[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 1.101.2.10

1.101.2.10! yamt        1: /*     $NetBSD: subr_pool.c,v 1.101.2.9 2008/02/04 09:24:16 yamt Exp $ */
1.1       pk          2:
                      3: /*-
1.101.2.6  yamt        4:  * Copyright (c) 1997, 1999, 2000, 2002, 2007 The NetBSD Foundation, Inc.
1.1       pk          5:  * All rights reserved.
                      6:  *
                      7:  * This code is derived from software contributed to The NetBSD Foundation
1.20      thorpej     8:  * by Paul Kranenburg; by Jason R. Thorpe of the Numerical Aerospace
1.101.2.6  yamt        9:  * Simulation Facility, NASA Ames Research Center, and by Andrew Doran.
1.1       pk         10:  *
                     11:  * Redistribution and use in source and binary forms, with or without
                     12:  * modification, are permitted provided that the following conditions
                     13:  * are met:
                     14:  * 1. Redistributions of source code must retain the above copyright
                     15:  *    notice, this list of conditions and the following disclaimer.
                     16:  * 2. Redistributions in binary form must reproduce the above copyright
                     17:  *    notice, this list of conditions and the following disclaimer in the
                     18:  *    documentation and/or other materials provided with the distribution.
                     19:  * 3. All advertising materials mentioning features or use of this software
                     20:  *    must display the following acknowledgement:
1.13      christos   21:  *     This product includes software developed by the NetBSD
                     22:  *     Foundation, Inc. and its contributors.
1.1       pk         23:  * 4. Neither the name of The NetBSD Foundation nor the names of its
                     24:  *    contributors may be used to endorse or promote products derived
                     25:  *    from this software without specific prior written permission.
                     26:  *
                     27:  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
                     28:  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
                     29:  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
                     30:  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
                     31:  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
                     32:  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
                     33:  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
                     34:  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
                     35:  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
                     36:  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
                     37:  * POSSIBILITY OF SUCH DAMAGE.
                     38:  */
1.64      lukem      39:
                     40: #include <sys/cdefs.h>
1.101.2.10! yamt       41: __KERNEL_RCSID(0, "$NetBSD: subr_pool.c,v 1.101.2.9 2008/02/04 09:24:16 yamt Exp $");
1.24      scottr     42:
1.101.2.8  yamt       43: #include "opt_ddb.h"
1.25      thorpej    44: #include "opt_pool.h"
1.24      scottr     45: #include "opt_poollog.h"
1.28      thorpej    46: #include "opt_lockdebug.h"
1.1       pk         47:
                     48: #include <sys/param.h>
                     49: #include <sys/systm.h>
1.101.2.6  yamt       50: #include <sys/bitops.h>
1.1       pk         51: #include <sys/proc.h>
                     52: #include <sys/errno.h>
                     53: #include <sys/kernel.h>
                     54: #include <sys/malloc.h>
                     55: #include <sys/pool.h>
1.20      thorpej    56: #include <sys/syslog.h>
1.101.2.3  yamt       57: #include <sys/debug.h>
1.101.2.6  yamt       58: #include <sys/lockdebug.h>
                     59: #include <sys/xcall.h>
                     60: #include <sys/cpu.h>
1.101.2.8  yamt       61: #include <sys/atomic.h>
1.3       pk         62:
                     63: #include <uvm/uvm.h>
                     64:
1.1       pk         65: /*
                     66:  * Pool resource management utility.
1.3       pk         67:  *
1.88      chs        68:  * Memory is allocated in pages which are split into pieces according to
                     69:  * the pool item size. Each page is kept on one of three lists in the
                     70:  * pool structure: `pr_emptypages', `pr_fullpages' and `pr_partpages',
                     71:  * for empty, full and partially-full pages respectively. The individual
                     72:  * pool items are on a linked list headed by `ph_itemlist' in each page
                     73:  * header. The memory for building the page list is either taken from
                     74:  * the allocated pages themselves (for small pool items) or taken from
                     75:  * an internal pool of page headers (`phpool').
1.1       pk         76:  */
                     77:
1.3       pk         78: /* List of all pools */
1.101.2.8  yamt       79: TAILQ_HEAD(,pool) pool_head = TAILQ_HEAD_INITIALIZER(pool_head);
1.101.2.6  yamt       80:
1.3       pk         81: /* Private pool for page header structures */
1.97      yamt       82: #define        PHPOOL_MAX      8
                     83: static struct pool phpool[PHPOOL_MAX];
1.101.2.6  yamt       84: #define        PHPOOL_FREELIST_NELEM(idx) \
                     85:        (((idx) == 0) ? 0 : BITMAP_SIZE * (1 << (idx)))
1.3       pk         86:
1.62      bjh21      87: #ifdef POOL_SUBPAGE
                     88: /* Pool of subpages for use by normal pools. */
                     89: static struct pool psppool;
                     90: #endif
                     91:
1.101.2.1  yamt       92: static SLIST_HEAD(, pool_allocator) pa_deferinitq =
                     93:     SLIST_HEAD_INITIALIZER(pa_deferinitq);
                     94:
1.98      yamt       95: static void *pool_page_alloc_meta(struct pool *, int);
                     96: static void pool_page_free_meta(struct pool *, void *);
                     97:
                     98: /* allocator for pool metadata */
1.101.2.6  yamt       99: struct pool_allocator pool_allocator_meta = {
1.101.2.1  yamt      100:        pool_page_alloc_meta, pool_page_free_meta,
                    101:        .pa_backingmapptr = &kmem_map,
1.98      yamt      102: };
                    103:
1.3       pk        104: /* # of seconds to retain page after last use */
                    105: int pool_inactive_time = 10;
                    106:
                    107: /* Next candidate for drainage (see pool_drain()) */
1.23      thorpej   108: static struct pool     *drainpp;
                    109:
1.101.2.6  yamt      110: /* This lock protects both pool_head and drainpp. */
                    111: static kmutex_t pool_head_lock;
                    112: static kcondvar_t pool_busy;
                    113:
                    114: typedef uint32_t pool_item_bitmap_t;
                    115: #define        BITMAP_SIZE     (CHAR_BIT * sizeof(pool_item_bitmap_t))
                    116: #define        BITMAP_MASK     (BITMAP_SIZE - 1)
1.99      yamt      117:
1.3       pk        118: struct pool_item_header {
                    119:        /* Page headers */
1.88      chs       120:        LIST_ENTRY(pool_item_header)
1.3       pk        121:                                ph_pagelist;    /* pool page list */
1.88      chs       122:        SPLAY_ENTRY(pool_item_header)
                    123:                                ph_node;        /* Off-page page headers */
1.101.2.4  yamt      124:        void *                  ph_page;        /* this page's address */
1.3       pk        125:        struct timeval          ph_time;        /* last referenced */
1.101.2.6  yamt      126:        uint16_t                ph_nmissing;    /* # of chunks in use */
1.101.2.8  yamt      127:        uint16_t                ph_off;         /* start offset in page */
1.97      yamt      128:        union {
                    129:                /* !PR_NOTOUCH */
                    130:                struct {
1.101.2.1  yamt      131:                        LIST_HEAD(, pool_item)
1.97      yamt      132:                                phu_itemlist;   /* chunk list for this page */
                    133:                } phu_normal;
                    134:                /* PR_NOTOUCH */
                    135:                struct {
1.101.2.8  yamt      136:                        pool_item_bitmap_t phu_bitmap[1];
1.97      yamt      137:                } phu_notouch;
                    138:        } ph_u;
1.3       pk        139: };
1.97      yamt      140: #define        ph_itemlist     ph_u.phu_normal.phu_itemlist
1.101.2.6  yamt      141: #define        ph_bitmap       ph_u.phu_notouch.phu_bitmap
1.3       pk        142:
1.1       pk        143: struct pool_item {
1.3       pk        144: #ifdef DIAGNOSTIC
1.82      thorpej   145:        u_int pi_magic;
1.33      chs       146: #endif
1.101.2.6  yamt      147: #define        PI_MAGIC 0xdeaddeadU
1.3       pk        148:        /* Other entries use only this list entry */
1.101.2.1  yamt      149:        LIST_ENTRY(pool_item)   pi_list;
1.3       pk        150: };
                    151:
1.53      thorpej   152: #define        POOL_NEEDS_CATCHUP(pp)                                          \
                    153:        ((pp)->pr_nitems < (pp)->pr_minitems)
                    154:
1.43      thorpej   155: /*
                    156:  * Pool cache management.
                    157:  *
                    158:  * Pool caches provide a way for constructed objects to be cached by the
                    159:  * pool subsystem.  This can lead to performance improvements by avoiding
                    160:  * needless object construction/destruction; it is deferred until absolutely
                    161:  * necessary.
                    162:  *
1.101.2.6  yamt      163:  * Caches are grouped into cache groups.  Each cache group references up
                    164:  * to PCG_NUMOBJECTS constructed objects.  When a cache allocates an
                    165:  * object from the pool, it calls the object's constructor and places it
                    166:  * into a cache group.  When a cache group frees an object back to the
                    167:  * pool, it first calls the object's destructor.  This allows the object
                    168:  * to persist in constructed form while freed to the cache.
                    169:  *
                    170:  * The pool references each cache, so that when a pool is drained by the
                    171:  * pagedaemon, it can drain each individual cache as well.  Each time a
                    172:  * cache is drained, the most idle cache group is freed to the pool in
                    173:  * its entirety.
1.43      thorpej   174:  *
                    175:  * Pool caches are layed on top of pools.  By layering them, we can avoid
                    176:  * the complexity of cache management for pools which would not benefit
                    177:  * from it.
                    178:  */
                    179:
1.101.2.8  yamt      180: static struct pool pcg_normal_pool;
                    181: static struct pool pcg_large_pool;
1.101.2.6  yamt      182: static struct pool cache_pool;
                    183: static struct pool cache_cpu_pool;
1.3       pk        184:
1.101.2.8  yamt      185: /* List of all caches. */
                    186: TAILQ_HEAD(,pool_cache) pool_cache_head =
                    187:     TAILQ_HEAD_INITIALIZER(pool_cache_head);
                    188:
                    189: int pool_cache_disable;
                    190:
                    191:
1.101.2.6  yamt      192: static pool_cache_cpu_t *pool_cache_put_slow(pool_cache_cpu_t *, int *,
                    193:                                             void *, paddr_t);
                    194: static pool_cache_cpu_t *pool_cache_get_slow(pool_cache_cpu_t *, int *,
                    195:                                             void **, paddr_t *, int);
                    196: static void    pool_cache_cpu_init1(struct cpu_info *, pool_cache_t);
                    197: static void    pool_cache_invalidate_groups(pool_cache_t, pcg_t *);
                    198: static void    pool_cache_xcall(pool_cache_t);
1.3       pk        199:
1.42      thorpej   200: static int     pool_catchup(struct pool *);
1.101.2.4  yamt      201: static void    pool_prime_page(struct pool *, void *,
1.55      thorpej   202:                    struct pool_item_header *);
1.88      chs       203: static void    pool_update_curpage(struct pool *);
1.66      thorpej   204:
1.101.2.1  yamt      205: static int     pool_grow(struct pool *, int);
                    206: static void    *pool_allocator_alloc(struct pool *, int);
                    207: static void    pool_allocator_free(struct pool *, void *);
1.3       pk        208:
1.97      yamt      209: static void pool_print_pagelist(struct pool *, struct pool_pagelist *,
1.88      chs       210:        void (*)(const char *, ...));
1.42      thorpej   211: static void pool_print1(struct pool *, const char *,
                    212:        void (*)(const char *, ...));
1.3       pk        213:
1.88      chs       214: static int pool_chk_page(struct pool *, const char *,
                    215:                         struct pool_item_header *);
                    216:
1.3       pk        217: /*
1.52      thorpej   218:  * Pool log entry. An array of these is allocated in pool_init().
1.3       pk        219:  */
                    220: struct pool_log {
                    221:        const char      *pl_file;
                    222:        long            pl_line;
                    223:        int             pl_action;
1.25      thorpej   224: #define        PRLOG_GET       1
                    225: #define        PRLOG_PUT       2
1.3       pk        226:        void            *pl_addr;
1.1       pk        227: };
                    228:
1.86      matt      229: #ifdef POOL_DIAGNOSTIC
1.3       pk        230: /* Number of entries in pool log buffers */
1.17      thorpej   231: #ifndef POOL_LOGSIZE
                    232: #define        POOL_LOGSIZE    10
                    233: #endif
                    234:
                    235: int pool_logsize = POOL_LOGSIZE;
1.1       pk        236:
1.101.2.1  yamt      237: static inline void
1.42      thorpej   238: pr_log(struct pool *pp, void *v, int action, const char *file, long line)
1.3       pk        239: {
                    240:        int n = pp->pr_curlogentry;
                    241:        struct pool_log *pl;
                    242:
1.20      thorpej   243:        if ((pp->pr_roflags & PR_LOGGING) == 0)
1.3       pk        244:                return;
                    245:
                    246:        /*
                    247:         * Fill in the current entry. Wrap around and overwrite
                    248:         * the oldest entry if necessary.
                    249:         */
                    250:        pl = &pp->pr_log[n];
                    251:        pl->pl_file = file;
                    252:        pl->pl_line = line;
                    253:        pl->pl_action = action;
                    254:        pl->pl_addr = v;
                    255:        if (++n >= pp->pr_logsize)
                    256:                n = 0;
                    257:        pp->pr_curlogentry = n;
                    258: }
                    259:
                    260: static void
1.42      thorpej   261: pr_printlog(struct pool *pp, struct pool_item *pi,
                    262:     void (*pr)(const char *, ...))
1.3       pk        263: {
                    264:        int i = pp->pr_logsize;
                    265:        int n = pp->pr_curlogentry;
                    266:
1.20      thorpej   267:        if ((pp->pr_roflags & PR_LOGGING) == 0)
1.3       pk        268:                return;
                    269:
                    270:        /*
                    271:         * Print all entries in this pool's log.
                    272:         */
                    273:        while (i-- > 0) {
                    274:                struct pool_log *pl = &pp->pr_log[n];
                    275:                if (pl->pl_action != 0) {
1.25      thorpej   276:                        if (pi == NULL || pi == pl->pl_addr) {
                    277:                                (*pr)("\tlog entry %d:\n", i);
                    278:                                (*pr)("\t\taction = %s, addr = %p\n",
                    279:                                    pl->pl_action == PRLOG_GET ? "get" : "put",
                    280:                                    pl->pl_addr);
                    281:                                (*pr)("\t\tfile: %s at line %lu\n",
                    282:                                    pl->pl_file, pl->pl_line);
                    283:                        }
1.3       pk        284:                }
                    285:                if (++n >= pp->pr_logsize)
                    286:                        n = 0;
                    287:        }
                    288: }
1.25      thorpej   289:
1.101.2.1  yamt      290: static inline void
1.42      thorpej   291: pr_enter(struct pool *pp, const char *file, long line)
1.25      thorpej   292: {
                    293:
1.34      thorpej   294:        if (__predict_false(pp->pr_entered_file != NULL)) {
1.25      thorpej   295:                printf("pool %s: reentrancy at file %s line %ld\n",
                    296:                    pp->pr_wchan, file, line);
                    297:                printf("         previous entry at file %s line %ld\n",
                    298:                    pp->pr_entered_file, pp->pr_entered_line);
                    299:                panic("pr_enter");
                    300:        }
                    301:
                    302:        pp->pr_entered_file = file;
                    303:        pp->pr_entered_line = line;
                    304: }
                    305:
1.101.2.1  yamt      306: static inline void
1.42      thorpej   307: pr_leave(struct pool *pp)
1.25      thorpej   308: {
                    309:
1.34      thorpej   310:        if (__predict_false(pp->pr_entered_file == NULL)) {
1.25      thorpej   311:                printf("pool %s not entered?\n", pp->pr_wchan);
                    312:                panic("pr_leave");
                    313:        }
                    314:
                    315:        pp->pr_entered_file = NULL;
                    316:        pp->pr_entered_line = 0;
                    317: }
                    318:
1.101.2.1  yamt      319: static inline void
1.42      thorpej   320: pr_enter_check(struct pool *pp, void (*pr)(const char *, ...))
1.25      thorpej   321: {
                    322:
                    323:        if (pp->pr_entered_file != NULL)
                    324:                (*pr)("\n\tcurrently entered from file %s line %ld\n",
                    325:                    pp->pr_entered_file, pp->pr_entered_line);
                    326: }
1.3       pk        327: #else
1.25      thorpej   328: #define        pr_log(pp, v, action, file, line)
                    329: #define        pr_printlog(pp, pi, pr)
                    330: #define        pr_enter(pp, file, line)
                    331: #define        pr_leave(pp)
                    332: #define        pr_enter_check(pp, pr)
1.59      thorpej   333: #endif /* POOL_DIAGNOSTIC */
1.3       pk        334:
1.101.2.6  yamt      335: static inline unsigned int
1.97      yamt      336: pr_item_notouch_index(const struct pool *pp, const struct pool_item_header *ph,
                    337:     const void *v)
                    338: {
                    339:        const char *cp = v;
1.101.2.6  yamt      340:        unsigned int idx;
1.97      yamt      341:
                    342:        KASSERT(pp->pr_roflags & PR_NOTOUCH);
1.101.2.4  yamt      343:        idx = (cp - (char *)ph->ph_page - ph->ph_off) / pp->pr_size;
1.97      yamt      344:        KASSERT(idx < pp->pr_itemsperpage);
                    345:        return idx;
                    346: }
                    347:
1.101.2.1  yamt      348: static inline void
1.97      yamt      349: pr_item_notouch_put(const struct pool *pp, struct pool_item_header *ph,
                    350:     void *obj)
                    351: {
1.101.2.6  yamt      352:        unsigned int idx = pr_item_notouch_index(pp, ph, obj);
                    353:        pool_item_bitmap_t *bitmap = ph->ph_bitmap + (idx / BITMAP_SIZE);
                    354:        pool_item_bitmap_t mask = 1 << (idx & BITMAP_MASK);
1.97      yamt      355:
1.101.2.6  yamt      356:        KASSERT((*bitmap & mask) == 0);
                    357:        *bitmap |= mask;
1.97      yamt      358: }
                    359:
1.101.2.1  yamt      360: static inline void *
1.97      yamt      361: pr_item_notouch_get(const struct pool *pp, struct pool_item_header *ph)
                    362: {
1.101.2.6  yamt      363:        pool_item_bitmap_t *bitmap = ph->ph_bitmap;
                    364:        unsigned int idx;
                    365:        int i;
                    366:
                    367:        for (i = 0; ; i++) {
                    368:                int bit;
                    369:
                    370:                KASSERT((i * BITMAP_SIZE) < pp->pr_itemsperpage);
                    371:                bit = ffs32(bitmap[i]);
                    372:                if (bit) {
                    373:                        pool_item_bitmap_t mask;
                    374:
                    375:                        bit--;
                    376:                        idx = (i * BITMAP_SIZE) + bit;
                    377:                        mask = 1 << bit;
                    378:                        KASSERT((bitmap[i] & mask) != 0);
                    379:                        bitmap[i] &= ~mask;
                    380:                        break;
                    381:                }
                    382:        }
                    383:        KASSERT(idx < pp->pr_itemsperpage);
                    384:        return (char *)ph->ph_page + ph->ph_off + idx * pp->pr_size;
                    385: }
1.97      yamt      386:
1.101.2.6  yamt      387: static inline void
                    388: pr_item_notouch_init(const struct pool *pp, struct pool_item_header *ph)
                    389: {
                    390:        pool_item_bitmap_t *bitmap = ph->ph_bitmap;
                    391:        const int n = howmany(pp->pr_itemsperpage, BITMAP_SIZE);
                    392:        int i;
1.97      yamt      393:
1.101.2.6  yamt      394:        for (i = 0; i < n; i++) {
                    395:                bitmap[i] = (pool_item_bitmap_t)-1;
                    396:        }
1.97      yamt      397: }
                    398:
1.101.2.1  yamt      399: static inline int
1.88      chs       400: phtree_compare(struct pool_item_header *a, struct pool_item_header *b)
                    401: {
1.101.2.2  yamt      402:
                    403:        /*
                    404:         * we consider pool_item_header with smaller ph_page bigger.
