[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 1.214

1.214   ! christos    1: /*     $NetBSD: subr_pool.c,v 1.213 2017/11/09 15:53:40 christos Exp $ */
1.1       pk          2:
                      3: /*-
1.204     maxv        4:  * Copyright (c) 1997, 1999, 2000, 2002, 2007, 2008, 2010, 2014, 2015
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
1.204     maxv       10:  * Simulation Facility, NASA Ames Research Center; by Andrew Doran, and by
                     11:  * Maxime Villard.
1.1       pk         12:  *
                     13:  * Redistribution and use in source and binary forms, with or without
                     14:  * modification, are permitted provided that the following conditions
                     15:  * are met:
                     16:  * 1. Redistributions of source code must retain the above copyright
                     17:  *    notice, this list of conditions and the following disclaimer.
                     18:  * 2. Redistributions in binary form must reproduce the above copyright
                     19:  *    notice, this list of conditions and the following disclaimer in the
                     20:  *    documentation and/or other materials provided with the distribution.
                     21:  *
                     22:  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
                     23:  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
                     24:  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
                     25:  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
                     26:  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
                     27:  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
                     28:  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
                     29:  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
                     30:  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
                     31:  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
                     32:  * POSSIBILITY OF SUCH DAMAGE.
                     33:  */
1.64      lukem      34:
                     35: #include <sys/cdefs.h>
1.214   ! christos   36: __KERNEL_RCSID(0, "$NetBSD: subr_pool.c,v 1.213 2017/11/09 15:53:40 christos Exp $");
1.24      scottr     37:
1.205     pooka      38: #ifdef _KERNEL_OPT
1.141     yamt       39: #include "opt_ddb.h"
1.28      thorpej    40: #include "opt_lockdebug.h"
1.205     pooka      41: #endif
1.1       pk         42:
                     43: #include <sys/param.h>
                     44: #include <sys/systm.h>
1.203     joerg      45: #include <sys/sysctl.h>
1.135     yamt       46: #include <sys/bitops.h>
1.1       pk         47: #include <sys/proc.h>
                     48: #include <sys/errno.h>
                     49: #include <sys/kernel.h>
1.191     para       50: #include <sys/vmem.h>
1.1       pk         51: #include <sys/pool.h>
1.20      thorpej    52: #include <sys/syslog.h>
1.125     ad         53: #include <sys/debug.h>
1.134     ad         54: #include <sys/lockdebug.h>
                     55: #include <sys/xcall.h>
                     56: #include <sys/cpu.h>
1.145     ad         57: #include <sys/atomic.h>
1.3       pk         58:
1.187     uebayasi   59: #include <uvm/uvm_extern.h>
1.3       pk         60:
1.1       pk         61: /*
                     62:  * Pool resource management utility.
1.3       pk         63:  *
1.88      chs        64:  * Memory is allocated in pages which are split into pieces according to
                     65:  * the pool item size. Each page is kept on one of three lists in the
                     66:  * pool structure: `pr_emptypages', `pr_fullpages' and `pr_partpages',
                     67:  * for empty, full and partially-full pages respectively. The individual
                     68:  * pool items are on a linked list headed by `ph_itemlist' in each page
                     69:  * header. The memory for building the page list is either taken from
                     70:  * the allocated pages themselves (for small pool items) or taken from
                     71:  * an internal pool of page headers (`phpool').
1.1       pk         72:  */
                     73:
1.202     abs        74: /* List of all pools. Non static as needed by 'vmstat -i' */
                     75: TAILQ_HEAD(, pool) pool_head = TAILQ_HEAD_INITIALIZER(pool_head);
1.134     ad         76:
1.3       pk         77: /* Private pool for page header structures */
1.97      yamt       78: #define        PHPOOL_MAX      8
                     79: static struct pool phpool[PHPOOL_MAX];
1.135     yamt       80: #define        PHPOOL_FREELIST_NELEM(idx) \
                     81:        (((idx) == 0) ? 0 : BITMAP_SIZE * (1 << (idx)))
1.3       pk         82:
1.62      bjh21      83: #ifdef POOL_SUBPAGE
                     84: /* Pool of subpages for use by normal pools. */
                     85: static struct pool psppool;
                     86: #endif
                     87:
1.204     maxv       88: #ifdef POOL_REDZONE
                     89: # define POOL_REDZONE_SIZE 2
                     90: static void pool_redzone_init(struct pool *, size_t);
                     91: static void pool_redzone_fill(struct pool *, void *);
                     92: static void pool_redzone_check(struct pool *, void *);
                     93: #else
                     94: # define pool_redzone_init(pp, sz)     /* NOTHING */
                     95: # define pool_redzone_fill(pp, ptr)    /* NOTHING */
                     96: # define pool_redzone_check(pp, ptr)   /* NOTHING */
                     97: #endif
                     98:
1.98      yamt       99: static void *pool_page_alloc_meta(struct pool *, int);
                    100: static void pool_page_free_meta(struct pool *, void *);
                    101:
                    102: /* allocator for pool metadata */
1.134     ad        103: struct pool_allocator pool_allocator_meta = {
1.191     para      104:        .pa_alloc = pool_page_alloc_meta,
                    105:        .pa_free = pool_page_free_meta,
                    106:        .pa_pagesz = 0
1.98      yamt      107: };
                    108:
1.208     chs       109: #define POOL_ALLOCATOR_BIG_BASE 13
                    110: extern struct pool_allocator pool_allocator_big[];
                    111: static int pool_bigidx(size_t);
                    112:
1.3       pk        113: /* # of seconds to retain page after last use */
                    114: int pool_inactive_time = 10;
                    115:
                    116: /* Next candidate for drainage (see pool_drain()) */
1.23      thorpej   117: static struct pool     *drainpp;
                    118:
1.134     ad        119: /* This lock protects both pool_head and drainpp. */
                    120: static kmutex_t pool_head_lock;
                    121: static kcondvar_t pool_busy;
1.3       pk        122:
1.178     elad      123: /* This lock protects initialization of a potentially shared pool allocator */
                    124: static kmutex_t pool_allocator_lock;
                    125:
1.135     yamt      126: typedef uint32_t pool_item_bitmap_t;
                    127: #define        BITMAP_SIZE     (CHAR_BIT * sizeof(pool_item_bitmap_t))
                    128: #define        BITMAP_MASK     (BITMAP_SIZE - 1)
1.99      yamt      129:
1.3       pk        130: struct pool_item_header {
                    131:        /* Page headers */
1.88      chs       132:        LIST_ENTRY(pool_item_header)
1.3       pk        133:                                ph_pagelist;    /* pool page list */
1.88      chs       134:        SPLAY_ENTRY(pool_item_header)
                    135:                                ph_node;        /* Off-page page headers */
1.128     christos  136:        void *                  ph_page;        /* this page's address */
1.151     yamt      137:        uint32_t                ph_time;        /* last referenced */
1.135     yamt      138:        uint16_t                ph_nmissing;    /* # of chunks in use */
1.141     yamt      139:        uint16_t                ph_off;         /* start offset in page */
1.97      yamt      140:        union {
                    141:                /* !PR_NOTOUCH */
                    142:                struct {
1.102     chs       143:                        LIST_HEAD(, pool_item)
1.97      yamt      144:                                phu_itemlist;   /* chunk list for this page */
                    145:                } phu_normal;
                    146:                /* PR_NOTOUCH */
                    147:                struct {
1.141     yamt      148:                        pool_item_bitmap_t phu_bitmap[1];
1.97      yamt      149:                } phu_notouch;
                    150:        } ph_u;
1.3       pk        151: };
1.97      yamt      152: #define        ph_itemlist     ph_u.phu_normal.phu_itemlist
1.135     yamt      153: #define        ph_bitmap       ph_u.phu_notouch.phu_bitmap
1.3       pk        154:
1.1       pk        155: struct pool_item {
1.3       pk        156: #ifdef DIAGNOSTIC
1.82      thorpej   157:        u_int pi_magic;
1.33      chs       158: #endif
1.134     ad        159: #define        PI_MAGIC 0xdeaddeadU
1.3       pk        160:        /* Other entries use only this list entry */
1.102     chs       161:        LIST_ENTRY(pool_item)   pi_list;
1.3       pk        162: };
                    163:
1.53      thorpej   164: #define        POOL_NEEDS_CATCHUP(pp)                                          \
                    165:        ((pp)->pr_nitems < (pp)->pr_minitems)
                    166:
1.43      thorpej   167: /*
                    168:  * Pool cache management.
                    169:  *
                    170:  * Pool caches provide a way for constructed objects to be cached by the
                    171:  * pool subsystem.  This can lead to performance improvements by avoiding
                    172:  * needless object construction/destruction; it is deferred until absolutely
                    173:  * necessary.
                    174:  *
1.134     ad        175:  * Caches are grouped into cache groups.  Each cache group references up
                    176:  * to PCG_NUMOBJECTS constructed objects.  When a cache allocates an
                    177:  * object from the pool, it calls the object's constructor and places it
                    178:  * into a cache group.  When a cache group frees an object back to the
                    179:  * pool, it first calls the object's destructor.  This allows the object
                    180:  * to persist in constructed form while freed to the cache.
                    181:  *
                    182:  * The pool references each cache, so that when a pool is drained by the
                    183:  * pagedaemon, it can drain each individual cache as well.  Each time a
                    184:  * cache is drained, the most idle cache group is freed to the pool in
                    185:  * its entirety.
1.43      thorpej   186:  *
                    187:  * Pool caches are layed on top of pools.  By layering them, we can avoid
                    188:  * the complexity of cache management for pools which would not benefit
                    189:  * from it.
                    190:  */
                    191:
1.142     ad        192: static struct pool pcg_normal_pool;
                    193: static struct pool pcg_large_pool;
1.134     ad        194: static struct pool cache_pool;
                    195: static struct pool cache_cpu_pool;
1.3       pk        196:
1.189     pooka     197: pool_cache_t pnbuf_cache;      /* pathname buffer cache */
                    198:
1.145     ad        199: /* List of all caches. */
                    200: TAILQ_HEAD(,pool_cache) pool_cache_head =
                    201:     TAILQ_HEAD_INITIALIZER(pool_cache_head);
                    202:
1.162     ad        203: int pool_cache_disable;                /* global disable for caching */
1.169     yamt      204: static const pcg_t pcg_dummy;  /* zero sized: always empty, yet always full */
1.145     ad        205:
1.162     ad        206: static bool    pool_cache_put_slow(pool_cache_cpu_t *, int,
                    207:                                    void *);
                    208: static bool    pool_cache_get_slow(pool_cache_cpu_t *, int,
                    209:                                    void **, paddr_t *, int);
1.134     ad        210: static void    pool_cache_cpu_init1(struct cpu_info *, pool_cache_t);
                    211: static void    pool_cache_invalidate_groups(pool_cache_t, pcg_t *);
1.175     jym       212: static void    pool_cache_invalidate_cpu(pool_cache_t, u_int);
1.196     jym       213: static void    pool_cache_transfer(pool_cache_t);
1.3       pk        214:
1.42      thorpej   215: static int     pool_catchup(struct pool *);
1.128     christos  216: static void    pool_prime_page(struct pool *, void *,
1.55      thorpej   217:                    struct pool_item_header *);
1.88      chs       218: static void    pool_update_curpage(struct pool *);
1.66      thorpej   219:
1.113     yamt      220: static int     pool_grow(struct pool *, int);
1.117     yamt      221: static void    *pool_allocator_alloc(struct pool *, int);
                    222: static void    pool_allocator_free(struct pool *, void *);
1.3       pk        223:
1.97      yamt      224: static void pool_print_pagelist(struct pool *, struct pool_pagelist *,
1.199     christos  225:        void (*)(const char *, ...) __printflike(1, 2));
1.42      thorpej   226: static void pool_print1(struct pool *, const char *,
1.199     christos  227:        void (*)(const char *, ...) __printflike(1, 2));
1.3       pk        228:
1.88      chs       229: static int pool_chk_page(struct pool *, const char *,
                    230:                         struct pool_item_header *);
                    231:
1.135     yamt      232: static inline unsigned int
1.97      yamt      233: pr_item_notouch_index(const struct pool *pp, const struct pool_item_header *ph,
                    234:     const void *v)
                    235: {
                    236:        const char *cp = v;
1.135     yamt      237:        unsigned int idx;
1.97      yamt      238:
                    239:        KASSERT(pp->pr_roflags & PR_NOTOUCH);
1.128     christos  240:        idx = (cp - (char *)ph->ph_page - ph->ph_off) / pp->pr_size;
1.97      yamt      241:        KASSERT(idx < pp->pr_itemsperpage);
                    242:        return idx;
                    243: }
                    244:
1.110     perry     245: static inline void
1.97      yamt      246: pr_item_notouch_put(const struct pool *pp, struct pool_item_header *ph,
                    247:     void *obj)
                    248: {
1.135     yamt      249:        unsigned int idx = pr_item_notouch_index(pp, ph, obj);
                    250:        pool_item_bitmap_t *bitmap = ph->ph_bitmap + (idx / BITMAP_SIZE);
                    251:        pool_item_bitmap_t mask = 1 << (idx & BITMAP_MASK);
1.97      yamt      252:
1.135     yamt      253:        KASSERT((*bitmap & mask) == 0);
                    254:        *bitmap |= mask;
1.97      yamt      255: }
                    256:
1.110     perry     257: static inline void *
1.97      yamt      258: pr_item_notouch_get(const struct pool *pp, struct pool_item_header *ph)
                    259: {
1.135     yamt      260:        pool_item_bitmap_t *bitmap = ph->ph_bitmap;
                    261:        unsigned int idx;
                    262:        int i;
1.97      yamt      263:
1.135     yamt      264:        for (i = 0; ; i++) {
                    265:                int bit;
1.97      yamt      266:
1.135     yamt      267:                KASSERT((i * BITMAP_SIZE) < pp->pr_itemsperpage);
                    268:                bit = ffs32(bitmap[i]);
                    269:                if (bit) {
                    270:                        pool_item_bitmap_t mask;
                    271:
                    272:                        bit--;
                    273:                        idx = (i * BITMAP_SIZE) + bit;
                    274:                        mask = 1 << bit;
                    275:                        KASSERT((bitmap[i] & mask) != 0);
                    276:                        bitmap[i] &= ~mask;
                    277:                        break;
                    278:                }
                    279:        }
                    280:        KASSERT(idx < pp->pr_itemsperpage);
1.128     christos  281:        return (char *)ph->ph_page + ph->ph_off + idx * pp->pr_size;
1.97      yamt      282: }
                    283:
1.135     yamt      284: static inline void
1.141     yamt      285: pr_item_notouch_init(const struct pool *pp, struct pool_item_header *ph)
1.135     yamt      286: {
                    287:        pool_item_bitmap_t *bitmap = ph->ph_bitmap;
                    288:        const int n = howmany(pp->pr_itemsperpage, BITMAP_SIZE);
                    289:        int i;
                    290:
                    291:        for (i = 0; i < n; i++) {
                    292:                bitmap[i] = (pool_item_bitmap_t)-1;
                    293:        }
                    294: }
                    295:
1.110     perry     296: static inline int
1.88      chs       297: phtree_compare(struct pool_item_header *a, struct pool_item_header *b)
                    298: {
1.121     yamt      299:
                    300:        /*
                    301:         * we consider pool_item_header with smaller ph_page bigger.
                    302:         * (this unnatural ordering is for the benefit of pr_find_pagehead.)
                    303:         */
                    304:
1.88      chs       305:        if (a->ph_page < b->ph_page)
1.121     yamt      306:                return (1);
                    307:        else if (a->ph_page > b->ph_page)
1.88      chs       308:                return (-1);
                    309:        else
                    310:                return (0);
                    311: }
                    312:
                    313: SPLAY_PROTOTYPE(phtree, pool_item_header, ph_node, phtree_compare);
                    314: SPLAY_GENERATE(phtree, pool_item_header, ph_node, phtree_compare);
                    315:
1.141     yamt      316: static inline struct pool_item_header *
                    317: pr_find_pagehead_noalign(struct pool *pp, void *v)
                    318: {
                    319:        struct pool_item_header *ph, tmp;
                    320:
                    321:        tmp.ph_page = (void *)(uintptr_t)v;
                    322:        ph = SPLAY_FIND(phtree, &pp->pr_phtree, &tmp);
                    323:        if (ph == NULL) {
                    324:                ph = SPLAY_ROOT(&pp->pr_phtree);
                    325:                if (ph != NULL && phtree_compare(&tmp, ph) >= 0) {
                    326:                        ph = SPLAY_NEXT(phtree, &pp->pr_phtree, ph);
                    327:                }
                    328:                KASSERT(ph == NULL || phtree_compare(&tmp, ph) < 0);
                    329:        }
                    330:
                    331:        return ph;
                    332: }
                    333:
1.3       pk        334: /*
1.121     yamt      335:  * Return the pool page header based on item address.
1.3       pk        336:  */
1.110     perry     337: static inline struct pool_item_header *
1.121     yamt      338: pr_find_pagehead(struct pool *pp, void *v)
1.3       pk        339: {
1.88      chs       340:        struct pool_item_header *ph, tmp;
1.3       pk        341:
1.121     yamt      342:        if ((pp->pr_roflags & PR_NOALIGN) != 0) {
1.141     yamt      343:                ph = pr_find_pagehead_noalign(pp, v);
1.121     yamt      344:        } else {
1.128     christos  345:                void *page =
                    346:                    (void *)((uintptr_t)v & pp->pr_alloc->pa_pagemask);
1.121     yamt      347:
                    348:                if ((pp->pr_roflags & PR_PHINPAGE) != 0) {
1.128     christos  349:                        ph = (struct pool_item_header *)((char *)page + pp->pr_phoffset);
1.121     yamt      350:                } else {
                    351:                        tmp.ph_page = page;
                    352:                        ph = SPLAY_FIND(phtree, &pp->pr_phtree, &tmp);
                    353:                }
                    354:        }
1.3       pk        355:
1.121     yamt      356:        KASSERT(ph == NULL || ((pp->pr_roflags & PR_PHINPAGE) != 0) ||
1.128     christos  357:            ((char *)ph->ph_page <= (char *)v &&
                    358:            (char *)v < (char *)ph->ph_page + pp->pr_alloc->pa_pagesz));
1.88      chs       359:        return ph;
1.3       pk        360: }
                    361:
1.101     thorpej   362: static void
                    363: pr_pagelist_free(struct pool *pp, struct pool_pagelist *pq)
                    364: {
                    365:        struct pool_item_header *ph;
                    366:
                    367:        while ((ph = LIST_FIRST(pq)) != NULL) {
                    368:                LIST_REMOVE(ph, ph_pagelist);
                    369:                pool_allocator_free(pp, ph->ph_page);
1.134     ad        370:                if ((pp->pr_roflags & PR_PHINPAGE) == 0)
1.101     thorpej   371:                        pool_put(pp->pr_phpool, ph);
                    372:        }
                    373: }
                    374:
1.3       pk        375: /*
                    376:  * Remove a page from the pool.
