[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 1.140

1.140   ! yamt        1: /*     $NetBSD: subr_pool.c,v 1.139 2007/12/11 19:07:27 ad Exp $       */
1.1       pk          2:
                      3: /*-
1.134     ad          4:  * Copyright (c) 1997, 1999, 2000, 2002, 2007 The NetBSD Foundation, Inc.
1.1       pk          5:  * All rights reserved.
                      6:  *
                      7:  * This code is derived from software contributed to The NetBSD Foundation
1.20      thorpej     8:  * by Paul Kranenburg; by Jason R. Thorpe of the Numerical Aerospace
1.134     ad          9:  * Simulation Facility, NASA Ames Research Center, and by Andrew Doran.
1.1       pk         10:  *
                     11:  * Redistribution and use in source and binary forms, with or without
                     12:  * modification, are permitted provided that the following conditions
                     13:  * are met:
                     14:  * 1. Redistributions of source code must retain the above copyright
                     15:  *    notice, this list of conditions and the following disclaimer.
                     16:  * 2. Redistributions in binary form must reproduce the above copyright
                     17:  *    notice, this list of conditions and the following disclaimer in the
                     18:  *    documentation and/or other materials provided with the distribution.
                     19:  * 3. All advertising materials mentioning features or use of this software
                     20:  *    must display the following acknowledgement:
1.13      christos   21:  *     This product includes software developed by the NetBSD
                     22:  *     Foundation, Inc. and its contributors.
1.1       pk         23:  * 4. Neither the name of The NetBSD Foundation nor the names of its
                     24:  *    contributors may be used to endorse or promote products derived
                     25:  *    from this software without specific prior written permission.
                     26:  *
                     27:  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
                     28:  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
                     29:  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
                     30:  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
                     31:  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
                     32:  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
                     33:  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
                     34:  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
                     35:  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
                     36:  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
                     37:  * POSSIBILITY OF SUCH DAMAGE.
                     38:  */
1.64      lukem      39:
                     40: #include <sys/cdefs.h>
1.140   ! yamt       41: __KERNEL_RCSID(0, "$NetBSD: subr_pool.c,v 1.139 2007/12/11 19:07:27 ad Exp $");
1.24      scottr     42:
1.25      thorpej    43: #include "opt_pool.h"
1.24      scottr     44: #include "opt_poollog.h"
1.28      thorpej    45: #include "opt_lockdebug.h"
1.1       pk         46:
                     47: #include <sys/param.h>
                     48: #include <sys/systm.h>
1.135     yamt       49: #include <sys/bitops.h>
1.1       pk         50: #include <sys/proc.h>
                     51: #include <sys/errno.h>
                     52: #include <sys/kernel.h>
                     53: #include <sys/malloc.h>
                     54: #include <sys/lock.h>
                     55: #include <sys/pool.h>
1.20      thorpej    56: #include <sys/syslog.h>
1.125     ad         57: #include <sys/debug.h>
1.134     ad         58: #include <sys/lockdebug.h>
                     59: #include <sys/xcall.h>
                     60: #include <sys/cpu.h>
1.3       pk         61:
                     62: #include <uvm/uvm.h>
                     63:
1.1       pk         64: /*
                     65:  * Pool resource management utility.
1.3       pk         66:  *
1.88      chs        67:  * Memory is allocated in pages which are split into pieces according to
                     68:  * the pool item size. Each page is kept on one of three lists in the
                     69:  * pool structure: `pr_emptypages', `pr_fullpages' and `pr_partpages',
                     70:  * for empty, full and partially-full pages respectively. The individual
                     71:  * pool items are on a linked list headed by `ph_itemlist' in each page
                     72:  * header. The memory for building the page list is either taken from
                     73:  * the allocated pages themselves (for small pool items) or taken from
                     74:  * an internal pool of page headers (`phpool').
1.1       pk         75:  */
                     76:
1.3       pk         77: /* List of all pools */
1.102     chs        78: LIST_HEAD(,pool) pool_head = LIST_HEAD_INITIALIZER(pool_head);
1.3       pk         79:
1.134     ad         80: /* List of all caches. */
                     81: LIST_HEAD(,pool_cache) pool_cache_head =
                     82:     LIST_HEAD_INITIALIZER(pool_cache_head);
                     83:
1.3       pk         84: /* Private pool for page header structures */
1.97      yamt       85: #define        PHPOOL_MAX      8
                     86: static struct pool phpool[PHPOOL_MAX];
1.135     yamt       87: #define        PHPOOL_FREELIST_NELEM(idx) \
                     88:        (((idx) == 0) ? 0 : BITMAP_SIZE * (1 << (idx)))
1.3       pk         89:
1.62      bjh21      90: #ifdef POOL_SUBPAGE
                     91: /* Pool of subpages for use by normal pools. */
                     92: static struct pool psppool;
                     93: #endif
                     94:
1.117     yamt       95: static SLIST_HEAD(, pool_allocator) pa_deferinitq =
                     96:     SLIST_HEAD_INITIALIZER(pa_deferinitq);
                     97:
1.98      yamt       98: static void *pool_page_alloc_meta(struct pool *, int);
                     99: static void pool_page_free_meta(struct pool *, void *);
                    100:
                    101: /* allocator for pool metadata */
1.134     ad        102: struct pool_allocator pool_allocator_meta = {
1.117     yamt      103:        pool_page_alloc_meta, pool_page_free_meta,
                    104:        .pa_backingmapptr = &kmem_map,
1.98      yamt      105: };
                    106:
1.3       pk        107: /* # of seconds to retain page after last use */
                    108: int pool_inactive_time = 10;
                    109:
                    110: /* Next candidate for drainage (see pool_drain()) */
1.23      thorpej   111: static struct pool     *drainpp;
                    112:
1.134     ad        113: /* This lock protects both pool_head and drainpp. */
                    114: static kmutex_t pool_head_lock;
                    115: static kcondvar_t pool_busy;
1.3       pk        116:
1.135     yamt      117: typedef uint32_t pool_item_bitmap_t;
                    118: #define        BITMAP_SIZE     (CHAR_BIT * sizeof(pool_item_bitmap_t))
                    119: #define        BITMAP_MASK     (BITMAP_SIZE - 1)
1.99      yamt      120:
1.3       pk        121: struct pool_item_header {
                    122:        /* Page headers */
1.88      chs       123:        LIST_ENTRY(pool_item_header)
1.3       pk        124:                                ph_pagelist;    /* pool page list */
1.88      chs       125:        SPLAY_ENTRY(pool_item_header)
                    126:                                ph_node;        /* Off-page page headers */
1.128     christos  127:        void *                  ph_page;        /* this page's address */
1.3       pk        128:        struct timeval          ph_time;        /* last referenced */
1.135     yamt      129:        uint16_t                ph_nmissing;    /* # of chunks in use */
1.97      yamt      130:        union {
                    131:                /* !PR_NOTOUCH */
                    132:                struct {
1.102     chs       133:                        LIST_HEAD(, pool_item)
1.97      yamt      134:                                phu_itemlist;   /* chunk list for this page */
                    135:                } phu_normal;
                    136:                /* PR_NOTOUCH */
                    137:                struct {
1.135     yamt      138:                        uint16_t phu_off;       /* start offset in page */
                    139:                        pool_item_bitmap_t phu_bitmap[];
1.97      yamt      140:                } phu_notouch;
                    141:        } ph_u;
1.3       pk        142: };
1.97      yamt      143: #define        ph_itemlist     ph_u.phu_normal.phu_itemlist
                    144: #define        ph_off          ph_u.phu_notouch.phu_off
1.135     yamt      145: #define        ph_bitmap       ph_u.phu_notouch.phu_bitmap
1.3       pk        146:
1.1       pk        147: struct pool_item {
1.3       pk        148: #ifdef DIAGNOSTIC
1.82      thorpej   149:        u_int pi_magic;
1.33      chs       150: #endif
1.134     ad        151: #define        PI_MAGIC 0xdeaddeadU
1.3       pk        152:        /* Other entries use only this list entry */
1.102     chs       153:        LIST_ENTRY(pool_item)   pi_list;
1.3       pk        154: };
                    155:
1.53      thorpej   156: #define        POOL_NEEDS_CATCHUP(pp)                                          \
                    157:        ((pp)->pr_nitems < (pp)->pr_minitems)
                    158:
1.43      thorpej   159: /*
                    160:  * Pool cache management.
                    161:  *
                    162:  * Pool caches provide a way for constructed objects to be cached by the
                    163:  * pool subsystem.  This can lead to performance improvements by avoiding
                    164:  * needless object construction/destruction; it is deferred until absolutely
                    165:  * necessary.
                    166:  *
1.134     ad        167:  * Caches are grouped into cache groups.  Each cache group references up
                    168:  * to PCG_NUMOBJECTS constructed objects.  When a cache allocates an
                    169:  * object from the pool, it calls the object's constructor and places it
                    170:  * into a cache group.  When a cache group frees an object back to the
                    171:  * pool, it first calls the object's destructor.  This allows the object
                    172:  * to persist in constructed form while freed to the cache.
                    173:  *
                    174:  * The pool references each cache, so that when a pool is drained by the
                    175:  * pagedaemon, it can drain each individual cache as well.  Each time a
                    176:  * cache is drained, the most idle cache group is freed to the pool in
                    177:  * its entirety.
1.43      thorpej   178:  *
                    179:  * Pool caches are layed on top of pools.  By layering them, we can avoid
                    180:  * the complexity of cache management for pools which would not benefit
                    181:  * from it.
                    182:  */
                    183:
                    184: static struct pool pcgpool;
1.134     ad        185: static struct pool cache_pool;
                    186: static struct pool cache_cpu_pool;
1.3       pk        187:
1.134     ad        188: static pool_cache_cpu_t *pool_cache_put_slow(pool_cache_cpu_t *, int *,
                    189:                                             void *, paddr_t);
                    190: static pool_cache_cpu_t *pool_cache_get_slow(pool_cache_cpu_t *, int *,
                    191:                                             void **, paddr_t *, int);
                    192: static void    pool_cache_cpu_init1(struct cpu_info *, pool_cache_t);
                    193: static void    pool_cache_invalidate_groups(pool_cache_t, pcg_t *);
                    194: static void    pool_cache_xcall(pool_cache_t);
1.3       pk        195:
1.42      thorpej   196: static int     pool_catchup(struct pool *);
1.128     christos  197: static void    pool_prime_page(struct pool *, void *,
1.55      thorpej   198:                    struct pool_item_header *);
1.88      chs       199: static void    pool_update_curpage(struct pool *);
1.66      thorpej   200:
1.113     yamt      201: static int     pool_grow(struct pool *, int);
1.117     yamt      202: static void    *pool_allocator_alloc(struct pool *, int);
                    203: static void    pool_allocator_free(struct pool *, void *);
1.3       pk        204:
1.97      yamt      205: static void pool_print_pagelist(struct pool *, struct pool_pagelist *,
1.88      chs       206:        void (*)(const char *, ...));
1.42      thorpej   207: static void pool_print1(struct pool *, const char *,
                    208:        void (*)(const char *, ...));
1.3       pk        209:
1.88      chs       210: static int pool_chk_page(struct pool *, const char *,
                    211:                         struct pool_item_header *);
                    212:
1.3       pk        213: /*
1.52      thorpej   214:  * Pool log entry. An array of these is allocated in pool_init().
1.3       pk        215:  */
                    216: struct pool_log {
                    217:        const char      *pl_file;
                    218:        long            pl_line;
                    219:        int             pl_action;
1.25      thorpej   220: #define        PRLOG_GET       1
                    221: #define        PRLOG_PUT       2
1.3       pk        222:        void            *pl_addr;
1.1       pk        223: };
                    224:
1.86      matt      225: #ifdef POOL_DIAGNOSTIC
1.3       pk        226: /* Number of entries in pool log buffers */
1.17      thorpej   227: #ifndef POOL_LOGSIZE
                    228: #define        POOL_LOGSIZE    10
                    229: #endif
                    230:
                    231: int pool_logsize = POOL_LOGSIZE;
1.1       pk        232:
1.110     perry     233: static inline void
1.42      thorpej   234: pr_log(struct pool *pp, void *v, int action, const char *file, long line)
1.3       pk        235: {
                    236:        int n = pp->pr_curlogentry;
                    237:        struct pool_log *pl;
                    238:
1.20      thorpej   239:        if ((pp->pr_roflags & PR_LOGGING) == 0)
1.3       pk        240:                return;
                    241:
                    242:        /*
                    243:         * Fill in the current entry. Wrap around and overwrite
                    244:         * the oldest entry if necessary.
                    245:         */
                    246:        pl = &pp->pr_log[n];
                    247:        pl->pl_file = file;
                    248:        pl->pl_line = line;
                    249:        pl->pl_action = action;
                    250:        pl->pl_addr = v;
                    251:        if (++n >= pp->pr_logsize)
                    252:                n = 0;
                    253:        pp->pr_curlogentry = n;
                    254: }
                    255:
                    256: static void
1.42      thorpej   257: pr_printlog(struct pool *pp, struct pool_item *pi,
                    258:     void (*pr)(const char *, ...))
1.3       pk        259: {
                    260:        int i = pp->pr_logsize;
                    261:        int n = pp->pr_curlogentry;
                    262:
1.20      thorpej   263:        if ((pp->pr_roflags & PR_LOGGING) == 0)
1.3       pk        264:                return;
                    265:
                    266:        /*
                    267:         * Print all entries in this pool's log.
                    268:         */
                    269:        while (i-- > 0) {
                    270:                struct pool_log *pl = &pp->pr_log[n];
                    271:                if (pl->pl_action != 0) {
1.25      thorpej   272:                        if (pi == NULL || pi == pl->pl_addr) {
                    273:                                (*pr)("\tlog entry %d:\n", i);
                    274:                                (*pr)("\t\taction = %s, addr = %p\n",
                    275:                                    pl->pl_action == PRLOG_GET ? "get" : "put",
                    276:                                    pl->pl_addr);
                    277:                                (*pr)("\t\tfile: %s at line %lu\n",
                    278:                                    pl->pl_file, pl->pl_line);
                    279:                        }
1.3       pk        280:                }
                    281:                if (++n >= pp->pr_logsize)
                    282:                        n = 0;
                    283:        }
                    284: }
1.25      thorpej   285:
1.110     perry     286: static inline void
1.42      thorpej   287: pr_enter(struct pool *pp, const char *file, long line)
1.25      thorpej   288: {
                    289:
1.34      thorpej   290:        if (__predict_false(pp->pr_entered_file != NULL)) {
1.25      thorpej   291:                printf("pool %s: reentrancy at file %s line %ld\n",
                    292:                    pp->pr_wchan, file, line);
                    293:                printf("         previous entry at file %s line %ld\n",
                    294:                    pp->pr_entered_file, pp->pr_entered_line);
                    295:                panic("pr_enter");
                    296:        }
                    297:
                    298:        pp->pr_entered_file = file;
                    299:        pp->pr_entered_line = line;
                    300: }
                    301:
1.110     perry     302: static inline void
1.42      thorpej   303: pr_leave(struct pool *pp)
1.25      thorpej   304: {
                    305:
1.34      thorpej   306:        if (__predict_false(pp->pr_entered_file == NULL)) {
1.25      thorpej   307:                printf("pool %s not entered?\n", pp->pr_wchan);
                    308:                panic("pr_leave");
                    309:        }
                    310:
                    311:        pp->pr_entered_file = NULL;
                    312:        pp->pr_entered_line = 0;
                    313: }
                    314:
1.110     perry     315: static inline void
1.42      thorpej   316: pr_enter_check(struct pool *pp, void (*pr)(const char *, ...))
1.25      thorpej   317: {
                    318:
                    319:        if (pp->pr_entered_file != NULL)
                    320:                (*pr)("\n\tcurrently entered from file %s line %ld\n",
                    321:                    pp->pr_entered_file, pp->pr_entered_line);
                    322: }
1.3       pk        323: #else
1.25      thorpej   324: #define        pr_log(pp, v, action, file, line)
                    325: #define        pr_printlog(pp, pi, pr)
                    326: #define        pr_enter(pp, file, line)
                    327: #define        pr_leave(pp)
                    328: #define        pr_enter_check(pp, pr)
1.59      thorpej   329: #endif /* POOL_DIAGNOSTIC */
1.3       pk        330:
1.135     yamt      331: static inline unsigned int
1.97      yamt      332: pr_item_notouch_index(const struct pool *pp, const struct pool_item_header *ph,
                    333:     const void *v)
                    334: {
                    335:        const char *cp = v;
1.135     yamt      336:        unsigned int idx;
1.97      yamt      337:
                    338:        KASSERT(pp->pr_roflags & PR_NOTOUCH);
1.128     christos  339:        idx = (cp - (char *)ph->ph_page - ph->ph_off) / pp->pr_size;
1.97      yamt      340:        KASSERT(idx < pp->pr_itemsperpage);
                    341:        return idx;
                    342: }
                    343:
1.110     perry     344: static inline void
1.97      yamt      345: pr_item_notouch_put(const struct pool *pp, struct pool_item_header *ph,
                    346:     void *obj)
                    347: {
1.135     yamt      348:        unsigned int idx = pr_item_notouch_index(pp, ph, obj);
                    349:        pool_item_bitmap_t *bitmap = ph->ph_bitmap + (idx / BITMAP_SIZE);
                    350:        pool_item_bitmap_t mask = 1 << (idx & BITMAP_MASK);
1.97      yamt      351:
1.135     yamt      352:        KASSERT((*bitmap & mask) == 0);
                    353:        *bitmap |= mask;
1.97      yamt      354: }
                    355:
1.110     perry     356: static inline void *
1.97      yamt      357: pr_item_notouch_get(const struct pool *pp, struct pool_item_header *ph)
                    358: {
1.135     yamt      359:        pool_item_bitmap_t *bitmap = ph->ph_bitmap;
                    360:        unsigned int idx;
                    361:        int i;
1.97      yamt      362:
1.135     yamt      363:        for (i = 0; ; i++) {
                    364:                int bit;
1.97      yamt      365:
1.135     yamt      366:                KASSERT((i * BITMAP_SIZE) < pp->pr_itemsperpage);
                    367:                bit = ffs32(bitmap[i]);
                    368:                if (bit) {
                    369:                        pool_item_bitmap_t mask;
                    370:
                    371:                        bit--;
                    372:                        idx = (i * BITMAP_SIZE) + bit;
                    373:                        mask = 1 << bit;
                    374:                        KASSERT((bitmap[i] & mask) != 0);
                    375:                        bitmap[i] &= ~mask;
                    376:                        break;
                    377:                }
                    378:        }
                    379:        KASSERT(idx < pp->pr_itemsperpage);
1.128     christos  380:        return (char *)ph->ph_page + ph->ph_off + idx * pp->pr_size;
1.97      yamt      381: }
                    382:
1.135     yamt      383: static inline void
1.140   ! yamt      384: pr_item_notouch_init(const struct pool *pp, struct pool_item_header *ph,
        !           385:     unsigned int offset)
1.135     yamt      386: {
                    387:        pool_item_bitmap_t *bitmap = ph->ph_bitmap;
                    388:        const int n = howmany(pp->pr_itemsperpage, BITMAP_SIZE);
                    389:        int i;
                    390:
1.140   ! yamt      391:        ph->ph_off = offset;
1.135     yamt      392:        for (i = 0; i < n; i++) {
                    393:                bitmap[i] = (pool_item_bitmap_t)-1;
                    394:        }
                    395: }
                    396:
1.110     perry     397: static inline int
1.88      chs       398: phtree_compare(struct pool_item_header *a, struct pool_item_header *b)
                    399: {
1.121     yamt      400:
                    401:        /*
                    402:         * we consider pool_item_header with smaller ph_page bigger.