                    405:         * (this unnatural ordering is for the benefit of pr_find_pagehead.)
                    406:         */
                    407:
1.88      chs       408:        if (a->ph_page < b->ph_page)
                    409:                return (1);
1.101.2.2  yamt      410:        else if (a->ph_page > b->ph_page)
                    411:                return (-1);
1.88      chs       412:        else
                    413:                return (0);
                    414: }
                    415:
                    416: SPLAY_PROTOTYPE(phtree, pool_item_header, ph_node, phtree_compare);
                    417: SPLAY_GENERATE(phtree, pool_item_header, ph_node, phtree_compare);
                    418:
1.101.2.8  yamt      419: static inline struct pool_item_header *
                    420: pr_find_pagehead_noalign(struct pool *pp, void *v)
                    421: {
                    422:        struct pool_item_header *ph, tmp;
                    423:
                    424:        tmp.ph_page = (void *)(uintptr_t)v;
                    425:        ph = SPLAY_FIND(phtree, &pp->pr_phtree, &tmp);
                    426:        if (ph == NULL) {
                    427:                ph = SPLAY_ROOT(&pp->pr_phtree);
                    428:                if (ph != NULL && phtree_compare(&tmp, ph) >= 0) {
                    429:                        ph = SPLAY_NEXT(phtree, &pp->pr_phtree, ph);
                    430:                }
                    431:                KASSERT(ph == NULL || phtree_compare(&tmp, ph) < 0);
                    432:        }
                    433:
                    434:        return ph;
                    435: }
                    436:
1.3       pk        437: /*
1.101.2.2  yamt      438:  * Return the pool page header based on item address.
1.3       pk        439:  */
1.101.2.1  yamt      440: static inline struct pool_item_header *
1.101.2.2  yamt      441: pr_find_pagehead(struct pool *pp, void *v)
1.3       pk        442: {
1.88      chs       443:        struct pool_item_header *ph, tmp;
1.3       pk        444:
1.101.2.2  yamt      445:        if ((pp->pr_roflags & PR_NOALIGN) != 0) {
1.101.2.8  yamt      446:                ph = pr_find_pagehead_noalign(pp, v);
1.101.2.2  yamt      447:        } else {
1.101.2.4  yamt      448:                void *page =
                    449:                    (void *)((uintptr_t)v & pp->pr_alloc->pa_pagemask);
1.101.2.2  yamt      450:
                    451:                if ((pp->pr_roflags & PR_PHINPAGE) != 0) {
1.101.2.4  yamt      452:                        ph = (struct pool_item_header *)((char *)page + pp->pr_phoffset);
1.101.2.2  yamt      453:                } else {
                    454:                        tmp.ph_page = page;
                    455:                        ph = SPLAY_FIND(phtree, &pp->pr_phtree, &tmp);
                    456:                }
                    457:        }
1.3       pk        458:
1.101.2.2  yamt      459:        KASSERT(ph == NULL || ((pp->pr_roflags & PR_PHINPAGE) != 0) ||
1.101.2.4  yamt      460:            ((char *)ph->ph_page <= (char *)v &&
                    461:            (char *)v < (char *)ph->ph_page + pp->pr_alloc->pa_pagesz));
1.88      chs       462:        return ph;
1.3       pk        463: }
                    464:
1.101     thorpej   465: static void
                    466: pr_pagelist_free(struct pool *pp, struct pool_pagelist *pq)
                    467: {
                    468:        struct pool_item_header *ph;
                    469:
                    470:        while ((ph = LIST_FIRST(pq)) != NULL) {
                    471:                LIST_REMOVE(ph, ph_pagelist);
                    472:                pool_allocator_free(pp, ph->ph_page);
1.101.2.6  yamt      473:                if ((pp->pr_roflags & PR_PHINPAGE) == 0)
1.101     thorpej   474:                        pool_put(pp->pr_phpool, ph);
                    475:        }
                    476: }
                    477:
1.3       pk        478: /*
                    479:  * Remove a page from the pool.
                    480:  */
1.101.2.1  yamt      481: static inline void
1.61      chs       482: pr_rmpage(struct pool *pp, struct pool_item_header *ph,
                    483:      struct pool_pagelist *pq)
1.3       pk        484: {
                    485:
1.101.2.6  yamt      486:        KASSERT(mutex_owned(&pp->pr_lock));
1.91      yamt      487:
1.3       pk        488:        /*
1.7       thorpej   489:         * If the page was idle, decrement the idle page count.
1.3       pk        490:         */
1.6       thorpej   491:        if (ph->ph_nmissing == 0) {
                    492: #ifdef DIAGNOSTIC
                    493:                if (pp->pr_nidle == 0)
                    494:                        panic("pr_rmpage: nidle inconsistent");
1.20      thorpej   495:                if (pp->pr_nitems < pp->pr_itemsperpage)
                    496:                        panic("pr_rmpage: nitems inconsistent");
1.6       thorpej   497: #endif
                    498:                pp->pr_nidle--;
                    499:        }
1.7       thorpej   500:
1.20      thorpej   501:        pp->pr_nitems -= pp->pr_itemsperpage;
                    502:
1.7       thorpej   503:        /*
1.101     thorpej   504:         * Unlink the page from the pool and queue it for release.
1.7       thorpej   505:         */
1.88      chs       506:        LIST_REMOVE(ph, ph_pagelist);
1.91      yamt      507:        if ((pp->pr_roflags & PR_PHINPAGE) == 0)
                    508:                SPLAY_REMOVE(phtree, &pp->pr_phtree, ph);
1.101     thorpej   509:        LIST_INSERT_HEAD(pq, ph, ph_pagelist);
                    510:
1.7       thorpej   511:        pp->pr_npages--;
                    512:        pp->pr_npagefree++;
1.6       thorpej   513:
1.88      chs       514:        pool_update_curpage(pp);
1.3       pk        515: }
                    516:
1.101.2.3  yamt      517: static bool
1.101.2.1  yamt      518: pa_starved_p(struct pool_allocator *pa)
                    519: {
                    520:
                    521:        if (pa->pa_backingmap != NULL) {
                    522:                return vm_map_starved_p(pa->pa_backingmap);
                    523:        }
1.101.2.3  yamt      524:        return false;
1.101.2.1  yamt      525: }
                    526:
                    527: static int
                    528: pool_reclaim_callback(struct callback_entry *ce, void *obj, void *arg)
                    529: {
                    530:        struct pool *pp = obj;
                    531:        struct pool_allocator *pa = pp->pr_alloc;
                    532:
                    533:        KASSERT(&pp->pr_reclaimerentry == ce);
                    534:        pool_reclaim(pp);
                    535:        if (!pa_starved_p(pa)) {
                    536:                return CALLBACK_CHAIN_ABORT;
                    537:        }
                    538:        return CALLBACK_CHAIN_CONTINUE;
                    539: }
                    540:
                    541: static void
                    542: pool_reclaim_register(struct pool *pp)
                    543: {
                    544:        struct vm_map *map = pp->pr_alloc->pa_backingmap;
                    545:        int s;
                    546:
                    547:        if (map == NULL) {
                    548:                return;
                    549:        }
                    550:
                    551:        s = splvm(); /* not necessary for INTRSAFE maps, but don't care. */
                    552:        callback_register(&vm_map_to_kernel(map)->vmk_reclaim_callback,
                    553:            &pp->pr_reclaimerentry, pp, pool_reclaim_callback);
                    554:        splx(s);
                    555: }
                    556:
                    557: static void
                    558: pool_reclaim_unregister(struct pool *pp)
                    559: {
                    560:        struct vm_map *map = pp->pr_alloc->pa_backingmap;
                    561:        int s;
                    562:
                    563:        if (map == NULL) {
                    564:                return;
                    565:        }
                    566:
                    567:        s = splvm(); /* not necessary for INTRSAFE maps, but don't care. */
                    568:        callback_unregister(&vm_map_to_kernel(map)->vmk_reclaim_callback,
                    569:            &pp->pr_reclaimerentry);
                    570:        splx(s);
                    571: }
                    572:
                    573: static void
                    574: pa_reclaim_register(struct pool_allocator *pa)
                    575: {
                    576:        struct vm_map *map = *pa->pa_backingmapptr;
                    577:        struct pool *pp;
                    578:
                    579:        KASSERT(pa->pa_backingmap == NULL);
                    580:        if (map == NULL) {
                    581:                SLIST_INSERT_HEAD(&pa_deferinitq, pa, pa_q);
                    582:                return;
                    583:        }
                    584:        pa->pa_backingmap = map;
                    585:        TAILQ_FOREACH(pp, &pa->pa_list, pr_alloc_list) {
                    586:                pool_reclaim_register(pp);
                    587:        }
                    588: }
                    589:
1.3       pk        590: /*
1.94      simonb    591:  * Initialize all the pools listed in the "pools" link set.
                    592:  */
                    593: void
1.101.2.1  yamt      594: pool_subsystem_init(void)
1.94      simonb    595: {
1.101.2.1  yamt      596:        struct pool_allocator *pa;
1.94      simonb    597:        __link_set_decl(pools, struct link_pool_init);
                    598:        struct link_pool_init * const *pi;
                    599:
1.101.2.6  yamt      600:        mutex_init(&pool_head_lock, MUTEX_DEFAULT, IPL_NONE);
                    601:        cv_init(&pool_busy, "poolbusy");
                    602:
1.94      simonb    603:        __link_set_foreach(pi, pools)
                    604:                pool_init((*pi)->pp, (*pi)->size, (*pi)->align,
                    605:                    (*pi)->align_offset, (*pi)->flags, (*pi)->wchan,
1.101.2.4  yamt      606:                    (*pi)->palloc, (*pi)->ipl);
1.101.2.1  yamt      607:
                    608:        while ((pa = SLIST_FIRST(&pa_deferinitq)) != NULL) {
                    609:                KASSERT(pa->pa_backingmapptr != NULL);
                    610:                KASSERT(*pa->pa_backingmapptr != NULL);
                    611:                SLIST_REMOVE_HEAD(&pa_deferinitq, pa_q);
                    612:                pa_reclaim_register(pa);
                    613:        }
1.101.2.6  yamt      614:
                    615:        pool_init(&cache_pool, sizeof(struct pool_cache), CACHE_LINE_SIZE,
                    616:            0, 0, "pcache", &pool_allocator_nointr, IPL_NONE);
                    617:
                    618:        pool_init(&cache_cpu_pool, sizeof(pool_cache_cpu_t), CACHE_LINE_SIZE,
                    619:            0, 0, "pcachecpu", &pool_allocator_nointr, IPL_NONE);
1.94      simonb    620: }
                    621:
                    622: /*
1.3       pk        623:  * Initialize the given pool resource structure.
                    624:  *
                    625:  * We export this routine to allow other kernel parts to declare
                    626:  * static pools that must be initialized before malloc() is available.
                    627:  */
                    628: void
1.42      thorpej   629: pool_init(struct pool *pp, size_t size, u_int align, u_int ioff, int flags,
1.101.2.4  yamt      630:     const char *wchan, struct pool_allocator *palloc, int ipl)
1.3       pk        631: {
1.101.2.1  yamt      632:        struct pool *pp1;
1.92      enami     633:        size_t trysize, phsize;
1.101.2.6  yamt      634:        int off, slack;
1.99      yamt      635:
1.101.2.1  yamt      636: #ifdef DEBUG
                    637:        /*
                    638:         * Check that the pool hasn't already been initialised and
                    639:         * added to the list of all pools.
                    640:         */
1.101.2.8  yamt      641:        TAILQ_FOREACH(pp1, &pool_head, pr_poollist) {
1.101.2.1  yamt      642:                if (pp == pp1)
                    643:                        panic("pool_init: pool %s already initialised",
                    644:                            wchan);
                    645:        }
                    646: #endif
                    647:
1.25      thorpej   648: #ifdef POOL_DIAGNOSTIC
                    649:        /*
                    650:         * Always log if POOL_DIAGNOSTIC is defined.
                    651:         */
                    652:        if (pool_logsize != 0)
                    653:                flags |= PR_LOGGING;
                    654: #endif
                    655:
1.66      thorpej   656:        if (palloc == NULL)
                    657:                palloc = &pool_allocator_kmem;
1.101.2.1  yamt      658: #ifdef POOL_SUBPAGE
                    659:        if (size > palloc->pa_pagesz) {
                    660:                if (palloc == &pool_allocator_kmem)
                    661:                        palloc = &pool_allocator_kmem_fullpage;
                    662:                else if (palloc == &pool_allocator_nointr)
                    663:                        palloc = &pool_allocator_nointr_fullpage;
                    664:        }
1.66      thorpej   665: #endif /* POOL_SUBPAGE */
                    666:        if ((palloc->pa_flags & PA_INITIALIZED) == 0) {
1.101.2.1  yamt      667:                if (palloc->pa_pagesz == 0)
1.66      thorpej   668:                        palloc->pa_pagesz = PAGE_SIZE;
                    669:
                    670:                TAILQ_INIT(&palloc->pa_list);
                    671:
1.101.2.6  yamt      672:                mutex_init(&palloc->pa_lock, MUTEX_DEFAULT, IPL_VM);
1.66      thorpej   673:                palloc->pa_pagemask = ~(palloc->pa_pagesz - 1);
                    674:                palloc->pa_pageshift = ffs(palloc->pa_pagesz) - 1;
1.101.2.1  yamt      675:
                    676:                if (palloc->pa_backingmapptr != NULL) {
                    677:                        pa_reclaim_register(palloc);
                    678:                }
1.66      thorpej   679:                palloc->pa_flags |= PA_INITIALIZED;
1.4       thorpej   680:        }
1.3       pk        681:
                    682:        if (align == 0)
                    683:                align = ALIGN(1);
1.14      thorpej   684:
1.101.2.2  yamt      685:        if ((flags & PR_NOTOUCH) == 0 && size < sizeof(struct pool_item))
1.14      thorpej   686:                size = sizeof(struct pool_item);
1.3       pk        687:
1.78      thorpej   688:        size = roundup(size, align);
1.66      thorpej   689: #ifdef DIAGNOSTIC
                    690:        if (size > palloc->pa_pagesz)
1.101.2.2  yamt      691:                panic("pool_init: pool item size (%zu) too large", size);
1.66      thorpej   692: #endif
1.35      pk        693:
1.3       pk        694:        /*
                    695:         * Initialize the pool structure.
                    696:         */
1.88      chs       697:        LIST_INIT(&pp->pr_emptypages);
                    698:        LIST_INIT(&pp->pr_fullpages);
                    699:        LIST_INIT(&pp->pr_partpages);
1.101.2.6  yamt      700:        pp->pr_cache = NULL;
1.3       pk        701:        pp->pr_curpage = NULL;
                    702:        pp->pr_npages = 0;
                    703:        pp->pr_minitems = 0;
                    704:        pp->pr_minpages = 0;
                    705:        pp->pr_maxpages = UINT_MAX;
1.20      thorpej   706:        pp->pr_roflags = flags;
                    707:        pp->pr_flags = 0;
1.35      pk        708:        pp->pr_size = size;
1.3       pk        709:        pp->pr_align = align;
                    710:        pp->pr_wchan = wchan;
1.66      thorpej   711:        pp->pr_alloc = palloc;
1.20      thorpej   712:        pp->pr_nitems = 0;
                    713:        pp->pr_nout = 0;
                    714:        pp->pr_hardlimit = UINT_MAX;
                    715:        pp->pr_hardlimit_warning = NULL;
1.31      thorpej   716:        pp->pr_hardlimit_ratecap.tv_sec = 0;
                    717:        pp->pr_hardlimit_ratecap.tv_usec = 0;
                    718:        pp->pr_hardlimit_warning_last.tv_sec = 0;
                    719:        pp->pr_hardlimit_warning_last.tv_usec = 0;
1.68      thorpej   720:        pp->pr_drain_hook = NULL;
                    721:        pp->pr_drain_hook_arg = NULL;
1.101.2.3  yamt      722:        pp->pr_freecheck = NULL;
1.3       pk        723:
                    724:        /*
                    725:         * Decide whether to put the page header off page to avoid
1.92      enami     726:         * wasting too large a part of the page or too big item.
                    727:         * Off-page page headers go on a hash table, so we can match
                    728:         * a returned item with its header based on the page address.
                    729:         * We use 1/16 of the page size and about 8 times of the item
                    730:         * size as the threshold (XXX: tune)
                    731:         *
                    732:         * However, we'll put the header into the page if we can put
                    733:         * it without wasting any items.
                    734:         *
                    735:         * Silently enforce `0 <= ioff < align'.
1.3       pk        736:         */
1.92      enami     737:        pp->pr_itemoffset = ioff %= align;
                    738:        /* See the comment below about reserved bytes. */
                    739:        trysize = palloc->pa_pagesz - ((align - ioff) % align);
                    740:        phsize = ALIGN(sizeof(struct pool_item_header));
1.101.2.2  yamt      741:        if ((pp->pr_roflags & (PR_NOTOUCH | PR_NOALIGN)) == 0 &&
1.97      yamt      742:            (pp->pr_size < MIN(palloc->pa_pagesz / 16, phsize << 3) ||
                    743:            trysize / pp->pr_size == (trysize - phsize) / pp->pr_size)) {
1.3       pk        744:                /* Use the end of the page for the page header */
1.20      thorpej   745:                pp->pr_roflags |= PR_PHINPAGE;
1.92      enami     746:                pp->pr_phoffset = off = palloc->pa_pagesz - phsize;
1.2       pk        747:        } else {
1.3       pk        748:                /* The page header will be taken from our page header pool */
                    749:                pp->pr_phoffset = 0;
1.66      thorpej   750:                off = palloc->pa_pagesz;
1.88      chs       751:                SPLAY_INIT(&pp->pr_phtree);
1.2       pk        752:        }
1.1       pk        753:
1.3       pk        754:        /*
                    755:         * Alignment is to take place at `ioff' within the item. This means
                    756:         * we must reserve up to `align - 1' bytes on the page to allow
                    757:         * appropriate positioning of each item.
                    758:         */
                    759:        pp->pr_itemsperpage = (off - ((align - ioff) % align)) / pp->pr_size;
1.43      thorpej   760:        KASSERT(pp->pr_itemsperpage != 0);
1.97      yamt      761:        if ((pp->pr_roflags & PR_NOTOUCH)) {
                    762:                int idx;
                    763:
                    764:                for (idx = 0; pp->pr_itemsperpage > PHPOOL_FREELIST_NELEM(idx);
                    765:                    idx++) {
                    766:                        /* nothing */
                    767:                }
                    768:                if (idx >= PHPOOL_MAX) {
                    769:                        /*
                    770:                         * if you see this panic, consider to tweak
                    771:                         * PHPOOL_MAX and PHPOOL_FREELIST_NELEM.
