[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 1.137.2.4

1.137.2.4! ad          1: /*     $NetBSD: subr_pool.c,v 1.137.2.3 2007/12/15 02:22:22 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>
1.137.2.4! ad         41: __KERNEL_RCSID(0, "$NetBSD: subr_pool.c,v 1.137.2.3 2007/12/15 02:22:22 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>
1.137.2.4! ad         61: #include <sys/atomic.h>
1.3       pk         62:
                     63: #include <uvm/uvm.h>
                     64:
1.1       pk         65: /*
                     66:  * Pool resource management utility.
1.3       pk         67:  *
1.88      chs        68:  * Memory is allocated in pages which are split into pieces according to
                     69:  * the pool item size. Each page is kept on one of three lists in the
                     70:  * pool structure: `pr_emptypages', `pr_fullpages' and `pr_partpages',
                     71:  * for empty, full and partially-full pages respectively. The individual
                     72:  * pool items are on a linked list headed by `ph_itemlist' in each page
                     73:  * header. The memory for building the page list is either taken from
                     74:  * the allocated pages themselves (for small pool items) or taken from
                     75:  * an internal pool of page headers (`phpool').
1.1       pk         76:  */
                     77:
1.3       pk         78: /* List of all pools */
1.137.2.3  ad         79: TAILQ_HEAD(,pool) pool_head = TAILQ_HEAD_INITIALIZER(pool_head);
1.3       pk         80:
                     81: /* Private pool for page header structures */
1.97      yamt       82: #define        PHPOOL_MAX      8
                     83: static struct pool phpool[PHPOOL_MAX];
1.135     yamt       84: #define        PHPOOL_FREELIST_NELEM(idx) \
                     85:        (((idx) == 0) ? 0 : BITMAP_SIZE * (1 << (idx)))
1.3       pk         86:
1.62      bjh21      87: #ifdef POOL_SUBPAGE
                     88: /* Pool of subpages for use by normal pools. */
                     89: static struct pool psppool;
                     90: #endif
                     91:
1.117     yamt       92: static SLIST_HEAD(, pool_allocator) pa_deferinitq =
                     93:     SLIST_HEAD_INITIALIZER(pa_deferinitq);
                     94:
1.98      yamt       95: static void *pool_page_alloc_meta(struct pool *, int);
                     96: static void pool_page_free_meta(struct pool *, void *);
                     97:
                     98: /* allocator for pool metadata */
1.134     ad         99: struct pool_allocator pool_allocator_meta = {
1.117     yamt      100:        pool_page_alloc_meta, pool_page_free_meta,
                    101:        .pa_backingmapptr = &kmem_map,
1.98      yamt      102: };
                    103:
1.3       pk        104: /* # of seconds to retain page after last use */
                    105: int pool_inactive_time = 10;
                    106:
                    107: /* Next candidate for drainage (see pool_drain()) */
1.23      thorpej   108: static struct pool     *drainpp;
                    109:
1.134     ad        110: /* This lock protects both pool_head and drainpp. */
                    111: static kmutex_t pool_head_lock;
                    112: static kcondvar_t pool_busy;
1.3       pk        113:
1.135     yamt      114: typedef uint32_t pool_item_bitmap_t;
                    115: #define        BITMAP_SIZE     (CHAR_BIT * sizeof(pool_item_bitmap_t))
                    116: #define        BITMAP_MASK     (BITMAP_SIZE - 1)
1.99      yamt      117:
1.3       pk        118: struct pool_item_header {
                    119:        /* Page headers */
1.88      chs       120:        LIST_ENTRY(pool_item_header)
1.3       pk        121:                                ph_pagelist;    /* pool page list */
1.88      chs       122:        SPLAY_ENTRY(pool_item_header)
                    123:                                ph_node;        /* Off-page page headers */
1.128     christos  124:        void *                  ph_page;        /* this page's address */
1.3       pk        125:        struct timeval          ph_time;        /* last referenced */
1.135     yamt      126:        uint16_t                ph_nmissing;    /* # of chunks in use */
1.97      yamt      127:        union {
                    128:                /* !PR_NOTOUCH */
                    129:                struct {
1.102     chs       130:                        LIST_HEAD(, pool_item)
1.97      yamt      131:                                phu_itemlist;   /* chunk list for this page */
                    132:                } phu_normal;
                    133:                /* PR_NOTOUCH */
                    134:                struct {
1.135     yamt      135:                        uint16_t phu_off;       /* start offset in page */
                    136:                        pool_item_bitmap_t phu_bitmap[];
1.97      yamt      137:                } phu_notouch;
                    138:        } ph_u;
1.3       pk        139: };
1.97      yamt      140: #define        ph_itemlist     ph_u.phu_normal.phu_itemlist
                    141: #define        ph_off          ph_u.phu_notouch.phu_off
1.135     yamt      142: #define        ph_bitmap       ph_u.phu_notouch.phu_bitmap
1.3       pk        143:
1.1       pk        144: struct pool_item {
1.3       pk        145: #ifdef DIAGNOSTIC
1.82      thorpej   146:        u_int pi_magic;
1.33      chs       147: #endif
1.134     ad        148: #define        PI_MAGIC 0xdeaddeadU
1.3       pk        149:        /* Other entries use only this list entry */
1.102     chs       150:        LIST_ENTRY(pool_item)   pi_list;
1.3       pk        151: };
                    152:
1.53      thorpej   153: #define        POOL_NEEDS_CATCHUP(pp)                                          \
                    154:        ((pp)->pr_nitems < (pp)->pr_minitems)
                    155:
1.43      thorpej   156: /*
                    157:  * Pool cache management.
                    158:  *
                    159:  * Pool caches provide a way for constructed objects to be cached by the
                    160:  * pool subsystem.  This can lead to performance improvements by avoiding
                    161:  * needless object construction/destruction; it is deferred until absolutely
                    162:  * necessary.
                    163:  *
1.134     ad        164:  * Caches are grouped into cache groups.  Each cache group references up
                    165:  * to PCG_NUMOBJECTS constructed objects.  When a cache allocates an
                    166:  * object from the pool, it calls the object's constructor and places it
                    167:  * into a cache group.  When a cache group frees an object back to the
                    168:  * pool, it first calls the object's destructor.  This allows the object
                    169:  * to persist in constructed form while freed to the cache.
                    170:  *
                    171:  * The pool references each cache, so that when a pool is drained by the
                    172:  * pagedaemon, it can drain each individual cache as well.  Each time a
                    173:  * cache is drained, the most idle cache group is freed to the pool in
                    174:  * its entirety.
1.43      thorpej   175:  *
                    176:  * Pool caches are layed on top of pools.  By layering them, we can avoid
                    177:  * the complexity of cache management for pools which would not benefit
                    178:  * from it.
                    179:  */
                    180:
                    181: static struct pool pcgpool;
1.134     ad        182: static struct pool cache_pool;
                    183: static struct pool cache_cpu_pool;
1.3       pk        184:
1.137.2.2  ad        185: /* List of all caches. */
1.137.2.3  ad        186: TAILQ_HEAD(,pool_cache) pool_cache_head =
                    187:     TAILQ_HEAD_INITIALIZER(pool_cache_head);
1.137.2.2  ad        188:
                    189: int pool_cache_disable;
                    190:
                    191:
1.134     ad        192: static pool_cache_cpu_t *pool_cache_put_slow(pool_cache_cpu_t *, int *,
                    193:                                             void *, paddr_t);
                    194: static pool_cache_cpu_t *pool_cache_get_slow(pool_cache_cpu_t *, int *,
                    195:                                             void **, paddr_t *, int);
                    196: static void    pool_cache_cpu_init1(struct cpu_info *, pool_cache_t);
                    197: static void    pool_cache_invalidate_groups(pool_cache_t, pcg_t *);
                    198: static void    pool_cache_xcall(pool_cache_t);
1.3       pk        199:
1.42      thorpej   200: static int     pool_catchup(struct pool *);
1.128     christos  201: static void    pool_prime_page(struct pool *, void *,
1.55      thorpej   202:                    struct pool_item_header *);
1.88      chs       203: static void    pool_update_curpage(struct pool *);
1.66      thorpej   204:
1.113     yamt      205: static int     pool_grow(struct pool *, int);
1.117     yamt      206: static void    *pool_allocator_alloc(struct pool *, int);
                    207: static void    pool_allocator_free(struct pool *, void *);
1.3       pk        208:
1.97      yamt      209: static void pool_print_pagelist(struct pool *, struct pool_pagelist *,
1.88      chs       210:        void (*)(const char *, ...));
1.42      thorpej   211: static void pool_print1(struct pool *, const char *,
                    212:        void (*)(const char *, ...));
1.3       pk        213:
1.88      chs       214: static int pool_chk_page(struct pool *, const char *,
                    215:                         struct pool_item_header *);
                    216:
1.3       pk        217: /*
1.52      thorpej   218:  * Pool log entry. An array of these is allocated in pool_init().
1.3       pk        219:  */
                    220: struct pool_log {
                    221:        const char      *pl_file;
                    222:        long            pl_line;
                    223:        int             pl_action;
1.25      thorpej   224: #define        PRLOG_GET       1
                    225: #define        PRLOG_PUT       2
1.3       pk        226:        void            *pl_addr;
1.1       pk        227: };
                    228:
1.86      matt      229: #ifdef POOL_DIAGNOSTIC
1.3       pk        230: /* Number of entries in pool log buffers */
1.17      thorpej   231: #ifndef POOL_LOGSIZE
                    232: #define        POOL_LOGSIZE    10
                    233: #endif
                    234:
                    235: int pool_logsize = POOL_LOGSIZE;
1.1       pk        236:
1.110     perry     237: static inline void
1.42      thorpej   238: pr_log(struct pool *pp, void *v, int action, const char *file, long line)
1.3       pk        239: {
                    240:        int n = pp->pr_curlogentry;
                    241:        struct pool_log *pl;
                    242:
1.20      thorpej   243:        if ((pp->pr_roflags & PR_LOGGING) == 0)
1.3       pk        244:                return;
                    245:
                    246:        /*
                    247:         * Fill in the current entry. Wrap around and overwrite
                    248:         * the oldest entry if necessary.
                    249:         */
                    250:        pl = &pp->pr_log[n];
                    251:        pl->pl_file = file;
                    252:        pl->pl_line = line;
                    253:        pl->pl_action = action;
                    254:        pl->pl_addr = v;
                    255:        if (++n >= pp->pr_logsize)
                    256:                n = 0;
                    257:        pp->pr_curlogentry = n;
                    258: }
                    259:
                    260: static void
1.42      thorpej   261: pr_printlog(struct pool *pp, struct pool_item *pi,
                    262:     void (*pr)(const char *, ...))
1.3       pk        263: {
                    264:        int i = pp->pr_logsize;
                    265:        int n = pp->pr_curlogentry;
                    266:
1.20      thorpej   267:        if ((pp->pr_roflags & PR_LOGGING) == 0)
1.3       pk        268:                return;
                    269:
                    270:        /*
                    271:         * Print all entries in this pool's log.
                    272:         */
                    273:        while (i-- > 0) {
                    274:                struct pool_log *pl = &pp->pr_log[n];
                    275:                if (pl->pl_action != 0) {
1.25      thorpej   276:                        if (pi == NULL || pi == pl->pl_addr) {
                    277:                                (*pr)("\tlog entry %d:\n", i);
                    278:                                (*pr)("\t\taction = %s, addr = %p\n",
                    279:                                    pl->pl_action == PRLOG_GET ? "get" : "put",
                    280:                                    pl->pl_addr);
                    281:                                (*pr)("\t\tfile: %s at line %lu\n",
                    282:                                    pl->pl_file, pl->pl_line);
                    283:                        }
1.3       pk        284:                }
                    285:                if (++n >= pp->pr_logsize)
                    286:                        n = 0;
                    287:        }
                    288: }
1.25      thorpej   289:
1.110     perry     290: static inline void
1.42      thorpej   291: pr_enter(struct pool *pp, const char *file, long line)
1.25      thorpej   292: {
                    293:
1.34      thorpej   294:        if (__predict_false(pp->pr_entered_file != NULL)) {
1.25      thorpej   295:                printf("pool %s: reentrancy at file %s line %ld\n",
                    296:                    pp->pr_wchan, file, line);
                    297:                printf("         previous entry at file %s line %ld\n",
                    298:                    pp->pr_entered_file, pp->pr_entered_line);
                    299:                panic("pr_enter");
                    300:        }
                    301:
                    302:        pp->pr_entered_file = file;
                    303:        pp->pr_entered_line = line;
                    304: }
                    305:
1.110     perry     306: static inline void
1.42      thorpej   307: pr_leave(struct pool *pp)
1.25      thorpej   308: {
                    309:
1.34      thorpej   310:        if (__predict_false(pp->pr_entered_file == NULL)) {
1.25      thorpej   311:                printf("pool %s not entered?\n", pp->pr_wchan);
                    312:                panic("pr_leave");
                    313:        }
                    314:
                    315:        pp->pr_entered_file = NULL;
                    316:        pp->pr_entered_line = 0;
                    317: }
                    318:
1.110     perry     319: static inline void
1.42      thorpej   320: pr_enter_check(struct pool *pp, void (*pr)(const char *, ...))
1.25      thorpej   321: {
                    322:
                    323:        if (pp->pr_entered_file != NULL)
                    324:                (*pr)("\n\tcurrently entered from file %s line %ld\n",
                    325:                    pp->pr_entered_file, pp->pr_entered_line);
                    326: }
1.3       pk        327: #else
1.25      thorpej   328: #define        pr_log(pp, v, action, file, line)
                    329: #define        pr_printlog(pp, pi, pr)
                    330: #define        pr_enter(pp, file, line)
                    331: #define        pr_leave(pp)
                    332: #define        pr_enter_check(pp, pr)
1.59      thorpej   333: #endif /* POOL_DIAGNOSTIC */
1.3       pk        334:
1.135     yamt      335: static inline unsigned int
1.97      yamt      336: pr_item_notouch_index(const struct pool *pp, const struct pool_item_header *ph,
                    337:     const void *v)
                    338: {
                    339:        const char *cp = v;
1.135     yamt      340:        unsigned int idx;
1.97      yamt      341:
                    342:        KASSERT(pp->pr_roflags & PR_NOTOUCH);
1.128     christos  343:        idx = (cp - (char *)ph->ph_page - ph->ph_off) / pp->pr_size;
1.97      yamt      344:        KASSERT(idx < pp->pr_itemsperpage);
                    345:        return idx;
                    346: }
                    347:
1.110     perry     348: static inline void
1.97      yamt      349: pr_item_notouch_put(const struct pool *pp, struct pool_item_header *ph,
                    350:     void *obj)
                    351: {
1.135     yamt      352:        unsigned int idx = pr_item_notouch_index(pp, ph, obj);
                    353:        pool_item_bitmap_t *bitmap = ph->ph_bitmap + (idx / BITMAP_SIZE);
                    354:        pool_item_bitmap_t mask = 1 << (idx & BITMAP_MASK);
1.97      yamt      355:
1.135     yamt      356:        KASSERT((*bitmap & mask) == 0);
                    357:        *bitmap |= mask;
1.97      yamt      358: }
                    359:
1.110     perry     360: static inline void *
1.97      yamt      361: pr_item_notouch_get(const struct pool *pp, struct pool_item_header *ph)
                    362: {
1.135     yamt      363:        pool_item_bitmap_t *bitmap = ph->ph_bitmap;
                    364:        unsigned int idx;
                    365:        int i;
1.97      yamt      366:
1.135     yamt      367:        for (i = 0; ; i++) {
                    368:                int bit;
1.97      yamt      369:
1.135     yamt      370:                KASSERT((i * BITMAP_SIZE) < pp->pr_itemsperpage);
                    371:                bit = ffs32(bitmap[i]);
                    372:                if (bit) {
                    373:                        pool_item_bitmap_t mask;
                    374:
                    375:                        bit--;
                    376:                        idx = (i * BITMAP_SIZE) + bit;
                    377:                        mask = 1 << bit;
                    378:                        KASSERT((bitmap[i] & mask) != 0);
                    379:                        bitmap[i] &= ~mask;
                    380:                        break;
                    381:                }
                    382:        }
                    383:        KASSERT(idx < pp->pr_itemsperpage);
1.128     christos  384:        return (char *)ph->ph_page + ph->ph_off + idx * pp->pr_size;
1.97      yamt      385: }
                    386:
1.135     yamt      387: static inline void
                    388: pr_item_notouch_init(const struct pool *pp, struct pool_item_header *ph)
                    389: {
                    390:        pool_item_bitmap_t *bitmap = ph->ph_bitmap;
                    391:        const int n = howmany(pp->pr_itemsperpage, BITMAP_SIZE);
                    392:        int i;
                    393:
                    394:        for (i = 0; i < n; i++) {
                    395:                bitmap[i] = (pool_item_bitmap_t)-1;
                    396:        }
                    397: }
                    398:
1.110     perry     399: static inline int
1.88      chs       400: phtree_compare(struct pool_item_header *a, struct pool_item_header *b)
                    401: {
1.121     yamt      402:
                    403:        /*
                    404:         * we consider pool_item_header with smaller ph_page bigger.
                    405:         * (this unnatural ordering is for the benefit of pr_find_pagehead.)
                    406:         */
                    407:
1.88      chs       408:        if (a->ph_page < b->ph_page)
1.121     yamt      409:                return (1);
                    410:        else if (a->ph_page > b->ph_page)
1.88      chs       411:                return (-1);
                    412:        else
                    413:                return (0);
                    414: }
                    415:
                    416: SPLAY_PROTOTYPE(phtree, pool_item_header, ph_node, phtree_compare);
                    417: SPLAY_GENERATE(phtree, pool_item_header, ph_node, phtree_compare);
                    418:
1.3       pk        419: /*
1.121     yamt      420:  * Return the pool page header based on item address.