                    377:  */
1.110     perry     378: static inline void
1.61      chs       379: pr_rmpage(struct pool *pp, struct pool_item_header *ph,
                    380:      struct pool_pagelist *pq)
1.3       pk        381: {
                    382:
1.134     ad        383:        KASSERT(mutex_owned(&pp->pr_lock));
1.91      yamt      384:
1.3       pk        385:        /*
1.7       thorpej   386:         * If the page was idle, decrement the idle page count.
1.3       pk        387:         */
1.6       thorpej   388:        if (ph->ph_nmissing == 0) {
1.207     riastrad  389:                KASSERT(pp->pr_nidle != 0);
                    390:                KASSERTMSG((pp->pr_nitems >= pp->pr_itemsperpage),
                    391:                    "nitems=%u < itemsperpage=%u",
                    392:                    pp->pr_nitems, pp->pr_itemsperpage);
1.6       thorpej   393:                pp->pr_nidle--;
                    394:        }
1.7       thorpej   395:
1.20      thorpej   396:        pp->pr_nitems -= pp->pr_itemsperpage;
                    397:
1.7       thorpej   398:        /*
1.101     thorpej   399:         * Unlink the page from the pool and queue it for release.
1.7       thorpej   400:         */
1.88      chs       401:        LIST_REMOVE(ph, ph_pagelist);
1.91      yamt      402:        if ((pp->pr_roflags & PR_PHINPAGE) == 0)
                    403:                SPLAY_REMOVE(phtree, &pp->pr_phtree, ph);
1.101     thorpej   404:        LIST_INSERT_HEAD(pq, ph, ph_pagelist);
                    405:
1.7       thorpej   406:        pp->pr_npages--;
                    407:        pp->pr_npagefree++;
1.6       thorpej   408:
1.88      chs       409:        pool_update_curpage(pp);
1.3       pk        410: }
                    411:
                    412: /*
1.94      simonb    413:  * Initialize all the pools listed in the "pools" link set.
                    414:  */
                    415: void
1.117     yamt      416: pool_subsystem_init(void)
1.94      simonb    417: {
1.192     rmind     418:        size_t size;
1.191     para      419:        int idx;
1.94      simonb    420:
1.134     ad        421:        mutex_init(&pool_head_lock, MUTEX_DEFAULT, IPL_NONE);
1.179     mlelstv   422:        mutex_init(&pool_allocator_lock, MUTEX_DEFAULT, IPL_NONE);
1.134     ad        423:        cv_init(&pool_busy, "poolbusy");
                    424:
1.191     para      425:        /*
                    426:         * Initialize private page header pool and cache magazine pool if we
                    427:         * haven't done so yet.
                    428:         */
                    429:        for (idx = 0; idx < PHPOOL_MAX; idx++) {
                    430:                static char phpool_names[PHPOOL_MAX][6+1+6+1];
                    431:                int nelem;
                    432:                size_t sz;
                    433:
                    434:                nelem = PHPOOL_FREELIST_NELEM(idx);
                    435:                snprintf(phpool_names[idx], sizeof(phpool_names[idx]),
                    436:                    "phpool-%d", nelem);
                    437:                sz = sizeof(struct pool_item_header);
                    438:                if (nelem) {
                    439:                        sz = offsetof(struct pool_item_header,
                    440:                            ph_bitmap[howmany(nelem, BITMAP_SIZE)]);
                    441:                }
                    442:                pool_init(&phpool[idx], sz, 0, 0, 0,
                    443:                    phpool_names[idx], &pool_allocator_meta, IPL_VM);
1.117     yamt      444:        }
1.191     para      445: #ifdef POOL_SUBPAGE
                    446:        pool_init(&psppool, POOL_SUBPAGE, POOL_SUBPAGE, 0,
                    447:            PR_RECURSIVE, "psppool", &pool_allocator_meta, IPL_VM);
                    448: #endif
                    449:
                    450:        size = sizeof(pcg_t) +
                    451:            (PCG_NOBJECTS_NORMAL - 1) * sizeof(pcgpair_t);
                    452:        pool_init(&pcg_normal_pool, size, coherency_unit, 0, 0,
                    453:            "pcgnormal", &pool_allocator_meta, IPL_VM);
                    454:
                    455:        size = sizeof(pcg_t) +
                    456:            (PCG_NOBJECTS_LARGE - 1) * sizeof(pcgpair_t);
                    457:        pool_init(&pcg_large_pool, size, coherency_unit, 0, 0,
                    458:            "pcglarge", &pool_allocator_meta, IPL_VM);
1.134     ad        459:
1.156     ad        460:        pool_init(&cache_pool, sizeof(struct pool_cache), coherency_unit,
1.191     para      461:            0, 0, "pcache", &pool_allocator_meta, IPL_NONE);
1.134     ad        462:
1.156     ad        463:        pool_init(&cache_cpu_pool, sizeof(pool_cache_cpu_t), coherency_unit,
1.191     para      464:            0, 0, "pcachecpu", &pool_allocator_meta, IPL_NONE);
1.94      simonb    465: }
                    466:
                    467: /*
1.3       pk        468:  * Initialize the given pool resource structure.
                    469:  *
                    470:  * We export this routine to allow other kernel parts to declare
1.195     rmind     471:  * static pools that must be initialized before kmem(9) is available.
1.3       pk        472:  */
                    473: void
1.42      thorpej   474: pool_init(struct pool *pp, size_t size, u_int align, u_int ioff, int flags,
1.129     ad        475:     const char *wchan, struct pool_allocator *palloc, int ipl)
1.3       pk        476: {
1.116     simonb    477:        struct pool *pp1;
1.204     maxv      478:        size_t trysize, phsize, prsize;
1.134     ad        479:        int off, slack;
1.3       pk        480:
1.116     simonb    481: #ifdef DEBUG
1.198     christos  482:        if (__predict_true(!cold))
                    483:                mutex_enter(&pool_head_lock);
1.116     simonb    484:        /*
                    485:         * Check that the pool hasn't already been initialised and
                    486:         * added to the list of all pools.
                    487:         */
1.145     ad        488:        TAILQ_FOREACH(pp1, &pool_head, pr_poollist) {
1.116     simonb    489:                if (pp == pp1)
1.213     christos  490:                        panic("%s: [%s] already initialised", __func__,
1.116     simonb    491:                            wchan);
                    492:        }
1.198     christos  493:        if (__predict_true(!cold))
                    494:                mutex_exit(&pool_head_lock);
1.116     simonb    495: #endif
                    496:
1.66      thorpej   497:        if (palloc == NULL)
                    498:                palloc = &pool_allocator_kmem;
1.112     bjh21     499: #ifdef POOL_SUBPAGE
                    500:        if (size > palloc->pa_pagesz) {
                    501:                if (palloc == &pool_allocator_kmem)
                    502:                        palloc = &pool_allocator_kmem_fullpage;
                    503:                else if (palloc == &pool_allocator_nointr)
                    504:                        palloc = &pool_allocator_nointr_fullpage;
                    505:        }
1.66      thorpej   506: #endif /* POOL_SUBPAGE */
1.180     mlelstv   507:        if (!cold)
                    508:                mutex_enter(&pool_allocator_lock);
1.178     elad      509:        if (palloc->pa_refcnt++ == 0) {
1.112     bjh21     510:                if (palloc->pa_pagesz == 0)
1.66      thorpej   511:                        palloc->pa_pagesz = PAGE_SIZE;
                    512:
                    513:                TAILQ_INIT(&palloc->pa_list);
                    514:
1.134     ad        515:                mutex_init(&palloc->pa_lock, MUTEX_DEFAULT, IPL_VM);
1.66      thorpej   516:                palloc->pa_pagemask = ~(palloc->pa_pagesz - 1);
                    517:                palloc->pa_pageshift = ffs(palloc->pa_pagesz) - 1;
1.4       thorpej   518:        }
1.180     mlelstv   519:        if (!cold)
                    520:                mutex_exit(&pool_allocator_lock);
1.3       pk        521:
                    522:        if (align == 0)
                    523:                align = ALIGN(1);
1.14      thorpej   524:
1.204     maxv      525:        prsize = size;
                    526:        if ((flags & PR_NOTOUCH) == 0 && prsize < sizeof(struct pool_item))
                    527:                prsize = sizeof(struct pool_item);
1.3       pk        528:
1.204     maxv      529:        prsize = roundup(prsize, align);
1.207     riastrad  530:        KASSERTMSG((prsize <= palloc->pa_pagesz),
1.213     christos  531:            "%s: [%s] pool item size (%zu) larger than page size (%u)",
                    532:            __func__, wchan, prsize, palloc->pa_pagesz);
1.35      pk        533:
1.3       pk        534:        /*
                    535:         * Initialize the pool structure.
                    536:         */
1.88      chs       537:        LIST_INIT(&pp->pr_emptypages);
                    538:        LIST_INIT(&pp->pr_fullpages);
                    539:        LIST_INIT(&pp->pr_partpages);
1.134     ad        540:        pp->pr_cache = NULL;
1.3       pk        541:        pp->pr_curpage = NULL;
                    542:        pp->pr_npages = 0;
                    543:        pp->pr_minitems = 0;
                    544:        pp->pr_minpages = 0;
                    545:        pp->pr_maxpages = UINT_MAX;
1.20      thorpej   546:        pp->pr_roflags = flags;
                    547:        pp->pr_flags = 0;
1.204     maxv      548:        pp->pr_size = prsize;
1.3       pk        549:        pp->pr_align = align;
                    550:        pp->pr_wchan = wchan;
1.66      thorpej   551:        pp->pr_alloc = palloc;
1.20      thorpej   552:        pp->pr_nitems = 0;
                    553:        pp->pr_nout = 0;
                    554:        pp->pr_hardlimit = UINT_MAX;
                    555:        pp->pr_hardlimit_warning = NULL;
1.31      thorpej   556:        pp->pr_hardlimit_ratecap.tv_sec = 0;
                    557:        pp->pr_hardlimit_ratecap.tv_usec = 0;
                    558:        pp->pr_hardlimit_warning_last.tv_sec = 0;
                    559:        pp->pr_hardlimit_warning_last.tv_usec = 0;
1.68      thorpej   560:        pp->pr_drain_hook = NULL;
                    561:        pp->pr_drain_hook_arg = NULL;
1.125     ad        562:        pp->pr_freecheck = NULL;
1.204     maxv      563:        pool_redzone_init(pp, size);
1.3       pk        564:
                    565:        /*
                    566:         * Decide whether to put the page header off page to avoid
1.92      enami     567:         * wasting too large a part of the page or too big item.
                    568:         * Off-page page headers go on a hash table, so we can match
                    569:         * a returned item with its header based on the page address.
                    570:         * We use 1/16 of the page size and about 8 times of the item
                    571:         * size as the threshold (XXX: tune)
                    572:         *
                    573:         * However, we'll put the header into the page if we can put
                    574:         * it without wasting any items.
                    575:         *
                    576:         * Silently enforce `0 <= ioff < align'.
1.3       pk        577:         */
1.92      enami     578:        pp->pr_itemoffset = ioff %= align;
                    579:        /* See the comment below about reserved bytes. */
                    580:        trysize = palloc->pa_pagesz - ((align - ioff) % align);
                    581:        phsize = ALIGN(sizeof(struct pool_item_header));
1.201     para      582:        if (pp->pr_roflags & PR_PHINPAGE ||
                    583:            ((pp->pr_roflags & (PR_NOTOUCH | PR_NOALIGN)) == 0 &&
1.97      yamt      584:            (pp->pr_size < MIN(palloc->pa_pagesz / 16, phsize << 3) ||
1.201     para      585:            trysize / pp->pr_size == (trysize - phsize) / pp->pr_size))) {
1.3       pk        586:                /* Use the end of the page for the page header */
1.20      thorpej   587:                pp->pr_roflags |= PR_PHINPAGE;
1.92      enami     588:                pp->pr_phoffset = off = palloc->pa_pagesz - phsize;
1.2       pk        589:        } else {
1.3       pk        590:                /* The page header will be taken from our page header pool */
                    591:                pp->pr_phoffset = 0;
1.66      thorpej   592:                off = palloc->pa_pagesz;
1.88      chs       593:                SPLAY_INIT(&pp->pr_phtree);
1.2       pk        594:        }
1.1       pk        595:
1.3       pk        596:        /*
                    597:         * Alignment is to take place at `ioff' within the item. This means
                    598:         * we must reserve up to `align - 1' bytes on the page to allow
                    599:         * appropriate positioning of each item.
                    600:         */
                    601:        pp->pr_itemsperpage = (off - ((align - ioff) % align)) / pp->pr_size;
1.43      thorpej   602:        KASSERT(pp->pr_itemsperpage != 0);
1.97      yamt      603:        if ((pp->pr_roflags & PR_NOTOUCH)) {
                    604:                int idx;
                    605:
                    606:                for (idx = 0; pp->pr_itemsperpage > PHPOOL_FREELIST_NELEM(idx);
                    607:                    idx++) {
                    608:                        /* nothing */
                    609:                }
                    610:                if (idx >= PHPOOL_MAX) {
                    611:                        /*
                    612:                         * if you see this panic, consider to tweak
                    613:                         * PHPOOL_MAX and PHPOOL_FREELIST_NELEM.
                    614:                         */
1.213     christos  615:                        panic("%s: [%s] too large itemsperpage(%d) for "
                    616:                            "PR_NOTOUCH", __func__,
1.97      yamt      617:                            pp->pr_wchan, pp->pr_itemsperpage);
                    618:                }
                    619:                pp->pr_phpool = &phpool[idx];
                    620:        } else if ((pp->pr_roflags & PR_PHINPAGE) == 0) {
                    621:                pp->pr_phpool = &phpool[0];
                    622:        }
                    623: #if defined(DIAGNOSTIC)
                    624:        else {
                    625:                pp->pr_phpool = NULL;
                    626:        }
                    627: #endif
1.3       pk        628:
                    629:        /*
                    630:         * Use the slack between the chunks and the page header
                    631:         * for "cache coloring".
                    632:         */
                    633:        slack = off - pp->pr_itemsperpage * pp->pr_size;
                    634:        pp->pr_maxcolor = (slack / align) * align;
                    635:        pp->pr_curcolor = 0;
                    636:
                    637:        pp->pr_nget = 0;
                    638:        pp->pr_nfail = 0;
                    639:        pp->pr_nput = 0;
                    640:        pp->pr_npagealloc = 0;
                    641:        pp->pr_npagefree = 0;
1.1       pk        642:        pp->pr_hiwat = 0;
1.8       thorpej   643:        pp->pr_nidle = 0;
1.134     ad        644:        pp->pr_refcnt = 0;
1.3       pk        645:
1.157     ad        646:        mutex_init(&pp->pr_lock, MUTEX_DEFAULT, ipl);
1.134     ad        647:        cv_init(&pp->pr_cv, wchan);
                    648:        pp->pr_ipl = ipl;
1.1       pk        649:
1.145     ad        650:        /* Insert into the list of all pools. */
1.181     mlelstv   651:        if (!cold)
1.134     ad        652:                mutex_enter(&pool_head_lock);
1.145     ad        653:        TAILQ_FOREACH(pp1, &pool_head, pr_poollist) {
                    654:                if (strcmp(pp1->pr_wchan, pp->pr_wchan) > 0)
                    655:                        break;
                    656:        }
                    657:        if (pp1 == NULL)
                    658:                TAILQ_INSERT_TAIL(&pool_head, pp, pr_poollist);
                    659:        else
                    660:                TAILQ_INSERT_BEFORE(pp1, pp, pr_poollist);
1.181     mlelstv   661:        if (!cold)
1.134     ad        662:                mutex_exit(&pool_head_lock);
                    663:
1.167     skrll     664:        /* Insert this into the list of pools using this allocator. */
1.181     mlelstv   665:        if (!cold)
1.134     ad        666:                mutex_enter(&palloc->pa_lock);
1.145     ad        667:        TAILQ_INSERT_TAIL(&palloc->pa_list, pp, pr_alloc_list);
1.181     mlelstv   668:        if (!cold)
1.134     ad        669:                mutex_exit(&palloc->pa_lock);
1.1       pk        670: }
                    671:
                    672: /*
                    673:  * De-commision a pool resource.
                    674:  */
                    675: void
1.42      thorpej   676: pool_destroy(struct pool *pp)
1.1       pk        677: {
1.101     thorpej   678:        struct pool_pagelist pq;
1.3       pk        679:        struct pool_item_header *ph;
1.43      thorpej   680:
1.101     thorpej   681:        /* Remove from global pool list */
1.134     ad        682:        mutex_enter(&pool_head_lock);
                    683:        while (pp->pr_refcnt != 0)
                    684:                cv_wait(&pool_busy, &pool_head_lock);
1.145     ad        685:        TAILQ_REMOVE(&pool_head, pp, pr_poollist);
1.101     thorpej   686:        if (drainpp == pp)
                    687:                drainpp = NULL;
1.134     ad        688:        mutex_exit(&pool_head_lock);
1.101     thorpej   689:
                    690:        /* Remove this pool from its allocator's list of pools. */
1.134     ad        691:        mutex_enter(&pp->pr_alloc->pa_lock);
1.66      thorpej   692:        TAILQ_REMOVE(&pp->pr_alloc->pa_list, pp, pr_alloc_list);
1.134     ad        693:        mutex_exit(&pp->pr_alloc->pa_lock);
1.66      thorpej   694:
1.178     elad      695:        mutex_enter(&pool_allocator_lock);
                    696:        if (--pp->pr_alloc->pa_refcnt == 0)
                    697:                mutex_destroy(&pp->pr_alloc->pa_lock);
                    698:        mutex_exit(&pool_allocator_lock);
                    699:
1.134     ad        700:        mutex_enter(&pp->pr_lock);
1.101     thorpej   701:
1.134     ad        702:        KASSERT(pp->pr_cache == NULL);
1.207     riastrad  703:        KASSERTMSG((pp->pr_nout == 0),
1.213     christos  704:            "%s: pool busy: still out: %u", __func__, pp->pr_nout);
1.101     thorpej   705:        KASSERT(LIST_EMPTY(&pp->pr_fullpages));
                    706:        KASSERT(LIST_EMPTY(&pp->pr_partpages));
                    707:
1.3       pk        708:        /* Remove all pages */
1.101     thorpej   709:        LIST_INIT(&pq);
1.88      chs       710:        while ((ph = LIST_FIRST(&pp->pr_emptypages)) != NULL)
1.101     thorpej   711:                pr_rmpage(pp, ph, &pq);
                    712:
1.134     ad        713:        mutex_exit(&pp->pr_lock);
1.3       pk        714:
1.101     thorpej   715:        pr_pagelist_free(pp, &pq);
1.134     ad        716:        cv_destroy(&pp->pr_cv);
                    717:        mutex_destroy(&pp->pr_lock);
1.1       pk        718: }
                    719:
1.68      thorpej   720: void
                    721: pool_set_drain_hook(struct pool *pp, void (*fn)(void *, int), void *arg)
                    722: {
                    723:
                    724:        /* XXX no locking -- must be used just after pool_init() */
1.207     riastrad  725:        KASSERTMSG((pp->pr_drain_hook == NULL),
1.213     christos  726:            "%s: [%s] already set", __func__, pp->pr_wchan);
1.68      thorpej   727:        pp->pr_drain_hook = fn;
                    728:        pp->pr_drain_hook_arg = arg;
                    729: }
                    730:
1.88      chs       731: static struct pool_item_header *
1.128     christos  732: pool_alloc_item_header(struct pool *pp, void *storage, int flags)
1.55      thorpej   733: {
                    734:        struct pool_item_header *ph;
                    735:
                    736:        if ((pp->pr_roflags & PR_PHINPAGE) != 0)
1.213     christos  737:                ph = (void *)((char *)storage + pp->pr_phoffset);
1.134     ad        738:        else
1.97      yamt      739:                ph = pool_get(pp->pr_phpool, flags);
1.55      thorpej   740:
                    741:        return (ph);
                    742: }
1.1       pk        743:
                    744: /*
1.134     ad        745:  * Grab an item from the pool.