                    772:                         */
                    773:                        panic("%s: too large itemsperpage(%d) for PR_NOTOUCH",
                    774:                            pp->pr_wchan, pp->pr_itemsperpage);
                    775:                }
                    776:                pp->pr_phpool = &phpool[idx];
                    777:        } else if ((pp->pr_roflags & PR_PHINPAGE) == 0) {
                    778:                pp->pr_phpool = &phpool[0];
                    779:        }
                    780: #if defined(DIAGNOSTIC)
                    781:        else {
                    782:                pp->pr_phpool = NULL;
                    783:        }
                    784: #endif
1.3       pk        785:
                    786:        /*
                    787:         * Use the slack between the chunks and the page header
                    788:         * for "cache coloring".
                    789:         */
                    790:        slack = off - pp->pr_itemsperpage * pp->pr_size;
                    791:        pp->pr_maxcolor = (slack / align) * align;
                    792:        pp->pr_curcolor = 0;
                    793:
                    794:        pp->pr_nget = 0;
                    795:        pp->pr_nfail = 0;
                    796:        pp->pr_nput = 0;
                    797:        pp->pr_npagealloc = 0;
                    798:        pp->pr_npagefree = 0;
1.1       pk        799:        pp->pr_hiwat = 0;
1.8       thorpej   800:        pp->pr_nidle = 0;
1.101.2.6  yamt      801:        pp->pr_refcnt = 0;
1.3       pk        802:
1.59      thorpej   803: #ifdef POOL_DIAGNOSTIC
1.25      thorpej   804:        if (flags & PR_LOGGING) {
                    805:                if (kmem_map == NULL ||
                    806:                    (pp->pr_log = malloc(pool_logsize * sizeof(struct pool_log),
                    807:                     M_TEMP, M_NOWAIT)) == NULL)
1.20      thorpej   808:                        pp->pr_roflags &= ~PR_LOGGING;
1.3       pk        809:                pp->pr_curlogentry = 0;
                    810:                pp->pr_logsize = pool_logsize;
                    811:        }
1.59      thorpej   812: #endif
1.25      thorpej   813:
                    814:        pp->pr_entered_file = NULL;
                    815:        pp->pr_entered_line = 0;
1.3       pk        816:
1.101.2.7  yamt      817:        /*
                    818:         * XXXAD hack to prevent IP input processing from blocking.
                    819:         */
                    820:        if (ipl == IPL_SOFTNET) {
                    821:                mutex_init(&pp->pr_lock, MUTEX_DEFAULT, IPL_VM);
                    822:        } else {
                    823:                mutex_init(&pp->pr_lock, MUTEX_DEFAULT, ipl);
                    824:        }
1.101.2.6  yamt      825:        cv_init(&pp->pr_cv, wchan);
                    826:        pp->pr_ipl = ipl;
1.1       pk        827:
1.3       pk        828:        /*
1.43      thorpej   829:         * Initialize private page header pool and cache magazine pool if we
                    830:         * haven't done so yet.
1.23      thorpej   831:         * XXX LOCKING.
1.3       pk        832:         */
1.97      yamt      833:        if (phpool[0].pr_size == 0) {
                    834:                int idx;
                    835:                for (idx = 0; idx < PHPOOL_MAX; idx++) {
                    836:                        static char phpool_names[PHPOOL_MAX][6+1+6+1];
                    837:                        int nelem;
                    838:                        size_t sz;
                    839:
                    840:                        nelem = PHPOOL_FREELIST_NELEM(idx);
                    841:                        snprintf(phpool_names[idx], sizeof(phpool_names[idx]),
                    842:                            "phpool-%d", nelem);
                    843:                        sz = sizeof(struct pool_item_header);
                    844:                        if (nelem) {
1.101.2.6  yamt      845:                                sz = offsetof(struct pool_item_header,
                    846:                                    ph_bitmap[howmany(nelem, BITMAP_SIZE)]);
1.97      yamt      847:                        }
                    848:                        pool_init(&phpool[idx], sz, 0, 0, 0,
1.101.2.4  yamt      849:                            phpool_names[idx], &pool_allocator_meta, IPL_VM);
1.97      yamt      850:                }
1.62      bjh21     851: #ifdef POOL_SUBPAGE
                    852:                pool_init(&psppool, POOL_SUBPAGE, POOL_SUBPAGE, 0,
1.101.2.4  yamt      853:                    PR_RECURSIVE, "psppool", &pool_allocator_meta, IPL_VM);
1.62      bjh21     854: #endif
1.101.2.8  yamt      855:
                    856:                size = sizeof(pcg_t) +
                    857:                    (PCG_NOBJECTS_NORMAL - 1) * sizeof(pcgpair_t);
                    858:                pool_init(&pcg_normal_pool, size, CACHE_LINE_SIZE, 0, 0,
                    859:                    "pcgnormal", &pool_allocator_meta, IPL_VM);
                    860:
                    861:                size = sizeof(pcg_t) +
                    862:                    (PCG_NOBJECTS_LARGE - 1) * sizeof(pcgpair_t);
                    863:                pool_init(&pcg_large_pool, size, CACHE_LINE_SIZE, 0, 0,
                    864:                    "pcglarge", &pool_allocator_meta, IPL_VM);
1.1       pk        865:        }
                    866:
1.101.2.8  yamt      867:        /* Insert into the list of all pools. */
                    868:        if (__predict_true(!cold))
1.101.2.6  yamt      869:                mutex_enter(&pool_head_lock);
1.101.2.8  yamt      870:        TAILQ_FOREACH(pp1, &pool_head, pr_poollist) {
                    871:                if (strcmp(pp1->pr_wchan, pp->pr_wchan) > 0)
                    872:                        break;
                    873:        }
                    874:        if (pp1 == NULL)
                    875:                TAILQ_INSERT_TAIL(&pool_head, pp, pr_poollist);
                    876:        else
                    877:                TAILQ_INSERT_BEFORE(pp1, pp, pr_poollist);
                    878:        if (__predict_true(!cold))
1.101.2.6  yamt      879:                mutex_exit(&pool_head_lock);
                    880:
                    881:                /* Insert this into the list of pools using this allocator. */
1.101.2.8  yamt      882:        if (__predict_true(!cold))
1.101.2.6  yamt      883:                mutex_enter(&palloc->pa_lock);
1.101.2.8  yamt      884:        TAILQ_INSERT_TAIL(&palloc->pa_list, pp, pr_alloc_list);
                    885:        if (__predict_true(!cold))
1.101.2.6  yamt      886:                mutex_exit(&palloc->pa_lock);
1.66      thorpej   887:
1.101.2.1  yamt      888:        pool_reclaim_register(pp);
1.1       pk        889: }
                    890:
                    891: /*
                    892:  * De-commision a pool resource.
                    893:  */
                    894: void
1.42      thorpej   895: pool_destroy(struct pool *pp)
1.1       pk        896: {
1.101     thorpej   897:        struct pool_pagelist pq;
1.3       pk        898:        struct pool_item_header *ph;
1.43      thorpej   899:
1.101     thorpej   900:        /* Remove from global pool list */
1.101.2.6  yamt      901:        mutex_enter(&pool_head_lock);
                    902:        while (pp->pr_refcnt != 0)
                    903:                cv_wait(&pool_busy, &pool_head_lock);
1.101.2.8  yamt      904:        TAILQ_REMOVE(&pool_head, pp, pr_poollist);
1.101     thorpej   905:        if (drainpp == pp)
                    906:                drainpp = NULL;
1.101.2.6  yamt      907:        mutex_exit(&pool_head_lock);
1.101     thorpej   908:
                    909:        /* Remove this pool from its allocator's list of pools. */
1.101.2.1  yamt      910:        pool_reclaim_unregister(pp);
1.101.2.6  yamt      911:        mutex_enter(&pp->pr_alloc->pa_lock);
1.66      thorpej   912:        TAILQ_REMOVE(&pp->pr_alloc->pa_list, pp, pr_alloc_list);
1.101.2.6  yamt      913:        mutex_exit(&pp->pr_alloc->pa_lock);
1.66      thorpej   914:
1.101.2.6  yamt      915:        mutex_enter(&pp->pr_lock);
1.101     thorpej   916:
1.101.2.6  yamt      917:        KASSERT(pp->pr_cache == NULL);
1.3       pk        918:
                    919: #ifdef DIAGNOSTIC
1.20      thorpej   920:        if (pp->pr_nout != 0) {
1.25      thorpej   921:                pr_printlog(pp, NULL, printf);
1.80      provos    922:                panic("pool_destroy: pool busy: still out: %u",
1.20      thorpej   923:                    pp->pr_nout);
1.3       pk        924:        }
                    925: #endif
1.1       pk        926:
1.101     thorpej   927:        KASSERT(LIST_EMPTY(&pp->pr_fullpages));
                    928:        KASSERT(LIST_EMPTY(&pp->pr_partpages));
                    929:
1.3       pk        930:        /* Remove all pages */
1.101     thorpej   931:        LIST_INIT(&pq);
1.88      chs       932:        while ((ph = LIST_FIRST(&pp->pr_emptypages)) != NULL)
1.101     thorpej   933:                pr_rmpage(pp, ph, &pq);
                    934:
1.101.2.6  yamt      935:        mutex_exit(&pp->pr_lock);
1.3       pk        936:
1.101     thorpej   937:        pr_pagelist_free(pp, &pq);
1.3       pk        938:
1.59      thorpej   939: #ifdef POOL_DIAGNOSTIC
1.20      thorpej   940:        if ((pp->pr_roflags & PR_LOGGING) != 0)
1.3       pk        941:                free(pp->pr_log, M_TEMP);
1.59      thorpej   942: #endif
1.101.2.6  yamt      943:
                    944:        cv_destroy(&pp->pr_cv);
                    945:        mutex_destroy(&pp->pr_lock);
1.1       pk        946: }
                    947:
1.68      thorpej   948: void
                    949: pool_set_drain_hook(struct pool *pp, void (*fn)(void *, int), void *arg)
                    950: {
                    951:
                    952:        /* XXX no locking -- must be used just after pool_init() */
                    953: #ifdef DIAGNOSTIC
                    954:        if (pp->pr_drain_hook != NULL)
                    955:                panic("pool_set_drain_hook(%s): already set", pp->pr_wchan);
                    956: #endif
                    957:        pp->pr_drain_hook = fn;
                    958:        pp->pr_drain_hook_arg = arg;
                    959: }
                    960:
1.88      chs       961: static struct pool_item_header *
1.101.2.4  yamt      962: pool_alloc_item_header(struct pool *pp, void *storage, int flags)
1.55      thorpej   963: {
                    964:        struct pool_item_header *ph;
                    965:
                    966:        if ((pp->pr_roflags & PR_PHINPAGE) != 0)
1.101.2.4  yamt      967:                ph = (struct pool_item_header *) ((char *)storage + pp->pr_phoffset);
1.101.2.6  yamt      968:        else
1.97      yamt      969:                ph = pool_get(pp->pr_phpool, flags);
1.55      thorpej   970:
                    971:        return (ph);
                    972: }
1.1       pk        973:
                    974: /*
1.101.2.6  yamt      975:  * Grab an item from the pool.
1.1       pk        976:  */
1.3       pk        977: void *
1.59      thorpej   978: #ifdef POOL_DIAGNOSTIC
1.42      thorpej   979: _pool_get(struct pool *pp, int flags, const char *file, long line)
1.56      sommerfe  980: #else
                    981: pool_get(struct pool *pp, int flags)
                    982: #endif
1.1       pk        983: {
                    984:        struct pool_item *pi;
1.3       pk        985:        struct pool_item_header *ph;
1.55      thorpej   986:        void *v;
1.1       pk        987:
1.2       pk        988: #ifdef DIAGNOSTIC
1.95      atatat    989:        if (__predict_false(pp->pr_itemsperpage == 0))
                    990:                panic("pool_get: pool %p: pr_itemsperpage is zero, "
                    991:                    "pool not initialized?", pp);
1.84      thorpej   992:        if (__predict_false(curlwp == NULL && doing_shutdown == 0 &&
1.37      sommerfe  993:                            (flags & PR_WAITOK) != 0))
1.77      matt      994:                panic("pool_get: %s: must have NOWAIT", pp->pr_wchan);
1.58      thorpej   995:
1.101.2.1  yamt      996: #endif /* DIAGNOSTIC */
1.58      thorpej   997: #ifdef LOCKDEBUG
                    998:        if (flags & PR_WAITOK)
1.101.2.2  yamt      999:                ASSERT_SLEEPABLE(NULL, "pool_get(PR_WAITOK)");
1.56      sommerfe 1000: #endif
1.1       pk       1001:
1.101.2.6  yamt     1002:        mutex_enter(&pp->pr_lock);
1.25      thorpej  1003:        pr_enter(pp, file, line);
1.20      thorpej  1004:
                   1005:  startover:
                   1006:        /*
                   1007:         * Check to see if we've reached the hard limit.  If we have,
                   1008:         * and we can wait, then wait until an item has been returned to
                   1009:         * the pool.
                   1010:         */
                   1011: #ifdef DIAGNOSTIC
1.34      thorpej  1012:        if (__predict_false(pp->pr_nout > pp->pr_hardlimit)) {
1.25      thorpej  1013:                pr_leave(pp);
1.101.2.6  yamt     1014:                mutex_exit(&pp->pr_lock);
1.20      thorpej  1015:                panic("pool_get: %s: crossed hard limit", pp->pr_wchan);
                   1016:        }
                   1017: #endif
1.34      thorpej  1018:        if (__predict_false(pp->pr_nout == pp->pr_hardlimit)) {
1.68      thorpej  1019:                if (pp->pr_drain_hook != NULL) {
                   1020:                        /*
                   1021:                         * Since the drain hook is going to free things
                   1022:                         * back to the pool, unlock, call the hook, re-lock,
                   1023:                         * and check the hardlimit condition again.
                   1024:                         */
                   1025:                        pr_leave(pp);
1.101.2.6  yamt     1026:                        mutex_exit(&pp->pr_lock);
1.68      thorpej  1027:                        (*pp->pr_drain_hook)(pp->pr_drain_hook_arg, flags);
1.101.2.6  yamt     1028:                        mutex_enter(&pp->pr_lock);
1.68      thorpej  1029:                        pr_enter(pp, file, line);
                   1030:                        if (pp->pr_nout < pp->pr_hardlimit)
                   1031:                                goto startover;
                   1032:                }
                   1033:
1.29      sommerfe 1034:                if ((flags & PR_WAITOK) && !(flags & PR_LIMITFAIL)) {
1.20      thorpej  1035:                        /*
                   1036:                         * XXX: A warning isn't logged in this case.  Should
                   1037:                         * it be?
                   1038:                         */
                   1039:                        pp->pr_flags |= PR_WANTED;
1.25      thorpej  1040:                        pr_leave(pp);
1.101.2.6  yamt     1041:                        cv_wait(&pp->pr_cv, &pp->pr_lock);
1.25      thorpej  1042:                        pr_enter(pp, file, line);
1.20      thorpej  1043:                        goto startover;
                   1044:                }
1.31      thorpej  1045:
                   1046:                /*
                   1047:                 * Log a message that the hard limit has been hit.
                   1048:                 */
                   1049:                if (pp->pr_hardlimit_warning != NULL &&
                   1050:                    ratecheck(&pp->pr_hardlimit_warning_last,
                   1051:                              &pp->pr_hardlimit_ratecap))
                   1052:                        log(LOG_ERR, "%s\n", pp->pr_hardlimit_warning);
1.21      thorpej  1053:
                   1054:                pp->pr_nfail++;
                   1055:
1.25      thorpej  1056:                pr_leave(pp);
1.101.2.6  yamt     1057:                mutex_exit(&pp->pr_lock);
1.20      thorpej  1058:                return (NULL);
                   1059:        }
                   1060:
1.3       pk       1061:        /*
                   1062:         * The convention we use is that if `curpage' is not NULL, then
                   1063:         * it points at a non-empty bucket. In particular, `curpage'
                   1064:         * never points at a page header which has PR_PHINPAGE set and
                   1065:         * has no items in its bucket.
                   1066:         */
1.20      thorpej  1067:        if ((ph = pp->pr_curpage) == NULL) {
1.101.2.1  yamt     1068:                int error;
                   1069:
1.20      thorpej  1070: #ifdef DIAGNOSTIC
                   1071:                if (pp->pr_nitems != 0) {
1.101.2.6  yamt     1072:                        mutex_exit(&pp->pr_lock);
1.20      thorpej  1073:                        printf("pool_get: %s: curpage NULL, nitems %u\n",
                   1074:                            pp->pr_wchan, pp->pr_nitems);
1.80      provos   1075:                        panic("pool_get: nitems inconsistent");
1.20      thorpej  1076:                }
                   1077: #endif
                   1078:
1.21      thorpej  1079:                /*
                   1080:                 * Call the back-end page allocator for more memory.
                   1081:                 * Release the pool lock, as the back-end page allocator
                   1082:                 * may block.
                   1083:                 */
1.25      thorpej  1084:                pr_leave(pp);
1.101.2.1  yamt     1085:                error = pool_grow(pp, flags);
                   1086:                pr_enter(pp, file, line);
                   1087:                if (error != 0) {
1.21      thorpej  1088:                        /*
1.55      thorpej  1089:                         * We were unable to allocate a page or item
                   1090:                         * header, but we released the lock during
                   1091:                         * allocation, so perhaps items were freed
                   1092:                         * back to the pool.  Check for this case.
1.21      thorpej  1093:                         */
                   1094:                        if (pp->pr_curpage != NULL)
                   1095:                                goto startover;
1.15      pk       1096:
1.101.2.1  yamt     1097:                        pp->pr_nfail++;
1.25      thorpej  1098:                        pr_leave(pp);
1.101.2.6  yamt     1099:                        mutex_exit(&pp->pr_lock);
1.101.2.1  yamt     1100:                        return (NULL);
1.1       pk       1101:                }
1.3       pk       1102:
1.20      thorpej  1103:                /* Start the allocation process over. */
                   1104:                goto startover;
1.3       pk       1105:        }
1.97      yamt     1106:        if (pp->pr_roflags & PR_NOTOUCH) {
                   1107: #ifdef DIAGNOSTIC
                   1108:                if (__predict_false(ph->ph_nmissing == pp->pr_itemsperpage)) {
                   1109:                        pr_leave(pp);
1.101.2.6  yamt     1110:                        mutex_exit(&pp->pr_lock);
1.97      yamt     1111:                        panic("pool_get: %s: page empty", pp->pr_wchan);
                   1112:                }
                   1113: #endif
                   1114:                v = pr_item_notouch_get(pp, ph);
                   1115: #ifdef POOL_DIAGNOSTIC
                   1116:                pr_log(pp, v, PRLOG_GET, file, line);
                   1117: #endif
                   1118:        } else {
1.101.2.1  yamt     1119:                v = pi = LIST_FIRST(&ph->ph_itemlist);
1.97      yamt     1120:                if (__predict_false(v == NULL)) {
                   1121:                        pr_leave(pp);
1.101.2.6  yamt     1122:                        mutex_exit(&pp->pr_lock);
1.97      yamt     1123:                        panic("pool_get: %s: page empty", pp->pr_wchan);
                   1124:                }
1.20      thorpej  1125: #ifdef DIAGNOSTIC
1.97      yamt     1126:                if (__predict_false(pp->pr_nitems == 0)) {
                   1127:                        pr_leave(pp);
1.101.2.6  yamt     1128:                        mutex_exit(&pp->pr_lock);
1.97      yamt     1129:                        printf("pool_get: %s: items on itemlist, nitems %u\n",
                   1130:                            pp->pr_wchan, pp->pr_nitems);
                   1131:                        panic("pool_get: nitems inconsistent");
                   1132:                }
1.65      enami    1133: #endif
1.56      sommerfe 1134:
1.65      enami    1135: #ifdef POOL_DIAGNOSTIC
1.97      yamt     1136:                pr_log(pp, v, PRLOG_GET, file, line);
1.65      enami    1137: #endif
1.3       pk       1138:
1.65      enami    1139: #ifdef DIAGNOSTIC
1.97      yamt     1140:                if (__predict_false(pi->pi_magic != PI_MAGIC)) {
                   1141:                        pr_printlog(pp, pi, printf);
                   1142:                        panic("pool_get(%s): free list modified: "
                   1143:                            "magic=%x; page %p; item addr %p\n",
                   1144:                            pp->pr_wchan, pi->pi_magic, ph->ph_page, pi);
                   1145:                }
1.3       pk       1146: #endif
                   1147:
1.97      yamt     1148:                /*
                   1149:                 * Remove from item list.
