1.134     ad       1025:                        cv_wait(&pp->pr_cv, &pp->pr_lock);
1.25      thorpej  1026:                        pr_enter(pp, file, line);
1.20      thorpej  1027:                        goto startover;
                   1028:                }
1.31      thorpej  1029:
                   1030:                /*
                   1031:                 * Log a message that the hard limit has been hit.
                   1032:                 */
                   1033:                if (pp->pr_hardlimit_warning != NULL &&
                   1034:                    ratecheck(&pp->pr_hardlimit_warning_last,
                   1035:                              &pp->pr_hardlimit_ratecap))
                   1036:                        log(LOG_ERR, "%s\n", pp->pr_hardlimit_warning);
1.21      thorpej  1037:
                   1038:                pp->pr_nfail++;
                   1039:
1.25      thorpej  1040:                pr_leave(pp);
1.134     ad       1041:                mutex_exit(&pp->pr_lock);
1.20      thorpej  1042:                return (NULL);
                   1043:        }
                   1044:
1.3       pk       1045:        /*
                   1046:         * The convention we use is that if `curpage' is not NULL, then
                   1047:         * it points at a non-empty bucket. In particular, `curpage'
                   1048:         * never points at a page header which has PR_PHINPAGE set and
                   1049:         * has no items in its bucket.
                   1050:         */
1.20      thorpej  1051:        if ((ph = pp->pr_curpage) == NULL) {
1.113     yamt     1052:                int error;
                   1053:
1.20      thorpej  1054: #ifdef DIAGNOSTIC
                   1055:                if (pp->pr_nitems != 0) {
1.134     ad       1056:                        mutex_exit(&pp->pr_lock);
1.20      thorpej  1057:                        printf("pool_get: %s: curpage NULL, nitems %u\n",
                   1058:                            pp->pr_wchan, pp->pr_nitems);
1.80      provos   1059:                        panic("pool_get: nitems inconsistent");
1.20      thorpej  1060:                }
                   1061: #endif
                   1062:
1.21      thorpej  1063:                /*
                   1064:                 * Call the back-end page allocator for more memory.
                   1065:                 * Release the pool lock, as the back-end page allocator
                   1066:                 * may block.
                   1067:                 */
1.25      thorpej  1068:                pr_leave(pp);
1.113     yamt     1069:                error = pool_grow(pp, flags);
                   1070:                pr_enter(pp, file, line);
                   1071:                if (error != 0) {
1.21      thorpej  1072:                        /*
1.55      thorpej  1073:                         * We were unable to allocate a page or item
                   1074:                         * header, but we released the lock during
                   1075:                         * allocation, so perhaps items were freed
                   1076:                         * back to the pool.  Check for this case.
1.21      thorpej  1077:                         */
                   1078:                        if (pp->pr_curpage != NULL)
                   1079:                                goto startover;
1.15      pk       1080:
1.117     yamt     1081:                        pp->pr_nfail++;
1.25      thorpej  1082:                        pr_leave(pp);
1.134     ad       1083:                        mutex_exit(&pp->pr_lock);
1.117     yamt     1084:                        return (NULL);
1.1       pk       1085:                }
1.3       pk       1086:
1.20      thorpej  1087:                /* Start the allocation process over. */
                   1088:                goto startover;
1.3       pk       1089:        }
1.97      yamt     1090:        if (pp->pr_roflags & PR_NOTOUCH) {
                   1091: #ifdef DIAGNOSTIC
                   1092:                if (__predict_false(ph->ph_nmissing == pp->pr_itemsperpage)) {
                   1093:                        pr_leave(pp);
1.134     ad       1094:                        mutex_exit(&pp->pr_lock);
1.97      yamt     1095:                        panic("pool_get: %s: page empty", pp->pr_wchan);
                   1096:                }
                   1097: #endif
                   1098:                v = pr_item_notouch_get(pp, ph);
                   1099: #ifdef POOL_DIAGNOSTIC
                   1100:                pr_log(pp, v, PRLOG_GET, file, line);
                   1101: #endif
                   1102:        } else {
1.102     chs      1103:                v = pi = LIST_FIRST(&ph->ph_itemlist);
1.97      yamt     1104:                if (__predict_false(v == NULL)) {
                   1105:                        pr_leave(pp);
1.134     ad       1106:                        mutex_exit(&pp->pr_lock);
1.97      yamt     1107:                        panic("pool_get: %s: page empty", pp->pr_wchan);
                   1108:                }
1.20      thorpej  1109: #ifdef DIAGNOSTIC
1.97      yamt     1110:                if (__predict_false(pp->pr_nitems == 0)) {
                   1111:                        pr_leave(pp);
1.134     ad       1112:                        mutex_exit(&pp->pr_lock);
1.97      yamt     1113:                        printf("pool_get: %s: items on itemlist, nitems %u\n",
                   1114:                            pp->pr_wchan, pp->pr_nitems);
                   1115:                        panic("pool_get: nitems inconsistent");
                   1116:                }
1.65      enami    1117: #endif
1.56      sommerfe 1118:
1.65      enami    1119: #ifdef POOL_DIAGNOSTIC
1.97      yamt     1120:                pr_log(pp, v, PRLOG_GET, file, line);
1.65      enami    1121: #endif
1.3       pk       1122:
1.65      enami    1123: #ifdef DIAGNOSTIC
1.97      yamt     1124:                if (__predict_false(pi->pi_magic != PI_MAGIC)) {
                   1125:                        pr_printlog(pp, pi, printf);
                   1126:                        panic("pool_get(%s): free list modified: "
                   1127:                            "magic=%x; page %p; item addr %p\n",
                   1128:                            pp->pr_wchan, pi->pi_magic, ph->ph_page, pi);
                   1129:                }
1.3       pk       1130: #endif
                   1131:
1.97      yamt     1132:                /*
                   1133:                 * Remove from item list.
                   1134:                 */
1.102     chs      1135:                LIST_REMOVE(pi, pi_list);
1.97      yamt     1136:        }
1.20      thorpej  1137:        pp->pr_nitems--;
                   1138:        pp->pr_nout++;
1.6       thorpej  1139:        if (ph->ph_nmissing == 0) {
                   1140: #ifdef DIAGNOSTIC
1.34      thorpej  1141:                if (__predict_false(pp->pr_nidle == 0))
1.6       thorpej  1142:                        panic("pool_get: nidle inconsistent");
                   1143: #endif
                   1144:                pp->pr_nidle--;
1.88      chs      1145:
                   1146:                /*
                   1147:                 * This page was previously empty.  Move it to the list of
                   1148:                 * partially-full pages.  This page is already curpage.
                   1149:                 */
                   1150:                LIST_REMOVE(ph, ph_pagelist);
                   1151:                LIST_INSERT_HEAD(&pp->pr_partpages, ph, ph_pagelist);
1.6       thorpej  1152:        }
1.3       pk       1153:        ph->ph_nmissing++;
1.97      yamt     1154:        if (ph->ph_nmissing == pp->pr_itemsperpage) {
1.21      thorpej  1155: #ifdef DIAGNOSTIC
1.97      yamt     1156:                if (__predict_false((pp->pr_roflags & PR_NOTOUCH) == 0 &&
1.102     chs      1157:                    !LIST_EMPTY(&ph->ph_itemlist))) {
1.25      thorpej  1158:                        pr_leave(pp);
1.134     ad       1159:                        mutex_exit(&pp->pr_lock);
1.21      thorpej  1160:                        panic("pool_get: %s: nmissing inconsistent",
                   1161:                            pp->pr_wchan);
                   1162:                }
                   1163: #endif
1.3       pk       1164:                /*
1.88      chs      1165:                 * This page is now full.  Move it to the full list
                   1166:                 * and select a new current page.
1.3       pk       1167:                 */
1.88      chs      1168:                LIST_REMOVE(ph, ph_pagelist);
                   1169:                LIST_INSERT_HEAD(&pp->pr_fullpages, ph, ph_pagelist);
                   1170:                pool_update_curpage(pp);
1.1       pk       1171:        }
1.3       pk       1172:
                   1173:        pp->pr_nget++;
1.111     christos 1174:        pr_leave(pp);
1.20      thorpej  1175:
                   1176:        /*
                   1177:         * If we have a low water mark and we are now below that low
                   1178:         * water mark, add more items to the pool.
                   1179:         */
1.53      thorpej  1180:        if (POOL_NEEDS_CATCHUP(pp) && pool_catchup(pp) != 0) {
1.20      thorpej  1181:                /*
                   1182:                 * XXX: Should we log a warning?  Should we set up a timeout
                   1183:                 * to try again in a second or so?  The latter could break
                   1184:                 * a caller's assumptions about interrupt protection, etc.
                   1185:                 */
                   1186:        }
                   1187:
1.134     ad       1188:        mutex_exit(&pp->pr_lock);
1.125     ad       1189:        KASSERT((((vaddr_t)v + pp->pr_itemoffset) & (pp->pr_align - 1)) == 0);
                   1190:        FREECHECK_OUT(&pp->pr_freecheck, v);
1.1       pk       1191:        return (v);
                   1192: }
                   1193:
                   1194: /*
1.43      thorpej  1195:  * Internal version of pool_put().  Pool is already locked/entered.
1.1       pk       1196:  */
1.43      thorpej  1197: static void
1.101     thorpej  1198: pool_do_put(struct pool *pp, void *v, struct pool_pagelist *pq)
1.1       pk       1199: {
                   1200:        struct pool_item *pi = v;
1.3       pk       1201:        struct pool_item_header *ph;
                   1202:
1.134     ad       1203:        KASSERT(mutex_owned(&pp->pr_lock));
1.125     ad       1204:        FREECHECK_IN(&pp->pr_freecheck, v);
1.134     ad       1205:        LOCKDEBUG_MEM_CHECK(v, pp->pr_size);
1.61      chs      1206:
1.30      thorpej  1207: #ifdef DIAGNOSTIC
1.34      thorpej  1208:        if (__predict_false(pp->pr_nout == 0)) {
1.30      thorpej  1209:                printf("pool %s: putting with none out\n",
                   1210:                    pp->pr_wchan);
                   1211:                panic("pool_put");
                   1212:        }
                   1213: #endif
1.3       pk       1214:
1.121     yamt     1215:        if (__predict_false((ph = pr_find_pagehead(pp, v)) == NULL)) {
1.25      thorpej  1216:                pr_printlog(pp, NULL, printf);
1.3       pk       1217:                panic("pool_put: %s: page header missing", pp->pr_wchan);
                   1218:        }
1.28      thorpej  1219:
1.3       pk       1220:        /*
                   1221:         * Return to item list.
                   1222:         */
1.97      yamt     1223:        if (pp->pr_roflags & PR_NOTOUCH) {
                   1224:                pr_item_notouch_put(pp, ph, v);
                   1225:        } else {
1.2       pk       1226: #ifdef DIAGNOSTIC
1.97      yamt     1227:                pi->pi_magic = PI_MAGIC;
1.3       pk       1228: #endif
1.32      chs      1229: #ifdef DEBUG
1.97      yamt     1230:                {
                   1231:                        int i, *ip = v;
1.32      chs      1232:
1.97      yamt     1233:                        for (i = 0; i < pp->pr_size / sizeof(int); i++) {
                   1234:                                *ip++ = PI_MAGIC;
                   1235:                        }
1.32      chs      1236:                }
                   1237: #endif
                   1238:
1.102     chs      1239:                LIST_INSERT_HEAD(&ph->ph_itemlist, pi, pi_list);
1.97      yamt     1240:        }
1.79      thorpej  1241:        KDASSERT(ph->ph_nmissing != 0);
1.3       pk       1242:        ph->ph_nmissing--;
                   1243:        pp->pr_nput++;
1.20      thorpej  1244:        pp->pr_nitems++;
                   1245:        pp->pr_nout--;
1.3       pk       1246:
                   1247:        /* Cancel "pool empty" condition if it exists */
                   1248:        if (pp->pr_curpage == NULL)
                   1249:                pp->pr_curpage = ph;
                   1250:
                   1251:        if (pp->pr_flags & PR_WANTED) {
                   1252:                pp->pr_flags &= ~PR_WANTED;
1.15      pk       1253:                if (ph->ph_nmissing == 0)
                   1254:                        pp->pr_nidle++;
1.134     ad       1255:                cv_broadcast(&pp->pr_cv);
1.3       pk       1256:                return;
                   1257:        }
                   1258:
                   1259:        /*
1.88      chs      1260:         * If this page is now empty, do one of two things:
1.21      thorpej  1261:         *
1.88      chs      1262:         *      (1) If we have more pages than the page high water mark,
1.96      thorpej  1263:         *          free the page back to the system.  ONLY CONSIDER
1.90      thorpej  1264:         *          FREEING BACK A PAGE IF WE HAVE MORE THAN OUR MINIMUM PAGE
                   1265:         *          CLAIM.
