[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 1.137.2.3

1.137.2.3! ad          1: /*     $NetBSD: subr_pool.c,v 1.137.2.2 2007/12/12 22:03:31 ad 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>
1.137.2.3! ad         41: __KERNEL_RCSID(0, "$NetBSD: subr_pool.c,v 1.137.2.2 2007/12/12 22:03:31 ad Exp $");
1.24      scottr     42:
1.25      thorpej    43: #include "opt_pool.h"
1.24      scottr     44: #include "opt_poollog.h"
1.28      thorpej    45: #include "opt_lockdebug.h"
1.1       pk         46:
                     47: #include <sys/param.h>
                     48: #include <sys/systm.h>
1.135     yamt       49: #include <sys/bitops.h>
1.1       pk         50: #include <sys/proc.h>
                     51: #include <sys/errno.h>
                     52: #include <sys/kernel.h>
                     53: #include <sys/malloc.h>
                     54: #include <sys/lock.h>
                     55: #include <sys/pool.h>
1.20      thorpej    56: #include <sys/syslog.h>
1.125     ad         57: #include <sys/debug.h>
1.134     ad         58: #include <sys/lockdebug.h>
                     59: #include <sys/xcall.h>
                     60: #include <sys/cpu.h>
1.3       pk         61:
                     62: #include <uvm/uvm.h>
                     63:
1.1       pk         64: /*
                     65:  * Pool resource management utility.
1.3       pk         66:  *
1.88      chs        67:  * Memory is allocated in pages which are split into pieces according to
                     68:  * the pool item size. Each page is kept on one of three lists in the
                     69:  * pool structure: `pr_emptypages', `pr_fullpages' and `pr_partpages',
                     70:  * for empty, full and partially-full pages respectively. The individual
                     71:  * pool items are on a linked list headed by `ph_itemlist' in each page
                     72:  * header. The memory for building the page list is either taken from
                     73:  * the allocated pages themselves (for small pool items) or taken from
                     74:  * an internal pool of page headers (`phpool').
1.1       pk         75:  */
                     76:
1.3       pk         77: /* List of all pools */
1.137.2.3! ad         78: TAILQ_HEAD(,pool) pool_head = TAILQ_HEAD_INITIALIZER(pool_head);
1.3       pk         79:
                     80: /* Private pool for page header structures */
1.97      yamt       81: #define        PHPOOL_MAX      8
                     82: static struct pool phpool[PHPOOL_MAX];
1.135     yamt       83: #define        PHPOOL_FREELIST_NELEM(idx) \
                     84:        (((idx) == 0) ? 0 : BITMAP_SIZE * (1 << (idx)))
1.3       pk         85:
1.62      bjh21      86: #ifdef POOL_SUBPAGE
                     87: /* Pool of subpages for use by normal pools. */
                     88: static struct pool psppool;
                     89: #endif
                     90:
1.117     yamt       91: static SLIST_HEAD(, pool_allocator) pa_deferinitq =
                     92:     SLIST_HEAD_INITIALIZER(pa_deferinitq);
                     93:
1.98      yamt       94: static void *pool_page_alloc_meta(struct pool *, int);
                     95: static void pool_page_free_meta(struct pool *, void *);
                     96:
                     97: /* allocator for pool metadata */
1.134     ad         98: struct pool_allocator pool_allocator_meta = {
1.117     yamt       99:        pool_page_alloc_meta, pool_page_free_meta,
                    100:        .pa_backingmapptr = &kmem_map,
1.98      yamt      101: };
                    102:
1.3       pk        103: /* # of seconds to retain page after last use */
                    104: int pool_inactive_time = 10;
                    105:
                    106: /* Next candidate for drainage (see pool_drain()) */
1.23      thorpej   107: static struct pool     *drainpp;
                    108:
1.134     ad        109: /* This lock protects both pool_head and drainpp. */
                    110: static kmutex_t pool_head_lock;
                    111: static kcondvar_t pool_busy;
1.3       pk        112:
1.135     yamt      113: typedef uint32_t pool_item_bitmap_t;
                    114: #define        BITMAP_SIZE     (CHAR_BIT * sizeof(pool_item_bitmap_t))
                    115: #define        BITMAP_MASK     (BITMAP_SIZE - 1)
1.99      yamt      116:
1.3       pk        117: struct pool_item_header {
                    118:        /* Page headers */
1.88      chs       119:        LIST_ENTRY(pool_item_header)
1.3       pk        120:                                ph_pagelist;    /* pool page list */
1.88      chs       121:        SPLAY_ENTRY(pool_item_header)
                    122:                                ph_node;        /* Off-page page headers */
1.128     christos  123:        void *                  ph_page;        /* this page's address */
1.3       pk        124:        struct timeval          ph_time;        /* last referenced */
1.135     yamt      125:        uint16_t                ph_nmissing;    /* # of chunks in use */
1.97      yamt      126:        union {
                    127:                /* !PR_NOTOUCH */
                    128:                struct {
1.102     chs       129:                        LIST_HEAD(, pool_item)
1.97      yamt      130:                                phu_itemlist;   /* chunk list for this page */
                    131:                } phu_normal;
                    132:                /* PR_NOTOUCH */
                    133:                struct {
1.135     yamt      134:                        uint16_t phu_off;       /* start offset in page */
                    135:                        pool_item_bitmap_t phu_bitmap[];
1.97      yamt      136:                } phu_notouch;
                    137:        } ph_u;
1.3       pk        138: };
1.97      yamt      139: #define        ph_itemlist     ph_u.phu_normal.phu_itemlist
                    140: #define        ph_off          ph_u.phu_notouch.phu_off
1.135     yamt      141: #define        ph_bitmap       ph_u.phu_notouch.phu_bitmap
1.3       pk        142:
1.1       pk        143: struct pool_item {
1.3       pk        144: #ifdef DIAGNOSTIC
1.82      thorpej   145:        u_int pi_magic;
1.33      chs       146: #endif
1.134     ad        147: #define        PI_MAGIC 0xdeaddeadU
1.3       pk        148:        /* Other entries use only this list entry */
1.102     chs       149:        LIST_ENTRY(pool_item)   pi_list;
1.3       pk        150: };
                    151:
1.53      thorpej   152: #define        POOL_NEEDS_CATCHUP(pp)                                          \
                    153:        ((pp)->pr_nitems < (pp)->pr_minitems)
                    154:
1.43      thorpej   155: /*
                    156:  * Pool cache management.
                    157:  *
                    158:  * Pool caches provide a way for constructed objects to be cached by the
                    159:  * pool subsystem.  This can lead to performance improvements by avoiding
                    160:  * needless object construction/destruction; it is deferred until absolutely
                    161:  * necessary.
                    162:  *
1.134     ad        163:  * Caches are grouped into cache groups.  Each cache group references up
                    164:  * to PCG_NUMOBJECTS constructed objects.  When a cache allocates an
                    165:  * object from the pool, it calls the object's constructor and places it
                    166:  * into a cache group.  When a cache group frees an object back to the
                    167:  * pool, it first calls the object's destructor.  This allows the object
                    168:  * to persist in constructed form while freed to the cache.
                    169:  *
                    170:  * The pool references each cache, so that when a pool is drained by the
                    171:  * pagedaemon, it can drain each individual cache as well.  Each time a
                    172:  * cache is drained, the most idle cache group is freed to the pool in
                    173:  * its entirety.
1.43      thorpej   174:  *
                    175:  * Pool caches are layed on top of pools.  By layering them, we can avoid
                    176:  * the complexity of cache management for pools which would not benefit
                    177:  * from it.
                    178:  */
                    179:
                    180: static struct pool pcgpool;
1.134     ad        181: static struct pool cache_pool;
                    182: static struct pool cache_cpu_pool;
1.3       pk        183:
1.137.2.2  ad        184: /* List of all caches. */
1.137.2.3! ad        185: TAILQ_HEAD(,pool_cache) pool_cache_head =
        !           186:     TAILQ_HEAD_INITIALIZER(pool_cache_head);
1.137.2.2  ad        187:
                    188: int pool_cache_disable;
                    189:
                    190:
1.134     ad        191: static pool_cache_cpu_t *pool_cache_put_slow(pool_cache_cpu_t *, int *,
                    192:                                             void *, paddr_t);
                    193: static pool_cache_cpu_t *pool_cache_get_slow(pool_cache_cpu_t *, int *,
                    194:                                             void **, paddr_t *, int);
                    195: static void    pool_cache_cpu_init1(struct cpu_info *, pool_cache_t);
                    196: static void    pool_cache_invalidate_groups(pool_cache_t, pcg_t *);
                    197: static void    pool_cache_xcall(pool_cache_t);
1.3       pk        198:
1.42      thorpej   199: static int     pool_catchup(struct pool *);
1.128     christos  200: static void    pool_prime_page(struct pool *, void *,
1.55      thorpej   201:                    struct pool_item_header *);
1.88      chs       202: static void    pool_update_curpage(struct pool *);
1.66      thorpej   203:
1.113     yamt      204: static int     pool_grow(struct pool *, int);
1.117     yamt      205: static void    *pool_allocator_alloc(struct pool *, int);
                    206: static void    pool_allocator_free(struct pool *, void *);
1.3       pk        207:
1.97      yamt      208: static void pool_print_pagelist(struct pool *, struct pool_pagelist *,
1.88      chs       209:        void (*)(const char *, ...));
1.42      thorpej   210: static void pool_print1(struct pool *, const char *,
                    211:        void (*)(const char *, ...));
1.3       pk        212:
1.88      chs       213: static int pool_chk_page(struct pool *, const char *,
                    214:                         struct pool_item_header *);
                    215:
1.3       pk        216: /*
1.52      thorpej   217:  * Pool log entry. An array of these is allocated in pool_init().
1.3       pk        218:  */
                    219: struct pool_log {
                    220:        const char      *pl_file;
                    221:        long            pl_line;
                    222:        int             pl_action;
1.25      thorpej   223: #define        PRLOG_GET       1
                    224: #define        PRLOG_PUT       2
1.3       pk        225:        void            *pl_addr;
1.1       pk        226: };
                    227:
1.86      matt      228: #ifdef POOL_DIAGNOSTIC
1.3       pk        229: /* Number of entries in pool log buffers */
1.17      thorpej   230: #ifndef POOL_LOGSIZE
                    231: #define        POOL_LOGSIZE    10
                    232: #endif
                    233:
                    234: int pool_logsize = POOL_LOGSIZE;
1.1       pk        235:
1.110     perry     236: static inline void
1.42      thorpej   237: pr_log(struct pool *pp, void *v, int action, const char *file, long line)
1.3       pk        238: {
                    239:        int n = pp->pr_curlogentry;
                    240:        struct pool_log *pl;
                    241:
1.20      thorpej   242:        if ((pp->pr_roflags & PR_LOGGING) == 0)
1.3       pk        243:                return;
                    244:
                    245:        /*
                    246:         * Fill in the current entry. Wrap around and overwrite
                    247:         * the oldest entry if necessary.
                    248:         */
                    249:        pl = &pp->pr_log[n];
                    250:        pl->pl_file = file;
                    251:        pl->pl_line = line;
                    252:        pl->pl_action = action;
                    253:        pl->pl_addr = v;
                    254:        if (++n >= pp->pr_logsize)
                    255:                n = 0;
                    256:        pp->pr_curlogentry = n;
                    257: }
                    258:
                    259: static void
1.42      thorpej   260: pr_printlog(struct pool *pp, struct pool_item *pi,
                    261:     void (*pr)(const char *, ...))
1.3       pk        262: {
                    263:        int i = pp->pr_logsize;
                    264:        int n = pp->pr_curlogentry;
                    265:
1.20      thorpej   266:        if ((pp->pr_roflags & PR_LOGGING) == 0)
1.3       pk        267:                return;
                    268:
                    269:        /*
                    270:         * Print all entries in this pool's log.
                    271:         */
                    272:        while (i-- > 0) {
                    273:                struct pool_log *pl = &pp->pr_log[n];
                    274:                if (pl->pl_action != 0) {
1.25      thorpej   275:                        if (pi == NULL || pi == pl->pl_addr) {
                    276:                                (*pr)("\tlog entry %d:\n", i);
                    277:                                (*pr)("\t\taction = %s, addr = %p\n",
                    278:                                    pl->pl_action == PRLOG_GET ? "get" : "put",
                    279:                                    pl->pl_addr);
                    280:                                (*pr)("\t\tfile: %s at line %lu\n",
                    281:                                    pl->pl_file, pl->pl_line);
                    282:                        }
1.3       pk        283:                }
                    284:                if (++n >= pp->pr_logsize)
                    285:                        n = 0;
                    286:        }
                    287: }
1.25      thorpej   288:
1.110     perry     289: static inline void
1.42      thorpej   290: pr_enter(struct pool *pp, const char *file, long line)
1.25      thorpej   291: {
                    292:
1.34      thorpej   293:        if (__predict_false(pp->pr_entered_file != NULL)) {
1.25      thorpej   294:                printf("pool %s: reentrancy at file %s line %ld\n",
                    295:                    pp->pr_wchan, file, line);
                    296:                printf("         previous entry at file %s line %ld\n",
                    297:                    pp->pr_entered_file, pp->pr_entered_line);
                    298:                panic("pr_enter");
                    299:        }
                    300:
                    301:        pp->pr_entered_file = file;
                    302:        pp->pr_entered_line = line;
                    303: }
                    304:
1.110     perry     305: static inline void
1.42      thorpej   306: pr_leave(struct pool *pp)
1.25      thorpej   307: {
                    308:
1.34      thorpej   309:        if (__predict_false(pp->pr_entered_file == NULL)) {
1.25      thorpej   310:                printf("pool %s not entered?\n", pp->pr_wchan);
                    311:                panic("pr_leave");
                    312:        }
                    313:
                    314:        pp->pr_entered_file = NULL;
                    315:        pp->pr_entered_line = 0;
                    316: }
                    317:
1.110     perry     318: static inline void
1.42      thorpej   319: pr_enter_check(struct pool *pp, void (*pr)(const char *, ...))
1.25      thorpej   320: {
                    321:
                    322:        if (pp->pr_entered_file != NULL)
                    323:                (*pr)("\n\tcurrently entered from file %s line %ld\n",
                    324:                    pp->pr_entered_file, pp->pr_entered_line);
                    325: }
1.3       pk        326: #else
1.25      thorpej   327: #define        pr_log(pp, v, action, file, line)
                    328: #define        pr_printlog(pp, pi, pr)
                    329: #define        pr_enter(pp, file, line)
                    330: #define        pr_leave(pp)
                    331: #define        pr_enter_check(pp, pr)
1.59      thorpej   332: #endif /* POOL_DIAGNOSTIC */
1.3       pk        333:
1.135     yamt      334: static inline unsigned int
1.97      yamt      335: pr_item_notouch_index(const struct pool *pp, const struct pool_item_header *ph,
                    336:     const void *v)
                    337: {
                    338:        const char *cp = v;
1.135     yamt      339:        unsigned int idx;
1.97      yamt      340:
                    341:        KASSERT(pp->pr_roflags & PR_NOTOUCH);
1.128     christos  342:        idx = (cp - (char *)ph->ph_page - ph->ph_off) / pp->pr_size;
1.97      yamt      343:        KASSERT(idx < pp->pr_itemsperpage);
                    344:        return idx;
                    345: }
                    346:
1.110     perry     347: static inline void
1.97      yamt      348: pr_item_notouch_put(const struct pool *pp, struct pool_item_header *ph,
                    349:     void *obj)
                    350: {
1.135     yamt      351:        unsigned int idx = pr_item_notouch_index(pp, ph, obj);
                    352:        pool_item_bitmap_t *bitmap = ph->ph_bitmap + (idx / BITMAP_SIZE);
                    353:        pool_item_bitmap_t mask = 1 << (idx & BITMAP_MASK);
1.97      yamt      354:
1.135     yamt      355:        KASSERT((*bitmap & mask) == 0);
                    356:        *bitmap |= mask;
1.97      yamt      357: }
                    358:
1.110     perry     359: static inline void *
1.97      yamt      360: pr_item_notouch_get(const struct pool *pp, struct pool_item_header *ph)
                    361: {
1.135     yamt      362:        pool_item_bitmap_t *bitmap = ph->ph_bitmap;
                    363:        unsigned int idx;
                    364:        int i;
1.97      yamt      365:
1.135     yamt      366:        for (i = 0; ; i++) {
                    367:                int bit;
1.97      yamt      368:
1.135     yamt      369:                KASSERT((i * BITMAP_SIZE) < pp->pr_itemsperpage);
                    370:                bit = ffs32(bitmap[i]);
                    371:                if (bit) {
                    372:                        pool_item_bitmap_t mask;
                    373:
                    374:                        bit--;
                    375:                        idx = (i * BITMAP_SIZE) + bit;
                    376:                        mask = 1 << bit;
                    377:                        KASSERT((bitmap[i] & mask) != 0);
                    378:                        bitmap[i] &= ~mask;
                    379:                        break;
                    380:                }
                    381:        }
                    382:        KASSERT(idx < pp->pr_itemsperpage);
1.128     christos  383:        return (char *)ph->ph_page + ph->ph_off + idx * pp->pr_size;
1.97      yamt      384: }
                    385:
1.135     yamt      386: static inline void
                    387: pr_item_notouch_init(const struct pool *pp, struct pool_item_header *ph)
                    388: {
                    389:        pool_item_bitmap_t *bitmap = ph->ph_bitmap;
                    390:        const int n = howmany(pp->pr_itemsperpage, BITMAP_SIZE);
                    391:        int i;
                    392:
                    393:        for (i = 0; i < n; i++) {
                    394:                bitmap[i] = (pool_item_bitmap_t)-1;
                    395:        }
                    396: }
                    397:
1.110     perry     398: static inline int
1.88      chs       399: phtree_compare(struct pool_item_header *a, struct pool_item_header *b)
                    400: {
1.121     yamt      401:
                    402:        /*
                    403:         * we consider pool_item_header with smaller ph_page bigger.