                   1150:                 */
1.101.2.1  yamt     1151:                LIST_REMOVE(pi, pi_list);
1.97      yamt     1152:        }
1.20      thorpej  1153:        pp->pr_nitems--;
                   1154:        pp->pr_nout++;
1.6       thorpej  1155:        if (ph->ph_nmissing == 0) {
                   1156: #ifdef DIAGNOSTIC
1.34      thorpej  1157:                if (__predict_false(pp->pr_nidle == 0))
1.6       thorpej  1158:                        panic("pool_get: nidle inconsistent");
                   1159: #endif
                   1160:                pp->pr_nidle--;
1.88      chs      1161:
                   1162:                /*
                   1163:                 * This page was previously empty.  Move it to the list of
                   1164:                 * partially-full pages.  This page is already curpage.
                   1165:                 */
                   1166:                LIST_REMOVE(ph, ph_pagelist);
                   1167:                LIST_INSERT_HEAD(&pp->pr_partpages, ph, ph_pagelist);
1.6       thorpej  1168:        }
1.3       pk       1169:        ph->ph_nmissing++;
1.97      yamt     1170:        if (ph->ph_nmissing == pp->pr_itemsperpage) {
1.21      thorpej  1171: #ifdef DIAGNOSTIC
1.97      yamt     1172:                if (__predict_false((pp->pr_roflags & PR_NOTOUCH) == 0 &&
1.101.2.1  yamt     1173:                    !LIST_EMPTY(&ph->ph_itemlist))) {
1.25      thorpej  1174:                        pr_leave(pp);
1.101.2.6  yamt     1175:                        mutex_exit(&pp->pr_lock);
1.21      thorpej  1176:                        panic("pool_get: %s: nmissing inconsistent",
                   1177:                            pp->pr_wchan);
                   1178:                }
                   1179: #endif
1.3       pk       1180:                /*
1.88      chs      1181:                 * This page is now full.  Move it to the full list
                   1182:                 * and select a new current page.
1.3       pk       1183:                 */
1.88      chs      1184:                LIST_REMOVE(ph, ph_pagelist);
                   1185:                LIST_INSERT_HEAD(&pp->pr_fullpages, ph, ph_pagelist);
                   1186:                pool_update_curpage(pp);
1.1       pk       1187:        }
1.3       pk       1188:
                   1189:        pp->pr_nget++;
1.101.2.1  yamt     1190:        pr_leave(pp);
1.20      thorpej  1191:
                   1192:        /*
                   1193:         * If we have a low water mark and we are now below that low
                   1194:         * water mark, add more items to the pool.
                   1195:         */
1.53      thorpej  1196:        if (POOL_NEEDS_CATCHUP(pp) && pool_catchup(pp) != 0) {
1.20      thorpej  1197:                /*
                   1198:                 * XXX: Should we log a warning?  Should we set up a timeout
                   1199:                 * to try again in a second or so?  The latter could break
                   1200:                 * a caller's assumptions about interrupt protection, etc.
                   1201:                 */
                   1202:        }
                   1203:
1.101.2.6  yamt     1204:        mutex_exit(&pp->pr_lock);
1.101.2.3  yamt     1205:        KASSERT((((vaddr_t)v + pp->pr_itemoffset) & (pp->pr_align - 1)) == 0);
                   1206:        FREECHECK_OUT(&pp->pr_freecheck, v);
1.1       pk       1207:        return (v);
                   1208: }
                   1209:
                   1210: /*
1.43      thorpej  1211:  * Internal version of pool_put().  Pool is already locked/entered.
1.1       pk       1212:  */
1.43      thorpej  1213: static void
1.101     thorpej  1214: pool_do_put(struct pool *pp, void *v, struct pool_pagelist *pq)
1.1       pk       1215: {
                   1216:        struct pool_item *pi = v;
1.3       pk       1217:        struct pool_item_header *ph;
                   1218:
1.101.2.6  yamt     1219:        KASSERT(mutex_owned(&pp->pr_lock));
1.101.2.3  yamt     1220:        FREECHECK_IN(&pp->pr_freecheck, v);
1.101.2.6  yamt     1221:        LOCKDEBUG_MEM_CHECK(v, pp->pr_size);
1.61      chs      1222:
1.30      thorpej  1223: #ifdef DIAGNOSTIC
1.34      thorpej  1224:        if (__predict_false(pp->pr_nout == 0)) {
1.30      thorpej  1225:                printf("pool %s: putting with none out\n",
                   1226:                    pp->pr_wchan);
                   1227:                panic("pool_put");
                   1228:        }
                   1229: #endif
1.3       pk       1230:
1.101.2.2  yamt     1231:        if (__predict_false((ph = pr_find_pagehead(pp, v)) == NULL)) {
1.25      thorpej  1232:                pr_printlog(pp, NULL, printf);
1.3       pk       1233:                panic("pool_put: %s: page header missing", pp->pr_wchan);
                   1234:        }
1.28      thorpej  1235:
1.3       pk       1236:        /*
                   1237:         * Return to item list.
                   1238:         */
1.97      yamt     1239:        if (pp->pr_roflags & PR_NOTOUCH) {
                   1240:                pr_item_notouch_put(pp, ph, v);
                   1241:        } else {
1.2       pk       1242: #ifdef DIAGNOSTIC
1.97      yamt     1243:                pi->pi_magic = PI_MAGIC;
1.3       pk       1244: #endif
1.32      chs      1245: #ifdef DEBUG
1.97      yamt     1246:                {
                   1247:                        int i, *ip = v;
1.32      chs      1248:
1.97      yamt     1249:                        for (i = 0; i < pp->pr_size / sizeof(int); i++) {
                   1250:                                *ip++ = PI_MAGIC;
                   1251:                        }
1.32      chs      1252:                }
                   1253: #endif
                   1254:
1.101.2.1  yamt     1255:                LIST_INSERT_HEAD(&ph->ph_itemlist, pi, pi_list);
1.97      yamt     1256:        }
1.79      thorpej  1257:        KDASSERT(ph->ph_nmissing != 0);
1.3       pk       1258:        ph->ph_nmissing--;
                   1259:        pp->pr_nput++;
1.20      thorpej  1260:        pp->pr_nitems++;
                   1261:        pp->pr_nout--;
1.3       pk       1262:
                   1263:        /* Cancel "pool empty" condition if it exists */
                   1264:        if (pp->pr_curpage == NULL)
                   1265:                pp->pr_curpage = ph;
                   1266:
                   1267:        if (pp->pr_flags & PR_WANTED) {
                   1268:                pp->pr_flags &= ~PR_WANTED;
1.15      pk       1269:                if (ph->ph_nmissing == 0)
                   1270:                        pp->pr_nidle++;
1.101.2.6  yamt     1271:                cv_broadcast(&pp->pr_cv);
1.3       pk       1272:                return;
                   1273:        }
                   1274:
                   1275:        /*
1.88      chs      1276:         * If this page is now empty, do one of two things:
1.21      thorpej  1277:         *
1.88      chs      1278:         *      (1) If we have more pages than the page high water mark,
1.96      thorpej  1279:         *          free the page back to the system.  ONLY CONSIDER
1.90      thorpej  1280:         *          FREEING BACK A PAGE IF WE HAVE MORE THAN OUR MINIMUM PAGE
                   1281:         *          CLAIM.
1.21      thorpej  1282:         *
1.88      chs      1283:         *      (2) Otherwise, move the page to the empty page list.
                   1284:         *
                   1285:         * Either way, select a new current page (so we use a partially-full
                   1286:         * page if one is available).
1.3       pk       1287:         */
                   1288:        if (ph->ph_nmissing == 0) {
1.6       thorpej  1289:                pp->pr_nidle++;
1.90      thorpej  1290:                if (pp->pr_npages > pp->pr_minpages &&
                   1291:                    (pp->pr_npages > pp->pr_maxpages ||
1.101.2.1  yamt     1292:                     pa_starved_p(pp->pr_alloc))) {
1.101     thorpej  1293:                        pr_rmpage(pp, ph, pq);
1.3       pk       1294:                } else {
1.88      chs      1295:                        LIST_REMOVE(ph, ph_pagelist);
                   1296:                        LIST_INSERT_HEAD(&pp->pr_emptypages, ph, ph_pagelist);
1.3       pk       1297:
1.21      thorpej  1298:                        /*
                   1299:                         * Update the timestamp on the page.  A page must
                   1300:                         * be idle for some period of time before it can
                   1301:                         * be reclaimed by the pagedaemon.  This minimizes
                   1302:                         * ping-pong'ing for memory.
                   1303:                         */
1.101.2.1  yamt     1304:                        getmicrotime(&ph->ph_time);
1.1       pk       1305:                }
1.88      chs      1306:                pool_update_curpage(pp);
1.1       pk       1307:        }
1.88      chs      1308:
1.21      thorpej  1309:        /*
1.88      chs      1310:         * If the page was previously completely full, move it to the
                   1311:         * partially-full list and make it the current page.  The next
                   1312:         * allocation will get the item from this page, instead of
                   1313:         * further fragmenting the pool.
1.21      thorpej  1314:         */
                   1315:        else if (ph->ph_nmissing == (pp->pr_itemsperpage - 1)) {
1.88      chs      1316:                LIST_REMOVE(ph, ph_pagelist);
                   1317:                LIST_INSERT_HEAD(&pp->pr_partpages, ph, ph_pagelist);
1.21      thorpej  1318:                pp->pr_curpage = ph;
                   1319:        }
1.43      thorpej  1320: }
                   1321:
                   1322: /*
1.101.2.6  yamt     1323:  * Return resource to the pool.
1.43      thorpej  1324:  */
1.59      thorpej  1325: #ifdef POOL_DIAGNOSTIC
1.43      thorpej  1326: void
                   1327: _pool_put(struct pool *pp, void *v, const char *file, long line)
                   1328: {
1.101     thorpej  1329:        struct pool_pagelist pq;
                   1330:
                   1331:        LIST_INIT(&pq);
1.43      thorpej  1332:
1.101.2.6  yamt     1333:        mutex_enter(&pp->pr_lock);
1.43      thorpej  1334:        pr_enter(pp, file, line);
                   1335:
1.56      sommerfe 1336:        pr_log(pp, v, PRLOG_PUT, file, line);
                   1337:
1.101     thorpej  1338:        pool_do_put(pp, v, &pq);
1.21      thorpej  1339:
1.25      thorpej  1340:        pr_leave(pp);
1.101.2.6  yamt     1341:        mutex_exit(&pp->pr_lock);
1.101     thorpej  1342:
1.101.2.1  yamt     1343:        pr_pagelist_free(pp, &pq);
1.1       pk       1344: }
1.57      sommerfe 1345: #undef pool_put
1.59      thorpej  1346: #endif /* POOL_DIAGNOSTIC */
1.1       pk       1347:
1.56      sommerfe 1348: void
                   1349: pool_put(struct pool *pp, void *v)
                   1350: {
1.101     thorpej  1351:        struct pool_pagelist pq;
                   1352:
                   1353:        LIST_INIT(&pq);
1.56      sommerfe 1354:
1.101.2.6  yamt     1355:        mutex_enter(&pp->pr_lock);
1.101     thorpej  1356:        pool_do_put(pp, v, &pq);
1.101.2.6  yamt     1357:        mutex_exit(&pp->pr_lock);
1.56      sommerfe 1358:
1.101.2.1  yamt     1359:        pr_pagelist_free(pp, &pq);
1.56      sommerfe 1360: }
1.57      sommerfe 1361:
1.59      thorpej  1362: #ifdef POOL_DIAGNOSTIC
1.57      sommerfe 1363: #define                pool_put(h, v)  _pool_put((h), (v), __FILE__, __LINE__)
1.56      sommerfe 1364: #endif
1.74      thorpej  1365:
                   1366: /*
1.101.2.1  yamt     1367:  * pool_grow: grow a pool by a page.
                   1368:  *
                   1369:  * => called with pool locked.
                   1370:  * => unlock and relock the pool.
                   1371:  * => return with pool locked.
                   1372:  */
                   1373:
                   1374: static int
                   1375: pool_grow(struct pool *pp, int flags)
                   1376: {
                   1377:        struct pool_item_header *ph = NULL;
                   1378:        char *cp;
                   1379:
1.101.2.6  yamt     1380:        mutex_exit(&pp->pr_lock);
1.101.2.1  yamt     1381:        cp = pool_allocator_alloc(pp, flags);
                   1382:        if (__predict_true(cp != NULL)) {
                   1383:                ph = pool_alloc_item_header(pp, cp, flags);
                   1384:        }
                   1385:        if (__predict_false(cp == NULL || ph == NULL)) {
                   1386:                if (cp != NULL) {
                   1387:                        pool_allocator_free(pp, cp);
                   1388:                }
1.101.2.6  yamt     1389:                mutex_enter(&pp->pr_lock);
1.101.2.1  yamt     1390:                return ENOMEM;
                   1391:        }
                   1392:
1.101.2.6  yamt     1393:        mutex_enter(&pp->pr_lock);
1.101.2.1  yamt     1394:        pool_prime_page(pp, cp, ph);
                   1395:        pp->pr_npagealloc++;
                   1396:        return 0;
                   1397: }
                   1398:
                   1399: /*
1.74      thorpej  1400:  * Add N items to the pool.
                   1401:  */
                   1402: int
                   1403: pool_prime(struct pool *pp, int n)
                   1404: {
1.75      simonb   1405:        int newpages;
1.101.2.1  yamt     1406:        int error = 0;
1.74      thorpej  1407:
1.101.2.6  yamt     1408:        mutex_enter(&pp->pr_lock);
1.74      thorpej  1409:
                   1410:        newpages = roundup(n, pp->pr_itemsperpage) / pp->pr_itemsperpage;
                   1411:
                   1412:        while (newpages-- > 0) {
1.101.2.1  yamt     1413:                error = pool_grow(pp, PR_NOWAIT);
                   1414:                if (error) {
1.74      thorpej  1415:                        break;
                   1416:                }
                   1417:                pp->pr_minpages++;
                   1418:        }
                   1419:
                   1420:        if (pp->pr_minpages >= pp->pr_maxpages)
                   1421:                pp->pr_maxpages = pp->pr_minpages + 1;  /* XXX */
                   1422:
1.101.2.6  yamt     1423:        mutex_exit(&pp->pr_lock);
1.101.2.1  yamt     1424:        return error;
1.74      thorpej  1425: }
1.55      thorpej  1426:
                   1427: /*
1.3       pk       1428:  * Add a page worth of items to the pool.
1.21      thorpej  1429:  *
                   1430:  * Note, we must be called with the pool descriptor LOCKED.
1.3       pk       1431:  */
1.55      thorpej  1432: static void
1.101.2.4  yamt     1433: pool_prime_page(struct pool *pp, void *storage, struct pool_item_header *ph)
1.3       pk       1434: {
                   1435:        struct pool_item *pi;
1.101.2.4  yamt     1436:        void *cp = storage;
1.101.2.3  yamt     1437:        const unsigned int align = pp->pr_align;
                   1438:        const unsigned int ioff = pp->pr_itemoffset;
1.55      thorpej  1439:        int n;
1.36      pk       1440:
1.101.2.6  yamt     1441:        KASSERT(mutex_owned(&pp->pr_lock));
1.91      yamt     1442:
1.66      thorpej  1443: #ifdef DIAGNOSTIC
1.101.2.2  yamt     1444:        if ((pp->pr_roflags & PR_NOALIGN) == 0 &&
1.101.2.10! yamt     1445:            ((uintptr_t)cp & (pp->pr_alloc->pa_pagesz - 1)) != 0)
1.36      pk       1446:                panic("pool_prime_page: %s: unaligned page", pp->pr_wchan);
1.66      thorpej  1447: #endif
1.3       pk       1448:
                   1449:        /*
                   1450:         * Insert page header.
                   1451:         */
1.88      chs      1452:        LIST_INSERT_HEAD(&pp->pr_emptypages, ph, ph_pagelist);
1.101.2.1  yamt     1453:        LIST_INIT(&ph->ph_itemlist);
1.3       pk       1454:        ph->ph_page = storage;
                   1455:        ph->ph_nmissing = 0;
1.101.2.1  yamt     1456:        getmicrotime(&ph->ph_time);
1.88      chs      1457:        if ((pp->pr_roflags & PR_PHINPAGE) == 0)
                   1458:                SPLAY_INSERT(phtree, &pp->pr_phtree, ph);
1.3       pk       1459:
1.6       thorpej  1460:        pp->pr_nidle++;
                   1461:
1.3       pk       1462:        /*
                   1463:         * Color this page.
                   1464:         */
1.101.2.8  yamt     1465:        ph->ph_off = pp->pr_curcolor;
                   1466:        cp = (char *)cp + ph->ph_off;
1.3       pk       1467:        if ((pp->pr_curcolor += align) > pp->pr_maxcolor)
                   1468:                pp->pr_curcolor = 0;
                   1469:
                   1470:        /*
                   1471:         * Adjust storage to apply aligment to `pr_itemoffset' in each item.
                   1472:         */
                   1473:        if (ioff != 0)
1.101.2.4  yamt     1474:                cp = (char *)cp + align - ioff;
1.3       pk       1475:
1.101.2.3  yamt     1476:        KASSERT((((vaddr_t)cp + ioff) & (align - 1)) == 0);
                   1477:
1.3       pk       1478:        /*
                   1479:         * Insert remaining chunks on the bucket list.
                   1480:         */
                   1481:        n = pp->pr_itemsperpage;
1.20      thorpej  1482:        pp->pr_nitems += n;
1.3       pk       1483:
1.97      yamt     1484:        if (pp->pr_roflags & PR_NOTOUCH) {
1.101.2.6  yamt     1485:                pr_item_notouch_init(pp, ph);
1.97      yamt     1486:        } else {
                   1487:                while (n--) {
                   1488:                        pi = (struct pool_item *)cp;
1.78      thorpej  1489:
1.97      yamt     1490:                        KASSERT(((((vaddr_t)pi) + ioff) & (align - 1)) == 0);
1.3       pk       1491:
1.97      yamt     1492:                        /* Insert on page list */
1.101.2.1  yamt     1493:                        LIST_INSERT_HEAD(&ph->ph_itemlist, pi, pi_list);
1.3       pk       1494: #ifdef DIAGNOSTIC
1.97      yamt     1495:                        pi->pi_magic = PI_MAGIC;
1.3       pk       1496: #endif
1.101.2.4  yamt     1497:                        cp = (char *)cp + pp->pr_size;
1.101.2.3  yamt     1498:
                   1499:                        KASSERT((((vaddr_t)cp + ioff) & (align - 1)) == 0);
1.97      yamt     1500:                }
1.3       pk       1501:        }
                   1502:
                   1503:        /*
                   1504:         * If the pool was depleted, point at the new page.