1.1       pk        746:  */
1.3       pk        747: void *
1.56      sommerfe  748: pool_get(struct pool *pp, int flags)
1.1       pk        749: {
                    750:        struct pool_item *pi;
1.3       pk        751:        struct pool_item_header *ph;
1.55      thorpej   752:        void *v;
1.1       pk        753:
1.207     riastrad  754:        KASSERTMSG((pp->pr_itemsperpage != 0),
1.213     christos  755:            "%s: [%s] pr_itemsperpage is zero, "
                    756:            "pool not initialized?", __func__, pp->pr_wchan);
1.207     riastrad  757:        KASSERTMSG((!(cpu_intr_p() || cpu_softintr_p())
                    758:                || pp->pr_ipl != IPL_NONE || cold || panicstr != NULL),
1.213     christos  759:            "%s: [%s] is IPL_NONE, but called from interrupt context",
                    760:            __func__, pp->pr_wchan);
1.155     ad        761:        if (flags & PR_WAITOK) {
1.154     yamt      762:                ASSERT_SLEEPABLE();
1.155     ad        763:        }
1.1       pk        764:
1.134     ad        765:        mutex_enter(&pp->pr_lock);
1.20      thorpej   766:  startover:
                    767:        /*
                    768:         * Check to see if we've reached the hard limit.  If we have,
                    769:         * and we can wait, then wait until an item has been returned to
                    770:         * the pool.
                    771:         */
1.207     riastrad  772:        KASSERTMSG((pp->pr_nout <= pp->pr_hardlimit),
1.213     christos  773:            "%s: %s: crossed hard limit", __func__, pp->pr_wchan);
1.34      thorpej   774:        if (__predict_false(pp->pr_nout == pp->pr_hardlimit)) {
1.68      thorpej   775:                if (pp->pr_drain_hook != NULL) {
                    776:                        /*
                    777:                         * Since the drain hook is going to free things
                    778:                         * back to the pool, unlock, call the hook, re-lock,
                    779:                         * and check the hardlimit condition again.
                    780:                         */
1.134     ad        781:                        mutex_exit(&pp->pr_lock);
1.68      thorpej   782:                        (*pp->pr_drain_hook)(pp->pr_drain_hook_arg, flags);
1.134     ad        783:                        mutex_enter(&pp->pr_lock);
1.68      thorpej   784:                        if (pp->pr_nout < pp->pr_hardlimit)
                    785:                                goto startover;
                    786:                }
                    787:
1.29      sommerfe  788:                if ((flags & PR_WAITOK) && !(flags & PR_LIMITFAIL)) {
1.20      thorpej   789:                        /*
                    790:                         * XXX: A warning isn't logged in this case.  Should
                    791:                         * it be?
                    792:                         */
                    793:                        pp->pr_flags |= PR_WANTED;
1.212     christos  794:                        do {
                    795:                                cv_wait(&pp->pr_cv, &pp->pr_lock);
                    796:                        } while (pp->pr_flags & PR_WANTED);
1.20      thorpej   797:                        goto startover;
                    798:                }
1.31      thorpej   799:
                    800:                /*
                    801:                 * Log a message that the hard limit has been hit.
                    802:                 */
                    803:                if (pp->pr_hardlimit_warning != NULL &&
                    804:                    ratecheck(&pp->pr_hardlimit_warning_last,
                    805:                              &pp->pr_hardlimit_ratecap))
                    806:                        log(LOG_ERR, "%s\n", pp->pr_hardlimit_warning);
1.21      thorpej   807:
                    808:                pp->pr_nfail++;
                    809:
1.134     ad        810:                mutex_exit(&pp->pr_lock);
1.211     riastrad  811:                KASSERT((flags & (PR_WAITOK|PR_NOWAIT)) == PR_NOWAIT);
1.20      thorpej   812:                return (NULL);
                    813:        }
                    814:
1.3       pk        815:        /*
                    816:         * The convention we use is that if `curpage' is not NULL, then
                    817:         * it points at a non-empty bucket. In particular, `curpage'
                    818:         * never points at a page header which has PR_PHINPAGE set and
                    819:         * has no items in its bucket.
                    820:         */
1.20      thorpej   821:        if ((ph = pp->pr_curpage) == NULL) {
1.113     yamt      822:                int error;
                    823:
1.207     riastrad  824:                KASSERTMSG((pp->pr_nitems == 0),
1.213     christos  825:                    "%s: [%s] curpage NULL, inconsistent nitems %u",
                    826:                    __func__, pp->pr_wchan, pp->pr_nitems);
1.20      thorpej   827:
1.21      thorpej   828:                /*
                    829:                 * Call the back-end page allocator for more memory.
                    830:                 * Release the pool lock, as the back-end page allocator
                    831:                 * may block.
                    832:                 */
1.113     yamt      833:                error = pool_grow(pp, flags);
                    834:                if (error != 0) {
1.21      thorpej   835:                        /*
1.210     mlelstv   836:                         * pool_grow aborts when another thread
                    837:                         * is allocating a new page. Retry if it
                    838:                         * waited for it.
                    839:                         */
                    840:                        if (error == ERESTART)
                    841:                                goto startover;
                    842:
                    843:                        /*
1.55      thorpej   844:                         * We were unable to allocate a page or item
                    845:                         * header, but we released the lock during
                    846:                         * allocation, so perhaps items were freed
                    847:                         * back to the pool.  Check for this case.
1.21      thorpej   848:                         */
                    849:                        if (pp->pr_curpage != NULL)
                    850:                                goto startover;
1.15      pk        851:
1.117     yamt      852:                        pp->pr_nfail++;
1.134     ad        853:                        mutex_exit(&pp->pr_lock);
1.211     riastrad  854:                        KASSERT((flags & (PR_WAITOK|PR_NOWAIT)) == PR_NOWAIT);
1.117     yamt      855:                        return (NULL);
1.1       pk        856:                }
1.3       pk        857:
1.20      thorpej   858:                /* Start the allocation process over. */
                    859:                goto startover;
1.3       pk        860:        }
1.97      yamt      861:        if (pp->pr_roflags & PR_NOTOUCH) {
1.207     riastrad  862:                KASSERTMSG((ph->ph_nmissing < pp->pr_itemsperpage),
1.213     christos  863:                    "%s: %s: page empty", __func__, pp->pr_wchan);
1.97      yamt      864:                v = pr_item_notouch_get(pp, ph);
                    865:        } else {
1.102     chs       866:                v = pi = LIST_FIRST(&ph->ph_itemlist);
1.97      yamt      867:                if (__predict_false(v == NULL)) {
1.134     ad        868:                        mutex_exit(&pp->pr_lock);
1.213     christos  869:                        panic("%s: [%s] page empty", __func__, pp->pr_wchan);
1.97      yamt      870:                }
1.207     riastrad  871:                KASSERTMSG((pp->pr_nitems > 0),
1.213     christos  872:                    "%s: [%s] nitems %u inconsistent on itemlist",
                    873:                    __func__, pp->pr_wchan, pp->pr_nitems);
1.207     riastrad  874:                KASSERTMSG((pi->pi_magic == PI_MAGIC),
1.213     christos  875:                    "%s: [%s] free list modified: "
                    876:                    "magic=%x; page %p; item addr %p", __func__,
1.207     riastrad  877:                    pp->pr_wchan, pi->pi_magic, ph->ph_page, pi);
1.3       pk        878:
1.97      yamt      879:                /*
                    880:                 * Remove from item list.
                    881:                 */
1.102     chs       882:                LIST_REMOVE(pi, pi_list);
1.97      yamt      883:        }
1.20      thorpej   884:        pp->pr_nitems--;
                    885:        pp->pr_nout++;
1.6       thorpej   886:        if (ph->ph_nmissing == 0) {
1.207     riastrad  887:                KASSERT(pp->pr_nidle > 0);
1.6       thorpej   888:                pp->pr_nidle--;
1.88      chs       889:
                    890:                /*
                    891:                 * This page was previously empty.  Move it to the list of
                    892:                 * partially-full pages.  This page is already curpage.
                    893:                 */
                    894:                LIST_REMOVE(ph, ph_pagelist);
                    895:                LIST_INSERT_HEAD(&pp->pr_partpages, ph, ph_pagelist);
1.6       thorpej   896:        }
1.3       pk        897:        ph->ph_nmissing++;
1.97      yamt      898:        if (ph->ph_nmissing == pp->pr_itemsperpage) {
1.207     riastrad  899:                KASSERTMSG(((pp->pr_roflags & PR_NOTOUCH) ||
                    900:                        LIST_EMPTY(&ph->ph_itemlist)),
1.213     christos  901:                    "%s: [%s] nmissing (%u) inconsistent", __func__,
                    902:                        pp->pr_wchan, ph->ph_nmissing);
1.3       pk        903:                /*
1.88      chs       904:                 * This page is now full.  Move it to the full list
                    905:                 * and select a new current page.
1.3       pk        906:                 */
1.88      chs       907:                LIST_REMOVE(ph, ph_pagelist);
                    908:                LIST_INSERT_HEAD(&pp->pr_fullpages, ph, ph_pagelist);
                    909:                pool_update_curpage(pp);
1.1       pk        910:        }
1.3       pk        911:
                    912:        pp->pr_nget++;
1.20      thorpej   913:
                    914:        /*
                    915:         * If we have a low water mark and we are now below that low
                    916:         * water mark, add more items to the pool.
                    917:         */
1.53      thorpej   918:        if (POOL_NEEDS_CATCHUP(pp) && pool_catchup(pp) != 0) {
1.20      thorpej   919:                /*
                    920:                 * XXX: Should we log a warning?  Should we set up a timeout
                    921:                 * to try again in a second or so?  The latter could break
                    922:                 * a caller's assumptions about interrupt protection, etc.
                    923:                 */
                    924:        }
                    925:
1.134     ad        926:        mutex_exit(&pp->pr_lock);
1.125     ad        927:        KASSERT((((vaddr_t)v + pp->pr_itemoffset) & (pp->pr_align - 1)) == 0);
                    928:        FREECHECK_OUT(&pp->pr_freecheck, v);
1.204     maxv      929:        pool_redzone_fill(pp, v);
1.1       pk        930:        return (v);
                    931: }
                    932:
                    933: /*
1.43      thorpej   934:  * Internal version of pool_put().  Pool is already locked/entered.
1.1       pk        935:  */
1.43      thorpej   936: static void
1.101     thorpej   937: pool_do_put(struct pool *pp, void *v, struct pool_pagelist *pq)
1.1       pk        938: {
                    939:        struct pool_item *pi = v;
1.3       pk        940:        struct pool_item_header *ph;
                    941:
1.134     ad        942:        KASSERT(mutex_owned(&pp->pr_lock));
1.204     maxv      943:        pool_redzone_check(pp, v);
1.125     ad        944:        FREECHECK_IN(&pp->pr_freecheck, v);
1.134     ad        945:        LOCKDEBUG_MEM_CHECK(v, pp->pr_size);
1.61      chs       946:
1.207     riastrad  947:        KASSERTMSG((pp->pr_nout > 0),
1.213     christos  948:            "%s: [%s] putting with none out", __func__, pp->pr_wchan);
1.3       pk        949:
1.121     yamt      950:        if (__predict_false((ph = pr_find_pagehead(pp, v)) == NULL)) {
1.213     christos  951:                panic("%s: [%s] page header missing", __func__,  pp->pr_wchan);
1.3       pk        952:        }
1.28      thorpej   953:
1.3       pk        954:        /*
                    955:         * Return to item list.
                    956:         */
1.97      yamt      957:        if (pp->pr_roflags & PR_NOTOUCH) {
                    958:                pr_item_notouch_put(pp, ph, v);
                    959:        } else {
1.2       pk        960: #ifdef DIAGNOSTIC
1.97      yamt      961:                pi->pi_magic = PI_MAGIC;
1.3       pk        962: #endif
1.32      chs       963: #ifdef DEBUG
1.97      yamt      964:                {
                    965:                        int i, *ip = v;
1.32      chs       966:
1.97      yamt      967:                        for (i = 0; i < pp->pr_size / sizeof(int); i++) {
                    968:                                *ip++ = PI_MAGIC;
                    969:                        }
1.32      chs       970:                }
                    971: #endif
                    972:
1.102     chs       973:                LIST_INSERT_HEAD(&ph->ph_itemlist, pi, pi_list);
1.97      yamt      974:        }
1.79      thorpej   975:        KDASSERT(ph->ph_nmissing != 0);
1.3       pk        976:        ph->ph_nmissing--;
                    977:        pp->pr_nput++;
1.20      thorpej   978:        pp->pr_nitems++;
                    979:        pp->pr_nout--;
1.3       pk        980:
                    981:        /* Cancel "pool empty" condition if it exists */
                    982:        if (pp->pr_curpage == NULL)
                    983:                pp->pr_curpage = ph;
                    984:
                    985:        if (pp->pr_flags & PR_WANTED) {
                    986:                pp->pr_flags &= ~PR_WANTED;
1.134     ad        987:                cv_broadcast(&pp->pr_cv);
1.3       pk        988:        }
                    989:
                    990:        /*
1.88      chs       991:         * If this page is now empty, do one of two things:
1.21      thorpej   992:         *
1.88      chs       993:         *      (1) If we have more pages than the page high water mark,
1.96      thorpej   994:         *          free the page back to the system.  ONLY CONSIDER
1.90      thorpej   995:         *          FREEING BACK A PAGE IF WE HAVE MORE THAN OUR MINIMUM PAGE
                    996:         *          CLAIM.
1.21      thorpej   997:         *
1.88      chs       998:         *      (2) Otherwise, move the page to the empty page list.
                    999:         *
                   1000:         * Either way, select a new current page (so we use a partially-full
                   1001:         * page if one is available).
1.3       pk       1002:         */
                   1003:        if (ph->ph_nmissing == 0) {
1.6       thorpej  1004:                pp->pr_nidle++;
1.90      thorpej  1005:                if (pp->pr_npages > pp->pr_minpages &&
1.152     yamt     1006:                    pp->pr_npages > pp->pr_maxpages) {
1.101     thorpej  1007:                        pr_rmpage(pp, ph, pq);
1.3       pk       1008:                } else {
1.88      chs      1009:                        LIST_REMOVE(ph, ph_pagelist);
                   1010:                        LIST_INSERT_HEAD(&pp->pr_emptypages, ph, ph_pagelist);
1.3       pk       1011:
1.21      thorpej  1012:                        /*
                   1013:                         * Update the timestamp on the page.  A page must
                   1014:                         * be idle for some period of time before it can
                   1015:                         * be reclaimed by the pagedaemon.  This minimizes
                   1016:                         * ping-pong'ing for memory.
1.151     yamt     1017:                         *
                   1018:                         * note for 64-bit time_t: truncating to 32-bit is not
                   1019:                         * a problem for our usage.
1.21      thorpej  1020:                         */
1.151     yamt     1021:                        ph->ph_time = time_uptime;
1.1       pk       1022:                }
1.88      chs      1023:                pool_update_curpage(pp);
1.1       pk       1024:        }
1.88      chs      1025:
1.21      thorpej  1026:        /*
1.88      chs      1027:         * If the page was previously completely full, move it to the
                   1028:         * partially-full list and make it the current page.  The next
                   1029:         * allocation will get the item from this page, instead of
                   1030:         * further fragmenting the pool.
1.21      thorpej  1031:         */
                   1032:        else if (ph->ph_nmissing == (pp->pr_itemsperpage - 1)) {
1.88      chs      1033:                LIST_REMOVE(ph, ph_pagelist);
                   1034:                LIST_INSERT_HEAD(&pp->pr_partpages, ph, ph_pagelist);
1.21      thorpej  1035:                pp->pr_curpage = ph;
                   1036:        }
1.43      thorpej  1037: }
                   1038:
1.56      sommerfe 1039: void
                   1040: pool_put(struct pool *pp, void *v)
                   1041: {
1.101     thorpej  1042:        struct pool_pagelist pq;
                   1043:
                   1044:        LIST_INIT(&pq);
1.56      sommerfe 1045:
1.134     ad       1046:        mutex_enter(&pp->pr_lock);
1.101     thorpej  1047:        pool_do_put(pp, v, &pq);
1.134     ad       1048:        mutex_exit(&pp->pr_lock);
1.56      sommerfe 1049:
1.102     chs      1050:        pr_pagelist_free(pp, &pq);
1.56      sommerfe 1051: }
1.57      sommerfe 1052:
1.74      thorpej  1053: /*
1.113     yamt     1054:  * pool_grow: grow a pool by a page.
                   1055:  *
                   1056:  * => called with pool locked.
                   1057:  * => unlock and relock the pool.
                   1058:  * => return with pool locked.
                   1059:  */
                   1060:
                   1061: static int
                   1062: pool_grow(struct pool *pp, int flags)
                   1063: {
                   1064:        struct pool_item_header *ph = NULL;
                   1065:        char *cp;
1.209     riastrad 1066:        int error;
                   1067:
                   1068:        /*
                   1069:         * If there's a pool_grow in progress, wait for it to complete
                   1070:         * and try again from the top.
                   1071:         */
                   1072:        if (pp->pr_flags & PR_GROWING) {
                   1073:                if (flags & PR_WAITOK) {
                   1074:                        do {
                   1075:                                cv_wait(&pp->pr_cv, &pp->pr_lock);
                   1076:                        } while (pp->pr_flags & PR_GROWING);
                   1077:                        return ERESTART;
                   1078:                } else {
                   1079:                        return EWOULDBLOCK;
                   1080:                }
                   1081:        }
                   1082:        pp->pr_flags |= PR_GROWING;
1.113     yamt     1083:
1.134     ad       1084:        mutex_exit(&pp->pr_lock);
1.113     yamt     1085:        cp = pool_allocator_alloc(pp, flags);
                   1086:        if (__predict_true(cp != NULL)) {
                   1087:                ph = pool_alloc_item_header(pp, cp, flags);
                   1088:        }
                   1089:        if (__predict_false(cp == NULL || ph == NULL)) {
                   1090:                if (cp != NULL) {
                   1091:                        pool_allocator_free(pp, cp);
                   1092:                }
1.134     ad       1093:                mutex_enter(&pp->pr_lock);
1.209     riastrad 1094:                error = ENOMEM;
                   1095:                goto out;
1.113     yamt     1096:        }
                   1097:
1.134     ad       1098:        mutex_enter(&pp->pr_lock);
1.113     yamt     1099:        pool_prime_page(pp, cp, ph);
                   1100:        pp->pr_npagealloc++;
1.209     riastrad 1101:        error = 0;
                   1102:
                   1103: out:
                   1104:        /*
                   1105:         * If anyone was waiting for pool_grow, notify them that we
                   1106:         * may have just done it.
