[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 1.63

1.63    ! chs         1: /*     $NetBSD: subr_pool.c,v 1.62 2001/10/07 12:44:06 bjh21 Exp $     */
1.1       pk          2:
                      3: /*-
1.43      thorpej     4:  * Copyright (c) 1997, 1999, 2000 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
                      9:  * Simulation Facility, NASA Ames Research Center.
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:  */
1.24      scottr     39:
1.25      thorpej    40: #include "opt_pool.h"
1.24      scottr     41: #include "opt_poollog.h"
1.28      thorpej    42: #include "opt_lockdebug.h"
1.1       pk         43:
                     44: #include <sys/param.h>
                     45: #include <sys/systm.h>
                     46: #include <sys/proc.h>
                     47: #include <sys/errno.h>
                     48: #include <sys/kernel.h>
                     49: #include <sys/malloc.h>
                     50: #include <sys/lock.h>
                     51: #include <sys/pool.h>
1.20      thorpej    52: #include <sys/syslog.h>
1.3       pk         53:
                     54: #include <uvm/uvm.h>
                     55:
1.1       pk         56: /*
                     57:  * Pool resource management utility.
1.3       pk         58:  *
                     59:  * Memory is allocated in pages which are split into pieces according
                     60:  * to the pool item size. Each page is kept on a list headed by `pr_pagelist'
                     61:  * in the pool structure and the individual pool items are on a linked list
                     62:  * headed by `ph_itemlist' in each page header. The memory for building
                     63:  * the page list is either taken from the allocated pages themselves (for
                     64:  * small pool items) or taken from an internal pool of page headers (`phpool').
1.1       pk         65:  */
                     66:
1.3       pk         67: /* List of all pools */
1.5       thorpej    68: TAILQ_HEAD(,pool) pool_head = TAILQ_HEAD_INITIALIZER(pool_head);
1.3       pk         69:
                     70: /* Private pool for page header structures */
                     71: static struct pool phpool;
                     72:
1.62      bjh21      73: #ifdef POOL_SUBPAGE
                     74: /* Pool of subpages for use by normal pools. */
                     75: static struct pool psppool;
                     76: #endif
                     77:
1.3       pk         78: /* # of seconds to retain page after last use */
                     79: int pool_inactive_time = 10;
                     80:
                     81: /* Next candidate for drainage (see pool_drain()) */
1.23      thorpej    82: static struct pool     *drainpp;
                     83:
                     84: /* This spin lock protects both pool_head and drainpp. */
                     85: struct simplelock pool_head_slock = SIMPLELOCK_INITIALIZER;
1.3       pk         86:
                     87: struct pool_item_header {
                     88:        /* Page headers */
                     89:        TAILQ_ENTRY(pool_item_header)
                     90:                                ph_pagelist;    /* pool page list */
                     91:        TAILQ_HEAD(,pool_item)  ph_itemlist;    /* chunk list for this page */
                     92:        LIST_ENTRY(pool_item_header)
                     93:                                ph_hashlist;    /* Off-page page headers */
                     94:        int                     ph_nmissing;    /* # of chunks in use */
                     95:        caddr_t                 ph_page;        /* this page's address */
                     96:        struct timeval          ph_time;        /* last referenced */
                     97: };
1.61      chs        98: TAILQ_HEAD(pool_pagelist,pool_item_header);
1.3       pk         99:
1.1       pk        100: struct pool_item {
1.3       pk        101: #ifdef DIAGNOSTIC
                    102:        int pi_magic;
1.33      chs       103: #endif
1.25      thorpej   104: #define        PI_MAGIC 0xdeadbeef
1.3       pk        105:        /* Other entries use only this list entry */
                    106:        TAILQ_ENTRY(pool_item)  pi_list;
                    107: };
                    108:
1.25      thorpej   109: #define        PR_HASH_INDEX(pp,addr) \
1.3       pk        110:        (((u_long)(addr) >> (pp)->pr_pageshift) & (PR_HASHTABSIZE - 1))
                    111:
1.53      thorpej   112: #define        POOL_NEEDS_CATCHUP(pp)                                          \
                    113:        ((pp)->pr_nitems < (pp)->pr_minitems)
                    114:
1.43      thorpej   115: /*
                    116:  * Pool cache management.