                   1505:         */
                   1506:        if (pp->pr_curpage == NULL)
                   1507:                pp->pr_curpage = ph;
                   1508:
                   1509:        if (++pp->pr_npages > pp->pr_hiwat)
                   1510:                pp->pr_hiwat = pp->pr_npages;
                   1511: }
                   1512:
1.20      thorpej  1513: /*
1.52      thorpej  1514:  * Used by pool_get() when nitems drops below the low water mark.  This
1.88      chs      1515:  * is used to catch up pr_nitems with the low water mark.
1.20      thorpej  1516:  *
1.21      thorpej  1517:  * Note 1, we never wait for memory here, we let the caller decide what to do.
1.20      thorpej  1518:  *
1.73      thorpej  1519:  * Note 2, we must be called with the pool already locked, and we return
1.20      thorpej  1520:  * with it locked.
                   1521:  */
                   1522: static int
1.42      thorpej  1523: pool_catchup(struct pool *pp)
1.20      thorpej  1524: {
                   1525:        int error = 0;
                   1526:
1.54      thorpej  1527:        while (POOL_NEEDS_CATCHUP(pp)) {
1.101.2.1  yamt     1528:                error = pool_grow(pp, PR_NOWAIT);
                   1529:                if (error) {
1.20      thorpej  1530:                        break;
                   1531:                }
                   1532:        }
1.101.2.1  yamt     1533:        return error;
1.20      thorpej  1534: }
                   1535:
1.88      chs      1536: static void
                   1537: pool_update_curpage(struct pool *pp)
                   1538: {
                   1539:
                   1540:        pp->pr_curpage = LIST_FIRST(&pp->pr_partpages);
                   1541:        if (pp->pr_curpage == NULL) {
                   1542:                pp->pr_curpage = LIST_FIRST(&pp->pr_emptypages);
                   1543:        }
                   1544: }
                   1545:
1.3       pk       1546: void
1.42      thorpej  1547: pool_setlowat(struct pool *pp, int n)
1.3       pk       1548: {
1.15      pk       1549:
1.101.2.6  yamt     1550:        mutex_enter(&pp->pr_lock);
1.21      thorpej  1551:
1.3       pk       1552:        pp->pr_minitems = n;
1.15      pk       1553:        pp->pr_minpages = (n == 0)
                   1554:                ? 0
1.18      thorpej  1555:                : roundup(n, pp->pr_itemsperpage) / pp->pr_itemsperpage;
1.20      thorpej  1556:
                   1557:        /* Make sure we're caught up with the newly-set low water mark. */
1.75      simonb   1558:        if (POOL_NEEDS_CATCHUP(pp) && pool_catchup(pp) != 0) {
1.20      thorpej  1559:                /*
                   1560:                 * XXX: Should we log a warning?  Should we set up a timeout
                   1561:                 * to try again in a second or so?  The latter could break
                   1562:                 * a caller's assumptions about interrupt protection, etc.
                   1563:                 */
                   1564:        }
1.21      thorpej  1565:
1.101.2.6  yamt     1566:        mutex_exit(&pp->pr_lock);
1.3       pk       1567: }
                   1568:
                   1569: void
1.42      thorpej  1570: pool_sethiwat(struct pool *pp, int n)
1.3       pk       1571: {
1.15      pk       1572:
1.101.2.6  yamt     1573:        mutex_enter(&pp->pr_lock);
1.21      thorpej  1574:
1.15      pk       1575:        pp->pr_maxpages = (n == 0)
                   1576:                ? 0
1.18      thorpej  1577:                : roundup(n, pp->pr_itemsperpage) / pp->pr_itemsperpage;
1.21      thorpej  1578:
1.101.2.6  yamt     1579:        mutex_exit(&pp->pr_lock);
1.3       pk       1580: }
                   1581:
1.20      thorpej  1582: void
1.42      thorpej  1583: pool_sethardlimit(struct pool *pp, int n, const char *warnmess, int ratecap)
1.20      thorpej  1584: {
                   1585:
1.101.2.6  yamt     1586:        mutex_enter(&pp->pr_lock);
1.20      thorpej  1587:
                   1588:        pp->pr_hardlimit = n;
                   1589:        pp->pr_hardlimit_warning = warnmess;
1.31      thorpej  1590:        pp->pr_hardlimit_ratecap.tv_sec = ratecap;
                   1591:        pp->pr_hardlimit_warning_last.tv_sec = 0;
                   1592:        pp->pr_hardlimit_warning_last.tv_usec = 0;
1.20      thorpej  1593:
                   1594:        /*
1.21      thorpej  1595:         * In-line version of pool_sethiwat(), because we don't want to
                   1596:         * release the lock.
1.20      thorpej  1597:         */
                   1598:        pp->pr_maxpages = (n == 0)
                   1599:                ? 0
                   1600:                : roundup(n, pp->pr_itemsperpage) / pp->pr_itemsperpage;
1.21      thorpej  1601:
1.101.2.6  yamt     1602:        mutex_exit(&pp->pr_lock);
1.20      thorpej  1603: }
1.3       pk       1604:
                   1605: /*
                   1606:  * Release all complete pages that have not been used recently.
                   1607:  */
1.66      thorpej  1608: int
1.59      thorpej  1609: #ifdef POOL_DIAGNOSTIC
1.42      thorpej  1610: _pool_reclaim(struct pool *pp, const char *file, long line)
1.56      sommerfe 1611: #else
                   1612: pool_reclaim(struct pool *pp)
                   1613: #endif
1.3       pk       1614: {
                   1615:        struct pool_item_header *ph, *phnext;
1.61      chs      1616:        struct pool_pagelist pq;
1.101.2.1  yamt     1617:        struct timeval curtime, diff;
1.101.2.6  yamt     1618:        bool klock;
                   1619:        int rv;
1.3       pk       1620:
1.68      thorpej  1621:        if (pp->pr_drain_hook != NULL) {
                   1622:                /*
                   1623:                 * The drain hook must be called with the pool unlocked.
                   1624:                 */
                   1625:                (*pp->pr_drain_hook)(pp->pr_drain_hook_arg, PR_NOWAIT);
                   1626:        }
                   1627:
1.101.2.6  yamt     1628:        /*
                   1629:         * XXXSMP Because mutexes at IPL_SOFTXXX are still spinlocks,
                   1630:         * and we are called from the pagedaemon without kernel_lock.
                   1631:         * Does not apply to IPL_SOFTBIO.
                   1632:         */
                   1633:        if (pp->pr_ipl == IPL_SOFTNET || pp->pr_ipl == IPL_SOFTCLOCK ||
                   1634:            pp->pr_ipl == IPL_SOFTSERIAL) {
                   1635:                KERNEL_LOCK(1, NULL);
                   1636:                klock = true;
                   1637:        } else
                   1638:                klock = false;
                   1639:
                   1640:        /* Reclaim items from the pool's cache (if any). */
                   1641:        if (pp->pr_cache != NULL)
                   1642:                pool_cache_invalidate(pp->pr_cache);
                   1643:
                   1644:        if (mutex_tryenter(&pp->pr_lock) == 0) {
                   1645:                if (klock) {
                   1646:                        KERNEL_UNLOCK_ONE(NULL);
                   1647:                }
1.66      thorpej  1648:                return (0);
1.101.2.6  yamt     1649:        }
1.25      thorpej  1650:        pr_enter(pp, file, line);
1.68      thorpej  1651:
1.88      chs      1652:        LIST_INIT(&pq);
1.43      thorpej  1653:
1.101.2.1  yamt     1654:        getmicrotime(&curtime);
1.21      thorpej  1655:
1.88      chs      1656:        for (ph = LIST_FIRST(&pp->pr_emptypages); ph != NULL; ph = phnext) {
                   1657:                phnext = LIST_NEXT(ph, ph_pagelist);
1.3       pk       1658:
                   1659:                /* Check our minimum page claim */
                   1660:                if (pp->pr_npages <= pp->pr_minpages)
                   1661:                        break;
                   1662:
1.88      chs      1663:                KASSERT(ph->ph_nmissing == 0);
                   1664:                timersub(&curtime, &ph->ph_time, &diff);
1.101.2.1  yamt     1665:                if (diff.tv_sec < pool_inactive_time
                   1666:                    && !pa_starved_p(pp->pr_alloc))
1.88      chs      1667:                        continue;
1.21      thorpej  1668:
1.88      chs      1669:                /*
                   1670:                 * If freeing this page would put us below
                   1671:                 * the low water mark, stop now.
                   1672:                 */
                   1673:                if ((pp->pr_nitems - pp->pr_itemsperpage) <
                   1674:                    pp->pr_minitems)
                   1675:                        break;
1.21      thorpej  1676:
1.88      chs      1677:                pr_rmpage(pp, ph, &pq);
1.3       pk       1678:        }
                   1679:
1.25      thorpej  1680:        pr_leave(pp);
1.101.2.6  yamt     1681:        mutex_exit(&pp->pr_lock);
1.66      thorpej  1682:
1.101.2.6  yamt     1683:        if (LIST_EMPTY(&pq))
                   1684:                rv = 0;
                   1685:        else {
                   1686:                pr_pagelist_free(pp, &pq);
                   1687:                rv = 1;
                   1688:        }
                   1689:
                   1690:        if (klock) {
                   1691:                KERNEL_UNLOCK_ONE(NULL);
                   1692:        }
                   1693:
                   1694:        return (rv);
1.3       pk       1695: }
                   1696:
                   1697: /*
1.101.2.6  yamt     1698:  * Drain pools, one at a time.  This is a two stage process;
                   1699:  * drain_start kicks off a cross call to drain CPU-level caches
                   1700:  * if the pool has an associated pool_cache.  drain_end waits
                   1701:  * for those cross calls to finish, and then drains the cache
                   1702:  * (if any) and pool.
1.101.2.4  yamt     1703:  *
1.101.2.6  yamt     1704:  * Note, must never be called from interrupt context.
1.3       pk       1705:  */
                   1706: void
1.101.2.6  yamt     1707: pool_drain_start(struct pool **ppp, uint64_t *wp)
1.3       pk       1708: {
                   1709:        struct pool *pp;
1.101.2.6  yamt     1710:
1.101.2.8  yamt     1711:        KASSERT(!TAILQ_EMPTY(&pool_head));
1.3       pk       1712:
1.61      chs      1713:        pp = NULL;
1.101.2.6  yamt     1714:
                   1715:        /* Find next pool to drain, and add a reference. */
                   1716:        mutex_enter(&pool_head_lock);
                   1717:        do {
                   1718:                if (drainpp == NULL) {
1.101.2.8  yamt     1719:                        drainpp = TAILQ_FIRST(&pool_head);
1.101.2.6  yamt     1720:                }
                   1721:                if (drainpp != NULL) {
                   1722:                        pp = drainpp;
1.101.2.8  yamt     1723:                        drainpp = TAILQ_NEXT(pp, pr_poollist);
1.101.2.6  yamt     1724:                }
                   1725:                /*
                   1726:                 * Skip completely idle pools.  We depend on at least
                   1727:                 * one pool in the system being active.
                   1728:                 */
                   1729:        } while (pp == NULL || pp->pr_npages == 0);
                   1730:        pp->pr_refcnt++;
                   1731:        mutex_exit(&pool_head_lock);
                   1732:
                   1733:        /* If there is a pool_cache, drain CPU level caches. */
                   1734:        *ppp = pp;
                   1735:        if (pp->pr_cache != NULL) {
                   1736:                *wp = xc_broadcast(0, (xcfunc_t)pool_cache_xcall,
                   1737:                    pp->pr_cache, NULL);
                   1738:        }
                   1739: }
                   1740:
                   1741: void
                   1742: pool_drain_end(struct pool *pp, uint64_t where)
                   1743: {
                   1744:
                   1745:        if (pp == NULL)
                   1746:                return;
                   1747:
                   1748:        KASSERT(pp->pr_refcnt > 0);
                   1749:
                   1750:        /* Wait for remote draining to complete. */
                   1751:        if (pp->pr_cache != NULL)
                   1752:                xc_wait(where);
                   1753:
                   1754:        /* Drain the cache (if any) and pool.. */
                   1755:        pool_reclaim(pp);
                   1756:
                   1757:        /* Finally, unlock the pool. */
                   1758:        mutex_enter(&pool_head_lock);
                   1759:        pp->pr_refcnt--;
                   1760:        cv_broadcast(&pool_busy);
                   1761:        mutex_exit(&pool_head_lock);
1.3       pk       1762: }
                   1763:
                   1764: /*
                   1765:  * Diagnostic helpers.
                   1766:  */
                   1767: void
1.42      thorpej  1768: pool_print(struct pool *pp, const char *modif)
1.21      thorpej  1769: {
                   1770:
1.25      thorpej  1771:        pool_print1(pp, modif, printf);
1.21      thorpej  1772: }
                   1773:
1.25      thorpej  1774: void
1.101.2.1  yamt     1775: pool_printall(const char *modif, void (*pr)(const char *, ...))
                   1776: {
                   1777:        struct pool *pp;
                   1778:
1.101.2.8  yamt     1779:        TAILQ_FOREACH(pp, &pool_head, pr_poollist) {
1.101.2.1  yamt     1780:                pool_printit(pp, modif, pr);
                   1781:        }
                   1782: }
                   1783:
                   1784: void
1.42      thorpej  1785: pool_printit(struct pool *pp, const char *modif, void (*pr)(const char *, ...))
1.25      thorpej  1786: {
                   1787:
                   1788:        if (pp == NULL) {
                   1789:                (*pr)("Must specify a pool to print.\n");
                   1790:                return;
                   1791:        }
                   1792:
                   1793:        pool_print1(pp, modif, pr);
                   1794: }
                   1795:
1.21      thorpej  1796: static void
1.97      yamt     1797: pool_print_pagelist(struct pool *pp, struct pool_pagelist *pl,
                   1798:     void (*pr)(const char *, ...))
1.88      chs      1799: {
                   1800:        struct pool_item_header *ph;
                   1801: #ifdef DIAGNOSTIC
                   1802:        struct pool_item *pi;
                   1803: #endif
                   1804:
                   1805:        LIST_FOREACH(ph, pl, ph_pagelist) {
                   1806:                (*pr)("\t\tpage %p, nmissing %d, time %lu,%lu\n",
                   1807:                    ph->ph_page, ph->ph_nmissing,
                   1808:                    (u_long)ph->ph_time.tv_sec,
                   1809:                    (u_long)ph->ph_time.tv_usec);
                   1810: #ifdef DIAGNOSTIC
1.97      yamt     1811:                if (!(pp->pr_roflags & PR_NOTOUCH)) {
1.101.2.1  yamt     1812:                        LIST_FOREACH(pi, &ph->ph_itemlist, pi_list) {
1.97      yamt     1813:                                if (pi->pi_magic != PI_MAGIC) {
                   1814:                                        (*pr)("\t\t\titem %p, magic 0x%x\n",
                   1815:                                            pi, pi->pi_magic);
                   1816:                                }
1.88      chs      1817:                        }
                   1818:                }
                   1819: #endif
                   1820:        }
                   1821: }
                   1822:
                   1823: static void
1.42      thorpej  1824: pool_print1(struct pool *pp, const char *modif, void (*pr)(const char *, ...))