1.21      thorpej  1266:         *
1.88      chs      1267:         *      (2) Otherwise, move the page to the empty page list.
                   1268:         *
                   1269:         * Either way, select a new current page (so we use a partially-full
                   1270:         * page if one is available).
1.3       pk       1271:         */
                   1272:        if (ph->ph_nmissing == 0) {
1.6       thorpej  1273:                pp->pr_nidle++;
1.90      thorpej  1274:                if (pp->pr_npages > pp->pr_minpages &&
                   1275:                    (pp->pr_npages > pp->pr_maxpages ||
1.117     yamt     1276:                     pa_starved_p(pp->pr_alloc))) {
1.101     thorpej  1277:                        pr_rmpage(pp, ph, pq);
1.3       pk       1278:                } else {
1.88      chs      1279:                        LIST_REMOVE(ph, ph_pagelist);
                   1280:                        LIST_INSERT_HEAD(&pp->pr_emptypages, ph, ph_pagelist);
1.3       pk       1281:
1.21      thorpej  1282:                        /*
                   1283:                         * Update the timestamp on the page.  A page must
                   1284:                         * be idle for some period of time before it can
                   1285:                         * be reclaimed by the pagedaemon.  This minimizes
                   1286:                         * ping-pong'ing for memory.
                   1287:                         */
1.118     kardel   1288:                        getmicrotime(&ph->ph_time);
1.1       pk       1289:                }
1.88      chs      1290:                pool_update_curpage(pp);
1.1       pk       1291:        }
1.88      chs      1292:
1.21      thorpej  1293:        /*
1.88      chs      1294:         * If the page was previously completely full, move it to the
                   1295:         * partially-full list and make it the current page.  The next
                   1296:         * allocation will get the item from this page, instead of
                   1297:         * further fragmenting the pool.
1.21      thorpej  1298:         */
                   1299:        else if (ph->ph_nmissing == (pp->pr_itemsperpage - 1)) {
1.88      chs      1300:                LIST_REMOVE(ph, ph_pagelist);
                   1301:                LIST_INSERT_HEAD(&pp->pr_partpages, ph, ph_pagelist);
1.21      thorpej  1302:                pp->pr_curpage = ph;
                   1303:        }
1.43      thorpej  1304: }
                   1305:
                   1306: /*
1.134     ad       1307:  * Return resource to the pool.
1.43      thorpej  1308:  */
1.59      thorpej  1309: #ifdef POOL_DIAGNOSTIC
1.43      thorpej  1310: void
                   1311: _pool_put(struct pool *pp, void *v, const char *file, long line)
                   1312: {
1.101     thorpej  1313:        struct pool_pagelist pq;
                   1314:
                   1315:        LIST_INIT(&pq);
1.43      thorpej  1316:
1.134     ad       1317:        mutex_enter(&pp->pr_lock);
1.43      thorpej  1318:        pr_enter(pp, file, line);
                   1319:
1.56      sommerfe 1320:        pr_log(pp, v, PRLOG_PUT, file, line);
                   1321:
1.101     thorpej  1322:        pool_do_put(pp, v, &pq);
1.21      thorpej  1323:
1.25      thorpej  1324:        pr_leave(pp);
1.134     ad       1325:        mutex_exit(&pp->pr_lock);
1.101     thorpej  1326:
1.102     chs      1327:        pr_pagelist_free(pp, &pq);
1.1       pk       1328: }
1.57      sommerfe 1329: #undef pool_put
1.59      thorpej  1330: #endif /* POOL_DIAGNOSTIC */
1.1       pk       1331:
1.56      sommerfe 1332: void
                   1333: pool_put(struct pool *pp, void *v)
                   1334: {
1.101     thorpej  1335:        struct pool_pagelist pq;
                   1336:
                   1337:        LIST_INIT(&pq);
1.56      sommerfe 1338:
1.134     ad       1339:        mutex_enter(&pp->pr_lock);
1.101     thorpej  1340:        pool_do_put(pp, v, &pq);
1.134     ad       1341:        mutex_exit(&pp->pr_lock);
1.56      sommerfe 1342:
1.102     chs      1343:        pr_pagelist_free(pp, &pq);
1.56      sommerfe 1344: }
1.57      sommerfe 1345:
1.59      thorpej  1346: #ifdef POOL_DIAGNOSTIC
1.57      sommerfe 1347: #define                pool_put(h, v)  _pool_put((h), (v), __FILE__, __LINE__)
1.56      sommerfe 1348: #endif
1.74      thorpej  1349:
                   1350: /*
1.113     yamt     1351:  * pool_grow: grow a pool by a page.
                   1352:  *
                   1353:  * => called with pool locked.
                   1354:  * => unlock and relock the pool.
                   1355:  * => return with pool locked.
                   1356:  */
                   1357:
                   1358: static int
                   1359: pool_grow(struct pool *pp, int flags)
                   1360: {
                   1361:        struct pool_item_header *ph = NULL;
                   1362:        char *cp;
                   1363:
1.134     ad       1364:        mutex_exit(&pp->pr_lock);
1.113     yamt     1365:        cp = pool_allocator_alloc(pp, flags);
                   1366:        if (__predict_true(cp != NULL)) {
                   1367:                ph = pool_alloc_item_header(pp, cp, flags);
                   1368:        }
                   1369:        if (__predict_false(cp == NULL || ph == NULL)) {
                   1370:                if (cp != NULL) {
                   1371:                        pool_allocator_free(pp, cp);
                   1372:                }
1.134     ad       1373:                mutex_enter(&pp->pr_lock);
1.113     yamt     1374:                return ENOMEM;
                   1375:        }
                   1376:
1.134     ad       1377:        mutex_enter(&pp->pr_lock);
1.113     yamt     1378:        pool_prime_page(pp, cp, ph);
                   1379:        pp->pr_npagealloc++;
                   1380:        return 0;
                   1381: }
                   1382:
                   1383: /*
1.74      thorpej  1384:  * Add N items to the pool.
                   1385:  */
                   1386: int
                   1387: pool_prime(struct pool *pp, int n)
                   1388: {
1.75      simonb   1389:        int newpages;
1.113     yamt     1390:        int error = 0;
1.74      thorpej  1391:
1.134     ad       1392:        mutex_enter(&pp->pr_lock);
1.74      thorpej  1393:
                   1394:        newpages = roundup(n, pp->pr_itemsperpage) / pp->pr_itemsperpage;
                   1395:
                   1396:        while (newpages-- > 0) {
1.113     yamt     1397:                error = pool_grow(pp, PR_NOWAIT);
                   1398:                if (error) {
1.74      thorpej  1399:                        break;
                   1400:                }
                   1401:                pp->pr_minpages++;
                   1402:        }
                   1403:
                   1404:        if (pp->pr_minpages >= pp->pr_maxpages)
                   1405:                pp->pr_maxpages = pp->pr_minpages + 1;  /* XXX */
                   1406:
1.134     ad       1407:        mutex_exit(&pp->pr_lock);
1.113     yamt     1408:        return error;
1.74      thorpej  1409: }
1.55      thorpej  1410:
                   1411: /*
1.3       pk       1412:  * Add a page worth of items to the pool.
1.21      thorpej  1413:  *
                   1414:  * Note, we must be called with the pool descriptor LOCKED.
1.3       pk       1415:  */
1.55      thorpej  1416: static void
1.128     christos 1417: pool_prime_page(struct pool *pp, void *storage, struct pool_item_header *ph)
1.3       pk       1418: {
                   1419:        struct pool_item *pi;
1.128     christos 1420:        void *cp = storage;
1.125     ad       1421:        const unsigned int align = pp->pr_align;
                   1422:        const unsigned int ioff = pp->pr_itemoffset;
1.55      thorpej  1423:        int n;
1.36      pk       1424:
1.134     ad       1425:        KASSERT(mutex_owned(&pp->pr_lock));
1.91      yamt     1426:
1.66      thorpej  1427: #ifdef DIAGNOSTIC
1.121     yamt     1428:        if ((pp->pr_roflags & PR_NOALIGN) == 0 &&
                   1429:            ((uintptr_t)cp & (pp->pr_alloc->pa_pagesz - 1)) != 0)
1.36      pk       1430:                panic("pool_prime_page: %s: unaligned page", pp->pr_wchan);
1.66      thorpej  1431: #endif
1.3       pk       1432:
                   1433:        /*
                   1434:         * Insert page header.
                   1435:         */
1.88      chs      1436:        LIST_INSERT_HEAD(&pp->pr_emptypages, ph, ph_pagelist);
1.102     chs      1437:        LIST_INIT(&ph->ph_itemlist);
1.3       pk       1438:        ph->ph_page = storage;
                   1439:        ph->ph_nmissing = 0;
1.118     kardel   1440:        getmicrotime(&ph->ph_time);
1.88      chs      1441:        if ((pp->pr_roflags & PR_PHINPAGE) == 0)
                   1442:                SPLAY_INSERT(phtree, &pp->pr_phtree, ph);
1.3       pk       1443:
1.6       thorpej  1444:        pp->pr_nidle++;
                   1445:
1.3       pk       1446:        /*
                   1447:         * Color this page.
                   1448:         */
1.141   ! yamt     1449:        ph->ph_off = pp->pr_curcolor;
        !          1450:        cp = (char *)cp + ph->ph_off;
1.3       pk       1451:        if ((pp->pr_curcolor += align) > pp->pr_maxcolor)
                   1452:                pp->pr_curcolor = 0;
                   1453:
                   1454:        /*
                   1455:         * Adjust storage to apply aligment to `pr_itemoffset' in each item.
                   1456:         */
                   1457:        if (ioff != 0)
1.128     christos 1458:                cp = (char *)cp + align - ioff;
1.3       pk       1459:
1.125     ad       1460:        KASSERT((((vaddr_t)cp + ioff) & (align - 1)) == 0);
                   1461:
1.3       pk       1462:        /*
                   1463:         * Insert remaining chunks on the bucket list.
                   1464:         */
                   1465:        n = pp->pr_itemsperpage;
1.20      thorpej  1466:        pp->pr_nitems += n;
1.3       pk       1467:
1.97      yamt     1468:        if (pp->pr_roflags & PR_NOTOUCH) {
1.141   ! yamt     1469:                pr_item_notouch_init(pp, ph);
1.97      yamt     1470:        } else {
                   1471:                while (n--) {
                   1472:                        pi = (struct pool_item *)cp;
1.78      thorpej  1473:
1.97      yamt     1474:                        KASSERT(((((vaddr_t)pi) + ioff) & (align - 1)) == 0);
1.3       pk       1475:
1.97      yamt     1476:                        /* Insert on page list */
1.102     chs      1477:                        LIST_INSERT_HEAD(&ph->ph_itemlist, pi, pi_list);
1.3       pk       1478: #ifdef DIAGNOSTIC
1.97      yamt     1479:                        pi->pi_magic = PI_MAGIC;
1.3       pk       1480: #endif
1.128     christos 1481:                        cp = (char *)cp + pp->pr_size;
1.125     ad       1482:
                   1483:                        KASSERT((((vaddr_t)cp + ioff) & (align - 1)) == 0);
1.97      yamt     1484:                }
1.3       pk       1485:        }
                   1486:
                   1487:        /*
                   1488:         * If the pool was depleted, point at the new page.