                    117:  *
                    118:  * Pool caches provide a way for constructed objects to be cached by the
                    119:  * pool subsystem.  This can lead to performance improvements by avoiding
                    120:  * needless object construction/destruction; it is deferred until absolutely
                    121:  * necessary.
                    122:  *
                    123:  * Caches are grouped into cache groups.  Each cache group references
                    124:  * up to 16 constructed objects.  When a cache allocates an object
                    125:  * from the pool, it calls the object's constructor and places it into
                    126:  * a cache group.  When a cache group frees an object back to the pool,
                    127:  * it first calls the object's destructor.  This allows the object to
                    128:  * persist in constructed form while freed to the cache.
                    129:  *
                    130:  * Multiple caches may exist for each pool.  This allows a single
                    131:  * object type to have multiple constructed forms.  The pool references
                    132:  * each cache, so that when a pool is drained by the pagedaemon, it can
                    133:  * drain each individual cache as well.  Each time a cache is drained,
                    134:  * the most idle cache group is freed to the pool in its entirety.
                    135:  *
                    136:  * Pool caches are layed on top of pools.  By layering them, we can avoid
                    137:  * the complexity of cache management for pools which would not benefit
                    138:  * from it.
                    139:  */
                    140:
                    141: /* The cache group pool. */
                    142: static struct pool pcgpool;
                    143:
                    144: /* The pool cache group. */
                    145: #define        PCG_NOBJECTS            16
                    146: struct pool_cache_group {
                    147:        TAILQ_ENTRY(pool_cache_group)
                    148:                pcg_list;       /* link in the pool cache's group list */
                    149:        u_int   pcg_avail;      /* # available objects */
                    150:                                /* pointers to the objects */
                    151:        void    *pcg_objects[PCG_NOBJECTS];
                    152: };
1.3       pk        153:
1.43      thorpej   154: static void    pool_cache_reclaim(struct pool_cache *);
1.3       pk        155:
1.42      thorpej   156: static int     pool_catchup(struct pool *);
1.55      thorpej   157: static void    pool_prime_page(struct pool *, caddr_t,
                    158:                    struct pool_item_header *);
1.42      thorpej   159: static void    *pool_page_alloc(unsigned long, int, int);
                    160: static void    pool_page_free(void *, unsigned long, int);
1.62      bjh21     161: #ifdef POOL_SUBPAGE
                    162: static void    *pool_subpage_alloc(unsigned long, int, int);
                    163: static void    pool_subpage_free(void *, unsigned long, int);
                    164: #endif
1.3       pk        165:
1.42      thorpej   166: static void pool_print1(struct pool *, const char *,
                    167:        void (*)(const char *, ...));
1.3       pk        168:
                    169: /*
1.52      thorpej   170:  * Pool log entry. An array of these is allocated in pool_init().
1.3       pk        171:  */
                    172: struct pool_log {
                    173:        const char      *pl_file;
                    174:        long            pl_line;
                    175:        int             pl_action;
1.25      thorpej   176: #define        PRLOG_GET       1
                    177: #define        PRLOG_PUT       2
1.3       pk        178:        void            *pl_addr;
1.1       pk        179: };
                    180:
1.3       pk        181: /* Number of entries in pool log buffers */
1.17      thorpej   182: #ifndef POOL_LOGSIZE
                    183: #define        POOL_LOGSIZE    10
                    184: #endif
                    185:
                    186: int pool_logsize = POOL_LOGSIZE;
1.1       pk        187:
1.59      thorpej   188: #ifdef POOL_DIAGNOSTIC
1.42      thorpej   189: static __inline void
                    190: pr_log(struct pool *pp, void *v, int action, const char *file, long line)
1.3       pk        191: {
                    192:        int n = pp->pr_curlogentry;
                    193:        struct pool_log *pl;
                    194:
1.20      thorpej   195:        if ((pp->pr_roflags & PR_LOGGING) == 0)
1.3       pk        196:                return;
                    197:
                    198:        /*
                    199:         * Fill in the current entry. Wrap around and overwrite
                    200:         * the oldest entry if necessary.