                    404:         * (this unnatural ordering is for the benefit of pr_find_pagehead.)
                    405:         */
                    406:
1.88      chs       407:        if (a->ph_page < b->ph_page)
1.121     yamt      408:                return (1);
                    409:        else if (a->ph_page > b->ph_page)
1.88      chs       410:                return (-1);
                    411:        else
                    412:                return (0);
                    413: }
                    414:
                    415: SPLAY_PROTOTYPE(phtree, pool_item_header, ph_node, phtree_compare);
                    416: SPLAY_GENERATE(phtree, pool_item_header, ph_node, phtree_compare);
                    417:
1.3       pk        418: /*
1.121     yamt      419:  * Return the pool page header based on item address.
1.3       pk        420:  */
1.110     perry     421: static inline struct pool_item_header *
1.121     yamt      422: pr_find_pagehead(struct pool *pp, void *v)
1.3       pk        423: {
1.88      chs       424:        struct pool_item_header *ph, tmp;
1.3       pk        425:
1.121     yamt      426:        if ((pp->pr_roflags & PR_NOALIGN) != 0) {
1.128     christos  427:                tmp.ph_page = (void *)(uintptr_t)v;
1.121     yamt      428:                ph = SPLAY_FIND(phtree, &pp->pr_phtree, &tmp);
                    429:                if (ph == NULL) {
                    430:                        ph = SPLAY_ROOT(&pp->pr_phtree);
                    431:                        if (ph != NULL && phtree_compare(&tmp, ph) >= 0) {
                    432:                                ph = SPLAY_NEXT(phtree, &pp->pr_phtree, ph);
                    433:                        }
                    434:                        KASSERT(ph == NULL || phtree_compare(&tmp, ph) < 0);
                    435:                }
                    436:        } else {
1.128     christos  437:                void *page =
                    438:                    (void *)((uintptr_t)v & pp->pr_alloc->pa_pagemask);
1.121     yamt      439:
                    440:                if ((pp->pr_roflags & PR_PHINPAGE) != 0) {
1.128     christos  441:                        ph = (struct pool_item_header *)((char *)page + pp->pr_phoffset);
1.121     yamt      442:                } else {
                    443:                        tmp.ph_page = page;
                    444:                        ph = SPLAY_FIND(phtree, &pp->pr_phtree, &tmp);
                    445:                }
                    446:        }
1.3       pk        447:
1.121     yamt      448:        KASSERT(ph == NULL || ((pp->pr_roflags & PR_PHINPAGE) != 0) ||
1.128     christos  449:            ((char *)ph->ph_page <= (char *)v &&
                    450:            (char *)v < (char *)ph->ph_page + pp->pr_alloc->pa_pagesz));
1.88      chs       451:        return ph;
1.3       pk        452: }
                    453:
1.101     thorpej   454: static void
                    455: pr_pagelist_free(struct pool *pp, struct pool_pagelist *pq)
                    456: {
                    457:        struct pool_item_header *ph;
                    458:
                    459:        while ((ph = LIST_FIRST(pq)) != NULL) {
                    460:                LIST_REMOVE(ph, ph_pagelist);
                    461:                pool_allocator_free(pp, ph->ph_page);
1.134     ad        462:                if ((pp->pr_roflags & PR_PHINPAGE) == 0)
1.101     thorpej   463:                        pool_put(pp->pr_phpool, ph);
                    464:        }
                    465: }
                    466:
1.3       pk        467: /*
                    468:  * Remove a page from the pool.
                    469:  */
1.110     perry     470: static inline void
1.61      chs       471: pr_rmpage(struct pool *pp, struct pool_item_header *ph,
                    472:      struct pool_pagelist *pq)
1.3       pk        473: {
                    474:
1.134     ad        475:        KASSERT(mutex_owned(&pp->pr_lock));
1.91      yamt      476:
1.3       pk        477:        /*
1.7       thorpej   478:         * If the page was idle, decrement the idle page count.
1.3       pk        479:         */
1.6       thorpej   480:        if (ph->ph_nmissing == 0) {
                    481: #ifdef DIAGNOSTIC
                    482:                if (pp->pr_nidle == 0)
                    483:                        panic("pr_rmpage: nidle inconsistent");
1.20      thorpej   484:                if (pp->pr_nitems < pp->pr_itemsperpage)
                    485:                        panic("pr_rmpage: nitems inconsistent");
1.6       thorpej   486: #endif
                    487:                pp->pr_nidle--;
                    488:        }
1.7       thorpej   489:
1.20      thorpej   490:        pp->pr_nitems -= pp->pr_itemsperpage;
                    491:
1.7       thorpej   492:        /*
1.101     thorpej   493:         * Unlink the page from the pool and queue it for release.
1.7       thorpej   494:         */
1.88      chs       495:        LIST_REMOVE(ph, ph_pagelist);
1.91      yamt      496:        if ((pp->pr_roflags & PR_PHINPAGE) == 0)
                    497:                SPLAY_REMOVE(phtree, &pp->pr_phtree, ph);
1.101     thorpej   498:        LIST_INSERT_HEAD(pq, ph, ph_pagelist);
                    499:
1.7       thorpej   500:        pp->pr_npages--;
                    501:        pp->pr_npagefree++;
1.6       thorpej   502:
1.88      chs       503:        pool_update_curpage(pp);
1.3       pk        504: }
                    505:
1.126     thorpej   506: static bool
1.117     yamt      507: pa_starved_p(struct pool_allocator *pa)
                    508: {
                    509:
                    510:        if (pa->pa_backingmap != NULL) {
                    511:                return vm_map_starved_p(pa->pa_backingmap);
                    512:        }
1.127     thorpej   513:        return false;
1.117     yamt      514: }
                    515:
                    516: static int
1.124     yamt      517: pool_reclaim_callback(struct callback_entry *ce, void *obj, void *arg)
1.117     yamt      518: {
                    519:        struct pool *pp = obj;
                    520:        struct pool_allocator *pa = pp->pr_alloc;
                    521:
                    522:        KASSERT(&pp->pr_reclaimerentry == ce);
                    523:        pool_reclaim(pp);
                    524:        if (!pa_starved_p(pa)) {
                    525:                return CALLBACK_CHAIN_ABORT;
                    526:        }
                    527:        return CALLBACK_CHAIN_CONTINUE;
                    528: }
                    529:
                    530: static void
                    531: pool_reclaim_register(struct pool *pp)
                    532: {
                    533:        struct vm_map *map = pp->pr_alloc->pa_backingmap;
                    534:        int s;
                    535:
                    536:        if (map == NULL) {
                    537:                return;
                    538:        }
                    539:
                    540:        s = splvm(); /* not necessary for INTRSAFE maps, but don't care. */
                    541:        callback_register(&vm_map_to_kernel(map)->vmk_reclaim_callback,
                    542:            &pp->pr_reclaimerentry, pp, pool_reclaim_callback);
                    543:        splx(s);
                    544: }
                    545:
                    546: static void
                    547: pool_reclaim_unregister(struct pool *pp)
                    548: {
                    549:        struct vm_map *map = pp->pr_alloc->pa_backingmap;
                    550:        int s;
                    551:
                    552:        if (map == NULL) {
                    553:                return;
                    554:        }
                    555:
                    556:        s = splvm(); /* not necessary for INTRSAFE maps, but don't care. */
                    557:        callback_unregister(&vm_map_to_kernel(map)->vmk_reclaim_callback,
                    558:            &pp->pr_reclaimerentry);
                    559:        splx(s);
                    560: }
                    561:
                    562: static void
                    563: pa_reclaim_register(struct pool_allocator *pa)
                    564: {
                    565:        struct vm_map *map = *pa->pa_backingmapptr;
                    566:        struct pool *pp;
                    567:
                    568:        KASSERT(pa->pa_backingmap == NULL);
                    569:        if (map == NULL) {
                    570:                SLIST_INSERT_HEAD(&pa_deferinitq, pa, pa_q);
                    571:                return;
                    572:        }
                    573:        pa->pa_backingmap = map;
                    574:        TAILQ_FOREACH(pp, &pa->pa_list, pr_alloc_list) {
                    575:                pool_reclaim_register(pp);
                    576:        }
                    577: }
                    578:
1.3       pk        579: /*
1.94      simonb    580:  * Initialize all the pools listed in the "pools" link set.
                    581:  */
                    582: void
1.117     yamt      583: pool_subsystem_init(void)
1.94      simonb    584: {
1.117     yamt      585:        struct pool_allocator *pa;
1.94      simonb    586:        __link_set_decl(pools, struct link_pool_init);
                    587:        struct link_pool_init * const *pi;
                    588:
1.134     ad        589:        mutex_init(&pool_head_lock, MUTEX_DEFAULT, IPL_NONE);
                    590:        cv_init(&pool_busy, "poolbusy");
                    591:
1.94      simonb    592:        __link_set_foreach(pi, pools)
                    593:                pool_init((*pi)->pp, (*pi)->size, (*pi)->align,
                    594:                    (*pi)->align_offset, (*pi)->flags, (*pi)->wchan,
1.129     ad        595:                    (*pi)->palloc, (*pi)->ipl);
1.117     yamt      596:
                    597:        while ((pa = SLIST_FIRST(&pa_deferinitq)) != NULL) {
                    598:                KASSERT(pa->pa_backingmapptr != NULL);
                    599:                KASSERT(*pa->pa_backingmapptr != NULL);
                    600:                SLIST_REMOVE_HEAD(&pa_deferinitq, pa_q);
                    601:                pa_reclaim_register(pa);
                    602:        }
1.134     ad        603:
                    604:        pool_init(&cache_pool, sizeof(struct pool_cache), CACHE_LINE_SIZE,
                    605:            0, 0, "pcache", &pool_allocator_nointr, IPL_NONE);
                    606:
                    607:        pool_init(&cache_cpu_pool, sizeof(pool_cache_cpu_t), CACHE_LINE_SIZE,
                    608:            0, 0, "pcachecpu", &pool_allocator_nointr, IPL_NONE);
1.94      simonb    609: }
                    610:
                    611: /*
1.3       pk        612:  * Initialize the given pool resource structure.
                    613:  *
                    614:  * We export this routine to allow other kernel parts to declare
                    615:  * static pools that must be initialized before malloc() is available.
                    616:  */
                    617: void
1.42      thorpej   618: pool_init(struct pool *pp, size_t size, u_int align, u_int ioff, int flags,
1.129     ad        619:     const char *wchan, struct pool_allocator *palloc, int ipl)
1.3       pk        620: {
1.116     simonb    621:        struct pool *pp1;
1.92      enami     622:        size_t trysize, phsize;
1.134     ad        623:        int off, slack;
1.3       pk        624:
1.116     simonb    625: #ifdef DEBUG
                    626:        /*
                    627:         * Check that the pool hasn't already been initialised and
                    628:         * added to the list of all pools.
                    629:         */
1.137.2.3! ad        630:        TAILQ_FOREACH(pp1, &pool_head, pr_poollist) {
1.116     simonb    631:                if (pp == pp1)
                    632:                        panic("pool_init: pool %s already initialised",
                    633:                            wchan);
                    634:        }
                    635: #endif
                    636:
1.25      thorpej   637: #ifdef POOL_DIAGNOSTIC
                    638:        /*
                    639:         * Always log if POOL_DIAGNOSTIC is defined.
                    640:         */
                    641:        if (pool_logsize != 0)
                    642:                flags |= PR_LOGGING;
                    643: #endif
                    644:
1.66      thorpej   645:        if (palloc == NULL)
                    646:                palloc = &pool_allocator_kmem;
1.112     bjh21     647: #ifdef POOL_SUBPAGE
                    648:        if (size > palloc->pa_pagesz) {
                    649:                if (palloc == &pool_allocator_kmem)
                    650:                        palloc = &pool_allocator_kmem_fullpage;
                    651:                else if (palloc == &pool_allocator_nointr)
                    652:                        palloc = &pool_allocator_nointr_fullpage;
                    653:        }
1.66      thorpej   654: #endif /* POOL_SUBPAGE */
                    655:        if ((palloc->pa_flags & PA_INITIALIZED) == 0) {
1.112     bjh21     656:                if (palloc->pa_pagesz == 0)
1.66      thorpej   657:                        palloc->pa_pagesz = PAGE_SIZE;
                    658:
                    659:                TAILQ_INIT(&palloc->pa_list);
                    660:
1.134     ad        661:                mutex_init(&palloc->pa_lock, MUTEX_DEFAULT, IPL_VM);
1.66      thorpej   662:                palloc->pa_pagemask = ~(palloc->pa_pagesz - 1);
                    663:                palloc->pa_pageshift = ffs(palloc->pa_pagesz) - 1;
1.117     yamt      664:
                    665:                if (palloc->pa_backingmapptr != NULL) {
                    666:                        pa_reclaim_register(palloc);
                    667:                }
1.66      thorpej   668:                palloc->pa_flags |= PA_INITIALIZED;
1.4       thorpej   669:        }
1.3       pk        670:
                    671:        if (align == 0)
                    672:                align = ALIGN(1);
1.14      thorpej   673:
1.120     yamt      674:        if ((flags & PR_NOTOUCH) == 0 && size < sizeof(struct pool_item))
1.14      thorpej   675:                size = sizeof(struct pool_item);
1.3       pk        676:
1.78      thorpej   677:        size = roundup(size, align);
1.66      thorpej   678: #ifdef DIAGNOSTIC
                    679:        if (size > palloc->pa_pagesz)
1.121     yamt      680:                panic("pool_init: pool item size (%zu) too large", size);
1.66      thorpej   681: #endif
1.35      pk        682:
1.3       pk        683:        /*
                    684:         * Initialize the pool structure.
                    685:         */
1.88      chs       686:        LIST_INIT(&pp->pr_emptypages);
                    687:        LIST_INIT(&pp->pr_fullpages);
                    688:        LIST_INIT(&pp->pr_partpages);
1.134     ad        689:        pp->pr_cache = NULL;
1.3       pk        690:        pp->pr_curpage = NULL;
                    691:        pp->pr_npages = 0;
                    692:        pp->pr_minitems = 0;
                    693:        pp->pr_minpages = 0;
                    694:        pp->pr_maxpages = UINT_MAX;
1.20      thorpej   695:        pp->pr_roflags = flags;
                    696:        pp->pr_flags = 0;
1.35      pk        697:        pp->pr_size = size;
1.3       pk        698:        pp->pr_align = align;
                    699:        pp->pr_wchan = wchan;
1.66      thorpej   700:        pp->pr_alloc = palloc;
1.20      thorpej   701:        pp->pr_nitems = 0;
                    702:        pp->pr_nout = 0;
                    703:        pp->pr_hardlimit = UINT_MAX;
                    704:        pp->pr_hardlimit_warning = NULL;
1.31      thorpej   705:        pp->pr_hardlimit_ratecap.tv_sec = 0;
                    706:        pp->pr_hardlimit_ratecap.tv_usec = 0;
                    707:        pp->pr_hardlimit_warning_last.tv_sec = 0;
                    708:        pp->pr_hardlimit_warning_last.tv_usec = 0;
1.68      thorpej   709:        pp->pr_drain_hook = NULL;
                    710:        pp->pr_drain_hook_arg = NULL;
1.125     ad        711:        pp->pr_freecheck = NULL;
1.3       pk        712:
                    713:        /*
                    714:         * Decide whether to put the page header off page to avoid
1.92      enami     715:         * wasting too large a part of the page or too big item.
                    716:         * Off-page page headers go on a hash table, so we can match
                    717:         * a returned item with its header based on the page address.
                    718:         * We use 1/16 of the page size and about 8 times of the item
                    719:         * size as the threshold (XXX: tune)
                    720:         *
                    721:         * However, we'll put the header into the page if we can put
                    722:         * it without wasting any items.
                    723:         *
                    724:         * Silently enforce `0 <= ioff < align'.
1.3       pk        725:         */
1.92      enami     726:        pp->pr_itemoffset = ioff %= align;
                    727:        /* See the comment below about reserved bytes. */
                    728:        trysize = palloc->pa_pagesz - ((align - ioff) % align);
                    729:        phsize = ALIGN(sizeof(struct pool_item_header));
1.121     yamt      730:        if ((pp->pr_roflags & (PR_NOTOUCH | PR_NOALIGN)) == 0 &&
1.97      yamt      731:            (pp->pr_size < MIN(palloc->pa_pagesz / 16, phsize << 3) ||
                    732:            trysize / pp->pr_size == (trysize - phsize) / pp->pr_size)) {
1.3       pk        733:                /* Use the end of the page for the page header */
1.20      thorpej   734:                pp->pr_roflags |= PR_PHINPAGE;
1.92      enami     735:                pp->pr_phoffset = off = palloc->pa_pagesz - phsize;
1.2       pk        736:        } else {
1.3       pk        737:                /* The page header will be taken from our page header pool */
                    738:                pp->pr_phoffset = 0;
1.66      thorpej   739:                off = palloc->pa_pagesz;
1.88      chs       740:                SPLAY_INIT(&pp->pr_phtree);
1.2       pk        741:        }
1.1       pk        742:
1.3       pk        743:        /*
                    744:         * Alignment is to take place at `ioff' within the item. This means
                    745:         * we must reserve up to `align - 1' bytes on the page to allow
                    746:         * appropriate positioning of each item.