1.3       pk        421:  */
1.110     perry     422: static inline struct pool_item_header *
1.121     yamt      423: pr_find_pagehead(struct pool *pp, void *v)
1.3       pk        424: {
1.88      chs       425:        struct pool_item_header *ph, tmp;
1.3       pk        426:
1.121     yamt      427:        if ((pp->pr_roflags & PR_NOALIGN) != 0) {
1.128     christos  428:                tmp.ph_page = (void *)(uintptr_t)v;
1.121     yamt      429:                ph = SPLAY_FIND(phtree, &pp->pr_phtree, &tmp);
                    430:                if (ph == NULL) {
                    431:                        ph = SPLAY_ROOT(&pp->pr_phtree);
                    432:                        if (ph != NULL && phtree_compare(&tmp, ph) >= 0) {
                    433:                                ph = SPLAY_NEXT(phtree, &pp->pr_phtree, ph);
                    434:                        }
                    435:                        KASSERT(ph == NULL || phtree_compare(&tmp, ph) < 0);
                    436:                }
                    437:        } else {
1.128     christos  438:                void *page =
                    439:                    (void *)((uintptr_t)v & pp->pr_alloc->pa_pagemask);
1.121     yamt      440:
                    441:                if ((pp->pr_roflags & PR_PHINPAGE) != 0) {
1.128     christos  442:                        ph = (struct pool_item_header *)((char *)page + pp->pr_phoffset);
1.121     yamt      443:                } else {
                    444:                        tmp.ph_page = page;
                    445:                        ph = SPLAY_FIND(phtree, &pp->pr_phtree, &tmp);
                    446:                }
                    447:        }
1.3       pk        448:
1.121     yamt      449:        KASSERT(ph == NULL || ((pp->pr_roflags & PR_PHINPAGE) != 0) ||
1.128     christos  450:            ((char *)ph->ph_page <= (char *)v &&
                    451:            (char *)v < (char *)ph->ph_page + pp->pr_alloc->pa_pagesz));
1.88      chs       452:        return ph;
1.3       pk        453: }
                    454:
1.101     thorpej   455: static void
                    456: pr_pagelist_free(struct pool *pp, struct pool_pagelist *pq)
                    457: {
                    458:        struct pool_item_header *ph;
                    459:
                    460:        while ((ph = LIST_FIRST(pq)) != NULL) {
                    461:                LIST_REMOVE(ph, ph_pagelist);
                    462:                pool_allocator_free(pp, ph->ph_page);
1.134     ad        463:                if ((pp->pr_roflags & PR_PHINPAGE) == 0)
1.101     thorpej   464:                        pool_put(pp->pr_phpool, ph);
                    465:        }
                    466: }
                    467:
1.3       pk        468: /*
                    469:  * Remove a page from the pool.
                    470:  */
1.110     perry     471: static inline void
1.61      chs       472: pr_rmpage(struct pool *pp, struct pool_item_header *ph,
                    473:      struct pool_pagelist *pq)
1.3       pk        474: {
                    475:
1.134     ad        476:        KASSERT(mutex_owned(&pp->pr_lock));
1.91      yamt      477:
1.3       pk        478:        /*
1.7       thorpej   479:         * If the page was idle, decrement the idle page count.
1.3       pk        480:         */
1.6       thorpej   481:        if (ph->ph_nmissing == 0) {
                    482: #ifdef DIAGNOSTIC
                    483:                if (pp->pr_nidle == 0)
                    484:                        panic("pr_rmpage: nidle inconsistent");
1.20      thorpej   485:                if (pp->pr_nitems < pp->pr_itemsperpage)
                    486:                        panic("pr_rmpage: nitems inconsistent");
1.6       thorpej   487: #endif
                    488:                pp->pr_nidle--;
                    489:        }
1.7       thorpej   490:
1.20      thorpej   491:        pp->pr_nitems -= pp->pr_itemsperpage;
                    492:
1.7       thorpej   493:        /*
1.101     thorpej   494:         * Unlink the page from the pool and queue it for release.
1.7       thorpej   495:         */
1.88      chs       496:        LIST_REMOVE(ph, ph_pagelist);
1.91      yamt      497:        if ((pp->pr_roflags & PR_PHINPAGE) == 0)
                    498:                SPLAY_REMOVE(phtree, &pp->pr_phtree, ph);
1.101     thorpej   499:        LIST_INSERT_HEAD(pq, ph, ph_pagelist);
                    500:
1.7       thorpej   501:        pp->pr_npages--;
                    502:        pp->pr_npagefree++;
1.6       thorpej   503:
1.88      chs       504:        pool_update_curpage(pp);
1.3       pk        505: }
                    506:
1.126     thorpej   507: static bool
1.117     yamt      508: pa_starved_p(struct pool_allocator *pa)
                    509: {
                    510:
                    511:        if (pa->pa_backingmap != NULL) {
                    512:                return vm_map_starved_p(pa->pa_backingmap);
                    513:        }
1.127     thorpej   514:        return false;
1.117     yamt      515: }
                    516:
                    517: static int
1.124     yamt      518: pool_reclaim_callback(struct callback_entry *ce, void *obj, void *arg)
1.117     yamt      519: {
                    520:        struct pool *pp = obj;
                    521:        struct pool_allocator *pa = pp->pr_alloc;
                    522:
                    523:        KASSERT(&pp->pr_reclaimerentry == ce);
                    524:        pool_reclaim(pp);
                    525:        if (!pa_starved_p(pa)) {
                    526:                return CALLBACK_CHAIN_ABORT;
                    527:        }
                    528:        return CALLBACK_CHAIN_CONTINUE;
                    529: }
                    530:
                    531: static void
                    532: pool_reclaim_register(struct pool *pp)
                    533: {
                    534:        struct vm_map *map = pp->pr_alloc->pa_backingmap;
                    535:        int s;
                    536:
                    537:        if (map == NULL) {
                    538:                return;
                    539:        }
                    540:
                    541:        s = splvm(); /* not necessary for INTRSAFE maps, but don't care. */
                    542:        callback_register(&vm_map_to_kernel(map)->vmk_reclaim_callback,
                    543:            &pp->pr_reclaimerentry, pp, pool_reclaim_callback);
                    544:        splx(s);
                    545: }
                    546:
                    547: static void
                    548: pool_reclaim_unregister(struct pool *pp)
                    549: {
                    550:        struct vm_map *map = pp->pr_alloc->pa_backingmap;
                    551:        int s;
                    552:
                    553:        if (map == NULL) {
                    554:                return;
                    555:        }
                    556:
                    557:        s = splvm(); /* not necessary for INTRSAFE maps, but don't care. */
                    558:        callback_unregister(&vm_map_to_kernel(map)->vmk_reclaim_callback,
                    559:            &pp->pr_reclaimerentry);
                    560:        splx(s);
                    561: }
                    562:
                    563: static void
                    564: pa_reclaim_register(struct pool_allocator *pa)
                    565: {
                    566:        struct vm_map *map = *pa->pa_backingmapptr;
                    567:        struct pool *pp;
                    568:
                    569:        KASSERT(pa->pa_backingmap == NULL);
                    570:        if (map == NULL) {
                    571:                SLIST_INSERT_HEAD(&pa_deferinitq, pa, pa_q);
                    572:                return;
                    573:        }
                    574:        pa->pa_backingmap = map;
                    575:        TAILQ_FOREACH(pp, &pa->pa_list, pr_alloc_list) {
                    576:                pool_reclaim_register(pp);
                    577:        }
                    578: }
                    579:
1.3       pk        580: /*
1.94      simonb    581:  * Initialize all the pools listed in the "pools" link set.
                    582:  */
                    583: void
1.117     yamt      584: pool_subsystem_init(void)
1.94      simonb    585: {
1.117     yamt      586:        struct pool_allocator *pa;
1.94      simonb    587:        __link_set_decl(pools, struct link_pool_init);
                    588:        struct link_pool_init * const *pi;
                    589:
1.134     ad        590:        mutex_init(&pool_head_lock, MUTEX_DEFAULT, IPL_NONE);
                    591:        cv_init(&pool_busy, "poolbusy");
                    592:
1.94      simonb    593:        __link_set_foreach(pi, pools)
                    594:                pool_init((*pi)->pp, (*pi)->size, (*pi)->align,
                    595:                    (*pi)->align_offset, (*pi)->flags, (*pi)->wchan,
1.129     ad        596:                    (*pi)->palloc, (*pi)->ipl);
1.117     yamt      597:
                    598:        while ((pa = SLIST_FIRST(&pa_deferinitq)) != NULL) {
                    599:                KASSERT(pa->pa_backingmapptr != NULL);
                    600:                KASSERT(*pa->pa_backingmapptr != NULL);
                    601:                SLIST_REMOVE_HEAD(&pa_deferinitq, pa_q);
                    602:                pa_reclaim_register(pa);
                    603:        }
1.134     ad        604:
                    605:        pool_init(&cache_pool, sizeof(struct pool_cache), CACHE_LINE_SIZE,
                    606:            0, 0, "pcache", &pool_allocator_nointr, IPL_NONE);
                    607:
                    608:        pool_init(&cache_cpu_pool, sizeof(pool_cache_cpu_t), CACHE_LINE_SIZE,
                    609:            0, 0, "pcachecpu", &pool_allocator_nointr, IPL_NONE);
1.94      simonb    610: }
                    611:
                    612: /*
1.3       pk        613:  * Initialize the given pool resource structure.
                    614:  *
                    615:  * We export this routine to allow other kernel parts to declare
                    616:  * static pools that must be initialized before malloc() is available.
                    617:  */
                    618: void
1.42      thorpej   619: pool_init(struct pool *pp, size_t size, u_int align, u_int ioff, int flags,
1.129     ad        620:     const char *wchan, struct pool_allocator *palloc, int ipl)
1.3       pk        621: {
1.116     simonb    622:        struct pool *pp1;
1.92      enami     623:        size_t trysize, phsize;
1.134     ad        624:        int off, slack;
1.3       pk        625:
1.116     simonb    626: #ifdef DEBUG
                    627:        /*
                    628:         * Check that the pool hasn't already been initialised and
                    629:         * added to the list of all pools.
                    630:         */
1.137.2.3  ad        631:        TAILQ_FOREACH(pp1, &pool_head, pr_poollist) {
1.116     simonb    632:                if (pp == pp1)
                    633:                        panic("pool_init: pool %s already initialised",
                    634:                            wchan);
                    635:        }
                    636: #endif
                    637:
1.25      thorpej   638: #ifdef POOL_DIAGNOSTIC
                    639:        /*
                    640:         * Always log if POOL_DIAGNOSTIC is defined.
                    641:         */
                    642:        if (pool_logsize != 0)
                    643:                flags |= PR_LOGGING;
                    644: #endif
                    645:
1.66      thorpej   646:        if (palloc == NULL)
                    647:                palloc = &pool_allocator_kmem;
1.112     bjh21     648: #ifdef POOL_SUBPAGE
                    649:        if (size > palloc->pa_pagesz) {
                    650:                if (palloc == &pool_allocator_kmem)
                    651:                        palloc = &pool_allocator_kmem_fullpage;
                    652:                else if (palloc == &pool_allocator_nointr)
                    653:                        palloc = &pool_allocator_nointr_fullpage;
                    654:        }
1.66      thorpej   655: #endif /* POOL_SUBPAGE */
                    656:        if ((palloc->pa_flags & PA_INITIALIZED) == 0) {
1.112     bjh21     657:                if (palloc->pa_pagesz == 0)
1.66      thorpej   658:                        palloc->pa_pagesz = PAGE_SIZE;
                    659:
                    660:                TAILQ_INIT(&palloc->pa_list);
                    661:
1.134     ad        662:                mutex_init(&palloc->pa_lock, MUTEX_DEFAULT, IPL_VM);
1.66      thorpej   663:                palloc->pa_pagemask = ~(palloc->pa_pagesz - 1);
                    664:                palloc->pa_pageshift = ffs(palloc->pa_pagesz) - 1;
1.117     yamt      665:
                    666:                if (palloc->pa_backingmapptr != NULL) {
                    667:                        pa_reclaim_register(palloc);
                    668:                }
1.66      thorpej   669:                palloc->pa_flags |= PA_INITIALIZED;
1.4       thorpej   670:        }
1.3       pk        671:
                    672:        if (align == 0)
                    673:                align = ALIGN(1);
1.14      thorpej   674:
1.120     yamt      675:        if ((flags & PR_NOTOUCH) == 0 && size < sizeof(struct pool_item))
1.14      thorpej   676:                size = sizeof(struct pool_item);
1.3       pk        677:
1.78      thorpej   678:        size = roundup(size, align);
1.66      thorpej   679: #ifdef DIAGNOSTIC
                    680:        if (size > palloc->pa_pagesz)
1.121     yamt      681:                panic("pool_init: pool item size (%zu) too large", size);
1.66      thorpej   682: #endif
1.35      pk        683:
1.3       pk        684:        /*
                    685:         * Initialize the pool structure.
                    686:         */
1.88      chs       687:        LIST_INIT(&pp->pr_emptypages);
                    688:        LIST_INIT(&pp->pr_fullpages);
                    689:        LIST_INIT(&pp->pr_partpages);
1.134     ad        690:        pp->pr_cache = NULL;
1.3       pk        691:        pp->pr_curpage = NULL;
                    692:        pp->pr_npages = 0;
                    693:        pp->pr_minitems = 0;
                    694:        pp->pr_minpages = 0;
                    695:        pp->pr_maxpages = UINT_MAX;
1.20      thorpej   696:        pp->pr_roflags = flags;
                    697:        pp->pr_flags = 0;
1.35      pk        698:        pp->pr_size = size;
1.3       pk        699:        pp->pr_align = align;
                    700:        pp->pr_wchan = wchan;
1.66      thorpej   701:        pp->pr_alloc = palloc;
1.20      thorpej   702:        pp->pr_nitems = 0;
                    703:        pp->pr_nout = 0;
                    704:        pp->pr_hardlimit = UINT_MAX;
                    705:        pp->pr_hardlimit_warning = NULL;
1.31      thorpej   706:        pp->pr_hardlimit_ratecap.tv_sec = 0;
                    707:        pp->pr_hardlimit_ratecap.tv_usec = 0;
                    708:        pp->pr_hardlimit_warning_last.tv_sec = 0;
                    709:        pp->pr_hardlimit_warning_last.tv_usec = 0;
1.68      thorpej   710:        pp->pr_drain_hook = NULL;
                    711:        pp->pr_drain_hook_arg = NULL;
1.125     ad        712:        pp->pr_freecheck = NULL;
1.3       pk        713:
                    714:        /*
                    715:         * Decide whether to put the page header off page to avoid
1.92      enami     716:         * wasting too large a part of the page or too big item.
                    717:         * Off-page page headers go on a hash table, so we can match
                    718:         * a returned item with its header based on the page address.
                    719:         * We use 1/16 of the page size and about 8 times of the item
                    720:         * size as the threshold (XXX: tune)
                    721:         *
                    722:         * However, we'll put the header into the page if we can put
                    723:         * it without wasting any items.
                    724:         *
                    725:         * Silently enforce `0 <= ioff < align'.
1.3       pk        726:         */
1.92      enami     727:        pp->pr_itemoffset = ioff %= align;
                    728:        /* See the comment below about reserved bytes. */
                    729:        trysize = palloc->pa_pagesz - ((align - ioff) % align);
                    730:        phsize = ALIGN(sizeof(struct pool_item_header));
1.121     yamt      731:        if ((pp->pr_roflags & (PR_NOTOUCH | PR_NOALIGN)) == 0 &&
1.97      yamt      732:            (pp->pr_size < MIN(palloc->pa_pagesz / 16, phsize << 3) ||
                    733:            trysize / pp->pr_size == (trysize - phsize) / pp->pr_size)) {
1.3       pk        734:                /* Use the end of the page for the page header */
1.20      thorpej   735:                pp->pr_roflags |= PR_PHINPAGE;
1.92      enami     736:                pp->pr_phoffset = off = palloc->pa_pagesz - phsize;
1.2       pk        737:        } else {
1.3       pk        738:                /* The page header will be taken from our page header pool */
                    739:                pp->pr_phoffset = 0;
1.66      thorpej   740:                off = palloc->pa_pagesz;
1.88      chs       741:                SPLAY_INIT(&pp->pr_phtree);
1.2       pk        742:        }
1.1       pk        743:
1.3       pk        744:        /*
                    745:         * Alignment is to take place at `ioff' within the item. This means
                    746:         * we must reserve up to `align - 1' bytes on the page to allow
                    747:         * appropriate positioning of each item.
                    748:         */
                    749:        pp->pr_itemsperpage = (off - ((align - ioff) % align)) / pp->pr_size;
1.43      thorpej   750:        KASSERT(pp->pr_itemsperpage != 0);
1.97      yamt      751:        if ((pp->pr_roflags & PR_NOTOUCH)) {
                    752:                int idx;
                    753:
                    754:                for (idx = 0; pp->pr_itemsperpage > PHPOOL_FREELIST_NELEM(idx);
                    755:                    idx++) {
                    756:                        /* nothing */
                    757:                }
                    758:                if (idx >= PHPOOL_MAX) {
                    759:                        /*
                    760:                         * if you see this panic, consider to tweak
                    761:                         * PHPOOL_MAX and PHPOOL_FREELIST_NELEM.