                   1107:         */
                   1108:        KASSERT(pp->pr_flags & PR_GROWING);
                   1109:        pp->pr_flags &= ~PR_GROWING;
                   1110:        cv_broadcast(&pp->pr_cv);
                   1111:
                   1112:        return error;
1.113     yamt     1113: }
                   1114:
                   1115: /*
1.74      thorpej  1116:  * Add N items to the pool.
                   1117:  */
                   1118: int
                   1119: pool_prime(struct pool *pp, int n)
                   1120: {
1.75      simonb   1121:        int newpages;
1.113     yamt     1122:        int error = 0;
1.74      thorpej  1123:
1.134     ad       1124:        mutex_enter(&pp->pr_lock);
1.74      thorpej  1125:
                   1126:        newpages = roundup(n, pp->pr_itemsperpage) / pp->pr_itemsperpage;
                   1127:
                   1128:        while (newpages-- > 0) {
1.113     yamt     1129:                error = pool_grow(pp, PR_NOWAIT);
                   1130:                if (error) {
1.214   ! christos 1131:                        if (error == ERESTART)
        !          1132:                                continue;
1.74      thorpej  1133:                        break;
                   1134:                }
                   1135:                pp->pr_minpages++;
                   1136:        }
                   1137:
                   1138:        if (pp->pr_minpages >= pp->pr_maxpages)
                   1139:                pp->pr_maxpages = pp->pr_minpages + 1;  /* XXX */
                   1140:
1.134     ad       1141:        mutex_exit(&pp->pr_lock);
1.113     yamt     1142:        return error;
1.74      thorpej  1143: }
1.55      thorpej  1144:
                   1145: /*
1.3       pk       1146:  * Add a page worth of items to the pool.
1.21      thorpej  1147:  *
                   1148:  * Note, we must be called with the pool descriptor LOCKED.
1.3       pk       1149:  */
1.55      thorpej  1150: static void
1.128     christos 1151: pool_prime_page(struct pool *pp, void *storage, struct pool_item_header *ph)
1.3       pk       1152: {
                   1153:        struct pool_item *pi;
1.128     christos 1154:        void *cp = storage;
1.125     ad       1155:        const unsigned int align = pp->pr_align;
                   1156:        const unsigned int ioff = pp->pr_itemoffset;
1.55      thorpej  1157:        int n;
1.36      pk       1158:
1.134     ad       1159:        KASSERT(mutex_owned(&pp->pr_lock));
1.207     riastrad 1160:        KASSERTMSG(((pp->pr_roflags & PR_NOALIGN) ||
                   1161:                (((uintptr_t)cp & (pp->pr_alloc->pa_pagesz - 1)) == 0)),
1.213     christos 1162:            "%s: [%s] unaligned page: %p", __func__, pp->pr_wchan, cp);
1.3       pk       1163:
                   1164:        /*
                   1165:         * Insert page header.
                   1166:         */
1.88      chs      1167:        LIST_INSERT_HEAD(&pp->pr_emptypages, ph, ph_pagelist);
1.102     chs      1168:        LIST_INIT(&ph->ph_itemlist);
1.3       pk       1169:        ph->ph_page = storage;
                   1170:        ph->ph_nmissing = 0;
1.151     yamt     1171:        ph->ph_time = time_uptime;
1.88      chs      1172:        if ((pp->pr_roflags & PR_PHINPAGE) == 0)
                   1173:                SPLAY_INSERT(phtree, &pp->pr_phtree, ph);
1.3       pk       1174:
1.6       thorpej  1175:        pp->pr_nidle++;
                   1176:
1.3       pk       1177:        /*
                   1178:         * Color this page.
                   1179:         */
1.141     yamt     1180:        ph->ph_off = pp->pr_curcolor;
                   1181:        cp = (char *)cp + ph->ph_off;
1.3       pk       1182:        if ((pp->pr_curcolor += align) > pp->pr_maxcolor)
                   1183:                pp->pr_curcolor = 0;
                   1184:
                   1185:        /*
                   1186:         * Adjust storage to apply aligment to `pr_itemoffset' in each item.
                   1187:         */
                   1188:        if (ioff != 0)
1.128     christos 1189:                cp = (char *)cp + align - ioff;
1.3       pk       1190:
1.125     ad       1191:        KASSERT((((vaddr_t)cp + ioff) & (align - 1)) == 0);
                   1192:
1.3       pk       1193:        /*
                   1194:         * Insert remaining chunks on the bucket list.
                   1195:         */
                   1196:        n = pp->pr_itemsperpage;
1.20      thorpej  1197:        pp->pr_nitems += n;
1.3       pk       1198:
1.97      yamt     1199:        if (pp->pr_roflags & PR_NOTOUCH) {
1.141     yamt     1200:                pr_item_notouch_init(pp, ph);
1.97      yamt     1201:        } else {
                   1202:                while (n--) {
                   1203:                        pi = (struct pool_item *)cp;
1.78      thorpej  1204:
1.97      yamt     1205:                        KASSERT(((((vaddr_t)pi) + ioff) & (align - 1)) == 0);
1.3       pk       1206:
1.97      yamt     1207:                        /* Insert on page list */
1.102     chs      1208:                        LIST_INSERT_HEAD(&ph->ph_itemlist, pi, pi_list);
1.3       pk       1209: #ifdef DIAGNOSTIC
1.97      yamt     1210:                        pi->pi_magic = PI_MAGIC;
1.3       pk       1211: #endif
1.128     christos 1212:                        cp = (char *)cp + pp->pr_size;
1.125     ad       1213:
                   1214:                        KASSERT((((vaddr_t)cp + ioff) & (align - 1)) == 0);
1.97      yamt     1215:                }
1.3       pk       1216:        }
                   1217:
                   1218:        /*
                   1219:         * If the pool was depleted, point at the new page.
                   1220:         */
                   1221:        if (pp->pr_curpage == NULL)
                   1222:                pp->pr_curpage = ph;
                   1223:
                   1224:        if (++pp->pr_npages > pp->pr_hiwat)
                   1225:                pp->pr_hiwat = pp->pr_npages;
                   1226: }
                   1227:
1.20      thorpej  1228: /*
1.52      thorpej  1229:  * Used by pool_get() when nitems drops below the low water mark.  This
1.88      chs      1230:  * is used to catch up pr_nitems with the low water mark.
1.20      thorpej  1231:  *
1.21      thorpej  1232:  * Note 1, we never wait for memory here, we let the caller decide what to do.
1.20      thorpej  1233:  *
1.73      thorpej  1234:  * Note 2, we must be called with the pool already locked, and we return
1.20      thorpej  1235:  * with it locked.
                   1236:  */
                   1237: static int
1.42      thorpej  1238: pool_catchup(struct pool *pp)
1.20      thorpej  1239: {
                   1240:        int error = 0;
                   1241:
1.54      thorpej  1242:        while (POOL_NEEDS_CATCHUP(pp)) {
1.113     yamt     1243:                error = pool_grow(pp, PR_NOWAIT);
                   1244:                if (error) {
1.214   ! christos 1245:                        if (error == ERESTART)
        !          1246:                                continue;
1.20      thorpej  1247:                        break;
                   1248:                }
                   1249:        }
1.113     yamt     1250:        return error;
1.20      thorpej  1251: }
                   1252:
1.88      chs      1253: static void
                   1254: pool_update_curpage(struct pool *pp)
                   1255: {
                   1256:
                   1257:        pp->pr_curpage = LIST_FIRST(&pp->pr_partpages);
                   1258:        if (pp->pr_curpage == NULL) {
                   1259:                pp->pr_curpage = LIST_FIRST(&pp->pr_emptypages);
                   1260:        }
1.168     yamt     1261:        KASSERT((pp->pr_curpage == NULL && pp->pr_nitems == 0) ||
                   1262:            (pp->pr_curpage != NULL && pp->pr_nitems > 0));
1.88      chs      1263: }
                   1264:
1.3       pk       1265: void
1.42      thorpej  1266: pool_setlowat(struct pool *pp, int n)
1.3       pk       1267: {
1.15      pk       1268:
1.134     ad       1269:        mutex_enter(&pp->pr_lock);
1.21      thorpej  1270:
1.3       pk       1271:        pp->pr_minitems = n;
1.15      pk       1272:        pp->pr_minpages = (n == 0)
                   1273:                ? 0
1.18      thorpej  1274:                : roundup(n, pp->pr_itemsperpage) / pp->pr_itemsperpage;
1.20      thorpej  1275:
                   1276:        /* Make sure we're caught up with the newly-set low water mark. */
1.75      simonb   1277:        if (POOL_NEEDS_CATCHUP(pp) && pool_catchup(pp) != 0) {
1.20      thorpej  1278:                /*
                   1279:                 * XXX: Should we log a warning?  Should we set up a timeout
                   1280:                 * to try again in a second or so?  The latter could break
                   1281:                 * a caller's assumptions about interrupt protection, etc.
                   1282:                 */
                   1283:        }
1.21      thorpej  1284:
1.134     ad       1285:        mutex_exit(&pp->pr_lock);
1.3       pk       1286: }
                   1287:
                   1288: void
1.42      thorpej  1289: pool_sethiwat(struct pool *pp, int n)
1.3       pk       1290: {
1.15      pk       1291:
1.134     ad       1292:        mutex_enter(&pp->pr_lock);
1.21      thorpej  1293:
1.15      pk       1294:        pp->pr_maxpages = (n == 0)
                   1295:                ? 0
1.18      thorpej  1296:                : roundup(n, pp->pr_itemsperpage) / pp->pr_itemsperpage;
1.21      thorpej  1297:
1.134     ad       1298:        mutex_exit(&pp->pr_lock);
1.3       pk       1299: }
                   1300:
1.20      thorpej  1301: void
1.42      thorpej  1302: pool_sethardlimit(struct pool *pp, int n, const char *warnmess, int ratecap)
1.20      thorpej  1303: {
                   1304:
1.134     ad       1305:        mutex_enter(&pp->pr_lock);
1.20      thorpej  1306:
                   1307:        pp->pr_hardlimit = n;
                   1308:        pp->pr_hardlimit_warning = warnmess;
1.31      thorpej  1309:        pp->pr_hardlimit_ratecap.tv_sec = ratecap;
                   1310:        pp->pr_hardlimit_warning_last.tv_sec = 0;
                   1311:        pp->pr_hardlimit_warning_last.tv_usec = 0;
1.20      thorpej  1312:
                   1313:        /*
1.21      thorpej  1314:         * In-line version of pool_sethiwat(), because we don't want to
                   1315:         * release the lock.
1.20      thorpej  1316:         */
                   1317:        pp->pr_maxpages = (n == 0)
                   1318:                ? 0
                   1319:                : roundup(n, pp->pr_itemsperpage) / pp->pr_itemsperpage;
1.21      thorpej  1320:
1.134     ad       1321:        mutex_exit(&pp->pr_lock);
1.20      thorpej  1322: }
1.3       pk       1323:
                   1324: /*
                   1325:  * Release all complete pages that have not been used recently.
1.184     rmind    1326:  *
1.197     jym      1327:  * Must not be called from interrupt context.
1.3       pk       1328:  */
1.66      thorpej  1329: int
1.56      sommerfe 1330: pool_reclaim(struct pool *pp)
1.3       pk       1331: {
                   1332:        struct pool_item_header *ph, *phnext;
1.61      chs      1333:        struct pool_pagelist pq;
1.151     yamt     1334:        uint32_t curtime;
1.134     ad       1335:        bool klock;
                   1336:        int rv;
1.3       pk       1337:
1.197     jym      1338:        KASSERT(!cpu_intr_p() && !cpu_softintr_p());
1.184     rmind    1339:
1.68      thorpej  1340:        if (pp->pr_drain_hook != NULL) {
                   1341:                /*
                   1342:                 * The drain hook must be called with the pool unlocked.
                   1343:                 */
                   1344:                (*pp->pr_drain_hook)(pp->pr_drain_hook_arg, PR_NOWAIT);
                   1345:        }
                   1346:
1.134     ad       1347:        /*
1.157     ad       1348:         * XXXSMP Because we do not want to cause non-MPSAFE code
                   1349:         * to block.
1.134     ad       1350:         */
                   1351:        if (pp->pr_ipl == IPL_SOFTNET || pp->pr_ipl == IPL_SOFTCLOCK ||
                   1352:            pp->pr_ipl == IPL_SOFTSERIAL) {
                   1353:                KERNEL_LOCK(1, NULL);
                   1354:                klock = true;
                   1355:        } else
                   1356:                klock = false;
                   1357:
                   1358:        /* Reclaim items from the pool's cache (if any). */
                   1359:        if (pp->pr_cache != NULL)
                   1360:                pool_cache_invalidate(pp->pr_cache);
                   1361:
                   1362:        if (mutex_tryenter(&pp->pr_lock) == 0) {
                   1363:                if (klock) {
                   1364:                        KERNEL_UNLOCK_ONE(NULL);
                   1365:                }
1.66      thorpej  1366:                return (0);
1.134     ad       1367:        }
1.68      thorpej  1368:
1.88      chs      1369:        LIST_INIT(&pq);
1.43      thorpej  1370:
1.151     yamt     1371:        curtime = time_uptime;
1.21      thorpej  1372:
1.88      chs      1373:        for (ph = LIST_FIRST(&pp->pr_emptypages); ph != NULL; ph = phnext) {
                   1374:                phnext = LIST_NEXT(ph, ph_pagelist);
1.3       pk       1375:
                   1376:                /* Check our minimum page claim */
                   1377:                if (pp->pr_npages <= pp->pr_minpages)
                   1378:                        break;
                   1379:
1.88      chs      1380:                KASSERT(ph->ph_nmissing == 0);
1.191     para     1381:                if (curtime - ph->ph_time < pool_inactive_time)
1.88      chs      1382:                        continue;
1.21      thorpej  1383:
1.88      chs      1384:                /*
                   1385:                 * If freeing this page would put us below
                   1386:                 * the low water mark, stop now.
                   1387:                 */
                   1388:                if ((pp->pr_nitems - pp->pr_itemsperpage) <
                   1389:                    pp->pr_minitems)
                   1390:                        break;
1.21      thorpej  1391:
1.88      chs      1392:                pr_rmpage(pp, ph, &pq);
1.3       pk       1393:        }
                   1394:
1.134     ad       1395:        mutex_exit(&pp->pr_lock);
                   1396:
                   1397:        if (LIST_EMPTY(&pq))
                   1398:                rv = 0;
                   1399:        else {
                   1400:                pr_pagelist_free(pp, &pq);
                   1401:                rv = 1;
                   1402:        }
                   1403:
                   1404:        if (klock) {
                   1405:                KERNEL_UNLOCK_ONE(NULL);
                   1406:        }
1.66      thorpej  1407:
1.134     ad       1408:        return (rv);
1.3       pk       1409: }
                   1410:
                   1411: /*
1.197     jym      1412:  * Drain pools, one at a time. The drained pool is returned within ppp.
1.131     ad       1413:  *
1.134     ad       1414:  * Note, must never be called from interrupt context.
1.3       pk       1415:  */
1.197     jym      1416: bool
                   1417: pool_drain(struct pool **ppp)
1.3       pk       1418: {
1.197     jym      1419:        bool reclaimed;
1.3       pk       1420:        struct pool *pp;
1.134     ad       1421:
1.145     ad       1422:        KASSERT(!TAILQ_EMPTY(&pool_head));
1.3       pk       1423:
1.61      chs      1424:        pp = NULL;
1.134     ad       1425:
                   1426:        /* Find next pool to drain, and add a reference. */
                   1427:        mutex_enter(&pool_head_lock);
                   1428:        do {
                   1429:                if (drainpp == NULL) {
1.145     ad       1430:                        drainpp = TAILQ_FIRST(&pool_head);
1.134     ad       1431:                }
                   1432:                if (drainpp != NULL) {
                   1433:                        pp = drainpp;
1.145     ad       1434:                        drainpp = TAILQ_NEXT(pp, pr_poollist);
1.134     ad       1435:                }
                   1436:                /*
                   1437:                 * Skip completely idle pools.  We depend on at least
                   1438:                 * one pool in the system being active.
                   1439:                 */
                   1440:        } while (pp == NULL || pp->pr_npages == 0);
                   1441:        pp->pr_refcnt++;
                   1442:        mutex_exit(&pool_head_lock);
                   1443:
                   1444:        /* Drain the cache (if any) and pool.. */
1.186     pooka    1445:        reclaimed = pool_reclaim(pp);
1.134     ad       1446:
                   1447:        /* Finally, unlock the pool. */
                   1448:        mutex_enter(&pool_head_lock);
                   1449:        pp->pr_refcnt--;
                   1450:        cv_broadcast(&pool_busy);
                   1451:        mutex_exit(&pool_head_lock);
1.186     pooka    1452:
1.197     jym      1453:        if (ppp != NULL)
                   1454:                *ppp = pp;
                   1455:
1.186     pooka    1456:        return reclaimed;
1.3       pk       1457: }
                   1458:
                   1459: /*
                   1460:  * Diagnostic helpers.
                   1461:  */
1.21      thorpej  1462:
1.25      thorpej  1463: void
1.108     yamt     1464: pool_printall(const char *modif, void (*pr)(const char *, ...))
                   1465: {
                   1466:        struct pool *pp;
                   1467:
1.145     ad       1468:        TAILQ_FOREACH(pp, &pool_head, pr_poollist) {
1.108     yamt     1469:                pool_printit(pp, modif, pr);
                   1470:        }
                   1471: }
                   1472:
                   1473: void
1.42      thorpej  1474: pool_printit(struct pool *pp, const char *modif, void (*pr)(const char *, ...))
1.25      thorpej  1475: {
                   1476:
                   1477:        if (pp == NULL) {
                   1478:                (*pr)("Must specify a pool to print.\n");
                   1479:                return;
                   1480:        }
                   1481:
                   1482:        pool_print1(pp, modif, pr);
                   1483: }
                   1484:
1.21      thorpej  1485: static void
1.124     yamt     1486: pool_print_pagelist(struct pool *pp, struct pool_pagelist *pl,
1.97      yamt     1487:     void (*pr)(const char *, ...))