                    747:         */
                    748:        pp->pr_itemsperpage = (off - ((align - ioff) % align)) / pp->pr_size;
1.43      thorpej   749:        KASSERT(pp->pr_itemsperpage != 0);
1.97      yamt      750:        if ((pp->pr_roflags & PR_NOTOUCH)) {
                    751:                int idx;
                    752:
                    753:                for (idx = 0; pp->pr_itemsperpage > PHPOOL_FREELIST_NELEM(idx);
                    754:                    idx++) {
                    755:                        /* nothing */
                    756:                }
                    757:                if (idx >= PHPOOL_MAX) {
                    758:                        /*
                    759:                         * if you see this panic, consider to tweak
                    760:                         * PHPOOL_MAX and PHPOOL_FREELIST_NELEM.
                    761:                         */
                    762:                        panic("%s: too large itemsperpage(%d) for PR_NOTOUCH",
                    763:                            pp->pr_wchan, pp->pr_itemsperpage);
                    764:                }
                    765:                pp->pr_phpool = &phpool[idx];
                    766:        } else if ((pp->pr_roflags & PR_PHINPAGE) == 0) {
                    767:                pp->pr_phpool = &phpool[0];
                    768:        }
                    769: #if defined(DIAGNOSTIC)
                    770:        else {
                    771:                pp->pr_phpool = NULL;
                    772:        }
                    773: #endif
1.3       pk        774:
                    775:        /*
                    776:         * Use the slack between the chunks and the page header
                    777:         * for "cache coloring".
                    778:         */
                    779:        slack = off - pp->pr_itemsperpage * pp->pr_size;
                    780:        pp->pr_maxcolor = (slack / align) * align;
                    781:        pp->pr_curcolor = 0;
                    782:
                    783:        pp->pr_nget = 0;
                    784:        pp->pr_nfail = 0;
                    785:        pp->pr_nput = 0;
                    786:        pp->pr_npagealloc = 0;
                    787:        pp->pr_npagefree = 0;
1.1       pk        788:        pp->pr_hiwat = 0;
1.8       thorpej   789:        pp->pr_nidle = 0;
1.134     ad        790:        pp->pr_refcnt = 0;
1.3       pk        791:
1.59      thorpej   792: #ifdef POOL_DIAGNOSTIC
1.25      thorpej   793:        if (flags & PR_LOGGING) {
                    794:                if (kmem_map == NULL ||
                    795:                    (pp->pr_log = malloc(pool_logsize * sizeof(struct pool_log),
                    796:                     M_TEMP, M_NOWAIT)) == NULL)
1.20      thorpej   797:                        pp->pr_roflags &= ~PR_LOGGING;
1.3       pk        798:                pp->pr_curlogentry = 0;
                    799:                pp->pr_logsize = pool_logsize;
                    800:        }
1.59      thorpej   801: #endif
1.25      thorpej   802:
                    803:        pp->pr_entered_file = NULL;
                    804:        pp->pr_entered_line = 0;
1.3       pk        805:
1.137.2.1  ad        806:        /*
                    807:         * XXXAD hack to prevent IP input processing from blocking.
                    808:         */
                    809:        if (ipl == IPL_SOFTNET) {
                    810:                mutex_init(&pp->pr_lock, MUTEX_DEFAULT, IPL_VM);
                    811:        } else {
                    812:                mutex_init(&pp->pr_lock, MUTEX_DEFAULT, ipl);
                    813:        }
1.134     ad        814:        cv_init(&pp->pr_cv, wchan);
                    815:        pp->pr_ipl = ipl;
1.1       pk        816:
1.3       pk        817:        /*
1.43      thorpej   818:         * Initialize private page header pool and cache magazine pool if we
                    819:         * haven't done so yet.
1.23      thorpej   820:         * XXX LOCKING.
1.3       pk        821:         */
1.97      yamt      822:        if (phpool[0].pr_size == 0) {
                    823:                int idx;
                    824:                for (idx = 0; idx < PHPOOL_MAX; idx++) {
                    825:                        static char phpool_names[PHPOOL_MAX][6+1+6+1];
                    826:                        int nelem;
                    827:                        size_t sz;
                    828:
                    829:                        nelem = PHPOOL_FREELIST_NELEM(idx);
                    830:                        snprintf(phpool_names[idx], sizeof(phpool_names[idx]),
                    831:                            "phpool-%d", nelem);
                    832:                        sz = sizeof(struct pool_item_header);
                    833:                        if (nelem) {
1.135     yamt      834:                                sz = offsetof(struct pool_item_header,
                    835:                                    ph_bitmap[howmany(nelem, BITMAP_SIZE)]);
1.97      yamt      836:                        }
                    837:                        pool_init(&phpool[idx], sz, 0, 0, 0,
1.129     ad        838:                            phpool_names[idx], &pool_allocator_meta, IPL_VM);
1.97      yamt      839:                }
1.62      bjh21     840: #ifdef POOL_SUBPAGE
                    841:                pool_init(&psppool, POOL_SUBPAGE, POOL_SUBPAGE, 0,
1.129     ad        842:                    PR_RECURSIVE, "psppool", &pool_allocator_meta, IPL_VM);
1.62      bjh21     843: #endif
1.134     ad        844:                pool_init(&pcgpool, sizeof(pcg_t), CACHE_LINE_SIZE, 0, 0,
                    845:                    "cachegrp", &pool_allocator_meta, IPL_VM);
1.1       pk        846:        }
                    847:
1.137.2.3! ad        848:        /* Insert into the list of all pools. */
        !           849:        if (__predict_true(!cold))
1.134     ad        850:                mutex_enter(&pool_head_lock);
1.137.2.3! ad        851:        TAILQ_FOREACH(pp1, &pool_head, pr_poollist) {
        !           852:                if (strcmp(pp1->pr_wchan, pp->pr_wchan) > 0)
        !           853:                        break;
        !           854:        }
        !           855:        if (pp1 == NULL)
        !           856:                TAILQ_INSERT_TAIL(&pool_head, pp, pr_poollist);
        !           857:        else
        !           858:                TAILQ_INSERT_BEFORE(pp1, pp, pr_poollist);
        !           859:        if (__predict_true(!cold))
1.134     ad        860:                mutex_exit(&pool_head_lock);
                    861:
                    862:                /* Insert this into the list of pools using this allocator. */
1.137.2.3! ad        863:        if (__predict_true(!cold))
1.134     ad        864:                mutex_enter(&palloc->pa_lock);
1.137.2.3! ad        865:        TAILQ_INSERT_TAIL(&palloc->pa_list, pp, pr_alloc_list);
        !           866:        if (__predict_true(!cold))
1.134     ad        867:                mutex_exit(&palloc->pa_lock);
1.66      thorpej   868:
1.117     yamt      869:        pool_reclaim_register(pp);
1.1       pk        870: }
                    871:
                    872: /*
                    873:  * De-commision a pool resource.
                    874:  */
                    875: void
1.42      thorpej   876: pool_destroy(struct pool *pp)
1.1       pk        877: {
1.101     thorpej   878:        struct pool_pagelist pq;
1.3       pk        879:        struct pool_item_header *ph;
1.43      thorpej   880:
1.101     thorpej   881:        /* Remove from global pool list */
1.134     ad        882:        mutex_enter(&pool_head_lock);
                    883:        while (pp->pr_refcnt != 0)
                    884:                cv_wait(&pool_busy, &pool_head_lock);
1.137.2.3! ad        885:        TAILQ_REMOVE(&pool_head, pp, pr_poollist);
1.101     thorpej   886:        if (drainpp == pp)
                    887:                drainpp = NULL;
1.134     ad        888:        mutex_exit(&pool_head_lock);
1.101     thorpej   889:
                    890:        /* Remove this pool from its allocator's list of pools. */
1.117     yamt      891:        pool_reclaim_unregister(pp);
1.134     ad        892:        mutex_enter(&pp->pr_alloc->pa_lock);
1.66      thorpej   893:        TAILQ_REMOVE(&pp->pr_alloc->pa_list, pp, pr_alloc_list);
1.134     ad        894:        mutex_exit(&pp->pr_alloc->pa_lock);
1.66      thorpej   895:
1.134     ad        896:        mutex_enter(&pp->pr_lock);
1.101     thorpej   897:
1.134     ad        898:        KASSERT(pp->pr_cache == NULL);
1.3       pk        899:
                    900: #ifdef DIAGNOSTIC
1.20      thorpej   901:        if (pp->pr_nout != 0) {
1.25      thorpej   902:                pr_printlog(pp, NULL, printf);
1.80      provos    903:                panic("pool_destroy: pool busy: still out: %u",
1.20      thorpej   904:                    pp->pr_nout);
1.3       pk        905:        }
                    906: #endif
1.1       pk        907:
1.101     thorpej   908:        KASSERT(LIST_EMPTY(&pp->pr_fullpages));
                    909:        KASSERT(LIST_EMPTY(&pp->pr_partpages));
                    910:
1.3       pk        911:        /* Remove all pages */
1.101     thorpej   912:        LIST_INIT(&pq);
1.88      chs       913:        while ((ph = LIST_FIRST(&pp->pr_emptypages)) != NULL)
1.101     thorpej   914:                pr_rmpage(pp, ph, &pq);
                    915:
1.134     ad        916:        mutex_exit(&pp->pr_lock);
1.3       pk        917:
1.101     thorpej   918:        pr_pagelist_free(pp, &pq);
1.3       pk        919:
1.59      thorpej   920: #ifdef POOL_DIAGNOSTIC
1.20      thorpej   921:        if ((pp->pr_roflags & PR_LOGGING) != 0)
1.3       pk        922:                free(pp->pr_log, M_TEMP);
1.59      thorpej   923: #endif
1.134     ad        924:
                    925:        cv_destroy(&pp->pr_cv);
                    926:        mutex_destroy(&pp->pr_lock);
1.1       pk        927: }
                    928:
1.68      thorpej   929: void
                    930: pool_set_drain_hook(struct pool *pp, void (*fn)(void *, int), void *arg)
                    931: {
                    932:
                    933:        /* XXX no locking -- must be used just after pool_init() */
                    934: #ifdef DIAGNOSTIC
                    935:        if (pp->pr_drain_hook != NULL)
                    936:                panic("pool_set_drain_hook(%s): already set", pp->pr_wchan);
                    937: #endif
                    938:        pp->pr_drain_hook = fn;
                    939:        pp->pr_drain_hook_arg = arg;
                    940: }
                    941:
1.88      chs       942: static struct pool_item_header *
1.128     christos  943: pool_alloc_item_header(struct pool *pp, void *storage, int flags)
1.55      thorpej   944: {
                    945:        struct pool_item_header *ph;
                    946:
                    947:        if ((pp->pr_roflags & PR_PHINPAGE) != 0)
1.128     christos  948:                ph = (struct pool_item_header *) ((char *)storage + pp->pr_phoffset);
1.134     ad        949:        else
1.97      yamt      950:                ph = pool_get(pp->pr_phpool, flags);
1.55      thorpej   951:
                    952:        return (ph);
                    953: }
1.1       pk        954:
                    955: /*
1.134     ad        956:  * Grab an item from the pool.
1.1       pk        957:  */
1.3       pk        958: void *
1.59      thorpej   959: #ifdef POOL_DIAGNOSTIC
1.42      thorpej   960: _pool_get(struct pool *pp, int flags, const char *file, long line)
1.56      sommerfe  961: #else
                    962: pool_get(struct pool *pp, int flags)
                    963: #endif
1.1       pk        964: {
                    965:        struct pool_item *pi;
1.3       pk        966:        struct pool_item_header *ph;
1.55      thorpej   967:        void *v;
1.1       pk        968:
1.2       pk        969: #ifdef DIAGNOSTIC
1.95      atatat    970:        if (__predict_false(pp->pr_itemsperpage == 0))
                    971:                panic("pool_get: pool %p: pr_itemsperpage is zero, "
                    972:                    "pool not initialized?", pp);
1.84      thorpej   973:        if (__predict_false(curlwp == NULL && doing_shutdown == 0 &&
1.37      sommerfe  974:                            (flags & PR_WAITOK) != 0))
1.77      matt      975:                panic("pool_get: %s: must have NOWAIT", pp->pr_wchan);
1.58      thorpej   976:
1.102     chs       977: #endif /* DIAGNOSTIC */
1.58      thorpej   978: #ifdef LOCKDEBUG
                    979:        if (flags & PR_WAITOK)
1.119     yamt      980:                ASSERT_SLEEPABLE(NULL, "pool_get(PR_WAITOK)");
1.56      sommerfe  981: #endif
1.1       pk        982:
1.134     ad        983:        mutex_enter(&pp->pr_lock);
1.25      thorpej   984:        pr_enter(pp, file, line);
1.20      thorpej   985:
                    986:  startover:
                    987:        /*
                    988:         * Check to see if we've reached the hard limit.  If we have,
                    989:         * and we can wait, then wait until an item has been returned to
                    990:         * the pool.
                    991:         */
                    992: #ifdef DIAGNOSTIC
1.34      thorpej   993:        if (__predict_false(pp->pr_nout > pp->pr_hardlimit)) {
1.25      thorpej   994:                pr_leave(pp);
1.134     ad        995:                mutex_exit(&pp->pr_lock);
1.20      thorpej   996:                panic("pool_get: %s: crossed hard limit", pp->pr_wchan);
                    997:        }
                    998: #endif
1.34      thorpej   999:        if (__predict_false(pp->pr_nout == pp->pr_hardlimit)) {
1.68      thorpej  1000:                if (pp->pr_drain_hook != NULL) {
                   1001:                        /*
                   1002:                         * Since the drain hook is going to free things
                   1003:                         * back to the pool, unlock, call the hook, re-lock,
                   1004:                         * and check the hardlimit condition again.
                   1005:                         */
                   1006:                        pr_leave(pp);
1.134     ad       1007:                        mutex_exit(&pp->pr_lock);
1.68      thorpej  1008:                        (*pp->pr_drain_hook)(pp->pr_drain_hook_arg, flags);
1.134     ad       1009:                        mutex_enter(&pp->pr_lock);
1.68      thorpej  1010:                        pr_enter(pp, file, line);
                   1011:                        if (pp->pr_nout < pp->pr_hardlimit)
                   1012:                                goto startover;
                   1013:                }
                   1014:
1.29      sommerfe 1015:                if ((flags & PR_WAITOK) && !(flags & PR_LIMITFAIL)) {
1.20      thorpej  1016:                        /*
                   1017:                         * XXX: A warning isn't logged in this case.  Should
                   1018:                         * it be?
                   1019:                         */
                   1020:                        pp->pr_flags |= PR_WANTED;
1.25      thorpej  1021:                        pr_leave(pp);
1.134     ad       1022:                        cv_wait(&pp->pr_cv, &pp->pr_lock);
1.25      thorpej  1023:                        pr_enter(pp, file, line);
1.20      thorpej  1024:                        goto startover;
                   1025:                }
1.31      thorpej  1026:
                   1027:                /*
                   1028:                 * Log a message that the hard limit has been hit.
                   1029:                 */
                   1030:                if (pp->pr_hardlimit_warning != NULL &&
                   1031:                    ratecheck(&pp->pr_hardlimit_warning_last,
                   1032:                              &pp->pr_hardlimit_ratecap))
                   1033:                        log(LOG_ERR, "%s\n", pp->pr_hardlimit_warning);
1.21      thorpej  1034:
                   1035:                pp->pr_nfail++;
                   1036:
1.25      thorpej  1037:                pr_leave(pp);
1.134     ad       1038:                mutex_exit(&pp->pr_lock);
1.20      thorpej  1039:                return (NULL);
                   1040:        }
                   1041:
1.3       pk       1042:        /*
                   1043:         * The convention we use is that if `curpage' is not NULL, then
                   1044:         * it points at a non-empty bucket. In particular, `curpage'
                   1045:         * never points at a page header which has PR_PHINPAGE set and
                   1046:         * has no items in its bucket.
                   1047:         */
1.20      thorpej  1048:        if ((ph = pp->pr_curpage) == NULL) {
1.113     yamt     1049:                int error;
                   1050:
1.20      thorpej  1051: #ifdef DIAGNOSTIC
                   1052:                if (pp->pr_nitems != 0) {
1.134     ad       1053:                        mutex_exit(&pp->pr_lock);
1.20      thorpej  1054:                        printf("pool_get: %s: curpage NULL, nitems %u\n",
                   1055:                            pp->pr_wchan, pp->pr_nitems);
1.80      provos   1056:                        panic("pool_get: nitems inconsistent");
1.20      thorpej  1057:                }
                   1058: #endif
                   1059:
1.21      thorpej  1060:                /*
                   1061:                 * Call the back-end page allocator for more memory.
                   1062:                 * Release the pool lock, as the back-end page allocator
                   1063:                 * may block.
                   1064:                 */
1.25      thorpej  1065:                pr_leave(pp);
1.113     yamt     1066:                error = pool_grow(pp, flags);
                   1067:                pr_enter(pp, file, line);
                   1068:                if (error != 0) {
1.21      thorpej  1069:                        /*
1.55      thorpej  1070:                         * We were unable to allocate a page or item
                   1071:                         * header, but we released the lock during
                   1072:                         * allocation, so perhaps items were freed
                   1073:                         * back to the pool.  Check for this case.