                    762:                         */
                    763:                        panic("%s: too large itemsperpage(%d) for PR_NOTOUCH",
                    764:                            pp->pr_wchan, pp->pr_itemsperpage);
                    765:                }
                    766:                pp->pr_phpool = &phpool[idx];
                    767:        } else if ((pp->pr_roflags & PR_PHINPAGE) == 0) {
                    768:                pp->pr_phpool = &phpool[0];
                    769:        }
                    770: #if defined(DIAGNOSTIC)
                    771:        else {
                    772:                pp->pr_phpool = NULL;
                    773:        }
                    774: #endif
1.3       pk        775:
                    776:        /*
                    777:         * Use the slack between the chunks and the page header
                    778:         * for "cache coloring".
                    779:         */
                    780:        slack = off - pp->pr_itemsperpage * pp->pr_size;
                    781:        pp->pr_maxcolor = (slack / align) * align;
                    782:        pp->pr_curcolor = 0;
                    783:
                    784:        pp->pr_nget = 0;
                    785:        pp->pr_nfail = 0;
                    786:        pp->pr_nput = 0;
                    787:        pp->pr_npagealloc = 0;
                    788:        pp->pr_npagefree = 0;
1.1       pk        789:        pp->pr_hiwat = 0;
1.8       thorpej   790:        pp->pr_nidle = 0;
1.134     ad        791:        pp->pr_refcnt = 0;
1.3       pk        792:
1.59      thorpej   793: #ifdef POOL_DIAGNOSTIC
1.25      thorpej   794:        if (flags & PR_LOGGING) {
                    795:                if (kmem_map == NULL ||
                    796:                    (pp->pr_log = malloc(pool_logsize * sizeof(struct pool_log),
                    797:                     M_TEMP, M_NOWAIT)) == NULL)
1.20      thorpej   798:                        pp->pr_roflags &= ~PR_LOGGING;
1.3       pk        799:                pp->pr_curlogentry = 0;
                    800:                pp->pr_logsize = pool_logsize;
                    801:        }
1.59      thorpej   802: #endif
1.25      thorpej   803:
                    804:        pp->pr_entered_file = NULL;
                    805:        pp->pr_entered_line = 0;
1.3       pk        806:
1.137.2.1  ad        807:        /*
                    808:         * XXXAD hack to prevent IP input processing from blocking.
                    809:         */
                    810:        if (ipl == IPL_SOFTNET) {
                    811:                mutex_init(&pp->pr_lock, MUTEX_DEFAULT, IPL_VM);
                    812:        } else {
                    813:                mutex_init(&pp->pr_lock, MUTEX_DEFAULT, ipl);
                    814:        }
1.134     ad        815:        cv_init(&pp->pr_cv, wchan);
                    816:        pp->pr_ipl = ipl;
1.1       pk        817:
1.3       pk        818:        /*
1.43      thorpej   819:         * Initialize private page header pool and cache magazine pool if we
                    820:         * haven't done so yet.
1.23      thorpej   821:         * XXX LOCKING.
1.3       pk        822:         */
1.97      yamt      823:        if (phpool[0].pr_size == 0) {
                    824:                int idx;
                    825:                for (idx = 0; idx < PHPOOL_MAX; idx++) {
                    826:                        static char phpool_names[PHPOOL_MAX][6+1+6+1];
                    827:                        int nelem;
                    828:                        size_t sz;
                    829:
                    830:                        nelem = PHPOOL_FREELIST_NELEM(idx);
                    831:                        snprintf(phpool_names[idx], sizeof(phpool_names[idx]),
                    832:                            "phpool-%d", nelem);
                    833:                        sz = sizeof(struct pool_item_header);
                    834:                        if (nelem) {
1.135     yamt      835:                                sz = offsetof(struct pool_item_header,
                    836:                                    ph_bitmap[howmany(nelem, BITMAP_SIZE)]);
1.97      yamt      837:                        }
                    838:                        pool_init(&phpool[idx], sz, 0, 0, 0,
1.129     ad        839:                            phpool_names[idx], &pool_allocator_meta, IPL_VM);
1.97      yamt      840:                }
1.62      bjh21     841: #ifdef POOL_SUBPAGE
                    842:                pool_init(&psppool, POOL_SUBPAGE, POOL_SUBPAGE, 0,
1.129     ad        843:                    PR_RECURSIVE, "psppool", &pool_allocator_meta, IPL_VM);
1.62      bjh21     844: #endif
1.134     ad        845:                pool_init(&pcgpool, sizeof(pcg_t), CACHE_LINE_SIZE, 0, 0,
                    846:                    "cachegrp", &pool_allocator_meta, IPL_VM);
1.1       pk        847:        }
                    848:
1.137.2.3  ad        849:        /* Insert into the list of all pools. */
                    850:        if (__predict_true(!cold))
1.134     ad        851:                mutex_enter(&pool_head_lock);
1.137.2.3  ad        852:        TAILQ_FOREACH(pp1, &pool_head, pr_poollist) {
                    853:                if (strcmp(pp1->pr_wchan, pp->pr_wchan) > 0)
                    854:                        break;
                    855:        }
                    856:        if (pp1 == NULL)
                    857:                TAILQ_INSERT_TAIL(&pool_head, pp, pr_poollist);
                    858:        else
                    859:                TAILQ_INSERT_BEFORE(pp1, pp, pr_poollist);
                    860:        if (__predict_true(!cold))
1.134     ad        861:                mutex_exit(&pool_head_lock);
                    862:
                    863:                /* Insert this into the list of pools using this allocator. */
1.137.2.3  ad        864:        if (__predict_true(!cold))
1.134     ad        865:                mutex_enter(&palloc->pa_lock);
1.137.2.3  ad        866:        TAILQ_INSERT_TAIL(&palloc->pa_list, pp, pr_alloc_list);
                    867:        if (__predict_true(!cold))
1.134     ad        868:                mutex_exit(&palloc->pa_lock);
1.66      thorpej   869:
1.117     yamt      870:        pool_reclaim_register(pp);
1.1       pk        871: }
                    872:
                    873: /*
                    874:  * De-commision a pool resource.
                    875:  */
                    876: void
1.42      thorpej   877: pool_destroy(struct pool *pp)
1.1       pk        878: {
1.101     thorpej   879:        struct pool_pagelist pq;
1.3       pk        880:        struct pool_item_header *ph;
1.43      thorpej   881:
1.101     thorpej   882:        /* Remove from global pool list */
1.134     ad        883:        mutex_enter(&pool_head_lock);
                    884:        while (pp->pr_refcnt != 0)
                    885:                cv_wait(&pool_busy, &pool_head_lock);
1.137.2.3  ad        886:        TAILQ_REMOVE(&pool_head, pp, pr_poollist);
1.101     thorpej   887:        if (drainpp == pp)
                    888:                drainpp = NULL;
1.134     ad        889:        mutex_exit(&pool_head_lock);
1.101     thorpej   890:
                    891:        /* Remove this pool from its allocator's list of pools. */
1.117     yamt      892:        pool_reclaim_unregister(pp);
1.134     ad        893:        mutex_enter(&pp->pr_alloc->pa_lock);
1.66      thorpej   894:        TAILQ_REMOVE(&pp->pr_alloc->pa_list, pp, pr_alloc_list);
1.134     ad        895:        mutex_exit(&pp->pr_alloc->pa_lock);
1.66      thorpej   896:
1.134     ad        897:        mutex_enter(&pp->pr_lock);
1.101     thorpej   898:
1.134     ad        899:        KASSERT(pp->pr_cache == NULL);
1.3       pk        900:
                    901: #ifdef DIAGNOSTIC
1.20      thorpej   902:        if (pp->pr_nout != 0) {
1.25      thorpej   903:                pr_printlog(pp, NULL, printf);
1.80      provos    904:                panic("pool_destroy: pool busy: still out: %u",
1.20      thorpej   905:                    pp->pr_nout);
1.3       pk        906:        }
                    907: #endif
1.1       pk        908:
1.101     thorpej   909:        KASSERT(LIST_EMPTY(&pp->pr_fullpages));
                    910:        KASSERT(LIST_EMPTY(&pp->pr_partpages));
                    911:
1.3       pk        912:        /* Remove all pages */
1.101     thorpej   913:        LIST_INIT(&pq);
1.88      chs       914:        while ((ph = LIST_FIRST(&pp->pr_emptypages)) != NULL)
1.101     thorpej   915:                pr_rmpage(pp, ph, &pq);
                    916:
1.134     ad        917:        mutex_exit(&pp->pr_lock);
1.3       pk        918:
1.101     thorpej   919:        pr_pagelist_free(pp, &pq);
1.3       pk        920:
1.59      thorpej   921: #ifdef POOL_DIAGNOSTIC
1.20      thorpej   922:        if ((pp->pr_roflags & PR_LOGGING) != 0)
1.3       pk        923:                free(pp->pr_log, M_TEMP);
1.59      thorpej   924: #endif
1.134     ad        925:
                    926:        cv_destroy(&pp->pr_cv);
                    927:        mutex_destroy(&pp->pr_lock);
1.1       pk        928: }
                    929:
1.68      thorpej   930: void
                    931: pool_set_drain_hook(struct pool *pp, void (*fn)(void *, int), void *arg)
                    932: {
                    933:
                    934:        /* XXX no locking -- must be used just after pool_init() */
                    935: #ifdef DIAGNOSTIC
                    936:        if (pp->pr_drain_hook != NULL)
                    937:                panic("pool_set_drain_hook(%s): already set", pp->pr_wchan);
                    938: #endif
                    939:        pp->pr_drain_hook = fn;
                    940:        pp->pr_drain_hook_arg = arg;
                    941: }
                    942:
1.88      chs       943: static struct pool_item_header *
1.128     christos  944: pool_alloc_item_header(struct pool *pp, void *storage, int flags)
1.55      thorpej   945: {
                    946:        struct pool_item_header *ph;
                    947:
                    948:        if ((pp->pr_roflags & PR_PHINPAGE) != 0)
1.128     christos  949:                ph = (struct pool_item_header *) ((char *)storage + pp->pr_phoffset);
1.134     ad        950:        else
1.97      yamt      951:                ph = pool_get(pp->pr_phpool, flags);
1.55      thorpej   952:
                    953:        return (ph);
                    954: }
1.1       pk        955:
                    956: /*
1.134     ad        957:  * Grab an item from the pool.
1.1       pk        958:  */
1.3       pk        959: void *
1.59      thorpej   960: #ifdef POOL_DIAGNOSTIC
1.42      thorpej   961: _pool_get(struct pool *pp, int flags, const char *file, long line)
1.56      sommerfe  962: #else
                    963: pool_get(struct pool *pp, int flags)
                    964: #endif
1.1       pk        965: {
                    966:        struct pool_item *pi;
1.3       pk        967:        struct pool_item_header *ph;
1.55      thorpej   968:        void *v;
1.1       pk        969:
1.2       pk        970: #ifdef DIAGNOSTIC
1.95      atatat    971:        if (__predict_false(pp->pr_itemsperpage == 0))
                    972:                panic("pool_get: pool %p: pr_itemsperpage is zero, "
                    973:                    "pool not initialized?", pp);
1.84      thorpej   974:        if (__predict_false(curlwp == NULL && doing_shutdown == 0 &&
1.37      sommerfe  975:                            (flags & PR_WAITOK) != 0))
1.77      matt      976:                panic("pool_get: %s: must have NOWAIT", pp->pr_wchan);
1.58      thorpej   977:
1.102     chs       978: #endif /* DIAGNOSTIC */
1.58      thorpej   979: #ifdef LOCKDEBUG
                    980:        if (flags & PR_WAITOK)
1.119     yamt      981:                ASSERT_SLEEPABLE(NULL, "pool_get(PR_WAITOK)");
1.56      sommerfe  982: #endif
1.1       pk        983:
1.134     ad        984:        mutex_enter(&pp->pr_lock);
1.25      thorpej   985:        pr_enter(pp, file, line);
1.20      thorpej   986:
                    987:  startover:
                    988:        /*
                    989:         * Check to see if we've reached the hard limit.  If we have,
                    990:         * and we can wait, then wait until an item has been returned to
                    991:         * the pool.
                    992:         */
                    993: #ifdef DIAGNOSTIC
1.34      thorpej   994:        if (__predict_false(pp->pr_nout > pp->pr_hardlimit)) {
1.25      thorpej   995:                pr_leave(pp);
1.134     ad        996:                mutex_exit(&pp->pr_lock);
1.20      thorpej   997:                panic("pool_get: %s: crossed hard limit", pp->pr_wchan);
                    998:        }
                    999: #endif
1.34      thorpej  1000:        if (__predict_false(pp->pr_nout == pp->pr_hardlimit)) {
1.68      thorpej  1001:                if (pp->pr_drain_hook != NULL) {
                   1002:                        /*
                   1003:                         * Since the drain hook is going to free things
                   1004:                         * back to the pool, unlock, call the hook, re-lock,
                   1005:                         * and check the hardlimit condition again.
                   1006:                         */
                   1007:                        pr_leave(pp);
1.134     ad       1008:                        mutex_exit(&pp->pr_lock);
1.68      thorpej  1009:                        (*pp->pr_drain_hook)(pp->pr_drain_hook_arg, flags);
1.134     ad       1010:                        mutex_enter(&pp->pr_lock);
1.68      thorpej  1011:                        pr_enter(pp, file, line);
                   1012:                        if (pp->pr_nout < pp->pr_hardlimit)
                   1013:                                goto startover;
                   1014:                }
                   1015:
1.29      sommerfe 1016:                if ((flags & PR_WAITOK) && !(flags & PR_LIMITFAIL)) {
1.20      thorpej  1017:                        /*
                   1018:                         * XXX: A warning isn't logged in this case.  Should
                   1019:                         * it be?
                   1020:                         */
                   1021:                        pp->pr_flags |= PR_WANTED;
1.25      thorpej  1022:                        pr_leave(pp);
1.134     ad       1023:                        cv_wait(&pp->pr_cv, &pp->pr_lock);
1.25      thorpej  1024:                        pr_enter(pp, file, line);
1.20      thorpej  1025:                        goto startover;
                   1026:                }
1.31      thorpej  1027:
                   1028:                /*
                   1029:                 * Log a message that the hard limit has been hit.
                   1030:                 */
                   1031:                if (pp->pr_hardlimit_warning != NULL &&
                   1032:                    ratecheck(&pp->pr_hardlimit_warning_last,
                   1033:                              &pp->pr_hardlimit_ratecap))
                   1034:                        log(LOG_ERR, "%s\n", pp->pr_hardlimit_warning);
1.21      thorpej  1035:
                   1036:                pp->pr_nfail++;
                   1037:
1.25      thorpej  1038:                pr_leave(pp);
1.134     ad       1039:                mutex_exit(&pp->pr_lock);
1.20      thorpej  1040:                return (NULL);
                   1041:        }
                   1042:
1.3       pk       1043:        /*
                   1044:         * The convention we use is that if `curpage' is not NULL, then
                   1045:         * it points at a non-empty bucket. In particular, `curpage'
                   1046:         * never points at a page header which has PR_PHINPAGE set and
                   1047:         * has no items in its bucket.
                   1048:         */
1.20      thorpej  1049:        if ((ph = pp->pr_curpage) == NULL) {
1.113     yamt     1050:                int error;
                   1051:
1.20      thorpej  1052: #ifdef DIAGNOSTIC
                   1053:                if (pp->pr_nitems != 0) {
1.134     ad       1054:                        mutex_exit(&pp->pr_lock);
1.20      thorpej  1055:                        printf("pool_get: %s: curpage NULL, nitems %u\n",
                   1056:                            pp->pr_wchan, pp->pr_nitems);
1.80      provos   1057:                        panic("pool_get: nitems inconsistent");
1.20      thorpej  1058:                }
                   1059: #endif
                   1060:
1.21      thorpej  1061:                /*
                   1062:                 * Call the back-end page allocator for more memory.
                   1063:                 * Release the pool lock, as the back-end page allocator
                   1064:                 * may block.
                   1065:                 */
1.25      thorpej  1066:                pr_leave(pp);
1.113     yamt     1067:                error = pool_grow(pp, flags);
                   1068:                pr_enter(pp, file, line);
                   1069:                if (error != 0) {
1.21      thorpej  1070:                        /*
1.55      thorpej  1071:                         * We were unable to allocate a page or item
                   1072:                         * header, but we released the lock during
                   1073:                         * allocation, so perhaps items were freed
                   1074:                         * back to the pool.  Check for this case.