                    403:         * (this unnatural ordering is for the benefit of pr_find_pagehead.)
                    404:         */
                    405:
1.88      chs       406:        if (a->ph_page < b->ph_page)
1.121     yamt      407:                return (1);
                    408:        else if (a->ph_page > b->ph_page)
1.88      chs       409:                return (-1);
                    410:        else
                    411:                return (0);
                    412: }
                    413:
                    414: SPLAY_PROTOTYPE(phtree, pool_item_header, ph_node, phtree_compare);
                    415: SPLAY_GENERATE(phtree, pool_item_header, ph_node, phtree_compare);
                    416:
1.3       pk        417: /*
1.121     yamt      418:  * Return the pool page header based on item address.
1.3       pk        419:  */
1.110     perry     420: static inline struct pool_item_header *
1.121     yamt      421: pr_find_pagehead(struct pool *pp, void *v)
1.3       pk        422: {
1.88      chs       423:        struct pool_item_header *ph, tmp;
1.3       pk        424:
1.121     yamt      425:        if ((pp->pr_roflags & PR_NOALIGN) != 0) {
1.128     christos  426:                tmp.ph_page = (void *)(uintptr_t)v;
1.121     yamt      427:                ph = SPLAY_FIND(phtree, &pp->pr_phtree, &tmp);
                    428:                if (ph == NULL) {
                    429:                        ph = SPLAY_ROOT(&pp->pr_phtree);
                    430:                        if (ph != NULL && phtree_compare(&tmp, ph) >= 0) {
                    431:                                ph = SPLAY_NEXT(phtree, &pp->pr_phtree, ph);
                    432:                        }
                    433:                        KASSERT(ph == NULL || phtree_compare(&tmp, ph) < 0);
                    434:                }
                    435:        } else {
1.128     christos  436:                void *page =
                    437:                    (void *)((uintptr_t)v & pp->pr_alloc->pa_pagemask);
1.121     yamt      438:
                    439:                if ((pp->pr_roflags & PR_PHINPAGE) != 0) {
1.128     christos  440:                        ph = (struct pool_item_header *)((char *)page + pp->pr_phoffset);
1.121     yamt      441:                } else {
                    442:                        tmp.ph_page = page;
                    443:                        ph = SPLAY_FIND(phtree, &pp->pr_phtree, &tmp);
                    444:                }
                    445:        }
1.3       pk        446:
1.121     yamt      447:        KASSERT(ph == NULL || ((pp->pr_roflags & PR_PHINPAGE) != 0) ||
1.128     christos  448:            ((char *)ph->ph_page <= (char *)v &&
                    449:            (char *)v < (char *)ph->ph_page + pp->pr_alloc->pa_pagesz));
1.88      chs       450:        return ph;
1.3       pk        451: }
                    452:
1.101     thorpej   453: static void
                    454: pr_pagelist_free(struct pool *pp, struct pool_pagelist *pq)
                    455: {
                    456:        struct pool_item_header *ph;
                    457:
                    458:        while ((ph = LIST_FIRST(pq)) != NULL) {
                    459:                LIST_REMOVE(ph, ph_pagelist);
                    460:                pool_allocator_free(pp, ph->ph_page);
1.134     ad        461:                if ((pp->pr_roflags & PR_PHINPAGE) == 0)
1.101     thorpej   462:                        pool_put(pp->pr_phpool, ph);
                    463:        }
                    464: }
                    465:
1.3       pk        466: /*
                    467:  * Remove a page from the pool.
                    468:  */
1.110     perry     469: static inline void
1.61      chs       470: pr_rmpage(struct pool *pp, struct pool_item_header *ph,
                    471:      struct pool_pagelist *pq)
1.3       pk        472: {
                    473:
1.134     ad        474:        KASSERT(mutex_owned(&pp->pr_lock));
1.91      yamt      475:
1.3       pk        476:        /*
1.7       thorpej   477:         * If the page was idle, decrement the idle page count.
1.3       pk        478:         */
1.6       thorpej   479:        if (ph->ph_nmissing == 0) {
                    480: #ifdef DIAGNOSTIC
                    481:                if (pp->pr_nidle == 0)
                    482:                        panic("pr_rmpage: nidle inconsistent");
1.20      thorpej   483:                if (pp->pr_nitems < pp->pr_itemsperpage)
                    484:                        panic("pr_rmpage: nitems inconsistent");
1.6       thorpej   485: #endif
                    486:                pp->pr_nidle--;
                    487:        }
1.7       thorpej   488:
1.20      thorpej   489:        pp->pr_nitems -= pp->pr_itemsperpage;
                    490:
1.7       thorpej   491:        /*
1.101     thorpej   492:         * Unlink the page from the pool and queue it for release.
1.7       thorpej   493:         */
1.88      chs       494:        LIST_REMOVE(ph, ph_pagelist);
1.91      yamt      495:        if ((pp->pr_roflags & PR_PHINPAGE) == 0)
                    496:                SPLAY_REMOVE(phtree, &pp->pr_phtree, ph);
1.101     thorpej   497:        LIST_INSERT_HEAD(pq, ph, ph_pagelist);
                    498:
1.7       thorpej   499:        pp->pr_npages--;
                    500:        pp->pr_npagefree++;
1.6       thorpej   501:
1.88      chs       502:        pool_update_curpage(pp);
1.3       pk        503: }
                    504:
1.126     thorpej   505: static bool
1.117     yamt      506: pa_starved_p(struct pool_allocator *pa)
                    507: {
                    508:
                    509:        if (pa->pa_backingmap != NULL) {
                    510:                return vm_map_starved_p(pa->pa_backingmap);
                    511:        }
1.127     thorpej   512:        return false;
1.117     yamt      513: }
                    514:
                    515: static int
1.124     yamt      516: pool_reclaim_callback(struct callback_entry *ce, void *obj, void *arg)
1.117     yamt      517: {
                    518:        struct pool *pp = obj;
                    519:        struct pool_allocator *pa = pp->pr_alloc;
                    520:
                    521:        KASSERT(&pp->pr_reclaimerentry == ce);
                    522:        pool_reclaim(pp);
                    523:        if (!pa_starved_p(pa)) {
                    524:                return CALLBACK_CHAIN_ABORT;
                    525:        }
                    526:        return CALLBACK_CHAIN_CONTINUE;
                    527: }
                    528:
                    529: static void
                    530: pool_reclaim_register(struct pool *pp)
                    531: {
                    532:        struct vm_map *map = pp->pr_alloc->pa_backingmap;
                    533:        int s;
                    534:
                    535:        if (map == NULL) {
                    536:                return;
                    537:        }
                    538:
                    539:        s = splvm(); /* not necessary for INTRSAFE maps, but don't care. */
                    540:        callback_register(&vm_map_to_kernel(map)->vmk_reclaim_callback,
                    541:            &pp->pr_reclaimerentry, pp, pool_reclaim_callback);
                    542:        splx(s);
                    543: }
                    544:
                    545: static void
                    546: pool_reclaim_unregister(struct pool *pp)
                    547: {
                    548:        struct vm_map *map = pp->pr_alloc->pa_backingmap;
                    549:        int s;
                    550:
                    551:        if (map == NULL) {
                    552:                return;
                    553:        }
                    554:
                    555:        s = splvm(); /* not necessary for INTRSAFE maps, but don't care. */
                    556:        callback_unregister(&vm_map_to_kernel(map)->vmk_reclaim_callback,
                    557:            &pp->pr_reclaimerentry);
                    558:        splx(s);
                    559: }
                    560:
                    561: static void
                    562: pa_reclaim_register(struct pool_allocator *pa)
                    563: {
                    564:        struct vm_map *map = *pa->pa_backingmapptr;
                    565:        struct pool *pp;
                    566:
                    567:        KASSERT(pa->pa_backingmap == NULL);
                    568:        if (map == NULL) {
                    569:                SLIST_INSERT_HEAD(&pa_deferinitq, pa, pa_q);
                    570:                return;
                    571:        }
                    572:        pa->pa_backingmap = map;
                    573:        TAILQ_FOREACH(pp, &pa->pa_list, pr_alloc_list) {
                    574:                pool_reclaim_register(pp);
                    575:        }
                    576: }
                    577:
1.3       pk        578: /*
1.94      simonb    579:  * Initialize all the pools listed in the "pools" link set.
                    580:  */
                    581: void
1.117     yamt      582: pool_subsystem_init(void)
1.94      simonb    583: {
1.117     yamt      584:        struct pool_allocator *pa;
1.94      simonb    585:        __link_set_decl(pools, struct link_pool_init);
                    586:        struct link_pool_init * const *pi;
                    587:
1.134     ad        588:        mutex_init(&pool_head_lock, MUTEX_DEFAULT, IPL_NONE);
                    589:        cv_init(&pool_busy, "poolbusy");
                    590:
1.94      simonb    591:        __link_set_foreach(pi, pools)
                    592:                pool_init((*pi)->pp, (*pi)->size, (*pi)->align,
                    593:                    (*pi)->align_offset, (*pi)->flags, (*pi)->wchan,
1.129     ad        594:                    (*pi)->palloc, (*pi)->ipl);
1.117     yamt      595:
                    596:        while ((pa = SLIST_FIRST(&pa_deferinitq)) != NULL) {
                    597:                KASSERT(pa->pa_backingmapptr != NULL);
                    598:                KASSERT(*pa->pa_backingmapptr != NULL);
                    599:                SLIST_REMOVE_HEAD(&pa_deferinitq, pa_q);
                    600:                pa_reclaim_register(pa);
                    601:        }
1.134     ad        602:
                    603:        pool_init(&cache_pool, sizeof(struct pool_cache), CACHE_LINE_SIZE,
                    604:            0, 0, "pcache", &pool_allocator_nointr, IPL_NONE);
                    605:
                    606:        pool_init(&cache_cpu_pool, sizeof(pool_cache_cpu_t), CACHE_LINE_SIZE,
                    607:            0, 0, "pcachecpu", &pool_allocator_nointr, IPL_NONE);
1.94      simonb    608: }
                    609:
                    610: /*
1.3       pk        611:  * Initialize the given pool resource structure.
                    612:  *
                    613:  * We export this routine to allow other kernel parts to declare
                    614:  * static pools that must be initialized before malloc() is available.
                    615:  */
                    616: void
1.42      thorpej   617: pool_init(struct pool *pp, size_t size, u_int align, u_int ioff, int flags,
1.129     ad        618:     const char *wchan, struct pool_allocator *palloc, int ipl)
1.3       pk        619: {
1.116     simonb    620: #ifdef DEBUG
                    621:        struct pool *pp1;
                    622: #endif
1.92      enami     623:        size_t trysize, phsize;
1.134     ad        624:        int off, slack;
1.3       pk        625:
1.116     simonb    626: #ifdef DEBUG
                    627:        /*
                    628:         * Check that the pool hasn't already been initialised and
                    629:         * added to the list of all pools.
                    630:         */
                    631:        LIST_FOREACH(pp1, &pool_head, pr_poollist) {
                    632:                if (pp == pp1)
                    633:                        panic("pool_init: pool %s already initialised",
                    634:                            wchan);
                    635:        }
                    636: #endif
                    637:
1.25      thorpej   638: #ifdef POOL_DIAGNOSTIC
                    639:        /*
                    640:         * Always log if POOL_DIAGNOSTIC is defined.
                    641:         */
                    642:        if (pool_logsize != 0)
                    643:                flags |= PR_LOGGING;
                    644: #endif
                    645:
1.66      thorpej   646:        if (palloc == NULL)
                    647:                palloc = &pool_allocator_kmem;
1.112     bjh21     648: #ifdef POOL_SUBPAGE
                    649:        if (size > palloc->pa_pagesz) {
                    650:                if (palloc == &pool_allocator_kmem)
                    651:                        palloc = &pool_allocator_kmem_fullpage;
                    652:                else if (palloc == &pool_allocator_nointr)
                    653:                        palloc = &pool_allocator_nointr_fullpage;
                    654:        }
1.66      thorpej   655: #endif /* POOL_SUBPAGE */
                    656:        if ((palloc->pa_flags & PA_INITIALIZED) == 0) {
1.112     bjh21     657:                if (palloc->pa_pagesz == 0)
1.66      thorpej   658:                        palloc->pa_pagesz = PAGE_SIZE;
                    659:
                    660:                TAILQ_INIT(&palloc->pa_list);
                    661:
1.134     ad        662:                mutex_init(&palloc->pa_lock, MUTEX_DEFAULT, IPL_VM);
1.66      thorpej   663:                palloc->pa_pagemask = ~(palloc->pa_pagesz - 1);
                    664:                palloc->pa_pageshift = ffs(palloc->pa_pagesz) - 1;
1.117     yamt      665:
                    666:                if (palloc->pa_backingmapptr != NULL) {
                    667:                        pa_reclaim_register(palloc);
                    668:                }
1.66      thorpej   669:                palloc->pa_flags |= PA_INITIALIZED;
1.4       thorpej   670:        }
1.3       pk        671:
                    672:        if (align == 0)
                    673:                align = ALIGN(1);
1.14      thorpej   674:
1.120     yamt      675:        if ((flags & PR_NOTOUCH) == 0 && size < sizeof(struct pool_item))
1.14      thorpej   676:                size = sizeof(struct pool_item);
1.3       pk        677:
1.78      thorpej   678:        size = roundup(size, align);
1.66      thorpej   679: #ifdef DIAGNOSTIC
                    680:        if (size > palloc->pa_pagesz)
1.121     yamt      681:                panic("pool_init: pool item size (%zu) too large", size);
1.66      thorpej   682: #endif
1.35      pk        683:
1.3       pk        684:        /*
                    685:         * Initialize the pool structure.
                    686:         */
1.88      chs       687:        LIST_INIT(&pp->pr_emptypages);
                    688:        LIST_INIT(&pp->pr_fullpages);
                    689:        LIST_INIT(&pp->pr_partpages);
1.134     ad        690:        pp->pr_cache = NULL;
1.3       pk        691:        pp->pr_curpage = NULL;
                    692:        pp->pr_npages = 0;
                    693:        pp->pr_minitems = 0;
                    694:        pp->pr_minpages = 0;
                    695:        pp->pr_maxpages = UINT_MAX;
1.20      thorpej   696:        pp->pr_roflags = flags;
                    697:        pp->pr_flags = 0;
1.35      pk        698:        pp->pr_size = size;
1.3       pk        699:        pp->pr_align = align;
                    700:        pp->pr_wchan = wchan;
1.66      thorpej   701:        pp->pr_alloc = palloc;
1.20      thorpej   702:        pp->pr_nitems = 0;
                    703:        pp->pr_nout = 0;
                    704:        pp->pr_hardlimit = UINT_MAX;
                    705:        pp->pr_hardlimit_warning = NULL;
1.31      thorpej   706:        pp->pr_hardlimit_ratecap.tv_sec = 0;
                    707:        pp->pr_hardlimit_ratecap.tv_usec = 0;
                    708:        pp->pr_hardlimit_warning_last.tv_sec = 0;
                    709:        pp->pr_hardlimit_warning_last.tv_usec = 0;
1.68      thorpej   710:        pp->pr_drain_hook = NULL;
                    711:        pp->pr_drain_hook_arg = NULL;
1.125     ad        712:        pp->pr_freecheck = NULL;
1.3       pk        713:
                    714:        /*
                    715:         * Decide whether to put the page header off page to avoid
1.92      enami     716:         * wasting too large a part of the page or too big item.
                    717:         * Off-page page headers go on a hash table, so we can match
                    718:         * a returned item with its header based on the page address.
                    719:         * We use 1/16 of the page size and about 8 times of the item
                    720:         * size as the threshold (XXX: tune)
                    721:         *
                    722:         * However, we'll put the header into the page if we can put
                    723:         * it without wasting any items.
                    724:         *
                    725:         * Silently enforce `0 <= ioff < align'.