1.88      chs      1488: {
                   1489:        struct pool_item_header *ph;
1.207     riastrad 1490:        struct pool_item *pi __diagused;
1.88      chs      1491:
                   1492:        LIST_FOREACH(ph, pl, ph_pagelist) {
1.151     yamt     1493:                (*pr)("\t\tpage %p, nmissing %d, time %" PRIu32 "\n",
                   1494:                    ph->ph_page, ph->ph_nmissing, ph->ph_time);
1.88      chs      1495: #ifdef DIAGNOSTIC
1.97      yamt     1496:                if (!(pp->pr_roflags & PR_NOTOUCH)) {
1.102     chs      1497:                        LIST_FOREACH(pi, &ph->ph_itemlist, pi_list) {
1.97      yamt     1498:                                if (pi->pi_magic != PI_MAGIC) {
                   1499:                                        (*pr)("\t\t\titem %p, magic 0x%x\n",
                   1500:                                            pi, pi->pi_magic);
                   1501:                                }
1.88      chs      1502:                        }
                   1503:                }
                   1504: #endif
                   1505:        }
                   1506: }
                   1507:
                   1508: static void
1.42      thorpej  1509: pool_print1(struct pool *pp, const char *modif, void (*pr)(const char *, ...))
1.3       pk       1510: {
1.25      thorpej  1511:        struct pool_item_header *ph;
1.134     ad       1512:        pool_cache_t pc;
                   1513:        pcg_t *pcg;
                   1514:        pool_cache_cpu_t *cc;
                   1515:        uint64_t cpuhit, cpumiss;
1.44      thorpej  1516:        int i, print_log = 0, print_pagelist = 0, print_cache = 0;
1.25      thorpej  1517:        char c;
                   1518:
                   1519:        while ((c = *modif++) != '\0') {
                   1520:                if (c == 'l')
                   1521:                        print_log = 1;
                   1522:                if (c == 'p')
                   1523:                        print_pagelist = 1;
1.44      thorpej  1524:                if (c == 'c')
                   1525:                        print_cache = 1;
1.25      thorpej  1526:        }
                   1527:
1.134     ad       1528:        if ((pc = pp->pr_cache) != NULL) {
                   1529:                (*pr)("POOL CACHE");
                   1530:        } else {
                   1531:                (*pr)("POOL");
                   1532:        }
                   1533:
                   1534:        (*pr)(" %s: size %u, align %u, ioff %u, roflags 0x%08x\n",
1.25      thorpej  1535:            pp->pr_wchan, pp->pr_size, pp->pr_align, pp->pr_itemoffset,
                   1536:            pp->pr_roflags);
1.66      thorpej  1537:        (*pr)("\talloc %p\n", pp->pr_alloc);
1.25      thorpej  1538:        (*pr)("\tminitems %u, minpages %u, maxpages %u, npages %u\n",
                   1539:            pp->pr_minitems, pp->pr_minpages, pp->pr_maxpages, pp->pr_npages);
                   1540:        (*pr)("\titemsperpage %u, nitems %u, nout %u, hardlimit %u\n",
                   1541:            pp->pr_itemsperpage, pp->pr_nitems, pp->pr_nout, pp->pr_hardlimit);
                   1542:
1.134     ad       1543:        (*pr)("\tnget %lu, nfail %lu, nput %lu\n",
1.25      thorpej  1544:            pp->pr_nget, pp->pr_nfail, pp->pr_nput);
                   1545:        (*pr)("\tnpagealloc %lu, npagefree %lu, hiwat %u, nidle %lu\n",
                   1546:            pp->pr_npagealloc, pp->pr_npagefree, pp->pr_hiwat, pp->pr_nidle);
                   1547:
                   1548:        if (print_pagelist == 0)
                   1549:                goto skip_pagelist;
                   1550:
1.88      chs      1551:        if ((ph = LIST_FIRST(&pp->pr_emptypages)) != NULL)
                   1552:                (*pr)("\n\tempty page list:\n");
1.97      yamt     1553:        pool_print_pagelist(pp, &pp->pr_emptypages, pr);
1.88      chs      1554:        if ((ph = LIST_FIRST(&pp->pr_fullpages)) != NULL)
                   1555:                (*pr)("\n\tfull page list:\n");
1.97      yamt     1556:        pool_print_pagelist(pp, &pp->pr_fullpages, pr);
1.88      chs      1557:        if ((ph = LIST_FIRST(&pp->pr_partpages)) != NULL)
                   1558:                (*pr)("\n\tpartial-page list:\n");
1.97      yamt     1559:        pool_print_pagelist(pp, &pp->pr_partpages, pr);
1.88      chs      1560:
1.25      thorpej  1561:        if (pp->pr_curpage == NULL)
                   1562:                (*pr)("\tno current page\n");
                   1563:        else
                   1564:                (*pr)("\tcurpage %p\n", pp->pr_curpage->ph_page);
                   1565:
                   1566:  skip_pagelist:
                   1567:        if (print_log == 0)
                   1568:                goto skip_log;
                   1569:
                   1570:        (*pr)("\n");
1.3       pk       1571:
1.25      thorpej  1572:  skip_log:
1.44      thorpej  1573:
1.102     chs      1574: #define PR_GROUPLIST(pcg)                                              \
                   1575:        (*pr)("\t\tgroup %p: avail %d\n", pcg, pcg->pcg_avail);         \
1.142     ad       1576:        for (i = 0; i < pcg->pcg_size; i++) {                           \
1.102     chs      1577:                if (pcg->pcg_objects[i].pcgo_pa !=                      \
                   1578:                    POOL_PADDR_INVALID) {                               \
                   1579:                        (*pr)("\t\t\t%p, 0x%llx\n",                     \
                   1580:                            pcg->pcg_objects[i].pcgo_va,                \
                   1581:                            (unsigned long long)                        \
                   1582:                            pcg->pcg_objects[i].pcgo_pa);               \
                   1583:                } else {                                                \
                   1584:                        (*pr)("\t\t\t%p\n",                             \
                   1585:                            pcg->pcg_objects[i].pcgo_va);               \
                   1586:                }                                                       \
                   1587:        }
                   1588:
1.134     ad       1589:        if (pc != NULL) {
                   1590:                cpuhit = 0;
                   1591:                cpumiss = 0;
1.183     ad       1592:                for (i = 0; i < __arraycount(pc->pc_cpus); i++) {
1.134     ad       1593:                        if ((cc = pc->pc_cpus[i]) == NULL)
                   1594:                                continue;
                   1595:                        cpuhit += cc->cc_hits;
                   1596:                        cpumiss += cc->cc_misses;
                   1597:                }
                   1598:                (*pr)("\tcpu layer hits %llu misses %llu\n", cpuhit, cpumiss);
                   1599:                (*pr)("\tcache layer hits %llu misses %llu\n",
                   1600:                    pc->pc_hits, pc->pc_misses);
                   1601:                (*pr)("\tcache layer entry uncontended %llu contended %llu\n",
                   1602:                    pc->pc_hits + pc->pc_misses - pc->pc_contended,
                   1603:                    pc->pc_contended);
                   1604:                (*pr)("\tcache layer empty groups %u full groups %u\n",
                   1605:                    pc->pc_nempty, pc->pc_nfull);
                   1606:                if (print_cache) {
                   1607:                        (*pr)("\tfull cache groups:\n");
                   1608:                        for (pcg = pc->pc_fullgroups; pcg != NULL;
                   1609:                            pcg = pcg->pcg_next) {
                   1610:                                PR_GROUPLIST(pcg);
                   1611:                        }
                   1612:                        (*pr)("\tempty cache groups:\n");
                   1613:                        for (pcg = pc->pc_emptygroups; pcg != NULL;
                   1614:                            pcg = pcg->pcg_next) {
                   1615:                                PR_GROUPLIST(pcg);
                   1616:                        }
1.103     chs      1617:                }
1.44      thorpej  1618:        }
1.102     chs      1619: #undef PR_GROUPLIST
1.88      chs      1620: }
                   1621:
                   1622: static int
                   1623: pool_chk_page(struct pool *pp, const char *label, struct pool_item_header *ph)
                   1624: {
                   1625:        struct pool_item *pi;
1.128     christos 1626:        void *page;
1.88      chs      1627:        int n;
                   1628:
1.121     yamt     1629:        if ((pp->pr_roflags & PR_NOALIGN) == 0) {
1.128     christos 1630:                page = (void *)((uintptr_t)ph & pp->pr_alloc->pa_pagemask);
1.121     yamt     1631:                if (page != ph->ph_page &&
                   1632:                    (pp->pr_roflags & PR_PHINPAGE) != 0) {
                   1633:                        if (label != NULL)
                   1634:                                printf("%s: ", label);
                   1635:                        printf("pool(%p:%s): page inconsistency: page %p;"
                   1636:                               " at page head addr %p (p %p)\n", pp,
                   1637:                                pp->pr_wchan, ph->ph_page,
                   1638:                                ph, page);
                   1639:                        return 1;
                   1640:                }
1.88      chs      1641:        }
1.3       pk       1642:
1.97      yamt     1643:        if ((pp->pr_roflags & PR_NOTOUCH) != 0)
                   1644:                return 0;
                   1645:
1.102     chs      1646:        for (pi = LIST_FIRST(&ph->ph_itemlist), n = 0;
1.88      chs      1647:             pi != NULL;
1.102     chs      1648:             pi = LIST_NEXT(pi,pi_list), n++) {
1.88      chs      1649:
                   1650: #ifdef DIAGNOSTIC
                   1651:                if (pi->pi_magic != PI_MAGIC) {
                   1652:                        if (label != NULL)
                   1653:                                printf("%s: ", label);
                   1654:                        printf("pool(%s): free list modified: magic=%x;"
1.121     yamt     1655:                               " page %p; item ordinal %d; addr %p\n",
1.88      chs      1656:                                pp->pr_wchan, pi->pi_magic, ph->ph_page,
1.121     yamt     1657:                                n, pi);
1.88      chs      1658:                        panic("pool");
                   1659:                }
                   1660: #endif
1.121     yamt     1661:                if ((pp->pr_roflags & PR_NOALIGN) != 0) {
                   1662:                        continue;
                   1663:                }
1.128     christos 1664:                page = (void *)((uintptr_t)pi & pp->pr_alloc->pa_pagemask);
1.88      chs      1665:                if (page == ph->ph_page)
                   1666:                        continue;
                   1667:
                   1668:                if (label != NULL)
                   1669:                        printf("%s: ", label);
                   1670:                printf("pool(%p:%s): page inconsistency: page %p;"
                   1671:                       " item ordinal %d; addr %p (p %p)\n", pp,
                   1672:                        pp->pr_wchan, ph->ph_page,
                   1673:                        n, pi, page);
                   1674:                return 1;
                   1675:        }
                   1676:        return 0;
1.3       pk       1677: }
                   1678:
1.88      chs      1679:
1.3       pk       1680: int
1.42      thorpej  1681: pool_chk(struct pool *pp, const char *label)
1.3       pk       1682: {
                   1683:        struct pool_item_header *ph;
                   1684:        int r = 0;
                   1685:
1.134     ad       1686:        mutex_enter(&pp->pr_lock);
1.88      chs      1687:        LIST_FOREACH(ph, &pp->pr_emptypages, ph_pagelist) {
                   1688:                r = pool_chk_page(pp, label, ph);
                   1689:                if (r) {
                   1690:                        goto out;
                   1691:                }
                   1692:        }
                   1693:        LIST_FOREACH(ph, &pp->pr_fullpages, ph_pagelist) {
                   1694:                r = pool_chk_page(pp, label, ph);
                   1695:                if (r) {
1.3       pk       1696:                        goto out;
                   1697:                }
1.88      chs      1698:        }
                   1699:        LIST_FOREACH(ph, &pp->pr_partpages, ph_pagelist) {
                   1700:                r = pool_chk_page(pp, label, ph);
                   1701:                if (r) {
1.3       pk       1702:                        goto out;
                   1703:                }
                   1704:        }
1.88      chs      1705:
1.3       pk       1706: out:
1.134     ad       1707:        mutex_exit(&pp->pr_lock);
1.3       pk       1708:        return (r);
1.43      thorpej  1709: }
                   1710:
                   1711: /*
                   1712:  * pool_cache_init:
                   1713:  *
                   1714:  *     Initialize a pool cache.
1.134     ad       1715:  */
                   1716: pool_cache_t
                   1717: pool_cache_init(size_t size, u_int align, u_int align_offset, u_int flags,
                   1718:     const char *wchan, struct pool_allocator *palloc, int ipl,
                   1719:     int (*ctor)(void *, void *, int), void (*dtor)(void *, void *), void *arg)
                   1720: {
                   1721:        pool_cache_t pc;
                   1722:
                   1723:        pc = pool_get(&cache_pool, PR_WAITOK);
                   1724:        if (pc == NULL)
                   1725:                return NULL;
                   1726:
                   1727:        pool_cache_bootstrap(pc, size, align, align_offset, flags, wchan,
                   1728:           palloc, ipl, ctor, dtor, arg);
                   1729:
                   1730:        return pc;
                   1731: }
                   1732:
                   1733: /*
                   1734:  * pool_cache_bootstrap:
1.43      thorpej  1735:  *
1.134     ad       1736:  *     Kernel-private version of pool_cache_init().  The caller
                   1737:  *     provides initial storage.
1.43      thorpej  1738:  */
                   1739: void
1.134     ad       1740: pool_cache_bootstrap(pool_cache_t pc, size_t size, u_int align,
                   1741:     u_int align_offset, u_int flags, const char *wchan,
                   1742:     struct pool_allocator *palloc, int ipl,
                   1743:     int (*ctor)(void *, void *, int), void (*dtor)(void *, void *),
1.43      thorpej  1744:     void *arg)
                   1745: {
1.134     ad       1746:        CPU_INFO_ITERATOR cii;
1.145     ad       1747:        pool_cache_t pc1;
1.134     ad       1748:        struct cpu_info *ci;
                   1749:        struct pool *pp;
                   1750:
                   1751:        pp = &pc->pc_pool;
1.208     chs      1752:        if (palloc == NULL && ipl == IPL_NONE) {
                   1753:                if (size > PAGE_SIZE) {
                   1754:                        int bigidx = pool_bigidx(size);
                   1755:
                   1756:                        palloc = &pool_allocator_big[bigidx];
                   1757:                } else
                   1758:                        palloc = &pool_allocator_nointr;
                   1759:        }
1.134     ad       1760:        pool_init(pp, size, align, align_offset, flags, wchan, palloc, ipl);
1.157     ad       1761:        mutex_init(&pc->pc_lock, MUTEX_DEFAULT, ipl);
1.43      thorpej  1762:
1.134     ad       1763:        if (ctor == NULL) {
                   1764:                ctor = (int (*)(void *, void *, int))nullop;
                   1765:        }
                   1766:        if (dtor == NULL) {
                   1767:                dtor = (void (*)(void *, void *))nullop;
                   1768:        }
1.43      thorpej  1769:
1.134     ad       1770:        pc->pc_emptygroups = NULL;
                   1771:        pc->pc_fullgroups = NULL;
                   1772:        pc->pc_partgroups = NULL;
1.43      thorpej  1773:        pc->pc_ctor = ctor;
                   1774:        pc->pc_dtor = dtor;
                   1775:        pc->pc_arg  = arg;
1.134     ad       1776:        pc->pc_hits  = 0;
1.48      thorpej  1777:        pc->pc_misses = 0;
1.134     ad       1778:        pc->pc_nempty = 0;
                   1779:        pc->pc_npart = 0;
                   1780:        pc->pc_nfull = 0;
                   1781:        pc->pc_contended = 0;
                   1782:        pc->pc_refcnt = 0;
1.136     yamt     1783:        pc->pc_freecheck = NULL;
1.134     ad       1784:
1.142     ad       1785:        if ((flags & PR_LARGECACHE) != 0) {
                   1786:                pc->pc_pcgsize = PCG_NOBJECTS_LARGE;
1.163     ad       1787:                pc->pc_pcgpool = &pcg_large_pool;
1.142     ad       1788:        } else {
                   1789:                pc->pc_pcgsize = PCG_NOBJECTS_NORMAL;
1.163     ad       1790:                pc->pc_pcgpool = &pcg_normal_pool;
1.142     ad       1791:        }
                   1792:
1.134     ad       1793:        /* Allocate per-CPU caches. */
                   1794:        memset(pc->pc_cpus, 0, sizeof(pc->pc_cpus));
                   1795:        pc->pc_ncpu = 0;
1.139     ad       1796:        if (ncpu < 2) {
1.137     ad       1797:                /* XXX For sparc: boot CPU is not attached yet. */
                   1798:                pool_cache_cpu_init1(curcpu(), pc);
                   1799:        } else {
                   1800:                for (CPU_INFO_FOREACH(cii, ci)) {
                   1801:                        pool_cache_cpu_init1(ci, pc);
                   1802:                }
1.134     ad       1803:        }
1.145     ad       1804:
                   1805:        /* Add to list of all pools. */
                   1806:        if (__predict_true(!cold))
1.134     ad       1807:                mutex_enter(&pool_head_lock);
1.145     ad       1808:        TAILQ_FOREACH(pc1, &pool_cache_head, pc_cachelist) {
                   1809:                if (strcmp(pc1->pc_pool.pr_wchan, pc->pc_pool.pr_wchan) > 0)
                   1810:                        break;
                   1811:        }
                   1812:        if (pc1 == NULL)
                   1813:                TAILQ_INSERT_TAIL(&pool_cache_head, pc, pc_cachelist);
                   1814:        else
                   1815:                TAILQ_INSERT_BEFORE(pc1, pc, pc_cachelist);
                   1816:        if (__predict_true(!cold))
1.134     ad       1817:                mutex_exit(&pool_head_lock);
1.145     ad       1818:
                   1819:        membar_sync();
                   1820:        pp->pr_cache = pc;
1.43      thorpej  1821: }
                   1822:
                   1823: /*
                   1824:  * pool_cache_destroy:
                   1825:  *
                   1826:  *     Destroy a pool cache.
                   1827:  */
                   1828: void
1.134     ad       1829: pool_cache_destroy(pool_cache_t pc)
1.43      thorpej  1830: {
1.191     para     1831:
                   1832:        pool_cache_bootstrap_destroy(pc);
                   1833:        pool_put(&cache_pool, pc);
                   1834: }
                   1835:
                   1836: /*
                   1837:  * pool_cache_bootstrap_destroy:
                   1838:  *
                   1839:  *     Destroy a pool cache.
                   1840:  */
                   1841: void
                   1842: pool_cache_bootstrap_destroy(pool_cache_t pc)
                   1843: {
1.134     ad       1844:        struct pool *pp = &pc->pc_pool;
1.175     jym      1845:        u_int i;
1.134     ad       1846:
                   1847:        /* Remove it from the global list. */
                   1848:        mutex_enter(&pool_head_lock);
                   1849:        while (pc->pc_refcnt != 0)
                   1850:                cv_wait(&pool_busy, &pool_head_lock);
1.145     ad       1851:        TAILQ_REMOVE(&pool_cache_head, pc, pc_cachelist);
1.134     ad       1852:        mutex_exit(&pool_head_lock);
1.43      thorpej  1853:
                   1854:        /* First, invalidate the entire cache. */
                   1855:        pool_cache_invalidate(pc);
                   1856:
1.134     ad       1857:        /* Disassociate it from the pool. */
                   1858:        mutex_enter(&pp->pr_lock);
                   1859:        pp->pr_cache = NULL;
                   1860:        mutex_exit(&pp->pr_lock);
                   1861:
                   1862:        /* Destroy per-CPU data */
1.183     ad       1863:        for (i = 0; i < __arraycount(pc->pc_cpus); i++)
1.175     jym      1864:                pool_cache_invalidate_cpu(pc, i);
1.134     ad       1865:
                   1866:        /* Finally, destroy it. */
                   1867:        mutex_destroy(&pc->pc_lock);
                   1868:        pool_destroy(pp);
                   1869: }
                   1870:
                   1871: /*
                   1872:  * pool_cache_cpu_init1:
                   1873:  *
                   1874:  *     Called for each pool_cache whenever a new CPU is attached.