1.21      thorpej  1075:                         */
                   1076:                        if (pp->pr_curpage != NULL)
                   1077:                                goto startover;
1.15      pk       1078:
1.117     yamt     1079:                        pp->pr_nfail++;
1.25      thorpej  1080:                        pr_leave(pp);
1.134     ad       1081:                        mutex_exit(&pp->pr_lock);
1.117     yamt     1082:                        return (NULL);
1.1       pk       1083:                }
1.3       pk       1084:
1.20      thorpej  1085:                /* Start the allocation process over. */
                   1086:                goto startover;
1.3       pk       1087:        }
1.97      yamt     1088:        if (pp->pr_roflags & PR_NOTOUCH) {
                   1089: #ifdef DIAGNOSTIC
                   1090:                if (__predict_false(ph->ph_nmissing == pp->pr_itemsperpage)) {
                   1091:                        pr_leave(pp);
1.134     ad       1092:                        mutex_exit(&pp->pr_lock);
1.97      yamt     1093:                        panic("pool_get: %s: page empty", pp->pr_wchan);
                   1094:                }
                   1095: #endif
                   1096:                v = pr_item_notouch_get(pp, ph);
                   1097: #ifdef POOL_DIAGNOSTIC
                   1098:                pr_log(pp, v, PRLOG_GET, file, line);
                   1099: #endif
                   1100:        } else {
1.102     chs      1101:                v = pi = LIST_FIRST(&ph->ph_itemlist);
1.97      yamt     1102:                if (__predict_false(v == NULL)) {
                   1103:                        pr_leave(pp);
1.134     ad       1104:                        mutex_exit(&pp->pr_lock);
1.97      yamt     1105:                        panic("pool_get: %s: page empty", pp->pr_wchan);
                   1106:                }
1.20      thorpej  1107: #ifdef DIAGNOSTIC
1.97      yamt     1108:                if (__predict_false(pp->pr_nitems == 0)) {
                   1109:                        pr_leave(pp);
1.134     ad       1110:                        mutex_exit(&pp->pr_lock);
1.97      yamt     1111:                        printf("pool_get: %s: items on itemlist, nitems %u\n",
                   1112:                            pp->pr_wchan, pp->pr_nitems);
                   1113:                        panic("pool_get: nitems inconsistent");
                   1114:                }
1.65      enami    1115: #endif
1.56      sommerfe 1116:
1.65      enami    1117: #ifdef POOL_DIAGNOSTIC
1.97      yamt     1118:                pr_log(pp, v, PRLOG_GET, file, line);
1.65      enami    1119: #endif
1.3       pk       1120:
1.65      enami    1121: #ifdef DIAGNOSTIC
1.97      yamt     1122:                if (__predict_false(pi->pi_magic != PI_MAGIC)) {
                   1123:                        pr_printlog(pp, pi, printf);
                   1124:                        panic("pool_get(%s): free list modified: "
                   1125:                            "magic=%x; page %p; item addr %p\n",
                   1126:                            pp->pr_wchan, pi->pi_magic, ph->ph_page, pi);
                   1127:                }
1.3       pk       1128: #endif
                   1129:
1.97      yamt     1130:                /*
                   1131:                 * Remove from item list.
                   1132:                 */
1.102     chs      1133:                LIST_REMOVE(pi, pi_list);
1.97      yamt     1134:        }
1.20      thorpej  1135:        pp->pr_nitems--;
                   1136:        pp->pr_nout++;
1.6       thorpej  1137:        if (ph->ph_nmissing == 0) {
                   1138: #ifdef DIAGNOSTIC
1.34      thorpej  1139:                if (__predict_false(pp->pr_nidle == 0))
1.6       thorpej  1140:                        panic("pool_get: nidle inconsistent");
                   1141: #endif
                   1142:                pp->pr_nidle--;
1.88      chs      1143:
                   1144:                /*
                   1145:                 * This page was previously empty.  Move it to the list of
                   1146:                 * partially-full pages.  This page is already curpage.
                   1147:                 */
                   1148:                LIST_REMOVE(ph, ph_pagelist);
                   1149:                LIST_INSERT_HEAD(&pp->pr_partpages, ph, ph_pagelist);
1.6       thorpej  1150:        }
1.3       pk       1151:        ph->ph_nmissing++;
1.97      yamt     1152:        if (ph->ph_nmissing == pp->pr_itemsperpage) {
1.21      thorpej  1153: #ifdef DIAGNOSTIC
1.97      yamt     1154:                if (__predict_false((pp->pr_roflags & PR_NOTOUCH) == 0 &&
1.102     chs      1155:                    !LIST_EMPTY(&ph->ph_itemlist))) {
1.25      thorpej  1156:                        pr_leave(pp);
1.134     ad       1157:                        mutex_exit(&pp->pr_lock);
1.21      thorpej  1158:                        panic("pool_get: %s: nmissing inconsistent",
                   1159:                            pp->pr_wchan);
                   1160:                }
                   1161: #endif
1.3       pk       1162:                /*
1.88      chs      1163:                 * This page is now full.  Move it to the full list
                   1164:                 * and select a new current page.
1.3       pk       1165:                 */
1.88      chs      1166:                LIST_REMOVE(ph, ph_pagelist);
                   1167:                LIST_INSERT_HEAD(&pp->pr_fullpages, ph, ph_pagelist);
                   1168:                pool_update_curpage(pp);
1.1       pk       1169:        }
1.3       pk       1170:
                   1171:        pp->pr_nget++;
1.111     christos 1172:        pr_leave(pp);
1.20      thorpej  1173:
                   1174:        /*
                   1175:         * If we have a low water mark and we are now below that low
                   1176:         * water mark, add more items to the pool.
                   1177:         */
1.53      thorpej  1178:        if (POOL_NEEDS_CATCHUP(pp) && pool_catchup(pp) != 0) {
1.20      thorpej  1179:                /*
                   1180:                 * XXX: Should we log a warning?  Should we set up a timeout
                   1181:                 * to try again in a second or so?  The latter could break
                   1182:                 * a caller's assumptions about interrupt protection, etc.
                   1183:                 */
                   1184:        }
                   1185:
1.134     ad       1186:        mutex_exit(&pp->pr_lock);
1.125     ad       1187:        KASSERT((((vaddr_t)v + pp->pr_itemoffset) & (pp->pr_align - 1)) == 0);
                   1188:        FREECHECK_OUT(&pp->pr_freecheck, v);
1.1       pk       1189:        return (v);
                   1190: }
                   1191:
                   1192: /*
1.43      thorpej  1193:  * Internal version of pool_put().  Pool is already locked/entered.
1.1       pk       1194:  */
1.43      thorpej  1195: static void
1.101     thorpej  1196: pool_do_put(struct pool *pp, void *v, struct pool_pagelist *pq)
1.1       pk       1197: {
                   1198:        struct pool_item *pi = v;
1.3       pk       1199:        struct pool_item_header *ph;
                   1200:
1.134     ad       1201:        KASSERT(mutex_owned(&pp->pr_lock));
1.125     ad       1202:        FREECHECK_IN(&pp->pr_freecheck, v);
1.134     ad       1203:        LOCKDEBUG_MEM_CHECK(v, pp->pr_size);
1.61      chs      1204:
1.30      thorpej  1205: #ifdef DIAGNOSTIC
1.34      thorpej  1206:        if (__predict_false(pp->pr_nout == 0)) {
1.30      thorpej  1207:                printf("pool %s: putting with none out\n",
                   1208:                    pp->pr_wchan);
                   1209:                panic("pool_put");
                   1210:        }
                   1211: #endif
1.3       pk       1212:
1.121     yamt     1213:        if (__predict_false((ph = pr_find_pagehead(pp, v)) == NULL)) {
1.25      thorpej  1214:                pr_printlog(pp, NULL, printf);
1.3       pk       1215:                panic("pool_put: %s: page header missing", pp->pr_wchan);
                   1216:        }
1.28      thorpej  1217:
1.3       pk       1218:        /*
                   1219:         * Return to item list.
                   1220:         */
1.97      yamt     1221:        if (pp->pr_roflags & PR_NOTOUCH) {
                   1222:                pr_item_notouch_put(pp, ph, v);
                   1223:        } else {
1.2       pk       1224: #ifdef DIAGNOSTIC
1.97      yamt     1225:                pi->pi_magic = PI_MAGIC;
1.3       pk       1226: #endif
1.32      chs      1227: #ifdef DEBUG
1.97      yamt     1228:                {
                   1229:                        int i, *ip = v;
1.32      chs      1230:
1.97      yamt     1231:                        for (i = 0; i < pp->pr_size / sizeof(int); i++) {
                   1232:                                *ip++ = PI_MAGIC;
                   1233:                        }
1.32      chs      1234:                }
                   1235: #endif
                   1236:
1.102     chs      1237:                LIST_INSERT_HEAD(&ph->ph_itemlist, pi, pi_list);
1.97      yamt     1238:        }
1.79      thorpej  1239:        KDASSERT(ph->ph_nmissing != 0);
1.3       pk       1240:        ph->ph_nmissing--;
                   1241:        pp->pr_nput++;
1.20      thorpej  1242:        pp->pr_nitems++;
                   1243:        pp->pr_nout--;
1.3       pk       1244:
                   1245:        /* Cancel "pool empty" condition if it exists */
                   1246:        if (pp->pr_curpage == NULL)
                   1247:                pp->pr_curpage = ph;
                   1248:
                   1249:        if (pp->pr_flags & PR_WANTED) {
                   1250:                pp->pr_flags &= ~PR_WANTED;
1.15      pk       1251:                if (ph->ph_nmissing == 0)
                   1252:                        pp->pr_nidle++;
1.134     ad       1253:                cv_broadcast(&pp->pr_cv);
1.3       pk       1254:                return;
                   1255:        }
                   1256:
                   1257:        /*
1.88      chs      1258:         * If this page is now empty, do one of two things:
1.21      thorpej  1259:         *
1.88      chs      1260:         *      (1) If we have more pages than the page high water mark,
1.96      thorpej  1261:         *          free the page back to the system.  ONLY CONSIDER
1.90      thorpej  1262:         *          FREEING BACK A PAGE IF WE HAVE MORE THAN OUR MINIMUM PAGE
                   1263:         *          CLAIM.
1.21      thorpej  1264:         *
1.88      chs      1265:         *      (2) Otherwise, move the page to the empty page list.
                   1266:         *
                   1267:         * Either way, select a new current page (so we use a partially-full
                   1268:         * page if one is available).
1.3       pk       1269:         */
                   1270:        if (ph->ph_nmissing == 0) {
1.6       thorpej  1271:                pp->pr_nidle++;
1.90      thorpej  1272:                if (pp->pr_npages > pp->pr_minpages &&
                   1273:                    (pp->pr_npages > pp->pr_maxpages ||
1.117     yamt     1274:                     pa_starved_p(pp->pr_alloc))) {
1.101     thorpej  1275:                        pr_rmpage(pp, ph, pq);
1.3       pk       1276:                } else {
1.88      chs      1277:                        LIST_REMOVE(ph, ph_pagelist);
                   1278:                        LIST_INSERT_HEAD(&pp->pr_emptypages, ph, ph_pagelist);
1.3       pk       1279:
1.21      thorpej  1280:                        /*
                   1281:                         * Update the timestamp on the page.  A page must
                   1282:                         * be idle for some period of time before it can
                   1283:                         * be reclaimed by the pagedaemon.  This minimizes
                   1284:                         * ping-pong'ing for memory.
                   1285:                         */
1.118     kardel   1286:                        getmicrotime(&ph->ph_time);
1.1       pk       1287:                }
1.88      chs      1288:                pool_update_curpage(pp);
1.1       pk       1289:        }
1.88      chs      1290:
1.21      thorpej  1291:        /*
1.88      chs      1292:         * If the page was previously completely full, move it to the
                   1293:         * partially-full list and make it the current page.  The next
                   1294:         * allocation will get the item from this page, instead of
                   1295:         * further fragmenting the pool.
1.21      thorpej  1296:         */
                   1297:        else if (ph->ph_nmissing == (pp->pr_itemsperpage - 1)) {
1.88      chs      1298:                LIST_REMOVE(ph, ph_pagelist);
                   1299:                LIST_INSERT_HEAD(&pp->pr_partpages, ph, ph_pagelist);
1.21      thorpej  1300:                pp->pr_curpage = ph;
                   1301:        }
1.43      thorpej  1302: }
                   1303:
                   1304: /*
1.134     ad       1305:  * Return resource to the pool.
1.43      thorpej  1306:  */
1.59      thorpej  1307: #ifdef POOL_DIAGNOSTIC
1.43      thorpej  1308: void
                   1309: _pool_put(struct pool *pp, void *v, const char *file, long line)
                   1310: {
1.101     thorpej  1311:        struct pool_pagelist pq;
                   1312:
                   1313:        LIST_INIT(&pq);
1.43      thorpej  1314:
1.134     ad       1315:        mutex_enter(&pp->pr_lock);
1.43      thorpej  1316:        pr_enter(pp, file, line);
                   1317:
1.56      sommerfe 1318:        pr_log(pp, v, PRLOG_PUT, file, line);
                   1319:
1.101     thorpej  1320:        pool_do_put(pp, v, &pq);
1.21      thorpej  1321:
1.25      thorpej  1322:        pr_leave(pp);
1.134     ad       1323:        mutex_exit(&pp->pr_lock);
1.101     thorpej  1324:
1.102     chs      1325:        pr_pagelist_free(pp, &pq);
1.1       pk       1326: }
1.57      sommerfe 1327: #undef pool_put
1.59      thorpej  1328: #endif /* POOL_DIAGNOSTIC */
1.1       pk       1329:
1.56      sommerfe 1330: void
                   1331: pool_put(struct pool *pp, void *v)
                   1332: {
1.101     thorpej  1333:        struct pool_pagelist pq;
                   1334:
                   1335:        LIST_INIT(&pq);
1.56      sommerfe 1336:
1.134     ad       1337:        mutex_enter(&pp->pr_lock);
1.101     thorpej  1338:        pool_do_put(pp, v, &pq);
1.134     ad       1339:        mutex_exit(&pp->pr_lock);
1.56      sommerfe 1340:
1.102     chs      1341:        pr_pagelist_free(pp, &pq);
1.56      sommerfe 1342: }
1.57      sommerfe 1343:
1.59      thorpej  1344: #ifdef POOL_DIAGNOSTIC
1.57      sommerfe 1345: #define                pool_put(h, v)  _pool_put((h), (v), __FILE__, __LINE__)
1.56      sommerfe 1346: #endif
1.74      thorpej  1347:
                   1348: /*
1.113     yamt     1349:  * pool_grow: grow a pool by a page.
                   1350:  *
                   1351:  * => called with pool locked.
                   1352:  * => unlock and relock the pool.
                   1353:  * => return with pool locked.
                   1354:  */
                   1355:
                   1356: static int
                   1357: pool_grow(struct pool *pp, int flags)
                   1358: {
                   1359:        struct pool_item_header *ph = NULL;
                   1360:        char *cp;
                   1361:
1.134     ad       1362:        mutex_exit(&pp->pr_lock);
1.113     yamt     1363:        cp = pool_allocator_alloc(pp, flags);
                   1364:        if (__predict_true(cp != NULL)) {
                   1365:                ph = pool_alloc_item_header(pp, cp, flags);
                   1366:        }
                   1367:        if (__predict_false(cp == NULL || ph == NULL)) {
                   1368:                if (cp != NULL) {
                   1369:                        pool_allocator_free(pp, cp);
                   1370:                }
1.134     ad       1371:                mutex_enter(&pp->pr_lock);
1.113     yamt     1372:                return ENOMEM;
                   1373:        }
                   1374:
1.134     ad       1375:        mutex_enter(&pp->pr_lock);
1.113     yamt     1376:        pool_prime_page(pp, cp, ph);
                   1377:        pp->pr_npagealloc++;
                   1378:        return 0;
                   1379: }
                   1380:
                   1381: /*
1.74      thorpej  1382:  * Add N items to the pool.
                   1383:  */
                   1384: int
                   1385: pool_prime(struct pool *pp, int n)
                   1386: {
1.75      simonb   1387:        int newpages;
1.113     yamt     1388:        int error = 0;
1.74      thorpej  1389:
1.134     ad       1390:        mutex_enter(&pp->pr_lock);
1.74      thorpej  1391:
                   1392:        newpages = roundup(n, pp->pr_itemsperpage) / pp->pr_itemsperpage;
                   1393:
                   1394:        while (newpages-- > 0) {
1.113     yamt     1395:                error = pool_grow(pp, PR_NOWAIT);
                   1396:                if (error) {
1.74      thorpej  1397:                        break;
                   1398:                }
                   1399:                pp->pr_minpages++;
                   1400:        }
                   1401:
                   1402:        if (pp->pr_minpages >= pp->pr_maxpages)
                   1403:                pp->pr_maxpages = pp->pr_minpages + 1;  /* XXX */
                   1404:
1.134     ad       1405:        mutex_exit(&pp->pr_lock);
1.113     yamt     1406:        return error;
1.74      thorpej  1407: }
1.55      thorpej  1408:
                   1409: /*
1.3       pk       1410:  * Add a page worth of items to the pool.
1.21      thorpej  1411:  *
                   1412:  * Note, we must be called with the pool descriptor LOCKED.
1.3       pk       1413:  */
1.55      thorpej  1414: static void
1.128     christos 1415: pool_prime_page(struct pool *pp, void *storage, struct pool_item_header *ph)
1.3       pk       1416: {
                   1417:        struct pool_item *pi;
1.128     christos 1418:        void *cp = storage;
1.125     ad       1419:        const unsigned int align = pp->pr_align;
                   1420:        const unsigned int ioff = pp->pr_itemoffset;
1.55      thorpej  1421:        int n;
1.36      pk       1422:
1.134     ad       1423:        KASSERT(mutex_owned(&pp->pr_lock));
1.91      yamt     1424:
1.66      thorpej  1425: #ifdef DIAGNOSTIC
1.121     yamt     1426:        if ((pp->pr_roflags & PR_NOALIGN) == 0 &&
                   1427:            ((uintptr_t)cp & (pp->pr_alloc->pa_pagesz - 1)) != 0)
1.36      pk       1428:                panic("pool_prime_page: %s: unaligned page", pp->pr_wchan);
1.66      thorpej  1429: #endif
1.3       pk       1430:
                   1431:        /*
                   1432:         * Insert page header.
                   1433:         */
1.88      chs      1434:        LIST_INSERT_HEAD(&pp->pr_emptypages, ph, ph_pagelist);
1.102     chs      1435:        LIST_INIT(&ph->ph_itemlist);
1.3       pk       1436:        ph->ph_page = storage;
                   1437:        ph->ph_nmissing = 0;
1.118     kardel   1438:        getmicrotime(&ph->ph_time);
1.88      chs      1439:        if ((pp->pr_roflags & PR_PHINPAGE) == 0)
                   1440:                SPLAY_INSERT(phtree, &pp->pr_phtree, ph);
1.3       pk       1441:
1.6       thorpej  1442:        pp->pr_nidle++;
                   1443:
1.3       pk       1444:        /*
                   1445:         * Color this page.