                    201:         */
                    202:        pl = &pp->pr_log[n];
                    203:        pl->pl_file = file;
                    204:        pl->pl_line = line;
                    205:        pl->pl_action = action;
                    206:        pl->pl_addr = v;
                    207:        if (++n >= pp->pr_logsize)
                    208:                n = 0;
                    209:        pp->pr_curlogentry = n;
                    210: }
                    211:
                    212: static void
1.42      thorpej   213: pr_printlog(struct pool *pp, struct pool_item *pi,
                    214:     void (*pr)(const char *, ...))
1.3       pk        215: {
                    216:        int i = pp->pr_logsize;
                    217:        int n = pp->pr_curlogentry;
                    218:
1.20      thorpej   219:        if ((pp->pr_roflags & PR_LOGGING) == 0)
1.3       pk        220:                return;
                    221:
                    222:        /*
                    223:         * Print all entries in this pool's log.
                    224:         */
                    225:        while (i-- > 0) {
                    226:                struct pool_log *pl = &pp->pr_log[n];
                    227:                if (pl->pl_action != 0) {
1.25      thorpej   228:                        if (pi == NULL || pi == pl->pl_addr) {
                    229:                                (*pr)("\tlog entry %d:\n", i);
                    230:                                (*pr)("\t\taction = %s, addr = %p\n",
                    231:                                    pl->pl_action == PRLOG_GET ? "get" : "put",
                    232:                                    pl->pl_addr);
                    233:                                (*pr)("\t\tfile: %s at line %lu\n",
                    234:                                    pl->pl_file, pl->pl_line);
                    235:                        }
1.3       pk        236:                }
                    237:                if (++n >= pp->pr_logsize)
                    238:                        n = 0;
                    239:        }
                    240: }
1.25      thorpej   241:
1.42      thorpej   242: static __inline void
                    243: pr_enter(struct pool *pp, const char *file, long line)
1.25      thorpej   244: {
                    245:
1.34      thorpej   246:        if (__predict_false(pp->pr_entered_file != NULL)) {
1.25      thorpej   247:                printf("pool %s: reentrancy at file %s line %ld\n",
                    248:                    pp->pr_wchan, file, line);
                    249:                printf("         previous entry at file %s line %ld\n",
                    250:                    pp->pr_entered_file, pp->pr_entered_line);
                    251:                panic("pr_enter");
                    252:        }
                    253:
                    254:        pp->pr_entered_file = file;
                    255:        pp->pr_entered_line = line;
                    256: }
                    257:
1.42      thorpej   258: static __inline void
                    259: pr_leave(struct pool *pp)
1.25      thorpej   260: {
                    261:
1.34      thorpej   262:        if (__predict_false(pp->pr_entered_file == NULL)) {
1.25      thorpej   263:                printf("pool %s not entered?\n", pp->pr_wchan);
                    264:                panic("pr_leave");
                    265:        }
                    266:
                    267:        pp->pr_entered_file = NULL;
                    268:        pp->pr_entered_line = 0;
                    269: }
                    270:
1.42      thorpej   271: static __inline void
                    272: pr_enter_check(struct pool *pp, void (*pr)(const char *, ...))
1.25      thorpej   273: {
                    274:
                    275:        if (pp->pr_entered_file != NULL)
                    276:                (*pr)("\n\tcurrently entered from file %s line %ld\n",
                    277:                    pp->pr_entered_file, pp->pr_entered_line);
                    278: }
1.3       pk        279: #else
1.25      thorpej   280: #define        pr_log(pp, v, action, file, line)
                    281: #define        pr_printlog(pp, pi, pr)
                    282: #define        pr_enter(pp, file, line)
                    283: #define        pr_leave(pp)
                    284: #define        pr_enter_check(pp, pr)
1.59      thorpej   285: #endif /* POOL_DIAGNOSTIC */
1.3       pk        286:
                    287: /*
                    288:  * Return the pool page header based on page address.