                   1489:         */
                   1490:        if (pp->pr_curpage == NULL)
                   1491:                pp->pr_curpage = ph;
                   1492:
                   1493:        if (++pp->pr_npages > pp->pr_hiwat)
                   1494:                pp->pr_hiwat = pp->pr_npages;
                   1495: }
                   1496:
1.20      thorpej  1497: /*
1.52      thorpej  1498:  * Used by pool_get() when nitems drops below the low water mark.  This
1.88      chs      1499:  * is used to catch up pr_nitems with the low water mark.
1.20      thorpej  1500:  *
1.21      thorpej  1501:  * Note 1, we never wait for memory here, we let the caller decide what to do.
1.20      thorpej  1502:  *
1.73      thorpej  1503:  * Note 2, we must be called with the pool already locked, and we return
1.20      thorpej  1504:  * with it locked.
                   1505:  */
                   1506: static int
1.42      thorpej  1507: pool_catchup(struct pool *pp)
1.20      thorpej  1508: {
                   1509:        int error = 0;
                   1510:
1.54      thorpej  1511:        while (POOL_NEEDS_CATCHUP(pp)) {
1.113     yamt     1512:                error = pool_grow(pp, PR_NOWAIT);
                   1513:                if (error) {
1.20      thorpej  1514:                        break;
                   1515:                }
                   1516:        }
1.113     yamt     1517:        return error;
1.20      thorpej  1518: }
                   1519:
1.88      chs      1520: static void
                   1521: pool_update_curpage(struct pool *pp)
                   1522: {
                   1523:
                   1524:        pp->pr_curpage = LIST_FIRST(&pp->pr_partpages);
                   1525:        if (pp->pr_curpage == NULL) {
                   1526:                pp->pr_curpage = LIST_FIRST(&pp->pr_emptypages);
                   1527:        }
                   1528: }
                   1529:
1.3       pk       1530: void
1.42      thorpej  1531: pool_setlowat(struct pool *pp, int n)
1.3       pk       1532: {
1.15      pk       1533:
1.134     ad       1534:        mutex_enter(&pp->pr_lock);
1.21      thorpej  1535:
1.3       pk       1536:        pp->pr_minitems = n;
1.15      pk       1537:        pp->pr_minpages = (n == 0)
                   1538:                ? 0
1.18      thorpej  1539:                : roundup(n, pp->pr_itemsperpage) / pp->pr_itemsperpage;
1.20      thorpej  1540:
                   1541:        /* Make sure we're caught up with the newly-set low water mark. */
1.75      simonb   1542:        if (POOL_NEEDS_CATCHUP(pp) && pool_catchup(pp) != 0) {
1.20      thorpej  1543:                /*
                   1544:                 * XXX: Should we log a warning?  Should we set up a timeout
                   1545:                 * to try again in a second or so?  The latter could break
                   1546:                 * a caller's assumptions about interrupt protection, etc.
                   1547:                 */
                   1548:        }
1.21      thorpej  1549:
1.134     ad       1550:        mutex_exit(&pp->pr_lock);
1.3       pk       1551: }
                   1552:
                   1553: void
1.42      thorpej  1554: pool_sethiwat(struct pool *pp, int n)
1.3       pk       1555: {
1.15      pk       1556:
1.134     ad       1557:        mutex_enter(&pp->pr_lock);
1.21      thorpej  1558:
1.15      pk       1559:        pp->pr_maxpages = (n == 0)
                   1560:                ? 0
1.18      thorpej  1561:                : roundup(n, pp->pr_itemsperpage) / pp->pr_itemsperpage;
1.21      thorpej  1562:
1.134     ad       1563:        mutex_exit(&pp->pr_lock);
1.3       pk       1564: }
                   1565:
1.20      thorpej  1566: void
1.42      thorpej  1567: pool_sethardlimit(struct pool *pp, int n, const char *warnmess, int ratecap)
1.20      thorpej  1568: {
                   1569:
1.134     ad       1570:        mutex_enter(&pp->pr_lock);
1.20      thorpej  1571:
                   1572:        pp->pr_hardlimit = n;
                   1573:        pp->pr_hardlimit_warning = warnmess;
1.31      thorpej  1574:        pp->pr_hardlimit_ratecap.tv_sec = ratecap;
                   1575:        pp->pr_hardlimit_warning_last.tv_sec = 0;
                   1576:        pp->pr_hardlimit_warning_last.tv_usec = 0;
1.20      thorpej  1577:
                   1578:        /*
1.21      thorpej  1579:         * In-line version of pool_sethiwat(), because we don't want to
                   1580:         * release the lock.
1.20      thorpej  1581:         */
                   1582:        pp->pr_maxpages = (n == 0)
                   1583:                ? 0
                   1584:                : roundup(n, pp->pr_itemsperpage) / pp->pr_itemsperpage;
1.21      thorpej  1585:
1.134     ad       1586:        mutex_exit(&pp->pr_lock);
1.20      thorpej  1587: }
1.3       pk       1588:
                   1589: /*
                   1590:  * Release all complete pages that have not been used recently.
                   1591:  */
1.66      thorpej  1592: int
1.59      thorpej  1593: #ifdef POOL_DIAGNOSTIC
1.42      thorpej  1594: _pool_reclaim(struct pool *pp, const char *file, long line)
1.56      sommerfe 1595: #else
                   1596: pool_reclaim(struct pool *pp)
                   1597: #endif
1.3       pk       1598: {
                   1599:        struct pool_item_header *ph, *phnext;
1.61      chs      1600:        struct pool_pagelist pq;
1.102     chs      1601:        struct timeval curtime, diff;
1.134     ad       1602:        bool klock;
                   1603:        int rv;
1.3       pk       1604:
1.68      thorpej  1605:        if (pp->pr_drain_hook != NULL) {
                   1606:                /*
                   1607:                 * The drain hook must be called with the pool unlocked.
                   1608:                 */
                   1609:                (*pp->pr_drain_hook)(pp->pr_drain_hook_arg, PR_NOWAIT);
                   1610:        }
                   1611:
1.134     ad       1612:        /*
                   1613:         * XXXSMP Because mutexes at IPL_SOFTXXX are still spinlocks,
                   1614:         * and we are called from the pagedaemon without kernel_lock.
                   1615:         * Does not apply to IPL_SOFTBIO.
                   1616:         */
                   1617:        if (pp->pr_ipl == IPL_SOFTNET || pp->pr_ipl == IPL_SOFTCLOCK ||
                   1618:            pp->pr_ipl == IPL_SOFTSERIAL) {
                   1619:                KERNEL_LOCK(1, NULL);
                   1620:                klock = true;
                   1621:        } else
                   1622:                klock = false;
                   1623:
                   1624:        /* Reclaim items from the pool's cache (if any). */
                   1625:        if (pp->pr_cache != NULL)
                   1626:                pool_cache_invalidate(pp->pr_cache);
                   1627:
                   1628:        if (mutex_tryenter(&pp->pr_lock) == 0) {
                   1629:                if (klock) {
                   1630:                        KERNEL_UNLOCK_ONE(NULL);
                   1631:                }
1.66      thorpej  1632:                return (0);
1.134     ad       1633:        }
1.25      thorpej  1634:        pr_enter(pp, file, line);
1.68      thorpej  1635:
1.88      chs      1636:        LIST_INIT(&pq);
1.43      thorpej  1637:
1.118     kardel   1638:        getmicrotime(&curtime);
1.21      thorpej  1639:
1.88      chs      1640:        for (ph = LIST_FIRST(&pp->pr_emptypages); ph != NULL; ph = phnext) {
                   1641:                phnext = LIST_NEXT(ph, ph_pagelist);
1.3       pk       1642:
                   1643:                /* Check our minimum page claim */
                   1644:                if (pp->pr_npages <= pp->pr_minpages)
                   1645:                        break;
                   1646:
1.88      chs      1647:                KASSERT(ph->ph_nmissing == 0);
                   1648:                timersub(&curtime, &ph->ph_time, &diff);
1.117     yamt     1649:                if (diff.tv_sec < pool_inactive_time
                   1650:                    && !pa_starved_p(pp->pr_alloc))
1.88      chs      1651:                        continue;
1.21      thorpej  1652:
1.88      chs      1653:                /*
                   1654:                 * If freeing this page would put us below
                   1655:                 * the low water mark, stop now.
                   1656:                 */
                   1657:                if ((pp->pr_nitems - pp->pr_itemsperpage) <
                   1658:                    pp->pr_minitems)
                   1659:                        break;
1.21      thorpej  1660:
1.88      chs      1661:                pr_rmpage(pp, ph, &pq);
1.3       pk       1662:        }
                   1663:
1.25      thorpej  1664:        pr_leave(pp);
1.134     ad       1665:        mutex_exit(&pp->pr_lock);
                   1666:
                   1667:        if (LIST_EMPTY(&pq))
                   1668:                rv = 0;
                   1669:        else {
                   1670:                pr_pagelist_free(pp, &pq);
                   1671:                rv = 1;
                   1672:        }
                   1673:
                   1674:        if (klock) {
                   1675:                KERNEL_UNLOCK_ONE(NULL);
                   1676:        }
1.66      thorpej  1677:
1.134     ad       1678:        return (rv);
1.3       pk       1679: }
                   1680:
                   1681: /*
1.134     ad       1682:  * Drain pools, one at a time.  This is a two stage process;
                   1683:  * drain_start kicks off a cross call to drain CPU-level caches
                   1684:  * if the pool has an associated pool_cache.  drain_end waits
                   1685:  * for those cross calls to finish, and then drains the cache
                   1686:  * (if any) and pool.
1.131     ad       1687:  *
1.134     ad       1688:  * Note, must never be called from interrupt context.
1.3       pk       1689:  */
                   1690: void
1.134     ad       1691: pool_drain_start(struct pool **ppp, uint64_t *wp)
1.3       pk       1692: {
                   1693:        struct pool *pp;
1.134     ad       1694:
                   1695:        KASSERT(!LIST_EMPTY(&pool_head));
1.3       pk       1696:
1.61      chs      1697:        pp = NULL;
1.134     ad       1698:
                   1699:        /* Find next pool to drain, and add a reference. */
                   1700:        mutex_enter(&pool_head_lock);
                   1701:        do {
                   1702:                if (drainpp == NULL) {
                   1703:                        drainpp = LIST_FIRST(&pool_head);
                   1704:                }
                   1705:                if (drainpp != NULL) {
                   1706:                        pp = drainpp;
                   1707:                        drainpp = LIST_NEXT(pp, pr_poollist);
                   1708:                }
                   1709:                /*
                   1710:                 * Skip completely idle pools.  We depend on at least
                   1711:                 * one pool in the system being active.