                   1446:         */
1.128     christos 1447:        cp = (char *)cp + pp->pr_curcolor;
1.3       pk       1448:        if ((pp->pr_curcolor += align) > pp->pr_maxcolor)
                   1449:                pp->pr_curcolor = 0;
                   1450:
                   1451:        /*
                   1452:         * Adjust storage to apply aligment to `pr_itemoffset' in each item.
                   1453:         */
                   1454:        if (ioff != 0)
1.128     christos 1455:                cp = (char *)cp + align - ioff;
1.3       pk       1456:
1.125     ad       1457:        KASSERT((((vaddr_t)cp + ioff) & (align - 1)) == 0);
                   1458:
1.3       pk       1459:        /*
                   1460:         * Insert remaining chunks on the bucket list.
                   1461:         */
                   1462:        n = pp->pr_itemsperpage;
1.20      thorpej  1463:        pp->pr_nitems += n;
1.3       pk       1464:
1.97      yamt     1465:        if (pp->pr_roflags & PR_NOTOUCH) {
1.135     yamt     1466:                pr_item_notouch_init(pp, ph);
1.97      yamt     1467:        } else {
                   1468:                while (n--) {
                   1469:                        pi = (struct pool_item *)cp;
1.78      thorpej  1470:
1.97      yamt     1471:                        KASSERT(((((vaddr_t)pi) + ioff) & (align - 1)) == 0);
1.3       pk       1472:
1.97      yamt     1473:                        /* Insert on page list */
1.102     chs      1474:                        LIST_INSERT_HEAD(&ph->ph_itemlist, pi, pi_list);
1.3       pk       1475: #ifdef DIAGNOSTIC
1.97      yamt     1476:                        pi->pi_magic = PI_MAGIC;
1.3       pk       1477: #endif
1.128     christos 1478:                        cp = (char *)cp + pp->pr_size;
1.125     ad       1479:
                   1480:                        KASSERT((((vaddr_t)cp + ioff) & (align - 1)) == 0);
1.97      yamt     1481:                }
1.3       pk       1482:        }
                   1483:
                   1484:        /*
                   1485:         * If the pool was depleted, point at the new page.
                   1486:         */
                   1487:        if (pp->pr_curpage == NULL)
                   1488:                pp->pr_curpage = ph;
                   1489:
                   1490:        if (++pp->pr_npages > pp->pr_hiwat)
                   1491:                pp->pr_hiwat = pp->pr_npages;
                   1492: }
                   1493:
1.20      thorpej  1494: /*
1.52      thorpej  1495:  * Used by pool_get() when nitems drops below the low water mark.  This
1.88      chs      1496:  * is used to catch up pr_nitems with the low water mark.
1.20      thorpej  1497:  *
1.21      thorpej  1498:  * Note 1, we never wait for memory here, we let the caller decide what to do.
1.20      thorpej  1499:  *
1.73      thorpej  1500:  * Note 2, we must be called with the pool already locked, and we return
1.20      thorpej  1501:  * with it locked.
                   1502:  */
                   1503: static int
1.42      thorpej  1504: pool_catchup(struct pool *pp)
1.20      thorpej  1505: {
                   1506:        int error = 0;
                   1507:
1.54      thorpej  1508:        while (POOL_NEEDS_CATCHUP(pp)) {
1.113     yamt     1509:                error = pool_grow(pp, PR_NOWAIT);
                   1510:                if (error) {
1.20      thorpej  1511:                        break;
                   1512:                }
                   1513:        }
1.113     yamt     1514:        return error;
1.20      thorpej  1515: }
                   1516:
1.88      chs      1517: static void
                   1518: pool_update_curpage(struct pool *pp)
                   1519: {
                   1520:
                   1521:        pp->pr_curpage = LIST_FIRST(&pp->pr_partpages);
                   1522:        if (pp->pr_curpage == NULL) {
                   1523:                pp->pr_curpage = LIST_FIRST(&pp->pr_emptypages);
                   1524:        }
                   1525: }
                   1526:
1.3       pk       1527: void
1.42      thorpej  1528: pool_setlowat(struct pool *pp, int n)
1.3       pk       1529: {
1.15      pk       1530:
1.134     ad       1531:        mutex_enter(&pp->pr_lock);
1.21      thorpej  1532:
1.3       pk       1533:        pp->pr_minitems = n;
1.15      pk       1534:        pp->pr_minpages = (n == 0)
                   1535:                ? 0
1.18      thorpej  1536:                : roundup(n, pp->pr_itemsperpage) / pp->pr_itemsperpage;
1.20      thorpej  1537:
                   1538:        /* Make sure we're caught up with the newly-set low water mark. */
1.75      simonb   1539:        if (POOL_NEEDS_CATCHUP(pp) && pool_catchup(pp) != 0) {
1.20      thorpej  1540:                /*
                   1541:                 * XXX: Should we log a warning?  Should we set up a timeout
                   1542:                 * to try again in a second or so?  The latter could break
                   1543:                 * a caller's assumptions about interrupt protection, etc.
                   1544:                 */
                   1545:        }
1.21      thorpej  1546:
1.134     ad       1547:        mutex_exit(&pp->pr_lock);
1.3       pk       1548: }
                   1549:
                   1550: void
1.42      thorpej  1551: pool_sethiwat(struct pool *pp, int n)
1.3       pk       1552: {
1.15      pk       1553:
1.134     ad       1554:        mutex_enter(&pp->pr_lock);
1.21      thorpej  1555:
1.15      pk       1556:        pp->pr_maxpages = (n == 0)
                   1557:                ? 0
1.18      thorpej  1558:                : roundup(n, pp->pr_itemsperpage) / pp->pr_itemsperpage;
1.21      thorpej  1559:
1.134     ad       1560:        mutex_exit(&pp->pr_lock);
1.3       pk       1561: }
                   1562:
1.20      thorpej  1563: void
1.42      thorpej  1564: pool_sethardlimit(struct pool *pp, int n, const char *warnmess, int ratecap)
1.20      thorpej  1565: {
                   1566:
1.134     ad       1567:        mutex_enter(&pp->pr_lock);
1.20      thorpej  1568:
                   1569:        pp->pr_hardlimit = n;
                   1570:        pp->pr_hardlimit_warning = warnmess;
1.31      thorpej  1571:        pp->pr_hardlimit_ratecap.tv_sec = ratecap;
                   1572:        pp->pr_hardlimit_warning_last.tv_sec = 0;
                   1573:        pp->pr_hardlimit_warning_last.tv_usec = 0;
1.20      thorpej  1574:
                   1575:        /*
1.21      thorpej  1576:         * In-line version of pool_sethiwat(), because we don't want to
                   1577:         * release the lock.
1.20      thorpej  1578:         */
                   1579:        pp->pr_maxpages = (n == 0)
                   1580:                ? 0
                   1581:                : roundup(n, pp->pr_itemsperpage) / pp->pr_itemsperpage;
1.21      thorpej  1582:
1.134     ad       1583:        mutex_exit(&pp->pr_lock);
1.20      thorpej  1584: }
1.3       pk       1585:
                   1586: /*
                   1587:  * Release all complete pages that have not been used recently.
                   1588:  */
1.66      thorpej  1589: int
1.59      thorpej  1590: #ifdef POOL_DIAGNOSTIC
1.42      thorpej  1591: _pool_reclaim(struct pool *pp, const char *file, long line)
1.56      sommerfe 1592: #else
                   1593: pool_reclaim(struct pool *pp)
                   1594: #endif
1.3       pk       1595: {
                   1596:        struct pool_item_header *ph, *phnext;
1.61      chs      1597:        struct pool_pagelist pq;
1.102     chs      1598:        struct timeval curtime, diff;
1.134     ad       1599:        bool klock;
                   1600:        int rv;
1.3       pk       1601:
1.68      thorpej  1602:        if (pp->pr_drain_hook != NULL) {
                   1603:                /*
                   1604:                 * The drain hook must be called with the pool unlocked.
                   1605:                 */
                   1606:                (*pp->pr_drain_hook)(pp->pr_drain_hook_arg, PR_NOWAIT);
                   1607:        }
                   1608:
1.134     ad       1609:        /*
                   1610:         * XXXSMP Because mutexes at IPL_SOFTXXX are still spinlocks,
                   1611:         * and we are called from the pagedaemon without kernel_lock.
                   1612:         * Does not apply to IPL_SOFTBIO.
                   1613:         */
                   1614:        if (pp->pr_ipl == IPL_SOFTNET || pp->pr_ipl == IPL_SOFTCLOCK ||
                   1615:            pp->pr_ipl == IPL_SOFTSERIAL) {
                   1616:                KERNEL_LOCK(1, NULL);
                   1617:                klock = true;
                   1618:        } else
                   1619:                klock = false;
                   1620:
                   1621:        /* Reclaim items from the pool's cache (if any). */
                   1622:        if (pp->pr_cache != NULL)
                   1623:                pool_cache_invalidate(pp->pr_cache);
                   1624:
                   1625:        if (mutex_tryenter(&pp->pr_lock) == 0) {
                   1626:                if (klock) {
                   1627:                        KERNEL_UNLOCK_ONE(NULL);
                   1628:                }
1.66      thorpej  1629:                return (0);
1.134     ad       1630:        }
1.25      thorpej  1631:        pr_enter(pp, file, line);
1.68      thorpej  1632:
1.88      chs      1633:        LIST_INIT(&pq);
1.43      thorpej  1634:
1.118     kardel   1635:        getmicrotime(&curtime);
1.21      thorpej  1636:
1.88      chs      1637:        for (ph = LIST_FIRST(&pp->pr_emptypages); ph != NULL; ph = phnext) {
                   1638:                phnext = LIST_NEXT(ph, ph_pagelist);
1.3       pk       1639:
                   1640:                /* Check our minimum page claim */
                   1641:                if (pp->pr_npages <= pp->pr_minpages)
                   1642:                        break;
                   1643:
1.88      chs      1644:                KASSERT(ph->ph_nmissing == 0);
                   1645:                timersub(&curtime, &ph->ph_time, &diff);
1.117     yamt     1646:                if (diff.tv_sec < pool_inactive_time
                   1647:                    && !pa_starved_p(pp->pr_alloc))
1.88      chs      1648:                        continue;
1.21      thorpej  1649:
1.88      chs      1650:                /*
                   1651:                 * If freeing this page would put us below
                   1652:                 * the low water mark, stop now.
                   1653:                 */
                   1654:                if ((pp->pr_nitems - pp->pr_itemsperpage) <
                   1655:                    pp->pr_minitems)
                   1656:                        break;
1.21      thorpej  1657:
1.88      chs      1658:                pr_rmpage(pp, ph, &pq);
1.3       pk       1659:        }
                   1660:
1.25      thorpej  1661:        pr_leave(pp);
1.134     ad       1662:        mutex_exit(&pp->pr_lock);
                   1663:
                   1664:        if (LIST_EMPTY(&pq))
                   1665:                rv = 0;
                   1666:        else {
                   1667:                pr_pagelist_free(pp, &pq);
                   1668:                rv = 1;
                   1669:        }
                   1670:
                   1671:        if (klock) {
                   1672:                KERNEL_UNLOCK_ONE(NULL);
                   1673:        }
1.66      thorpej  1674:
1.134     ad       1675:        return (rv);
1.3       pk       1676: }
                   1677:
                   1678: /*
1.134     ad       1679:  * Drain pools, one at a time.  This is a two stage process;
                   1680:  * drain_start kicks off a cross call to drain CPU-level caches
                   1681:  * if the pool has an associated pool_cache.  drain_end waits
                   1682:  * for those cross calls to finish, and then drains the cache
                   1683:  * (if any) and pool.
1.131     ad       1684:  *
1.134     ad       1685:  * Note, must never be called from interrupt context.
1.3       pk       1686:  */
                   1687: void
1.134     ad       1688: pool_drain_start(struct pool **ppp, uint64_t *wp)
1.3       pk       1689: {
                   1690:        struct pool *pp;
1.134     ad       1691:
1.137.2.3  ad       1692:        KASSERT(!TAILQ_EMPTY(&pool_head));
1.3       pk       1693:
1.61      chs      1694:        pp = NULL;
1.134     ad       1695:
                   1696:        /* Find next pool to drain, and add a reference. */
                   1697:        mutex_enter(&pool_head_lock);
                   1698:        do {
                   1699:                if (drainpp == NULL) {
1.137.2.3  ad       1700:                        drainpp = TAILQ_FIRST(&pool_head);
1.134     ad       1701:                }
                   1702:                if (drainpp != NULL) {
                   1703:                        pp = drainpp;
1.137.2.3  ad       1704:                        drainpp = TAILQ_NEXT(pp, pr_poollist);
1.134     ad       1705:                }
                   1706:                /*
                   1707:                 * Skip completely idle pools.  We depend on at least
                   1708:                 * one pool in the system being active.
                   1709:                 */
                   1710:        } while (pp == NULL || pp->pr_npages == 0);
                   1711:        pp->pr_refcnt++;
                   1712:        mutex_exit(&pool_head_lock);
                   1713:
                   1714:        /* If there is a pool_cache, drain CPU level caches. */
                   1715:        *ppp = pp;
                   1716:        if (pp->pr_cache != NULL) {
                   1717:                *wp = xc_broadcast(0, (xcfunc_t)pool_cache_xcall,
                   1718:                    pp->pr_cache, NULL);
                   1719:        }
                   1720: }
                   1721:
                   1722: void
                   1723: pool_drain_end(struct pool *pp, uint64_t where)
                   1724: {
                   1725:
                   1726:        if (pp == NULL)
                   1727:                return;
                   1728:
                   1729:        KASSERT(pp->pr_refcnt > 0);
                   1730:
                   1731:        /* Wait for remote draining to complete. */
                   1732:        if (pp->pr_cache != NULL)
                   1733:                xc_wait(where);
                   1734:
                   1735:        /* Drain the cache (if any) and pool.. */
                   1736:        pool_reclaim(pp);
                   1737:
                   1738:        /* Finally, unlock the pool. */
                   1739:        mutex_enter(&pool_head_lock);
                   1740:        pp->pr_refcnt--;
                   1741:        cv_broadcast(&pool_busy);
                   1742:        mutex_exit(&pool_head_lock);
1.3       pk       1743: }
                   1744:
                   1745: /*
                   1746:  * Diagnostic helpers.
                   1747:  */
                   1748: void
1.42      thorpej  1749: pool_print(struct pool *pp, const char *modif)
1.21      thorpej  1750: {
                   1751:
1.25      thorpej  1752:        pool_print1(pp, modif, printf);
1.21      thorpej  1753: }
                   1754:
1.25      thorpej  1755: void
1.108     yamt     1756: pool_printall(const char *modif, void (*pr)(const char *, ...))
                   1757: {
                   1758:        struct pool *pp;
                   1759:
1.137.2.3  ad       1760:        TAILQ_FOREACH(pp, &pool_head, pr_poollist) {
1.108     yamt     1761:                pool_printit(pp, modif, pr);
                   1762:        }
                   1763: }
                   1764:
                   1765: void
1.42      thorpej  1766: pool_printit(struct pool *pp, const char *modif, void (*pr)(const char *, ...))
1.25      thorpej  1767: {
                   1768:
                   1769:        if (pp == NULL) {
                   1770:                (*pr)("Must specify a pool to print.\n");
                   1771:                return;
                   1772:        }
                   1773:
                   1774:        pool_print1(pp, modif, pr);
                   1775: }
                   1776:
1.21      thorpej  1777: static void
1.124     yamt     1778: pool_print_pagelist(struct pool *pp, struct pool_pagelist *pl,
1.97      yamt     1779:     void (*pr)(const char *, ...))
1.88      chs      1780: {
                   1781:        struct pool_item_header *ph;
                   1782: #ifdef DIAGNOSTIC
                   1783:        struct pool_item *pi;
                   1784: #endif
                   1785:
                   1786:        LIST_FOREACH(ph, pl, ph_pagelist) {
                   1787:                (*pr)("\t\tpage %p, nmissing %d, time %lu,%lu\n",
                   1788:                    ph->ph_page, ph->ph_nmissing,
                   1789:                    (u_long)ph->ph_time.tv_sec,
                   1790:                    (u_long)ph->ph_time.tv_usec);
                   1791: #ifdef DIAGNOSTIC
1.97      yamt     1792:                if (!(pp->pr_roflags & PR_NOTOUCH)) {
1.102     chs      1793:                        LIST_FOREACH(pi, &ph->ph_itemlist, pi_list) {
1.97      yamt     1794:                                if (pi->pi_magic != PI_MAGIC) {
                   1795:                                        (*pr)("\t\t\titem %p, magic 0x%x\n",
                   1796:                                            pi, pi->pi_magic);
                   1797:                                }
1.88      chs      1798:                        }
                   1799:                }
                   1800: #endif
                   1801:        }
                   1802: }
                   1803:
                   1804: static void
1.42      thorpej  1805: pool_print1(struct pool *pp, const char *modif, void (*pr)(const char *, ...))