1.3       pk        726:         */
1.92      enami     727:        pp->pr_itemoffset = ioff %= align;
                    728:        /* See the comment below about reserved bytes. */
                    729:        trysize = palloc->pa_pagesz - ((align - ioff) % align);
                    730:        phsize = ALIGN(sizeof(struct pool_item_header));
1.121     yamt      731:        if ((pp->pr_roflags & (PR_NOTOUCH | PR_NOALIGN)) == 0 &&
1.97      yamt      732:            (pp->pr_size < MIN(palloc->pa_pagesz / 16, phsize << 3) ||
                    733:            trysize / pp->pr_size == (trysize - phsize) / pp->pr_size)) {
1.3       pk        734:                /* Use the end of the page for the page header */
1.20      thorpej   735:                pp->pr_roflags |= PR_PHINPAGE;
1.92      enami     736:                pp->pr_phoffset = off = palloc->pa_pagesz - phsize;
1.2       pk        737:        } else {
1.3       pk        738:                /* The page header will be taken from our page header pool */
                    739:                pp->pr_phoffset = 0;
1.66      thorpej   740:                off = palloc->pa_pagesz;
1.88      chs       741:                SPLAY_INIT(&pp->pr_phtree);
1.2       pk        742:        }
1.1       pk        743:
1.3       pk        744:        /*
                    745:         * Alignment is to take place at `ioff' within the item. This means
                    746:         * we must reserve up to `align - 1' bytes on the page to allow
                    747:         * appropriate positioning of each item.
                    748:         */
                    749:        pp->pr_itemsperpage = (off - ((align - ioff) % align)) / pp->pr_size;
1.43      thorpej   750:        KASSERT(pp->pr_itemsperpage != 0);
1.97      yamt      751:        if ((pp->pr_roflags & PR_NOTOUCH)) {
                    752:                int idx;
                    753:
                    754:                for (idx = 0; pp->pr_itemsperpage > PHPOOL_FREELIST_NELEM(idx);
                    755:                    idx++) {
                    756:                        /* nothing */
                    757:                }
                    758:                if (idx >= PHPOOL_MAX) {
                    759:                        /*
                    760:                         * if you see this panic, consider to tweak
                    761:                         * PHPOOL_MAX and PHPOOL_FREELIST_NELEM.
                    762:                         */
                    763:                        panic("%s: too large itemsperpage(%d) for PR_NOTOUCH",
                    764:                            pp->pr_wchan, pp->pr_itemsperpage);
                    765:                }
                    766:                pp->pr_phpool = &phpool[idx];
                    767:        } else if ((pp->pr_roflags & PR_PHINPAGE) == 0) {
                    768:                pp->pr_phpool = &phpool[0];
                    769:        }
                    770: #if defined(DIAGNOSTIC)
                    771:        else {
                    772:                pp->pr_phpool = NULL;
                    773:        }
                    774: #endif
1.3       pk        775:
                    776:        /*
                    777:         * Use the slack between the chunks and the page header
                    778:         * for "cache coloring".
                    779:         */
                    780:        slack = off - pp->pr_itemsperpage * pp->pr_size;
                    781:        pp->pr_maxcolor = (slack / align) * align;
                    782:        pp->pr_curcolor = 0;
                    783:
                    784:        pp->pr_nget = 0;
                    785:        pp->pr_nfail = 0;
                    786:        pp->pr_nput = 0;
                    787:        pp->pr_npagealloc = 0;
                    788:        pp->pr_npagefree = 0;
1.1       pk        789:        pp->pr_hiwat = 0;
1.8       thorpej   790:        pp->pr_nidle = 0;
1.134     ad        791:        pp->pr_refcnt = 0;
1.3       pk        792:
1.59      thorpej   793: #ifdef POOL_DIAGNOSTIC
1.25      thorpej   794:        if (flags & PR_LOGGING) {
                    795:                if (kmem_map == NULL ||
                    796:                    (pp->pr_log = malloc(pool_logsize * sizeof(struct pool_log),
                    797:                     M_TEMP, M_NOWAIT)) == NULL)
1.20      thorpej   798:                        pp->pr_roflags &= ~PR_LOGGING;
1.3       pk        799:                pp->pr_curlogentry = 0;
                    800:                pp->pr_logsize = pool_logsize;
                    801:        }
1.59      thorpej   802: #endif
1.25      thorpej   803:
                    804:        pp->pr_entered_file = NULL;
                    805:        pp->pr_entered_line = 0;
1.3       pk        806:
1.138     ad        807:        /*
                    808:         * XXXAD hack to prevent IP input processing from blocking.
                    809:         */
                    810:        if (ipl == IPL_SOFTNET) {
                    811:                mutex_init(&pp->pr_lock, MUTEX_DEFAULT, IPL_VM);
                    812:        } else {
                    813:                mutex_init(&pp->pr_lock, MUTEX_DEFAULT, ipl);
                    814:        }
1.134     ad        815:        cv_init(&pp->pr_cv, wchan);
                    816:        pp->pr_ipl = ipl;
1.1       pk        817:
1.3       pk        818:        /*
1.43      thorpej   819:         * Initialize private page header pool and cache magazine pool if we
                    820:         * haven't done so yet.
1.23      thorpej   821:         * XXX LOCKING.
1.3       pk        822:         */
1.97      yamt      823:        if (phpool[0].pr_size == 0) {
                    824:                int idx;
                    825:                for (idx = 0; idx < PHPOOL_MAX; idx++) {
                    826:                        static char phpool_names[PHPOOL_MAX][6+1+6+1];
                    827:                        int nelem;
                    828:                        size_t sz;
                    829:
                    830:                        nelem = PHPOOL_FREELIST_NELEM(idx);
                    831:                        snprintf(phpool_names[idx], sizeof(phpool_names[idx]),
                    832:                            "phpool-%d", nelem);
                    833:                        sz = sizeof(struct pool_item_header);
                    834:                        if (nelem) {
1.135     yamt      835:                                sz = offsetof(struct pool_item_header,
                    836:                                    ph_bitmap[howmany(nelem, BITMAP_SIZE)]);
1.97      yamt      837:                        }
                    838:                        pool_init(&phpool[idx], sz, 0, 0, 0,
1.129     ad        839:                            phpool_names[idx], &pool_allocator_meta, IPL_VM);
1.97      yamt      840:                }
1.62      bjh21     841: #ifdef POOL_SUBPAGE
                    842:                pool_init(&psppool, POOL_SUBPAGE, POOL_SUBPAGE, 0,
1.129     ad        843:                    PR_RECURSIVE, "psppool", &pool_allocator_meta, IPL_VM);
1.62      bjh21     844: #endif
1.134     ad        845:                pool_init(&pcgpool, sizeof(pcg_t), CACHE_LINE_SIZE, 0, 0,
                    846:                    "cachegrp", &pool_allocator_meta, IPL_VM);
1.1       pk        847:        }
                    848:
1.134     ad        849:        if (__predict_true(!cold)) {
                    850:                /* Insert into the list of all pools. */
                    851:                mutex_enter(&pool_head_lock);
                    852:                LIST_INSERT_HEAD(&pool_head, pp, pr_poollist);
                    853:                mutex_exit(&pool_head_lock);
                    854:
                    855:                /* Insert this into the list of pools using this allocator. */
                    856:                mutex_enter(&palloc->pa_lock);
                    857:                TAILQ_INSERT_TAIL(&palloc->pa_list, pp, pr_alloc_list);
                    858:                mutex_exit(&palloc->pa_lock);
                    859:        } else {
                    860:                LIST_INSERT_HEAD(&pool_head, pp, pr_poollist);
                    861:                TAILQ_INSERT_TAIL(&palloc->pa_list, pp, pr_alloc_list);
                    862:        }
1.66      thorpej   863:
1.117     yamt      864:        pool_reclaim_register(pp);
1.1       pk        865: }
                    866:
                    867: /*
                    868:  * De-commision a pool resource.
                    869:  */
                    870: void
1.42      thorpej   871: pool_destroy(struct pool *pp)
1.1       pk        872: {
1.101     thorpej   873:        struct pool_pagelist pq;
1.3       pk        874:        struct pool_item_header *ph;
1.43      thorpej   875:
1.101     thorpej   876:        /* Remove from global pool list */
1.134     ad        877:        mutex_enter(&pool_head_lock);
                    878:        while (pp->pr_refcnt != 0)
                    879:                cv_wait(&pool_busy, &pool_head_lock);
1.102     chs       880:        LIST_REMOVE(pp, pr_poollist);
1.101     thorpej   881:        if (drainpp == pp)
                    882:                drainpp = NULL;
1.134     ad        883:        mutex_exit(&pool_head_lock);
1.101     thorpej   884:
                    885:        /* Remove this pool from its allocator's list of pools. */
1.117     yamt      886:        pool_reclaim_unregister(pp);
1.134     ad        887:        mutex_enter(&pp->pr_alloc->pa_lock);
1.66      thorpej   888:        TAILQ_REMOVE(&pp->pr_alloc->pa_list, pp, pr_alloc_list);
1.134     ad        889:        mutex_exit(&pp->pr_alloc->pa_lock);
1.66      thorpej   890:
1.134     ad        891:        mutex_enter(&pp->pr_lock);
1.101     thorpej   892:
1.134     ad        893:        KASSERT(pp->pr_cache == NULL);
1.3       pk        894:
                    895: #ifdef DIAGNOSTIC
1.20      thorpej   896:        if (pp->pr_nout != 0) {
1.25      thorpej   897:                pr_printlog(pp, NULL, printf);
1.80      provos    898:                panic("pool_destroy: pool busy: still out: %u",
1.20      thorpej   899:                    pp->pr_nout);
1.3       pk        900:        }
                    901: #endif
1.1       pk        902:
1.101     thorpej   903:        KASSERT(LIST_EMPTY(&pp->pr_fullpages));
                    904:        KASSERT(LIST_EMPTY(&pp->pr_partpages));
                    905:
1.3       pk        906:        /* Remove all pages */
1.101     thorpej   907:        LIST_INIT(&pq);
1.88      chs       908:        while ((ph = LIST_FIRST(&pp->pr_emptypages)) != NULL)
1.101     thorpej   909:                pr_rmpage(pp, ph, &pq);
                    910:
1.134     ad        911:        mutex_exit(&pp->pr_lock);
1.3       pk        912:
1.101     thorpej   913:        pr_pagelist_free(pp, &pq);
1.3       pk        914:
1.59      thorpej   915: #ifdef POOL_DIAGNOSTIC
1.20      thorpej   916:        if ((pp->pr_roflags & PR_LOGGING) != 0)
1.3       pk        917:                free(pp->pr_log, M_TEMP);
1.59      thorpej   918: #endif
1.134     ad        919:
                    920:        cv_destroy(&pp->pr_cv);
                    921:        mutex_destroy(&pp->pr_lock);
1.1       pk        922: }
                    923:
1.68      thorpej   924: void
                    925: pool_set_drain_hook(struct pool *pp, void (*fn)(void *, int), void *arg)
                    926: {
                    927:
                    928:        /* XXX no locking -- must be used just after pool_init() */
                    929: #ifdef DIAGNOSTIC
                    930:        if (pp->pr_drain_hook != NULL)
                    931:                panic("pool_set_drain_hook(%s): already set", pp->pr_wchan);
                    932: #endif
                    933:        pp->pr_drain_hook = fn;
                    934:        pp->pr_drain_hook_arg = arg;
                    935: }
                    936:
1.88      chs       937: static struct pool_item_header *
1.128     christos  938: pool_alloc_item_header(struct pool *pp, void *storage, int flags)
1.55      thorpej   939: {
                    940:        struct pool_item_header *ph;
                    941:
                    942:        if ((pp->pr_roflags & PR_PHINPAGE) != 0)
1.128     christos  943:                ph = (struct pool_item_header *) ((char *)storage + pp->pr_phoffset);
1.134     ad        944:        else
1.97      yamt      945:                ph = pool_get(pp->pr_phpool, flags);
1.55      thorpej   946:
                    947:        return (ph);
                    948: }
1.1       pk        949:
                    950: /*
1.134     ad        951:  * Grab an item from the pool.
1.1       pk        952:  */
1.3       pk        953: void *
1.59      thorpej   954: #ifdef POOL_DIAGNOSTIC
1.42      thorpej   955: _pool_get(struct pool *pp, int flags, const char *file, long line)
1.56      sommerfe  956: #else
                    957: pool_get(struct pool *pp, int flags)
                    958: #endif
1.1       pk        959: {
                    960:        struct pool_item *pi;
1.3       pk        961:        struct pool_item_header *ph;
1.55      thorpej   962:        void *v;
1.1       pk        963:
1.2       pk        964: #ifdef DIAGNOSTIC
1.95      atatat    965:        if (__predict_false(pp->pr_itemsperpage == 0))
                    966:                panic("pool_get: pool %p: pr_itemsperpage is zero, "
                    967:                    "pool not initialized?", pp);
1.84      thorpej   968:        if (__predict_false(curlwp == NULL && doing_shutdown == 0 &&
1.37      sommerfe  969:                            (flags & PR_WAITOK) != 0))
1.77      matt      970:                panic("pool_get: %s: must have NOWAIT", pp->pr_wchan);
1.58      thorpej   971:
1.102     chs       972: #endif /* DIAGNOSTIC */
1.58      thorpej   973: #ifdef LOCKDEBUG
                    974:        if (flags & PR_WAITOK)
1.119     yamt      975:                ASSERT_SLEEPABLE(NULL, "pool_get(PR_WAITOK)");
1.56      sommerfe  976: #endif
1.1       pk        977:
1.134     ad        978:        mutex_enter(&pp->pr_lock);
1.25      thorpej   979:        pr_enter(pp, file, line);
1.20      thorpej   980:
                    981:  startover:
                    982:        /*
                    983:         * Check to see if we've reached the hard limit.  If we have,
                    984:         * and we can wait, then wait until an item has been returned to
                    985:         * the pool.
                    986:         */
                    987: #ifdef DIAGNOSTIC
1.34      thorpej   988:        if (__predict_false(pp->pr_nout > pp->pr_hardlimit)) {
1.25      thorpej   989:                pr_leave(pp);
1.134     ad        990:                mutex_exit(&pp->pr_lock);
1.20      thorpej   991:                panic("pool_get: %s: crossed hard limit", pp->pr_wchan);
                    992:        }
                    993: #endif
1.34      thorpej   994:        if (__predict_false(pp->pr_nout == pp->pr_hardlimit)) {
1.68      thorpej   995:                if (pp->pr_drain_hook != NULL) {
                    996:                        /*
                    997:                         * Since the drain hook is going to free things
                    998:                         * back to the pool, unlock, call the hook, re-lock,
                    999:                         * and check the hardlimit condition again.
                   1000:                         */
                   1001:                        pr_leave(pp);
1.134     ad       1002:                        mutex_exit(&pp->pr_lock);
1.68      thorpej  1003:                        (*pp->pr_drain_hook)(pp->pr_drain_hook_arg, flags);
1.134     ad       1004:                        mutex_enter(&pp->pr_lock);
1.68      thorpej  1005:                        pr_enter(pp, file, line);
                   1006:                        if (pp->pr_nout < pp->pr_hardlimit)
                   1007:                                goto startover;
                   1008:                }
                   1009:
1.29      sommerfe 1010:                if ((flags & PR_WAITOK) && !(flags & PR_LIMITFAIL)) {
1.20      thorpej  1011:                        /*
                   1012:                         * XXX: A warning isn't logged in this case.  Should
                   1013:                         * it be?
                   1014:                         */
                   1015:                        pp->pr_flags |= PR_WANTED;
1.25      thorpej  1016:                        pr_leave(pp);
1.134     ad       1017:                        cv_wait(&pp->pr_cv, &pp->pr_lock);
1.25      thorpej  1018:                        pr_enter(pp, file, line);
1.20      thorpej  1019:                        goto startover;
                   1020:                }
1.31      thorpej  1021:
                   1022:                /*
                   1023:                 * Log a message that the hard limit has been hit.
                   1024:                 */
                   1025:                if (pp->pr_hardlimit_warning != NULL &&
                   1026:                    ratecheck(&pp->pr_hardlimit_warning_last,
                   1027:                              &pp->pr_hardlimit_ratecap))
                   1028:                        log(LOG_ERR, "%s\n", pp->pr_hardlimit_warning);
1.21      thorpej  1029:
                   1030:                pp->pr_nfail++;
                   1031:
1.25      thorpej  1032:                pr_leave(pp);
1.134     ad       1033:                mutex_exit(&pp->pr_lock);
1.20      thorpej  1034:                return (NULL);
                   1035:        }
                   1036:
1.3       pk       1037:        /*
                   1038:         * The convention we use is that if `curpage' is not NULL, then
                   1039:         * it points at a non-empty bucket. In particular, `curpage'
                   1040:         * never points at a page header which has PR_PHINPAGE set and
                   1041:         * has no items in its bucket.
                   1042:         */
1.20      thorpej  1043:        if ((ph = pp->pr_curpage) == NULL) {
1.113     yamt     1044:                int error;
                   1045:
1.20      thorpej  1046: #ifdef DIAGNOSTIC
                   1047:                if (pp->pr_nitems != 0) {
1.134     ad       1048:                        mutex_exit(&pp->pr_lock);
1.20      thorpej  1049:                        printf("pool_get: %s: curpage NULL, nitems %u\n",
                   1050:                            pp->pr_wchan, pp->pr_nitems);
1.80      provos   1051:                        panic("pool_get: nitems inconsistent");
1.20      thorpej  1052:                }
                   1053: #endif
                   1054:
1.21      thorpej  1055:                /*
                   1056:                 * Call the back-end page allocator for more memory.
                   1057:                 * Release the pool lock, as the back-end page allocator
                   1058:                 * may block.
