[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 1.198.2.3

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

CVSweb <webmaster@jp.NetBSD.org>