1.3       pk       1806: {
1.25      thorpej  1807:        struct pool_item_header *ph;
1.134     ad       1808:        pool_cache_t pc;
                   1809:        pcg_t *pcg;
                   1810:        pool_cache_cpu_t *cc;
                   1811:        uint64_t cpuhit, cpumiss;
1.44      thorpej  1812:        int i, print_log = 0, print_pagelist = 0, print_cache = 0;
1.25      thorpej  1813:        char c;
                   1814:
                   1815:        while ((c = *modif++) != '\0') {
                   1816:                if (c == 'l')
                   1817:                        print_log = 1;
                   1818:                if (c == 'p')
                   1819:                        print_pagelist = 1;
1.44      thorpej  1820:                if (c == 'c')
                   1821:                        print_cache = 1;
1.25      thorpej  1822:        }
                   1823:
1.134     ad       1824:        if ((pc = pp->pr_cache) != NULL) {
                   1825:                (*pr)("POOL CACHE");
                   1826:        } else {
                   1827:                (*pr)("POOL");
                   1828:        }
                   1829:
                   1830:        (*pr)(" %s: size %u, align %u, ioff %u, roflags 0x%08x\n",
1.25      thorpej  1831:            pp->pr_wchan, pp->pr_size, pp->pr_align, pp->pr_itemoffset,
                   1832:            pp->pr_roflags);
1.66      thorpej  1833:        (*pr)("\talloc %p\n", pp->pr_alloc);
1.25      thorpej  1834:        (*pr)("\tminitems %u, minpages %u, maxpages %u, npages %u\n",
                   1835:            pp->pr_minitems, pp->pr_minpages, pp->pr_maxpages, pp->pr_npages);
                   1836:        (*pr)("\titemsperpage %u, nitems %u, nout %u, hardlimit %u\n",
                   1837:            pp->pr_itemsperpage, pp->pr_nitems, pp->pr_nout, pp->pr_hardlimit);
                   1838:
1.134     ad       1839:        (*pr)("\tnget %lu, nfail %lu, nput %lu\n",
1.25      thorpej  1840:            pp->pr_nget, pp->pr_nfail, pp->pr_nput);
                   1841:        (*pr)("\tnpagealloc %lu, npagefree %lu, hiwat %u, nidle %lu\n",
                   1842:            pp->pr_npagealloc, pp->pr_npagefree, pp->pr_hiwat, pp->pr_nidle);
                   1843:
                   1844:        if (print_pagelist == 0)
                   1845:                goto skip_pagelist;
                   1846:
1.88      chs      1847:        if ((ph = LIST_FIRST(&pp->pr_emptypages)) != NULL)
                   1848:                (*pr)("\n\tempty page list:\n");
1.97      yamt     1849:        pool_print_pagelist(pp, &pp->pr_emptypages, pr);
1.88      chs      1850:        if ((ph = LIST_FIRST(&pp->pr_fullpages)) != NULL)
                   1851:                (*pr)("\n\tfull page list:\n");
1.97      yamt     1852:        pool_print_pagelist(pp, &pp->pr_fullpages, pr);
1.88      chs      1853:        if ((ph = LIST_FIRST(&pp->pr_partpages)) != NULL)
                   1854:                (*pr)("\n\tpartial-page list:\n");
1.97      yamt     1855:        pool_print_pagelist(pp, &pp->pr_partpages, pr);
1.88      chs      1856:
1.25      thorpej  1857:        if (pp->pr_curpage == NULL)
                   1858:                (*pr)("\tno current page\n");
                   1859:        else
                   1860:                (*pr)("\tcurpage %p\n", pp->pr_curpage->ph_page);
                   1861:
                   1862:  skip_pagelist:
                   1863:        if (print_log == 0)
                   1864:                goto skip_log;
                   1865:
                   1866:        (*pr)("\n");
                   1867:        if ((pp->pr_roflags & PR_LOGGING) == 0)
                   1868:                (*pr)("\tno log\n");
1.122     christos 1869:        else {
1.25      thorpej  1870:                pr_printlog(pp, NULL, pr);
1.122     christos 1871:        }
1.3       pk       1872:
1.25      thorpej  1873:  skip_log:
1.44      thorpej  1874:
1.102     chs      1875: #define PR_GROUPLIST(pcg)                                              \
                   1876:        (*pr)("\t\tgroup %p: avail %d\n", pcg, pcg->pcg_avail);         \
                   1877:        for (i = 0; i < PCG_NOBJECTS; i++) {                            \
                   1878:                if (pcg->pcg_objects[i].pcgo_pa !=                      \
                   1879:                    POOL_PADDR_INVALID) {                               \
                   1880:                        (*pr)("\t\t\t%p, 0x%llx\n",                     \
                   1881:                            pcg->pcg_objects[i].pcgo_va,                \
                   1882:                            (unsigned long long)                        \
                   1883:                            pcg->pcg_objects[i].pcgo_pa);               \
                   1884:                } else {                                                \
                   1885:                        (*pr)("\t\t\t%p\n",                             \
                   1886:                            pcg->pcg_objects[i].pcgo_va);               \
                   1887:                }                                                       \
                   1888:        }
                   1889:
1.134     ad       1890:        if (pc != NULL) {
                   1891:                cpuhit = 0;
                   1892:                cpumiss = 0;
                   1893:                for (i = 0; i < MAXCPUS; i++) {
                   1894:                        if ((cc = pc->pc_cpus[i]) == NULL)
                   1895:                                continue;
                   1896:                        cpuhit += cc->cc_hits;
                   1897:                        cpumiss += cc->cc_misses;
                   1898:                }
                   1899:                (*pr)("\tcpu layer hits %llu misses %llu\n", cpuhit, cpumiss);
                   1900:                (*pr)("\tcache layer hits %llu misses %llu\n",
                   1901:                    pc->pc_hits, pc->pc_misses);
                   1902:                (*pr)("\tcache layer entry uncontended %llu contended %llu\n",
                   1903:                    pc->pc_hits + pc->pc_misses - pc->pc_contended,
                   1904:                    pc->pc_contended);
                   1905:                (*pr)("\tcache layer empty groups %u full groups %u\n",
                   1906:                    pc->pc_nempty, pc->pc_nfull);
                   1907:                if (print_cache) {
                   1908:                        (*pr)("\tfull cache groups:\n");
                   1909:                        for (pcg = pc->pc_fullgroups; pcg != NULL;
                   1910:                            pcg = pcg->pcg_next) {
                   1911:                                PR_GROUPLIST(pcg);
                   1912:                        }
                   1913:                        (*pr)("\tempty cache groups:\n");
                   1914:                        for (pcg = pc->pc_emptygroups; pcg != NULL;
                   1915:                            pcg = pcg->pcg_next) {
                   1916:                                PR_GROUPLIST(pcg);
                   1917:                        }
1.103     chs      1918:                }
1.44      thorpej  1919:        }
1.102     chs      1920: #undef PR_GROUPLIST
1.44      thorpej  1921:
1.88      chs      1922:        pr_enter_check(pp, pr);
                   1923: }
                   1924:
                   1925: static int
                   1926: pool_chk_page(struct pool *pp, const char *label, struct pool_item_header *ph)
                   1927: {
                   1928:        struct pool_item *pi;
1.128     christos 1929:        void *page;
1.88      chs      1930:        int n;
                   1931:
1.121     yamt     1932:        if ((pp->pr_roflags & PR_NOALIGN) == 0) {
1.128     christos 1933:                page = (void *)((uintptr_t)ph & pp->pr_alloc->pa_pagemask);
1.121     yamt     1934:                if (page != ph->ph_page &&
                   1935:                    (pp->pr_roflags & PR_PHINPAGE) != 0) {
                   1936:                        if (label != NULL)
                   1937:                                printf("%s: ", label);
                   1938:                        printf("pool(%p:%s): page inconsistency: page %p;"
                   1939:                               " at page head addr %p (p %p)\n", pp,
                   1940:                                pp->pr_wchan, ph->ph_page,
                   1941:                                ph, page);
                   1942:                        return 1;
                   1943:                }
1.88      chs      1944:        }
1.3       pk       1945:
1.97      yamt     1946:        if ((pp->pr_roflags & PR_NOTOUCH) != 0)
                   1947:                return 0;
                   1948:
1.102     chs      1949:        for (pi = LIST_FIRST(&ph->ph_itemlist), n = 0;
1.88      chs      1950:             pi != NULL;
1.102     chs      1951:             pi = LIST_NEXT(pi,pi_list), n++) {
1.88      chs      1952:
                   1953: #ifdef DIAGNOSTIC
                   1954:                if (pi->pi_magic != PI_MAGIC) {
                   1955:                        if (label != NULL)
                   1956:                                printf("%s: ", label);
                   1957:                        printf("pool(%s): free list modified: magic=%x;"
1.121     yamt     1958:                               " page %p; item ordinal %d; addr %p\n",
1.88      chs      1959:                                pp->pr_wchan, pi->pi_magic, ph->ph_page,
1.121     yamt     1960:                                n, pi);
1.88      chs      1961:                        panic("pool");
                   1962:                }
                   1963: #endif
1.121     yamt     1964:                if ((pp->pr_roflags & PR_NOALIGN) != 0) {
                   1965:                        continue;
                   1966:                }
1.128     christos 1967:                page = (void *)((uintptr_t)pi & pp->pr_alloc->pa_pagemask);
1.88      chs      1968:                if (page == ph->ph_page)
                   1969:                        continue;
                   1970:
                   1971:                if (label != NULL)
                   1972:                        printf("%s: ", label);
                   1973:                printf("pool(%p:%s): page inconsistency: page %p;"
                   1974:                       " item ordinal %d; addr %p (p %p)\n", pp,
                   1975:                        pp->pr_wchan, ph->ph_page,
                   1976:                        n, pi, page);
                   1977:                return 1;
                   1978:        }
                   1979:        return 0;
1.3       pk       1980: }
                   1981:
1.88      chs      1982:
1.3       pk       1983: int
1.42      thorpej  1984: pool_chk(struct pool *pp, const char *label)
1.3       pk       1985: {
                   1986:        struct pool_item_header *ph;
                   1987:        int r = 0;
                   1988:
1.134     ad       1989:        mutex_enter(&pp->pr_lock);
1.88      chs      1990:        LIST_FOREACH(ph, &pp->pr_emptypages, ph_pagelist) {
                   1991:                r = pool_chk_page(pp, label, ph);
                   1992:                if (r) {
                   1993:                        goto out;
                   1994:                }
                   1995:        }
                   1996:        LIST_FOREACH(ph, &pp->pr_fullpages, ph_pagelist) {
                   1997:                r = pool_chk_page(pp, label, ph);
                   1998:                if (r) {
1.3       pk       1999:                        goto out;
                   2000:                }
1.88      chs      2001:        }
                   2002:        LIST_FOREACH(ph, &pp->pr_partpages, ph_pagelist) {
                   2003:                r = pool_chk_page(pp, label, ph);
                   2004:                if (r) {
1.3       pk       2005:                        goto out;
                   2006:                }
                   2007:        }
1.88      chs      2008:
1.3       pk       2009: out:
1.134     ad       2010:        mutex_exit(&pp->pr_lock);
1.3       pk       2011:        return (r);
1.43      thorpej  2012: }
                   2013:
                   2014: /*
                   2015:  * pool_cache_init:
                   2016:  *
                   2017:  *     Initialize a pool cache.
1.134     ad       2018:  */
                   2019: pool_cache_t
                   2020: pool_cache_init(size_t size, u_int align, u_int align_offset, u_int flags,
                   2021:     const char *wchan, struct pool_allocator *palloc, int ipl,
                   2022:     int (*ctor)(void *, void *, int), void (*dtor)(void *, void *), void *arg)
                   2023: {
                   2024:        pool_cache_t pc;
                   2025:
                   2026:        pc = pool_get(&cache_pool, PR_WAITOK);
                   2027:        if (pc == NULL)
                   2028:                return NULL;
                   2029:
                   2030:        pool_cache_bootstrap(pc, size, align, align_offset, flags, wchan,
                   2031:           palloc, ipl, ctor, dtor, arg);
                   2032:
                   2033:        return pc;
                   2034: }
                   2035:
                   2036: /*
                   2037:  * pool_cache_bootstrap:
1.43      thorpej  2038:  *
1.134     ad       2039:  *     Kernel-private version of pool_cache_init().  The caller
                   2040:  *     provides initial storage.
1.43      thorpej  2041:  */
                   2042: void
1.134     ad       2043: pool_cache_bootstrap(pool_cache_t pc, size_t size, u_int align,
                   2044:     u_int align_offset, u_int flags, const char *wchan,
                   2045:     struct pool_allocator *palloc, int ipl,
                   2046:     int (*ctor)(void *, void *, int), void (*dtor)(void *, void *),
1.43      thorpej  2047:     void *arg)
                   2048: {
1.134     ad       2049:        CPU_INFO_ITERATOR cii;
1.137.2.3  ad       2050:        pool_cache_t pc1;
1.134     ad       2051:        struct cpu_info *ci;
                   2052:        struct pool *pp;
                   2053:
                   2054:        pp = &pc->pc_pool;
                   2055:        if (palloc == NULL && ipl == IPL_NONE)
                   2056:                palloc = &pool_allocator_nointr;
                   2057:        pool_init(pp, size, align, align_offset, flags, wchan, palloc, ipl);
1.43      thorpej  2058:
1.137.2.1  ad       2059:        /*
                   2060:         * XXXAD hack to prevent IP input processing from blocking.
                   2061:         */
                   2062:        if (ipl == IPL_SOFTNET) {
                   2063:                mutex_init(&pc->pc_lock, MUTEX_DEFAULT, IPL_VM);
                   2064:        } else {
                   2065:                mutex_init(&pc->pc_lock, MUTEX_DEFAULT, ipl);
                   2066:        }
1.43      thorpej  2067:
1.134     ad       2068:        if (ctor == NULL) {
                   2069:                ctor = (int (*)(void *, void *, int))nullop;
                   2070:        }
                   2071:        if (dtor == NULL) {
                   2072:                dtor = (void (*)(void *, void *))nullop;
                   2073:        }
1.43      thorpej  2074:
1.134     ad       2075:        pc->pc_emptygroups = NULL;
                   2076:        pc->pc_fullgroups = NULL;
                   2077:        pc->pc_partgroups = NULL;
1.43      thorpej  2078:        pc->pc_ctor = ctor;
                   2079:        pc->pc_dtor = dtor;
                   2080:        pc->pc_arg  = arg;
1.134     ad       2081:        pc->pc_hits  = 0;
1.48      thorpej  2082:        pc->pc_misses = 0;
1.134     ad       2083:        pc->pc_nempty = 0;
                   2084:        pc->pc_npart = 0;
                   2085:        pc->pc_nfull = 0;
                   2086:        pc->pc_contended = 0;
                   2087:        pc->pc_refcnt = 0;
1.136     yamt     2088:        pc->pc_freecheck = NULL;
1.134     ad       2089:
                   2090:        /* Allocate per-CPU caches. */
                   2091:        memset(pc->pc_cpus, 0, sizeof(pc->pc_cpus));
                   2092:        pc->pc_ncpu = 0;
1.137     ad       2093:        if (ncpu == 0) {
                   2094:                /* XXX For sparc: boot CPU is not attached yet. */
                   2095:                pool_cache_cpu_init1(curcpu(), pc);
                   2096:        } else {
                   2097:                for (CPU_INFO_FOREACH(cii, ci)) {
                   2098:                        pool_cache_cpu_init1(ci, pc);
                   2099:                }
1.134     ad       2100:        }
1.137.2.3  ad       2101:
                   2102:        /* Add to list of all pools. */
                   2103:        if (__predict_true(!cold))
1.134     ad       2104:                mutex_enter(&pool_head_lock);
1.137.2.3  ad       2105:        TAILQ_FOREACH(pc1, &pool_cache_head, pc_cachelist) {
                   2106:                if (strcmp(pc1->pc_pool.pr_wchan, pc->pc_pool.pr_wchan) > 0)
                   2107:                        break;
1.134     ad       2108:        }
1.137.2.3  ad       2109:        if (pc1 == NULL)
                   2110:                TAILQ_INSERT_TAIL(&pool_cache_head, pc, pc_cachelist);
                   2111:        else
                   2112:                TAILQ_INSERT_BEFORE(pc1, pc, pc_cachelist);
                   2113:        if (__predict_true(!cold))
                   2114:                mutex_exit(&pool_head_lock);
                   2115:
                   2116:        membar_sync();
                   2117:        pp->pr_cache = pc;
1.43      thorpej  2118: }
                   2119:
                   2120: /*
                   2121:  * pool_cache_destroy:
                   2122:  *
                   2123:  *     Destroy a pool cache.
                   2124:  */
                   2125: void
1.134     ad       2126: pool_cache_destroy(pool_cache_t pc)
1.43      thorpej  2127: {
1.134     ad       2128:        struct pool *pp = &pc->pc_pool;
                   2129:        pool_cache_cpu_t *cc;
                   2130:        pcg_t *pcg;
                   2131:        int i;
                   2132:
                   2133:        /* Remove it from the global list. */
                   2134:        mutex_enter(&pool_head_lock);
                   2135:        while (pc->pc_refcnt != 0)
                   2136:                cv_wait(&pool_busy, &pool_head_lock);
1.137.2.3  ad       2137:        TAILQ_REMOVE(&pool_cache_head, pc, pc_cachelist);
1.134     ad       2138:        mutex_exit(&pool_head_lock);
1.43      thorpej  2139:
                   2140:        /* First, invalidate the entire cache. */
                   2141:        pool_cache_invalidate(pc);
                   2142:
1.134     ad       2143:        /* Disassociate it from the pool. */
                   2144:        mutex_enter(&pp->pr_lock);
                   2145:        pp->pr_cache = NULL;
                   2146:        mutex_exit(&pp->pr_lock);
                   2147:
                   2148:        /* Destroy per-CPU data */
                   2149:        for (i = 0; i < MAXCPUS; i++) {
                   2150:                if ((cc = pc->pc_cpus[i]) == NULL)
                   2151:                        continue;
                   2152:                if ((pcg = cc->cc_current) != NULL) {
                   2153:                        pcg->pcg_next = NULL;
                   2154:                        pool_cache_invalidate_groups(pc, pcg);
                   2155:                }
                   2156:                if ((pcg = cc->cc_previous) != NULL) {
                   2157:                        pcg->pcg_next = NULL;
                   2158:                        pool_cache_invalidate_groups(pc, pcg);
                   2159:                }
                   2160:                if (cc != &pc->pc_cpu0)
                   2161:                        pool_put(&cache_cpu_pool, cc);
                   2162:        }
                   2163:
                   2164:        /* Finally, destroy it. */
                   2165:        mutex_destroy(&pc->pc_lock);
                   2166:        pool_destroy(pp);
                   2167:        pool_put(&cache_pool, pc);
                   2168: }
                   2169:
                   2170: /*
                   2171:  * pool_cache_cpu_init1:
                   2172:  *
                   2173:  *     Called for each pool_cache whenever a new CPU is attached.