                   1712:                 */
                   1713:        } while (pp == NULL || pp->pr_npages == 0);
                   1714:        pp->pr_refcnt++;
                   1715:        mutex_exit(&pool_head_lock);
                   1716:
                   1717:        /* If there is a pool_cache, drain CPU level caches. */
                   1718:        *ppp = pp;
                   1719:        if (pp->pr_cache != NULL) {
                   1720:                *wp = xc_broadcast(0, (xcfunc_t)pool_cache_xcall,
                   1721:                    pp->pr_cache, NULL);
                   1722:        }
                   1723: }
                   1724:
                   1725: void
                   1726: pool_drain_end(struct pool *pp, uint64_t where)
                   1727: {
                   1728:
                   1729:        if (pp == NULL)
                   1730:                return;
                   1731:
                   1732:        KASSERT(pp->pr_refcnt > 0);
                   1733:
                   1734:        /* Wait for remote draining to complete. */
                   1735:        if (pp->pr_cache != NULL)
                   1736:                xc_wait(where);
                   1737:
                   1738:        /* Drain the cache (if any) and pool.. */
                   1739:        pool_reclaim(pp);
                   1740:
                   1741:        /* Finally, unlock the pool. */
                   1742:        mutex_enter(&pool_head_lock);
                   1743:        pp->pr_refcnt--;
                   1744:        cv_broadcast(&pool_busy);
                   1745:        mutex_exit(&pool_head_lock);
1.3       pk       1746: }
                   1747:
                   1748: /*
                   1749:  * Diagnostic helpers.
                   1750:  */
                   1751: void
1.42      thorpej  1752: pool_print(struct pool *pp, const char *modif)
1.21      thorpej  1753: {
                   1754:
1.25      thorpej  1755:        pool_print1(pp, modif, printf);
1.21      thorpej  1756: }
                   1757:
1.25      thorpej  1758: void
1.108     yamt     1759: pool_printall(const char *modif, void (*pr)(const char *, ...))
                   1760: {
                   1761:        struct pool *pp;
                   1762:
                   1763:        LIST_FOREACH(pp, &pool_head, pr_poollist) {
                   1764:                pool_printit(pp, modif, pr);
                   1765:        }
                   1766: }
                   1767:
                   1768: void
1.42      thorpej  1769: pool_printit(struct pool *pp, const char *modif, void (*pr)(const char *, ...))
1.25      thorpej  1770: {
                   1771:
                   1772:        if (pp == NULL) {
                   1773:                (*pr)("Must specify a pool to print.\n");
                   1774:                return;
                   1775:        }
                   1776:
                   1777:        pool_print1(pp, modif, pr);
                   1778: }
                   1779:
1.21      thorpej  1780: static void
1.124     yamt     1781: pool_print_pagelist(struct pool *pp, struct pool_pagelist *pl,
1.97      yamt     1782:     void (*pr)(const char *, ...))
1.88      chs      1783: {
                   1784:        struct pool_item_header *ph;
                   1785: #ifdef DIAGNOSTIC
                   1786:        struct pool_item *pi;
                   1787: #endif
                   1788:
                   1789:        LIST_FOREACH(ph, pl, ph_pagelist) {
                   1790:                (*pr)("\t\tpage %p, nmissing %d, time %lu,%lu\n",
                   1791:                    ph->ph_page, ph->ph_nmissing,
                   1792:                    (u_long)ph->ph_time.tv_sec,
                   1793:                    (u_long)ph->ph_time.tv_usec);
                   1794: #ifdef DIAGNOSTIC
1.97      yamt     1795:                if (!(pp->pr_roflags & PR_NOTOUCH)) {
1.102     chs      1796:                        LIST_FOREACH(pi, &ph->ph_itemlist, pi_list) {
1.97      yamt     1797:                                if (pi->pi_magic != PI_MAGIC) {
                   1798:                                        (*pr)("\t\t\titem %p, magic 0x%x\n",
                   1799:                                            pi, pi->pi_magic);
                   1800:                                }
1.88      chs      1801:                        }
                   1802:                }
                   1803: #endif
                   1804:        }
                   1805: }
                   1806:
                   1807: static void
1.42      thorpej  1808: pool_print1(struct pool *pp, const char *modif, void (*pr)(const char *, ...))
1.3       pk       1809: {
1.25      thorpej  1810:        struct pool_item_header *ph;
1.134     ad       1811:        pool_cache_t pc;
                   1812:        pcg_t *pcg;
                   1813:        pool_cache_cpu_t *cc;
                   1814:        uint64_t cpuhit, cpumiss;
1.44      thorpej  1815:        int i, print_log = 0, print_pagelist = 0, print_cache = 0;
1.25      thorpej  1816:        char c;
                   1817:
                   1818:        while ((c = *modif++) != '\0') {
                   1819:                if (c == 'l')
                   1820:                        print_log = 1;
                   1821:                if (c == 'p')
                   1822:                        print_pagelist = 1;
1.44      thorpej  1823:                if (c == 'c')
                   1824:                        print_cache = 1;
1.25      thorpej  1825:        }
                   1826:
1.134     ad       1827:        if ((pc = pp->pr_cache) != NULL) {
                   1828:                (*pr)("POOL CACHE");
                   1829:        } else {
                   1830:                (*pr)("POOL");
                   1831:        }
                   1832:
                   1833:        (*pr)(" %s: size %u, align %u, ioff %u, roflags 0x%08x\n",
1.25      thorpej  1834:            pp->pr_wchan, pp->pr_size, pp->pr_align, pp->pr_itemoffset,
                   1835:            pp->pr_roflags);
1.66      thorpej  1836:        (*pr)("\talloc %p\n", pp->pr_alloc);
1.25      thorpej  1837:        (*pr)("\tminitems %u, minpages %u, maxpages %u, npages %u\n",
                   1838:            pp->pr_minitems, pp->pr_minpages, pp->pr_maxpages, pp->pr_npages);
                   1839:        (*pr)("\titemsperpage %u, nitems %u, nout %u, hardlimit %u\n",
                   1840:            pp->pr_itemsperpage, pp->pr_nitems, pp->pr_nout, pp->pr_hardlimit);
                   1841:
1.134     ad       1842:        (*pr)("\tnget %lu, nfail %lu, nput %lu\n",
1.25      thorpej  1843:            pp->pr_nget, pp->pr_nfail, pp->pr_nput);
                   1844:        (*pr)("\tnpagealloc %lu, npagefree %lu, hiwat %u, nidle %lu\n",
                   1845:            pp->pr_npagealloc, pp->pr_npagefree, pp->pr_hiwat, pp->pr_nidle);
                   1846:
                   1847:        if (print_pagelist == 0)
                   1848:                goto skip_pagelist;
                   1849:
1.88      chs      1850:        if ((ph = LIST_FIRST(&pp->pr_emptypages)) != NULL)
                   1851:                (*pr)("\n\tempty page list:\n");
1.97      yamt     1852:        pool_print_pagelist(pp, &pp->pr_emptypages, pr);
1.88      chs      1853:        if ((ph = LIST_FIRST(&pp->pr_fullpages)) != NULL)
                   1854:                (*pr)("\n\tfull page list:\n");
1.97      yamt     1855:        pool_print_pagelist(pp, &pp->pr_fullpages, pr);
1.88      chs      1856:        if ((ph = LIST_FIRST(&pp->pr_partpages)) != NULL)
                   1857:                (*pr)("\n\tpartial-page list:\n");
1.97      yamt     1858:        pool_print_pagelist(pp, &pp->pr_partpages, pr);
1.88      chs      1859:
1.25      thorpej  1860:        if (pp->pr_curpage == NULL)
                   1861:                (*pr)("\tno current page\n");
                   1862:        else
                   1863:                (*pr)("\tcurpage %p\n", pp->pr_curpage->ph_page);
                   1864:
                   1865:  skip_pagelist:
                   1866:        if (print_log == 0)
                   1867:                goto skip_log;
                   1868:
                   1869:        (*pr)("\n");
                   1870:        if ((pp->pr_roflags & PR_LOGGING) == 0)
                   1871:                (*pr)("\tno log\n");
1.122     christos 1872:        else {
1.25      thorpej  1873:                pr_printlog(pp, NULL, pr);
1.122     christos 1874:        }
1.3       pk       1875:
1.25      thorpej  1876:  skip_log:
1.44      thorpej  1877:
1.102     chs      1878: #define PR_GROUPLIST(pcg)                                              \
                   1879:        (*pr)("\t\tgroup %p: avail %d\n", pcg, pcg->pcg_avail);         \
                   1880:        for (i = 0; i < PCG_NOBJECTS; i++) {                            \
                   1881:                if (pcg->pcg_objects[i].pcgo_pa !=                      \
                   1882:                    POOL_PADDR_INVALID) {                               \
                   1883:                        (*pr)("\t\t\t%p, 0x%llx\n",                     \
                   1884:                            pcg->pcg_objects[i].pcgo_va,                \
                   1885:                            (unsigned long long)                        \
                   1886:                            pcg->pcg_objects[i].pcgo_pa);               \
                   1887:                } else {                                                \
                   1888:                        (*pr)("\t\t\t%p\n",                             \
                   1889:                            pcg->pcg_objects[i].pcgo_va);               \
                   1890:                }                                                       \
                   1891:        }
                   1892:
1.134     ad       1893:        if (pc != NULL) {
                   1894:                cpuhit = 0;
                   1895:                cpumiss = 0;
                   1896:                for (i = 0; i < MAXCPUS; i++) {
                   1897:                        if ((cc = pc->pc_cpus[i]) == NULL)
                   1898:                                continue;
                   1899:                        cpuhit += cc->cc_hits;
                   1900:                        cpumiss += cc->cc_misses;
                   1901:                }
                   1902:                (*pr)("\tcpu layer hits %llu misses %llu\n", cpuhit, cpumiss);
                   1903:                (*pr)("\tcache layer hits %llu misses %llu\n",
                   1904:                    pc->pc_hits, pc->pc_misses);
                   1905:                (*pr)("\tcache layer entry uncontended %llu contended %llu\n",
                   1906:                    pc->pc_hits + pc->pc_misses - pc->pc_contended,
                   1907:                    pc->pc_contended);
                   1908:                (*pr)("\tcache layer empty groups %u full groups %u\n",
                   1909:                    pc->pc_nempty, pc->pc_nfull);
                   1910:                if (print_cache) {
                   1911:                        (*pr)("\tfull cache groups:\n");
                   1912:                        for (pcg = pc->pc_fullgroups; pcg != NULL;
                   1913:                            pcg = pcg->pcg_next) {
                   1914:                                PR_GROUPLIST(pcg);
                   1915:                        }
                   1916:                        (*pr)("\tempty cache groups:\n");
                   1917:                        for (pcg = pc->pc_emptygroups; pcg != NULL;
                   1918:                            pcg = pcg->pcg_next) {
                   1919:                                PR_GROUPLIST(pcg);
                   1920:                        }
1.103     chs      1921:                }
1.44      thorpej  1922:        }
1.102     chs      1923: #undef PR_GROUPLIST
1.44      thorpej  1924:
1.88      chs      1925:        pr_enter_check(pp, pr);
                   1926: }
                   1927:
                   1928: static int
                   1929: pool_chk_page(struct pool *pp, const char *label, struct pool_item_header *ph)
                   1930: {
                   1931:        struct pool_item *pi;
1.128     christos 1932:        void *page;
1.88      chs      1933:        int n;
                   1934:
1.121     yamt     1935:        if ((pp->pr_roflags & PR_NOALIGN) == 0) {
1.128     christos 1936:                page = (void *)((uintptr_t)ph & pp->pr_alloc->pa_pagemask);
1.121     yamt     1937:                if (page != ph->ph_page &&
                   1938:                    (pp->pr_roflags & PR_PHINPAGE) != 0) {
                   1939:                        if (label != NULL)
                   1940:                                printf("%s: ", label);
                   1941:                        printf("pool(%p:%s): page inconsistency: page %p;"
                   1942:                               " at page head addr %p (p %p)\n", pp,
                   1943:                                pp->pr_wchan, ph->ph_page,
                   1944:                                ph, page);
                   1945:                        return 1;
                   1946:                }
1.88      chs      1947:        }
1.3       pk       1948:
1.97      yamt     1949:        if ((pp->pr_roflags & PR_NOTOUCH) != 0)
                   1950:                return 0;
                   1951:
1.102     chs      1952:        for (pi = LIST_FIRST(&ph->ph_itemlist), n = 0;
1.88      chs      1953:             pi != NULL;
1.102     chs      1954:             pi = LIST_NEXT(pi,pi_list), n++) {
1.88      chs      1955:
                   1956: #ifdef DIAGNOSTIC
                   1957:                if (pi->pi_magic != PI_MAGIC) {
                   1958:                        if (label != NULL)
                   1959:                                printf("%s: ", label);
                   1960:                        printf("pool(%s): free list modified: magic=%x;"
1.121     yamt     1961:                               " page %p; item ordinal %d; addr %p\n",
1.88      chs      1962:                                pp->pr_wchan, pi->pi_magic, ph->ph_page,
1.121     yamt     1963:                                n, pi);
1.88      chs      1964:                        panic("pool");
                   1965:                }
                   1966: #endif
1.121     yamt     1967:                if ((pp->pr_roflags & PR_NOALIGN) != 0) {
                   1968:                        continue;
                   1969:                }
1.128     christos 1970:                page = (void *)((uintptr_t)pi & pp->pr_alloc->pa_pagemask);
1.88      chs      1971:                if (page == ph->ph_page)
                   1972:                        continue;
                   1973:
                   1974:                if (label != NULL)
                   1975:                        printf("%s: ", label);
                   1976:                printf("pool(%p:%s): page inconsistency: page %p;"
                   1977:                       " item ordinal %d; addr %p (p %p)\n", pp,
                   1978:                        pp->pr_wchan, ph->ph_page,
                   1979:                        n, pi, page);
                   1980:                return 1;
                   1981:        }
                   1982:        return 0;
1.3       pk       1983: }
                   1984:
1.88      chs      1985:
1.3       pk       1986: int
1.42      thorpej  1987: pool_chk(struct pool *pp, const char *label)
1.3       pk       1988: {
                   1989:        struct pool_item_header *ph;
                   1990:        int r = 0;
                   1991:
1.134     ad       1992:        mutex_enter(&pp->pr_lock);
1.88      chs      1993:        LIST_FOREACH(ph, &pp->pr_emptypages, ph_pagelist) {
                   1994:                r = pool_chk_page(pp, label, ph);
                   1995:                if (r) {
                   1996:                        goto out;
                   1997:                }
                   1998:        }
                   1999:        LIST_FOREACH(ph, &pp->pr_fullpages, ph_pagelist) {
                   2000:                r = pool_chk_page(pp, label, ph);
                   2001:                if (r) {
1.3       pk       2002:                        goto out;
                   2003:                }
1.88      chs      2004:        }
                   2005:        LIST_FOREACH(ph, &pp->pr_partpages, ph_pagelist) {
                   2006:                r = pool_chk_page(pp, label, ph);
                   2007:                if (r) {
1.3       pk       2008:                        goto out;
                   2009:                }
                   2010:        }
1.88      chs      2011:
1.3       pk       2012: out:
1.134     ad       2013:        mutex_exit(&pp->pr_lock);
1.3       pk       2014:        return (r);
1.43      thorpej  2015: }
                   2016:
                   2017: /*
                   2018:  * pool_cache_init:
                   2019:  *
                   2020:  *     Initialize a pool cache.