                   1059:                 */
1.25      thorpej  1060:                pr_leave(pp);
1.113     yamt     1061:                error = pool_grow(pp, flags);
                   1062:                pr_enter(pp, file, line);
                   1063:                if (error != 0) {
1.21      thorpej  1064:                        /*
1.55      thorpej  1065:                         * We were unable to allocate a page or item
                   1066:                         * header, but we released the lock during
                   1067:                         * allocation, so perhaps items were freed
                   1068:                         * back to the pool.  Check for this case.
1.21      thorpej  1069:                         */
                   1070:                        if (pp->pr_curpage != NULL)
                   1071:                                goto startover;
1.15      pk       1072:
1.117     yamt     1073:                        pp->pr_nfail++;
1.25      thorpej  1074:                        pr_leave(pp);
1.134     ad       1075:                        mutex_exit(&pp->pr_lock);
1.117     yamt     1076:                        return (NULL);
1.1       pk       1077:                }
1.3       pk       1078:
1.20      thorpej  1079:                /* Start the allocation process over. */
                   1080:                goto startover;
1.3       pk       1081:        }
1.97      yamt     1082:        if (pp->pr_roflags & PR_NOTOUCH) {
                   1083: #ifdef DIAGNOSTIC
                   1084:                if (__predict_false(ph->ph_nmissing == pp->pr_itemsperpage)) {
                   1085:                        pr_leave(pp);
1.134     ad       1086:                        mutex_exit(&pp->pr_lock);
1.97      yamt     1087:                        panic("pool_get: %s: page empty", pp->pr_wchan);
                   1088:                }
                   1089: #endif
                   1090:                v = pr_item_notouch_get(pp, ph);
                   1091: #ifdef POOL_DIAGNOSTIC
                   1092:                pr_log(pp, v, PRLOG_GET, file, line);
                   1093: #endif
                   1094:        } else {
1.102     chs      1095:                v = pi = LIST_FIRST(&ph->ph_itemlist);
1.97      yamt     1096:                if (__predict_false(v == NULL)) {
                   1097:                        pr_leave(pp);
1.134     ad       1098:                        mutex_exit(&pp->pr_lock);
1.97      yamt     1099:                        panic("pool_get: %s: page empty", pp->pr_wchan);
                   1100:                }
1.20      thorpej  1101: #ifdef DIAGNOSTIC
1.97      yamt     1102:                if (__predict_false(pp->pr_nitems == 0)) {
                   1103:                        pr_leave(pp);
1.134     ad       1104:                        mutex_exit(&pp->pr_lock);
1.97      yamt     1105:                        printf("pool_get: %s: items on itemlist, nitems %u\n",
                   1106:                            pp->pr_wchan, pp->pr_nitems);
                   1107:                        panic("pool_get: nitems inconsistent");
                   1108:                }
1.65      enami    1109: #endif
1.56      sommerfe 1110:
1.65      enami    1111: #ifdef POOL_DIAGNOSTIC
1.97      yamt     1112:                pr_log(pp, v, PRLOG_GET, file, line);
1.65      enami    1113: #endif
1.3       pk       1114:
1.65      enami    1115: #ifdef DIAGNOSTIC
1.97      yamt     1116:                if (__predict_false(pi->pi_magic != PI_MAGIC)) {
                   1117:                        pr_printlog(pp, pi, printf);
                   1118:                        panic("pool_get(%s): free list modified: "
                   1119:                            "magic=%x; page %p; item addr %p\n",
                   1120:                            pp->pr_wchan, pi->pi_magic, ph->ph_page, pi);
                   1121:                }
1.3       pk       1122: #endif
                   1123:
1.97      yamt     1124:                /*
                   1125:                 * Remove from item list.
                   1126:                 */
1.102     chs      1127:                LIST_REMOVE(pi, pi_list);
1.97      yamt     1128:        }
1.20      thorpej  1129:        pp->pr_nitems--;
                   1130:        pp->pr_nout++;
1.6       thorpej  1131:        if (ph->ph_nmissing == 0) {
                   1132: #ifdef DIAGNOSTIC
1.34      thorpej  1133:                if (__predict_false(pp->pr_nidle == 0))
1.6       thorpej  1134:                        panic("pool_get: nidle inconsistent");
                   1135: #endif
                   1136:                pp->pr_nidle--;
1.88      chs      1137:
                   1138:                /*
                   1139:                 * This page was previously empty.  Move it to the list of
                   1140:                 * partially-full pages.  This page is already curpage.
                   1141:                 */
                   1142:                LIST_REMOVE(ph, ph_pagelist);
                   1143:                LIST_INSERT_HEAD(&pp->pr_partpages, ph, ph_pagelist);
1.6       thorpej  1144:        }
1.3       pk       1145:        ph->ph_nmissing++;
1.97      yamt     1146:        if (ph->ph_nmissing == pp->pr_itemsperpage) {
1.21      thorpej  1147: #ifdef DIAGNOSTIC
1.97      yamt     1148:                if (__predict_false((pp->pr_roflags & PR_NOTOUCH) == 0 &&
1.102     chs      1149:                    !LIST_EMPTY(&ph->ph_itemlist))) {
1.25      thorpej  1150:                        pr_leave(pp);
1.134     ad       1151:                        mutex_exit(&pp->pr_lock);
1.21      thorpej  1152:                        panic("pool_get: %s: nmissing inconsistent",
                   1153:                            pp->pr_wchan);
                   1154:                }
                   1155: #endif
1.3       pk       1156:                /*
1.88      chs      1157:                 * This page is now full.  Move it to the full list
                   1158:                 * and select a new current page.
1.3       pk       1159:                 */
1.88      chs      1160:                LIST_REMOVE(ph, ph_pagelist);
                   1161:                LIST_INSERT_HEAD(&pp->pr_fullpages, ph, ph_pagelist);
                   1162:                pool_update_curpage(pp);
1.1       pk       1163:        }
1.3       pk       1164:
                   1165:        pp->pr_nget++;
1.111     christos 1166:        pr_leave(pp);
1.20      thorpej  1167:
                   1168:        /*
                   1169:         * If we have a low water mark and we are now below that low
                   1170:         * water mark, add more items to the pool.
                   1171:         */
1.53      thorpej  1172:        if (POOL_NEEDS_CATCHUP(pp) && pool_catchup(pp) != 0) {
1.20      thorpej  1173:                /*
                   1174:                 * XXX: Should we log a warning?  Should we set up a timeout
                   1175:                 * to try again in a second or so?  The latter could break
                   1176:                 * a caller's assumptions about interrupt protection, etc.
                   1177:                 */
                   1178:        }
                   1179:
1.134     ad       1180:        mutex_exit(&pp->pr_lock);
1.125     ad       1181:        KASSERT((((vaddr_t)v + pp->pr_itemoffset) & (pp->pr_align - 1)) == 0);
                   1182:        FREECHECK_OUT(&pp->pr_freecheck, v);
1.1       pk       1183:        return (v);
                   1184: }
                   1185:
                   1186: /*
1.43      thorpej  1187:  * Internal version of pool_put().  Pool is already locked/entered.
1.1       pk       1188:  */
1.43      thorpej  1189: static void
1.101     thorpej  1190: pool_do_put(struct pool *pp, void *v, struct pool_pagelist *pq)
1.1       pk       1191: {
                   1192:        struct pool_item *pi = v;
1.3       pk       1193:        struct pool_item_header *ph;
                   1194:
1.134     ad       1195:        KASSERT(mutex_owned(&pp->pr_lock));
1.125     ad       1196:        FREECHECK_IN(&pp->pr_freecheck, v);
1.134     ad       1197:        LOCKDEBUG_MEM_CHECK(v, pp->pr_size);
1.61      chs      1198:
1.30      thorpej  1199: #ifdef DIAGNOSTIC
1.34      thorpej  1200:        if (__predict_false(pp->pr_nout == 0)) {
1.30      thorpej  1201:                printf("pool %s: putting with none out\n",
                   1202:                    pp->pr_wchan);
                   1203:                panic("pool_put");
                   1204:        }
                   1205: #endif
1.3       pk       1206:
1.121     yamt     1207:        if (__predict_false((ph = pr_find_pagehead(pp, v)) == NULL)) {
1.25      thorpej  1208:                pr_printlog(pp, NULL, printf);
1.3       pk       1209:                panic("pool_put: %s: page header missing", pp->pr_wchan);
                   1210:        }
1.28      thorpej  1211:
1.3       pk       1212:        /*
                   1213:         * Return to item list.
                   1214:         */
1.97      yamt     1215:        if (pp->pr_roflags & PR_NOTOUCH) {
                   1216:                pr_item_notouch_put(pp, ph, v);
                   1217:        } else {
1.2       pk       1218: #ifdef DIAGNOSTIC
1.97      yamt     1219:                pi->pi_magic = PI_MAGIC;
1.3       pk       1220: #endif
1.32      chs      1221: #ifdef DEBUG
1.97      yamt     1222:                {
                   1223:                        int i, *ip = v;
1.32      chs      1224:
1.97      yamt     1225:                        for (i = 0; i < pp->pr_size / sizeof(int); i++) {
                   1226:                                *ip++ = PI_MAGIC;
                   1227:                        }
1.32      chs      1228:                }
                   1229: #endif
                   1230:
1.102     chs      1231:                LIST_INSERT_HEAD(&ph->ph_itemlist, pi, pi_list);
1.97      yamt     1232:        }
1.79      thorpej  1233:        KDASSERT(ph->ph_nmissing != 0);
1.3       pk       1234:        ph->ph_nmissing--;
                   1235:        pp->pr_nput++;
1.20      thorpej  1236:        pp->pr_nitems++;
                   1237:        pp->pr_nout--;
1.3       pk       1238:
                   1239:        /* Cancel "pool empty" condition if it exists */
                   1240:        if (pp->pr_curpage == NULL)
                   1241:                pp->pr_curpage = ph;
                   1242:
                   1243:        if (pp->pr_flags & PR_WANTED) {
                   1244:                pp->pr_flags &= ~PR_WANTED;
1.15      pk       1245:                if (ph->ph_nmissing == 0)
                   1246:                        pp->pr_nidle++;
1.134     ad       1247:                cv_broadcast(&pp->pr_cv);
1.3       pk       1248:                return;
                   1249:        }
                   1250:
                   1251:        /*
1.88      chs      1252:         * If this page is now empty, do one of two things:
1.21      thorpej  1253:         *
1.88      chs      1254:         *      (1) If we have more pages than the page high water mark,
1.96      thorpej  1255:         *          free the page back to the system.  ONLY CONSIDER
1.90      thorpej  1256:         *          FREEING BACK A PAGE IF WE HAVE MORE THAN OUR MINIMUM PAGE
                   1257:         *          CLAIM.
1.21      thorpej  1258:         *
1.88      chs      1259:         *      (2) Otherwise, move the page to the empty page list.
                   1260:         *
                   1261:         * Either way, select a new current page (so we use a partially-full
                   1262:         * page if one is available).
1.3       pk       1263:         */
                   1264:        if (ph->ph_nmissing == 0) {
1.6       thorpej  1265:                pp->pr_nidle++;
1.90      thorpej  1266:                if (pp->pr_npages > pp->pr_minpages &&
                   1267:                    (pp->pr_npages > pp->pr_maxpages ||
1.117     yamt     1268:                     pa_starved_p(pp->pr_alloc))) {
1.101     thorpej  1269:                        pr_rmpage(pp, ph, pq);
1.3       pk       1270:                } else {
1.88      chs      1271:                        LIST_REMOVE(ph, ph_pagelist);
                   1272:                        LIST_INSERT_HEAD(&pp->pr_emptypages, ph, ph_pagelist);
1.3       pk       1273:
1.21      thorpej  1274:                        /*
                   1275:                         * Update the timestamp on the page.  A page must
                   1276:                         * be idle for some period of time before it can
                   1277:                         * be reclaimed by the pagedaemon.  This minimizes
                   1278:                         * ping-pong'ing for memory.
                   1279:                         */
1.118     kardel   1280:                        getmicrotime(&ph->ph_time);
1.1       pk       1281:                }
1.88      chs      1282:                pool_update_curpage(pp);
1.1       pk       1283:        }
1.88      chs      1284:
1.21      thorpej  1285:        /*
1.88      chs      1286:         * If the page was previously completely full, move it to the
                   1287:         * partially-full list and make it the current page.  The next
                   1288:         * allocation will get the item from this page, instead of
                   1289:         * further fragmenting the pool.
1.21      thorpej  1290:         */
                   1291:        else if (ph->ph_nmissing == (pp->pr_itemsperpage - 1)) {
1.88      chs      1292:                LIST_REMOVE(ph, ph_pagelist);
                   1293:                LIST_INSERT_HEAD(&pp->pr_partpages, ph, ph_pagelist);
1.21      thorpej  1294:                pp->pr_curpage = ph;
                   1295:        }
1.43      thorpej  1296: }
                   1297:
                   1298: /*
1.134     ad       1299:  * Return resource to the pool.
1.43      thorpej  1300:  */
1.59      thorpej  1301: #ifdef POOL_DIAGNOSTIC
1.43      thorpej  1302: void
                   1303: _pool_put(struct pool *pp, void *v, const char *file, long line)
                   1304: {
1.101     thorpej  1305:        struct pool_pagelist pq;
                   1306:
                   1307:        LIST_INIT(&pq);
1.43      thorpej  1308:
1.134     ad       1309:        mutex_enter(&pp->pr_lock);
1.43      thorpej  1310:        pr_enter(pp, file, line);
                   1311:
1.56      sommerfe 1312:        pr_log(pp, v, PRLOG_PUT, file, line);
                   1313:
1.101     thorpej  1314:        pool_do_put(pp, v, &pq);
1.21      thorpej  1315:
1.25      thorpej  1316:        pr_leave(pp);
1.134     ad       1317:        mutex_exit(&pp->pr_lock);
1.101     thorpej  1318:
1.102     chs      1319:        pr_pagelist_free(pp, &pq);
1.1       pk       1320: }
1.57      sommerfe 1321: #undef pool_put
1.59      thorpej  1322: #endif /* POOL_DIAGNOSTIC */
1.1       pk       1323:
1.56      sommerfe 1324: void
                   1325: pool_put(struct pool *pp, void *v)
                   1326: {
1.101     thorpej  1327:        struct pool_pagelist pq;
                   1328:
                   1329:        LIST_INIT(&pq);
1.56      sommerfe 1330:
1.134     ad       1331:        mutex_enter(&pp->pr_lock);
1.101     thorpej  1332:        pool_do_put(pp, v, &pq);
1.134     ad       1333:        mutex_exit(&pp->pr_lock);
1.56      sommerfe 1334:
1.102     chs      1335:        pr_pagelist_free(pp, &pq);
1.56      sommerfe 1336: }
1.57      sommerfe 1337:
1.59      thorpej  1338: #ifdef POOL_DIAGNOSTIC
1.57      sommerfe 1339: #define                pool_put(h, v)  _pool_put((h), (v), __FILE__, __LINE__)
1.56      sommerfe 1340: #endif
1.74      thorpej  1341:
                   1342: /*
1.113     yamt     1343:  * pool_grow: grow a pool by a page.
                   1344:  *
                   1345:  * => called with pool locked.
                   1346:  * => unlock and relock the pool.
                   1347:  * => return with pool locked.
                   1348:  */
                   1349:
                   1350: static int
                   1351: pool_grow(struct pool *pp, int flags)
                   1352: {
                   1353:        struct pool_item_header *ph = NULL;
                   1354:        char *cp;
                   1355:
1.134     ad       1356:        mutex_exit(&pp->pr_lock);
1.113     yamt     1357:        cp = pool_allocator_alloc(pp, flags);
                   1358:        if (__predict_true(cp != NULL)) {
                   1359:                ph = pool_alloc_item_header(pp, cp, flags);
                   1360:        }
                   1361:        if (__predict_false(cp == NULL || ph == NULL)) {
                   1362:                if (cp != NULL) {
                   1363:                        pool_allocator_free(pp, cp);
                   1364:                }
1.134     ad       1365:                mutex_enter(&pp->pr_lock);
1.113     yamt     1366:                return ENOMEM;
                   1367:        }
                   1368:
1.134     ad       1369:        mutex_enter(&pp->pr_lock);
1.113     yamt     1370:        pool_prime_page(pp, cp, ph);
                   1371:        pp->pr_npagealloc++;
                   1372:        return 0;
                   1373: }
                   1374:
                   1375: /*
1.74      thorpej  1376:  * Add N items to the pool.
                   1377:  */
                   1378: int
                   1379: pool_prime(struct pool *pp, int n)
                   1380: {
1.75      simonb   1381:        int newpages;
1.113     yamt     1382:        int error = 0;
1.74      thorpej  1383:
1.134     ad       1384:        mutex_enter(&pp->pr_lock);
1.74      thorpej  1385:
                   1386:        newpages = roundup(n, pp->pr_itemsperpage) / pp->pr_itemsperpage;
                   1387:
                   1388:        while (newpages-- > 0) {
1.113     yamt     1389:                error = pool_grow(pp, PR_NOWAIT);
                   1390:                if (error) {
1.74      thorpej  1391:                        break;
                   1392:                }
                   1393:                pp->pr_minpages++;
                   1394:        }
                   1395:
                   1396:        if (pp->pr_minpages >= pp->pr_maxpages)
                   1397:                pp->pr_maxpages = pp->pr_minpages + 1;  /* XXX */
                   1398:
1.134     ad       1399:        mutex_exit(&pp->pr_lock);
1.113     yamt     1400:        return error;
1.74      thorpej  1401: }
1.55      thorpej  1402:
                   1403: /*
1.3       pk       1404:  * Add a page worth of items to the pool.
1.21      thorpej  1405:  *
                   1406:  * Note, we must be called with the pool descriptor LOCKED.