                   1875:  */
                   1876: static void
                   1877: pool_cache_cpu_init1(struct cpu_info *ci, pool_cache_t pc)
                   1878: {
                   1879:        pool_cache_cpu_t *cc;
1.137     ad       1880:        int index;
1.134     ad       1881:
1.137     ad       1882:        index = ci->ci_index;
                   1883:
1.183     ad       1884:        KASSERT(index < __arraycount(pc->pc_cpus));
1.134     ad       1885:
1.137     ad       1886:        if ((cc = pc->pc_cpus[index]) != NULL) {
                   1887:                KASSERT(cc->cc_cpuindex == index);
1.134     ad       1888:                return;
                   1889:        }
                   1890:
                   1891:        /*
                   1892:         * The first CPU is 'free'.  This needs to be the case for
                   1893:         * bootstrap - we may not be able to allocate yet.
                   1894:         */
                   1895:        if (pc->pc_ncpu == 0) {
                   1896:                cc = &pc->pc_cpu0;
                   1897:                pc->pc_ncpu = 1;
                   1898:        } else {
                   1899:                mutex_enter(&pc->pc_lock);
                   1900:                pc->pc_ncpu++;
                   1901:                mutex_exit(&pc->pc_lock);
                   1902:                cc = pool_get(&cache_cpu_pool, PR_WAITOK);
                   1903:        }
                   1904:
                   1905:        cc->cc_ipl = pc->pc_pool.pr_ipl;
                   1906:        cc->cc_iplcookie = makeiplcookie(cc->cc_ipl);
                   1907:        cc->cc_cache = pc;
1.137     ad       1908:        cc->cc_cpuindex = index;
1.134     ad       1909:        cc->cc_hits = 0;
                   1910:        cc->cc_misses = 0;
1.169     yamt     1911:        cc->cc_current = __UNCONST(&pcg_dummy);
                   1912:        cc->cc_previous = __UNCONST(&pcg_dummy);
1.134     ad       1913:
1.137     ad       1914:        pc->pc_cpus[index] = cc;
1.43      thorpej  1915: }
                   1916:
1.134     ad       1917: /*
                   1918:  * pool_cache_cpu_init:
                   1919:  *
                   1920:  *     Called whenever a new CPU is attached.
                   1921:  */
                   1922: void
                   1923: pool_cache_cpu_init(struct cpu_info *ci)
1.43      thorpej  1924: {
1.134     ad       1925:        pool_cache_t pc;
                   1926:
                   1927:        mutex_enter(&pool_head_lock);
1.145     ad       1928:        TAILQ_FOREACH(pc, &pool_cache_head, pc_cachelist) {
1.134     ad       1929:                pc->pc_refcnt++;
                   1930:                mutex_exit(&pool_head_lock);
1.43      thorpej  1931:
1.134     ad       1932:                pool_cache_cpu_init1(ci, pc);
1.43      thorpej  1933:
1.134     ad       1934:                mutex_enter(&pool_head_lock);
                   1935:                pc->pc_refcnt--;
                   1936:                cv_broadcast(&pool_busy);
                   1937:        }
                   1938:        mutex_exit(&pool_head_lock);
1.43      thorpej  1939: }
                   1940:
1.134     ad       1941: /*
                   1942:  * pool_cache_reclaim:
                   1943:  *
                   1944:  *     Reclaim memory from a pool cache.
                   1945:  */
                   1946: bool
                   1947: pool_cache_reclaim(pool_cache_t pc)
1.43      thorpej  1948: {
                   1949:
1.134     ad       1950:        return pool_reclaim(&pc->pc_pool);
                   1951: }
1.43      thorpej  1952:
1.136     yamt     1953: static void
                   1954: pool_cache_destruct_object1(pool_cache_t pc, void *object)
                   1955: {
                   1956:
                   1957:        (*pc->pc_dtor)(pc->pc_arg, object);
                   1958:        pool_put(&pc->pc_pool, object);
                   1959: }
                   1960:
1.134     ad       1961: /*
                   1962:  * pool_cache_destruct_object:
                   1963:  *
                   1964:  *     Force destruction of an object and its release back into
                   1965:  *     the pool.
                   1966:  */
                   1967: void
                   1968: pool_cache_destruct_object(pool_cache_t pc, void *object)
                   1969: {
                   1970:
1.136     yamt     1971:        FREECHECK_IN(&pc->pc_freecheck, object);
                   1972:
                   1973:        pool_cache_destruct_object1(pc, object);
1.43      thorpej  1974: }
                   1975:
1.134     ad       1976: /*
                   1977:  * pool_cache_invalidate_groups:
                   1978:  *
                   1979:  *     Invalidate a chain of groups and destruct all objects.
                   1980:  */
1.102     chs      1981: static void
1.134     ad       1982: pool_cache_invalidate_groups(pool_cache_t pc, pcg_t *pcg)
1.102     chs      1983: {
1.134     ad       1984:        void *object;
                   1985:        pcg_t *next;
                   1986:        int i;
                   1987:
                   1988:        for (; pcg != NULL; pcg = next) {
                   1989:                next = pcg->pcg_next;
                   1990:
                   1991:                for (i = 0; i < pcg->pcg_avail; i++) {
                   1992:                        object = pcg->pcg_objects[i].pcgo_va;
1.136     yamt     1993:                        pool_cache_destruct_object1(pc, object);
1.134     ad       1994:                }
1.102     chs      1995:
1.142     ad       1996:                if (pcg->pcg_size == PCG_NOBJECTS_LARGE) {
                   1997:                        pool_put(&pcg_large_pool, pcg);
                   1998:                } else {
                   1999:                        KASSERT(pcg->pcg_size == PCG_NOBJECTS_NORMAL);
                   2000:                        pool_put(&pcg_normal_pool, pcg);
                   2001:                }
1.102     chs      2002:        }
                   2003: }
                   2004:
1.43      thorpej  2005: /*
1.134     ad       2006:  * pool_cache_invalidate:
1.43      thorpej  2007:  *
1.134     ad       2008:  *     Invalidate a pool cache (destruct and release all of the
                   2009:  *     cached objects).  Does not reclaim objects from the pool.
1.176     thorpej  2010:  *
                   2011:  *     Note: For pool caches that provide constructed objects, there
                   2012:  *     is an assumption that another level of synchronization is occurring
                   2013:  *     between the input to the constructor and the cache invalidation.
1.196     jym      2014:  *
                   2015:  *     Invalidation is a costly process and should not be called from
                   2016:  *     interrupt context.
1.43      thorpej  2017:  */
1.134     ad       2018: void
                   2019: pool_cache_invalidate(pool_cache_t pc)
                   2020: {
1.196     jym      2021:        uint64_t where;
1.134     ad       2022:        pcg_t *full, *empty, *part;
1.196     jym      2023:
                   2024:        KASSERT(!cpu_intr_p() && !cpu_softintr_p());
1.176     thorpej  2025:
1.177     jym      2026:        if (ncpu < 2 || !mp_online) {
1.176     thorpej  2027:                /*
                   2028:                 * We might be called early enough in the boot process
                   2029:                 * for the CPU data structures to not be fully initialized.
1.196     jym      2030:                 * In this case, transfer the content of the local CPU's
                   2031:                 * cache back into global cache as only this CPU is currently
                   2032:                 * running.
1.176     thorpej  2033:                 */
1.196     jym      2034:                pool_cache_transfer(pc);
1.176     thorpej  2035:        } else {
                   2036:                /*
1.196     jym      2037:                 * Signal all CPUs that they must transfer their local
                   2038:                 * cache back to the global pool then wait for the xcall to
                   2039:                 * complete.
1.176     thorpej  2040:                 */
1.196     jym      2041:                where = xc_broadcast(0, (xcfunc_t)pool_cache_transfer,
                   2042:                    pc, NULL);
1.176     thorpej  2043:                xc_wait(where);
                   2044:        }
1.196     jym      2045:
                   2046:        /* Empty pool caches, then invalidate objects */
1.134     ad       2047:        mutex_enter(&pc->pc_lock);
                   2048:        full = pc->pc_fullgroups;
                   2049:        empty = pc->pc_emptygroups;
                   2050:        part = pc->pc_partgroups;
                   2051:        pc->pc_fullgroups = NULL;
                   2052:        pc->pc_emptygroups = NULL;
                   2053:        pc->pc_partgroups = NULL;
                   2054:        pc->pc_nfull = 0;
                   2055:        pc->pc_nempty = 0;
                   2056:        pc->pc_npart = 0;
                   2057:        mutex_exit(&pc->pc_lock);
                   2058:
                   2059:        pool_cache_invalidate_groups(pc, full);
                   2060:        pool_cache_invalidate_groups(pc, empty);
                   2061:        pool_cache_invalidate_groups(pc, part);
                   2062: }
                   2063:
1.175     jym      2064: /*
                   2065:  * pool_cache_invalidate_cpu:
                   2066:  *
                   2067:  *     Invalidate all CPU-bound cached objects in pool cache, the CPU being
                   2068:  *     identified by its associated index.
                   2069:  *     It is caller's responsibility to ensure that no operation is
                   2070:  *     taking place on this pool cache while doing this invalidation.
                   2071:  *     WARNING: as no inter-CPU locking is enforced, trying to invalidate
                   2072:  *     pool cached objects from a CPU different from the one currently running
                   2073:  *     may result in an undefined behaviour.
                   2074:  */
                   2075: static void
                   2076: pool_cache_invalidate_cpu(pool_cache_t pc, u_int index)
                   2077: {
                   2078:        pool_cache_cpu_t *cc;
                   2079:        pcg_t *pcg;
                   2080:
                   2081:        if ((cc = pc->pc_cpus[index]) == NULL)
                   2082:                return;
                   2083:
                   2084:        if ((pcg = cc->cc_current) != &pcg_dummy) {
                   2085:                pcg->pcg_next = NULL;
                   2086:                pool_cache_invalidate_groups(pc, pcg);
                   2087:        }
                   2088:        if ((pcg = cc->cc_previous) != &pcg_dummy) {
                   2089:                pcg->pcg_next = NULL;
                   2090:                pool_cache_invalidate_groups(pc, pcg);
                   2091:        }
                   2092:        if (cc != &pc->pc_cpu0)
                   2093:                pool_put(&cache_cpu_pool, cc);
                   2094:
                   2095: }
                   2096:
1.134     ad       2097: void
                   2098: pool_cache_set_drain_hook(pool_cache_t pc, void (*fn)(void *, int), void *arg)
                   2099: {
                   2100:
                   2101:        pool_set_drain_hook(&pc->pc_pool, fn, arg);
                   2102: }
                   2103:
                   2104: void
                   2105: pool_cache_setlowat(pool_cache_t pc, int n)
                   2106: {
                   2107:
                   2108:        pool_setlowat(&pc->pc_pool, n);
                   2109: }
                   2110:
                   2111: void
                   2112: pool_cache_sethiwat(pool_cache_t pc, int n)
                   2113: {
                   2114:
                   2115:        pool_sethiwat(&pc->pc_pool, n);
                   2116: }
                   2117:
                   2118: void
                   2119: pool_cache_sethardlimit(pool_cache_t pc, int n, const char *warnmess, int ratecap)
                   2120: {
                   2121:
                   2122:        pool_sethardlimit(&pc->pc_pool, n, warnmess, ratecap);
                   2123: }
                   2124:
1.162     ad       2125: static bool __noinline
                   2126: pool_cache_get_slow(pool_cache_cpu_t *cc, int s, void **objectp,
1.134     ad       2127:                    paddr_t *pap, int flags)
1.43      thorpej  2128: {
1.134     ad       2129:        pcg_t *pcg, *cur;
                   2130:        uint64_t ncsw;
                   2131:        pool_cache_t pc;
1.43      thorpej  2132:        void *object;
1.58      thorpej  2133:
1.168     yamt     2134:        KASSERT(cc->cc_current->pcg_avail == 0);
                   2135:        KASSERT(cc->cc_previous->pcg_avail == 0);
                   2136:
1.134     ad       2137:        pc = cc->cc_cache;
                   2138:        cc->cc_misses++;
1.43      thorpej  2139:
1.134     ad       2140:        /*
                   2141:         * Nothing was available locally.  Try and grab a group
                   2142:         * from the cache.
                   2143:         */
1.162     ad       2144:        if (__predict_false(!mutex_tryenter(&pc->pc_lock))) {
1.134     ad       2145:                ncsw = curlwp->l_ncsw;
                   2146:                mutex_enter(&pc->pc_lock);
                   2147:                pc->pc_contended++;
1.43      thorpej  2148:
1.134     ad       2149:                /*
                   2150:                 * If we context switched while locking, then
                   2151:                 * our view of the per-CPU data is invalid:
                   2152:                 * retry.
                   2153:                 */
                   2154:                if (curlwp->l_ncsw != ncsw) {
                   2155:                        mutex_exit(&pc->pc_lock);
1.162     ad       2156:                        return true;
1.43      thorpej  2157:                }
1.102     chs      2158:        }
1.43      thorpej  2159:
1.162     ad       2160:        if (__predict_true((pcg = pc->pc_fullgroups) != NULL)) {
1.43      thorpej  2161:                /*
1.134     ad       2162:                 * If there's a full group, release our empty
                   2163:                 * group back to the cache.  Install the full
                   2164:                 * group as cc_current and return.
1.43      thorpej  2165:                 */
1.162     ad       2166:                if (__predict_true((cur = cc->cc_current) != &pcg_dummy)) {
1.134     ad       2167:                        KASSERT(cur->pcg_avail == 0);
                   2168:                        cur->pcg_next = pc->pc_emptygroups;
                   2169:                        pc->pc_emptygroups = cur;
                   2170:                        pc->pc_nempty++;
1.87      thorpej  2171:                }
1.142     ad       2172:                KASSERT(pcg->pcg_avail == pcg->pcg_size);
1.134     ad       2173:                cc->cc_current = pcg;
                   2174:                pc->pc_fullgroups = pcg->pcg_next;
                   2175:                pc->pc_hits++;
                   2176:                pc->pc_nfull--;
                   2177:                mutex_exit(&pc->pc_lock);
1.162     ad       2178:                return true;
1.134     ad       2179:        }
                   2180:
                   2181:        /*
                   2182:         * Nothing available locally or in cache.  Take the slow
                   2183:         * path: fetch a new object from the pool and construct
                   2184:         * it.
                   2185:         */
                   2186:        pc->pc_misses++;
                   2187:        mutex_exit(&pc->pc_lock);
1.162     ad       2188:        splx(s);
1.134     ad       2189:
                   2190:        object = pool_get(&pc->pc_pool, flags);
                   2191:        *objectp = object;
1.211     riastrad 2192:        if (__predict_false(object == NULL)) {
                   2193:                KASSERT((flags & (PR_WAITOK|PR_NOWAIT)) == PR_NOWAIT);
1.162     ad       2194:                return false;
1.211     riastrad 2195:        }
1.125     ad       2196:
1.162     ad       2197:        if (__predict_false((*pc->pc_ctor)(pc->pc_arg, object, flags) != 0)) {
1.134     ad       2198:                pool_put(&pc->pc_pool, object);
                   2199:                *objectp = NULL;
1.162     ad       2200:                return false;
1.43      thorpej  2201:        }
                   2202:
1.134     ad       2203:        KASSERT((((vaddr_t)object + pc->pc_pool.pr_itemoffset) &
                   2204:            (pc->pc_pool.pr_align - 1)) == 0);
1.43      thorpej  2205:
1.134     ad       2206:        if (pap != NULL) {
                   2207: #ifdef POOL_VTOPHYS
                   2208:                *pap = POOL_VTOPHYS(object);
                   2209: #else
                   2210:                *pap = POOL_PADDR_INVALID;
                   2211: #endif
1.102     chs      2212:        }
1.43      thorpej  2213:
1.125     ad       2214:        FREECHECK_OUT(&pc->pc_freecheck, object);
1.204     maxv     2215:        pool_redzone_fill(&pc->pc_pool, object);
1.162     ad       2216:        return false;
1.43      thorpej  2217: }
                   2218:
                   2219: /*
1.134     ad       2220:  * pool_cache_get{,_paddr}:
1.43      thorpej  2221:  *
1.134     ad       2222:  *     Get an object from a pool cache (optionally returning
                   2223:  *     the physical address of the object).
1.43      thorpej  2224:  */
1.134     ad       2225: void *
                   2226: pool_cache_get_paddr(pool_cache_t pc, int flags, paddr_t *pap)
1.43      thorpej  2227: {
1.134     ad       2228:        pool_cache_cpu_t *cc;
                   2229:        pcg_t *pcg;
                   2230:        void *object;
1.60      thorpej  2231:        int s;
1.43      thorpej  2232:
1.184     rmind    2233:        KASSERTMSG((!cpu_intr_p() && !cpu_softintr_p()) ||
1.185     rmind    2234:            (pc->pc_pool.pr_ipl != IPL_NONE || cold || panicstr != NULL),
1.213     christos 2235:            "%s: [%s] is IPL_NONE, but called from interrupt context",
                   2236:            __func__, pc->pc_pool.pr_wchan);
1.184     rmind    2237:
1.155     ad       2238:        if (flags & PR_WAITOK) {
1.154     yamt     2239:                ASSERT_SLEEPABLE();
1.155     ad       2240:        }
1.125     ad       2241:
1.162     ad       2242:        /* Lock out interrupts and disable preemption. */
                   2243:        s = splvm();
1.165     yamt     2244:        while (/* CONSTCOND */ true) {
1.134     ad       2245:                /* Try and allocate an object from the current group. */
1.162     ad       2246:                cc = pc->pc_cpus[curcpu()->ci_index];
                   2247:                KASSERT(cc->cc_cache == pc);
1.134     ad       2248:                pcg = cc->cc_current;
1.162     ad       2249:                if (__predict_true(pcg->pcg_avail > 0)) {
1.134     ad       2250:                        object = pcg->pcg_objects[--pcg->pcg_avail].pcgo_va;
1.162     ad       2251:                        if (__predict_false(pap != NULL))
1.134     ad       2252:                                *pap = pcg->pcg_objects[pcg->pcg_avail].pcgo_pa;
1.148     yamt     2253: #if defined(DIAGNOSTIC)
1.134     ad       2254:                        pcg->pcg_objects[pcg->pcg_avail].pcgo_va = NULL;
1.163     ad       2255:                        KASSERT(pcg->pcg_avail < pcg->pcg_size);
1.134     ad       2256:                        KASSERT(object != NULL);
1.163     ad       2257: #endif
1.134     ad       2258:                        cc->cc_hits++;
1.162     ad       2259:                        splx(s);
1.134     ad       2260:                        FREECHECK_OUT(&pc->pc_freecheck, object);
1.204     maxv     2261:                        pool_redzone_fill(&pc->pc_pool, object);
1.134     ad       2262:                        return object;
1.43      thorpej  2263:                }
                   2264:
                   2265:                /*
1.134     ad       2266:                 * That failed.  If the previous group isn't empty, swap
                   2267:                 * it with the current group and allocate from there.