1.21      thorpej  1074:                         */
                   1075:                        if (pp->pr_curpage != NULL)
                   1076:                                goto startover;
1.15      pk       1077:
1.117     yamt     1078:                        pp->pr_nfail++;
1.25      thorpej  1079:                        pr_leave(pp);
1.134     ad       1080:                        mutex_exit(&pp->pr_lock);
1.117     yamt     1081:                        return (NULL);
1.1       pk       1082:                }
1.3       pk       1083:
1.20      thorpej  1084:                /* Start the allocation process over. */
                   1085:                goto startover;
1.3       pk       1086:        }
1.97      yamt     1087:        if (pp->pr_roflags & PR_NOTOUCH) {
                   1088: #ifdef DIAGNOSTIC
                   1089:                if (__predict_false(ph->ph_nmissing == pp->pr_itemsperpage)) {
                   1090:                        pr_leave(pp);
1.134     ad       1091:                        mutex_exit(&pp->pr_lock);
1.97      yamt     1092:                        panic("pool_get: %s: page empty", pp->pr_wchan);
                   1093:                }
                   1094: #endif
                   1095:                v = pr_item_notouch_get(pp, ph);
                   1096: #ifdef POOL_DIAGNOSTIC
                   1097:                pr_log(pp, v, PRLOG_GET, file, line);
                   1098: #endif
                   1099:        } else {
1.102     chs      1100:                v = pi = LIST_FIRST(&ph->ph_itemlist);
1.97      yamt     1101:                if (__predict_false(v == NULL)) {
                   1102:                        pr_leave(pp);
1.134     ad       1103:                        mutex_exit(&pp->pr_lock);
1.97      yamt     1104:                        panic("pool_get: %s: page empty", pp->pr_wchan);
                   1105:                }
1.20      thorpej  1106: #ifdef DIAGNOSTIC
1.97      yamt     1107:                if (__predict_false(pp->pr_nitems == 0)) {
                   1108:                        pr_leave(pp);
1.134     ad       1109:                        mutex_exit(&pp->pr_lock);
1.97      yamt     1110:                        printf("pool_get: %s: items on itemlist, nitems %u\n",
                   1111:                            pp->pr_wchan, pp->pr_nitems);
                   1112:                        panic("pool_get: nitems inconsistent");
                   1113:                }
1.65      enami    1114: #endif
1.56      sommerfe 1115:
1.65      enami    1116: #ifdef POOL_DIAGNOSTIC
1.97      yamt     1117:                pr_log(pp, v, PRLOG_GET, file, line);
1.65      enami    1118: #endif
1.3       pk       1119:
1.65      enami    1120: #ifdef DIAGNOSTIC
1.97      yamt     1121:                if (__predict_false(pi->pi_magic != PI_MAGIC)) {
                   1122:                        pr_printlog(pp, pi, printf);
                   1123:                        panic("pool_get(%s): free list modified: "
                   1124:                            "magic=%x; page %p; item addr %p\n",
                   1125:                            pp->pr_wchan, pi->pi_magic, ph->ph_page, pi);
                   1126:                }
1.3       pk       1127: #endif
                   1128:
1.97      yamt     1129:                /*
                   1130:                 * Remove from item list.
                   1131:                 */
1.102     chs      1132:                LIST_REMOVE(pi, pi_list);
1.97      yamt     1133:        }
1.20      thorpej  1134:        pp->pr_nitems--;
                   1135:        pp->pr_nout++;
1.6       thorpej  1136:        if (ph->ph_nmissing == 0) {
                   1137: #ifdef DIAGNOSTIC
1.34      thorpej  1138:                if (__predict_false(pp->pr_nidle == 0))
1.6       thorpej  1139:                        panic("pool_get: nidle inconsistent");
                   1140: #endif
                   1141:                pp->pr_nidle--;
1.88      chs      1142:
                   1143:                /*
                   1144:                 * This page was previously empty.  Move it to the list of
                   1145:                 * partially-full pages.  This page is already curpage.
                   1146:                 */
                   1147:                LIST_REMOVE(ph, ph_pagelist);
                   1148:                LIST_INSERT_HEAD(&pp->pr_partpages, ph, ph_pagelist);
1.6       thorpej  1149:        }
1.3       pk       1150:        ph->ph_nmissing++;
1.97      yamt     1151:        if (ph->ph_nmissing == pp->pr_itemsperpage) {
1.21      thorpej  1152: #ifdef DIAGNOSTIC
1.97      yamt     1153:                if (__predict_false((pp->pr_roflags & PR_NOTOUCH) == 0 &&
1.102     chs      1154:                    !LIST_EMPTY(&ph->ph_itemlist))) {
1.25      thorpej  1155:                        pr_leave(pp);
1.134     ad       1156:                        mutex_exit(&pp->pr_lock);
1.21      thorpej  1157:                        panic("pool_get: %s: nmissing inconsistent",
                   1158:                            pp->pr_wchan);
                   1159:                }
                   1160: #endif
1.3       pk       1161:                /*
1.88      chs      1162:                 * This page is now full.  Move it to the full list
                   1163:                 * and select a new current page.
1.3       pk       1164:                 */
1.88      chs      1165:                LIST_REMOVE(ph, ph_pagelist);
                   1166:                LIST_INSERT_HEAD(&pp->pr_fullpages, ph, ph_pagelist);
                   1167:                pool_update_curpage(pp);
1.1       pk       1168:        }
1.3       pk       1169:
                   1170:        pp->pr_nget++;
1.111     christos 1171:        pr_leave(pp);
1.20      thorpej  1172:
                   1173:        /*
                   1174:         * If we have a low water mark and we are now below that low
                   1175:         * water mark, add more items to the pool.
                   1176:         */
1.53      thorpej  1177:        if (POOL_NEEDS_CATCHUP(pp) && pool_catchup(pp) != 0) {
1.20      thorpej  1178:                /*
                   1179:                 * XXX: Should we log a warning?  Should we set up a timeout
                   1180:                 * to try again in a second or so?  The latter could break
                   1181:                 * a caller's assumptions about interrupt protection, etc.
                   1182:                 */
                   1183:        }
                   1184:
1.134     ad       1185:        mutex_exit(&pp->pr_lock);
1.125     ad       1186:        KASSERT((((vaddr_t)v + pp->pr_itemoffset) & (pp->pr_align - 1)) == 0);
                   1187:        FREECHECK_OUT(&pp->pr_freecheck, v);
1.1       pk       1188:        return (v);
                   1189: }
                   1190:
                   1191: /*
1.43      thorpej  1192:  * Internal version of pool_put().  Pool is already locked/entered.
1.1       pk       1193:  */
1.43      thorpej  1194: static void
1.101     thorpej  1195: pool_do_put(struct pool *pp, void *v, struct pool_pagelist *pq)
1.1       pk       1196: {
                   1197:        struct pool_item *pi = v;
1.3       pk       1198:        struct pool_item_header *ph;
                   1199:
1.134     ad       1200:        KASSERT(mutex_owned(&pp->pr_lock));
1.125     ad       1201:        FREECHECK_IN(&pp->pr_freecheck, v);
1.134     ad       1202:        LOCKDEBUG_MEM_CHECK(v, pp->pr_size);
1.61      chs      1203:
1.30      thorpej  1204: #ifdef DIAGNOSTIC
1.34      thorpej  1205:        if (__predict_false(pp->pr_nout == 0)) {
1.30      thorpej  1206:                printf("pool %s: putting with none out\n",
                   1207:                    pp->pr_wchan);
                   1208:                panic("pool_put");
                   1209:        }
                   1210: #endif
1.3       pk       1211:
1.121     yamt     1212:        if (__predict_false((ph = pr_find_pagehead(pp, v)) == NULL)) {
1.25      thorpej  1213:                pr_printlog(pp, NULL, printf);
1.3       pk       1214:                panic("pool_put: %s: page header missing", pp->pr_wchan);
                   1215:        }
1.28      thorpej  1216:
1.3       pk       1217:        /*
                   1218:         * Return to item list.
                   1219:         */
1.97      yamt     1220:        if (pp->pr_roflags & PR_NOTOUCH) {
                   1221:                pr_item_notouch_put(pp, ph, v);
                   1222:        } else {
1.2       pk       1223: #ifdef DIAGNOSTIC
1.97      yamt     1224:                pi->pi_magic = PI_MAGIC;
1.3       pk       1225: #endif
1.32      chs      1226: #ifdef DEBUG
1.97      yamt     1227:                {
                   1228:                        int i, *ip = v;
1.32      chs      1229:
1.97      yamt     1230:                        for (i = 0; i < pp->pr_size / sizeof(int); i++) {
                   1231:                                *ip++ = PI_MAGIC;
                   1232:                        }
1.32      chs      1233:                }
                   1234: #endif
                   1235:
1.102     chs      1236:                LIST_INSERT_HEAD(&ph->ph_itemlist, pi, pi_list);
1.97      yamt     1237:        }
1.79      thorpej  1238:        KDASSERT(ph->ph_nmissing != 0);
1.3       pk       1239:        ph->ph_nmissing--;
                   1240:        pp->pr_nput++;
1.20      thorpej  1241:        pp->pr_nitems++;
                   1242:        pp->pr_nout--;
1.3       pk       1243:
                   1244:        /* Cancel "pool empty" condition if it exists */
                   1245:        if (pp->pr_curpage == NULL)
                   1246:                pp->pr_curpage = ph;
                   1247:
                   1248:        if (pp->pr_flags & PR_WANTED) {
                   1249:                pp->pr_flags &= ~PR_WANTED;
1.15      pk       1250:                if (ph->ph_nmissing == 0)
                   1251:                        pp->pr_nidle++;
1.134     ad       1252:                cv_broadcast(&pp->pr_cv);
1.3       pk       1253:                return;
                   1254:        }
                   1255:
                   1256:        /*
1.88      chs      1257:         * If this page is now empty, do one of two things:
1.21      thorpej  1258:         *
1.88      chs      1259:         *      (1) If we have more pages than the page high water mark,
1.96      thorpej  1260:         *          free the page back to the system.  ONLY CONSIDER
1.90      thorpej  1261:         *          FREEING BACK A PAGE IF WE HAVE MORE THAN OUR MINIMUM PAGE
                   1262:         *          CLAIM.
1.21      thorpej  1263:         *
1.88      chs      1264:         *      (2) Otherwise, move the page to the empty page list.
                   1265:         *
                   1266:         * Either way, select a new current page (so we use a partially-full
                   1267:         * page if one is available).
1.3       pk       1268:         */
                   1269:        if (ph->ph_nmissing == 0) {
1.6       thorpej  1270:                pp->pr_nidle++;
1.90      thorpej  1271:                if (pp->pr_npages > pp->pr_minpages &&
                   1272:                    (pp->pr_npages > pp->pr_maxpages ||
1.117     yamt     1273:                     pa_starved_p(pp->pr_alloc))) {
1.101     thorpej  1274:                        pr_rmpage(pp, ph, pq);
1.3       pk       1275:                } else {
1.88      chs      1276:                        LIST_REMOVE(ph, ph_pagelist);
                   1277:                        LIST_INSERT_HEAD(&pp->pr_emptypages, ph, ph_pagelist);
1.3       pk       1278:
1.21      thorpej  1279:                        /*
                   1280:                         * Update the timestamp on the page.  A page must
                   1281:                         * be idle for some period of time before it can
                   1282:                         * be reclaimed by the pagedaemon.  This minimizes
                   1283:                         * ping-pong'ing for memory.
                   1284:                         */
1.118     kardel   1285:                        getmicrotime(&ph->ph_time);
1.1       pk       1286:                }
1.88      chs      1287:                pool_update_curpage(pp);
1.1       pk       1288:        }
1.88      chs      1289:
1.21      thorpej  1290:        /*
1.88      chs      1291:         * If the page was previously completely full, move it to the
                   1292:         * partially-full list and make it the current page.  The next
                   1293:         * allocation will get the item from this page, instead of
                   1294:         * further fragmenting the pool.
1.21      thorpej  1295:         */
                   1296:        else if (ph->ph_nmissing == (pp->pr_itemsperpage - 1)) {
1.88      chs      1297:                LIST_REMOVE(ph, ph_pagelist);
                   1298:                LIST_INSERT_HEAD(&pp->pr_partpages, ph, ph_pagelist);
1.21      thorpej  1299:                pp->pr_curpage = ph;
                   1300:        }
1.43      thorpej  1301: }
                   1302:
                   1303: /*
1.134     ad       1304:  * Return resource to the pool.
1.43      thorpej  1305:  */
1.59      thorpej  1306: #ifdef POOL_DIAGNOSTIC
1.43      thorpej  1307: void
                   1308: _pool_put(struct pool *pp, void *v, const char *file, long line)
                   1309: {
1.101     thorpej  1310:        struct pool_pagelist pq;
                   1311:
                   1312:        LIST_INIT(&pq);
1.43      thorpej  1313:
1.134     ad       1314:        mutex_enter(&pp->pr_lock);
1.43      thorpej  1315:        pr_enter(pp, file, line);
                   1316:
1.56      sommerfe 1317:        pr_log(pp, v, PRLOG_PUT, file, line);
                   1318:
1.101     thorpej  1319:        pool_do_put(pp, v, &pq);
1.21      thorpej  1320:
1.25      thorpej  1321:        pr_leave(pp);
1.134     ad       1322:        mutex_exit(&pp->pr_lock);
1.101     thorpej  1323:
1.102     chs      1324:        pr_pagelist_free(pp, &pq);
1.1       pk       1325: }
1.57      sommerfe 1326: #undef pool_put
1.59      thorpej  1327: #endif /* POOL_DIAGNOSTIC */
1.1       pk       1328:
1.56      sommerfe 1329: void
                   1330: pool_put(struct pool *pp, void *v)
                   1331: {
1.101     thorpej  1332:        struct pool_pagelist pq;
                   1333:
                   1334:        LIST_INIT(&pq);
1.56      sommerfe 1335:
1.134     ad       1336:        mutex_enter(&pp->pr_lock);
1.101     thorpej  1337:        pool_do_put(pp, v, &pq);
1.134     ad       1338:        mutex_exit(&pp->pr_lock);
1.56      sommerfe 1339:
1.102     chs      1340:        pr_pagelist_free(pp, &pq);
1.56      sommerfe 1341: }
1.57      sommerfe 1342:
1.59      thorpej  1343: #ifdef POOL_DIAGNOSTIC
1.57      sommerfe 1344: #define                pool_put(h, v)  _pool_put((h), (v), __FILE__, __LINE__)
1.56      sommerfe 1345: #endif
1.74      thorpej  1346:
                   1347: /*
1.113     yamt     1348:  * pool_grow: grow a pool by a page.
                   1349:  *
                   1350:  * => called with pool locked.
                   1351:  * => unlock and relock the pool.
                   1352:  * => return with pool locked.
                   1353:  */
                   1354:
                   1355: static int
                   1356: pool_grow(struct pool *pp, int flags)
                   1357: {
                   1358:        struct pool_item_header *ph = NULL;
                   1359:        char *cp;
                   1360:
1.134     ad       1361:        mutex_exit(&pp->pr_lock);
1.113     yamt     1362:        cp = pool_allocator_alloc(pp, flags);
                   1363:        if (__predict_true(cp != NULL)) {
                   1364:                ph = pool_alloc_item_header(pp, cp, flags);
                   1365:        }
                   1366:        if (__predict_false(cp == NULL || ph == NULL)) {
                   1367:                if (cp != NULL) {
                   1368:                        pool_allocator_free(pp, cp);
                   1369:                }
1.134     ad       1370:                mutex_enter(&pp->pr_lock);
1.113     yamt     1371:                return ENOMEM;
                   1372:        }
                   1373:
1.134     ad       1374:        mutex_enter(&pp->pr_lock);
1.113     yamt     1375:        pool_prime_page(pp, cp, ph);
                   1376:        pp->pr_npagealloc++;
                   1377:        return 0;
                   1378: }
                   1379:
                   1380: /*
1.74      thorpej  1381:  * Add N items to the pool.
                   1382:  */
                   1383: int
                   1384: pool_prime(struct pool *pp, int n)
                   1385: {
1.75      simonb   1386:        int newpages;
1.113     yamt     1387:        int error = 0;
1.74      thorpej  1388:
1.134     ad       1389:        mutex_enter(&pp->pr_lock);
1.74      thorpej  1390:
                   1391:        newpages = roundup(n, pp->pr_itemsperpage) / pp->pr_itemsperpage;
                   1392:
                   1393:        while (newpages-- > 0) {
1.113     yamt     1394:                error = pool_grow(pp, PR_NOWAIT);
                   1395:                if (error) {
1.74      thorpej  1396:                        break;
                   1397:                }
                   1398:                pp->pr_minpages++;
                   1399:        }
                   1400:
                   1401:        if (pp->pr_minpages >= pp->pr_maxpages)
                   1402:                pp->pr_maxpages = pp->pr_minpages + 1;  /* XXX */
                   1403:
1.134     ad       1404:        mutex_exit(&pp->pr_lock);
1.113     yamt     1405:        return error;
1.74      thorpej  1406: }
1.55      thorpej  1407:
                   1408: /*
1.3       pk       1409:  * Add a page worth of items to the pool.
1.21      thorpej  1410:  *
                   1411:  * Note, we must be called with the pool descriptor LOCKED.
1.3       pk       1412:  */
1.55      thorpej  1413: static void
1.128     christos 1414: pool_prime_page(struct pool *pp, void *storage, struct pool_item_header *ph)
1.3       pk       1415: {
                   1416:        struct pool_item *pi;
1.128     christos 1417:        void *cp = storage;
1.125     ad       1418:        const unsigned int align = pp->pr_align;
                   1419:        const unsigned int ioff = pp->pr_itemoffset;
1.55      thorpej  1420:        int n;
1.36      pk       1421:
1.134     ad       1422:        KASSERT(mutex_owned(&pp->pr_lock));
1.91      yamt     1423:
1.66      thorpej  1424: #ifdef DIAGNOSTIC
1.121     yamt     1425:        if ((pp->pr_roflags & PR_NOALIGN) == 0 &&
                   1426:            ((uintptr_t)cp & (pp->pr_alloc->pa_pagesz - 1)) != 0)
1.36      pk       1427:                panic("pool_prime_page: %s: unaligned page", pp->pr_wchan);
1.66      thorpej  1428: #endif
1.3       pk       1429:
                   1430:        /*
                   1431:         * Insert page header.