1.3       pk       1407:  */
1.55      thorpej  1408: static void
1.128     christos 1409: pool_prime_page(struct pool *pp, void *storage, struct pool_item_header *ph)
1.3       pk       1410: {
                   1411:        struct pool_item *pi;
1.128     christos 1412:        void *cp = storage;
1.125     ad       1413:        const unsigned int align = pp->pr_align;
                   1414:        const unsigned int ioff = pp->pr_itemoffset;
1.55      thorpej  1415:        int n;
1.36      pk       1416:
1.134     ad       1417:        KASSERT(mutex_owned(&pp->pr_lock));
1.91      yamt     1418:
1.66      thorpej  1419: #ifdef DIAGNOSTIC
1.121     yamt     1420:        if ((pp->pr_roflags & PR_NOALIGN) == 0 &&
                   1421:            ((uintptr_t)cp & (pp->pr_alloc->pa_pagesz - 1)) != 0)
1.36      pk       1422:                panic("pool_prime_page: %s: unaligned page", pp->pr_wchan);
1.66      thorpej  1423: #endif
1.3       pk       1424:
                   1425:        /*
                   1426:         * Insert page header.
                   1427:         */
1.88      chs      1428:        LIST_INSERT_HEAD(&pp->pr_emptypages, ph, ph_pagelist);
1.102     chs      1429:        LIST_INIT(&ph->ph_itemlist);
1.3       pk       1430:        ph->ph_page = storage;
                   1431:        ph->ph_nmissing = 0;
1.118     kardel   1432:        getmicrotime(&ph->ph_time);
1.88      chs      1433:        if ((pp->pr_roflags & PR_PHINPAGE) == 0)
                   1434:                SPLAY_INSERT(phtree, &pp->pr_phtree, ph);
1.3       pk       1435:
1.6       thorpej  1436:        pp->pr_nidle++;
                   1437:
1.3       pk       1438:        /*
                   1439:         * Color this page.
                   1440:         */
1.128     christos 1441:        cp = (char *)cp + pp->pr_curcolor;
1.3       pk       1442:        if ((pp->pr_curcolor += align) > pp->pr_maxcolor)
                   1443:                pp->pr_curcolor = 0;
                   1444:
                   1445:        /*
                   1446:         * Adjust storage to apply aligment to `pr_itemoffset' in each item.
                   1447:         */
                   1448:        if (ioff != 0)
1.128     christos 1449:                cp = (char *)cp + align - ioff;
1.3       pk       1450:
1.125     ad       1451:        KASSERT((((vaddr_t)cp + ioff) & (align - 1)) == 0);
                   1452:
1.3       pk       1453:        /*
                   1454:         * Insert remaining chunks on the bucket list.
                   1455:         */
                   1456:        n = pp->pr_itemsperpage;
1.20      thorpej  1457:        pp->pr_nitems += n;
1.3       pk       1458:
1.97      yamt     1459:        if (pp->pr_roflags & PR_NOTOUCH) {
1.140   ! yamt     1460:                pr_item_notouch_init(pp, ph, (char *)cp - (char *)storage);
1.97      yamt     1461:        } else {
                   1462:                while (n--) {
                   1463:                        pi = (struct pool_item *)cp;
1.78      thorpej  1464:
1.97      yamt     1465:                        KASSERT(((((vaddr_t)pi) + ioff) & (align - 1)) == 0);
1.3       pk       1466:
1.97      yamt     1467:                        /* Insert on page list */
1.102     chs      1468:                        LIST_INSERT_HEAD(&ph->ph_itemlist, pi, pi_list);
1.3       pk       1469: #ifdef DIAGNOSTIC
1.97      yamt     1470:                        pi->pi_magic = PI_MAGIC;
1.3       pk       1471: #endif
1.128     christos 1472:                        cp = (char *)cp + pp->pr_size;
1.125     ad       1473:
                   1474:                        KASSERT((((vaddr_t)cp + ioff) & (align - 1)) == 0);
1.97      yamt     1475:                }
1.3       pk       1476:        }
                   1477:
                   1478:        /*
                   1479:         * If the pool was depleted, point at the new page.
                   1480:         */
                   1481:        if (pp->pr_curpage == NULL)
                   1482:                pp->pr_curpage = ph;
                   1483:
                   1484:        if (++pp->pr_npages > pp->pr_hiwat)
                   1485:                pp->pr_hiwat = pp->pr_npages;
                   1486: }
                   1487:
1.20      thorpej  1488: /*
1.52      thorpej  1489:  * Used by pool_get() when nitems drops below the low water mark.  This
1.88      chs      1490:  * is used to catch up pr_nitems with the low water mark.
1.20      thorpej  1491:  *
1.21      thorpej  1492:  * Note 1, we never wait for memory here, we let the caller decide what to do.
1.20      thorpej  1493:  *
1.73      thorpej  1494:  * Note 2, we must be called with the pool already locked, and we return
1.20      thorpej  1495:  * with it locked.
                   1496:  */
                   1497: static int
1.42      thorpej  1498: pool_catchup(struct pool *pp)
1.20      thorpej  1499: {
                   1500:        int error = 0;
                   1501:
1.54      thorpej  1502:        while (POOL_NEEDS_CATCHUP(pp)) {
1.113     yamt     1503:                error = pool_grow(pp, PR_NOWAIT);
                   1504:                if (error) {
1.20      thorpej  1505:                        break;
                   1506:                }
                   1507:        }
1.113     yamt     1508:        return error;
1.20      thorpej  1509: }
                   1510:
1.88      chs      1511: static void
                   1512: pool_update_curpage(struct pool *pp)
                   1513: {
                   1514:
                   1515:        pp->pr_curpage = LIST_FIRST(&pp->pr_partpages);
                   1516:        if (pp->pr_curpage == NULL) {
                   1517:                pp->pr_curpage = LIST_FIRST(&pp->pr_emptypages);
                   1518:        }
                   1519: }
                   1520:
1.3       pk       1521: void
1.42      thorpej  1522: pool_setlowat(struct pool *pp, int n)
1.3       pk       1523: {
1.15      pk       1524:
1.134     ad       1525:        mutex_enter(&pp->pr_lock);
1.21      thorpej  1526:
1.3       pk       1527:        pp->pr_minitems = n;
1.15      pk       1528:        pp->pr_minpages = (n == 0)
                   1529:                ? 0
1.18      thorpej  1530:                : roundup(n, pp->pr_itemsperpage) / pp->pr_itemsperpage;
1.20      thorpej  1531:
                   1532:        /* Make sure we're caught up with the newly-set low water mark. */
1.75      simonb   1533:        if (POOL_NEEDS_CATCHUP(pp) && pool_catchup(pp) != 0) {
1.20      thorpej  1534:                /*
                   1535:                 * XXX: Should we log a warning?  Should we set up a timeout
                   1536:                 * to try again in a second or so?  The latter could break
                   1537:                 * a caller's assumptions about interrupt protection, etc.
                   1538:                 */
                   1539:        }
1.21      thorpej  1540:
1.134     ad       1541:        mutex_exit(&pp->pr_lock);
1.3       pk       1542: }
                   1543:
                   1544: void
1.42      thorpej  1545: pool_sethiwat(struct pool *pp, int n)
1.3       pk       1546: {
1.15      pk       1547:
1.134     ad       1548:        mutex_enter(&pp->pr_lock);
1.21      thorpej  1549:
1.15      pk       1550:        pp->pr_maxpages = (n == 0)
                   1551:                ? 0
1.18      thorpej  1552:                : roundup(n, pp->pr_itemsperpage) / pp->pr_itemsperpage;
1.21      thorpej  1553:
1.134     ad       1554:        mutex_exit(&pp->pr_lock);
1.3       pk       1555: }
                   1556:
1.20      thorpej  1557: void
1.42      thorpej  1558: pool_sethardlimit(struct pool *pp, int n, const char *warnmess, int ratecap)
1.20      thorpej  1559: {
                   1560:
1.134     ad       1561:        mutex_enter(&pp->pr_lock);
1.20      thorpej  1562:
                   1563:        pp->pr_hardlimit = n;
                   1564:        pp->pr_hardlimit_warning = warnmess;
1.31      thorpej  1565:        pp->pr_hardlimit_ratecap.tv_sec = ratecap;
                   1566:        pp->pr_hardlimit_warning_last.tv_sec = 0;
                   1567:        pp->pr_hardlimit_warning_last.tv_usec = 0;
1.20      thorpej  1568:
                   1569:        /*
1.21      thorpej  1570:         * In-line version of pool_sethiwat(), because we don't want to
                   1571:         * release the lock.
1.20      thorpej  1572:         */
                   1573:        pp->pr_maxpages = (n == 0)
                   1574:                ? 0
                   1575:                : roundup(n, pp->pr_itemsperpage) / pp->pr_itemsperpage;
1.21      thorpej  1576:
1.134     ad       1577:        mutex_exit(&pp->pr_lock);
1.20      thorpej  1578: }
1.3       pk       1579:
                   1580: /*
                   1581:  * Release all complete pages that have not been used recently.
                   1582:  */
1.66      thorpej  1583: int
1.59      thorpej  1584: #ifdef POOL_DIAGNOSTIC
1.42      thorpej  1585: _pool_reclaim(struct pool *pp, const char *file, long line)
1.56      sommerfe 1586: #else
                   1587: pool_reclaim(struct pool *pp)
                   1588: #endif
1.3       pk       1589: {
                   1590:        struct pool_item_header *ph, *phnext;
1.61      chs      1591:        struct pool_pagelist pq;
1.102     chs      1592:        struct timeval curtime, diff;
1.134     ad       1593:        bool klock;
                   1594:        int rv;
1.3       pk       1595:
1.68      thorpej  1596:        if (pp->pr_drain_hook != NULL) {
                   1597:                /*
                   1598:                 * The drain hook must be called with the pool unlocked.
                   1599:                 */
                   1600:                (*pp->pr_drain_hook)(pp->pr_drain_hook_arg, PR_NOWAIT);
                   1601:        }
                   1602:
1.134     ad       1603:        /*
                   1604:         * XXXSMP Because mutexes at IPL_SOFTXXX are still spinlocks,
                   1605:         * and we are called from the pagedaemon without kernel_lock.
                   1606:         * Does not apply to IPL_SOFTBIO.
                   1607:         */
                   1608:        if (pp->pr_ipl == IPL_SOFTNET || pp->pr_ipl == IPL_SOFTCLOCK ||
                   1609:            pp->pr_ipl == IPL_SOFTSERIAL) {
                   1610:                KERNEL_LOCK(1, NULL);
                   1611:                klock = true;
                   1612:        } else
                   1613:                klock = false;
                   1614:
                   1615:        /* Reclaim items from the pool's cache (if any). */
                   1616:        if (pp->pr_cache != NULL)
                   1617:                pool_cache_invalidate(pp->pr_cache);
                   1618:
                   1619:        if (mutex_tryenter(&pp->pr_lock) == 0) {
                   1620:                if (klock) {
                   1621:                        KERNEL_UNLOCK_ONE(NULL);
                   1622:                }
1.66      thorpej  1623:                return (0);
1.134     ad       1624:        }
1.25      thorpej  1625:        pr_enter(pp, file, line);
1.68      thorpej  1626:
1.88      chs      1627:        LIST_INIT(&pq);
1.43      thorpej  1628:
1.118     kardel   1629:        getmicrotime(&curtime);
1.21      thorpej  1630:
1.88      chs      1631:        for (ph = LIST_FIRST(&pp->pr_emptypages); ph != NULL; ph = phnext) {
                   1632:                phnext = LIST_NEXT(ph, ph_pagelist);
1.3       pk       1633:
                   1634:                /* Check our minimum page claim */
                   1635:                if (pp->pr_npages <= pp->pr_minpages)
                   1636:                        break;
                   1637:
1.88      chs      1638:                KASSERT(ph->ph_nmissing == 0);
                   1639:                timersub(&curtime, &ph->ph_time, &diff);
1.117     yamt     1640:                if (diff.tv_sec < pool_inactive_time
                   1641:                    && !pa_starved_p(pp->pr_alloc))
1.88      chs      1642:                        continue;
1.21      thorpej  1643:
1.88      chs      1644:                /*
                   1645:                 * If freeing this page would put us below
                   1646:                 * the low water mark, stop now.
                   1647:                 */
                   1648:                if ((pp->pr_nitems - pp->pr_itemsperpage) <
                   1649:                    pp->pr_minitems)
                   1650:                        break;
1.21      thorpej  1651:
1.88      chs      1652:                pr_rmpage(pp, ph, &pq);
1.3       pk       1653:        }
                   1654:
1.25      thorpej  1655:        pr_leave(pp);
1.134     ad       1656:        mutex_exit(&pp->pr_lock);
                   1657:
                   1658:        if (LIST_EMPTY(&pq))
                   1659:                rv = 0;
                   1660:        else {
                   1661:                pr_pagelist_free(pp, &pq);
                   1662:                rv = 1;
                   1663:        }
                   1664:
                   1665:        if (klock) {
                   1666:                KERNEL_UNLOCK_ONE(NULL);
                   1667:        }
1.66      thorpej  1668:
1.134     ad       1669:        return (rv);
1.3       pk       1670: }
                   1671:
                   1672: /*
1.134     ad       1673:  * Drain pools, one at a time.  This is a two stage process;
                   1674:  * drain_start kicks off a cross call to drain CPU-level caches
                   1675:  * if the pool has an associated pool_cache.  drain_end waits
                   1676:  * for those cross calls to finish, and then drains the cache
                   1677:  * (if any) and pool.
1.131     ad       1678:  *
1.134     ad       1679:  * Note, must never be called from interrupt context.
1.3       pk       1680:  */
                   1681: void
1.134     ad       1682: pool_drain_start(struct pool **ppp, uint64_t *wp)
1.3       pk       1683: {
                   1684:        struct pool *pp;
1.134     ad       1685:
                   1686:        KASSERT(!LIST_EMPTY(&pool_head));
1.3       pk       1687:
1.61      chs      1688:        pp = NULL;
1.134     ad       1689:
                   1690:        /* Find next pool to drain, and add a reference. */
                   1691:        mutex_enter(&pool_head_lock);
                   1692:        do {
                   1693:                if (drainpp == NULL) {
                   1694:                        drainpp = LIST_FIRST(&pool_head);
                   1695:                }
                   1696:                if (drainpp != NULL) {
                   1697:                        pp = drainpp;
                   1698:                        drainpp = LIST_NEXT(pp, pr_poollist);
                   1699:                }
                   1700:                /*
                   1701:                 * Skip completely idle pools.  We depend on at least
                   1702:                 * one pool in the system being active.
                   1703:                 */
                   1704:        } while (pp == NULL || pp->pr_npages == 0);
                   1705:        pp->pr_refcnt++;
                   1706:        mutex_exit(&pool_head_lock);
                   1707:
                   1708:        /* If there is a pool_cache, drain CPU level caches. */
                   1709:        *ppp = pp;
                   1710:        if (pp->pr_cache != NULL) {
                   1711:                *wp = xc_broadcast(0, (xcfunc_t)pool_cache_xcall,
                   1712:                    pp->pr_cache, NULL);
                   1713:        }
                   1714: }
                   1715:
                   1716: void
                   1717: pool_drain_end(struct pool *pp, uint64_t where)
                   1718: {
                   1719:
                   1720:        if (pp == NULL)
                   1721:                return;
                   1722:
                   1723:        KASSERT(pp->pr_refcnt > 0);
                   1724:
                   1725:        /* Wait for remote draining to complete. */
                   1726:        if (pp->pr_cache != NULL)
                   1727:                xc_wait(where);
                   1728:
                   1729:        /* Drain the cache (if any) and pool.. */
                   1730:        pool_reclaim(pp);
                   1731:
                   1732:        /* Finally, unlock the pool. */
                   1733:        mutex_enter(&pool_head_lock);
                   1734:        pp->pr_refcnt--;
                   1735:        cv_broadcast(&pool_busy);
                   1736:        mutex_exit(&pool_head_lock);
1.3       pk       1737: }
                   1738:
                   1739: /*
                   1740:  * Diagnostic helpers.
                   1741:  */
                   1742: void
1.42      thorpej  1743: pool_print(struct pool *pp, const char *modif)
1.21      thorpej  1744: {
                   1745:
1.25      thorpej  1746:        pool_print1(pp, modif, printf);
1.21      thorpej  1747: }
                   1748:
1.25      thorpej  1749: void
1.108     yamt     1750: pool_printall(const char *modif, void (*pr)(const char *, ...))
                   1751: {
                   1752:        struct pool *pp;
                   1753:
                   1754:        LIST_FOREACH(pp, &pool_head, pr_poollist) {
                   1755:                pool_printit(pp, modif, pr);
                   1756:        }
                   1757: }
                   1758:
                   1759: void
1.42      thorpej  1760: pool_printit(struct pool *pp, const char *modif, void (*pr)(const char *, ...))
1.25      thorpej  1761: {
                   1762:
                   1763:        if (pp == NULL) {
                   1764:                (*pr)("Must specify a pool to print.\n");
                   1765:                return;
                   1766:        }
                   1767:
                   1768:        pool_print1(pp, modif, pr);
                   1769: }
                   1770:
1.21      thorpej  1771: static void
1.124     yamt     1772: pool_print_pagelist(struct pool *pp, struct pool_pagelist *pl,
1.97      yamt     1773:     void (*pr)(const char *, ...))