1.134     ad       2021:  */
                   2022: pool_cache_t
                   2023: pool_cache_init(size_t size, u_int align, u_int align_offset, u_int flags,
                   2024:     const char *wchan, struct pool_allocator *palloc, int ipl,
                   2025:     int (*ctor)(void *, void *, int), void (*dtor)(void *, void *), void *arg)
                   2026: {
                   2027:        pool_cache_t pc;
                   2028:
                   2029:        pc = pool_get(&cache_pool, PR_WAITOK);
                   2030:        if (pc == NULL)
                   2031:                return NULL;
                   2032:
                   2033:        pool_cache_bootstrap(pc, size, align, align_offset, flags, wchan,
                   2034:           palloc, ipl, ctor, dtor, arg);
                   2035:
                   2036:        return pc;
                   2037: }
                   2038:
                   2039: /*
                   2040:  * pool_cache_bootstrap:
1.43      thorpej  2041:  *
1.134     ad       2042:  *     Kernel-private version of pool_cache_init().  The caller
                   2043:  *     provides initial storage.
1.43      thorpej  2044:  */
                   2045: void
1.134     ad       2046: pool_cache_bootstrap(pool_cache_t pc, size_t size, u_int align,
                   2047:     u_int align_offset, u_int flags, const char *wchan,
                   2048:     struct pool_allocator *palloc, int ipl,
                   2049:     int (*ctor)(void *, void *, int), void (*dtor)(void *, void *),
1.43      thorpej  2050:     void *arg)
                   2051: {
1.134     ad       2052:        CPU_INFO_ITERATOR cii;
                   2053:        struct cpu_info *ci;
                   2054:        struct pool *pp;
                   2055:
                   2056:        pp = &pc->pc_pool;
                   2057:        if (palloc == NULL && ipl == IPL_NONE)
                   2058:                palloc = &pool_allocator_nointr;
                   2059:        pool_init(pp, size, align, align_offset, flags, wchan, palloc, ipl);
1.43      thorpej  2060:
1.138     ad       2061:        /*
                   2062:         * XXXAD hack to prevent IP input processing from blocking.
                   2063:         */
                   2064:        if (ipl == IPL_SOFTNET) {
                   2065:                mutex_init(&pc->pc_lock, MUTEX_DEFAULT, IPL_VM);
                   2066:        } else {
                   2067:                mutex_init(&pc->pc_lock, MUTEX_DEFAULT, ipl);
                   2068:        }
1.43      thorpej  2069:
1.134     ad       2070:        if (ctor == NULL) {
                   2071:                ctor = (int (*)(void *, void *, int))nullop;
                   2072:        }
                   2073:        if (dtor == NULL) {
                   2074:                dtor = (void (*)(void *, void *))nullop;
                   2075:        }
1.43      thorpej  2076:
1.134     ad       2077:        pc->pc_emptygroups = NULL;
                   2078:        pc->pc_fullgroups = NULL;
                   2079:        pc->pc_partgroups = NULL;
1.43      thorpej  2080:        pc->pc_ctor = ctor;
                   2081:        pc->pc_dtor = dtor;
                   2082:        pc->pc_arg  = arg;
1.134     ad       2083:        pc->pc_hits  = 0;
1.48      thorpej  2084:        pc->pc_misses = 0;
1.134     ad       2085:        pc->pc_nempty = 0;
                   2086:        pc->pc_npart = 0;
                   2087:        pc->pc_nfull = 0;
                   2088:        pc->pc_contended = 0;
                   2089:        pc->pc_refcnt = 0;
1.136     yamt     2090:        pc->pc_freecheck = NULL;
1.134     ad       2091:
                   2092:        /* Allocate per-CPU caches. */
                   2093:        memset(pc->pc_cpus, 0, sizeof(pc->pc_cpus));
                   2094:        pc->pc_ncpu = 0;
1.139     ad       2095:        if (ncpu < 2) {
1.137     ad       2096:                /* XXX For sparc: boot CPU is not attached yet. */
                   2097:                pool_cache_cpu_init1(curcpu(), pc);
                   2098:        } else {
                   2099:                for (CPU_INFO_FOREACH(cii, ci)) {
                   2100:                        pool_cache_cpu_init1(ci, pc);
                   2101:                }
1.134     ad       2102:        }
                   2103:
                   2104:        if (__predict_true(!cold)) {
                   2105:                mutex_enter(&pp->pr_lock);
                   2106:                pp->pr_cache = pc;
                   2107:                mutex_exit(&pp->pr_lock);
                   2108:                mutex_enter(&pool_head_lock);
                   2109:                LIST_INSERT_HEAD(&pool_cache_head, pc, pc_cachelist);
                   2110:                mutex_exit(&pool_head_lock);
                   2111:        } else {
                   2112:                pp->pr_cache = pc;
                   2113:                LIST_INSERT_HEAD(&pool_cache_head, pc, pc_cachelist);
                   2114:        }
1.43      thorpej  2115: }
                   2116:
                   2117: /*
                   2118:  * pool_cache_destroy:
                   2119:  *
                   2120:  *     Destroy a pool cache.
                   2121:  */
                   2122: void
1.134     ad       2123: pool_cache_destroy(pool_cache_t pc)
1.43      thorpej  2124: {
1.134     ad       2125:        struct pool *pp = &pc->pc_pool;
                   2126:        pool_cache_cpu_t *cc;
                   2127:        pcg_t *pcg;
                   2128:        int i;
                   2129:
                   2130:        /* Remove it from the global list. */
                   2131:        mutex_enter(&pool_head_lock);
                   2132:        while (pc->pc_refcnt != 0)
                   2133:                cv_wait(&pool_busy, &pool_head_lock);
                   2134:        LIST_REMOVE(pc, pc_cachelist);
                   2135:        mutex_exit(&pool_head_lock);
1.43      thorpej  2136:
                   2137:        /* First, invalidate the entire cache. */
                   2138:        pool_cache_invalidate(pc);
                   2139:
1.134     ad       2140:        /* Disassociate it from the pool. */
                   2141:        mutex_enter(&pp->pr_lock);
                   2142:        pp->pr_cache = NULL;
                   2143:        mutex_exit(&pp->pr_lock);
                   2144:
                   2145:        /* Destroy per-CPU data */
                   2146:        for (i = 0; i < MAXCPUS; i++) {
                   2147:                if ((cc = pc->pc_cpus[i]) == NULL)
                   2148:                        continue;
                   2149:                if ((pcg = cc->cc_current) != NULL) {
                   2150:                        pcg->pcg_next = NULL;
                   2151:                        pool_cache_invalidate_groups(pc, pcg);
                   2152:                }
                   2153:                if ((pcg = cc->cc_previous) != NULL) {
                   2154:                        pcg->pcg_next = NULL;
                   2155:                        pool_cache_invalidate_groups(pc, pcg);
                   2156:                }
                   2157:                if (cc != &pc->pc_cpu0)
                   2158:                        pool_put(&cache_cpu_pool, cc);
                   2159:        }
                   2160:
                   2161:        /* Finally, destroy it. */
                   2162:        mutex_destroy(&pc->pc_lock);
                   2163:        pool_destroy(pp);
                   2164:        pool_put(&cache_pool, pc);
                   2165: }
                   2166:
                   2167: /*
                   2168:  * pool_cache_cpu_init1:
                   2169:  *
                   2170:  *     Called for each pool_cache whenever a new CPU is attached.
                   2171:  */
                   2172: static void
                   2173: pool_cache_cpu_init1(struct cpu_info *ci, pool_cache_t pc)
                   2174: {
                   2175:        pool_cache_cpu_t *cc;
1.137     ad       2176:        int index;
1.134     ad       2177:
1.137     ad       2178:        index = ci->ci_index;
                   2179:
                   2180:        KASSERT(index < MAXCPUS);
1.134     ad       2181:        KASSERT(((uintptr_t)pc->pc_cpus & (CACHE_LINE_SIZE - 1)) == 0);
                   2182:
1.137     ad       2183:        if ((cc = pc->pc_cpus[index]) != NULL) {
                   2184:                KASSERT(cc->cc_cpuindex == index);
1.134     ad       2185:                return;
                   2186:        }
                   2187:
                   2188:        /*
                   2189:         * The first CPU is 'free'.  This needs to be the case for
                   2190:         * bootstrap - we may not be able to allocate yet.