1.3       pk       1825: {
1.25      thorpej  1826:        struct pool_item_header *ph;
1.101.2.6  yamt     1827:        pool_cache_t pc;
                   1828:        pcg_t *pcg;
                   1829:        pool_cache_cpu_t *cc;
                   1830:        uint64_t cpuhit, cpumiss;
1.44      thorpej  1831:        int i, print_log = 0, print_pagelist = 0, print_cache = 0;
1.25      thorpej  1832:        char c;
                   1833:
                   1834:        while ((c = *modif++) != '\0') {
                   1835:                if (c == 'l')
                   1836:                        print_log = 1;
                   1837:                if (c == 'p')
                   1838:                        print_pagelist = 1;
1.44      thorpej  1839:                if (c == 'c')
                   1840:                        print_cache = 1;
1.25      thorpej  1841:        }
                   1842:
1.101.2.6  yamt     1843:        if ((pc = pp->pr_cache) != NULL) {
                   1844:                (*pr)("POOL CACHE");
                   1845:        } else {
                   1846:                (*pr)("POOL");
                   1847:        }
                   1848:
                   1849:        (*pr)(" %s: size %u, align %u, ioff %u, roflags 0x%08x\n",
1.25      thorpej  1850:            pp->pr_wchan, pp->pr_size, pp->pr_align, pp->pr_itemoffset,
                   1851:            pp->pr_roflags);
1.66      thorpej  1852:        (*pr)("\talloc %p\n", pp->pr_alloc);
1.25      thorpej  1853:        (*pr)("\tminitems %u, minpages %u, maxpages %u, npages %u\n",
                   1854:            pp->pr_minitems, pp->pr_minpages, pp->pr_maxpages, pp->pr_npages);
                   1855:        (*pr)("\titemsperpage %u, nitems %u, nout %u, hardlimit %u\n",
                   1856:            pp->pr_itemsperpage, pp->pr_nitems, pp->pr_nout, pp->pr_hardlimit);
                   1857:
1.101.2.6  yamt     1858:        (*pr)("\tnget %lu, nfail %lu, nput %lu\n",
1.25      thorpej  1859:            pp->pr_nget, pp->pr_nfail, pp->pr_nput);
                   1860:        (*pr)("\tnpagealloc %lu, npagefree %lu, hiwat %u, nidle %lu\n",
                   1861:            pp->pr_npagealloc, pp->pr_npagefree, pp->pr_hiwat, pp->pr_nidle);
                   1862:
                   1863:        if (print_pagelist == 0)
                   1864:                goto skip_pagelist;
                   1865:
1.88      chs      1866:        if ((ph = LIST_FIRST(&pp->pr_emptypages)) != NULL)
                   1867:                (*pr)("\n\tempty page list:\n");
1.97      yamt     1868:        pool_print_pagelist(pp, &pp->pr_emptypages, pr);
1.88      chs      1869:        if ((ph = LIST_FIRST(&pp->pr_fullpages)) != NULL)
                   1870:                (*pr)("\n\tfull page list:\n");
1.97      yamt     1871:        pool_print_pagelist(pp, &pp->pr_fullpages, pr);
1.88      chs      1872:        if ((ph = LIST_FIRST(&pp->pr_partpages)) != NULL)
                   1873:                (*pr)("\n\tpartial-page list:\n");
1.97      yamt     1874:        pool_print_pagelist(pp, &pp->pr_partpages, pr);
1.88      chs      1875:
1.25      thorpej  1876:        if (pp->pr_curpage == NULL)
                   1877:                (*pr)("\tno current page\n");
                   1878:        else
                   1879:                (*pr)("\tcurpage %p\n", pp->pr_curpage->ph_page);
                   1880:
                   1881:  skip_pagelist:
                   1882:        if (print_log == 0)
                   1883:                goto skip_log;
                   1884:
                   1885:        (*pr)("\n");
                   1886:        if ((pp->pr_roflags & PR_LOGGING) == 0)
                   1887:                (*pr)("\tno log\n");
1.101.2.2  yamt     1888:        else {
1.25      thorpej  1889:                pr_printlog(pp, NULL, pr);
1.101.2.2  yamt     1890:        }
1.3       pk       1891:
1.25      thorpej  1892:  skip_log:
1.44      thorpej  1893:
1.101.2.1  yamt     1894: #define PR_GROUPLIST(pcg)                                              \
                   1895:        (*pr)("\t\tgroup %p: avail %d\n", pcg, pcg->pcg_avail);         \
1.101.2.8  yamt     1896:        for (i = 0; i < pcg->pcg_size; i++) {                           \
1.101.2.1  yamt     1897:                if (pcg->pcg_objects[i].pcgo_pa !=                      \
                   1898:                    POOL_PADDR_INVALID) {                               \
                   1899:                        (*pr)("\t\t\t%p, 0x%llx\n",                     \
                   1900:                            pcg->pcg_objects[i].pcgo_va,                \
                   1901:                            (unsigned long long)                        \
                   1902:                            pcg->pcg_objects[i].pcgo_pa);               \
                   1903:                } else {                                                \
                   1904:                        (*pr)("\t\t\t%p\n",                             \
                   1905:                            pcg->pcg_objects[i].pcgo_va);               \
                   1906:                }                                                       \
                   1907:        }
                   1908:
1.101.2.6  yamt     1909:        if (pc != NULL) {
                   1910:                cpuhit = 0;
                   1911:                cpumiss = 0;
                   1912:                for (i = 0; i < MAXCPUS; i++) {
                   1913:                        if ((cc = pc->pc_cpus[i]) == NULL)
                   1914:                                continue;
                   1915:                        cpuhit += cc->cc_hits;
                   1916:                        cpumiss += cc->cc_misses;
                   1917:                }
                   1918:                (*pr)("\tcpu layer hits %llu misses %llu\n", cpuhit, cpumiss);
                   1919:                (*pr)("\tcache layer hits %llu misses %llu\n",
                   1920:                    pc->pc_hits, pc->pc_misses);
                   1921:                (*pr)("\tcache layer entry uncontended %llu contended %llu\n",
                   1922:                    pc->pc_hits + pc->pc_misses - pc->pc_contended,
                   1923:                    pc->pc_contended);
                   1924:                (*pr)("\tcache layer empty groups %u full groups %u\n",
                   1925:                    pc->pc_nempty, pc->pc_nfull);
                   1926:                if (print_cache) {
                   1927:                        (*pr)("\tfull cache groups:\n");
                   1928:                        for (pcg = pc->pc_fullgroups; pcg != NULL;
                   1929:                            pcg = pcg->pcg_next) {
                   1930:                                PR_GROUPLIST(pcg);
                   1931:                        }
                   1932:                        (*pr)("\tempty cache groups:\n");
                   1933:                        for (pcg = pc->pc_emptygroups; pcg != NULL;
                   1934:                            pcg = pcg->pcg_next) {
                   1935:                                PR_GROUPLIST(pcg);
                   1936:                        }
1.44      thorpej  1937:                }
                   1938:        }
1.101.2.1  yamt     1939: #undef PR_GROUPLIST
1.44      thorpej  1940:
1.88      chs      1941:        pr_enter_check(pp, pr);
                   1942: }
                   1943:
                   1944: static int
                   1945: pool_chk_page(struct pool *pp, const char *label, struct pool_item_header *ph)
                   1946: {
                   1947:        struct pool_item *pi;
1.101.2.4  yamt     1948:        void *page;
1.88      chs      1949:        int n;
                   1950:
1.101.2.2  yamt     1951:        if ((pp->pr_roflags & PR_NOALIGN) == 0) {
1.101.2.4  yamt     1952:                page = (void *)((uintptr_t)ph & pp->pr_alloc->pa_pagemask);
1.101.2.2  yamt     1953:                if (page != ph->ph_page &&
                   1954:                    (pp->pr_roflags & PR_PHINPAGE) != 0) {
                   1955:                        if (label != NULL)
                   1956:                                printf("%s: ", label);
                   1957:                        printf("pool(%p:%s): page inconsistency: page %p;"
                   1958:                               " at page head addr %p (p %p)\n", pp,
                   1959:                                pp->pr_wchan, ph->ph_page,
                   1960:                                ph, page);
                   1961:                        return 1;
                   1962:                }
1.88      chs      1963:        }
1.3       pk       1964:
1.97      yamt     1965:        if ((pp->pr_roflags & PR_NOTOUCH) != 0)
                   1966:                return 0;
                   1967:
1.101.2.1  yamt     1968:        for (pi = LIST_FIRST(&ph->ph_itemlist), n = 0;
1.88      chs      1969:             pi != NULL;
1.101.2.1  yamt     1970:             pi = LIST_NEXT(pi,pi_list), n++) {
1.88      chs      1971:
                   1972: #ifdef DIAGNOSTIC
                   1973:                if (pi->pi_magic != PI_MAGIC) {
                   1974:                        if (label != NULL)
                   1975:                                printf("%s: ", label);
                   1976:                        printf("pool(%s): free list modified: magic=%x;"
1.101.2.2  yamt     1977:                               " page %p; item ordinal %d; addr %p\n",
1.88      chs      1978:                                pp->pr_wchan, pi->pi_magic, ph->ph_page,
1.101.2.2  yamt     1979:                                n, pi);
1.88      chs      1980:                        panic("pool");
                   1981:                }
                   1982: #endif
1.101.2.2  yamt     1983:                if ((pp->pr_roflags & PR_NOALIGN) != 0) {
                   1984:                        continue;
                   1985:                }
1.101.2.4  yamt     1986:                page = (void *)((uintptr_t)pi & pp->pr_alloc->pa_pagemask);
1.88      chs      1987:                if (page == ph->ph_page)
                   1988:                        continue;
                   1989:
                   1990:                if (label != NULL)
                   1991:                        printf("%s: ", label);
                   1992:                printf("pool(%p:%s): page inconsistency: page %p;"
                   1993:                       " item ordinal %d; addr %p (p %p)\n", pp,
                   1994:                        pp->pr_wchan, ph->ph_page,
                   1995:                        n, pi, page);
                   1996:                return 1;
                   1997:        }
                   1998:        return 0;
1.3       pk       1999: }
                   2000:
1.88      chs      2001:
1.3       pk       2002: int
1.42      thorpej  2003: pool_chk(struct pool *pp, const char *label)
1.3       pk       2004: {
                   2005:        struct pool_item_header *ph;
                   2006:        int r = 0;
                   2007:
1.101.2.6  yamt     2008:        mutex_enter(&pp->pr_lock);
1.88      chs      2009:        LIST_FOREACH(ph, &pp->pr_emptypages, ph_pagelist) {
                   2010:                r = pool_chk_page(pp, label, ph);
                   2011:                if (r) {
                   2012:                        goto out;
                   2013:                }
                   2014:        }
                   2015:        LIST_FOREACH(ph, &pp->pr_fullpages, ph_pagelist) {
                   2016:                r = pool_chk_page(pp, label, ph);
                   2017:                if (r) {
1.3       pk       2018:                        goto out;
                   2019:                }
1.88      chs      2020:        }
                   2021:        LIST_FOREACH(ph, &pp->pr_partpages, ph_pagelist) {
                   2022:                r = pool_chk_page(pp, label, ph);
                   2023:                if (r) {
1.3       pk       2024:                        goto out;
                   2025:                }
                   2026:        }
1.88      chs      2027:
1.3       pk       2028: out:
1.101.2.6  yamt     2029:        mutex_exit(&pp->pr_lock);
1.3       pk       2030:        return (r);
1.43      thorpej  2031: }
                   2032:
                   2033: /*
                   2034:  * pool_cache_init:
                   2035:  *
                   2036:  *     Initialize a pool cache.
1.101.2.6  yamt     2037:  */
                   2038: pool_cache_t
                   2039: pool_cache_init(size_t size, u_int align, u_int align_offset, u_int flags,
                   2040:     const char *wchan, struct pool_allocator *palloc, int ipl,
                   2041:     int (*ctor)(void *, void *, int), void (*dtor)(void *, void *), void *arg)
                   2042: {
                   2043:        pool_cache_t pc;
                   2044:
                   2045:        pc = pool_get(&cache_pool, PR_WAITOK);
                   2046:        if (pc == NULL)
                   2047:                return NULL;
                   2048:
                   2049:        pool_cache_bootstrap(pc, size, align, align_offset, flags, wchan,
                   2050:           palloc, ipl, ctor, dtor, arg);
                   2051:
                   2052:        return pc;
                   2053: }
                   2054:
                   2055: /*
                   2056:  * pool_cache_bootstrap:
1.43      thorpej  2057:  *
1.101.2.6  yamt     2058:  *     Kernel-private version of pool_cache_init().  The caller
                   2059:  *     provides initial storage.
1.43      thorpej  2060:  */
                   2061: void
1.101.2.6  yamt     2062: pool_cache_bootstrap(pool_cache_t pc, size_t size, u_int align,
                   2063:     u_int align_offset, u_int flags, const char *wchan,
                   2064:     struct pool_allocator *palloc, int ipl,
                   2065:     int (*ctor)(void *, void *, int), void (*dtor)(void *, void *),
1.43      thorpej  2066:     void *arg)
                   2067: {
1.101.2.6  yamt     2068:        CPU_INFO_ITERATOR cii;
1.101.2.8  yamt     2069:        pool_cache_t pc1;
1.101.2.6  yamt     2070:        struct cpu_info *ci;
                   2071:        struct pool *pp;
                   2072:
                   2073:        pp = &pc->pc_pool;
                   2074:        if (palloc == NULL && ipl == IPL_NONE)
                   2075:                palloc = &pool_allocator_nointr;
                   2076:        pool_init(pp, size, align, align_offset, flags, wchan, palloc, ipl);
1.43      thorpej  2077:
1.101.2.7  yamt     2078:        /*
                   2079:         * XXXAD hack to prevent IP input processing from blocking.
                   2080:         */
                   2081:        if (ipl == IPL_SOFTNET) {
                   2082:                mutex_init(&pc->pc_lock, MUTEX_DEFAULT, IPL_VM);
                   2083:        } else {
                   2084:                mutex_init(&pc->pc_lock, MUTEX_DEFAULT, ipl);
                   2085:        }
1.43      thorpej  2086:
1.101.2.6  yamt     2087:        if (ctor == NULL) {
                   2088:                ctor = (int (*)(void *, void *, int))nullop;
                   2089:        }
                   2090:        if (dtor == NULL) {
                   2091:                dtor = (void (*)(void *, void *))nullop;
                   2092:        }
1.43      thorpej  2093:
1.101.2.6  yamt     2094:        pc->pc_emptygroups = NULL;
                   2095:        pc->pc_fullgroups = NULL;
                   2096:        pc->pc_partgroups = NULL;
1.43      thorpej  2097:        pc->pc_ctor = ctor;
                   2098:        pc->pc_dtor = dtor;
                   2099:        pc->pc_arg  = arg;
1.101.2.6  yamt     2100:        pc->pc_hits  = 0;
1.48      thorpej  2101:        pc->pc_misses = 0;
1.101.2.6  yamt     2102:        pc->pc_nempty = 0;
                   2103:        pc->pc_npart = 0;
                   2104:        pc->pc_nfull = 0;
                   2105:        pc->pc_contended = 0;
                   2106:        pc->pc_refcnt = 0;
                   2107:        pc->pc_freecheck = NULL;
                   2108:
1.101.2.8  yamt     2109:        if ((flags & PR_LARGECACHE) != 0) {
                   2110:                pc->pc_pcgsize = PCG_NOBJECTS_LARGE;
                   2111:        } else {
                   2112:                pc->pc_pcgsize = PCG_NOBJECTS_NORMAL;
                   2113:        }
                   2114:
1.101.2.6  yamt     2115:        /* Allocate per-CPU caches. */
                   2116:        memset(pc->pc_cpus, 0, sizeof(pc->pc_cpus));
                   2117:        pc->pc_ncpu = 0;
1.101.2.8  yamt     2118:        if (ncpu < 2) {
1.101.2.7  yamt     2119:                /* XXX For sparc: boot CPU is not attached yet. */
                   2120:                pool_cache_cpu_init1(curcpu(), pc);
                   2121:        } else {
                   2122:                for (CPU_INFO_FOREACH(cii, ci)) {
                   2123:                        pool_cache_cpu_init1(ci, pc);
                   2124:                }
1.101.2.6  yamt     2125:        }
1.101.2.8  yamt     2126:
                   2127:        /* Add to list of all pools. */
                   2128:        if (__predict_true(!cold))
1.101.2.6  yamt     2129:                mutex_enter(&pool_head_lock);
1.101.2.8  yamt     2130:        TAILQ_FOREACH(pc1, &pool_cache_head, pc_cachelist) {
                   2131:                if (strcmp(pc1->pc_pool.pr_wchan, pc->pc_pool.pr_wchan) > 0)
                   2132:                        break;
1.101.2.6  yamt     2133:        }
1.101.2.8  yamt     2134:        if (pc1 == NULL)
                   2135:                TAILQ_INSERT_TAIL(&pool_cache_head, pc, pc_cachelist);
                   2136:        else
                   2137:                TAILQ_INSERT_BEFORE(pc1, pc, pc_cachelist);
                   2138:        if (__predict_true(!cold))
                   2139:                mutex_exit(&pool_head_lock);
                   2140:
                   2141:        membar_sync();
                   2142:        pp->pr_cache = pc;
1.43      thorpej  2143: }
                   2144:
                   2145: /*
                   2146:  * pool_cache_destroy:
                   2147:  *
                   2148:  *     Destroy a pool cache.
                   2149:  */
                   2150: void
1.101.2.6  yamt     2151: pool_cache_destroy(pool_cache_t pc)
1.43      thorpej  2152: {
1.101.2.6  yamt     2153:        struct pool *pp = &pc->pc_pool;
                   2154:        pool_cache_cpu_t *cc;
                   2155:        pcg_t *pcg;
                   2156:        int i;
                   2157:
                   2158:        /* Remove it from the global list. */
                   2159:        mutex_enter(&pool_head_lock);
                   2160:        while (pc->pc_refcnt != 0)
                   2161:                cv_wait(&pool_busy, &pool_head_lock);
1.101.2.8  yamt     2162:        TAILQ_REMOVE(&pool_cache_head, pc, pc_cachelist);
1.101.2.6  yamt     2163:        mutex_exit(&pool_head_lock);
1.43      thorpej  2164:
                   2165:        /* First, invalidate the entire cache. */
                   2166:        pool_cache_invalidate(pc);
                   2167:
1.101.2.6  yamt     2168:        /* Disassociate it from the pool. */
                   2169:        mutex_enter(&pp->pr_lock);
                   2170:        pp->pr_cache = NULL;
                   2171:        mutex_exit(&pp->pr_lock);
                   2172:
                   2173:        /* Destroy per-CPU data */
                   2174:        for (i = 0; i < MAXCPUS; i++) {
                   2175:                if ((cc = pc->pc_cpus[i]) == NULL)
                   2176:                        continue;
                   2177:                if ((pcg = cc->cc_current) != NULL) {
                   2178:                        pcg->pcg_next = NULL;
                   2179:                        pool_cache_invalidate_groups(pc, pcg);
                   2180:                }
                   2181:                if ((pcg = cc->cc_previous) != NULL) {
                   2182:                        pcg->pcg_next = NULL;
                   2183:                        pool_cache_invalidate_groups(pc, pcg);
                   2184:                }
                   2185:                if (cc != &pc->pc_cpu0)
                   2186:                        pool_put(&cache_cpu_pool, cc);
                   2187:        }
                   2188:
                   2189:        /* Finally, destroy it. */
                   2190:        mutex_destroy(&pc->pc_lock);
                   2191:        pool_destroy(pp);
                   2192:        pool_put(&cache_pool, pc);
1.43      thorpej  2193: }
                   2194:
1.101.2.6  yamt     2195: /*
                   2196:  * pool_cache_cpu_init1:
                   2197:  *
                   2198:  *     Called for each pool_cache whenever a new CPU is attached.
                   2199:  */
                   2200: static void
                   2201: pool_cache_cpu_init1(struct cpu_info *ci, pool_cache_t pc)
1.43      thorpej  2202: {
1.101.2.6  yamt     2203:        pool_cache_cpu_t *cc;
1.101.2.7  yamt     2204:        int index;
                   2205:
                   2206:        index = ci->ci_index;
1.43      thorpej  2207:
1.101.2.7  yamt     2208:        KASSERT(index < MAXCPUS);
1.101.2.6  yamt     2209:        KASSERT(((uintptr_t)pc->pc_cpus & (CACHE_LINE_SIZE - 1)) == 0);
1.43      thorpej  2210:
1.101.2.7  yamt     2211:        if ((cc = pc->pc_cpus[index]) != NULL) {
                   2212:                KASSERT(cc->cc_cpuindex == index);
1.101.2.6  yamt     2213:                return;
                   2214:        }
                   2215:
                   2216:        /*
                   2217:         * The first CPU is 'free'.  This needs to be the case for
                   2218:         * bootstrap - we may not be able to allocate yet.
                   2219:         */
                   2220:        if (pc->pc_ncpu == 0) {
                   2221:                cc = &pc->pc_cpu0;
                   2222:                pc->pc_ncpu = 1;
                   2223:        } else {
                   2224:                mutex_enter(&pc->pc_lock);
                   2225:                pc->pc_ncpu++;
                   2226:                mutex_exit(&pc->pc_lock);
                   2227:                cc = pool_get(&cache_cpu_pool, PR_WAITOK);
                   2228:        }
                   2229:
                   2230:        cc->cc_ipl = pc->pc_pool.pr_ipl;
                   2231:        cc->cc_iplcookie = makeiplcookie(cc->cc_ipl);
                   2232:        cc->cc_cache = pc;
1.101.2.7  yamt     2233:        cc->cc_cpuindex = index;
1.101.2.6  yamt     2234:        cc->cc_hits = 0;
                   2235:        cc->cc_misses = 0;
                   2236:        cc->cc_current = NULL;
                   2237:        cc->cc_previous = NULL;
                   2238:
1.101.2.7  yamt     2239:        pc->pc_cpus[index] = cc;
1.43      thorpej  2240: }
                   2241:
1.101.2.6  yamt     2242: /*
                   2243:  * pool_cache_cpu_init:
                   2244:  *
                   2245:  *     Called whenever a new CPU is attached.
                   2246:  */
                   2247: void
                   2248: pool_cache_cpu_init(struct cpu_info *ci)
1.43      thorpej  2249: {
1.101.2.6  yamt     2250:        pool_cache_t pc;
1.43      thorpej  2251:
1.101.2.6  yamt     2252:        mutex_enter(&pool_head_lock);
1.101.2.8  yamt     2253:        TAILQ_FOREACH(pc, &pool_cache_head, pc_cachelist) {
1.101.2.6  yamt     2254:                pc->pc_refcnt++;
                   2255:                mutex_exit(&pool_head_lock);
1.43      thorpej  2256:
1.101.2.6  yamt     2257:                pool_cache_cpu_init1(ci, pc);
                   2258:
                   2259:                mutex_enter(&pool_head_lock);
                   2260:                pc->pc_refcnt--;
                   2261:                cv_broadcast(&pool_busy);
                   2262:        }
                   2263:        mutex_exit(&pool_head_lock);
                   2264: }
                   2265:
                   2266: /*
                   2267:  * pool_cache_reclaim:
                   2268:  *
                   2269:  *     Reclaim memory from a pool cache.