1.88      chs      1774: {
                   1775:        struct pool_item_header *ph;
                   1776: #ifdef DIAGNOSTIC
                   1777:        struct pool_item *pi;
                   1778: #endif
                   1779:
                   1780:        LIST_FOREACH(ph, pl, ph_pagelist) {
                   1781:                (*pr)("\t\tpage %p, nmissing %d, time %lu,%lu\n",
                   1782:                    ph->ph_page, ph->ph_nmissing,
                   1783:                    (u_long)ph->ph_time.tv_sec,
                   1784:                    (u_long)ph->ph_time.tv_usec);
                   1785: #ifdef DIAGNOSTIC
1.97      yamt     1786:                if (!(pp->pr_roflags & PR_NOTOUCH)) {
1.102     chs      1787:                        LIST_FOREACH(pi, &ph->ph_itemlist, pi_list) {
1.97      yamt     1788:                                if (pi->pi_magic != PI_MAGIC) {
                   1789:                                        (*pr)("\t\t\titem %p, magic 0x%x\n",
                   1790:                                            pi, pi->pi_magic);
                   1791:                                }
1.88      chs      1792:                        }
                   1793:                }
                   1794: #endif
                   1795:        }
                   1796: }
                   1797:
                   1798: static void
1.42      thorpej  1799: pool_print1(struct pool *pp, const char *modif, void (*pr)(const char *, ...))
1.3       pk       1800: {
1.25      thorpej  1801:        struct pool_item_header *ph;
1.134     ad       1802:        pool_cache_t pc;
                   1803:        pcg_t *pcg;
                   1804:        pool_cache_cpu_t *cc;
                   1805:        uint64_t cpuhit, cpumiss;
1.44      thorpej  1806:        int i, print_log = 0, print_pagelist = 0, print_cache = 0;
1.25      thorpej  1807:        char c;
                   1808:
                   1809:        while ((c = *modif++) != '\0') {
                   1810:                if (c == 'l')
                   1811:                        print_log = 1;
                   1812:                if (c == 'p')
                   1813:                        print_pagelist = 1;
1.44      thorpej  1814:                if (c == 'c')
                   1815:                        print_cache = 1;
1.25      thorpej  1816:        }
                   1817:
1.134     ad       1818:        if ((pc = pp->pr_cache) != NULL) {
                   1819:                (*pr)("POOL CACHE");
                   1820:        } else {
                   1821:                (*pr)("POOL");
                   1822:        }
                   1823:
                   1824:        (*pr)(" %s: size %u, align %u, ioff %u, roflags 0x%08x\n",
1.25      thorpej  1825:            pp->pr_wchan, pp->pr_size, pp->pr_align, pp->pr_itemoffset,
                   1826:            pp->pr_roflags);
1.66      thorpej  1827:        (*pr)("\talloc %p\n", pp->pr_alloc);
1.25      thorpej  1828:        (*pr)("\tminitems %u, minpages %u, maxpages %u, npages %u\n",
                   1829:            pp->pr_minitems, pp->pr_minpages, pp->pr_maxpages, pp->pr_npages);
                   1830:        (*pr)("\titemsperpage %u, nitems %u, nout %u, hardlimit %u\n",
                   1831:            pp->pr_itemsperpage, pp->pr_nitems, pp->pr_nout, pp->pr_hardlimit);
                   1832:
1.134     ad       1833:        (*pr)("\tnget %lu, nfail %lu, nput %lu\n",
1.25      thorpej  1834:            pp->pr_nget, pp->pr_nfail, pp->pr_nput);
                   1835:        (*pr)("\tnpagealloc %lu, npagefree %lu, hiwat %u, nidle %lu\n",
                   1836:            pp->pr_npagealloc, pp->pr_npagefree, pp->pr_hiwat, pp->pr_nidle);
                   1837:
                   1838:        if (print_pagelist == 0)
                   1839:                goto skip_pagelist;
                   1840:
1.88      chs      1841:        if ((ph = LIST_FIRST(&pp->pr_emptypages)) != NULL)
                   1842:                (*pr)("\n\tempty page list:\n");
1.97      yamt     1843:        pool_print_pagelist(pp, &pp->pr_emptypages, pr);
1.88      chs      1844:        if ((ph = LIST_FIRST(&pp->pr_fullpages)) != NULL)
                   1845:                (*pr)("\n\tfull page list:\n");
1.97      yamt     1846:        pool_print_pagelist(pp, &pp->pr_fullpages, pr);
1.88      chs      1847:        if ((ph = LIST_FIRST(&pp->pr_partpages)) != NULL)
                   1848:                (*pr)("\n\tpartial-page list:\n");
1.97      yamt     1849:        pool_print_pagelist(pp, &pp->pr_partpages, pr);
1.88      chs      1850:
1.25      thorpej  1851:        if (pp->pr_curpage == NULL)
                   1852:                (*pr)("\tno current page\n");
                   1853:        else
                   1854:                (*pr)("\tcurpage %p\n", pp->pr_curpage->ph_page);
                   1855:
                   1856:  skip_pagelist:
                   1857:        if (print_log == 0)
                   1858:                goto skip_log;
                   1859:
                   1860:        (*pr)("\n");
                   1861:        if ((pp->pr_roflags & PR_LOGGING) == 0)
                   1862:                (*pr)("\tno log\n");
1.122     christos 1863:        else {
1.25      thorpej  1864:                pr_printlog(pp, NULL, pr);
1.122     christos 1865:        }
1.3       pk       1866:
1.25      thorpej  1867:  skip_log:
1.44      thorpej  1868:
1.102     chs      1869: #define PR_GROUPLIST(pcg)                                              \
                   1870:        (*pr)("\t\tgroup %p: avail %d\n", pcg, pcg->pcg_avail);         \
                   1871:        for (i = 0; i < PCG_NOBJECTS; i++) {                            \
                   1872:                if (pcg->pcg_objects[i].pcgo_pa !=                      \
                   1873:                    POOL_PADDR_INVALID) {                               \
                   1874:                        (*pr)("\t\t\t%p, 0x%llx\n",                     \
                   1875:                            pcg->pcg_objects[i].pcgo_va,                \
                   1876:                            (unsigned long long)                        \
                   1877:                            pcg->pcg_objects[i].pcgo_pa);               \
                   1878:                } else {                                                \
                   1879:                        (*pr)("\t\t\t%p\n",                             \
                   1880:                            pcg->pcg_objects[i].pcgo_va);               \
                   1881:                }                                                       \
                   1882:        }
                   1883:
1.134     ad       1884:        if (pc != NULL) {
                   1885:                cpuhit = 0;
                   1886:                cpumiss = 0;
                   1887:                for (i = 0; i < MAXCPUS; i++) {
                   1888:                        if ((cc = pc->pc_cpus[i]) == NULL)
                   1889:                                continue;
                   1890:                        cpuhit += cc->cc_hits;
                   1891:                        cpumiss += cc->cc_misses;
                   1892:                }
                   1893:                (*pr)("\tcpu layer hits %llu misses %llu\n", cpuhit, cpumiss);
                   1894:                (*pr)("\tcache layer hits %llu misses %llu\n",
                   1895:                    pc->pc_hits, pc->pc_misses);
                   1896:                (*pr)("\tcache layer entry uncontended %llu contended %llu\n",
                   1897:                    pc->pc_hits + pc->pc_misses - pc->pc_contended,
                   1898:                    pc->pc_contended);
                   1899:                (*pr)("\tcache layer empty groups %u full groups %u\n",
                   1900:                    pc->pc_nempty, pc->pc_nfull);
                   1901:                if (print_cache) {
                   1902:                        (*pr)("\tfull cache groups:\n");
                   1903:                        for (pcg = pc->pc_fullgroups; pcg != NULL;
                   1904:                            pcg = pcg->pcg_next) {
                   1905:                                PR_GROUPLIST(pcg);
                   1906:                        }
                   1907:                        (*pr)("\tempty cache groups:\n");
                   1908:                        for (pcg = pc->pc_emptygroups; pcg != NULL;
                   1909:                            pcg = pcg->pcg_next) {
                   1910:                                PR_GROUPLIST(pcg);
                   1911:                        }
1.103     chs      1912:                }
1.44      thorpej  1913:        }
1.102     chs      1914: #undef PR_GROUPLIST
1.44      thorpej  1915:
1.88      chs      1916:        pr_enter_check(pp, pr);
                   1917: }
                   1918:
                   1919: static int
                   1920: pool_chk_page(struct pool *pp, const char *label, struct pool_item_header *ph)
                   1921: {
                   1922:        struct pool_item *pi;
1.128     christos 1923:        void *page;
1.88      chs      1924:        int n;
                   1925:
1.121     yamt     1926:        if ((pp->pr_roflags & PR_NOALIGN) == 0) {
1.128     christos 1927:                page = (void *)((uintptr_t)ph & pp->pr_alloc->pa_pagemask);
1.121     yamt     1928:                if (page != ph->ph_page &&
                   1929:                    (pp->pr_roflags & PR_PHINPAGE) != 0) {
                   1930:                        if (label != NULL)
                   1931:                                printf("%s: ", label);
                   1932:                        printf("pool(%p:%s): page inconsistency: page %p;"
                   1933:                               " at page head addr %p (p %p)\n", pp,
                   1934:                                pp->pr_wchan, ph->ph_page,
                   1935:                                ph, page);
                   1936:                        return 1;
                   1937:                }
1.88      chs      1938:        }
1.3       pk       1939:
1.97      yamt     1940:        if ((pp->pr_roflags & PR_NOTOUCH) != 0)
                   1941:                return 0;
                   1942:
1.102     chs      1943:        for (pi = LIST_FIRST(&ph->ph_itemlist), n = 0;
1.88      chs      1944:             pi != NULL;
1.102     chs      1945:             pi = LIST_NEXT(pi,pi_list), n++) {
1.88      chs      1946:
                   1947: #ifdef DIAGNOSTIC
                   1948:                if (pi->pi_magic != PI_MAGIC) {
                   1949:                        if (label != NULL)
                   1950:                                printf("%s: ", label);
                   1951:                        printf("pool(%s): free list modified: magic=%x;"
1.121     yamt     1952:                               " page %p; item ordinal %d; addr %p\n",
1.88      chs      1953:                                pp->pr_wchan, pi->pi_magic, ph->ph_page,
1.121     yamt     1954:                                n, pi);
1.88      chs      1955:                        panic("pool");
                   1956:                }
                   1957: #endif
1.121     yamt     1958:                if ((pp->pr_roflags & PR_NOALIGN) != 0) {
                   1959:                        continue;
                   1960:                }
1.128     christos 1961:                page = (void *)((uintptr_t)pi & pp->pr_alloc->pa_pagemask);
1.88      chs      1962:                if (page == ph->ph_page)
                   1963:                        continue;
                   1964:
                   1965:                if (label != NULL)
                   1966:                        printf("%s: ", label);
                   1967:                printf("pool(%p:%s): page inconsistency: page %p;"
                   1968:                       " item ordinal %d; addr %p (p %p)\n", pp,
                   1969:                        pp->pr_wchan, ph->ph_page,
                   1970:                        n, pi, page);
                   1971:                return 1;
                   1972:        }
                   1973:        return 0;
1.3       pk       1974: }
                   1975:
1.88      chs      1976:
1.3       pk       1977: int
1.42      thorpej  1978: pool_chk(struct pool *pp, const char *label)
1.3       pk       1979: {
                   1980:        struct pool_item_header *ph;
                   1981:        int r = 0;
                   1982:
1.134     ad       1983:        mutex_enter(&pp->pr_lock);
1.88      chs      1984:        LIST_FOREACH(ph, &pp->pr_emptypages, ph_pagelist) {
                   1985:                r = pool_chk_page(pp, label, ph);
                   1986:                if (r) {
                   1987:                        goto out;
                   1988:                }
                   1989:        }
                   1990:        LIST_FOREACH(ph, &pp->pr_fullpages, ph_pagelist) {
                   1991:                r = pool_chk_page(pp, label, ph);
                   1992:                if (r) {
1.3       pk       1993:                        goto out;
                   1994:                }
1.88      chs      1995:        }
                   1996:        LIST_FOREACH(ph, &pp->pr_partpages, ph_pagelist) {
                   1997:                r = pool_chk_page(pp, label, ph);
                   1998:                if (r) {
1.3       pk       1999:                        goto out;
                   2000:                }
                   2001:        }
1.88      chs      2002:
1.3       pk       2003: out:
1.134     ad       2004:        mutex_exit(&pp->pr_lock);
1.3       pk       2005:        return (r);
1.43      thorpej  2006: }
                   2007:
                   2008: /*
                   2009:  * pool_cache_init:
                   2010:  *
                   2011:  *     Initialize a pool cache.
1.134     ad       2012:  */
                   2013: pool_cache_t
                   2014: pool_cache_init(size_t size, u_int align, u_int align_offset, u_int flags,
                   2015:     const char *wchan, struct pool_allocator *palloc, int ipl,
                   2016:     int (*ctor)(void *, void *, int), void (*dtor)(void *, void *), void *arg)
                   2017: {
                   2018:        pool_cache_t pc;
                   2019:
                   2020:        pc = pool_get(&cache_pool, PR_WAITOK);
                   2021:        if (pc == NULL)
                   2022:                return NULL;
                   2023:
                   2024:        pool_cache_bootstrap(pc, size, align, align_offset, flags, wchan,
                   2025:           palloc, ipl, ctor, dtor, arg);
                   2026:
                   2027:        return pc;
                   2028: }
                   2029:
                   2030: /*
                   2031:  * pool_cache_bootstrap:
1.43      thorpej  2032:  *
1.134     ad       2033:  *     Kernel-private version of pool_cache_init().  The caller
                   2034:  *     provides initial storage.
1.43      thorpej  2035:  */
                   2036: void
1.134     ad       2037: pool_cache_bootstrap(pool_cache_t pc, size_t size, u_int align,
                   2038:     u_int align_offset, u_int flags, const char *wchan,
                   2039:     struct pool_allocator *palloc, int ipl,
                   2040:     int (*ctor)(void *, void *, int), void (*dtor)(void *, void *),
1.43      thorpej  2041:     void *arg)
                   2042: {
1.134     ad       2043:        CPU_INFO_ITERATOR cii;
                   2044:        struct cpu_info *ci;
                   2045:        struct pool *pp;
                   2046:
                   2047:        pp = &pc->pc_pool;
                   2048:        if (palloc == NULL && ipl == IPL_NONE)
                   2049:                palloc = &pool_allocator_nointr;
                   2050:        pool_init(pp, size, align, align_offset, flags, wchan, palloc, ipl);
1.43      thorpej  2051:
1.138     ad       2052:        /*
                   2053:         * XXXAD hack to prevent IP input processing from blocking.
                   2054:         */
                   2055:        if (ipl == IPL_SOFTNET) {
                   2056:                mutex_init(&pc->pc_lock, MUTEX_DEFAULT, IPL_VM);
                   2057:        } else {
                   2058:                mutex_init(&pc->pc_lock, MUTEX_DEFAULT, ipl);
                   2059:        }
1.43      thorpej  2060:
1.134     ad       2061:        if (ctor == NULL) {
                   2062:                ctor = (int (*)(void *, void *, int))nullop;
                   2063:        }
                   2064:        if (dtor == NULL) {
                   2065:                dtor = (void (*)(void *, void *))nullop;
                   2066:        }
1.43      thorpej  2067:
1.134     ad       2068:        pc->pc_emptygroups = NULL;
                   2069:        pc->pc_fullgroups = NULL;
                   2070:        pc->pc_partgroups = NULL;
1.43      thorpej  2071:        pc->pc_ctor = ctor;
                   2072:        pc->pc_dtor = dtor;
                   2073:        pc->pc_arg  = arg;
1.134     ad       2074:        pc->pc_hits  = 0;
1.48      thorpej  2075:        pc->pc_misses = 0;
1.134     ad       2076:        pc->pc_nempty = 0;
                   2077:        pc->pc_npart = 0;
                   2078:        pc->pc_nfull = 0;
                   2079:        pc->pc_contended = 0;
                   2080:        pc->pc_refcnt = 0;
1.136     yamt     2081:        pc->pc_freecheck = NULL;
1.134     ad       2082:
                   2083:        /* Allocate per-CPU caches. */
                   2084:        memset(pc->pc_cpus, 0, sizeof(pc->pc_cpus));
                   2085:        pc->pc_ncpu = 0;
1.139     ad       2086:        if (ncpu < 2) {
1.137     ad       2087:                /* XXX For sparc: boot CPU is not attached yet. */
                   2088:                pool_cache_cpu_init1(curcpu(), pc);
                   2089:        } else {
                   2090:                for (CPU_INFO_FOREACH(cii, ci)) {
                   2091:                        pool_cache_cpu_init1(ci, pc);
                   2092:                }
1.134     ad       2093:        }
                   2094:
                   2095:        if (__predict_true(!cold)) {
                   2096:                mutex_enter(&pp->pr_lock);
                   2097:                pp->pr_cache = pc;
                   2098:                mutex_exit(&pp->pr_lock);
                   2099:                mutex_enter(&pool_head_lock);
                   2100:                LIST_INSERT_HEAD(&pool_cache_head, pc, pc_cachelist);
                   2101:                mutex_exit(&pool_head_lock);
                   2102:        } else {
                   2103:                pp->pr_cache = pc;
                   2104:                LIST_INSERT_HEAD(&pool_cache_head, pc, pc_cachelist);
                   2105:        }
1.43      thorpej  2106: }
                   2107:
                   2108: /*
                   2109:  * pool_cache_destroy:
                   2110:  *
                   2111:  *     Destroy a pool cache.