                   2174:  */
                   2175: static void
                   2176: pool_cache_cpu_init1(struct cpu_info *ci, pool_cache_t pc)
                   2177: {
                   2178:        pool_cache_cpu_t *cc;
1.137     ad       2179:        int index;
1.134     ad       2180:
1.137     ad       2181:        index = ci->ci_index;
                   2182:
                   2183:        KASSERT(index < MAXCPUS);
1.134     ad       2184:        KASSERT(((uintptr_t)pc->pc_cpus & (CACHE_LINE_SIZE - 1)) == 0);
                   2185:
1.137     ad       2186:        if ((cc = pc->pc_cpus[index]) != NULL) {
                   2187:                KASSERT(cc->cc_cpuindex == index);
1.134     ad       2188:                return;
                   2189:        }
                   2190:
                   2191:        /*
                   2192:         * The first CPU is 'free'.  This needs to be the case for
                   2193:         * bootstrap - we may not be able to allocate yet.
                   2194:         */
                   2195:        if (pc->pc_ncpu == 0) {
                   2196:                cc = &pc->pc_cpu0;
                   2197:                pc->pc_ncpu = 1;
                   2198:        } else {
                   2199:                mutex_enter(&pc->pc_lock);
                   2200:                pc->pc_ncpu++;
                   2201:                mutex_exit(&pc->pc_lock);
                   2202:                cc = pool_get(&cache_cpu_pool, PR_WAITOK);
                   2203:        }
                   2204:
                   2205:        cc->cc_ipl = pc->pc_pool.pr_ipl;
                   2206:        cc->cc_iplcookie = makeiplcookie(cc->cc_ipl);
                   2207:        cc->cc_cache = pc;
1.137     ad       2208:        cc->cc_cpuindex = index;
1.134     ad       2209:        cc->cc_hits = 0;
                   2210:        cc->cc_misses = 0;
                   2211:        cc->cc_current = NULL;
                   2212:        cc->cc_previous = NULL;
                   2213:
1.137     ad       2214:        pc->pc_cpus[index] = cc;
1.43      thorpej  2215: }
                   2216:
1.134     ad       2217: /*
                   2218:  * pool_cache_cpu_init:
                   2219:  *
                   2220:  *     Called whenever a new CPU is attached.
                   2221:  */
                   2222: void
                   2223: pool_cache_cpu_init(struct cpu_info *ci)
1.43      thorpej  2224: {
1.134     ad       2225:        pool_cache_t pc;
                   2226:
                   2227:        mutex_enter(&pool_head_lock);
1.137.2.3  ad       2228:        TAILQ_FOREACH(pc, &pool_cache_head, pc_cachelist) {
1.134     ad       2229:                pc->pc_refcnt++;
                   2230:                mutex_exit(&pool_head_lock);
1.43      thorpej  2231:
1.134     ad       2232:                pool_cache_cpu_init1(ci, pc);
1.43      thorpej  2233:
1.134     ad       2234:                mutex_enter(&pool_head_lock);
                   2235:                pc->pc_refcnt--;
                   2236:                cv_broadcast(&pool_busy);
                   2237:        }
                   2238:        mutex_exit(&pool_head_lock);
1.43      thorpej  2239: }
                   2240:
1.134     ad       2241: /*
                   2242:  * pool_cache_reclaim:
                   2243:  *
                   2244:  *     Reclaim memory from a pool cache.
                   2245:  */
                   2246: bool
                   2247: pool_cache_reclaim(pool_cache_t pc)
1.43      thorpej  2248: {
                   2249:
1.134     ad       2250:        return pool_reclaim(&pc->pc_pool);
                   2251: }
1.43      thorpej  2252:
1.136     yamt     2253: static void
                   2254: pool_cache_destruct_object1(pool_cache_t pc, void *object)
                   2255: {
                   2256:
                   2257:        (*pc->pc_dtor)(pc->pc_arg, object);
                   2258:        pool_put(&pc->pc_pool, object);
                   2259: }
                   2260:
1.134     ad       2261: /*
                   2262:  * pool_cache_destruct_object:
                   2263:  *
                   2264:  *     Force destruction of an object and its release back into
                   2265:  *     the pool.
                   2266:  */
                   2267: void
                   2268: pool_cache_destruct_object(pool_cache_t pc, void *object)
                   2269: {
                   2270:
1.136     yamt     2271:        FREECHECK_IN(&pc->pc_freecheck, object);
                   2272:
                   2273:        pool_cache_destruct_object1(pc, object);
1.43      thorpej  2274: }
                   2275:
1.134     ad       2276: /*
                   2277:  * pool_cache_invalidate_groups:
                   2278:  *
                   2279:  *     Invalidate a chain of groups and destruct all objects.
                   2280:  */
1.102     chs      2281: static void
1.134     ad       2282: pool_cache_invalidate_groups(pool_cache_t pc, pcg_t *pcg)
1.102     chs      2283: {
1.134     ad       2284:        void *object;
                   2285:        pcg_t *next;
                   2286:        int i;
                   2287:
                   2288:        for (; pcg != NULL; pcg = next) {
                   2289:                next = pcg->pcg_next;
                   2290:
                   2291:                for (i = 0; i < pcg->pcg_avail; i++) {
                   2292:                        object = pcg->pcg_objects[i].pcgo_va;
1.136     yamt     2293:                        pool_cache_destruct_object1(pc, object);
1.134     ad       2294:                }
1.102     chs      2295:
                   2296:                pool_put(&pcgpool, pcg);
                   2297:        }
                   2298: }
                   2299:
1.43      thorpej  2300: /*
1.134     ad       2301:  * pool_cache_invalidate:
1.43      thorpej  2302:  *
1.134     ad       2303:  *     Invalidate a pool cache (destruct and release all of the
                   2304:  *     cached objects).  Does not reclaim objects from the pool.
1.43      thorpej  2305:  */
1.134     ad       2306: void
                   2307: pool_cache_invalidate(pool_cache_t pc)
                   2308: {
                   2309:        pcg_t *full, *empty, *part;
                   2310:
                   2311:        mutex_enter(&pc->pc_lock);
                   2312:        full = pc->pc_fullgroups;
                   2313:        empty = pc->pc_emptygroups;
                   2314:        part = pc->pc_partgroups;
                   2315:        pc->pc_fullgroups = NULL;
                   2316:        pc->pc_emptygroups = NULL;
                   2317:        pc->pc_partgroups = NULL;
                   2318:        pc->pc_nfull = 0;
                   2319:        pc->pc_nempty = 0;
                   2320:        pc->pc_npart = 0;
                   2321:        mutex_exit(&pc->pc_lock);
                   2322:
                   2323:        pool_cache_invalidate_groups(pc, full);
                   2324:        pool_cache_invalidate_groups(pc, empty);
                   2325:        pool_cache_invalidate_groups(pc, part);
                   2326: }
                   2327:
                   2328: void
                   2329: pool_cache_set_drain_hook(pool_cache_t pc, void (*fn)(void *, int), void *arg)
                   2330: {
                   2331:
                   2332:        pool_set_drain_hook(&pc->pc_pool, fn, arg);
                   2333: }
                   2334:
                   2335: void
                   2336: pool_cache_setlowat(pool_cache_t pc, int n)
                   2337: {
                   2338:
                   2339:        pool_setlowat(&pc->pc_pool, n);
                   2340: }
                   2341:
                   2342: void
                   2343: pool_cache_sethiwat(pool_cache_t pc, int n)
                   2344: {
                   2345:
                   2346:        pool_sethiwat(&pc->pc_pool, n);
                   2347: }
                   2348:
                   2349: void
                   2350: pool_cache_sethardlimit(pool_cache_t pc, int n, const char *warnmess, int ratecap)
                   2351: {
                   2352:
                   2353:        pool_sethardlimit(&pc->pc_pool, n, warnmess, ratecap);
                   2354: }
                   2355:
                   2356: static inline pool_cache_cpu_t *
                   2357: pool_cache_cpu_enter(pool_cache_t pc, int *s)
                   2358: {
                   2359:        pool_cache_cpu_t *cc;
                   2360:
                   2361:        /*
                   2362:         * Prevent other users of the cache from accessing our
                   2363:         * CPU-local data.  To avoid touching shared state, we
                   2364:         * pull the neccessary information from CPU local data.
                   2365:         */
1.137     ad       2366:        crit_enter();
                   2367:        cc = pc->pc_cpus[curcpu()->ci_index];
1.134     ad       2368:        KASSERT(cc->cc_cache == pc);
1.137     ad       2369:        if (cc->cc_ipl != IPL_NONE) {
1.134     ad       2370:                *s = splraiseipl(cc->cc_iplcookie);
                   2371:        }
                   2372:        KASSERT(((uintptr_t)cc & (CACHE_LINE_SIZE - 1)) == 0);
                   2373:
                   2374:        return cc;
                   2375: }
                   2376:
                   2377: static inline void
                   2378: pool_cache_cpu_exit(pool_cache_cpu_t *cc, int *s)
                   2379: {
                   2380:
                   2381:        /* No longer need exclusive access to the per-CPU data. */
1.137     ad       2382:        if (cc->cc_ipl != IPL_NONE) {
1.134     ad       2383:                splx(*s);
                   2384:        }
1.137     ad       2385:        crit_exit();
1.134     ad       2386: }
                   2387:
                   2388: #if __GNUC_PREREQ__(3, 0)
                   2389: __attribute ((noinline))
                   2390: #endif
                   2391: pool_cache_cpu_t *
                   2392: pool_cache_get_slow(pool_cache_cpu_t *cc, int *s, void **objectp,
                   2393:                    paddr_t *pap, int flags)
1.43      thorpej  2394: {
1.134     ad       2395:        pcg_t *pcg, *cur;
                   2396:        uint64_t ncsw;
                   2397:        pool_cache_t pc;
1.43      thorpej  2398:        void *object;
1.58      thorpej  2399:
1.134     ad       2400:        pc = cc->cc_cache;
                   2401:        cc->cc_misses++;
1.43      thorpej  2402:
1.134     ad       2403:        /*
                   2404:         * Nothing was available locally.  Try and grab a group
                   2405:         * from the cache.
                   2406:         */
                   2407:        if (!mutex_tryenter(&pc->pc_lock)) {
                   2408:                ncsw = curlwp->l_ncsw;
                   2409:                mutex_enter(&pc->pc_lock);
                   2410:                pc->pc_contended++;
1.43      thorpej  2411:
1.134     ad       2412:                /*
                   2413:                 * If we context switched while locking, then
                   2414:                 * our view of the per-CPU data is invalid:
                   2415:                 * retry.
                   2416:                 */
                   2417:                if (curlwp->l_ncsw != ncsw) {
                   2418:                        mutex_exit(&pc->pc_lock);
                   2419:                        pool_cache_cpu_exit(cc, s);
                   2420:                        return pool_cache_cpu_enter(pc, s);
1.43      thorpej  2421:                }
1.102     chs      2422:        }
1.43      thorpej  2423:
1.134     ad       2424:        if ((pcg = pc->pc_fullgroups) != NULL) {
1.43      thorpej  2425:                /*
1.134     ad       2426:                 * If there's a full group, release our empty
                   2427:                 * group back to the cache.  Install the full
                   2428:                 * group as cc_current and return.
1.43      thorpej  2429:                 */
1.134     ad       2430:                if ((cur = cc->cc_current) != NULL) {
                   2431:                        KASSERT(cur->pcg_avail == 0);
                   2432:                        cur->pcg_next = pc->pc_emptygroups;
                   2433:                        pc->pc_emptygroups = cur;
                   2434:                        pc->pc_nempty++;
1.87      thorpej  2435:                }
1.134     ad       2436:                KASSERT(pcg->pcg_avail == PCG_NOBJECTS);
                   2437:                cc->cc_current = pcg;
                   2438:                pc->pc_fullgroups = pcg->pcg_next;
                   2439:                pc->pc_hits++;
                   2440:                pc->pc_nfull--;
                   2441:                mutex_exit(&pc->pc_lock);
                   2442:                return cc;
                   2443:        }
                   2444:
                   2445:        /*
                   2446:         * Nothing available locally or in cache.  Take the slow
                   2447:         * path: fetch a new object from the pool and construct
                   2448:         * it.
                   2449:         */
                   2450:        pc->pc_misses++;
                   2451:        mutex_exit(&pc->pc_lock);
                   2452:        pool_cache_cpu_exit(cc, s);
                   2453:
                   2454:        object = pool_get(&pc->pc_pool, flags);
                   2455:        *objectp = object;
                   2456:        if (object == NULL)
                   2457:                return NULL;
1.125     ad       2458:
1.134     ad       2459:        if ((*pc->pc_ctor)(pc->pc_arg, object, flags) != 0) {
                   2460:                pool_put(&pc->pc_pool, object);
                   2461:                *objectp = NULL;
                   2462:                return NULL;
1.43      thorpej  2463:        }
                   2464:
1.134     ad       2465:        KASSERT((((vaddr_t)object + pc->pc_pool.pr_itemoffset) &
                   2466:            (pc->pc_pool.pr_align - 1)) == 0);
1.43      thorpej  2467:
1.134     ad       2468:        if (pap != NULL) {
                   2469: #ifdef POOL_VTOPHYS
                   2470:                *pap = POOL_VTOPHYS(object);
                   2471: #else
                   2472:                *pap = POOL_PADDR_INVALID;
                   2473: #endif
1.102     chs      2474:        }
1.43      thorpej  2475:
1.125     ad       2476:        FREECHECK_OUT(&pc->pc_freecheck, object);
1.134     ad       2477:        return NULL;
1.43      thorpej  2478: }
                   2479:
                   2480: /*
1.134     ad       2481:  * pool_cache_get{,_paddr}:
1.43      thorpej  2482:  *
1.134     ad       2483:  *     Get an object from a pool cache (optionally returning
                   2484:  *     the physical address of the object).
1.43      thorpej  2485:  */
1.134     ad       2486: void *
                   2487: pool_cache_get_paddr(pool_cache_t pc, int flags, paddr_t *pap)
1.43      thorpej  2488: {
1.134     ad       2489:        pool_cache_cpu_t *cc;
                   2490:        pcg_t *pcg;
                   2491:        void *object;
1.60      thorpej  2492:        int s;
1.43      thorpej  2493:
1.134     ad       2494: #ifdef LOCKDEBUG
                   2495:        if (flags & PR_WAITOK)
                   2496:                ASSERT_SLEEPABLE(NULL, "pool_cache_get(PR_WAITOK)");
                   2497: #endif
1.125     ad       2498:
1.134     ad       2499:        cc = pool_cache_cpu_enter(pc, &s);
                   2500:        do {
                   2501:                /* Try and allocate an object from the current group. */
                   2502:                pcg = cc->cc_current;
                   2503:                if (pcg != NULL && pcg->pcg_avail > 0) {
                   2504:                        object = pcg->pcg_objects[--pcg->pcg_avail].pcgo_va;
                   2505:                        if (pap != NULL)
                   2506:                                *pap = pcg->pcg_objects[pcg->pcg_avail].pcgo_pa;
                   2507:                        pcg->pcg_objects[pcg->pcg_avail].pcgo_va = NULL;
                   2508:                        KASSERT(pcg->pcg_avail <= PCG_NOBJECTS);
                   2509:                        KASSERT(object != NULL);
                   2510:                        cc->cc_hits++;
                   2511:                        pool_cache_cpu_exit(cc, &s);
                   2512:                        FREECHECK_OUT(&pc->pc_freecheck, object);
                   2513:                        return object;
1.43      thorpej  2514:                }
                   2515:
                   2516:                /*
1.134     ad       2517:                 * That failed.  If the previous group isn't empty, swap
                   2518:                 * it with the current group and allocate from there.
1.43      thorpej  2519:                 */
1.134     ad       2520:                pcg = cc->cc_previous;
                   2521:                if (pcg != NULL && pcg->pcg_avail > 0) {
                   2522:                        cc->cc_previous = cc->cc_current;
                   2523:                        cc->cc_current = pcg;
                   2524:                        continue;
1.43      thorpej  2525:                }
                   2526:
1.134     ad       2527:                /*
                   2528:                 * Can't allocate from either group: try the slow path.
                   2529:                 * If get_slow() allocated an object for us, or if
                   2530:                 * no more objects are available, it will return NULL.
                   2531:                 * Otherwise, we need to retry.