                   2191:         */
                   2192:        if (pc->pc_ncpu == 0) {
                   2193:                cc = &pc->pc_cpu0;
                   2194:                pc->pc_ncpu = 1;
                   2195:        } else {
                   2196:                mutex_enter(&pc->pc_lock);
                   2197:                pc->pc_ncpu++;
                   2198:                mutex_exit(&pc->pc_lock);
                   2199:                cc = pool_get(&cache_cpu_pool, PR_WAITOK);
                   2200:        }
                   2201:
                   2202:        cc->cc_ipl = pc->pc_pool.pr_ipl;
                   2203:        cc->cc_iplcookie = makeiplcookie(cc->cc_ipl);
                   2204:        cc->cc_cache = pc;
1.137     ad       2205:        cc->cc_cpuindex = index;
1.134     ad       2206:        cc->cc_hits = 0;
                   2207:        cc->cc_misses = 0;
                   2208:        cc->cc_current = NULL;
                   2209:        cc->cc_previous = NULL;
                   2210:
1.137     ad       2211:        pc->pc_cpus[index] = cc;
1.43      thorpej  2212: }
                   2213:
1.134     ad       2214: /*
                   2215:  * pool_cache_cpu_init:
                   2216:  *
                   2217:  *     Called whenever a new CPU is attached.
                   2218:  */
                   2219: void
                   2220: pool_cache_cpu_init(struct cpu_info *ci)
1.43      thorpej  2221: {
1.134     ad       2222:        pool_cache_t pc;
                   2223:
                   2224:        mutex_enter(&pool_head_lock);
                   2225:        LIST_FOREACH(pc, &pool_cache_head, pc_cachelist) {
                   2226:                pc->pc_refcnt++;
                   2227:                mutex_exit(&pool_head_lock);
1.43      thorpej  2228:
1.134     ad       2229:                pool_cache_cpu_init1(ci, pc);
1.43      thorpej  2230:
1.134     ad       2231:                mutex_enter(&pool_head_lock);
                   2232:                pc->pc_refcnt--;
                   2233:                cv_broadcast(&pool_busy);
                   2234:        }
                   2235:        mutex_exit(&pool_head_lock);
1.43      thorpej  2236: }
                   2237:
1.134     ad       2238: /*
                   2239:  * pool_cache_reclaim:
                   2240:  *
                   2241:  *     Reclaim memory from a pool cache.
                   2242:  */
                   2243: bool
                   2244: pool_cache_reclaim(pool_cache_t pc)
1.43      thorpej  2245: {
                   2246:
1.134     ad       2247:        return pool_reclaim(&pc->pc_pool);
                   2248: }
1.43      thorpej  2249:
1.136     yamt     2250: static void
                   2251: pool_cache_destruct_object1(pool_cache_t pc, void *object)
                   2252: {
                   2253:
                   2254:        (*pc->pc_dtor)(pc->pc_arg, object);
                   2255:        pool_put(&pc->pc_pool, object);
                   2256: }
                   2257:
1.134     ad       2258: /*
                   2259:  * pool_cache_destruct_object:
                   2260:  *
                   2261:  *     Force destruction of an object and its release back into
                   2262:  *     the pool.
                   2263:  */
                   2264: void
                   2265: pool_cache_destruct_object(pool_cache_t pc, void *object)
                   2266: {
                   2267:
1.136     yamt     2268:        FREECHECK_IN(&pc->pc_freecheck, object);
                   2269:
                   2270:        pool_cache_destruct_object1(pc, object);
1.43      thorpej  2271: }
                   2272:
1.134     ad       2273: /*
                   2274:  * pool_cache_invalidate_groups:
                   2275:  *
                   2276:  *     Invalidate a chain of groups and destruct all objects.
                   2277:  */
1.102     chs      2278: static void
1.134     ad       2279: pool_cache_invalidate_groups(pool_cache_t pc, pcg_t *pcg)
1.102     chs      2280: {
1.134     ad       2281:        void *object;
                   2282:        pcg_t *next;
                   2283:        int i;
                   2284:
                   2285:        for (; pcg != NULL; pcg = next) {
                   2286:                next = pcg->pcg_next;
                   2287:
                   2288:                for (i = 0; i < pcg->pcg_avail; i++) {
                   2289:                        object = pcg->pcg_objects[i].pcgo_va;
1.136     yamt     2290:                        pool_cache_destruct_object1(pc, object);
1.134     ad       2291:                }
1.102     chs      2292:
                   2293:                pool_put(&pcgpool, pcg);
                   2294:        }
                   2295: }
                   2296:
1.43      thorpej  2297: /*
1.134     ad       2298:  * pool_cache_invalidate:
1.43      thorpej  2299:  *
1.134     ad       2300:  *     Invalidate a pool cache (destruct and release all of the
                   2301:  *     cached objects).  Does not reclaim objects from the pool.
1.43      thorpej  2302:  */
1.134     ad       2303: void
                   2304: pool_cache_invalidate(pool_cache_t pc)
                   2305: {
                   2306:        pcg_t *full, *empty, *part;
                   2307:
                   2308:        mutex_enter(&pc->pc_lock);
                   2309:        full = pc->pc_fullgroups;
                   2310:        empty = pc->pc_emptygroups;
                   2311:        part = pc->pc_partgroups;
                   2312:        pc->pc_fullgroups = NULL;
                   2313:        pc->pc_emptygroups = NULL;
                   2314:        pc->pc_partgroups = NULL;
                   2315:        pc->pc_nfull = 0;
                   2316:        pc->pc_nempty = 0;
                   2317:        pc->pc_npart = 0;
                   2318:        mutex_exit(&pc->pc_lock);
                   2319:
                   2320:        pool_cache_invalidate_groups(pc, full);
                   2321:        pool_cache_invalidate_groups(pc, empty);
                   2322:        pool_cache_invalidate_groups(pc, part);
                   2323: }
                   2324:
                   2325: void
                   2326: pool_cache_set_drain_hook(pool_cache_t pc, void (*fn)(void *, int), void *arg)
                   2327: {
                   2328:
                   2329:        pool_set_drain_hook(&pc->pc_pool, fn, arg);
                   2330: }
                   2331:
                   2332: void
                   2333: pool_cache_setlowat(pool_cache_t pc, int n)
                   2334: {
                   2335:
                   2336:        pool_setlowat(&pc->pc_pool, n);
                   2337: }
                   2338:
                   2339: void
                   2340: pool_cache_sethiwat(pool_cache_t pc, int n)
                   2341: {
                   2342:
                   2343:        pool_sethiwat(&pc->pc_pool, n);
                   2344: }
                   2345:
                   2346: void
                   2347: pool_cache_sethardlimit(pool_cache_t pc, int n, const char *warnmess, int ratecap)
                   2348: {
                   2349:
                   2350:        pool_sethardlimit(&pc->pc_pool, n, warnmess, ratecap);
                   2351: }
                   2352:
                   2353: static inline pool_cache_cpu_t *
                   2354: pool_cache_cpu_enter(pool_cache_t pc, int *s)
                   2355: {
                   2356:        pool_cache_cpu_t *cc;
                   2357:
                   2358:        /*
                   2359:         * Prevent other users of the cache from accessing our
                   2360:         * CPU-local data.  To avoid touching shared state, we
                   2361:         * pull the neccessary information from CPU local data.
                   2362:         */
1.137     ad       2363:        crit_enter();
                   2364:        cc = pc->pc_cpus[curcpu()->ci_index];
1.134     ad       2365:        KASSERT(cc->cc_cache == pc);
1.137     ad       2366:        if (cc->cc_ipl != IPL_NONE) {
1.134     ad       2367:                *s = splraiseipl(cc->cc_iplcookie);
                   2368:        }
                   2369:        KASSERT(((uintptr_t)cc & (CACHE_LINE_SIZE - 1)) == 0);
                   2370:
                   2371:        return cc;
                   2372: }
                   2373:
                   2374: static inline void
                   2375: pool_cache_cpu_exit(pool_cache_cpu_t *cc, int *s)
                   2376: {
                   2377:
                   2378:        /* No longer need exclusive access to the per-CPU data. */
1.137     ad       2379:        if (cc->cc_ipl != IPL_NONE) {
1.134     ad       2380:                splx(*s);
                   2381:        }
1.137     ad       2382:        crit_exit();
1.134     ad       2383: }
                   2384:
                   2385: #if __GNUC_PREREQ__(3, 0)
                   2386: __attribute ((noinline))
                   2387: #endif
                   2388: pool_cache_cpu_t *
                   2389: pool_cache_get_slow(pool_cache_cpu_t *cc, int *s, void **objectp,
                   2390:                    paddr_t *pap, int flags)
1.43      thorpej  2391: {
1.134     ad       2392:        pcg_t *pcg, *cur;
                   2393:        uint64_t ncsw;
                   2394:        pool_cache_t pc;
1.43      thorpej  2395:        void *object;
1.58      thorpej  2396:
1.134     ad       2397:        pc = cc->cc_cache;
                   2398:        cc->cc_misses++;
1.43      thorpej  2399:
1.134     ad       2400:        /*
                   2401:         * Nothing was available locally.  Try and grab a group
                   2402:         * from the cache.
                   2403:         */
                   2404:        if (!mutex_tryenter(&pc->pc_lock)) {
                   2405:                ncsw = curlwp->l_ncsw;
                   2406:                mutex_enter(&pc->pc_lock);
                   2407:                pc->pc_contended++;
1.43      thorpej  2408:
1.134     ad       2409:                /*
                   2410:                 * If we context switched while locking, then
                   2411:                 * our view of the per-CPU data is invalid:
                   2412:                 * retry.
                   2413:                 */
                   2414:                if (curlwp->l_ncsw != ncsw) {
                   2415:                        mutex_exit(&pc->pc_lock);
                   2416:                        pool_cache_cpu_exit(cc, s);
                   2417:                        return pool_cache_cpu_enter(pc, s);
1.43      thorpej  2418:                }
1.102     chs      2419:        }
1.43      thorpej  2420:
1.134     ad       2421:        if ((pcg = pc->pc_fullgroups) != NULL) {
1.43      thorpej  2422:                /*
1.134     ad       2423:                 * If there's a full group, release our empty
                   2424:                 * group back to the cache.  Install the full
                   2425:                 * group as cc_current and return.
1.43      thorpej  2426:                 */
1.134     ad       2427:                if ((cur = cc->cc_current) != NULL) {
                   2428:                        KASSERT(cur->pcg_avail == 0);
                   2429:                        cur->pcg_next = pc->pc_emptygroups;
                   2430:                        pc->pc_emptygroups = cur;
                   2431:                        pc->pc_nempty++;
1.87      thorpej  2432:                }
1.134     ad       2433:                KASSERT(pcg->pcg_avail == PCG_NOBJECTS);
                   2434:                cc->cc_current = pcg;
                   2435:                pc->pc_fullgroups = pcg->pcg_next;
                   2436:                pc->pc_hits++;
                   2437:                pc->pc_nfull--;
                   2438:                mutex_exit(&pc->pc_lock);
                   2439:                return cc;
                   2440:        }
                   2441:
                   2442:        /*
                   2443:         * Nothing available locally or in cache.  Take the slow
                   2444:         * path: fetch a new object from the pool and construct
                   2445:         * it.