                   2112:  */
                   2113: void
1.134     ad       2114: pool_cache_destroy(pool_cache_t pc)
1.43      thorpej  2115: {
1.134     ad       2116:        struct pool *pp = &pc->pc_pool;
                   2117:        pool_cache_cpu_t *cc;
                   2118:        pcg_t *pcg;
                   2119:        int i;
                   2120:
                   2121:        /* Remove it from the global list. */
                   2122:        mutex_enter(&pool_head_lock);
                   2123:        while (pc->pc_refcnt != 0)
                   2124:                cv_wait(&pool_busy, &pool_head_lock);
                   2125:        LIST_REMOVE(pc, pc_cachelist);
                   2126:        mutex_exit(&pool_head_lock);
1.43      thorpej  2127:
                   2128:        /* First, invalidate the entire cache. */
                   2129:        pool_cache_invalidate(pc);
                   2130:
1.134     ad       2131:        /* Disassociate it from the pool. */
                   2132:        mutex_enter(&pp->pr_lock);
                   2133:        pp->pr_cache = NULL;
                   2134:        mutex_exit(&pp->pr_lock);
                   2135:
                   2136:        /* Destroy per-CPU data */
                   2137:        for (i = 0; i < MAXCPUS; i++) {
                   2138:                if ((cc = pc->pc_cpus[i]) == NULL)
                   2139:                        continue;
                   2140:                if ((pcg = cc->cc_current) != NULL) {
                   2141:                        pcg->pcg_next = NULL;
                   2142:                        pool_cache_invalidate_groups(pc, pcg);
                   2143:                }
                   2144:                if ((pcg = cc->cc_previous) != NULL) {
                   2145:                        pcg->pcg_next = NULL;
                   2146:                        pool_cache_invalidate_groups(pc, pcg);
                   2147:                }
                   2148:                if (cc != &pc->pc_cpu0)
                   2149:                        pool_put(&cache_cpu_pool, cc);
                   2150:        }
                   2151:
                   2152:        /* Finally, destroy it. */
                   2153:        mutex_destroy(&pc->pc_lock);
                   2154:        pool_destroy(pp);
                   2155:        pool_put(&cache_pool, pc);
                   2156: }
                   2157:
                   2158: /*
                   2159:  * pool_cache_cpu_init1:
                   2160:  *
                   2161:  *     Called for each pool_cache whenever a new CPU is attached.
                   2162:  */
                   2163: static void
                   2164: pool_cache_cpu_init1(struct cpu_info *ci, pool_cache_t pc)
                   2165: {
                   2166:        pool_cache_cpu_t *cc;
1.137     ad       2167:        int index;
1.134     ad       2168:
1.137     ad       2169:        index = ci->ci_index;
                   2170:
                   2171:        KASSERT(index < MAXCPUS);
1.134     ad       2172:        KASSERT(((uintptr_t)pc->pc_cpus & (CACHE_LINE_SIZE - 1)) == 0);
                   2173:
1.137     ad       2174:        if ((cc = pc->pc_cpus[index]) != NULL) {
                   2175:                KASSERT(cc->cc_cpuindex == index);
1.134     ad       2176:                return;
                   2177:        }
                   2178:
                   2179:        /*
                   2180:         * The first CPU is 'free'.  This needs to be the case for
                   2181:         * bootstrap - we may not be able to allocate yet.
                   2182:         */
                   2183:        if (pc->pc_ncpu == 0) {
                   2184:                cc = &pc->pc_cpu0;
                   2185:                pc->pc_ncpu = 1;
                   2186:        } else {
                   2187:                mutex_enter(&pc->pc_lock);
                   2188:                pc->pc_ncpu++;
                   2189:                mutex_exit(&pc->pc_lock);
                   2190:                cc = pool_get(&cache_cpu_pool, PR_WAITOK);
                   2191:        }
                   2192:
                   2193:        cc->cc_ipl = pc->pc_pool.pr_ipl;
                   2194:        cc->cc_iplcookie = makeiplcookie(cc->cc_ipl);
                   2195:        cc->cc_cache = pc;
1.137     ad       2196:        cc->cc_cpuindex = index;
1.134     ad       2197:        cc->cc_hits = 0;
                   2198:        cc->cc_misses = 0;
                   2199:        cc->cc_current = NULL;
                   2200:        cc->cc_previous = NULL;
                   2201:
1.137     ad       2202:        pc->pc_cpus[index] = cc;
1.43      thorpej  2203: }
                   2204:
1.134     ad       2205: /*
                   2206:  * pool_cache_cpu_init:
                   2207:  *
                   2208:  *     Called whenever a new CPU is attached.
                   2209:  */
                   2210: void
                   2211: pool_cache_cpu_init(struct cpu_info *ci)
1.43      thorpej  2212: {
1.134     ad       2213:        pool_cache_t pc;
                   2214:
                   2215:        mutex_enter(&pool_head_lock);
                   2216:        LIST_FOREACH(pc, &pool_cache_head, pc_cachelist) {
                   2217:                pc->pc_refcnt++;
                   2218:                mutex_exit(&pool_head_lock);
1.43      thorpej  2219:
1.134     ad       2220:                pool_cache_cpu_init1(ci, pc);
1.43      thorpej  2221:
1.134     ad       2222:                mutex_enter(&pool_head_lock);
                   2223:                pc->pc_refcnt--;
                   2224:                cv_broadcast(&pool_busy);
                   2225:        }
                   2226:        mutex_exit(&pool_head_lock);
1.43      thorpej  2227: }
                   2228:
1.134     ad       2229: /*
                   2230:  * pool_cache_reclaim:
                   2231:  *
                   2232:  *     Reclaim memory from a pool cache.
                   2233:  */
                   2234: bool
                   2235: pool_cache_reclaim(pool_cache_t pc)
1.43      thorpej  2236: {
                   2237:
1.134     ad       2238:        return pool_reclaim(&pc->pc_pool);
                   2239: }
1.43      thorpej  2240:
1.136     yamt     2241: static void
                   2242: pool_cache_destruct_object1(pool_cache_t pc, void *object)
                   2243: {
                   2244:
                   2245:        (*pc->pc_dtor)(pc->pc_arg, object);
                   2246:        pool_put(&pc->pc_pool, object);
                   2247: }
                   2248:
1.134     ad       2249: /*
                   2250:  * pool_cache_destruct_object:
                   2251:  *
                   2252:  *     Force destruction of an object and its release back into
                   2253:  *     the pool.
                   2254:  */
                   2255: void
                   2256: pool_cache_destruct_object(pool_cache_t pc, void *object)
                   2257: {
                   2258:
1.136     yamt     2259:        FREECHECK_IN(&pc->pc_freecheck, object);
                   2260:
                   2261:        pool_cache_destruct_object1(pc, object);
1.43      thorpej  2262: }
                   2263:
1.134     ad       2264: /*
                   2265:  * pool_cache_invalidate_groups:
                   2266:  *
                   2267:  *     Invalidate a chain of groups and destruct all objects.
                   2268:  */
1.102     chs      2269: static void
1.134     ad       2270: pool_cache_invalidate_groups(pool_cache_t pc, pcg_t *pcg)
1.102     chs      2271: {
1.134     ad       2272:        void *object;
                   2273:        pcg_t *next;
                   2274:        int i;
                   2275:
                   2276:        for (; pcg != NULL; pcg = next) {
                   2277:                next = pcg->pcg_next;
                   2278:
                   2279:                for (i = 0; i < pcg->pcg_avail; i++) {
                   2280:                        object = pcg->pcg_objects[i].pcgo_va;
1.136     yamt     2281:                        pool_cache_destruct_object1(pc, object);
1.134     ad       2282:                }
1.102     chs      2283:
                   2284:                pool_put(&pcgpool, pcg);
                   2285:        }
                   2286: }
                   2287:
1.43      thorpej  2288: /*
1.134     ad       2289:  * pool_cache_invalidate:
1.43      thorpej  2290:  *
1.134     ad       2291:  *     Invalidate a pool cache (destruct and release all of the
                   2292:  *     cached objects).  Does not reclaim objects from the pool.
1.43      thorpej  2293:  */
1.134     ad       2294: void
                   2295: pool_cache_invalidate(pool_cache_t pc)
                   2296: {
                   2297:        pcg_t *full, *empty, *part;
                   2298:
                   2299:        mutex_enter(&pc->pc_lock);
                   2300:        full = pc->pc_fullgroups;
                   2301:        empty = pc->pc_emptygroups;
                   2302:        part = pc->pc_partgroups;
                   2303:        pc->pc_fullgroups = NULL;
                   2304:        pc->pc_emptygroups = NULL;
                   2305:        pc->pc_partgroups = NULL;
                   2306:        pc->pc_nfull = 0;
                   2307:        pc->pc_nempty = 0;
                   2308:        pc->pc_npart = 0;
                   2309:        mutex_exit(&pc->pc_lock);
                   2310:
                   2311:        pool_cache_invalidate_groups(pc, full);
                   2312:        pool_cache_invalidate_groups(pc, empty);
                   2313:        pool_cache_invalidate_groups(pc, part);
                   2314: }
                   2315:
                   2316: void
                   2317: pool_cache_set_drain_hook(pool_cache_t pc, void (*fn)(void *, int), void *arg)
                   2318: {
                   2319:
                   2320:        pool_set_drain_hook(&pc->pc_pool, fn, arg);
                   2321: }
                   2322:
                   2323: void
                   2324: pool_cache_setlowat(pool_cache_t pc, int n)
                   2325: {
                   2326:
                   2327:        pool_setlowat(&pc->pc_pool, n);
                   2328: }
                   2329:
                   2330: void
                   2331: pool_cache_sethiwat(pool_cache_t pc, int n)
                   2332: {
                   2333:
                   2334:        pool_sethiwat(&pc->pc_pool, n);
                   2335: }
                   2336:
                   2337: void
                   2338: pool_cache_sethardlimit(pool_cache_t pc, int n, const char *warnmess, int ratecap)
                   2339: {
                   2340:
                   2341:        pool_sethardlimit(&pc->pc_pool, n, warnmess, ratecap);
                   2342: }
                   2343:
                   2344: static inline pool_cache_cpu_t *
                   2345: pool_cache_cpu_enter(pool_cache_t pc, int *s)
                   2346: {
                   2347:        pool_cache_cpu_t *cc;
                   2348:
                   2349:        /*
                   2350:         * Prevent other users of the cache from accessing our
                   2351:         * CPU-local data.  To avoid touching shared state, we
                   2352:         * pull the neccessary information from CPU local data.
                   2353:         */
1.137     ad       2354:        crit_enter();
                   2355:        cc = pc->pc_cpus[curcpu()->ci_index];
1.134     ad       2356:        KASSERT(cc->cc_cache == pc);
1.137     ad       2357:        if (cc->cc_ipl != IPL_NONE) {
1.134     ad       2358:                *s = splraiseipl(cc->cc_iplcookie);
                   2359:        }
                   2360:        KASSERT(((uintptr_t)cc & (CACHE_LINE_SIZE - 1)) == 0);
                   2361:
                   2362:        return cc;
                   2363: }
                   2364:
                   2365: static inline void
                   2366: pool_cache_cpu_exit(pool_cache_cpu_t *cc, int *s)
                   2367: {
                   2368:
                   2369:        /* No longer need exclusive access to the per-CPU data. */
1.137     ad       2370:        if (cc->cc_ipl != IPL_NONE) {
1.134     ad       2371:                splx(*s);
                   2372:        }
1.137     ad       2373:        crit_exit();
1.134     ad       2374: }
                   2375:
                   2376: #if __GNUC_PREREQ__(3, 0)
                   2377: __attribute ((noinline))
                   2378: #endif
                   2379: pool_cache_cpu_t *
                   2380: pool_cache_get_slow(pool_cache_cpu_t *cc, int *s, void **objectp,
                   2381:                    paddr_t *pap, int flags)
1.43      thorpej  2382: {
1.134     ad       2383:        pcg_t *pcg, *cur;
                   2384:        uint64_t ncsw;
                   2385:        pool_cache_t pc;
1.43      thorpej  2386:        void *object;
1.58      thorpej  2387:
1.134     ad       2388:        pc = cc->cc_cache;
                   2389:        cc->cc_misses++;
1.43      thorpej  2390:
1.134     ad       2391:        /*
                   2392:         * Nothing was available locally.  Try and grab a group
                   2393:         * from the cache.
                   2394:         */
                   2395:        if (!mutex_tryenter(&pc->pc_lock)) {
                   2396:                ncsw = curlwp->l_ncsw;
                   2397:                mutex_enter(&pc->pc_lock);
                   2398:                pc->pc_contended++;
1.43      thorpej  2399:
1.134     ad       2400:                /*
                   2401:                 * If we context switched while locking, then
                   2402:                 * our view of the per-CPU data is invalid:
                   2403:                 * retry.
                   2404:                 */
                   2405:                if (curlwp->l_ncsw != ncsw) {
                   2406:                        mutex_exit(&pc->pc_lock);
                   2407:                        pool_cache_cpu_exit(cc, s);
                   2408:                        return pool_cache_cpu_enter(pc, s);
1.43      thorpej  2409:                }
1.102     chs      2410:        }
1.43      thorpej  2411:
1.134     ad       2412:        if ((pcg = pc->pc_fullgroups) != NULL) {
1.43      thorpej  2413:                /*
1.134     ad       2414:                 * If there's a full group, release our empty
                   2415:                 * group back to the cache.  Install the full
                   2416:                 * group as cc_current and return.
1.43      thorpej  2417:                 */
1.134     ad       2418:                if ((cur = cc->cc_current) != NULL) {
                   2419:                        KASSERT(cur->pcg_avail == 0);
                   2420:                        cur->pcg_next = pc->pc_emptygroups;
                   2421:                        pc->pc_emptygroups = cur;
                   2422:                        pc->pc_nempty++;
1.87      thorpej  2423:                }
1.134     ad       2424:                KASSERT(pcg->pcg_avail == PCG_NOBJECTS);
                   2425:                cc->cc_current = pcg;
                   2426:                pc->pc_fullgroups = pcg->pcg_next;
                   2427:                pc->pc_hits++;
                   2428:                pc->pc_nfull--;
                   2429:                mutex_exit(&pc->pc_lock);
                   2430:                return cc;
                   2431:        }
                   2432:
                   2433:        /*
                   2434:         * Nothing available locally or in cache.  Take the slow
                   2435:         * path: fetch a new object from the pool and construct
                   2436:         * it.
                   2437:         */
                   2438:        pc->pc_misses++;
                   2439:        mutex_exit(&pc->pc_lock);
                   2440:        pool_cache_cpu_exit(cc, s);
                   2441:
                   2442:        object = pool_get(&pc->pc_pool, flags);
                   2443:        *objectp = object;
                   2444:        if (object == NULL)
                   2445:                return NULL;
1.125     ad       2446:
1.134     ad       2447:        if ((*pc->pc_ctor)(pc->pc_arg, object, flags) != 0) {
                   2448:                pool_put(&pc->pc_pool, object);
                   2449:                *objectp = NULL;
                   2450:                return NULL;
1.43      thorpej  2451:        }
                   2452:
1.134     ad       2453:        KASSERT((((vaddr_t)object + pc->pc_pool.pr_itemoffset) &
                   2454:            (pc->pc_pool.pr_align - 1)) == 0);
1.43      thorpej  2455:
1.134     ad       2456:        if (pap != NULL) {
                   2457: #ifdef POOL_VTOPHYS
                   2458:                *pap = POOL_VTOPHYS(object);
                   2459: #else
                   2460:                *pap = POOL_PADDR_INVALID;
                   2461: #endif
1.102     chs      2462:        }
1.43      thorpej  2463:
1.125     ad       2464:        FREECHECK_OUT(&pc->pc_freecheck, object);
1.134     ad       2465:        return NULL;
1.43      thorpej  2466: }
                   2467:
                   2468: /*
1.134     ad       2469:  * pool_cache_get{,_paddr}:
1.43      thorpej  2470:  *
1.134     ad       2471:  *     Get an object from a pool cache (optionally returning
                   2472:  *     the physical address of the object).
1.43      thorpej  2473:  */
1.134     ad       2474: void *
                   2475: pool_cache_get_paddr(pool_cache_t pc, int flags, paddr_t *pap)
1.43      thorpej  2476: {
1.134     ad       2477:        pool_cache_cpu_t *cc;
                   2478:        pcg_t *pcg;
                   2479:        void *object;
1.60      thorpej  2480:        int s;
1.43      thorpej  2481:
1.134     ad       2482: #ifdef LOCKDEBUG
                   2483:        if (flags & PR_WAITOK)
                   2484:                ASSERT_SLEEPABLE(NULL, "pool_cache_get(PR_WAITOK)");
                   2485: #endif
1.125     ad       2486:
1.134     ad       2487:        cc = pool_cache_cpu_enter(pc, &s);
                   2488:        do {
                   2489:                /* Try and allocate an object from the current group. */
                   2490:                pcg = cc->cc_current;
                   2491:                if (pcg != NULL && pcg->pcg_avail > 0) {
                   2492:                        object = pcg->pcg_objects[--pcg->pcg_avail].pcgo_va;
                   2493:                        if (pap != NULL)
                   2494:                                *pap = pcg->pcg_objects[pcg->pcg_avail].pcgo_pa;
                   2495:                        pcg->pcg_objects[pcg->pcg_avail].pcgo_va = NULL;
                   2496:                        KASSERT(pcg->pcg_avail <= PCG_NOBJECTS);
                   2497:                        KASSERT(object != NULL);
                   2498:                        cc->cc_hits++;
                   2499:                        pool_cache_cpu_exit(cc, &s);
                   2500:                        FREECHECK_OUT(&pc->pc_freecheck, object);
                   2501:                        return object;
1.43      thorpej  2502:                }
                   2503:
                   2504:                /*
1.134     ad       2505:                 * That failed.  If the previous group isn't empty, swap
                   2506:                 * it with the current group and allocate from there.
1.43      thorpej  2507:                 */
1.134     ad       2508:                pcg = cc->cc_previous;
                   2509:                if (pcg != NULL && pcg->pcg_avail > 0) {
                   2510:                        cc->cc_previous = cc->cc_current;
                   2511:                        cc->cc_current = pcg;
                   2512:                        continue;
1.43      thorpej  2513:                }
                   2514:
1.134     ad       2515:                /*
                   2516:                 * Can't allocate from either group: try the slow path.
                   2517:                 * If get_slow() allocated an object for us, or if
                   2518:                 * no more objects are available, it will return NULL.