                   2270:  */
                   2271: bool
                   2272: pool_cache_reclaim(pool_cache_t pc)
                   2273: {
                   2274:
                   2275:        return pool_reclaim(&pc->pc_pool);
1.43      thorpej  2276: }
                   2277:
1.101.2.1  yamt     2278: static void
1.101.2.6  yamt     2279: pool_cache_destruct_object1(pool_cache_t pc, void *object)
1.101.2.1  yamt     2280: {
                   2281:
1.101.2.6  yamt     2282:        (*pc->pc_dtor)(pc->pc_arg, object);
                   2283:        pool_put(&pc->pc_pool, object);
1.101.2.1  yamt     2284: }
                   2285:
1.43      thorpej  2286: /*
1.101.2.6  yamt     2287:  * pool_cache_destruct_object:
1.43      thorpej  2288:  *
1.101.2.6  yamt     2289:  *     Force destruction of an object and its release back into
                   2290:  *     the pool.
1.43      thorpej  2291:  */
1.101.2.6  yamt     2292: void
                   2293: pool_cache_destruct_object(pool_cache_t pc, void *object)
1.43      thorpej  2294: {
1.58      thorpej  2295:
1.101.2.6  yamt     2296:        FREECHECK_IN(&pc->pc_freecheck, object);
1.43      thorpej  2297:
1.101.2.6  yamt     2298:        pool_cache_destruct_object1(pc, object);
                   2299: }
1.43      thorpej  2300:
1.101.2.6  yamt     2301: /*
                   2302:  * pool_cache_invalidate_groups:
                   2303:  *
                   2304:  *     Invalidate a chain of groups and destruct all objects.
                   2305:  */
                   2306: static void
                   2307: pool_cache_invalidate_groups(pool_cache_t pc, pcg_t *pcg)
                   2308: {
                   2309:        void *object;
                   2310:        pcg_t *next;
                   2311:        int i;
1.101.2.3  yamt     2312:
1.101.2.6  yamt     2313:        for (; pcg != NULL; pcg = next) {
                   2314:                next = pcg->pcg_next;
1.43      thorpej  2315:
1.101.2.6  yamt     2316:                for (i = 0; i < pcg->pcg_avail; i++) {
                   2317:                        object = pcg->pcg_objects[i].pcgo_va;
                   2318:                        pool_cache_destruct_object1(pc, object);
                   2319:                }
1.43      thorpej  2320:
1.101.2.8  yamt     2321:                if (pcg->pcg_size == PCG_NOBJECTS_LARGE) {
                   2322:                        pool_put(&pcg_large_pool, pcg);
                   2323:                } else {
                   2324:                        KASSERT(pcg->pcg_size == PCG_NOBJECTS_NORMAL);
                   2325:                        pool_put(&pcg_normal_pool, pcg);
                   2326:                }
1.101.2.1  yamt     2327:        }
1.43      thorpej  2328: }
                   2329:
                   2330: /*
1.101.2.6  yamt     2331:  * pool_cache_invalidate:
1.43      thorpej  2332:  *
1.101.2.6  yamt     2333:  *     Invalidate a pool cache (destruct and release all of the
                   2334:  *     cached objects).  Does not reclaim objects from the pool.
1.43      thorpej  2335:  */
                   2336: void
1.101.2.6  yamt     2337: pool_cache_invalidate(pool_cache_t pc)
1.43      thorpej  2338: {
1.101.2.6  yamt     2339:        pcg_t *full, *empty, *part;
1.43      thorpej  2340:
1.101.2.6  yamt     2341:        mutex_enter(&pc->pc_lock);
                   2342:        full = pc->pc_fullgroups;
                   2343:        empty = pc->pc_emptygroups;
                   2344:        part = pc->pc_partgroups;
                   2345:        pc->pc_fullgroups = NULL;
                   2346:        pc->pc_emptygroups = NULL;
                   2347:        pc->pc_partgroups = NULL;
                   2348:        pc->pc_nfull = 0;
                   2349:        pc->pc_nempty = 0;
                   2350:        pc->pc_npart = 0;
                   2351:        mutex_exit(&pc->pc_lock);
                   2352:
                   2353:        pool_cache_invalidate_groups(pc, full);
                   2354:        pool_cache_invalidate_groups(pc, empty);
                   2355:        pool_cache_invalidate_groups(pc, part);
                   2356: }
                   2357:
                   2358: void
                   2359: pool_cache_set_drain_hook(pool_cache_t pc, void (*fn)(void *, int), void *arg)
                   2360: {
                   2361:
                   2362:        pool_set_drain_hook(&pc->pc_pool, fn, arg);
                   2363: }
                   2364:
                   2365: void
                   2366: pool_cache_setlowat(pool_cache_t pc, int n)
                   2367: {
                   2368:
                   2369:        pool_setlowat(&pc->pc_pool, n);
                   2370: }
                   2371:
                   2372: void
                   2373: pool_cache_sethiwat(pool_cache_t pc, int n)
                   2374: {
                   2375:
                   2376:        pool_sethiwat(&pc->pc_pool, n);
                   2377: }
                   2378:
                   2379: void
                   2380: pool_cache_sethardlimit(pool_cache_t pc, int n, const char *warnmess, int ratecap)
                   2381: {
                   2382:
                   2383:        pool_sethardlimit(&pc->pc_pool, n, warnmess, ratecap);
                   2384: }
                   2385:
                   2386: static inline pool_cache_cpu_t *
                   2387: pool_cache_cpu_enter(pool_cache_t pc, int *s)
                   2388: {
                   2389:        pool_cache_cpu_t *cc;
                   2390:
                   2391:        /*
                   2392:         * Prevent other users of the cache from accessing our
                   2393:         * CPU-local data.  To avoid touching shared state, we
                   2394:         * pull the neccessary information from CPU local data.
                   2395:         */
1.101.2.7  yamt     2396:        crit_enter();
                   2397:        cc = pc->pc_cpus[curcpu()->ci_index];
1.101.2.6  yamt     2398:        KASSERT(cc->cc_cache == pc);
1.101.2.7  yamt     2399:        if (cc->cc_ipl != IPL_NONE) {
1.101.2.6  yamt     2400:                *s = splraiseipl(cc->cc_iplcookie);
                   2401:        }
                   2402:        KASSERT(((uintptr_t)cc & (CACHE_LINE_SIZE - 1)) == 0);
1.43      thorpej  2403:
1.101.2.6  yamt     2404:        return cc;
                   2405: }
                   2406:
                   2407: static inline void
                   2408: pool_cache_cpu_exit(pool_cache_cpu_t *cc, int *s)
                   2409: {
                   2410:
                   2411:        /* No longer need exclusive access to the per-CPU data. */
1.101.2.7  yamt     2412:        if (cc->cc_ipl != IPL_NONE) {
1.101.2.6  yamt     2413:                splx(*s);
1.101.2.1  yamt     2414:        }
1.101.2.7  yamt     2415:        crit_exit();
1.101.2.6  yamt     2416: }
                   2417:
                   2418: #if __GNUC_PREREQ__(3, 0)
                   2419: __attribute ((noinline))
                   2420: #endif
                   2421: pool_cache_cpu_t *
                   2422: pool_cache_get_slow(pool_cache_cpu_t *cc, int *s, void **objectp,
                   2423:                    paddr_t *pap, int flags)
                   2424: {
                   2425:        pcg_t *pcg, *cur;
                   2426:        uint64_t ncsw;
                   2427:        pool_cache_t pc;
                   2428:        void *object;
                   2429:
                   2430:        pc = cc->cc_cache;
                   2431:        cc->cc_misses++;
                   2432:
                   2433:        /*
                   2434:         * Nothing was available locally.  Try and grab a group
                   2435:         * from the cache.
                   2436:         */
                   2437:        if (!mutex_tryenter(&pc->pc_lock)) {
                   2438:                ncsw = curlwp->l_ncsw;
                   2439:                mutex_enter(&pc->pc_lock);
                   2440:                pc->pc_contended++;
1.43      thorpej  2441:
                   2442:                /*
1.101.2.6  yamt     2443:                 * If we context switched while locking, then
                   2444:                 * our view of the per-CPU data is invalid:
                   2445:                 * retry.
1.43      thorpej  2446:                 */
1.101.2.6  yamt     2447:                if (curlwp->l_ncsw != ncsw) {
                   2448:                        mutex_exit(&pc->pc_lock);
                   2449:                        pool_cache_cpu_exit(cc, s);
                   2450:                        return pool_cache_cpu_enter(pc, s);
                   2451:                }
                   2452:        }
1.43      thorpej  2453:
1.101.2.6  yamt     2454:        if ((pcg = pc->pc_fullgroups) != NULL) {
                   2455:                /*
                   2456:                 * If there's a full group, release our empty
                   2457:                 * group back to the cache.  Install the full
                   2458:                 * group as cc_current and return.
                   2459:                 */
                   2460:                if ((cur = cc->cc_current) != NULL) {
                   2461:                        KASSERT(cur->pcg_avail == 0);
                   2462:                        cur->pcg_next = pc->pc_emptygroups;
                   2463:                        pc->pc_emptygroups = cur;
                   2464:                        pc->pc_nempty++;
1.101.2.1  yamt     2465:                }
1.101.2.8  yamt     2466:                KASSERT(pcg->pcg_avail == pcg->pcg_size);
1.101.2.6  yamt     2467:                cc->cc_current = pcg;
                   2468:                pc->pc_fullgroups = pcg->pcg_next;
                   2469:                pc->pc_hits++;
                   2470:                pc->pc_nfull--;
                   2471:                mutex_exit(&pc->pc_lock);
                   2472:                return cc;
1.43      thorpej  2473:        }
                   2474:
1.101.2.6  yamt     2475:        /*
                   2476:         * Nothing available locally or in cache.  Take the slow
                   2477:         * path: fetch a new object from the pool and construct
                   2478:         * it.
                   2479:         */
                   2480:        pc->pc_misses++;
                   2481:        mutex_exit(&pc->pc_lock);
                   2482:        pool_cache_cpu_exit(cc, s);
1.43      thorpej  2483:
1.101.2.6  yamt     2484:        object = pool_get(&pc->pc_pool, flags);
                   2485:        *objectp = object;
                   2486:        if (object == NULL)
                   2487:                return NULL;
                   2488:
                   2489:        if ((*pc->pc_ctor)(pc->pc_arg, object, flags) != 0) {
                   2490:                pool_put(&pc->pc_pool, object);
                   2491:                *objectp = NULL;
                   2492:                return NULL;
1.101.2.1  yamt     2493:        }
1.51      thorpej  2494:
1.101.2.6  yamt     2495:        KASSERT((((vaddr_t)object + pc->pc_pool.pr_itemoffset) &
                   2496:            (pc->pc_pool.pr_align - 1)) == 0);
                   2497:
                   2498:        if (pap != NULL) {
                   2499: #ifdef POOL_VTOPHYS
                   2500:                *pap = POOL_VTOPHYS(object);
                   2501: #else
                   2502:                *pap = POOL_PADDR_INVALID;
                   2503: #endif
                   2504:        }
1.51      thorpej  2505:
1.101.2.6  yamt     2506:        FREECHECK_OUT(&pc->pc_freecheck, object);
                   2507:        return NULL;
1.43      thorpej  2508: }
                   2509:
1.101.2.4  yamt     2510: /*
1.101.2.6  yamt     2511:  * pool_cache_get{,_paddr}:
1.101.2.4  yamt     2512:  *
1.101.2.6  yamt     2513:  *     Get an object from a pool cache (optionally returning
                   2514:  *     the physical address of the object).
1.101.2.4  yamt     2515:  */
1.101.2.6  yamt     2516: void *
                   2517: pool_cache_get_paddr(pool_cache_t pc, int flags, paddr_t *pap)
1.101.2.1  yamt     2518: {
1.101.2.6  yamt     2519:        pool_cache_cpu_t *cc;
                   2520:        pcg_t *pcg;
1.101.2.1  yamt     2521:        void *object;
1.101.2.6  yamt     2522:        int s;
1.101.2.1  yamt     2523:
1.101.2.6  yamt     2524: #ifdef LOCKDEBUG
                   2525:        if (flags & PR_WAITOK)
                   2526:                ASSERT_SLEEPABLE(NULL, "pool_cache_get(PR_WAITOK)");
                   2527: #endif
1.101.2.4  yamt     2528:
1.101.2.6  yamt     2529:        cc = pool_cache_cpu_enter(pc, &s);
                   2530:        do {
                   2531:                /* Try and allocate an object from the current group. */
                   2532:                pcg = cc->cc_current;
                   2533:                if (pcg != NULL && pcg->pcg_avail > 0) {
                   2534:                        object = pcg->pcg_objects[--pcg->pcg_avail].pcgo_va;
                   2535:                        if (pap != NULL)
                   2536:                                *pap = pcg->pcg_objects[pcg->pcg_avail].pcgo_pa;
1.101.2.9  yamt     2537: #if defined(DIAGNOSTIC)
1.101.2.6  yamt     2538:                        pcg->pcg_objects[pcg->pcg_avail].pcgo_va = NULL;
1.101.2.9  yamt     2539: #endif /* defined(DIAGNOSTIC) */
1.101.2.8  yamt     2540:                        KASSERT(pcg->pcg_avail <= pcg->pcg_size);
1.101.2.6  yamt     2541:                        KASSERT(object != NULL);
                   2542:                        cc->cc_hits++;
                   2543:                        pool_cache_cpu_exit(cc, &s);
                   2544:                        FREECHECK_OUT(&pc->pc_freecheck, object);
                   2545:                        return object;
1.101.2.1  yamt     2546:                }
1.101.2.4  yamt     2547:
1.101.2.6  yamt     2548:                /*
                   2549:                 * That failed.  If the previous group isn't empty, swap
                   2550:                 * it with the current group and allocate from there.
                   2551:                 */
                   2552:                pcg = cc->cc_previous;
                   2553:                if (pcg != NULL && pcg->pcg_avail > 0) {
                   2554:                        cc->cc_previous = cc->cc_current;
                   2555:                        cc->cc_current = pcg;
                   2556:                        continue;
                   2557:                }
                   2558:
                   2559:                /*
                   2560:                 * Can't allocate from either group: try the slow path.
                   2561:                 * If get_slow() allocated an object for us, or if
                   2562:                 * no more objects are available, it will return NULL.
                   2563:                 * Otherwise, we need to retry.
                   2564:                 */
                   2565:                cc = pool_cache_get_slow(cc, &s, &object, pap, flags);
                   2566:        } while (cc != NULL);
                   2567:
                   2568:        return object;
1.101.2.1  yamt     2569: }
                   2570:
1.101.2.6  yamt     2571: #if __GNUC_PREREQ__(3, 0)
                   2572: __attribute ((noinline))
                   2573: #endif
                   2574: pool_cache_cpu_t *
                   2575: pool_cache_put_slow(pool_cache_cpu_t *cc, int *s, void *object, paddr_t pa)
1.101.2.1  yamt     2576: {
1.101.2.6  yamt     2577:        pcg_t *pcg, *cur;
                   2578:        uint64_t ncsw;
                   2579:        pool_cache_t pc;
1.101.2.8  yamt     2580:        u_int nobj;
1.101.2.1  yamt     2581:
1.101.2.6  yamt     2582:        pc = cc->cc_cache;
                   2583:        cc->cc_misses++;
1.101.2.1  yamt     2584:
1.101.2.6  yamt     2585:        /*
                   2586:         * No free slots locally.  Try to grab an empty, unused
                   2587:         * group from the cache.
                   2588:         */
                   2589:        if (!mutex_tryenter(&pc->pc_lock)) {
                   2590:                ncsw = curlwp->l_ncsw;
                   2591:                mutex_enter(&pc->pc_lock);
                   2592:                pc->pc_contended++;
1.101.2.1  yamt     2593:
1.101.2.6  yamt     2594:                /*
                   2595:                 * If we context switched while locking, then
                   2596:                 * our view of the per-CPU data is invalid:
                   2597:                 * retry.
                   2598:                 */
                   2599:                if (curlwp->l_ncsw != ncsw) {
                   2600:                        mutex_exit(&pc->pc_lock);
                   2601:                        pool_cache_cpu_exit(cc, s);
                   2602:                        return pool_cache_cpu_enter(pc, s);
                   2603:                }
                   2604:        }
                   2605:
                   2606:        if ((pcg = pc->pc_emptygroups) != NULL) {
                   2607:                /*
                   2608:                 * If there's a empty group, release our full
                   2609:                 * group back to the cache.  Install the empty
1.101.2.8  yamt     2610:                 * group and return.
1.101.2.6  yamt     2611:                 */
                   2612:                KASSERT(pcg->pcg_avail == 0);
                   2613:                pc->pc_emptygroups = pcg->pcg_next;
1.101.2.8  yamt     2614:                if (cc->cc_previous == NULL) {
                   2615:                        cc->cc_previous = pcg;
                   2616:                } else {
                   2617:                        if ((cur = cc->cc_current) != NULL) {
                   2618:                                KASSERT(cur->pcg_avail == pcg->pcg_size);
                   2619:                                cur->pcg_next = pc->pc_fullgroups;
                   2620:                                pc->pc_fullgroups = cur;
                   2621:                                pc->pc_nfull++;
                   2622:                        }
                   2623:                        cc->cc_current = pcg;
                   2624:                }
1.101.2.6  yamt     2625:                pc->pc_hits++;
                   2626:                pc->pc_nempty--;
                   2627:                mutex_exit(&pc->pc_lock);
                   2628:                return cc;
                   2629:        }
                   2630:
                   2631:        /*
                   2632:         * Nothing available locally or in cache.  Take the
                   2633:         * slow path and try to allocate a new group that we
                   2634:         * can release to.
                   2635:         */
                   2636:        pc->pc_misses++;
                   2637:        mutex_exit(&pc->pc_lock);
                   2638:        pool_cache_cpu_exit(cc, s);
                   2639:
                   2640:        /*
                   2641:         * If we can't allocate a new group, just throw the
                   2642:         * object away.
                   2643:         */
1.101.2.8  yamt     2644:        nobj = pc->pc_pcgsize;
                   2645:        if (pool_cache_disable) {
                   2646:                pcg = NULL;
                   2647:        } else if (nobj == PCG_NOBJECTS_LARGE) {
                   2648:                pcg = pool_get(&pcg_large_pool, PR_NOWAIT);
                   2649:        } else {
                   2650:                pcg = pool_get(&pcg_normal_pool, PR_NOWAIT);
                   2651:        }
1.101.2.6  yamt     2652:        if (pcg == NULL) {
                   2653:                pool_cache_destruct_object(pc, object);
                   2654:                return NULL;
                   2655:        }
                   2656:        pcg->pcg_avail = 0;
1.101.2.8  yamt     2657:        pcg->pcg_size = nobj;
1.101.2.6  yamt     2658:
                   2659:        /*
                   2660:         * Add the empty group to the cache and try again.
                   2661:         */
                   2662:        mutex_enter(&pc->pc_lock);
                   2663:        pcg->pcg_next = pc->pc_emptygroups;
                   2664:        pc->pc_emptygroups = pcg;
                   2665:        pc->pc_nempty++;
                   2666:        mutex_exit(&pc->pc_lock);
                   2667:
                   2668:        return pool_cache_cpu_enter(pc, s);
                   2669: }
1.101.2.1  yamt     2670:
1.43      thorpej  2671: /*
1.101.2.6  yamt     2672:  * pool_cache_put{,_paddr}:
1.43      thorpej  2673:  *
1.101.2.6  yamt     2674:  *     Put an object back to the pool cache (optionally caching the
                   2675:  *     physical address of the object).