                   1432:         */
1.88      chs      1433:        LIST_INSERT_HEAD(&pp->pr_emptypages, ph, ph_pagelist);
1.102     chs      1434:        LIST_INIT(&ph->ph_itemlist);
1.3       pk       1435:        ph->ph_page = storage;
                   1436:        ph->ph_nmissing = 0;
1.118     kardel   1437:        getmicrotime(&ph->ph_time);
1.88      chs      1438:        if ((pp->pr_roflags & PR_PHINPAGE) == 0)
                   1439:                SPLAY_INSERT(phtree, &pp->pr_phtree, ph);
1.3       pk       1440:
1.6       thorpej  1441:        pp->pr_nidle++;
                   1442:
1.3       pk       1443:        /*
                   1444:         * Color this page.
                   1445:         */
1.128     christos 1446:        cp = (char *)cp + pp->pr_curcolor;
1.3       pk       1447:        if ((pp->pr_curcolor += align) > pp->pr_maxcolor)
                   1448:                pp->pr_curcolor = 0;
                   1449:
                   1450:        /*
                   1451:         * Adjust storage to apply aligment to `pr_itemoffset' in each item.
                   1452:         */
                   1453:        if (ioff != 0)
1.128     christos 1454:                cp = (char *)cp + align - ioff;
1.3       pk       1455:
1.125     ad       1456:        KASSERT((((vaddr_t)cp + ioff) & (align - 1)) == 0);
                   1457:
1.3       pk       1458:        /*
                   1459:         * Insert remaining chunks on the bucket list.
                   1460:         */
                   1461:        n = pp->pr_itemsperpage;
1.20      thorpej  1462:        pp->pr_nitems += n;
1.3       pk       1463:
1.97      yamt     1464:        if (pp->pr_roflags & PR_NOTOUCH) {
1.135     yamt     1465:                pr_item_notouch_init(pp, ph);
1.97      yamt     1466:        } else {
                   1467:                while (n--) {
                   1468:                        pi = (struct pool_item *)cp;
1.78      thorpej  1469:
1.97      yamt     1470:                        KASSERT(((((vaddr_t)pi) + ioff) & (align - 1)) == 0);
1.3       pk       1471:
1.97      yamt     1472:                        /* Insert on page list */
1.102     chs      1473:                        LIST_INSERT_HEAD(&ph->ph_itemlist, pi, pi_list);
1.3       pk       1474: #ifdef DIAGNOSTIC
1.97      yamt     1475:                        pi->pi_magic = PI_MAGIC;
1.3       pk       1476: #endif
1.128     christos 1477:                        cp = (char *)cp + pp->pr_size;
1.125     ad       1478:
                   1479:                        KASSERT((((vaddr_t)cp + ioff) & (align - 1)) == 0);
1.97      yamt     1480:                }
1.3       pk       1481:        }
                   1482:
                   1483:        /*
                   1484:         * If the pool was depleted, point at the new page.
                   1485:         */
                   1486:        if (pp->pr_curpage == NULL)
                   1487:                pp->pr_curpage = ph;
                   1488:
                   1489:        if (++pp->pr_npages > pp->pr_hiwat)
                   1490:                pp->pr_hiwat = pp->pr_npages;
                   1491: }
                   1492:
1.20      thorpej  1493: /*
1.52      thorpej  1494:  * Used by pool_get() when nitems drops below the low water mark.  This
1.88      chs      1495:  * is used to catch up pr_nitems with the low water mark.
1.20      thorpej  1496:  *
1.21      thorpej  1497:  * Note 1, we never wait for memory here, we let the caller decide what to do.
1.20      thorpej  1498:  *
1.73      thorpej  1499:  * Note 2, we must be called with the pool already locked, and we return
1.20      thorpej  1500:  * with it locked.
                   1501:  */
                   1502: static int
1.42      thorpej  1503: pool_catchup(struct pool *pp)
1.20      thorpej  1504: {
                   1505:        int error = 0;
                   1506:
1.54      thorpej  1507:        while (POOL_NEEDS_CATCHUP(pp)) {
1.113     yamt     1508:                error = pool_grow(pp, PR_NOWAIT);
                   1509:                if (error) {
1.20      thorpej  1510:                        break;
                   1511:                }
                   1512:        }
1.113     yamt     1513:        return error;
1.20      thorpej  1514: }
                   1515:
1.88      chs      1516: static void
                   1517: pool_update_curpage(struct pool *pp)
                   1518: {
                   1519:
                   1520:        pp->pr_curpage = LIST_FIRST(&pp->pr_partpages);
                   1521:        if (pp->pr_curpage == NULL) {
                   1522:                pp->pr_curpage = LIST_FIRST(&pp->pr_emptypages);
                   1523:        }
                   1524: }
                   1525:
1.3       pk       1526: void
1.42      thorpej  1527: pool_setlowat(struct pool *pp, int n)
1.3       pk       1528: {
1.15      pk       1529:
1.134     ad       1530:        mutex_enter(&pp->pr_lock);
1.21      thorpej  1531:
1.3       pk       1532:        pp->pr_minitems = n;
1.15      pk       1533:        pp->pr_minpages = (n == 0)
                   1534:                ? 0
1.18      thorpej  1535:                : roundup(n, pp->pr_itemsperpage) / pp->pr_itemsperpage;
1.20      thorpej  1536:
                   1537:        /* Make sure we're caught up with the newly-set low water mark. */
1.75      simonb   1538:        if (POOL_NEEDS_CATCHUP(pp) && pool_catchup(pp) != 0) {
1.20      thorpej  1539:                /*
                   1540:                 * XXX: Should we log a warning?  Should we set up a timeout
                   1541:                 * to try again in a second or so?  The latter could break
                   1542:                 * a caller's assumptions about interrupt protection, etc.
                   1543:                 */
                   1544:        }
1.21      thorpej  1545:
1.134     ad       1546:        mutex_exit(&pp->pr_lock);
1.3       pk       1547: }
                   1548:
                   1549: void
1.42      thorpej  1550: pool_sethiwat(struct pool *pp, int n)
1.3       pk       1551: {
1.15      pk       1552:
1.134     ad       1553:        mutex_enter(&pp->pr_lock);
1.21      thorpej  1554:
1.15      pk       1555:        pp->pr_maxpages = (n == 0)
                   1556:                ? 0
1.18      thorpej  1557:                : roundup(n, pp->pr_itemsperpage) / pp->pr_itemsperpage;
1.21      thorpej  1558:
1.134     ad       1559:        mutex_exit(&pp->pr_lock);
1.3       pk       1560: }
                   1561:
1.20      thorpej  1562: void
1.42      thorpej  1563: pool_sethardlimit(struct pool *pp, int n, const char *warnmess, int ratecap)
1.20      thorpej  1564: {
                   1565:
1.134     ad       1566:        mutex_enter(&pp->pr_lock);
1.20      thorpej  1567:
                   1568:        pp->pr_hardlimit = n;
                   1569:        pp->pr_hardlimit_warning = warnmess;
1.31      thorpej  1570:        pp->pr_hardlimit_ratecap.tv_sec = ratecap;
                   1571:        pp->pr_hardlimit_warning_last.tv_sec = 0;
                   1572:        pp->pr_hardlimit_warning_last.tv_usec = 0;
1.20      thorpej  1573:
                   1574:        /*
1.21      thorpej  1575:         * In-line version of pool_sethiwat(), because we don't want to
                   1576:         * release the lock.
1.20      thorpej  1577:         */
                   1578:        pp->pr_maxpages = (n == 0)
                   1579:                ? 0
                   1580:                : roundup(n, pp->pr_itemsperpage) / pp->pr_itemsperpage;
1.21      thorpej  1581:
1.134     ad       1582:        mutex_exit(&pp->pr_lock);
1.20      thorpej  1583: }
1.3       pk       1584:
                   1585: /*
                   1586:  * Release all complete pages that have not been used recently.
                   1587:  */
1.66      thorpej  1588: int
1.59      thorpej  1589: #ifdef POOL_DIAGNOSTIC
1.42      thorpej  1590: _pool_reclaim(struct pool *pp, const char *file, long line)
1.56      sommerfe 1591: #else
                   1592: pool_reclaim(struct pool *pp)
                   1593: #endif
1.3       pk       1594: {
                   1595:        struct pool_item_header *ph, *phnext;
1.61      chs      1596:        struct pool_pagelist pq;
1.102     chs      1597:        struct timeval curtime, diff;
1.134     ad       1598:        bool klock;
                   1599:        int rv;
1.3       pk       1600:
1.68      thorpej  1601:        if (pp->pr_drain_hook != NULL) {
                   1602:                /*
                   1603:                 * The drain hook must be called with the pool unlocked.
                   1604:                 */
                   1605:                (*pp->pr_drain_hook)(pp->pr_drain_hook_arg, PR_NOWAIT);
                   1606:        }
                   1607:
1.134     ad       1608:        /*
                   1609:         * XXXSMP Because mutexes at IPL_SOFTXXX are still spinlocks,
                   1610:         * and we are called from the pagedaemon without kernel_lock.
                   1611:         * Does not apply to IPL_SOFTBIO.
                   1612:         */
                   1613:        if (pp->pr_ipl == IPL_SOFTNET || pp->pr_ipl == IPL_SOFTCLOCK ||
                   1614:            pp->pr_ipl == IPL_SOFTSERIAL) {
                   1615:                KERNEL_LOCK(1, NULL);
                   1616:                klock = true;
                   1617:        } else
                   1618:                klock = false;
                   1619:
                   1620:        /* Reclaim items from the pool's cache (if any). */
                   1621:        if (pp->pr_cache != NULL)
                   1622:                pool_cache_invalidate(pp->pr_cache);
                   1623:
                   1624:        if (mutex_tryenter(&pp->pr_lock) == 0) {
                   1625:                if (klock) {
                   1626:                        KERNEL_UNLOCK_ONE(NULL);
                   1627:                }
1.66      thorpej  1628:                return (0);
1.134     ad       1629:        }
1.25      thorpej  1630:        pr_enter(pp, file, line);
1.68      thorpej  1631:
1.88      chs      1632:        LIST_INIT(&pq);
1.43      thorpej  1633:
1.118     kardel   1634:        getmicrotime(&curtime);
1.21      thorpej  1635:
1.88      chs      1636:        for (ph = LIST_FIRST(&pp->pr_emptypages); ph != NULL; ph = phnext) {
                   1637:                phnext = LIST_NEXT(ph, ph_pagelist);
1.3       pk       1638:
                   1639:                /* Check our minimum page claim */
                   1640:                if (pp->pr_npages <= pp->pr_minpages)
                   1641:                        break;
                   1642:
1.88      chs      1643:                KASSERT(ph->ph_nmissing == 0);
                   1644:                timersub(&curtime, &ph->ph_time, &diff);
1.117     yamt     1645:                if (diff.tv_sec < pool_inactive_time
                   1646:                    && !pa_starved_p(pp->pr_alloc))
1.88      chs      1647:                        continue;
1.21      thorpej  1648:
1.88      chs      1649:                /*
                   1650:                 * If freeing this page would put us below
                   1651:                 * the low water mark, stop now.
                   1652:                 */
                   1653:                if ((pp->pr_nitems - pp->pr_itemsperpage) <
                   1654:                    pp->pr_minitems)
                   1655:                        break;
1.21      thorpej  1656:
1.88      chs      1657:                pr_rmpage(pp, ph, &pq);
1.3       pk       1658:        }
                   1659:
1.25      thorpej  1660:        pr_leave(pp);
1.134     ad       1661:        mutex_exit(&pp->pr_lock);
                   1662:
                   1663:        if (LIST_EMPTY(&pq))
                   1664:                rv = 0;
                   1665:        else {
                   1666:                pr_pagelist_free(pp, &pq);
                   1667:                rv = 1;
                   1668:        }
                   1669:
                   1670:        if (klock) {
                   1671:                KERNEL_UNLOCK_ONE(NULL);
                   1672:        }
1.66      thorpej  1673:
1.134     ad       1674:        return (rv);
1.3       pk       1675: }
                   1676:
                   1677: /*
1.134     ad       1678:  * Drain pools, one at a time.  This is a two stage process;
                   1679:  * drain_start kicks off a cross call to drain CPU-level caches
                   1680:  * if the pool has an associated pool_cache.  drain_end waits
                   1681:  * for those cross calls to finish, and then drains the cache
                   1682:  * (if any) and pool.
1.131     ad       1683:  *
1.134     ad       1684:  * Note, must never be called from interrupt context.
1.3       pk       1685:  */
                   1686: void
1.134     ad       1687: pool_drain_start(struct pool **ppp, uint64_t *wp)
1.3       pk       1688: {
                   1689:        struct pool *pp;
1.134     ad       1690:
1.137.2.3! ad       1691:        KASSERT(!TAILQ_EMPTY(&pool_head));
1.3       pk       1692:
1.61      chs      1693:        pp = NULL;
1.134     ad       1694:
                   1695:        /* Find next pool to drain, and add a reference. */
                   1696:        mutex_enter(&pool_head_lock);
                   1697:        do {
                   1698:                if (drainpp == NULL) {
1.137.2.3! ad       1699:                        drainpp = TAILQ_FIRST(&pool_head);
1.134     ad       1700:                }
                   1701:                if (drainpp != NULL) {
                   1702:                        pp = drainpp;
1.137.2.3! ad       1703:                        drainpp = TAILQ_NEXT(pp, pr_poollist);
1.134     ad       1704:                }
                   1705:                /*
                   1706:                 * Skip completely idle pools.  We depend on at least
                   1707:                 * one pool in the system being active.
                   1708:                 */
                   1709:        } while (pp == NULL || pp->pr_npages == 0);
                   1710:        pp->pr_refcnt++;
                   1711:        mutex_exit(&pool_head_lock);
                   1712:
                   1713:        /* If there is a pool_cache, drain CPU level caches. */
                   1714:        *ppp = pp;
                   1715:        if (pp->pr_cache != NULL) {
                   1716:                *wp = xc_broadcast(0, (xcfunc_t)pool_cache_xcall,
                   1717:                    pp->pr_cache, NULL);
                   1718:        }
                   1719: }
                   1720:
                   1721: void
                   1722: pool_drain_end(struct pool *pp, uint64_t where)
                   1723: {
                   1724:
                   1725:        if (pp == NULL)
                   1726:                return;
                   1727:
                   1728:        KASSERT(pp->pr_refcnt > 0);
                   1729:
                   1730:        /* Wait for remote draining to complete. */
                   1731:        if (pp->pr_cache != NULL)
                   1732:                xc_wait(where);
                   1733:
                   1734:        /* Drain the cache (if any) and pool.. */
                   1735:        pool_reclaim(pp);
                   1736:
                   1737:        /* Finally, unlock the pool. */
                   1738:        mutex_enter(&pool_head_lock);
                   1739:        pp->pr_refcnt--;
                   1740:        cv_broadcast(&pool_busy);
                   1741:        mutex_exit(&pool_head_lock);
1.3       pk       1742: }
                   1743:
                   1744: /*
                   1745:  * Diagnostic helpers.
                   1746:  */
                   1747: void
1.42      thorpej  1748: pool_print(struct pool *pp, const char *modif)
1.21      thorpej  1749: {
                   1750:
1.25      thorpej  1751:        pool_print1(pp, modif, printf);
1.21      thorpej  1752: }
                   1753:
1.25      thorpej  1754: void
1.108     yamt     1755: pool_printall(const char *modif, void (*pr)(const char *, ...))
                   1756: {
                   1757:        struct pool *pp;
                   1758:
1.137.2.3! ad       1759:        TAILQ_FOREACH(pp, &pool_head, pr_poollist) {
1.108     yamt     1760:                pool_printit(pp, modif, pr);
                   1761:        }
                   1762: }
                   1763:
                   1764: void
1.42      thorpej  1765: pool_printit(struct pool *pp, const char *modif, void (*pr)(const char *, ...))
1.25      thorpej  1766: {
                   1767:
                   1768:        if (pp == NULL) {
                   1769:                (*pr)("Must specify a pool to print.\n");
                   1770:                return;
                   1771:        }
                   1772:
                   1773:        pool_print1(pp, modif, pr);
                   1774: }
                   1775:
1.21      thorpej  1776: static void
1.124     yamt     1777: pool_print_pagelist(struct pool *pp, struct pool_pagelist *pl,
1.97      yamt     1778:     void (*pr)(const char *, ...))
1.88      chs      1779: {
                   1780:        struct pool_item_header *ph;
                   1781: #ifdef DIAGNOSTIC
                   1782:        struct pool_item *pi;
                   1783: #endif
                   1784:
                   1785:        LIST_FOREACH(ph, pl, ph_pagelist) {
                   1786:                (*pr)("\t\tpage %p, nmissing %d, time %lu,%lu\n",
                   1787:                    ph->ph_page, ph->ph_nmissing,
                   1788:                    (u_long)ph->ph_time.tv_sec,
                   1789:                    (u_long)ph->ph_time.tv_usec);
                   1790: #ifdef DIAGNOSTIC
1.97      yamt     1791:                if (!(pp->pr_roflags & PR_NOTOUCH)) {
1.102     chs      1792:                        LIST_FOREACH(pi, &ph->ph_itemlist, pi_list) {
1.97      yamt     1793:                                if (pi->pi_magic != PI_MAGIC) {
                   1794:                                        (*pr)("\t\t\titem %p, magic 0x%x\n",
                   1795:                                            pi, pi->pi_magic);
                   1796:                                }
1.88      chs      1797:                        }
                   1798:                }
                   1799: #endif
                   1800:        }
                   1801: }
                   1802:
                   1803: static void
1.42      thorpej  1804: pool_print1(struct pool *pp, const char *modif, void (*pr)(const char *, ...))