                   2446:         */
                   2447:        pc->pc_misses++;
                   2448:        mutex_exit(&pc->pc_lock);
                   2449:        pool_cache_cpu_exit(cc, s);
                   2450:
                   2451:        object = pool_get(&pc->pc_pool, flags);
                   2452:        *objectp = object;
                   2453:        if (object == NULL)
                   2454:                return NULL;
1.125     ad       2455:
1.134     ad       2456:        if ((*pc->pc_ctor)(pc->pc_arg, object, flags) != 0) {
                   2457:                pool_put(&pc->pc_pool, object);
                   2458:                *objectp = NULL;
                   2459:                return NULL;
1.43      thorpej  2460:        }
                   2461:
1.134     ad       2462:        KASSERT((((vaddr_t)object + pc->pc_pool.pr_itemoffset) &
                   2463:            (pc->pc_pool.pr_align - 1)) == 0);
1.43      thorpej  2464:
1.134     ad       2465:        if (pap != NULL) {
                   2466: #ifdef POOL_VTOPHYS
                   2467:                *pap = POOL_VTOPHYS(object);
                   2468: #else
                   2469:                *pap = POOL_PADDR_INVALID;
                   2470: #endif
1.102     chs      2471:        }
1.43      thorpej  2472:
1.125     ad       2473:        FREECHECK_OUT(&pc->pc_freecheck, object);
1.134     ad       2474:        return NULL;
1.43      thorpej  2475: }
                   2476:
                   2477: /*
1.134     ad       2478:  * pool_cache_get{,_paddr}:
1.43      thorpej  2479:  *
1.134     ad       2480:  *     Get an object from a pool cache (optionally returning
                   2481:  *     the physical address of the object).
1.43      thorpej  2482:  */
1.134     ad       2483: void *
                   2484: pool_cache_get_paddr(pool_cache_t pc, int flags, paddr_t *pap)
1.43      thorpej  2485: {
1.134     ad       2486:        pool_cache_cpu_t *cc;
                   2487:        pcg_t *pcg;
                   2488:        void *object;
1.60      thorpej  2489:        int s;
1.43      thorpej  2490:
1.134     ad       2491: #ifdef LOCKDEBUG
                   2492:        if (flags & PR_WAITOK)
                   2493:                ASSERT_SLEEPABLE(NULL, "pool_cache_get(PR_WAITOK)");
                   2494: #endif
1.125     ad       2495:
1.134     ad       2496:        cc = pool_cache_cpu_enter(pc, &s);
                   2497:        do {
                   2498:                /* Try and allocate an object from the current group. */
                   2499:                pcg = cc->cc_current;
                   2500:                if (pcg != NULL && pcg->pcg_avail > 0) {
                   2501:                        object = pcg->pcg_objects[--pcg->pcg_avail].pcgo_va;
                   2502:                        if (pap != NULL)
                   2503:                                *pap = pcg->pcg_objects[pcg->pcg_avail].pcgo_pa;
                   2504:                        pcg->pcg_objects[pcg->pcg_avail].pcgo_va = NULL;
                   2505:                        KASSERT(pcg->pcg_avail <= PCG_NOBJECTS);
                   2506:                        KASSERT(object != NULL);
                   2507:                        cc->cc_hits++;
                   2508:                        pool_cache_cpu_exit(cc, &s);
                   2509:                        FREECHECK_OUT(&pc->pc_freecheck, object);
                   2510:                        return object;
1.43      thorpej  2511:                }
                   2512:
                   2513:                /*
1.134     ad       2514:                 * That failed.  If the previous group isn't empty, swap
                   2515:                 * it with the current group and allocate from there.
1.43      thorpej  2516:                 */
1.134     ad       2517:                pcg = cc->cc_previous;
                   2518:                if (pcg != NULL && pcg->pcg_avail > 0) {
                   2519:                        cc->cc_previous = cc->cc_current;
                   2520:                        cc->cc_current = pcg;
                   2521:                        continue;
1.43      thorpej  2522:                }
                   2523:
1.134     ad       2524:                /*
                   2525:                 * Can't allocate from either group: try the slow path.
                   2526:                 * If get_slow() allocated an object for us, or if
                   2527:                 * no more objects are available, it will return NULL.
                   2528:                 * Otherwise, we need to retry.
                   2529:                 */
                   2530:                cc = pool_cache_get_slow(cc, &s, &object, pap, flags);
                   2531:        } while (cc != NULL);
1.43      thorpej  2532:
1.134     ad       2533:        return object;
1.51      thorpej  2534: }
                   2535:
1.134     ad       2536: #if __GNUC_PREREQ__(3, 0)
                   2537: __attribute ((noinline))
                   2538: #endif
                   2539: pool_cache_cpu_t *
                   2540: pool_cache_put_slow(pool_cache_cpu_t *cc, int *s, void *object, paddr_t pa)
1.51      thorpej  2541: {
1.134     ad       2542:        pcg_t *pcg, *cur;
                   2543:        uint64_t ncsw;
                   2544:        pool_cache_t pc;
1.51      thorpej  2545:
1.134     ad       2546:        pc = cc->cc_cache;
                   2547:        cc->cc_misses++;
1.43      thorpej  2548:
1.134     ad       2549:        /*
                   2550:         * No free slots locally.  Try to grab an empty, unused
                   2551:         * group from the cache.
                   2552:         */
                   2553:        if (!mutex_tryenter(&pc->pc_lock)) {
                   2554:                ncsw = curlwp->l_ncsw;
                   2555:                mutex_enter(&pc->pc_lock);
                   2556:                pc->pc_contended++;
1.102     chs      2557:
1.134     ad       2558:                /*
                   2559:                 * If we context switched while locking, then
                   2560:                 * our view of the per-CPU data is invalid:
                   2561:                 * retry.
                   2562:                 */
                   2563:                if (curlwp->l_ncsw != ncsw) {
                   2564:                        mutex_exit(&pc->pc_lock);
                   2565:                        pool_cache_cpu_exit(cc, s);
                   2566:                        return pool_cache_cpu_enter(pc, s);
                   2567:                }
                   2568:        }
1.130     ad       2569:
1.134     ad       2570:        if ((pcg = pc->pc_emptygroups) != NULL) {
                   2571:                /*
                   2572:                 * If there's a empty group, release our full
                   2573:                 * group back to the cache.  Install the empty
                   2574:                 * group as cc_current and return.
                   2575:                 */
                   2576:                if ((cur = cc->cc_current) != NULL) {
                   2577:                        KASSERT(cur->pcg_avail == PCG_NOBJECTS);
                   2578:                        cur->pcg_next = pc->pc_fullgroups;
                   2579:                        pc->pc_fullgroups = cur;
                   2580:                        pc->pc_nfull++;
1.102     chs      2581:                }
1.134     ad       2582:                KASSERT(pcg->pcg_avail == 0);
                   2583:                cc->cc_current = pcg;
                   2584:                pc->pc_emptygroups = pcg->pcg_next;
                   2585:                pc->pc_hits++;
                   2586:                pc->pc_nempty--;
                   2587:                mutex_exit(&pc->pc_lock);
                   2588:                return cc;
1.102     chs      2589:        }
1.105     christos 2590:
1.134     ad       2591:        /*
                   2592:         * Nothing available locally or in cache.  Take the
                   2593:         * slow path and try to allocate a new group that we
                   2594:         * can release to.
                   2595:         */
                   2596:        pc->pc_misses++;
                   2597:        mutex_exit(&pc->pc_lock);
                   2598:        pool_cache_cpu_exit(cc, s);
1.105     christos 2599:
1.134     ad       2600:        /*
                   2601:         * If we can't allocate a new group, just throw the
                   2602:         * object away.
                   2603:         */
                   2604:        pcg = pool_get(&pcgpool, PR_NOWAIT);
                   2605:        if (pcg == NULL) {
                   2606:                pool_cache_destruct_object(pc, object);
                   2607:                return NULL;
                   2608:        }
                   2609: #ifdef DIAGNOSTIC
                   2610:        memset(pcg, 0, sizeof(*pcg));
                   2611: #else
                   2612:        pcg->pcg_avail = 0;
                   2613: #endif
1.105     christos 2614:
1.134     ad       2615:        /*
                   2616:         * Add the empty group to the cache and try again.
                   2617:         */
                   2618:        mutex_enter(&pc->pc_lock);
                   2619:        pcg->pcg_next = pc->pc_emptygroups;
                   2620:        pc->pc_emptygroups = pcg;
                   2621:        pc->pc_nempty++;
                   2622:        mutex_exit(&pc->pc_lock);
1.103     chs      2623:
1.134     ad       2624:        return pool_cache_cpu_enter(pc, s);
                   2625: }
1.102     chs      2626:
1.43      thorpej  2627: /*
1.134     ad       2628:  * pool_cache_put{,_paddr}:
1.43      thorpej  2629:  *
1.134     ad       2630:  *     Put an object back to the pool cache (optionally caching the
                   2631:  *     physical address of the object).
1.43      thorpej  2632:  */
1.101     thorpej  2633: void
1.134     ad       2634: pool_cache_put_paddr(pool_cache_t pc, void *object, paddr_t pa)
1.43      thorpej  2635: {
1.134     ad       2636:        pool_cache_cpu_t *cc;
                   2637:        pcg_t *pcg;
                   2638:        int s;
1.101     thorpej  2639:
1.134     ad       2640:        FREECHECK_IN(&pc->pc_freecheck, object);
1.101     thorpej  2641:
1.134     ad       2642:        cc = pool_cache_cpu_enter(pc, &s);
                   2643:        do {
                   2644:                /* If the current group isn't full, release it there. */
                   2645:                pcg = cc->cc_current;
                   2646:                if (pcg != NULL && pcg->pcg_avail < PCG_NOBJECTS) {
                   2647:                        KASSERT(pcg->pcg_objects[pcg->pcg_avail].pcgo_va
                   2648:                            == NULL);
                   2649:                        pcg->pcg_objects[pcg->pcg_avail].pcgo_va = object;
                   2650:                        pcg->pcg_objects[pcg->pcg_avail].pcgo_pa = pa;
                   2651:                        pcg->pcg_avail++;
                   2652:                        cc->cc_hits++;
                   2653:                        pool_cache_cpu_exit(cc, &s);
                   2654:                        return;
                   2655:                }
1.43      thorpej  2656:
1.134     ad       2657:                /*
                   2658:                 * That failed.  If the previous group is empty, swap
                   2659:                 * it with the current group and try again.
                   2660:                 */
                   2661:                pcg = cc->cc_previous;
                   2662:                if (pcg != NULL && pcg->pcg_avail == 0) {
                   2663:                        cc->cc_previous = cc->cc_current;
                   2664:                        cc->cc_current = pcg;
                   2665:                        continue;
                   2666:                }
1.43      thorpej  2667:
1.134     ad       2668:                /*
                   2669:                 * Can't free to either group: try the slow path.
                   2670:                 * If put_slow() releases the object for us, it
                   2671:                 * will return NULL.  Otherwise we need to retry.
                   2672:                 */
                   2673:                cc = pool_cache_put_slow(cc, &s, object, pa);
                   2674:        } while (cc != NULL);
1.43      thorpej  2675: }
                   2676:
                   2677: /*
1.134     ad       2678:  * pool_cache_xcall:
1.43      thorpej  2679:  *
1.134     ad       2680:  *     Transfer objects from the per-CPU cache to the global cache.
                   2681:  *     Run within a cross-call thread.
1.43      thorpej  2682:  */
                   2683: static void
1.134     ad       2684: pool_cache_xcall(pool_cache_t pc)
1.43      thorpej  2685: {
1.134     ad       2686:        pool_cache_cpu_t *cc;
                   2687:        pcg_t *prev, *cur, **list;
                   2688:        int s = 0; /* XXXgcc */
                   2689:
                   2690:        cc = pool_cache_cpu_enter(pc, &s);
                   2691:        cur = cc->cc_current;
                   2692:        cc->cc_current = NULL;
                   2693:        prev = cc->cc_previous;
                   2694:        cc->cc_previous = NULL;
                   2695:        pool_cache_cpu_exit(cc, &s);
                   2696:
                   2697:        /*
                   2698:         * XXXSMP Go to splvm to prevent kernel_lock from being taken,
                   2699:         * because locks at IPL_SOFTXXX are still spinlocks.  Does not
                   2700:         * apply to IPL_SOFTBIO.  Cross-call threads do not take the
                   2701:         * kernel_lock.