1.43      thorpej  2676:  */
1.101     thorpej  2677: void
1.101.2.6  yamt     2678: pool_cache_put_paddr(pool_cache_t pc, void *object, paddr_t pa)
1.43      thorpej  2679: {
1.101.2.6  yamt     2680:        pool_cache_cpu_t *cc;
                   2681:        pcg_t *pcg;
                   2682:        int s;
1.101     thorpej  2683:
1.101.2.6  yamt     2684:        FREECHECK_IN(&pc->pc_freecheck, object);
1.43      thorpej  2685:
1.101.2.6  yamt     2686:        cc = pool_cache_cpu_enter(pc, &s);
                   2687:        do {
                   2688:                /* If the current group isn't full, release it there. */
                   2689:                pcg = cc->cc_current;
1.101.2.8  yamt     2690:                if (pcg != NULL && pcg->pcg_avail < pcg->pcg_size) {
1.101.2.6  yamt     2691:                        pcg->pcg_objects[pcg->pcg_avail].pcgo_va = object;
                   2692:                        pcg->pcg_objects[pcg->pcg_avail].pcgo_pa = pa;
                   2693:                        pcg->pcg_avail++;
                   2694:                        cc->cc_hits++;
                   2695:                        pool_cache_cpu_exit(cc, &s);
                   2696:                        return;
                   2697:                }
1.43      thorpej  2698:
1.101.2.6  yamt     2699:                /*
                   2700:                 * That failed.  If the previous group is empty, swap
                   2701:                 * it with the current group and try again.
                   2702:                 */
                   2703:                pcg = cc->cc_previous;
                   2704:                if (pcg != NULL && pcg->pcg_avail == 0) {
                   2705:                        cc->cc_previous = cc->cc_current;
                   2706:                        cc->cc_current = pcg;
                   2707:                        continue;
                   2708:                }
1.43      thorpej  2709:
1.101.2.6  yamt     2710:                /*
                   2711:                 * Can't free to either group: try the slow path.
                   2712:                 * If put_slow() releases the object for us, it
                   2713:                 * will return NULL.  Otherwise we need to retry.
                   2714:                 */
                   2715:                cc = pool_cache_put_slow(cc, &s, object, pa);
                   2716:        } while (cc != NULL);
1.43      thorpej  2717: }
                   2718:
                   2719: /*
1.101.2.6  yamt     2720:  * pool_cache_xcall:
1.43      thorpej  2721:  *
1.101.2.6  yamt     2722:  *     Transfer objects from the per-CPU cache to the global cache.
                   2723:  *     Run within a cross-call thread.
1.43      thorpej  2724:  */
                   2725: static void
1.101.2.6  yamt     2726: pool_cache_xcall(pool_cache_t pc)
1.43      thorpej  2727: {
1.101.2.6  yamt     2728:        pool_cache_cpu_t *cc;
                   2729:        pcg_t *prev, *cur, **list;
                   2730:        int s = 0; /* XXXgcc */
                   2731:
                   2732:        cc = pool_cache_cpu_enter(pc, &s);
                   2733:        cur = cc->cc_current;
                   2734:        cc->cc_current = NULL;
                   2735:        prev = cc->cc_previous;
                   2736:        cc->cc_previous = NULL;
                   2737:        pool_cache_cpu_exit(cc, &s);
                   2738:
                   2739:        /*
                   2740:         * XXXSMP Go to splvm to prevent kernel_lock from being taken,
                   2741:         * because locks at IPL_SOFTXXX are still spinlocks.  Does not
                   2742:         * apply to IPL_SOFTBIO.  Cross-call threads do not take the
                   2743:         * kernel_lock.
1.101     thorpej  2744:         */
1.101.2.6  yamt     2745:        s = splvm();
                   2746:        mutex_enter(&pc->pc_lock);
                   2747:        if (cur != NULL) {
1.101.2.8  yamt     2748:                if (cur->pcg_avail == cur->pcg_size) {
1.101.2.6  yamt     2749:                        list = &pc->pc_fullgroups;
                   2750:                        pc->pc_nfull++;
                   2751:                } else if (cur->pcg_avail == 0) {
                   2752:                        list = &pc->pc_emptygroups;
                   2753:                        pc->pc_nempty++;
                   2754:                } else {
                   2755:                        list = &pc->pc_partgroups;
                   2756:                        pc->pc_npart++;
                   2757:                }
                   2758:                cur->pcg_next = *list;
                   2759:                *list = cur;
                   2760:        }
                   2761:        if (prev != NULL) {
1.101.2.8  yamt     2762:                if (prev->pcg_avail == prev->pcg_size) {
1.101.2.6  yamt     2763:                        list = &pc->pc_fullgroups;
                   2764:                        pc->pc_nfull++;
                   2765:                } else if (prev->pcg_avail == 0) {
                   2766:                        list = &pc->pc_emptygroups;
                   2767:                        pc->pc_nempty++;
                   2768:                } else {
                   2769:                        list = &pc->pc_partgroups;
                   2770:                        pc->pc_npart++;
                   2771:                }
                   2772:                prev->pcg_next = *list;
                   2773:                *list = prev;
                   2774:        }
                   2775:        mutex_exit(&pc->pc_lock);
                   2776:        splx(s);
1.3       pk       2777: }
1.66      thorpej  2778:
                   2779: /*
                   2780:  * Pool backend allocators.
                   2781:  *
                   2782:  * Each pool has a backend allocator that handles allocation, deallocation,
                   2783:  * and any additional draining that might be needed.
                   2784:  *
                   2785:  * We provide two standard allocators:
                   2786:  *
                   2787:  *     pool_allocator_kmem - the default when no allocator is specified
                   2788:  *
                   2789:  *     pool_allocator_nointr - used for pools that will not be accessed
                   2790:  *     in interrupt context.
                   2791:  */
                   2792: void   *pool_page_alloc(struct pool *, int);
                   2793: void   pool_page_free(struct pool *, void *);
                   2794:
1.101.2.1  yamt     2795: #ifdef POOL_SUBPAGE
                   2796: struct pool_allocator pool_allocator_kmem_fullpage = {
                   2797:        pool_page_alloc, pool_page_free, 0,
                   2798:        .pa_backingmapptr = &kmem_map,
                   2799: };
                   2800: #else
1.66      thorpej  2801: struct pool_allocator pool_allocator_kmem = {
                   2802:        pool_page_alloc, pool_page_free, 0,
1.101.2.1  yamt     2803:        .pa_backingmapptr = &kmem_map,
1.66      thorpej  2804: };
1.101.2.1  yamt     2805: #endif
1.66      thorpej  2806:
                   2807: void   *pool_page_alloc_nointr(struct pool *, int);
                   2808: void   pool_page_free_nointr(struct pool *, void *);
                   2809:
1.101.2.1  yamt     2810: #ifdef POOL_SUBPAGE
                   2811: struct pool_allocator pool_allocator_nointr_fullpage = {
                   2812:        pool_page_alloc_nointr, pool_page_free_nointr, 0,
                   2813:        .pa_backingmapptr = &kernel_map,
                   2814: };
                   2815: #else
1.66      thorpej  2816: struct pool_allocator pool_allocator_nointr = {
                   2817:        pool_page_alloc_nointr, pool_page_free_nointr, 0,
1.101.2.1  yamt     2818:        .pa_backingmapptr = &kernel_map,
1.66      thorpej  2819: };
1.101.2.1  yamt     2820: #endif
1.66      thorpej  2821:
                   2822: #ifdef POOL_SUBPAGE
                   2823: void   *pool_subpage_alloc(struct pool *, int);
                   2824: void   pool_subpage_free(struct pool *, void *);
                   2825:
1.101.2.1  yamt     2826: struct pool_allocator pool_allocator_kmem = {
                   2827:        pool_subpage_alloc, pool_subpage_free, POOL_SUBPAGE,
                   2828:        .pa_backingmapptr = &kmem_map,
                   2829: };
                   2830:
                   2831: void   *pool_subpage_alloc_nointr(struct pool *, int);
                   2832: void   pool_subpage_free_nointr(struct pool *, void *);
                   2833:
                   2834: struct pool_allocator pool_allocator_nointr = {
                   2835:        pool_subpage_alloc, pool_subpage_free, POOL_SUBPAGE,
                   2836:        .pa_backingmapptr = &kmem_map,
1.66      thorpej  2837: };
                   2838: #endif /* POOL_SUBPAGE */
                   2839:
1.101.2.1  yamt     2840: static void *
                   2841: pool_allocator_alloc(struct pool *pp, int flags)
1.66      thorpej  2842: {
1.101.2.1  yamt     2843:        struct pool_allocator *pa = pp->pr_alloc;
1.66      thorpej  2844:        void *res;
                   2845:
1.101.2.1  yamt     2846:        res = (*pa->pa_alloc)(pp, flags);
                   2847:        if (res == NULL && (flags & PR_WAITOK) == 0) {
1.66      thorpej  2848:                /*
1.101.2.1  yamt     2849:                 * We only run the drain hook here if PR_NOWAIT.
                   2850:                 * In other cases, the hook will be run in
                   2851:                 * pool_reclaim().
1.66      thorpej  2852:                 */
1.101.2.1  yamt     2853:                if (pp->pr_drain_hook != NULL) {
                   2854:                        (*pp->pr_drain_hook)(pp->pr_drain_hook_arg, flags);
                   2855:                        res = (*pa->pa_alloc)(pp, flags);
1.66      thorpej  2856:                }
1.101.2.1  yamt     2857:        }
                   2858:        return res;
1.66      thorpej  2859: }
                   2860:
1.101.2.1  yamt     2861: static void
1.66      thorpej  2862: pool_allocator_free(struct pool *pp, void *v)
                   2863: {
                   2864:        struct pool_allocator *pa = pp->pr_alloc;
                   2865:
                   2866:        (*pa->pa_free)(pp, v);
                   2867: }
                   2868:
                   2869: void *
                   2870: pool_page_alloc(struct pool *pp, int flags)
                   2871: {
1.101.2.3  yamt     2872:        bool waitok = (flags & PR_WAITOK) ? true : false;
1.66      thorpej  2873:
1.100     yamt     2874:        return ((void *) uvm_km_alloc_poolpage_cache(kmem_map, waitok));
1.66      thorpej  2875: }
                   2876:
                   2877: void
                   2878: pool_page_free(struct pool *pp, void *v)
                   2879: {
                   2880:
1.98      yamt     2881:        uvm_km_free_poolpage_cache(kmem_map, (vaddr_t) v);
                   2882: }
                   2883:
                   2884: static void *
                   2885: pool_page_alloc_meta(struct pool *pp, int flags)
                   2886: {
1.101.2.3  yamt     2887:        bool waitok = (flags & PR_WAITOK) ? true : false;
1.98      yamt     2888:
1.100     yamt     2889:        return ((void *) uvm_km_alloc_poolpage(kmem_map, waitok));
1.98      yamt     2890: }
                   2891:
                   2892: static void
                   2893: pool_page_free_meta(struct pool *pp, void *v)
                   2894: {
                   2895:
1.100     yamt     2896:        uvm_km_free_poolpage(kmem_map, (vaddr_t) v);
1.66      thorpej  2897: }
                   2898:
                   2899: #ifdef POOL_SUBPAGE
                   2900: /* Sub-page allocator, for machines with large hardware pages. */
                   2901: void *
                   2902: pool_subpage_alloc(struct pool *pp, int flags)
                   2903: {
1.101.2.6  yamt     2904:        return pool_get(&psppool, flags);
1.66      thorpej  2905: }
                   2906:
                   2907: void
                   2908: pool_subpage_free(struct pool *pp, void *v)
                   2909: {
                   2910:        pool_put(&psppool, v);
                   2911: }
                   2912:
                   2913: /* We don't provide a real nointr allocator.  Maybe later. */
                   2914: void *
1.101.2.1  yamt     2915: pool_subpage_alloc_nointr(struct pool *pp, int flags)
1.66      thorpej  2916: {
                   2917:
                   2918:        return (pool_subpage_alloc(pp, flags));
                   2919: }
                   2920:
                   2921: void
1.101.2.1  yamt     2922: pool_subpage_free_nointr(struct pool *pp, void *v)
1.66      thorpej  2923: {
                   2924:
                   2925:        pool_subpage_free(pp, v);
                   2926: }
1.101.2.1  yamt     2927: #endif /* POOL_SUBPAGE */
1.66      thorpej  2928: void *
                   2929: pool_page_alloc_nointr(struct pool *pp, int flags)
                   2930: {
1.101.2.3  yamt     2931:        bool waitok = (flags & PR_WAITOK) ? true : false;
1.66      thorpej  2932:
1.100     yamt     2933:        return ((void *) uvm_km_alloc_poolpage_cache(kernel_map, waitok));
1.66      thorpej  2934: }
                   2935:
                   2936: void
                   2937: pool_page_free_nointr(struct pool *pp, void *v)
                   2938: {
                   2939:
1.98      yamt     2940:        uvm_km_free_poolpage_cache(kernel_map, (vaddr_t) v);
1.66      thorpej  2941: }
1.101.2.8  yamt     2942:
                   2943: #if defined(DDB)
                   2944: static bool
                   2945: pool_in_page(struct pool *pp, struct pool_item_header *ph, uintptr_t addr)
                   2946: {
                   2947:
                   2948:        return (uintptr_t)ph->ph_page <= addr &&
                   2949:            addr < (uintptr_t)ph->ph_page + pp->pr_alloc->pa_pagesz;
                   2950: }
                   2951:
                   2952: static bool
                   2953: pool_in_item(struct pool *pp, void *item, uintptr_t addr)
                   2954: {
                   2955:
                   2956:        return (uintptr_t)item <= addr && addr < (uintptr_t)item + pp->pr_size;
                   2957: }
                   2958:
                   2959: static bool
                   2960: pool_in_cg(struct pool *pp, struct pool_cache_group *pcg, uintptr_t addr)
                   2961: {
                   2962:        int i;
                   2963:
                   2964:        if (pcg == NULL) {
                   2965:                return false;
                   2966:        }
                   2967:        for (i = 0; i < pcg->pcg_avail; i++) {
                   2968:                if (pool_in_item(pp, pcg->pcg_objects[i].pcgo_va, addr)) {
                   2969:                        return true;
                   2970:                }
                   2971:        }
                   2972:        return false;
                   2973: }
                   2974:
                   2975: static bool
                   2976: pool_allocated(struct pool *pp, struct pool_item_header *ph, uintptr_t addr)
                   2977: {
                   2978:
                   2979:        if ((pp->pr_roflags & PR_NOTOUCH) != 0) {
                   2980:                unsigned int idx = pr_item_notouch_index(pp, ph, (void *)addr);
                   2981:                pool_item_bitmap_t *bitmap =
                   2982:                    ph->ph_bitmap + (idx / BITMAP_SIZE);
                   2983:                pool_item_bitmap_t mask = 1 << (idx & BITMAP_MASK);
                   2984:
                   2985:                return (*bitmap & mask) == 0;
                   2986:        } else {
                   2987:                struct pool_item *pi;
                   2988:
                   2989:                LIST_FOREACH(pi, &ph->ph_itemlist, pi_list) {
                   2990:                        if (pool_in_item(pp, pi, addr)) {
                   2991:                                return false;
                   2992:                        }
                   2993:                }
                   2994:                return true;
                   2995:        }
                   2996: }
                   2997:
                   2998: void
                   2999: pool_whatis(uintptr_t addr, void (*pr)(const char *, ...))
                   3000: {
                   3001:        struct pool *pp;
                   3002:
                   3003:        TAILQ_FOREACH(pp, &pool_head, pr_poollist) {
                   3004:                struct pool_item_header *ph;
                   3005:                uintptr_t item;
                   3006:                bool allocated = true;
                   3007:                bool incache = false;
                   3008:                bool incpucache = false;
                   3009:                char cpucachestr[32];
                   3010:
                   3011:                if ((pp->pr_roflags & PR_PHINPAGE) != 0) {
                   3012:                        LIST_FOREACH(ph, &pp->pr_fullpages, ph_pagelist) {
                   3013:                                if (pool_in_page(pp, ph, addr)) {
                   3014:                                        goto found;
                   3015:                                }
                   3016:                        }
                   3017:                        LIST_FOREACH(ph, &pp->pr_partpages, ph_pagelist) {
                   3018:                                if (pool_in_page(pp, ph, addr)) {
                   3019:                                        allocated =
                   3020:                                            pool_allocated(pp, ph, addr);
                   3021:                                        goto found;
                   3022:                                }
                   3023:                        }
                   3024:                        LIST_FOREACH(ph, &pp->pr_emptypages, ph_pagelist) {
                   3025:                                if (pool_in_page(pp, ph, addr)) {
                   3026:                                        allocated = false;
                   3027:                                        goto found;
                   3028:                                }
                   3029:                        }
                   3030:                        continue;
                   3031:                } else {
                   3032:                        ph = pr_find_pagehead_noalign(pp, (void *)addr);
                   3033:                        if (ph == NULL || !pool_in_page(pp, ph, addr)) {
                   3034:                                continue;
                   3035:                        }
                   3036:                        allocated = pool_allocated(pp, ph, addr);
                   3037:                }
                   3038: found:
                   3039:                if (allocated && pp->pr_cache) {
                   3040:                        pool_cache_t pc = pp->pr_cache;
                   3041:                        struct pool_cache_group *pcg;
                   3042:                        int i;
                   3043:
                   3044:                        for (pcg = pc->pc_fullgroups; pcg != NULL;
                   3045:                            pcg = pcg->pcg_next) {
                   3046:                                if (pool_in_cg(pp, pcg, addr)) {
                   3047:                                        incache = true;
                   3048:                                        goto print;
                   3049:                                }
                   3050:                        }
                   3051:                        for (i = 0; i < MAXCPUS; i++) {
                   3052:                                pool_cache_cpu_t *cc;
                   3053:
                   3054:                                if ((cc = pc->pc_cpus[i]) == NULL) {
                   3055:                                        continue;
                   3056:                                }
                   3057:                                if (pool_in_cg(pp, cc->cc_current, addr) ||
                   3058:                                    pool_in_cg(pp, cc->cc_previous, addr)) {
                   3059:                                        struct cpu_info *ci =
                   3060:                                            cpu_lookup_byindex(i);
                   3061:
                   3062:                                        incpucache = true;
                   3063:                                        snprintf(cpucachestr,
                   3064:                                            sizeof(cpucachestr),
                   3065:                                            "cached by CPU %u",
                   3066:                                            (u_int)ci->ci_cpuid);
                   3067:                                        goto print;
                   3068:                                }
                   3069:                        }
                   3070:                }
                   3071: print:
                   3072:                item = (uintptr_t)ph->ph_page + ph->ph_off;
                   3073:                item = item + rounddown(addr - item, pp->pr_size);
                   3074:                (*pr)("%p is %p+%zu in POOL '%s' (%s)\n",
                   3075:                    (void *)addr, item, (size_t)(addr - item),
                   3076:                    pp->pr_wchan,
                   3077:                    incpucache ? cpucachestr :
                   3078:                    incache ? "cached" : allocated ? "allocated" : "free");
                   3079:        }
                   3080: }
                   3081: #endif /* defined(DDB) */

CVSweb <webmaster@jp.NetBSD.org>