1.43      thorpej  2268:                 */
1.134     ad       2269:                pcg = cc->cc_previous;
1.162     ad       2270:                if (__predict_true(pcg->pcg_avail > 0)) {
1.134     ad       2271:                        cc->cc_previous = cc->cc_current;
                   2272:                        cc->cc_current = pcg;
                   2273:                        continue;
1.43      thorpej  2274:                }
                   2275:
1.134     ad       2276:                /*
                   2277:                 * Can't allocate from either group: try the slow path.
                   2278:                 * If get_slow() allocated an object for us, or if
1.162     ad       2279:                 * no more objects are available, it will return false.
1.134     ad       2280:                 * Otherwise, we need to retry.
                   2281:                 */
1.165     yamt     2282:                if (!pool_cache_get_slow(cc, s, &object, pap, flags))
                   2283:                        break;
                   2284:        }
1.43      thorpej  2285:
1.211     riastrad 2286:        /*
                   2287:         * We would like to KASSERT(object || (flags & PR_NOWAIT)), but
                   2288:         * pool_cache_get can fail even in the PR_WAITOK case, if the
                   2289:         * constructor fails.
                   2290:         */
1.134     ad       2291:        return object;
1.51      thorpej  2292: }
                   2293:
1.162     ad       2294: static bool __noinline
                   2295: pool_cache_put_slow(pool_cache_cpu_t *cc, int s, void *object)
1.51      thorpej  2296: {
1.200     pooka    2297:        struct lwp *l = curlwp;
1.163     ad       2298:        pcg_t *pcg, *cur;
1.134     ad       2299:        uint64_t ncsw;
                   2300:        pool_cache_t pc;
1.51      thorpej  2301:
1.168     yamt     2302:        KASSERT(cc->cc_current->pcg_avail == cc->cc_current->pcg_size);
                   2303:        KASSERT(cc->cc_previous->pcg_avail == cc->cc_previous->pcg_size);
                   2304:
1.134     ad       2305:        pc = cc->cc_cache;
1.171     ad       2306:        pcg = NULL;
1.134     ad       2307:        cc->cc_misses++;
1.200     pooka    2308:        ncsw = l->l_ncsw;
1.43      thorpej  2309:
1.171     ad       2310:        /*
                   2311:         * If there are no empty groups in the cache then allocate one
                   2312:         * while still unlocked.
                   2313:         */
                   2314:        if (__predict_false(pc->pc_emptygroups == NULL)) {
                   2315:                if (__predict_true(!pool_cache_disable)) {
                   2316:                        pcg = pool_get(pc->pc_pcgpool, PR_NOWAIT);
                   2317:                }
1.200     pooka    2318:                /*
                   2319:                 * If pool_get() blocked, then our view of
                   2320:                 * the per-CPU data is invalid: retry.
                   2321:                 */
                   2322:                if (__predict_false(l->l_ncsw != ncsw)) {
                   2323:                        if (pcg != NULL) {
                   2324:                                pool_put(pc->pc_pcgpool, pcg);
                   2325:                        }
                   2326:                        return true;
                   2327:                }
1.171     ad       2328:                if (__predict_true(pcg != NULL)) {
                   2329:                        pcg->pcg_avail = 0;
                   2330:                        pcg->pcg_size = pc->pc_pcgsize;
                   2331:                }
                   2332:        }
                   2333:
1.162     ad       2334:        /* Lock the cache. */
                   2335:        if (__predict_false(!mutex_tryenter(&pc->pc_lock))) {
1.134     ad       2336:                mutex_enter(&pc->pc_lock);
                   2337:                pc->pc_contended++;
1.162     ad       2338:
1.163     ad       2339:                /*
                   2340:                 * If we context switched while locking, then our view of
                   2341:                 * the per-CPU data is invalid: retry.
                   2342:                 */
1.200     pooka    2343:                if (__predict_false(l->l_ncsw != ncsw)) {
1.163     ad       2344:                        mutex_exit(&pc->pc_lock);
1.171     ad       2345:                        if (pcg != NULL) {
                   2346:                                pool_put(pc->pc_pcgpool, pcg);
                   2347:                        }
1.163     ad       2348:                        return true;
                   2349:                }
1.162     ad       2350:        }
1.102     chs      2351:
1.163     ad       2352:        /* If there are no empty groups in the cache then allocate one. */
1.171     ad       2353:        if (pcg == NULL && pc->pc_emptygroups != NULL) {
                   2354:                pcg = pc->pc_emptygroups;
1.163     ad       2355:                pc->pc_emptygroups = pcg->pcg_next;
                   2356:                pc->pc_nempty--;
1.134     ad       2357:        }
1.130     ad       2358:
1.162     ad       2359:        /*
                   2360:         * If there's a empty group, release our full group back
                   2361:         * to the cache.  Install the empty group to the local CPU
                   2362:         * and return.
                   2363:         */
1.163     ad       2364:        if (pcg != NULL) {
1.134     ad       2365:                KASSERT(pcg->pcg_avail == 0);
1.162     ad       2366:                if (__predict_false(cc->cc_previous == &pcg_dummy)) {
1.146     ad       2367:                        cc->cc_previous = pcg;
                   2368:                } else {
1.162     ad       2369:                        cur = cc->cc_current;
                   2370:                        if (__predict_true(cur != &pcg_dummy)) {
1.163     ad       2371:                                KASSERT(cur->pcg_avail == cur->pcg_size);
1.146     ad       2372:                                cur->pcg_next = pc->pc_fullgroups;
                   2373:                                pc->pc_fullgroups = cur;
                   2374:                                pc->pc_nfull++;
                   2375:                        }
                   2376:                        cc->cc_current = pcg;
                   2377:                }
1.163     ad       2378:                pc->pc_hits++;
1.134     ad       2379:                mutex_exit(&pc->pc_lock);
1.162     ad       2380:                return true;
1.102     chs      2381:        }
1.105     christos 2382:
1.134     ad       2383:        /*
1.162     ad       2384:         * Nothing available locally or in cache, and we didn't
                   2385:         * allocate an empty group.  Take the slow path and destroy
                   2386:         * the object here and now.
1.134     ad       2387:         */
                   2388:        pc->pc_misses++;
                   2389:        mutex_exit(&pc->pc_lock);
1.162     ad       2390:        splx(s);
                   2391:        pool_cache_destruct_object(pc, object);
1.105     christos 2392:
1.162     ad       2393:        return false;
1.134     ad       2394: }
1.102     chs      2395:
1.43      thorpej  2396: /*
1.134     ad       2397:  * pool_cache_put{,_paddr}:
1.43      thorpej  2398:  *
1.134     ad       2399:  *     Put an object back to the pool cache (optionally caching the
                   2400:  *     physical address of the object).
1.43      thorpej  2401:  */
1.101     thorpej  2402: void
1.134     ad       2403: pool_cache_put_paddr(pool_cache_t pc, void *object, paddr_t pa)
1.43      thorpej  2404: {
1.134     ad       2405:        pool_cache_cpu_t *cc;
                   2406:        pcg_t *pcg;
                   2407:        int s;
1.101     thorpej  2408:
1.172     yamt     2409:        KASSERT(object != NULL);
1.204     maxv     2410:        pool_redzone_check(&pc->pc_pool, object);
1.134     ad       2411:        FREECHECK_IN(&pc->pc_freecheck, object);
1.101     thorpej  2412:
1.162     ad       2413:        /* Lock out interrupts and disable preemption. */
                   2414:        s = splvm();
1.165     yamt     2415:        while (/* CONSTCOND */ true) {
1.134     ad       2416:                /* If the current group isn't full, release it there. */
1.162     ad       2417:                cc = pc->pc_cpus[curcpu()->ci_index];
                   2418:                KASSERT(cc->cc_cache == pc);
1.134     ad       2419:                pcg = cc->cc_current;
1.162     ad       2420:                if (__predict_true(pcg->pcg_avail < pcg->pcg_size)) {
1.134     ad       2421:                        pcg->pcg_objects[pcg->pcg_avail].pcgo_va = object;
                   2422:                        pcg->pcg_objects[pcg->pcg_avail].pcgo_pa = pa;
                   2423:                        pcg->pcg_avail++;
                   2424:                        cc->cc_hits++;
1.162     ad       2425:                        splx(s);
1.134     ad       2426:                        return;
                   2427:                }
1.43      thorpej  2428:
1.134     ad       2429:                /*
1.162     ad       2430:                 * That failed.  If the previous group isn't full, swap
1.134     ad       2431:                 * it with the current group and try again.
                   2432:                 */
                   2433:                pcg = cc->cc_previous;
1.162     ad       2434:                if (__predict_true(pcg->pcg_avail < pcg->pcg_size)) {
1.134     ad       2435:                        cc->cc_previous = cc->cc_current;
                   2436:                        cc->cc_current = pcg;
                   2437:                        continue;
                   2438:                }
1.43      thorpej  2439:
1.134     ad       2440:                /*
                   2441:                 * Can't free to either group: try the slow path.
                   2442:                 * If put_slow() releases the object for us, it
1.162     ad       2443:                 * will return false.  Otherwise we need to retry.
1.134     ad       2444:                 */
1.165     yamt     2445:                if (!pool_cache_put_slow(cc, s, object))
                   2446:                        break;
                   2447:        }
1.43      thorpej  2448: }
                   2449:
                   2450: /*
1.196     jym      2451:  * pool_cache_transfer:
1.43      thorpej  2452:  *
1.134     ad       2453:  *     Transfer objects from the per-CPU cache to the global cache.
                   2454:  *     Run within a cross-call thread.
1.43      thorpej  2455:  */
                   2456: static void
1.196     jym      2457: pool_cache_transfer(pool_cache_t pc)
1.43      thorpej  2458: {
1.134     ad       2459:        pool_cache_cpu_t *cc;
                   2460:        pcg_t *prev, *cur, **list;
1.162     ad       2461:        int s;
1.134     ad       2462:
1.162     ad       2463:        s = splvm();
                   2464:        mutex_enter(&pc->pc_lock);
                   2465:        cc = pc->pc_cpus[curcpu()->ci_index];
1.134     ad       2466:        cur = cc->cc_current;
1.169     yamt     2467:        cc->cc_current = __UNCONST(&pcg_dummy);
1.134     ad       2468:        prev = cc->cc_previous;
1.169     yamt     2469:        cc->cc_previous = __UNCONST(&pcg_dummy);
1.162     ad       2470:        if (cur != &pcg_dummy) {
1.142     ad       2471:                if (cur->pcg_avail == cur->pcg_size) {
1.134     ad       2472:                        list = &pc->pc_fullgroups;
                   2473:                        pc->pc_nfull++;
                   2474:                } else if (cur->pcg_avail == 0) {
                   2475:                        list = &pc->pc_emptygroups;
                   2476:                        pc->pc_nempty++;
                   2477:                } else {
                   2478:                        list = &pc->pc_partgroups;
                   2479:                        pc->pc_npart++;
                   2480:                }
                   2481:                cur->pcg_next = *list;
                   2482:                *list = cur;
                   2483:        }
1.162     ad       2484:        if (prev != &pcg_dummy) {
1.142     ad       2485:                if (prev->pcg_avail == prev->pcg_size) {
1.134     ad       2486:                        list = &pc->pc_fullgroups;
                   2487:                        pc->pc_nfull++;
                   2488:                } else if (prev->pcg_avail == 0) {
                   2489:                        list = &pc->pc_emptygroups;
                   2490:                        pc->pc_nempty++;
                   2491:                } else {
                   2492:                        list = &pc->pc_partgroups;
                   2493:                        pc->pc_npart++;
                   2494:                }
                   2495:                prev->pcg_next = *list;
                   2496:                *list = prev;
                   2497:        }
                   2498:        mutex_exit(&pc->pc_lock);
                   2499:        splx(s);
1.3       pk       2500: }
1.66      thorpej  2501:
                   2502: /*
                   2503:  * Pool backend allocators.
                   2504:  *
                   2505:  * Each pool has a backend allocator that handles allocation, deallocation,
                   2506:  * and any additional draining that might be needed.
                   2507:  *
                   2508:  * We provide two standard allocators:
                   2509:  *
                   2510:  *     pool_allocator_kmem - the default when no allocator is specified
                   2511:  *
                   2512:  *     pool_allocator_nointr - used for pools that will not be accessed
                   2513:  *     in interrupt context.
                   2514:  */
                   2515: void   *pool_page_alloc(struct pool *, int);
                   2516: void   pool_page_free(struct pool *, void *);
                   2517:
1.112     bjh21    2518: #ifdef POOL_SUBPAGE
                   2519: struct pool_allocator pool_allocator_kmem_fullpage = {
1.192     rmind    2520:        .pa_alloc = pool_page_alloc,
                   2521:        .pa_free = pool_page_free,
                   2522:        .pa_pagesz = 0
1.112     bjh21    2523: };
                   2524: #else
1.66      thorpej  2525: struct pool_allocator pool_allocator_kmem = {
1.191     para     2526:        .pa_alloc = pool_page_alloc,
                   2527:        .pa_free = pool_page_free,
                   2528:        .pa_pagesz = 0
1.66      thorpej  2529: };
1.112     bjh21    2530: #endif
1.66      thorpej  2531:
1.112     bjh21    2532: #ifdef POOL_SUBPAGE
                   2533: struct pool_allocator pool_allocator_nointr_fullpage = {
1.194     para     2534:        .pa_alloc = pool_page_alloc,
                   2535:        .pa_free = pool_page_free,
1.192     rmind    2536:        .pa_pagesz = 0
1.112     bjh21    2537: };
                   2538: #else
1.66      thorpej  2539: struct pool_allocator pool_allocator_nointr = {
1.191     para     2540:        .pa_alloc = pool_page_alloc,
                   2541:        .pa_free = pool_page_free,
                   2542:        .pa_pagesz = 0
1.66      thorpej  2543: };
1.112     bjh21    2544: #endif
1.66      thorpej  2545:
                   2546: #ifdef POOL_SUBPAGE
                   2547: void   *pool_subpage_alloc(struct pool *, int);
                   2548: void   pool_subpage_free(struct pool *, void *);
                   2549:
1.112     bjh21    2550: struct pool_allocator pool_allocator_kmem = {
1.193     he       2551:        .pa_alloc = pool_subpage_alloc,
                   2552:        .pa_free = pool_subpage_free,
                   2553:        .pa_pagesz = POOL_SUBPAGE
1.112     bjh21    2554: };
                   2555:
                   2556: struct pool_allocator pool_allocator_nointr = {
1.192     rmind    2557:        .pa_alloc = pool_subpage_alloc,
                   2558:        .pa_free = pool_subpage_free,
                   2559:        .pa_pagesz = POOL_SUBPAGE
1.66      thorpej  2560: };
                   2561: #endif /* POOL_SUBPAGE */
                   2562:
1.208     chs      2563: struct pool_allocator pool_allocator_big[] = {
                   2564:        {
                   2565:                .pa_alloc = pool_page_alloc,
                   2566:                .pa_free = pool_page_free,
                   2567:                .pa_pagesz = 1 << (POOL_ALLOCATOR_BIG_BASE + 0),
                   2568:        },
                   2569:        {
                   2570:                .pa_alloc = pool_page_alloc,
                   2571:                .pa_free = pool_page_free,
                   2572:                .pa_pagesz = 1 << (POOL_ALLOCATOR_BIG_BASE + 1),
                   2573:        },
                   2574:        {
                   2575:                .pa_alloc = pool_page_alloc,
                   2576:                .pa_free = pool_page_free,
                   2577:                .pa_pagesz = 1 << (POOL_ALLOCATOR_BIG_BASE + 2),
                   2578:        },
                   2579:        {
                   2580:                .pa_alloc = pool_page_alloc,
                   2581:                .pa_free = pool_page_free,
                   2582:                .pa_pagesz = 1 << (POOL_ALLOCATOR_BIG_BASE + 3),
                   2583:        },
                   2584:        {
                   2585:                .pa_alloc = pool_page_alloc,
                   2586:                .pa_free = pool_page_free,
                   2587:                .pa_pagesz = 1 << (POOL_ALLOCATOR_BIG_BASE + 4),
                   2588:        },
                   2589:        {
                   2590:                .pa_alloc = pool_page_alloc,
                   2591:                .pa_free = pool_page_free,
                   2592:                .pa_pagesz = 1 << (POOL_ALLOCATOR_BIG_BASE + 5),
                   2593:        },
                   2594:        {
                   2595:                .pa_alloc = pool_page_alloc,
                   2596:                .pa_free = pool_page_free,
                   2597:                .pa_pagesz = 1 << (POOL_ALLOCATOR_BIG_BASE + 6),
                   2598:        },
                   2599:        {
                   2600:                .pa_alloc = pool_page_alloc,
                   2601:                .pa_free = pool_page_free,
                   2602:                .pa_pagesz = 1 << (POOL_ALLOCATOR_BIG_BASE + 7),
                   2603:        }
                   2604: };
                   2605:
                   2606: static int
                   2607: pool_bigidx(size_t size)
                   2608: {
                   2609:        int i;
                   2610:
                   2611:        for (i = 0; i < __arraycount(pool_allocator_big); i++) {
                   2612:                if (1 << (i + POOL_ALLOCATOR_BIG_BASE) >= size)
                   2613:                        return i;
                   2614:        }
                   2615:        panic("pool item size %zu too large, use a custom allocator", size);
                   2616: }
                   2617:
1.117     yamt     2618: static void *
                   2619: pool_allocator_alloc(struct pool *pp, int flags)
1.66      thorpej  2620: {
1.117     yamt     2621:        struct pool_allocator *pa = pp->pr_alloc;
1.66      thorpej  2622:        void *res;
                   2623:
1.117     yamt     2624:        res = (*pa->pa_alloc)(pp, flags);
                   2625:        if (res == NULL && (flags & PR_WAITOK) == 0) {
1.66      thorpej  2626:                /*
1.117     yamt     2627:                 * We only run the drain hook here if PR_NOWAIT.
                   2628:                 * In other cases, the hook will be run in
                   2629:                 * pool_reclaim().