1.3       pk       1805: {
1.25      thorpej  1806:        struct pool_item_header *ph;
1.134     ad       1807:        pool_cache_t pc;
                   1808:        pcg_t *pcg;
                   1809:        pool_cache_cpu_t *cc;
                   1810:        uint64_t cpuhit, cpumiss;
1.44      thorpej  1811:        int i, print_log = 0, print_pagelist = 0, print_cache = 0;
1.25      thorpej  1812:        char c;
                   1813:
                   1814:        while ((c = *modif++) != '\0') {
                   1815:                if (c == 'l')
                   1816:                        print_log = 1;
                   1817:                if (c == 'p')
                   1818:                        print_pagelist = 1;
1.44      thorpej  1819:                if (c == 'c')
                   1820:                        print_cache = 1;
1.25      thorpej  1821:        }
                   1822:
1.134     ad       1823:        if ((pc = pp->pr_cache) != NULL) {
                   1824:                (*pr)("POOL CACHE");
                   1825:        } else {
                   1826:                (*pr)("POOL");
                   1827:        }
                   1828:
                   1829:        (*pr)(" %s: size %u, align %u, ioff %u, roflags 0x%08x\n",
1.25      thorpej  1830:            pp->pr_wchan, pp->pr_size, pp->pr_align, pp->pr_itemoffset,
                   1831:            pp->pr_roflags);
1.66      thorpej  1832:        (*pr)("\talloc %p\n", pp->pr_alloc);
1.25      thorpej  1833:        (*pr)("\tminitems %u, minpages %u, maxpages %u, npages %u\n",
                   1834:            pp->pr_minitems, pp->pr_minpages, pp->pr_maxpages, pp->pr_npages);
                   1835:        (*pr)("\titemsperpage %u, nitems %u, nout %u, hardlimit %u\n",
                   1836:            pp->pr_itemsperpage, pp->pr_nitems, pp->pr_nout, pp->pr_hardlimit);
                   1837:
1.134     ad       1838:        (*pr)("\tnget %lu, nfail %lu, nput %lu\n",
1.25      thorpej  1839:            pp->pr_nget, pp->pr_nfail, pp->pr_nput);
                   1840:        (*pr)("\tnpagealloc %lu, npagefree %lu, hiwat %u, nidle %lu\n",
                   1841:            pp->pr_npagealloc, pp->pr_npagefree, pp->pr_hiwat, pp->pr_nidle);
                   1842:
                   1843:        if (print_pagelist == 0)
                   1844:                goto skip_pagelist;
                   1845:
1.88      chs      1846:        if ((ph = LIST_FIRST(&pp->pr_emptypages)) != NULL)
                   1847:                (*pr)("\n\tempty page list:\n");
1.97      yamt     1848:        pool_print_pagelist(pp, &pp->pr_emptypages, pr);
1.88      chs      1849:        if ((ph = LIST_FIRST(&pp->pr_fullpages)) != NULL)
                   1850:                (*pr)("\n\tfull page list:\n");
1.97      yamt     1851:        pool_print_pagelist(pp, &pp->pr_fullpages, pr);
1.88      chs      1852:        if ((ph = LIST_FIRST(&pp->pr_partpages)) != NULL)
                   1853:                (*pr)("\n\tpartial-page list:\n");
1.97      yamt     1854:        pool_print_pagelist(pp, &pp->pr_partpages, pr);
1.88      chs      1855:
1.25      thorpej  1856:        if (pp->pr_curpage == NULL)
                   1857:                (*pr)("\tno current page\n");
                   1858:        else
                   1859:                (*pr)("\tcurpage %p\n", pp->pr_curpage->ph_page);
                   1860:
                   1861:  skip_pagelist:
                   1862:        if (print_log == 0)
                   1863:                goto skip_log;
                   1864:
                   1865:        (*pr)("\n");
                   1866:        if ((pp->pr_roflags & PR_LOGGING) == 0)
                   1867:                (*pr)("\tno log\n");
1.122     christos 1868:        else {
1.25      thorpej  1869:                pr_printlog(pp, NULL, pr);
1.122     christos 1870:        }
1.3       pk       1871:
1.25      thorpej  1872:  skip_log:
1.44      thorpej  1873:
1.102     chs      1874: #define PR_GROUPLIST(pcg)                                              \
                   1875:        (*pr)("\t\tgroup %p: avail %d\n", pcg, pcg->pcg_avail);         \
                   1876:        for (i = 0; i < PCG_NOBJECTS; i++) {                            \
                   1877:                if (pcg->pcg_objects[i].pcgo_pa !=                      \
                   1878:                    POOL_PADDR_INVALID) {                               \
                   1879:                        (*pr)("\t\t\t%p, 0x%llx\n",                     \
                   1880:                            pcg->pcg_objects[i].pcgo_va,                \
                   1881:                            (unsigned long long)                        \
                   1882:                            pcg->pcg_objects[i].pcgo_pa);               \
                   1883:                } else {                                                \
                   1884:                        (*pr)("\t\t\t%p\n",                             \
                   1885:                            pcg->pcg_objects[i].pcgo_va);               \
                   1886:                }                                                       \
                   1887:        }
                   1888:
1.134     ad       1889:        if (pc != NULL) {
                   1890:                cpuhit = 0;
                   1891:                cpumiss = 0;
                   1892:                for (i = 0; i < MAXCPUS; i++) {
                   1893:                        if ((cc = pc->pc_cpus[i]) == NULL)
                   1894:                                continue;
                   1895:                        cpuhit += cc->cc_hits;
                   1896:                        cpumiss += cc->cc_misses;
                   1897:                }
                   1898:                (*pr)("\tcpu layer hits %llu misses %llu\n", cpuhit, cpumiss);
                   1899:                (*pr)("\tcache layer hits %llu misses %llu\n",
                   1900:                    pc->pc_hits, pc->pc_misses);
                   1901:                (*pr)("\tcache layer entry uncontended %llu contended %llu\n",
                   1902:                    pc->pc_hits + pc->pc_misses - pc->pc_contended,
                   1903:                    pc->pc_contended);
                   1904:                (*pr)("\tcache layer empty groups %u full groups %u\n",
                   1905:                    pc->pc_nempty, pc->pc_nfull);
                   1906:                if (print_cache) {
                   1907:                        (*pr)("\tfull cache groups:\n");
                   1908:                        for (pcg = pc->pc_fullgroups; pcg != NULL;
                   1909:                            pcg = pcg->pcg_next) {
                   1910:                                PR_GROUPLIST(pcg);
                   1911:                        }
                   1912:                        (*pr)("\tempty cache groups:\n");
                   1913:                        for (pcg = pc->pc_emptygroups; pcg != NULL;
                   1914:                            pcg = pcg->pcg_next) {
                   1915:                                PR_GROUPLIST(pcg);
                   1916:                        }
1.103     chs      1917:                }
1.44      thorpej  1918:        }
1.102     chs      1919: #undef PR_GROUPLIST
1.44      thorpej  1920:
1.88      chs      1921:        pr_enter_check(pp, pr);
                   1922: }
                   1923:
                   1924: static int
                   1925: pool_chk_page(struct pool *pp, const char *label, struct pool_item_header *ph)
                   1926: {
                   1927:        struct pool_item *pi;
1.128     christos 1928:        void *page;
1.88      chs      1929:        int n;
                   1930:
1.121     yamt     1931:        if ((pp->pr_roflags & PR_NOALIGN) == 0) {
1.128     christos 1932:                page = (void *)((uintptr_t)ph & pp->pr_alloc->pa_pagemask);
1.121     yamt     1933:                if (page != ph->ph_page &&
                   1934:                    (pp->pr_roflags & PR_PHINPAGE) != 0) {
                   1935:                        if (label != NULL)
                   1936:                                printf("%s: ", label);
                   1937:                        printf("pool(%p:%s): page inconsistency: page %p;"
                   1938:                               " at page head addr %p (p %p)\n", pp,
                   1939:                                pp->pr_wchan, ph->ph_page,
                   1940:                                ph, page);
                   1941:                        return 1;
                   1942:                }
1.88      chs      1943:        }
1.3       pk       1944:
1.97      yamt     1945:        if ((pp->pr_roflags & PR_NOTOUCH) != 0)
                   1946:                return 0;
                   1947:
1.102     chs      1948:        for (pi = LIST_FIRST(&ph->ph_itemlist), n = 0;
1.88      chs      1949:             pi != NULL;
1.102     chs      1950:             pi = LIST_NEXT(pi,pi_list), n++) {
1.88      chs      1951:
                   1952: #ifdef DIAGNOSTIC
                   1953:                if (pi->pi_magic != PI_MAGIC) {
                   1954:                        if (label != NULL)
                   1955:                                printf("%s: ", label);
                   1956:                        printf("pool(%s): free list modified: magic=%x;"
1.121     yamt     1957:                               " page %p; item ordinal %d; addr %p\n",
1.88      chs      1958:                                pp->pr_wchan, pi->pi_magic, ph->ph_page,
1.121     yamt     1959:                                n, pi);
1.88      chs      1960:                        panic("pool");
                   1961:                }
                   1962: #endif
1.121     yamt     1963:                if ((pp->pr_roflags & PR_NOALIGN) != 0) {
                   1964:                        continue;
                   1965:                }
1.128     christos 1966:                page = (void *)((uintptr_t)pi & pp->pr_alloc->pa_pagemask);
1.88      chs      1967:                if (page == ph->ph_page)
                   1968:                        continue;
                   1969:
                   1970:                if (label != NULL)
                   1971:                        printf("%s: ", label);
                   1972:                printf("pool(%p:%s): page inconsistency: page %p;"
                   1973:                       " item ordinal %d; addr %p (p %p)\n", pp,
                   1974:                        pp->pr_wchan, ph->ph_page,
                   1975:                        n, pi, page);
                   1976:                return 1;
                   1977:        }
                   1978:        return 0;
1.3       pk       1979: }
                   1980:
1.88      chs      1981:
1.3       pk       1982: int
1.42      thorpej  1983: pool_chk(struct pool *pp, const char *label)
1.3       pk       1984: {
                   1985:        struct pool_item_header *ph;
                   1986:        int r = 0;
                   1987:
1.134     ad       1988:        mutex_enter(&pp->pr_lock);
1.88      chs      1989:        LIST_FOREACH(ph, &pp->pr_emptypages, ph_pagelist) {
                   1990:                r = pool_chk_page(pp, label, ph);
                   1991:                if (r) {
                   1992:                        goto out;
                   1993:                }
                   1994:        }
                   1995:        LIST_FOREACH(ph, &pp->pr_fullpages, ph_pagelist) {
                   1996:                r = pool_chk_page(pp, label, ph);
                   1997:                if (r) {
1.3       pk       1998:                        goto out;
                   1999:                }
1.88      chs      2000:        }
                   2001:        LIST_FOREACH(ph, &pp->pr_partpages, ph_pagelist) {
                   2002:                r = pool_chk_page(pp, label, ph);
                   2003:                if (r) {
1.3       pk       2004:                        goto out;
                   2005:                }
                   2006:        }
1.88      chs      2007:
1.3       pk       2008: out:
1.134     ad       2009:        mutex_exit(&pp->pr_lock);
1.3       pk       2010:        return (r);
1.43      thorpej  2011: }
                   2012:
                   2013: /*
                   2014:  * pool_cache_init:
                   2015:  *
                   2016:  *     Initialize a pool cache.
1.134     ad       2017:  */
                   2018: pool_cache_t
                   2019: pool_cache_init(size_t size, u_int align, u_int align_offset, u_int flags,
                   2020:     const char *wchan, struct pool_allocator *palloc, int ipl,
                   2021:     int (*ctor)(void *, void *, int), void (*dtor)(void *, void *), void *arg)
                   2022: {
                   2023:        pool_cache_t pc;
                   2024:
                   2025:        pc = pool_get(&cache_pool, PR_WAITOK);
                   2026:        if (pc == NULL)
                   2027:                return NULL;
                   2028:
                   2029:        pool_cache_bootstrap(pc, size, align, align_offset, flags, wchan,
                   2030:           palloc, ipl, ctor, dtor, arg);
                   2031:
                   2032:        return pc;
                   2033: }
                   2034:
                   2035: /*
                   2036:  * pool_cache_bootstrap:
1.43      thorpej  2037:  *
1.134     ad       2038:  *     Kernel-private version of pool_cache_init().  The caller
                   2039:  *     provides initial storage.
1.43      thorpej  2040:  */
                   2041: void
1.134     ad       2042: pool_cache_bootstrap(pool_cache_t pc, size_t size, u_int align,
                   2043:     u_int align_offset, u_int flags, const char *wchan,
                   2044:     struct pool_allocator *palloc, int ipl,
                   2045:     int (*ctor)(void *, void *, int), void (*dtor)(void *, void *),
1.43      thorpej  2046:     void *arg)
                   2047: {
1.134     ad       2048:        CPU_INFO_ITERATOR cii;
1.137.2.3! ad       2049:        pool_cache_t pc1;
1.134     ad       2050:        struct cpu_info *ci;
                   2051:        struct pool *pp;
                   2052:
                   2053:        pp = &pc->pc_pool;
                   2054:        if (palloc == NULL && ipl == IPL_NONE)
                   2055:                palloc = &pool_allocator_nointr;
                   2056:        pool_init(pp, size, align, align_offset, flags, wchan, palloc, ipl);
1.43      thorpej  2057:
1.137.2.1  ad       2058:        /*
                   2059:         * XXXAD hack to prevent IP input processing from blocking.
                   2060:         */
                   2061:        if (ipl == IPL_SOFTNET) {
                   2062:                mutex_init(&pc->pc_lock, MUTEX_DEFAULT, IPL_VM);
                   2063:        } else {
                   2064:                mutex_init(&pc->pc_lock, MUTEX_DEFAULT, ipl);
                   2065:        }
1.43      thorpej  2066:
1.134     ad       2067:        if (ctor == NULL) {
                   2068:                ctor = (int (*)(void *, void *, int))nullop;
                   2069:        }
                   2070:        if (dtor == NULL) {
                   2071:                dtor = (void (*)(void *, void *))nullop;
                   2072:        }
1.43      thorpej  2073:
1.134     ad       2074:        pc->pc_emptygroups = NULL;
                   2075:        pc->pc_fullgroups = NULL;
                   2076:        pc->pc_partgroups = NULL;
1.43      thorpej  2077:        pc->pc_ctor = ctor;
                   2078:        pc->pc_dtor = dtor;
                   2079:        pc->pc_arg  = arg;
1.134     ad       2080:        pc->pc_hits  = 0;
1.48      thorpej  2081:        pc->pc_misses = 0;
1.134     ad       2082:        pc->pc_nempty = 0;
                   2083:        pc->pc_npart = 0;
                   2084:        pc->pc_nfull = 0;
                   2085:        pc->pc_contended = 0;
                   2086:        pc->pc_refcnt = 0;
1.136     yamt     2087:        pc->pc_freecheck = NULL;
1.134     ad       2088:
                   2089:        /* Allocate per-CPU caches. */
                   2090:        memset(pc->pc_cpus, 0, sizeof(pc->pc_cpus));
                   2091:        pc->pc_ncpu = 0;
1.137     ad       2092:        if (ncpu == 0) {
                   2093:                /* XXX For sparc: boot CPU is not attached yet. */
                   2094:                pool_cache_cpu_init1(curcpu(), pc);
                   2095:        } else {
                   2096:                for (CPU_INFO_FOREACH(cii, ci)) {
                   2097:                        pool_cache_cpu_init1(ci, pc);
                   2098:                }
1.134     ad       2099:        }
1.137.2.3! ad       2100:
        !          2101:        /* Add to list of all pools. */
        !          2102:        if (__predict_true(!cold))
1.134     ad       2103:                mutex_enter(&pool_head_lock);
1.137.2.3! ad       2104:        TAILQ_FOREACH(pc1, &pool_cache_head, pc_cachelist) {
        !          2105:                if (strcmp(pc1->pc_pool.pr_wchan, pc->pc_pool.pr_wchan) > 0)
        !          2106:                        break;
1.134     ad       2107:        }
1.137.2.3! ad       2108:        if (pc1 == NULL)
        !          2109:                TAILQ_INSERT_TAIL(&pool_cache_head, pc, pc_cachelist);
        !          2110:        else
        !          2111:                TAILQ_INSERT_BEFORE(pc1, pc, pc_cachelist);
        !          2112:        if (__predict_true(!cold))
        !          2113:                mutex_exit(&pool_head_lock);
        !          2114:
        !          2115:        membar_sync();
        !          2116:        pp->pr_cache = pc;
1.43      thorpej  2117: }
                   2118:
                   2119: /*
                   2120:  * pool_cache_destroy:
                   2121:  *
                   2122:  *     Destroy a pool cache.