                   2519:                 * Otherwise, we need to retry.
                   2520:                 */
                   2521:                cc = pool_cache_get_slow(cc, &s, &object, pap, flags);
                   2522:        } while (cc != NULL);
1.43      thorpej  2523:
1.134     ad       2524:        return object;
1.51      thorpej  2525: }
                   2526:
1.134     ad       2527: #if __GNUC_PREREQ__(3, 0)
                   2528: __attribute ((noinline))
                   2529: #endif
                   2530: pool_cache_cpu_t *
                   2531: pool_cache_put_slow(pool_cache_cpu_t *cc, int *s, void *object, paddr_t pa)
1.51      thorpej  2532: {
1.134     ad       2533:        pcg_t *pcg, *cur;
                   2534:        uint64_t ncsw;
                   2535:        pool_cache_t pc;
1.51      thorpej  2536:
1.134     ad       2537:        pc = cc->cc_cache;
                   2538:        cc->cc_misses++;
1.43      thorpej  2539:
1.134     ad       2540:        /*
                   2541:         * No free slots locally.  Try to grab an empty, unused
                   2542:         * group from the cache.
                   2543:         */
                   2544:        if (!mutex_tryenter(&pc->pc_lock)) {
                   2545:                ncsw = curlwp->l_ncsw;
                   2546:                mutex_enter(&pc->pc_lock);
                   2547:                pc->pc_contended++;
1.102     chs      2548:
1.134     ad       2549:                /*
                   2550:                 * If we context switched while locking, then
                   2551:                 * our view of the per-CPU data is invalid:
                   2552:                 * retry.
                   2553:                 */
                   2554:                if (curlwp->l_ncsw != ncsw) {
                   2555:                        mutex_exit(&pc->pc_lock);
                   2556:                        pool_cache_cpu_exit(cc, s);
                   2557:                        return pool_cache_cpu_enter(pc, s);
                   2558:                }
                   2559:        }
1.130     ad       2560:
1.134     ad       2561:        if ((pcg = pc->pc_emptygroups) != NULL) {
                   2562:                /*
                   2563:                 * If there's a empty group, release our full
                   2564:                 * group back to the cache.  Install the empty
                   2565:                 * group as cc_current and return.
                   2566:                 */
                   2567:                if ((cur = cc->cc_current) != NULL) {
                   2568:                        KASSERT(cur->pcg_avail == PCG_NOBJECTS);
                   2569:                        cur->pcg_next = pc->pc_fullgroups;
                   2570:                        pc->pc_fullgroups = cur;
                   2571:                        pc->pc_nfull++;
1.102     chs      2572:                }
1.134     ad       2573:                KASSERT(pcg->pcg_avail == 0);
                   2574:                cc->cc_current = pcg;
                   2575:                pc->pc_emptygroups = pcg->pcg_next;
                   2576:                pc->pc_hits++;
                   2577:                pc->pc_nempty--;
                   2578:                mutex_exit(&pc->pc_lock);
                   2579:                return cc;
1.102     chs      2580:        }
1.105     christos 2581:
1.134     ad       2582:        /*
                   2583:         * Nothing available locally or in cache.  Take the
                   2584:         * slow path and try to allocate a new group that we
                   2585:         * can release to.
                   2586:         */
                   2587:        pc->pc_misses++;
                   2588:        mutex_exit(&pc->pc_lock);
                   2589:        pool_cache_cpu_exit(cc, s);
1.105     christos 2590:
1.134     ad       2591:        /*
                   2592:         * If we can't allocate a new group, just throw the
                   2593:         * object away.
                   2594:         */
                   2595:        pcg = pool_get(&pcgpool, PR_NOWAIT);
                   2596:        if (pcg == NULL) {
                   2597:                pool_cache_destruct_object(pc, object);
                   2598:                return NULL;
                   2599:        }
                   2600: #ifdef DIAGNOSTIC
                   2601:        memset(pcg, 0, sizeof(*pcg));
                   2602: #else
                   2603:        pcg->pcg_avail = 0;
                   2604: #endif
1.105     christos 2605:
1.134     ad       2606:        /*
                   2607:         * Add the empty group to the cache and try again.
                   2608:         */
                   2609:        mutex_enter(&pc->pc_lock);
                   2610:        pcg->pcg_next = pc->pc_emptygroups;
                   2611:        pc->pc_emptygroups = pcg;
                   2612:        pc->pc_nempty++;
                   2613:        mutex_exit(&pc->pc_lock);
1.103     chs      2614:
1.134     ad       2615:        return pool_cache_cpu_enter(pc, s);
                   2616: }
1.102     chs      2617:
1.43      thorpej  2618: /*
1.134     ad       2619:  * pool_cache_put{,_paddr}:
1.43      thorpej  2620:  *
1.134     ad       2621:  *     Put an object back to the pool cache (optionally caching the
                   2622:  *     physical address of the object).
1.43      thorpej  2623:  */
1.101     thorpej  2624: void
1.134     ad       2625: pool_cache_put_paddr(pool_cache_t pc, void *object, paddr_t pa)
1.43      thorpej  2626: {
1.134     ad       2627:        pool_cache_cpu_t *cc;
                   2628:        pcg_t *pcg;
                   2629:        int s;
1.101     thorpej  2630:
1.134     ad       2631:        FREECHECK_IN(&pc->pc_freecheck, object);
1.101     thorpej  2632:
1.134     ad       2633:        cc = pool_cache_cpu_enter(pc, &s);
                   2634:        do {
                   2635:                /* If the current group isn't full, release it there. */
                   2636:                pcg = cc->cc_current;
                   2637:                if (pcg != NULL && pcg->pcg_avail < PCG_NOBJECTS) {
                   2638:                        KASSERT(pcg->pcg_objects[pcg->pcg_avail].pcgo_va
                   2639:                            == NULL);
                   2640:                        pcg->pcg_objects[pcg->pcg_avail].pcgo_va = object;
                   2641:                        pcg->pcg_objects[pcg->pcg_avail].pcgo_pa = pa;
                   2642:                        pcg->pcg_avail++;
                   2643:                        cc->cc_hits++;
                   2644:                        pool_cache_cpu_exit(cc, &s);
                   2645:                        return;
                   2646:                }
1.43      thorpej  2647:
1.134     ad       2648:                /*
                   2649:                 * That failed.  If the previous group is empty, swap
                   2650:                 * it with the current group and try again.
                   2651:                 */
                   2652:                pcg = cc->cc_previous;
                   2653:                if (pcg != NULL && pcg->pcg_avail == 0) {
                   2654:                        cc->cc_previous = cc->cc_current;
                   2655:                        cc->cc_current = pcg;
                   2656:                        continue;
                   2657:                }
1.43      thorpej  2658:
1.134     ad       2659:                /*
                   2660:                 * Can't free to either group: try the slow path.
                   2661:                 * If put_slow() releases the object for us, it
                   2662:                 * will return NULL.  Otherwise we need to retry.
                   2663:                 */
                   2664:                cc = pool_cache_put_slow(cc, &s, object, pa);
                   2665:        } while (cc != NULL);
1.43      thorpej  2666: }
                   2667:
                   2668: /*
1.134     ad       2669:  * pool_cache_xcall:
1.43      thorpej  2670:  *
1.134     ad       2671:  *     Transfer objects from the per-CPU cache to the global cache.
                   2672:  *     Run within a cross-call thread.
1.43      thorpej  2673:  */
                   2674: static void
1.134     ad       2675: pool_cache_xcall(pool_cache_t pc)
1.43      thorpej  2676: {
1.134     ad       2677:        pool_cache_cpu_t *cc;
                   2678:        pcg_t *prev, *cur, **list;
                   2679:        int s = 0; /* XXXgcc */
                   2680:
                   2681:        cc = pool_cache_cpu_enter(pc, &s);
                   2682:        cur = cc->cc_current;
                   2683:        cc->cc_current = NULL;
                   2684:        prev = cc->cc_previous;
                   2685:        cc->cc_previous = NULL;
                   2686:        pool_cache_cpu_exit(cc, &s);
                   2687:
                   2688:        /*
                   2689:         * XXXSMP Go to splvm to prevent kernel_lock from being taken,
                   2690:         * because locks at IPL_SOFTXXX are still spinlocks.  Does not
                   2691:         * apply to IPL_SOFTBIO.  Cross-call threads do not take the
                   2692:         * kernel_lock.
1.101     thorpej  2693:         */
1.134     ad       2694:        s = splvm();
                   2695:        mutex_enter(&pc->pc_lock);
                   2696:        if (cur != NULL) {
                   2697:                if (cur->pcg_avail == PCG_NOBJECTS) {
                   2698:                        list = &pc->pc_fullgroups;
                   2699:                        pc->pc_nfull++;
                   2700:                } else if (cur->pcg_avail == 0) {
                   2701:                        list = &pc->pc_emptygroups;
                   2702:                        pc->pc_nempty++;
                   2703:                } else {
                   2704:                        list = &pc->pc_partgroups;
                   2705:                        pc->pc_npart++;
                   2706:                }
                   2707:                cur->pcg_next = *list;
                   2708:                *list = cur;
                   2709:        }
                   2710:        if (prev != NULL) {
                   2711:                if (prev->pcg_avail == PCG_NOBJECTS) {
                   2712:                        list = &pc->pc_fullgroups;
                   2713:                        pc->pc_nfull++;
                   2714:                } else if (prev->pcg_avail == 0) {
                   2715:                        list = &pc->pc_emptygroups;
                   2716:                        pc->pc_nempty++;
                   2717:                } else {
                   2718:                        list = &pc->pc_partgroups;
                   2719:                        pc->pc_npart++;
                   2720:                }
                   2721:                prev->pcg_next = *list;
                   2722:                *list = prev;
                   2723:        }
                   2724:        mutex_exit(&pc->pc_lock);
                   2725:        splx(s);
1.3       pk       2726: }
1.66      thorpej  2727:
                   2728: /*
                   2729:  * Pool backend allocators.
                   2730:  *
                   2731:  * Each pool has a backend allocator that handles allocation, deallocation,
                   2732:  * and any additional draining that might be needed.
                   2733:  *
                   2734:  * We provide two standard allocators:
                   2735:  *
                   2736:  *     pool_allocator_kmem - the default when no allocator is specified
                   2737:  *
                   2738:  *     pool_allocator_nointr - used for pools that will not be accessed
                   2739:  *     in interrupt context.
                   2740:  */
                   2741: void   *pool_page_alloc(struct pool *, int);
                   2742: void   pool_page_free(struct pool *, void *);
                   2743:
1.112     bjh21    2744: #ifdef POOL_SUBPAGE
                   2745: struct pool_allocator pool_allocator_kmem_fullpage = {
                   2746:        pool_page_alloc, pool_page_free, 0,
1.117     yamt     2747:        .pa_backingmapptr = &kmem_map,
1.112     bjh21    2748: };
                   2749: #else
1.66      thorpej  2750: struct pool_allocator pool_allocator_kmem = {
                   2751:        pool_page_alloc, pool_page_free, 0,
1.117     yamt     2752:        .pa_backingmapptr = &kmem_map,
1.66      thorpej  2753: };
1.112     bjh21    2754: #endif
1.66      thorpej  2755:
                   2756: void   *pool_page_alloc_nointr(struct pool *, int);
                   2757: void   pool_page_free_nointr(struct pool *, void *);
                   2758:
1.112     bjh21    2759: #ifdef POOL_SUBPAGE
                   2760: struct pool_allocator pool_allocator_nointr_fullpage = {
                   2761:        pool_page_alloc_nointr, pool_page_free_nointr, 0,
1.117     yamt     2762:        .pa_backingmapptr = &kernel_map,
1.112     bjh21    2763: };
                   2764: #else
1.66      thorpej  2765: struct pool_allocator pool_allocator_nointr = {
                   2766:        pool_page_alloc_nointr, pool_page_free_nointr, 0,
1.117     yamt     2767:        .pa_backingmapptr = &kernel_map,
1.66      thorpej  2768: };
1.112     bjh21    2769: #endif
1.66      thorpej  2770:
                   2771: #ifdef POOL_SUBPAGE
                   2772: void   *pool_subpage_alloc(struct pool *, int);
                   2773: void   pool_subpage_free(struct pool *, void *);
                   2774:
1.112     bjh21    2775: struct pool_allocator pool_allocator_kmem = {
                   2776:        pool_subpage_alloc, pool_subpage_free, POOL_SUBPAGE,
1.117     yamt     2777:        .pa_backingmapptr = &kmem_map,
1.112     bjh21    2778: };
                   2779:
                   2780: void   *pool_subpage_alloc_nointr(struct pool *, int);
                   2781: void   pool_subpage_free_nointr(struct pool *, void *);
                   2782:
                   2783: struct pool_allocator pool_allocator_nointr = {
                   2784:        pool_subpage_alloc, pool_subpage_free, POOL_SUBPAGE,
1.117     yamt     2785:        .pa_backingmapptr = &kmem_map,
1.66      thorpej  2786: };
                   2787: #endif /* POOL_SUBPAGE */
                   2788:
1.117     yamt     2789: static void *
                   2790: pool_allocator_alloc(struct pool *pp, int flags)
1.66      thorpej  2791: {
1.117     yamt     2792:        struct pool_allocator *pa = pp->pr_alloc;
1.66      thorpej  2793:        void *res;
                   2794:
1.117     yamt     2795:        res = (*pa->pa_alloc)(pp, flags);
                   2796:        if (res == NULL && (flags & PR_WAITOK) == 0) {
1.66      thorpej  2797:                /*
1.117     yamt     2798:                 * We only run the drain hook here if PR_NOWAIT.
                   2799:                 * In other cases, the hook will be run in
                   2800:                 * pool_reclaim().
1.66      thorpej  2801:                 */
1.117     yamt     2802:                if (pp->pr_drain_hook != NULL) {
                   2803:                        (*pp->pr_drain_hook)(pp->pr_drain_hook_arg, flags);
                   2804:                        res = (*pa->pa_alloc)(pp, flags);
1.66      thorpej  2805:                }
1.117     yamt     2806:        }
                   2807:        return res;
1.66      thorpej  2808: }
                   2809:
1.117     yamt     2810: static void
1.66      thorpej  2811: pool_allocator_free(struct pool *pp, void *v)
                   2812: {
                   2813:        struct pool_allocator *pa = pp->pr_alloc;
                   2814:
                   2815:        (*pa->pa_free)(pp, v);
                   2816: }
                   2817:
                   2818: void *
1.124     yamt     2819: pool_page_alloc(struct pool *pp, int flags)
1.66      thorpej  2820: {
1.127     thorpej  2821:        bool waitok = (flags & PR_WAITOK) ? true : false;
1.66      thorpej  2822:
1.100     yamt     2823:        return ((void *) uvm_km_alloc_poolpage_cache(kmem_map, waitok));
1.66      thorpej  2824: }
                   2825:
                   2826: void
1.124     yamt     2827: pool_page_free(struct pool *pp, void *v)
1.66      thorpej  2828: {
                   2829:
1.98      yamt     2830:        uvm_km_free_poolpage_cache(kmem_map, (vaddr_t) v);
                   2831: }
                   2832:
                   2833: static void *
1.124     yamt     2834: pool_page_alloc_meta(struct pool *pp, int flags)
1.98      yamt     2835: {
1.127     thorpej  2836:        bool waitok = (flags & PR_WAITOK) ? true : false;
1.98      yamt     2837:
1.100     yamt     2838:        return ((void *) uvm_km_alloc_poolpage(kmem_map, waitok));
1.98      yamt     2839: }
                   2840:
                   2841: static void
1.124     yamt     2842: pool_page_free_meta(struct pool *pp, void *v)
1.98      yamt     2843: {
                   2844:
1.100     yamt     2845:        uvm_km_free_poolpage(kmem_map, (vaddr_t) v);
1.66      thorpej  2846: }
                   2847:
                   2848: #ifdef POOL_SUBPAGE
                   2849: /* Sub-page allocator, for machines with large hardware pages. */
                   2850: void *
                   2851: pool_subpage_alloc(struct pool *pp, int flags)
                   2852: {
1.134     ad       2853:        return pool_get(&psppool, flags);
1.66      thorpej  2854: }
                   2855:
                   2856: void
                   2857: pool_subpage_free(struct pool *pp, void *v)
                   2858: {
                   2859:        pool_put(&psppool, v);
                   2860: }
                   2861:
                   2862: /* We don't provide a real nointr allocator.  Maybe later. */
                   2863: void *
1.112     bjh21    2864: pool_subpage_alloc_nointr(struct pool *pp, int flags)
1.66      thorpej  2865: {
                   2866:
                   2867:        return (pool_subpage_alloc(pp, flags));
                   2868: }
                   2869:
                   2870: void
1.112     bjh21    2871: pool_subpage_free_nointr(struct pool *pp, void *v)
1.66      thorpej  2872: {
                   2873:
                   2874:        pool_subpage_free(pp, v);
                   2875: }
1.112     bjh21    2876: #endif /* POOL_SUBPAGE */
1.66      thorpej  2877: void *
1.124     yamt     2878: pool_page_alloc_nointr(struct pool *pp, int flags)
1.66      thorpej  2879: {
1.127     thorpej  2880:        bool waitok = (flags & PR_WAITOK) ? true : false;
1.66      thorpej  2881:
1.100     yamt     2882:        return ((void *) uvm_km_alloc_poolpage_cache(kernel_map, waitok));
1.66      thorpej  2883: }
                   2884:
                   2885: void
1.124     yamt     2886: pool_page_free_nointr(struct pool *pp, void *v)
1.66      thorpej  2887: {
                   2888:
1.98      yamt     2889:        uvm_km_free_poolpage_cache(kernel_map, (vaddr_t) v);
1.66      thorpej  2890: }

CVSweb <webmaster@jp.NetBSD.org>