1.66      thorpej  2630:                 */
1.117     yamt     2631:                if (pp->pr_drain_hook != NULL) {
                   2632:                        (*pp->pr_drain_hook)(pp->pr_drain_hook_arg, flags);
                   2633:                        res = (*pa->pa_alloc)(pp, flags);
1.66      thorpej  2634:                }
1.117     yamt     2635:        }
                   2636:        return res;
1.66      thorpej  2637: }
                   2638:
1.117     yamt     2639: static void
1.66      thorpej  2640: pool_allocator_free(struct pool *pp, void *v)
                   2641: {
                   2642:        struct pool_allocator *pa = pp->pr_alloc;
                   2643:
                   2644:        (*pa->pa_free)(pp, v);
                   2645: }
                   2646:
                   2647: void *
1.124     yamt     2648: pool_page_alloc(struct pool *pp, int flags)
1.66      thorpej  2649: {
1.192     rmind    2650:        const vm_flag_t vflags = (flags & PR_WAITOK) ? VM_SLEEP: VM_NOSLEEP;
1.191     para     2651:        vmem_addr_t va;
1.192     rmind    2652:        int ret;
1.191     para     2653:
1.192     rmind    2654:        ret = uvm_km_kmem_alloc(kmem_va_arena, pp->pr_alloc->pa_pagesz,
                   2655:            vflags | VM_INSTANTFIT, &va);
1.66      thorpej  2656:
1.192     rmind    2657:        return ret ? NULL : (void *)va;
1.66      thorpej  2658: }
                   2659:
                   2660: void
1.124     yamt     2661: pool_page_free(struct pool *pp, void *v)
1.66      thorpej  2662: {
                   2663:
1.191     para     2664:        uvm_km_kmem_free(kmem_va_arena, (vaddr_t)v, pp->pr_alloc->pa_pagesz);
1.98      yamt     2665: }
                   2666:
                   2667: static void *
1.124     yamt     2668: pool_page_alloc_meta(struct pool *pp, int flags)
1.98      yamt     2669: {
1.192     rmind    2670:        const vm_flag_t vflags = (flags & PR_WAITOK) ? VM_SLEEP: VM_NOSLEEP;
                   2671:        vmem_addr_t va;
                   2672:        int ret;
1.191     para     2673:
1.192     rmind    2674:        ret = vmem_alloc(kmem_meta_arena, pp->pr_alloc->pa_pagesz,
                   2675:            vflags | VM_INSTANTFIT, &va);
1.98      yamt     2676:
1.192     rmind    2677:        return ret ? NULL : (void *)va;
1.98      yamt     2678: }
                   2679:
                   2680: static void
1.124     yamt     2681: pool_page_free_meta(struct pool *pp, void *v)
1.98      yamt     2682: {
                   2683:
1.192     rmind    2684:        vmem_free(kmem_meta_arena, (vmem_addr_t)v, pp->pr_alloc->pa_pagesz);
1.66      thorpej  2685: }
                   2686:
1.204     maxv     2687: #ifdef POOL_REDZONE
                   2688: #if defined(_LP64)
                   2689: # define PRIME 0x9e37fffffffc0000UL
                   2690: #else /* defined(_LP64) */
                   2691: # define PRIME 0x9e3779b1
                   2692: #endif /* defined(_LP64) */
                   2693: #define STATIC_BYTE    0xFE
                   2694: CTASSERT(POOL_REDZONE_SIZE > 1);
                   2695:
                   2696: static inline uint8_t
                   2697: pool_pattern_generate(const void *p)
                   2698: {
                   2699:        return (uint8_t)(((uintptr_t)p) * PRIME
                   2700:           >> ((sizeof(uintptr_t) - sizeof(uint8_t))) * CHAR_BIT);
                   2701: }
                   2702:
                   2703: static void
                   2704: pool_redzone_init(struct pool *pp, size_t requested_size)
                   2705: {
                   2706:        size_t nsz;
                   2707:
                   2708:        if (pp->pr_roflags & PR_NOTOUCH) {
                   2709:                pp->pr_reqsize = 0;
                   2710:                pp->pr_redzone = false;
                   2711:                return;
                   2712:        }
                   2713:
                   2714:        /*
                   2715:         * We may have extended the requested size earlier; check if
                   2716:         * there's naturally space in the padding for a red zone.
                   2717:         */
                   2718:        if (pp->pr_size - requested_size >= POOL_REDZONE_SIZE) {
                   2719:                pp->pr_reqsize = requested_size;
                   2720:                pp->pr_redzone = true;
                   2721:                return;
                   2722:        }
                   2723:
                   2724:        /*
                   2725:         * No space in the natural padding; check if we can extend a
                   2726:         * bit the size of the pool.
                   2727:         */
                   2728:        nsz = roundup(pp->pr_size + POOL_REDZONE_SIZE, pp->pr_align);
                   2729:        if (nsz <= pp->pr_alloc->pa_pagesz) {
                   2730:                /* Ok, we can */
                   2731:                pp->pr_size = nsz;
                   2732:                pp->pr_reqsize = requested_size;
                   2733:                pp->pr_redzone = true;
                   2734:        } else {
                   2735:                /* No space for a red zone... snif :'( */
                   2736:                pp->pr_reqsize = 0;
                   2737:                pp->pr_redzone = false;
                   2738:                printf("pool redzone disabled for '%s'\n", pp->pr_wchan);
                   2739:        }
                   2740: }
                   2741:
                   2742: static void
                   2743: pool_redzone_fill(struct pool *pp, void *p)
                   2744: {
                   2745:        uint8_t *cp, pat;
                   2746:        const uint8_t *ep;
                   2747:
                   2748:        if (!pp->pr_redzone)
                   2749:                return;
                   2750:
                   2751:        cp = (uint8_t *)p + pp->pr_reqsize;
                   2752:        ep = cp + POOL_REDZONE_SIZE;
                   2753:
                   2754:        /*
                   2755:         * We really don't want the first byte of the red zone to be '\0';
                   2756:         * an off-by-one in a string may not be properly detected.
                   2757:         */
                   2758:        pat = pool_pattern_generate(cp);
                   2759:        *cp = (pat == '\0') ? STATIC_BYTE: pat;
                   2760:        cp++;
                   2761:
                   2762:        while (cp < ep) {
                   2763:                *cp = pool_pattern_generate(cp);
                   2764:                cp++;
                   2765:        }
                   2766: }
                   2767:
                   2768: static void
                   2769: pool_redzone_check(struct pool *pp, void *p)
                   2770: {
                   2771:        uint8_t *cp, pat, expected;
                   2772:        const uint8_t *ep;
                   2773:
                   2774:        if (!pp->pr_redzone)
                   2775:                return;
                   2776:
                   2777:        cp = (uint8_t *)p + pp->pr_reqsize;
                   2778:        ep = cp + POOL_REDZONE_SIZE;
                   2779:
                   2780:        pat = pool_pattern_generate(cp);
                   2781:        expected = (pat == '\0') ? STATIC_BYTE: pat;
                   2782:        if (expected != *cp) {
                   2783:                panic("%s: %p: 0x%02x != 0x%02x\n",
                   2784:                   __func__, cp, *cp, expected);
                   2785:        }
                   2786:        cp++;
                   2787:
                   2788:        while (cp < ep) {
                   2789:                expected = pool_pattern_generate(cp);
                   2790:                if (*cp != expected) {
                   2791:                        panic("%s: %p: 0x%02x != 0x%02x\n",
                   2792:                           __func__, cp, *cp, expected);
                   2793:                }
                   2794:                cp++;
                   2795:        }
                   2796: }
                   2797:
                   2798: #endif /* POOL_REDZONE */
                   2799:
                   2800:
1.66      thorpej  2801: #ifdef POOL_SUBPAGE
                   2802: /* Sub-page allocator, for machines with large hardware pages. */
                   2803: void *
                   2804: pool_subpage_alloc(struct pool *pp, int flags)
                   2805: {
1.134     ad       2806:        return pool_get(&psppool, flags);
1.66      thorpej  2807: }
                   2808:
                   2809: void
                   2810: pool_subpage_free(struct pool *pp, void *v)
                   2811: {
                   2812:        pool_put(&psppool, v);
                   2813: }
                   2814:
1.112     bjh21    2815: #endif /* POOL_SUBPAGE */
1.141     yamt     2816:
                   2817: #if defined(DDB)
                   2818: static bool
                   2819: pool_in_page(struct pool *pp, struct pool_item_header *ph, uintptr_t addr)
                   2820: {
                   2821:
                   2822:        return (uintptr_t)ph->ph_page <= addr &&
                   2823:            addr < (uintptr_t)ph->ph_page + pp->pr_alloc->pa_pagesz;
                   2824: }
                   2825:
1.143     yamt     2826: static bool
                   2827: pool_in_item(struct pool *pp, void *item, uintptr_t addr)
                   2828: {
                   2829:
                   2830:        return (uintptr_t)item <= addr && addr < (uintptr_t)item + pp->pr_size;
                   2831: }
                   2832:
                   2833: static bool
                   2834: pool_in_cg(struct pool *pp, struct pool_cache_group *pcg, uintptr_t addr)
                   2835: {
                   2836:        int i;
                   2837:
                   2838:        if (pcg == NULL) {
                   2839:                return false;
                   2840:        }
1.144     yamt     2841:        for (i = 0; i < pcg->pcg_avail; i++) {
1.143     yamt     2842:                if (pool_in_item(pp, pcg->pcg_objects[i].pcgo_va, addr)) {
                   2843:                        return true;
                   2844:                }
                   2845:        }
                   2846:        return false;
                   2847: }
                   2848:
                   2849: static bool
                   2850: pool_allocated(struct pool *pp, struct pool_item_header *ph, uintptr_t addr)
                   2851: {
                   2852:
                   2853:        if ((pp->pr_roflags & PR_NOTOUCH) != 0) {
                   2854:                unsigned int idx = pr_item_notouch_index(pp, ph, (void *)addr);
                   2855:                pool_item_bitmap_t *bitmap =
                   2856:                    ph->ph_bitmap + (idx / BITMAP_SIZE);
                   2857:                pool_item_bitmap_t mask = 1 << (idx & BITMAP_MASK);
                   2858:
                   2859:                return (*bitmap & mask) == 0;
                   2860:        } else {
                   2861:                struct pool_item *pi;
                   2862:
                   2863:                LIST_FOREACH(pi, &ph->ph_itemlist, pi_list) {
                   2864:                        if (pool_in_item(pp, pi, addr)) {
                   2865:                                return false;
                   2866:                        }
                   2867:                }
                   2868:                return true;
                   2869:        }
                   2870: }
                   2871:
1.141     yamt     2872: void
                   2873: pool_whatis(uintptr_t addr, void (*pr)(const char *, ...))
                   2874: {
                   2875:        struct pool *pp;
                   2876:
1.145     ad       2877:        TAILQ_FOREACH(pp, &pool_head, pr_poollist) {
1.141     yamt     2878:                struct pool_item_header *ph;
                   2879:                uintptr_t item;
1.143     yamt     2880:                bool allocated = true;
                   2881:                bool incache = false;
                   2882:                bool incpucache = false;
                   2883:                char cpucachestr[32];
1.141     yamt     2884:
                   2885:                if ((pp->pr_roflags & PR_PHINPAGE) != 0) {
                   2886:                        LIST_FOREACH(ph, &pp->pr_fullpages, ph_pagelist) {
                   2887:                                if (pool_in_page(pp, ph, addr)) {
                   2888:                                        goto found;
                   2889:                                }
                   2890:                        }
                   2891:                        LIST_FOREACH(ph, &pp->pr_partpages, ph_pagelist) {
                   2892:                                if (pool_in_page(pp, ph, addr)) {
1.143     yamt     2893:                                        allocated =
                   2894:                                            pool_allocated(pp, ph, addr);
                   2895:                                        goto found;
                   2896:                                }
                   2897:                        }
                   2898:                        LIST_FOREACH(ph, &pp->pr_emptypages, ph_pagelist) {
                   2899:                                if (pool_in_page(pp, ph, addr)) {
                   2900:                                        allocated = false;
1.141     yamt     2901:                                        goto found;
                   2902:                                }
                   2903:                        }
                   2904:                        continue;
                   2905:                } else {
                   2906:                        ph = pr_find_pagehead_noalign(pp, (void *)addr);
                   2907:                        if (ph == NULL || !pool_in_page(pp, ph, addr)) {
                   2908:                                continue;
                   2909:                        }
1.143     yamt     2910:                        allocated = pool_allocated(pp, ph, addr);
1.141     yamt     2911:                }
                   2912: found:
1.143     yamt     2913:                if (allocated && pp->pr_cache) {
                   2914:                        pool_cache_t pc = pp->pr_cache;
                   2915:                        struct pool_cache_group *pcg;
                   2916:                        int i;
                   2917:
                   2918:                        for (pcg = pc->pc_fullgroups; pcg != NULL;
                   2919:                            pcg = pcg->pcg_next) {
                   2920:                                if (pool_in_cg(pp, pcg, addr)) {
                   2921:                                        incache = true;
                   2922:                                        goto print;
                   2923:                                }
                   2924:                        }
1.183     ad       2925:                        for (i = 0; i < __arraycount(pc->pc_cpus); i++) {
1.143     yamt     2926:                                pool_cache_cpu_t *cc;
                   2927:
                   2928:                                if ((cc = pc->pc_cpus[i]) == NULL) {
                   2929:                                        continue;
                   2930:                                }
                   2931:                                if (pool_in_cg(pp, cc->cc_current, addr) ||
                   2932:                                    pool_in_cg(pp, cc->cc_previous, addr)) {
                   2933:                                        struct cpu_info *ci =
1.170     ad       2934:                                            cpu_lookup(i);
1.143     yamt     2935:
                   2936:                                        incpucache = true;
                   2937:                                        snprintf(cpucachestr,
                   2938:                                            sizeof(cpucachestr),
                   2939:                                            "cached by CPU %u",
1.153     martin   2940:                                            ci->ci_index);
1.143     yamt     2941:                                        goto print;
                   2942:                                }
                   2943:                        }
                   2944:                }
                   2945: print:
1.141     yamt     2946:                item = (uintptr_t)ph->ph_page + ph->ph_off;
                   2947:                item = item + rounddown(addr - item, pp->pr_size);
1.143     yamt     2948:                (*pr)("%p is %p+%zu in POOL '%s' (%s)\n",
1.141     yamt     2949:                    (void *)addr, item, (size_t)(addr - item),
1.143     yamt     2950:                    pp->pr_wchan,
                   2951:                    incpucache ? cpucachestr :
                   2952:                    incache ? "cached" : allocated ? "allocated" : "free");
1.141     yamt     2953:        }
                   2954: }
                   2955: #endif /* defined(DDB) */
1.203     joerg    2956:
                   2957: static int
                   2958: pool_sysctl(SYSCTLFN_ARGS)
                   2959: {
                   2960:        struct pool_sysctl data;
                   2961:        struct pool *pp;
                   2962:        struct pool_cache *pc;
                   2963:        pool_cache_cpu_t *cc;
                   2964:        int error;
                   2965:        size_t i, written;
                   2966:
                   2967:        if (oldp == NULL) {
                   2968:                *oldlenp = 0;
                   2969:                TAILQ_FOREACH(pp, &pool_head, pr_poollist)
                   2970:                        *oldlenp += sizeof(data);
                   2971:                return 0;
                   2972:        }
                   2973:
                   2974:        memset(&data, 0, sizeof(data));
                   2975:        error = 0;
                   2976:        written = 0;
                   2977:        TAILQ_FOREACH(pp, &pool_head, pr_poollist) {
                   2978:                if (written + sizeof(data) > *oldlenp)
                   2979:                        break;
                   2980:                strlcpy(data.pr_wchan, pp->pr_wchan, sizeof(data.pr_wchan));
                   2981:                data.pr_pagesize = pp->pr_alloc->pa_pagesz;
                   2982:                data.pr_flags = pp->pr_roflags | pp->pr_flags;
                   2983: #define COPY(field) data.field = pp->field
                   2984:                COPY(pr_size);
                   2985:
                   2986:                COPY(pr_itemsperpage);
                   2987:                COPY(pr_nitems);
                   2988:                COPY(pr_nout);
                   2989:                COPY(pr_hardlimit);
                   2990:                COPY(pr_npages);
                   2991:                COPY(pr_minpages);
                   2992:                COPY(pr_maxpages);
                   2993:
                   2994:                COPY(pr_nget);
                   2995:                COPY(pr_nfail);
                   2996:                COPY(pr_nput);
                   2997:                COPY(pr_npagealloc);
                   2998:                COPY(pr_npagefree);
                   2999:                COPY(pr_hiwat);
                   3000:                COPY(pr_nidle);
                   3001: #undef COPY
                   3002:
                   3003:                data.pr_cache_nmiss_pcpu = 0;
                   3004:                data.pr_cache_nhit_pcpu = 0;
                   3005:                if (pp->pr_cache) {
                   3006:                        pc = pp->pr_cache;
                   3007:                        data.pr_cache_meta_size = pc->pc_pcgsize;
                   3008:                        data.pr_cache_nfull = pc->pc_nfull;
                   3009:                        data.pr_cache_npartial = pc->pc_npart;
                   3010:                        data.pr_cache_nempty = pc->pc_nempty;
                   3011:                        data.pr_cache_ncontended = pc->pc_contended;
                   3012:                        data.pr_cache_nmiss_global = pc->pc_misses;
                   3013:                        data.pr_cache_nhit_global = pc->pc_hits;
                   3014:                        for (i = 0; i < pc->pc_ncpu; ++i) {
                   3015:                                cc = pc->pc_cpus[i];
                   3016:                                if (cc == NULL)
                   3017:                                        continue;
1.206     knakahar 3018:                                data.pr_cache_nmiss_pcpu += cc->cc_misses;
                   3019:                                data.pr_cache_nhit_pcpu += cc->cc_hits;
1.203     joerg    3020:                        }
                   3021:                } else {
                   3022:                        data.pr_cache_meta_size = 0;
                   3023:                        data.pr_cache_nfull = 0;
                   3024:                        data.pr_cache_npartial = 0;
                   3025:                        data.pr_cache_nempty = 0;
                   3026:                        data.pr_cache_ncontended = 0;
                   3027:                        data.pr_cache_nmiss_global = 0;
                   3028:                        data.pr_cache_nhit_global = 0;
                   3029:                }
                   3030:
                   3031:                error = sysctl_copyout(l, &data, oldp, sizeof(data));
                   3032:                if (error)
                   3033:                        break;
                   3034:                written += sizeof(data);
                   3035:                oldp = (char *)oldp + sizeof(data);
                   3036:        }
                   3037:
                   3038:        *oldlenp = written;
                   3039:        return error;
                   3040: }
                   3041:
                   3042: SYSCTL_SETUP(sysctl_pool_setup, "sysctl kern.pool setup")
                   3043: {
                   3044:        const struct sysctlnode *rnode = NULL;
                   3045:
                   3046:        sysctl_createv(clog, 0, NULL, &rnode,
                   3047:                       CTLFLAG_PERMANENT,
                   3048:                       CTLTYPE_STRUCT, "pool",
                   3049:                       SYSCTL_DESCR("Get pool statistics"),
                   3050:                       pool_sysctl, 0, NULL, 0,
                   3051:                       CTL_KERN, CTL_CREATE, CTL_EOL);
                   3052: }

CVSweb <webmaster@jp.NetBSD.org>