                   2123:  */
                   2124: void
1.134     ad       2125: pool_cache_destroy(pool_cache_t pc)
1.43      thorpej  2126: {
1.134     ad       2127:        struct pool *pp = &pc->pc_pool;
                   2128:        pool_cache_cpu_t *cc;
                   2129:        pcg_t *pcg;
                   2130:        int i;
                   2131:
                   2132:        /* Remove it from the global list. */
                   2133:        mutex_enter(&pool_head_lock);
                   2134:        while (pc->pc_refcnt != 0)
                   2135:                cv_wait(&pool_busy, &pool_head_lock);
1.137.2.3! ad       2136:        TAILQ_REMOVE(&pool_cache_head, pc, pc_cachelist);
1.134     ad       2137:        mutex_exit(&pool_head_lock);
1.43      thorpej  2138:
                   2139:        /* First, invalidate the entire cache. */
                   2140:        pool_cache_invalidate(pc);
                   2141:
1.134     ad       2142:        /* Disassociate it from the pool. */
                   2143:        mutex_enter(&pp->pr_lock);
                   2144:        pp->pr_cache = NULL;
                   2145:        mutex_exit(&pp->pr_lock);
                   2146:
                   2147:        /* Destroy per-CPU data */
                   2148:        for (i = 0; i < MAXCPUS; i++) {
                   2149:                if ((cc = pc->pc_cpus[i]) == NULL)
                   2150:                        continue;
                   2151:                if ((pcg = cc->cc_current) != NULL) {
                   2152:                        pcg->pcg_next = NULL;
                   2153:                        pool_cache_invalidate_groups(pc, pcg);
                   2154:                }
                   2155:                if ((pcg = cc->cc_previous) != NULL) {
                   2156:                        pcg->pcg_next = NULL;
                   2157:                        pool_cache_invalidate_groups(pc, pcg);
                   2158:                }
                   2159:                if (cc != &pc->pc_cpu0)
                   2160:                        pool_put(&cache_cpu_pool, cc);
                   2161:        }
                   2162:
                   2163:        /* Finally, destroy it. */
                   2164:        mutex_destroy(&pc->pc_lock);
                   2165:        pool_destroy(pp);
                   2166:        pool_put(&cache_pool, pc);
                   2167: }
                   2168:
                   2169: /*
                   2170:  * pool_cache_cpu_init1:
                   2171:  *
                   2172:  *     Called for each pool_cache whenever a new CPU is attached.
                   2173:  */
                   2174: static void
                   2175: pool_cache_cpu_init1(struct cpu_info *ci, pool_cache_t pc)
                   2176: {
                   2177:        pool_cache_cpu_t *cc;
1.137     ad       2178:        int index;
1.134     ad       2179:
1.137     ad       2180:        index = ci->ci_index;
                   2181:
                   2182:        KASSERT(index < MAXCPUS);
1.134     ad       2183:        KASSERT(((uintptr_t)pc->pc_cpus & (CACHE_LINE_SIZE - 1)) == 0);
                   2184:
1.137     ad       2185:        if ((cc = pc->pc_cpus[index]) != NULL) {
                   2186:                KASSERT(cc->cc_cpuindex == index);
1.134     ad       2187:                return;
                   2188:        }
                   2189:
                   2190:        /*
                   2191:         * The first CPU is 'free'.  This needs to be the case for
                   2192:         * bootstrap - we may not be able to allocate yet.
                   2193:         */
                   2194:        if (pc->pc_ncpu == 0) {
                   2195:                cc = &pc->pc_cpu0;
                   2196:                pc->pc_ncpu = 1;
                   2197:        } else {
                   2198:                mutex_enter(&pc->pc_lock);
                   2199:                pc->pc_ncpu++;
                   2200:                mutex_exit(&pc->pc_lock);
                   2201:                cc = pool_get(&cache_cpu_pool, PR_WAITOK);
                   2202:        }
                   2203:
                   2204:        cc->cc_ipl = pc->pc_pool.pr_ipl;
                   2205:        cc->cc_iplcookie = makeiplcookie(cc->cc_ipl);
                   2206:        cc->cc_cache = pc;
1.137     ad       2207:        cc->cc_cpuindex = index;
1.134     ad       2208:        cc->cc_hits = 0;
                   2209:        cc->cc_misses = 0;
                   2210:        cc->cc_current = NULL;
                   2211:        cc->cc_previous = NULL;
                   2212:
1.137     ad       2213:        pc->pc_cpus[index] = cc;
1.43      thorpej  2214: }
                   2215:
1.134     ad       2216: /*
                   2217:  * pool_cache_cpu_init:
                   2218:  *
                   2219:  *     Called whenever a new CPU is attached.
                   2220:  */
                   2221: void
                   2222: pool_cache_cpu_init(struct cpu_info *ci)
1.43      thorpej  2223: {
1.134     ad       2224:        pool_cache_t pc;
                   2225:
                   2226:        mutex_enter(&pool_head_lock);
1.137.2.3! ad       2227:        TAILQ_FOREACH(pc, &pool_cache_head, pc_cachelist) {
1.134     ad       2228:                pc->pc_refcnt++;
                   2229:                mutex_exit(&pool_head_lock);
1.43      thorpej  2230:
1.134     ad       2231:                pool_cache_cpu_init1(ci, pc);
1.43      thorpej  2232:
1.134     ad       2233:                mutex_enter(&pool_head_lock);
                   2234:                pc->pc_refcnt--;
                   2235:                cv_broadcast(&pool_busy);
                   2236:        }
                   2237:        mutex_exit(&pool_head_lock);
1.43      thorpej  2238: }
                   2239:
1.134     ad       2240: /*
                   2241:  * pool_cache_reclaim:
                   2242:  *
                   2243:  *     Reclaim memory from a pool cache.
                   2244:  */
                   2245: bool
                   2246: pool_cache_reclaim(pool_cache_t pc)
1.43      thorpej  2247: {
                   2248:
1.134     ad       2249:        return pool_reclaim(&pc->pc_pool);
                   2250: }
1.43      thorpej  2251:
1.136     yamt     2252: static void
                   2253: pool_cache_destruct_object1(pool_cache_t pc, void *object)
                   2254: {
                   2255:
                   2256:        (*pc->pc_dtor)(pc->pc_arg, object);
                   2257:        pool_put(&pc->pc_pool, object);
                   2258: }
                   2259:
1.134     ad       2260: /*
                   2261:  * pool_cache_destruct_object:
                   2262:  *
                   2263:  *     Force destruction of an object and its release back into
                   2264:  *     the pool.
                   2265:  */
                   2266: void
                   2267: pool_cache_destruct_object(pool_cache_t pc, void *object)
                   2268: {
                   2269:
1.136     yamt     2270:        FREECHECK_IN(&pc->pc_freecheck, object);
                   2271:
                   2272:        pool_cache_destruct_object1(pc, object);
1.43      thorpej  2273: }
                   2274:
1.134     ad       2275: /*
                   2276:  * pool_cache_invalidate_groups:
                   2277:  *
                   2278:  *     Invalidate a chain of groups and destruct all objects.
                   2279:  */
1.102     chs      2280: static void
1.134     ad       2281: pool_cache_invalidate_groups(pool_cache_t pc, pcg_t *pcg)
1.102     chs      2282: {
1.134     ad       2283:        void *object;
                   2284:        pcg_t *next;
                   2285:        int i;
                   2286:
                   2287:        for (; pcg != NULL; pcg = next) {
                   2288:                next = pcg->pcg_next;
                   2289:
                   2290:                for (i = 0; i < pcg->pcg_avail; i++) {
                   2291:                        object = pcg->pcg_objects[i].pcgo_va;
1.136     yamt     2292:                        pool_cache_destruct_object1(pc, object);
1.134     ad       2293:                }
1.102     chs      2294:
                   2295:                pool_put(&pcgpool, pcg);
                   2296:        }
                   2297: }
                   2298:
1.43      thorpej  2299: /*
1.134     ad       2300:  * pool_cache_invalidate:
1.43      thorpej  2301:  *
1.134     ad       2302:  *     Invalidate a pool cache (destruct and release all of the
                   2303:  *     cached objects).  Does not reclaim objects from the pool.
1.43      thorpej  2304:  */
1.134     ad       2305: void
                   2306: pool_cache_invalidate(pool_cache_t pc)
                   2307: {
                   2308:        pcg_t *full, *empty, *part;
                   2309:
                   2310:        mutex_enter(&pc->pc_lock);
                   2311:        full = pc->pc_fullgroups;
                   2312:        empty = pc->pc_emptygroups;
                   2313:        part = pc->pc_partgroups;
                   2314:        pc->pc_fullgroups = NULL;
                   2315:        pc->pc_emptygroups = NULL;
                   2316:        pc->pc_partgroups = NULL;
                   2317:        pc->pc_nfull = 0;
                   2318:        pc->pc_nempty = 0;
                   2319:        pc->pc_npart = 0;
                   2320:        mutex_exit(&pc->pc_lock);
                   2321:
                   2322:        pool_cache_invalidate_groups(pc, full);
                   2323:        pool_cache_invalidate_groups(pc, empty);
                   2324:        pool_cache_invalidate_groups(pc, part);
                   2325: }
                   2326:
                   2327: void
                   2328: pool_cache_set_drain_hook(pool_cache_t pc, void (*fn)(void *, int), void *arg)
                   2329: {
                   2330:
                   2331:        pool_set_drain_hook(&pc->pc_pool, fn, arg);
                   2332: }
                   2333:
                   2334: void
                   2335: pool_cache_setlowat(pool_cache_t pc, int n)
                   2336: {
                   2337:
                   2338:        pool_setlowat(&pc->pc_pool, n);
                   2339: }
                   2340:
                   2341: void
                   2342: pool_cache_sethiwat(pool_cache_t pc, int n)
                   2343: {
                   2344:
                   2345:        pool_sethiwat(&pc->pc_pool, n);
                   2346: }
                   2347:
                   2348: void
                   2349: pool_cache_sethardlimit(pool_cache_t pc, int n, const char *warnmess, int ratecap)
                   2350: {
                   2351:
                   2352:        pool_sethardlimit(&pc->pc_pool, n, warnmess, ratecap);
                   2353: }
                   2354:
                   2355: static inline pool_cache_cpu_t *
                   2356: pool_cache_cpu_enter(pool_cache_t pc, int *s)
                   2357: {
                   2358:        pool_cache_cpu_t *cc;
                   2359:
                   2360:        /*
                   2361:         * Prevent other users of the cache from accessing our
                   2362:         * CPU-local data.  To avoid touching shared state, we
                   2363:         * pull the neccessary information from CPU local data.
                   2364:         */
1.137     ad       2365:        crit_enter();
                   2366:        cc = pc->pc_cpus[curcpu()->ci_index];
1.134     ad       2367:        KASSERT(cc->cc_cache == pc);
1.137     ad       2368:        if (cc->cc_ipl != IPL_NONE) {
1.134     ad       2369:                *s = splraiseipl(cc->cc_iplcookie);
                   2370:        }
                   2371:        KASSERT(((uintptr_t)cc & (CACHE_LINE_SIZE - 1)) == 0);
                   2372:
                   2373:        return cc;
                   2374: }
                   2375:
                   2376: static inline void
                   2377: pool_cache_cpu_exit(pool_cache_cpu_t *cc, int *s)
                   2378: {
                   2379:
                   2380:        /* No longer need exclusive access to the per-CPU data. */
1.137     ad       2381:        if (cc->cc_ipl != IPL_NONE) {
1.134     ad       2382:                splx(*s);
                   2383:        }
1.137     ad       2384:        crit_exit();
1.134     ad       2385: }
                   2386:
                   2387: #if __GNUC_PREREQ__(3, 0)
                   2388: __attribute ((noinline))
                   2389: #endif
                   2390: pool_cache_cpu_t *
                   2391: pool_cache_get_slow(pool_cache_cpu_t *cc, int *s, void **objectp,
                   2392:                    paddr_t *pap, int flags)
1.43      thorpej  2393: {
1.134     ad       2394:        pcg_t *pcg, *cur;
                   2395:        uint64_t ncsw;
                   2396:        pool_cache_t pc;
1.43      thorpej  2397:        void *object;
1.58      thorpej  2398:
1.134     ad       2399:        pc = cc->cc_cache;
                   2400:        cc->cc_misses++;
1.43      thorpej  2401:
1.134     ad       2402:        /*
                   2403:         * Nothing was available locally.  Try and grab a group
                   2404:         * from the cache.
                   2405:         */
                   2406:        if (!mutex_tryenter(&pc->pc_lock)) {
                   2407:                ncsw = curlwp->l_ncsw;
                   2408:                mutex_enter(&pc->pc_lock);
                   2409:                pc->pc_contended++;
1.43      thorpej  2410:
1.134     ad       2411:                /*
                   2412:                 * If we context switched while locking, then
                   2413:                 * our view of the per-CPU data is invalid:
                   2414:                 * retry.
                   2415:                 */
                   2416:                if (curlwp->l_ncsw != ncsw) {
                   2417:                        mutex_exit(&pc->pc_lock);
                   2418:                        pool_cache_cpu_exit(cc, s);
                   2419:                        return pool_cache_cpu_enter(pc, s);
1.43      thorpej  2420:                }
1.102     chs      2421:        }
1.43      thorpej  2422:
1.134     ad       2423:        if ((pcg = pc->pc_fullgroups) != NULL) {
1.43      thorpej  2424:                /*
1.134     ad       2425:                 * If there's a full group, release our empty
                   2426:                 * group back to the cache.  Install the full
                   2427:                 * group as cc_current and return.
1.43      thorpej  2428:                 */
1.134     ad       2429:                if ((cur = cc->cc_current) != NULL) {
                   2430:                        KASSERT(cur->pcg_avail == 0);
                   2431:                        cur->pcg_next = pc->pc_emptygroups;
                   2432:                        pc->pc_emptygroups = cur;
                   2433:                        pc->pc_nempty++;
1.87      thorpej  2434:                }
1.134     ad       2435:                KASSERT(pcg->pcg_avail == PCG_NOBJECTS);
                   2436:                cc->cc_current = pcg;
                   2437:                pc->pc_fullgroups = pcg->pcg_next;
                   2438:                pc->pc_hits++;
                   2439:                pc->pc_nfull--;
                   2440:                mutex_exit(&pc->pc_lock);
                   2441:                return cc;
                   2442:        }
                   2443:
                   2444:        /*
                   2445:         * Nothing available locally or in cache.  Take the slow
                   2446:         * path: fetch a new object from the pool and construct
                   2447:         * it.
                   2448:         */
                   2449:        pc->pc_misses++;
                   2450:        mutex_exit(&pc->pc_lock);
                   2451:        pool_cache_cpu_exit(cc, s);
                   2452:
                   2453:        object = pool_get(&pc->pc_pool, flags);
                   2454:        *objectp = object;
                   2455:        if (object == NULL)
                   2456:                return NULL;
1.125     ad       2457:
1.134     ad       2458:        if ((*pc->pc_ctor)(pc->pc_arg, object, flags) != 0) {
                   2459:                pool_put(&pc->pc_pool, object);
                   2460:                *objectp = NULL;
                   2461:                return NULL;
1.43      thorpej  2462:        }
                   2463:
1.134     ad       2464:        KASSERT((((vaddr_t)object + pc->pc_pool.pr_itemoffset) &
                   2465:            (pc->pc_pool.pr_align - 1)) == 0);
1.43      thorpej  2466:
1.134     ad       2467:        if (pap != NULL) {
                   2468: #ifdef POOL_VTOPHYS
                   2469:                *pap = POOL_VTOPHYS(object);
                   2470: #else
                   2471:                *pap = POOL_PADDR_INVALID;
                   2472: #endif
1.102     chs      2473:        }
1.43      thorpej  2474:
1.125     ad       2475:        FREECHECK_OUT(&pc->pc_freecheck, object);
1.134     ad       2476:        return NULL;
1.43      thorpej  2477: }
                   2478:
                   2479: /*
1.134     ad       2480:  * pool_cache_get{,_paddr}:
1.43      thorpej  2481:  *
1.134     ad       2482:  *     Get an object from a pool cache (optionally returning
                   2483:  *     the physical address of the object).
1.43      thorpej  2484:  */
1.134     ad       2485: void *
                   2486: pool_cache_get_paddr(pool_cache_t pc, int flags, paddr_t *pap)
1.43      thorpej  2487: {
1.134     ad       2488:        pool_cache_cpu_t *cc;
                   2489:        pcg_t *pcg;
                   2490:        void *object;
1.60      thorpej  2491:        int s;
1.43      thorpej  2492:
1.134     ad       2493: #ifdef LOCKDEBUG
                   2494:        if (flags & PR_WAITOK)
                   2495:                ASSERT_SLEEPABLE(NULL, "pool_cache_get(PR_WAITOK)");
                   2496: #endif
1.125     ad       2497:
1.134     ad       2498:        cc = pool_cache_cpu_enter(pc, &s);
                   2499:        do {
                   2500:                /* Try and allocate an object from the current group. */
                   2501:                pcg = cc->cc_current;
                   2502:                if (pcg != NULL && pcg->pcg_avail > 0) {
                   2503:                        object = pcg->pcg_objects[--pcg->pcg_avail].pcgo_va;
                   2504:                        if (pap != NULL)
                   2505:                                *pap = pcg->pcg_objects[pcg->pcg_avail].pcgo_pa;
                   2506:                        pcg->pcg_objects[pcg->pcg_avail].pcgo_va = NULL;
                   2507:                        KASSERT(pcg->pcg_avail <= PCG_NOBJECTS);
                   2508:                        KASSERT(object != NULL);
                   2509:                        cc->cc_hits++;
                   2510:                        pool_cache_cpu_exit(cc, &s);
                   2511:                        FREECHECK_OUT(&pc->pc_freecheck, object);
                   2512:                        return object;
1.43      thorpej  2513:                }
                   2514:
                   2515:                /*
1.134     ad       2516:                 * That failed.  If the previous group isn't empty, swap
                   2517:                 * it with the current group and allocate from there.
1.43      thorpej  2518:                 */
1.134     ad       2519:                pcg = cc->cc_previous;
                   2520:                if (pcg != NULL && pcg->pcg_avail > 0) {
                   2521:                        cc->cc_previous = cc->cc_current;
                   2522:                        cc->cc_current = pcg;
                   2523:                        continue;
1.43      thorpej  2524:                }
                   2525:
1.134     ad       2526:                /*
                   2527:                 * Can't allocate from either group: try the slow path.
                   2528:                 * If get_slow() allocated an object for us, or if
                   2529:                 * no more objects are available, it will return NULL.
                   2530:                 * Otherwise, we need to retry.