                   2532:                 */
                   2533:                cc = pool_cache_get_slow(cc, &s, &object, pap, flags);
                   2534:        } while (cc != NULL);
1.43      thorpej  2535:
1.134     ad       2536:        return object;
1.51      thorpej  2537: }
                   2538:
1.134     ad       2539: #if __GNUC_PREREQ__(3, 0)
                   2540: __attribute ((noinline))
                   2541: #endif
                   2542: pool_cache_cpu_t *
                   2543: pool_cache_put_slow(pool_cache_cpu_t *cc, int *s, void *object, paddr_t pa)
1.51      thorpej  2544: {
1.134     ad       2545:        pcg_t *pcg, *cur;
                   2546:        uint64_t ncsw;
                   2547:        pool_cache_t pc;
1.51      thorpej  2548:
1.134     ad       2549:        pc = cc->cc_cache;
                   2550:        cc->cc_misses++;
1.43      thorpej  2551:
1.134     ad       2552:        /*
                   2553:         * No free slots locally.  Try to grab an empty, unused
                   2554:         * group from the cache.
                   2555:         */
                   2556:        if (!mutex_tryenter(&pc->pc_lock)) {
                   2557:                ncsw = curlwp->l_ncsw;
                   2558:                mutex_enter(&pc->pc_lock);
                   2559:                pc->pc_contended++;
1.102     chs      2560:
1.134     ad       2561:                /*
                   2562:                 * If we context switched while locking, then
                   2563:                 * our view of the per-CPU data is invalid:
                   2564:                 * retry.
                   2565:                 */
                   2566:                if (curlwp->l_ncsw != ncsw) {
                   2567:                        mutex_exit(&pc->pc_lock);
                   2568:                        pool_cache_cpu_exit(cc, s);
                   2569:                        return pool_cache_cpu_enter(pc, s);
                   2570:                }
                   2571:        }
1.130     ad       2572:
1.134     ad       2573:        if ((pcg = pc->pc_emptygroups) != NULL) {
                   2574:                /*
                   2575:                 * If there's a empty group, release our full
                   2576:                 * group back to the cache.  Install the empty
                   2577:                 * group as cc_current and return.
                   2578:                 */
                   2579:                if ((cur = cc->cc_current) != NULL) {
                   2580:                        KASSERT(cur->pcg_avail == PCG_NOBJECTS);
                   2581:                        cur->pcg_next = pc->pc_fullgroups;
                   2582:                        pc->pc_fullgroups = cur;
                   2583:                        pc->pc_nfull++;
1.102     chs      2584:                }
1.134     ad       2585:                KASSERT(pcg->pcg_avail == 0);
                   2586:                cc->cc_current = pcg;
                   2587:                pc->pc_emptygroups = pcg->pcg_next;
                   2588:                pc->pc_hits++;
                   2589:                pc->pc_nempty--;
                   2590:                mutex_exit(&pc->pc_lock);
                   2591:                return cc;
1.102     chs      2592:        }
1.105     christos 2593:
1.134     ad       2594:        /*
                   2595:         * Nothing available locally or in cache.  Take the
                   2596:         * slow path and try to allocate a new group that we
                   2597:         * can release to.
                   2598:         */
                   2599:        pc->pc_misses++;
                   2600:        mutex_exit(&pc->pc_lock);
                   2601:        pool_cache_cpu_exit(cc, s);
1.105     christos 2602:
1.134     ad       2603:        /*
                   2604:         * If we can't allocate a new group, just throw the
                   2605:         * object away.
                   2606:         */
1.137.2.2  ad       2607:        if (pool_cache_disable) {
                   2608:                pcg = NULL;
                   2609:        } else {
                   2610:                pcg = pool_get(&pcgpool, PR_NOWAIT);
                   2611:        }
1.134     ad       2612:        if (pcg == NULL) {
                   2613:                pool_cache_destruct_object(pc, object);
                   2614:                return NULL;
                   2615:        }
                   2616: #ifdef DIAGNOSTIC
                   2617:        memset(pcg, 0, sizeof(*pcg));
                   2618: #else
                   2619:        pcg->pcg_avail = 0;
                   2620: #endif
1.105     christos 2621:
1.134     ad       2622:        /*
                   2623:         * Add the empty group to the cache and try again.
                   2624:         */
                   2625:        mutex_enter(&pc->pc_lock);
                   2626:        pcg->pcg_next = pc->pc_emptygroups;
                   2627:        pc->pc_emptygroups = pcg;
                   2628:        pc->pc_nempty++;
                   2629:        mutex_exit(&pc->pc_lock);
1.103     chs      2630:
1.134     ad       2631:        return pool_cache_cpu_enter(pc, s);
                   2632: }
1.102     chs      2633:
1.43      thorpej  2634: /*
1.134     ad       2635:  * pool_cache_put{,_paddr}:
1.43      thorpej  2636:  *
1.134     ad       2637:  *     Put an object back to the pool cache (optionally caching the
                   2638:  *     physical address of the object).
1.43      thorpej  2639:  */
1.101     thorpej  2640: void
1.134     ad       2641: pool_cache_put_paddr(pool_cache_t pc, void *object, paddr_t pa)
1.43      thorpej  2642: {
1.134     ad       2643:        pool_cache_cpu_t *cc;
                   2644:        pcg_t *pcg;
                   2645:        int s;
1.101     thorpej  2646:
1.134     ad       2647:        FREECHECK_IN(&pc->pc_freecheck, object);
1.101     thorpej  2648:
1.134     ad       2649:        cc = pool_cache_cpu_enter(pc, &s);
                   2650:        do {
                   2651:                /* If the current group isn't full, release it there. */
                   2652:                pcg = cc->cc_current;
                   2653:                if (pcg != NULL && pcg->pcg_avail < PCG_NOBJECTS) {
                   2654:                        KASSERT(pcg->pcg_objects[pcg->pcg_avail].pcgo_va
                   2655:                            == NULL);
                   2656:                        pcg->pcg_objects[pcg->pcg_avail].pcgo_va = object;
                   2657:                        pcg->pcg_objects[pcg->pcg_avail].pcgo_pa = pa;
                   2658:                        pcg->pcg_avail++;
                   2659:                        cc->cc_hits++;
                   2660:                        pool_cache_cpu_exit(cc, &s);
                   2661:                        return;
                   2662:                }
1.43      thorpej  2663:
1.134     ad       2664:                /*
                   2665:                 * That failed.  If the previous group is empty, swap
                   2666:                 * it with the current group and try again.
                   2667:                 */
                   2668:                pcg = cc->cc_previous;
                   2669:                if (pcg != NULL && pcg->pcg_avail == 0) {
                   2670:                        cc->cc_previous = cc->cc_current;
                   2671:                        cc->cc_current = pcg;
                   2672:                        continue;
                   2673:                }
1.43      thorpej  2674:
1.134     ad       2675:                /*
                   2676:                 * Can't free to either group: try the slow path.
                   2677:                 * If put_slow() releases the object for us, it
                   2678:                 * will return NULL.  Otherwise we need to retry.
                   2679:                 */
                   2680:                cc = pool_cache_put_slow(cc, &s, object, pa);
                   2681:        } while (cc != NULL);
1.43      thorpej  2682: }
                   2683:
                   2684: /*
1.134     ad       2685:  * pool_cache_xcall:
1.43      thorpej  2686:  *
1.134     ad       2687:  *     Transfer objects from the per-CPU cache to the global cache.
                   2688:  *     Run within a cross-call thread.
1.43      thorpej  2689:  */
                   2690: static void
1.134     ad       2691: pool_cache_xcall(pool_cache_t pc)
1.43      thorpej  2692: {
1.134     ad       2693:        pool_cache_cpu_t *cc;
                   2694:        pcg_t *prev, *cur, **list;
                   2695:        int s = 0; /* XXXgcc */
                   2696:
                   2697:        cc = pool_cache_cpu_enter(pc, &s);
                   2698:        cur = cc->cc_current;
                   2699:        cc->cc_current = NULL;
                   2700:        prev = cc->cc_previous;
                   2701:        cc->cc_previous = NULL;
                   2702:        pool_cache_cpu_exit(cc, &s);
                   2703:
                   2704:        /*
                   2705:         * XXXSMP Go to splvm to prevent kernel_lock from being taken,
                   2706:         * because locks at IPL_SOFTXXX are still spinlocks.  Does not
                   2707:         * apply to IPL_SOFTBIO.  Cross-call threads do not take the
                   2708:         * kernel_lock.
1.101     thorpej  2709:         */
1.134     ad       2710:        s = splvm();
                   2711:        mutex_enter(&pc->pc_lock);
                   2712:        if (cur != NULL) {
                   2713:                if (cur->pcg_avail == PCG_NOBJECTS) {
                   2714:                        list = &pc->pc_fullgroups;
                   2715:                        pc->pc_nfull++;
                   2716:                } else if (cur->pcg_avail == 0) {
                   2717:                        list = &pc->pc_emptygroups;
                   2718:                        pc->pc_nempty++;
                   2719:                } else {
                   2720:                        list = &pc->pc_partgroups;
                   2721:                        pc->pc_npart++;
                   2722:                }
                   2723:                cur->pcg_next = *list;
                   2724:                *list = cur;
                   2725:        }
                   2726:        if (prev != NULL) {
                   2727:                if (prev->pcg_avail == PCG_NOBJECTS) {
                   2728:                        list = &pc->pc_fullgroups;
                   2729:                        pc->pc_nfull++;
                   2730:                } else if (prev->pcg_avail == 0) {
                   2731:                        list = &pc->pc_emptygroups;
                   2732:                        pc->pc_nempty++;
                   2733:                } else {
                   2734:                        list = &pc->pc_partgroups;
                   2735:                        pc->pc_npart++;
                   2736:                }
                   2737:                prev->pcg_next = *list;
                   2738:                *list = prev;
                   2739:        }
                   2740:        mutex_exit(&pc->pc_lock);
                   2741:        splx(s);
1.3       pk       2742: }
1.66      thorpej  2743:
                   2744: /*
                   2745:  * Pool backend allocators.
                   2746:  *
                   2747:  * Each pool has a backend allocator that handles allocation, deallocation,
                   2748:  * and any additional draining that might be needed.
                   2749:  *
                   2750:  * We provide two standard allocators:
                   2751:  *
                   2752:  *     pool_allocator_kmem - the default when no allocator is specified
                   2753:  *
                   2754:  *     pool_allocator_nointr - used for pools that will not be accessed
                   2755:  *     in interrupt context.
                   2756:  */
                   2757: void   *pool_page_alloc(struct pool *, int);
                   2758: void   pool_page_free(struct pool *, void *);
                   2759:
1.112     bjh21    2760: #ifdef POOL_SUBPAGE
                   2761: struct pool_allocator pool_allocator_kmem_fullpage = {
                   2762:        pool_page_alloc, pool_page_free, 0,
1.117     yamt     2763:        .pa_backingmapptr = &kmem_map,
1.112     bjh21    2764: };
                   2765: #else
1.66      thorpej  2766: struct pool_allocator pool_allocator_kmem = {
                   2767:        pool_page_alloc, pool_page_free, 0,
1.117     yamt     2768:        .pa_backingmapptr = &kmem_map,
1.66      thorpej  2769: };
1.112     bjh21    2770: #endif
1.66      thorpej  2771:
                   2772: void   *pool_page_alloc_nointr(struct pool *, int);
                   2773: void   pool_page_free_nointr(struct pool *, void *);
                   2774:
1.112     bjh21    2775: #ifdef POOL_SUBPAGE
                   2776: struct pool_allocator pool_allocator_nointr_fullpage = {
                   2777:        pool_page_alloc_nointr, pool_page_free_nointr, 0,
1.117     yamt     2778:        .pa_backingmapptr = &kernel_map,
1.112     bjh21    2779: };
                   2780: #else
1.66      thorpej  2781: struct pool_allocator pool_allocator_nointr = {
                   2782:        pool_page_alloc_nointr, pool_page_free_nointr, 0,
1.117     yamt     2783:        .pa_backingmapptr = &kernel_map,
1.66      thorpej  2784: };
1.112     bjh21    2785: #endif
1.66      thorpej  2786:
                   2787: #ifdef POOL_SUBPAGE
                   2788: void   *pool_subpage_alloc(struct pool *, int);
                   2789: void   pool_subpage_free(struct pool *, void *);
                   2790:
1.112     bjh21    2791: struct pool_allocator pool_allocator_kmem = {
                   2792:        pool_subpage_alloc, pool_subpage_free, POOL_SUBPAGE,
1.117     yamt     2793:        .pa_backingmapptr = &kmem_map,
1.112     bjh21    2794: };
                   2795:
                   2796: void   *pool_subpage_alloc_nointr(struct pool *, int);
                   2797: void   pool_subpage_free_nointr(struct pool *, void *);
                   2798:
                   2799: struct pool_allocator pool_allocator_nointr = {
                   2800:        pool_subpage_alloc, pool_subpage_free, POOL_SUBPAGE,
1.117     yamt     2801:        .pa_backingmapptr = &kmem_map,
1.66      thorpej  2802: };
                   2803: #endif /* POOL_SUBPAGE */
                   2804:
1.117     yamt     2805: static void *
                   2806: pool_allocator_alloc(struct pool *pp, int flags)
1.66      thorpej  2807: {
1.117     yamt     2808:        struct pool_allocator *pa = pp->pr_alloc;
1.66      thorpej  2809:        void *res;
                   2810:
1.117     yamt     2811:        res = (*pa->pa_alloc)(pp, flags);
                   2812:        if (res == NULL && (flags & PR_WAITOK) == 0) {
1.66      thorpej  2813:                /*
1.117     yamt     2814:                 * We only run the drain hook here if PR_NOWAIT.
                   2815:                 * In other cases, the hook will be run in
                   2816:                 * pool_reclaim().
1.66      thorpej  2817:                 */
1.117     yamt     2818:                if (pp->pr_drain_hook != NULL) {
                   2819:                        (*pp->pr_drain_hook)(pp->pr_drain_hook_arg, flags);
                   2820:                        res = (*pa->pa_alloc)(pp, flags);
1.66      thorpej  2821:                }
1.117     yamt     2822:        }
                   2823:        return res;
1.66      thorpej  2824: }
                   2825:
1.117     yamt     2826: static void
1.66      thorpej  2827: pool_allocator_free(struct pool *pp, void *v)
                   2828: {
                   2829:        struct pool_allocator *pa = pp->pr_alloc;
                   2830:
                   2831:        (*pa->pa_free)(pp, v);
                   2832: }
                   2833:
                   2834: void *
1.124     yamt     2835: pool_page_alloc(struct pool *pp, int flags)
1.66      thorpej  2836: {
1.127     thorpej  2837:        bool waitok = (flags & PR_WAITOK) ? true : false;
1.66      thorpej  2838:
1.100     yamt     2839:        return ((void *) uvm_km_alloc_poolpage_cache(kmem_map, waitok));
1.66      thorpej  2840: }
                   2841:
                   2842: void
1.124     yamt     2843: pool_page_free(struct pool *pp, void *v)
1.66      thorpej  2844: {
                   2845:
1.98      yamt     2846:        uvm_km_free_poolpage_cache(kmem_map, (vaddr_t) v);
                   2847: }
                   2848:
                   2849: static void *
1.124     yamt     2850: pool_page_alloc_meta(struct pool *pp, int flags)
1.98      yamt     2851: {
1.127     thorpej  2852:        bool waitok = (flags & PR_WAITOK) ? true : false;
1.98      yamt     2853:
1.100     yamt     2854:        return ((void *) uvm_km_alloc_poolpage(kmem_map, waitok));
1.98      yamt     2855: }
                   2856:
                   2857: static void
1.124     yamt     2858: pool_page_free_meta(struct pool *pp, void *v)
1.98      yamt     2859: {
                   2860:
1.100     yamt     2861:        uvm_km_free_poolpage(kmem_map, (vaddr_t) v);
1.66      thorpej  2862: }
                   2863:
                   2864: #ifdef POOL_SUBPAGE
                   2865: /* Sub-page allocator, for machines with large hardware pages. */
                   2866: void *
                   2867: pool_subpage_alloc(struct pool *pp, int flags)
                   2868: {
1.134     ad       2869:        return pool_get(&psppool, flags);
1.66      thorpej  2870: }
                   2871:
                   2872: void
                   2873: pool_subpage_free(struct pool *pp, void *v)
                   2874: {
                   2875:        pool_put(&psppool, v);
                   2876: }
                   2877:
                   2878: /* We don't provide a real nointr allocator.  Maybe later. */
                   2879: void *
1.112     bjh21    2880: pool_subpage_alloc_nointr(struct pool *pp, int flags)
1.66      thorpej  2881: {
                   2882:
                   2883:        return (pool_subpage_alloc(pp, flags));
                   2884: }
                   2885:
                   2886: void
1.112     bjh21    2887: pool_subpage_free_nointr(struct pool *pp, void *v)
1.66      thorpej  2888: {
                   2889:
                   2890:        pool_subpage_free(pp, v);
                   2891: }
1.112     bjh21    2892: #endif /* POOL_SUBPAGE */
1.66      thorpej  2893: void *
1.124     yamt     2894: pool_page_alloc_nointr(struct pool *pp, int flags)
1.66      thorpej  2895: {
1.127     thorpej  2896:        bool waitok = (flags & PR_WAITOK) ? true : false;
1.66      thorpej  2897:
1.100     yamt     2898:        return ((void *) uvm_km_alloc_poolpage_cache(kernel_map, waitok));
1.66      thorpej  2899: }
                   2900:
                   2901: void
1.124     yamt     2902: pool_page_free_nointr(struct pool *pp, void *v)
1.66      thorpej  2903: {
                   2904:
1.98      yamt     2905:        uvm_km_free_poolpage_cache(kernel_map, (vaddr_t) v);
1.66      thorpej  2906: }

CVSweb <webmaster@jp.NetBSD.org>