1.101     thorpej  2702:         */
1.134     ad       2703:        s = splvm();
                   2704:        mutex_enter(&pc->pc_lock);
                   2705:        if (cur != NULL) {
                   2706:                if (cur->pcg_avail == PCG_NOBJECTS) {
                   2707:                        list = &pc->pc_fullgroups;
                   2708:                        pc->pc_nfull++;
                   2709:                } else if (cur->pcg_avail == 0) {
                   2710:                        list = &pc->pc_emptygroups;
                   2711:                        pc->pc_nempty++;
                   2712:                } else {
                   2713:                        list = &pc->pc_partgroups;
                   2714:                        pc->pc_npart++;
                   2715:                }
                   2716:                cur->pcg_next = *list;
                   2717:                *list = cur;
                   2718:        }
                   2719:        if (prev != NULL) {
                   2720:                if (prev->pcg_avail == PCG_NOBJECTS) {
                   2721:                        list = &pc->pc_fullgroups;
                   2722:                        pc->pc_nfull++;
                   2723:                } else if (prev->pcg_avail == 0) {
                   2724:                        list = &pc->pc_emptygroups;
                   2725:                        pc->pc_nempty++;
                   2726:                } else {
                   2727:                        list = &pc->pc_partgroups;
                   2728:                        pc->pc_npart++;
                   2729:                }
                   2730:                prev->pcg_next = *list;
                   2731:                *list = prev;
                   2732:        }
                   2733:        mutex_exit(&pc->pc_lock);
                   2734:        splx(s);
1.3       pk       2735: }
1.66      thorpej  2736:
                   2737: /*
                   2738:  * Pool backend allocators.
                   2739:  *
                   2740:  * Each pool has a backend allocator that handles allocation, deallocation,
                   2741:  * and any additional draining that might be needed.
                   2742:  *
                   2743:  * We provide two standard allocators:
                   2744:  *
                   2745:  *     pool_allocator_kmem - the default when no allocator is specified
                   2746:  *
                   2747:  *     pool_allocator_nointr - used for pools that will not be accessed
                   2748:  *     in interrupt context.
                   2749:  */
                   2750: void   *pool_page_alloc(struct pool *, int);
                   2751: void   pool_page_free(struct pool *, void *);
                   2752:
1.112     bjh21    2753: #ifdef POOL_SUBPAGE
                   2754: struct pool_allocator pool_allocator_kmem_fullpage = {
                   2755:        pool_page_alloc, pool_page_free, 0,
1.117     yamt     2756:        .pa_backingmapptr = &kmem_map,
1.112     bjh21    2757: };
                   2758: #else
1.66      thorpej  2759: struct pool_allocator pool_allocator_kmem = {
                   2760:        pool_page_alloc, pool_page_free, 0,
1.117     yamt     2761:        .pa_backingmapptr = &kmem_map,
1.66      thorpej  2762: };
1.112     bjh21    2763: #endif
1.66      thorpej  2764:
                   2765: void   *pool_page_alloc_nointr(struct pool *, int);
                   2766: void   pool_page_free_nointr(struct pool *, void *);
                   2767:
1.112     bjh21    2768: #ifdef POOL_SUBPAGE
                   2769: struct pool_allocator pool_allocator_nointr_fullpage = {
                   2770:        pool_page_alloc_nointr, pool_page_free_nointr, 0,
1.117     yamt     2771:        .pa_backingmapptr = &kernel_map,
1.112     bjh21    2772: };
                   2773: #else
1.66      thorpej  2774: struct pool_allocator pool_allocator_nointr = {
                   2775:        pool_page_alloc_nointr, pool_page_free_nointr, 0,
1.117     yamt     2776:        .pa_backingmapptr = &kernel_map,
1.66      thorpej  2777: };
1.112     bjh21    2778: #endif
1.66      thorpej  2779:
                   2780: #ifdef POOL_SUBPAGE
                   2781: void   *pool_subpage_alloc(struct pool *, int);
                   2782: void   pool_subpage_free(struct pool *, void *);
                   2783:
1.112     bjh21    2784: struct pool_allocator pool_allocator_kmem = {
                   2785:        pool_subpage_alloc, pool_subpage_free, POOL_SUBPAGE,
1.117     yamt     2786:        .pa_backingmapptr = &kmem_map,
1.112     bjh21    2787: };
                   2788:
                   2789: void   *pool_subpage_alloc_nointr(struct pool *, int);
                   2790: void   pool_subpage_free_nointr(struct pool *, void *);
                   2791:
                   2792: struct pool_allocator pool_allocator_nointr = {
                   2793:        pool_subpage_alloc, pool_subpage_free, POOL_SUBPAGE,
1.117     yamt     2794:        .pa_backingmapptr = &kmem_map,
1.66      thorpej  2795: };
                   2796: #endif /* POOL_SUBPAGE */
                   2797:
1.117     yamt     2798: static void *
                   2799: pool_allocator_alloc(struct pool *pp, int flags)
1.66      thorpej  2800: {
1.117     yamt     2801:        struct pool_allocator *pa = pp->pr_alloc;
1.66      thorpej  2802:        void *res;
                   2803:
1.117     yamt     2804:        res = (*pa->pa_alloc)(pp, flags);
                   2805:        if (res == NULL && (flags & PR_WAITOK) == 0) {
1.66      thorpej  2806:                /*
1.117     yamt     2807:                 * We only run the drain hook here if PR_NOWAIT.
                   2808:                 * In other cases, the hook will be run in
                   2809:                 * pool_reclaim().
1.66      thorpej  2810:                 */
1.117     yamt     2811:                if (pp->pr_drain_hook != NULL) {
                   2812:                        (*pp->pr_drain_hook)(pp->pr_drain_hook_arg, flags);
                   2813:                        res = (*pa->pa_alloc)(pp, flags);
1.66      thorpej  2814:                }
1.117     yamt     2815:        }
                   2816:        return res;
1.66      thorpej  2817: }
                   2818:
1.117     yamt     2819: static void
1.66      thorpej  2820: pool_allocator_free(struct pool *pp, void *v)
                   2821: {
                   2822:        struct pool_allocator *pa = pp->pr_alloc;
                   2823:
                   2824:        (*pa->pa_free)(pp, v);
                   2825: }
                   2826:
                   2827: void *
1.124     yamt     2828: pool_page_alloc(struct pool *pp, int flags)
1.66      thorpej  2829: {
1.127     thorpej  2830:        bool waitok = (flags & PR_WAITOK) ? true : false;
1.66      thorpej  2831:
1.100     yamt     2832:        return ((void *) uvm_km_alloc_poolpage_cache(kmem_map, waitok));
1.66      thorpej  2833: }
                   2834:
                   2835: void
1.124     yamt     2836: pool_page_free(struct pool *pp, void *v)
1.66      thorpej  2837: {
                   2838:
1.98      yamt     2839:        uvm_km_free_poolpage_cache(kmem_map, (vaddr_t) v);
                   2840: }
                   2841:
                   2842: static void *
1.124     yamt     2843: pool_page_alloc_meta(struct pool *pp, int flags)
1.98      yamt     2844: {
1.127     thorpej  2845:        bool waitok = (flags & PR_WAITOK) ? true : false;
1.98      yamt     2846:
1.100     yamt     2847:        return ((void *) uvm_km_alloc_poolpage(kmem_map, waitok));
1.98      yamt     2848: }
                   2849:
                   2850: static void
1.124     yamt     2851: pool_page_free_meta(struct pool *pp, void *v)
1.98      yamt     2852: {
                   2853:
1.100     yamt     2854:        uvm_km_free_poolpage(kmem_map, (vaddr_t) v);
1.66      thorpej  2855: }
                   2856:
                   2857: #ifdef POOL_SUBPAGE
                   2858: /* Sub-page allocator, for machines with large hardware pages. */
                   2859: void *
                   2860: pool_subpage_alloc(struct pool *pp, int flags)
                   2861: {
1.134     ad       2862:        return pool_get(&psppool, flags);
1.66      thorpej  2863: }
                   2864:
                   2865: void
                   2866: pool_subpage_free(struct pool *pp, void *v)
                   2867: {
                   2868:        pool_put(&psppool, v);
                   2869: }
                   2870:
                   2871: /* We don't provide a real nointr allocator.  Maybe later. */
                   2872: void *
1.112     bjh21    2873: pool_subpage_alloc_nointr(struct pool *pp, int flags)
1.66      thorpej  2874: {
                   2875:
                   2876:        return (pool_subpage_alloc(pp, flags));
                   2877: }
                   2878:
                   2879: void
1.112     bjh21    2880: pool_subpage_free_nointr(struct pool *pp, void *v)
1.66      thorpej  2881: {
                   2882:
                   2883:        pool_subpage_free(pp, v);
                   2884: }
1.112     bjh21    2885: #endif /* POOL_SUBPAGE */
1.66      thorpej  2886: void *
1.124     yamt     2887: pool_page_alloc_nointr(struct pool *pp, int flags)
1.66      thorpej  2888: {
1.127     thorpej  2889:        bool waitok = (flags & PR_WAITOK) ? true : false;
1.66      thorpej  2890:
1.100     yamt     2891:        return ((void *) uvm_km_alloc_poolpage_cache(kernel_map, waitok));
1.66      thorpej  2892: }
                   2893:
                   2894: void
1.124     yamt     2895: pool_page_free_nointr(struct pool *pp, void *v)
1.66      thorpej  2896: {
                   2897:
1.98      yamt     2898:        uvm_km_free_poolpage_cache(kernel_map, (vaddr_t) v);
1.66      thorpej  2899: }
1.141   ! yamt     2900:
        !          2901: #if defined(DDB)
        !          2902: static bool
        !          2903: pool_in_page(struct pool *pp, struct pool_item_header *ph, uintptr_t addr)
        !          2904: {
        !          2905:
        !          2906:        return (uintptr_t)ph->ph_page <= addr &&
        !          2907:            addr < (uintptr_t)ph->ph_page + pp->pr_alloc->pa_pagesz;
        !          2908: }
        !          2909:
        !          2910: void
        !          2911: pool_whatis(uintptr_t addr, void (*pr)(const char *, ...))
        !          2912: {
        !          2913:        struct pool *pp;
        !          2914:
        !          2915:        LIST_FOREACH(pp, &pool_head, pr_poollist) {
        !          2916:                struct pool_item_header *ph;
        !          2917:                uintptr_t item;
        !          2918:
        !          2919:                if ((pp->pr_roflags & PR_PHINPAGE) != 0) {
        !          2920:                        LIST_FOREACH(ph, &pp->pr_fullpages, ph_pagelist) {
        !          2921:                                if (pool_in_page(pp, ph, addr)) {
        !          2922:                                        goto found;
        !          2923:                                }
        !          2924:                        }
        !          2925:                        LIST_FOREACH(ph, &pp->pr_partpages, ph_pagelist) {
        !          2926:                                if (pool_in_page(pp, ph, addr)) {
        !          2927:                                        goto found;
        !          2928:                                }
        !          2929:                        }
        !          2930:                        continue;
        !          2931:                } else {
        !          2932:                        ph = pr_find_pagehead_noalign(pp, (void *)addr);
        !          2933:                        if (ph == NULL || !pool_in_page(pp, ph, addr)) {
        !          2934:                                continue;
        !          2935:                        }
        !          2936:                }
        !          2937: found:
        !          2938:                item = (uintptr_t)ph->ph_page + ph->ph_off;
        !          2939:                item = item + rounddown(addr - item, pp->pr_size);
        !          2940:                (*pr)("%p is %p+%zu from POOL '%s'\n",
        !          2941:                    (void *)addr, item, (size_t)(addr - item),
        !          2942:                    pp->pr_wchan);
        !          2943:        }
        !          2944: }
        !          2945: #endif /* defined(DDB) */

CVSweb <webmaster@jp.NetBSD.org>