                   2531:                 */
                   2532:                cc = pool_cache_get_slow(cc, &s, &object, pap, flags);
                   2533:        } while (cc != NULL);
1.43      thorpej  2534:
1.134     ad       2535:        return object;
1.51      thorpej  2536: }
                   2537:
1.134     ad       2538: #if __GNUC_PREREQ__(3, 0)
                   2539: __attribute ((noinline))
                   2540: #endif
                   2541: pool_cache_cpu_t *
                   2542: pool_cache_put_slow(pool_cache_cpu_t *cc, int *s, void *object, paddr_t pa)
1.51      thorpej  2543: {
1.134     ad       2544:        pcg_t *pcg, *cur;
                   2545:        uint64_t ncsw;
                   2546:        pool_cache_t pc;
1.51      thorpej  2547:
1.134     ad       2548:        pc = cc->cc_cache;
                   2549:        cc->cc_misses++;
1.43      thorpej  2550:
1.134     ad       2551:        /*
                   2552:         * No free slots locally.  Try to grab an empty, unused
                   2553:         * group from the cache.
                   2554:         */
                   2555:        if (!mutex_tryenter(&pc->pc_lock)) {
                   2556:                ncsw = curlwp->l_ncsw;
                   2557:                mutex_enter(&pc->pc_lock);
                   2558:                pc->pc_contended++;
1.102     chs      2559:
1.134     ad       2560:                /*
                   2561:                 * If we context switched while locking, then
                   2562:                 * our view of the per-CPU data is invalid:
                   2563:                 * retry.
                   2564:                 */
                   2565:                if (curlwp->l_ncsw != ncsw) {
                   2566:                        mutex_exit(&pc->pc_lock);
                   2567:                        pool_cache_cpu_exit(cc, s);
                   2568:                        return pool_cache_cpu_enter(pc, s);
                   2569:                }
                   2570:        }
1.130     ad       2571:
1.134     ad       2572:        if ((pcg = pc->pc_emptygroups) != NULL) {
                   2573:                /*
                   2574:                 * If there's a empty group, release our full
                   2575:                 * group back to the cache.  Install the empty
                   2576:                 * group as cc_current and return.
                   2577:                 */
                   2578:                if ((cur = cc->cc_current) != NULL) {
                   2579:                        KASSERT(cur->pcg_avail == PCG_NOBJECTS);
                   2580:                        cur->pcg_next = pc->pc_fullgroups;
                   2581:                        pc->pc_fullgroups = cur;
                   2582:                        pc->pc_nfull++;
1.102     chs      2583:                }
1.134     ad       2584:                KASSERT(pcg->pcg_avail == 0);
                   2585:                cc->cc_current = pcg;
                   2586:                pc->pc_emptygroups = pcg->pcg_next;
                   2587:                pc->pc_hits++;
                   2588:                pc->pc_nempty--;
                   2589:                mutex_exit(&pc->pc_lock);
                   2590:                return cc;
1.102     chs      2591:        }
1.105     christos 2592:
1.134     ad       2593:        /*
                   2594:         * Nothing available locally or in cache.  Take the
                   2595:         * slow path and try to allocate a new group that we
                   2596:         * can release to.
                   2597:         */
                   2598:        pc->pc_misses++;
                   2599:        mutex_exit(&pc->pc_lock);
                   2600:        pool_cache_cpu_exit(cc, s);
1.105     christos 2601:
1.134     ad       2602:        /*
                   2603:         * If we can't allocate a new group, just throw the
                   2604:         * object away.
                   2605:         */
1.137.2.2  ad       2606:        if (pool_cache_disable) {
                   2607:                pcg = NULL;
                   2608:        } else {
                   2609:                pcg = pool_get(&pcgpool, PR_NOWAIT);
                   2610:        }
1.134     ad       2611:        if (pcg == NULL) {
                   2612:                pool_cache_destruct_object(pc, object);
                   2613:                return NULL;
                   2614:        }
                   2615: #ifdef DIAGNOSTIC
                   2616:        memset(pcg, 0, sizeof(*pcg));
                   2617: #else
                   2618:        pcg->pcg_avail = 0;
                   2619: #endif
1.105     christos 2620:
1.134     ad       2621:        /*
                   2622:         * Add the empty group to the cache and try again.
                   2623:         */
                   2624:        mutex_enter(&pc->pc_lock);
                   2625:        pcg->pcg_next = pc->pc_emptygroups;
                   2626:        pc->pc_emptygroups = pcg;
                   2627:        pc->pc_nempty++;
                   2628:        mutex_exit(&pc->pc_lock);
1.103     chs      2629:
1.134     ad       2630:        return pool_cache_cpu_enter(pc, s);
                   2631: }
1.102     chs      2632:
1.43      thorpej  2633: /*
1.134     ad       2634:  * pool_cache_put{,_paddr}:
1.43      thorpej  2635:  *
1.134     ad       2636:  *     Put an object back to the pool cache (optionally caching the
                   2637:  *     physical address of the object).
1.43      thorpej  2638:  */
1.101     thorpej  2639: void
1.134     ad       2640: pool_cache_put_paddr(pool_cache_t pc, void *object, paddr_t pa)
1.43      thorpej  2641: {
1.134     ad       2642:        pool_cache_cpu_t *cc;
                   2643:        pcg_t *pcg;
                   2644:        int s;
1.101     thorpej  2645:
1.134     ad       2646:        FREECHECK_IN(&pc->pc_freecheck, object);
1.101     thorpej  2647:
1.134     ad       2648:        cc = pool_cache_cpu_enter(pc, &s);
                   2649:        do {
                   2650:                /* If the current group isn't full, release it there. */
                   2651:                pcg = cc->cc_current;
                   2652:                if (pcg != NULL && pcg->pcg_avail < PCG_NOBJECTS) {
                   2653:                        KASSERT(pcg->pcg_objects[pcg->pcg_avail].pcgo_va
                   2654:                            == NULL);
                   2655:                        pcg->pcg_objects[pcg->pcg_avail].pcgo_va = object;
                   2656:                        pcg->pcg_objects[pcg->pcg_avail].pcgo_pa = pa;
                   2657:                        pcg->pcg_avail++;
                   2658:                        cc->cc_hits++;
                   2659:                        pool_cache_cpu_exit(cc, &s);
                   2660:                        return;
                   2661:                }
1.43      thorpej  2662:
1.134     ad       2663:                /*
                   2664:                 * That failed.  If the previous group is empty, swap
                   2665:                 * it with the current group and try again.
                   2666:                 */
                   2667:                pcg = cc->cc_previous;
                   2668:                if (pcg != NULL && pcg->pcg_avail == 0) {
                   2669:                        cc->cc_previous = cc->cc_current;
                   2670:                        cc->cc_current = pcg;
                   2671:                        continue;
                   2672:                }
1.43      thorpej  2673:
1.134     ad       2674:                /*
                   2675:                 * Can't free to either group: try the slow path.
                   2676:                 * If put_slow() releases the object for us, it
                   2677:                 * will return NULL.  Otherwise we need to retry.
                   2678:                 */
                   2679:                cc = pool_cache_put_slow(cc, &s, object, pa);
                   2680:        } while (cc != NULL);
1.43      thorpej  2681: }
                   2682:
                   2683: /*
1.134     ad       2684:  * pool_cache_xcall:
1.43      thorpej  2685:  *
1.134     ad       2686:  *     Transfer objects from the per-CPU cache to the global cache.
                   2687:  *     Run within a cross-call thread.
1.43      thorpej  2688:  */
                   2689: static void
1.134     ad       2690: pool_cache_xcall(pool_cache_t pc)
1.43      thorpej  2691: {
1.134     ad       2692:        pool_cache_cpu_t *cc;
                   2693:        pcg_t *prev, *cur, **list;
                   2694:        int s = 0; /* XXXgcc */
                   2695:
                   2696:        cc = pool_cache_cpu_enter(pc, &s);
                   2697:        cur = cc->cc_current;
                   2698:        cc->cc_current = NULL;
                   2699:        prev = cc->cc_previous;
                   2700:        cc->cc_previous = NULL;
                   2701:        pool_cache_cpu_exit(cc, &s);
                   2702:
                   2703:        /*
                   2704:         * XXXSMP Go to splvm to prevent kernel_lock from being taken,
                   2705:         * because locks at IPL_SOFTXXX are still spinlocks.  Does not
                   2706:         * apply to IPL_SOFTBIO.  Cross-call threads do not take the
                   2707:         * kernel_lock.
1.101     thorpej  2708:         */
1.134     ad       2709:        s = splvm();
                   2710:        mutex_enter(&pc->pc_lock);
                   2711:        if (cur != NULL) {
                   2712:                if (cur->pcg_avail == PCG_NOBJECTS) {
                   2713:                        list = &pc->pc_fullgroups;
                   2714:                        pc->pc_nfull++;
                   2715:                } else if (cur->pcg_avail == 0) {
                   2716:                        list = &pc->pc_emptygroups;
                   2717:                        pc->pc_nempty++;
                   2718:                } else {
                   2719:                        list = &pc->pc_partgroups;
                   2720:                        pc->pc_npart++;
                   2721:                }
                   2722:                cur->pcg_next = *list;
                   2723:                *list = cur;
                   2724:        }
                   2725:        if (prev != NULL) {
                   2726:                if (prev->pcg_avail == PCG_NOBJECTS) {
                   2727:                        list = &pc->pc_fullgroups;
                   2728:                        pc->pc_nfull++;
                   2729:                } else if (prev->pcg_avail == 0) {
                   2730:                        list = &pc->pc_emptygroups;
                   2731:                        pc->pc_nempty++;
                   2732:                } else {
                   2733:                        list = &pc->pc_partgroups;
                   2734:                        pc->pc_npart++;
                   2735:                }
                   2736:                prev->pcg_next = *list;
                   2737:                *list = prev;
                   2738:        }
                   2739:        mutex_exit(&pc->pc_lock);
                   2740:        splx(s);
1.3       pk       2741: }
1.66      thorpej  2742:
                   2743: /*
                   2744:  * Pool backend allocators.
                   2745:  *
                   2746:  * Each pool has a backend allocator that handles allocation, deallocation,
                   2747:  * and any additional draining that might be needed.
                   2748:  *
                   2749:  * We provide two standard allocators:
                   2750:  *
                   2751:  *     pool_allocator_kmem - the default when no allocator is specified
                   2752:  *
                   2753:  *     pool_allocator_nointr - used for pools that will not be accessed
                   2754:  *     in interrupt context.
                   2755:  */
                   2756: void   *pool_page_alloc(struct pool *, int);
                   2757: void   pool_page_free(struct pool *, void *);
                   2758:
1.112     bjh21    2759: #ifdef POOL_SUBPAGE
                   2760: struct pool_allocator pool_allocator_kmem_fullpage = {
                   2761:        pool_page_alloc, pool_page_free, 0,
1.117     yamt     2762:        .pa_backingmapptr = &kmem_map,
1.112     bjh21    2763: };
                   2764: #else
1.66      thorpej  2765: struct pool_allocator pool_allocator_kmem = {
                   2766:        pool_page_alloc, pool_page_free, 0,
1.117     yamt     2767:        .pa_backingmapptr = &kmem_map,
1.66      thorpej  2768: };
1.112     bjh21    2769: #endif
1.66      thorpej  2770:
                   2771: void   *pool_page_alloc_nointr(struct pool *, int);
                   2772: void   pool_page_free_nointr(struct pool *, void *);
                   2773:
1.112     bjh21    2774: #ifdef POOL_SUBPAGE
                   2775: struct pool_allocator pool_allocator_nointr_fullpage = {
                   2776:        pool_page_alloc_nointr, pool_page_free_nointr, 0,
1.117     yamt     2777:        .pa_backingmapptr = &kernel_map,
1.112     bjh21    2778: };
                   2779: #else
1.66      thorpej  2780: struct pool_allocator pool_allocator_nointr = {
                   2781:        pool_page_alloc_nointr, pool_page_free_nointr, 0,
1.117     yamt     2782:        .pa_backingmapptr = &kernel_map,
1.66      thorpej  2783: };
1.112     bjh21    2784: #endif
1.66      thorpej  2785:
                   2786: #ifdef POOL_SUBPAGE
                   2787: void   *pool_subpage_alloc(struct pool *, int);
                   2788: void   pool_subpage_free(struct pool *, void *);
                   2789:
1.112     bjh21    2790: struct pool_allocator pool_allocator_kmem = {
                   2791:        pool_subpage_alloc, pool_subpage_free, POOL_SUBPAGE,
1.117     yamt     2792:        .pa_backingmapptr = &kmem_map,
1.112     bjh21    2793: };
                   2794:
                   2795: void   *pool_subpage_alloc_nointr(struct pool *, int);
                   2796: void   pool_subpage_free_nointr(struct pool *, void *);
                   2797:
                   2798: struct pool_allocator pool_allocator_nointr = {
                   2799:        pool_subpage_alloc, pool_subpage_free, POOL_SUBPAGE,
1.117     yamt     2800:        .pa_backingmapptr = &kmem_map,
1.66      thorpej  2801: };
                   2802: #endif /* POOL_SUBPAGE */
                   2803:
1.117     yamt     2804: static void *
                   2805: pool_allocator_alloc(struct pool *pp, int flags)
1.66      thorpej  2806: {
1.117     yamt     2807:        struct pool_allocator *pa = pp->pr_alloc;
1.66      thorpej  2808:        void *res;
                   2809:
1.117     yamt     2810:        res = (*pa->pa_alloc)(pp, flags);
                   2811:        if (res == NULL && (flags & PR_WAITOK) == 0) {
1.66      thorpej  2812:                /*
1.117     yamt     2813:                 * We only run the drain hook here if PR_NOWAIT.
                   2814:                 * In other cases, the hook will be run in
                   2815:                 * pool_reclaim().
1.66      thorpej  2816:                 */
1.117     yamt     2817:                if (pp->pr_drain_hook != NULL) {
                   2818:                        (*pp->pr_drain_hook)(pp->pr_drain_hook_arg, flags);
                   2819:                        res = (*pa->pa_alloc)(pp, flags);
1.66      thorpej  2820:                }
1.117     yamt     2821:        }
                   2822:        return res;
1.66      thorpej  2823: }
                   2824:
1.117     yamt     2825: static void
1.66      thorpej  2826: pool_allocator_free(struct pool *pp, void *v)
                   2827: {
                   2828:        struct pool_allocator *pa = pp->pr_alloc;
                   2829:
                   2830:        (*pa->pa_free)(pp, v);
                   2831: }
                   2832:
                   2833: void *
1.124     yamt     2834: pool_page_alloc(struct pool *pp, int flags)
1.66      thorpej  2835: {
1.127     thorpej  2836:        bool waitok = (flags & PR_WAITOK) ? true : false;
1.66      thorpej  2837:
1.100     yamt     2838:        return ((void *) uvm_km_alloc_poolpage_cache(kmem_map, waitok));
1.66      thorpej  2839: }
                   2840:
                   2841: void
1.124     yamt     2842: pool_page_free(struct pool *pp, void *v)
1.66      thorpej  2843: {
                   2844:
1.98      yamt     2845:        uvm_km_free_poolpage_cache(kmem_map, (vaddr_t) v);
                   2846: }
                   2847:
                   2848: static void *
1.124     yamt     2849: pool_page_alloc_meta(struct pool *pp, int flags)
1.98      yamt     2850: {
1.127     thorpej  2851:        bool waitok = (flags & PR_WAITOK) ? true : false;
1.98      yamt     2852:
1.100     yamt     2853:        return ((void *) uvm_km_alloc_poolpage(kmem_map, waitok));
1.98      yamt     2854: }
                   2855:
                   2856: static void
1.124     yamt     2857: pool_page_free_meta(struct pool *pp, void *v)
1.98      yamt     2858: {
                   2859:
1.100     yamt     2860:        uvm_km_free_poolpage(kmem_map, (vaddr_t) v);
1.66      thorpej  2861: }
                   2862:
                   2863: #ifdef POOL_SUBPAGE
                   2864: /* Sub-page allocator, for machines with large hardware pages. */
                   2865: void *
                   2866: pool_subpage_alloc(struct pool *pp, int flags)
                   2867: {
1.134     ad       2868:        return pool_get(&psppool, flags);
1.66      thorpej  2869: }
                   2870:
                   2871: void
                   2872: pool_subpage_free(struct pool *pp, void *v)
                   2873: {
                   2874:        pool_put(&psppool, v);
                   2875: }
                   2876:
                   2877: /* We don't provide a real nointr allocator.  Maybe later. */
                   2878: void *
1.112     bjh21    2879: pool_subpage_alloc_nointr(struct pool *pp, int flags)
1.66      thorpej  2880: {
                   2881:
                   2882:        return (pool_subpage_alloc(pp, flags));
                   2883: }
                   2884:
                   2885: void
1.112     bjh21    2886: pool_subpage_free_nointr(struct pool *pp, void *v)
1.66      thorpej  2887: {
                   2888:
                   2889:        pool_subpage_free(pp, v);
                   2890: }
1.112     bjh21    2891: #endif /* POOL_SUBPAGE */
1.66      thorpej  2892: void *
1.124     yamt     2893: pool_page_alloc_nointr(struct pool *pp, int flags)
1.66      thorpej  2894: {
1.127     thorpej  2895:        bool waitok = (flags & PR_WAITOK) ? true : false;
1.66      thorpej  2896:
1.100     yamt     2897:        return ((void *) uvm_km_alloc_poolpage_cache(kernel_map, waitok));
1.66      thorpej  2898: }
                   2899:
                   2900: void
1.124     yamt     2901: pool_page_free_nointr(struct pool *pp, void *v)
1.66      thorpej  2902: {
                   2903:
1.98      yamt     2904:        uvm_km_free_poolpage_cache(kernel_map, (vaddr_t) v);
1.66      thorpej  2905: }

CVSweb <webmaster@jp.NetBSD.org>