                    289:  */
1.42      thorpej   290: static __inline struct pool_item_header *
                    291: pr_find_pagehead(struct pool *pp, caddr_t page)
1.3       pk        292: {
                    293:        struct pool_item_header *ph;
                    294:
1.20      thorpej   295:        if ((pp->pr_roflags & PR_PHINPAGE) != 0)
1.3       pk        296:                return ((struct pool_item_header *)(page + pp->pr_phoffset));
                    297:
                    298:        for (ph = LIST_FIRST(&pp->pr_hashtab[PR_HASH_INDEX(pp, page)]);
                    299:             ph != NULL;
                    300:             ph = LIST_NEXT(ph, ph_hashlist)) {
                    301:                if (ph->ph_page == page)
                    302:                        return (ph);
                    303:        }
                    304:        return (NULL);
                    305: }
                    306:
                    307: /*
                    308:  * Remove a page from the pool.
                    309:  */
1.42      thorpej   310: static __inline void
1.61      chs       311: pr_rmpage(struct pool *pp, struct pool_item_header *ph,
                    312:      struct pool_pagelist *pq)
1.3       pk        313: {
1.61      chs       314:        int s;
1.3       pk        315:
                    316:        /*
1.7       thorpej   317:         * If the page was idle, decrement the idle page count.
1.3       pk        318:         */
1.6       thorpej   319:        if (ph->ph_nmissing == 0) {
                    320: #ifdef DIAGNOSTIC
                    321:                if (pp->pr_nidle == 0)
                    322:                        panic("pr_rmpage: nidle inconsistent");
1.20      thorpej   323:                if (pp->pr_nitems < pp->pr_itemsperpage)
                    324:                        panic("pr_rmpage: nitems inconsistent");
1.6       thorpej   325: #endif
                    326:                pp->pr_nidle--;
                    327:        }
1.7       thorpej   328:
1.20      thorpej   329:        pp->pr_nitems -= pp->pr_itemsperpage;
                    330:
1.7       thorpej   331:        /*
1.61      chs       332:         * Unlink a page from the pool and release it (or queue it for release).
1.7       thorpej   333:         */
                    334:        TAILQ_REMOVE(&pp->pr_pagelist, ph, ph_pagelist);
1.61      chs       335:        if (pq) {
                    336:                TAILQ_INSERT_HEAD(pq, ph, ph_pagelist);
                    337:        } else {
                    338:                (*pp->pr_free)(ph->ph_page, pp->pr_pagesz, pp->pr_mtype);
                    339:                if ((pp->pr_roflags & PR_PHINPAGE) == 0) {
                    340:                        LIST_REMOVE(ph, ph_hashlist);
                    341:                        s = splhigh();
                    342:                        pool_put(&phpool, ph);
                    343:                        splx(s);
                    344:                }
                    345:        }
1.7       thorpej   346:        pp->pr_npages--;
                    347:        pp->pr_npagefree++;
1.6       thorpej   348:
1.3       pk        349:        if (pp->pr_curpage == ph) {
                    350:                /*
                    351:                 * Find a new non-empty page header, if any.
                    352:                 * Start search from the page head, to increase the
                    353:                 * chance for "high water" pages to be freed.
                    354:                 */
1.61      chs       355:                TAILQ_FOREACH(ph, &pp->pr_pagelist, ph_pagelist)
1.3       pk        356:                        if (TAILQ_FIRST(&ph->ph_itemlist) != NULL)
                    357:                                break;
                    358:
                    359:                pp->pr_curpage = ph;
1.21      thorpej   360:        }
1.3       pk        361: }
                    362:
                    363: /*
                    364:  * Initialize the given pool resource structure.
                    365:  *
                    366:  * We export this routine to allow other kernel parts to declare
                    367:  * static pools that must be initialized before malloc() is available.
                    368:  */
                    369: void
1.42      thorpej   370: pool_init(struct pool *pp, size_t size, u_int align, u_int ioff, int flags,
                    371:     const char *wchan, size_t pagesz,
                    372:     void *(*alloc)(unsigned long, int, int),
                    373:     void (*release)(void *, unsigned long, int),
                    374:     int mtype)
1.3       pk        375: {
1.16      briggs    376:        int off, slack, i;
1.3       pk        377:
1.25      thorpej   378: #ifdef POOL_DIAGNOSTIC
                    379:        /*
                    380:         * Always log if POOL_DIAGNOSTIC is defined.
                    381:         */
                    382:        if (pool_logsize != 0)
                    383:                flags |= PR_LOGGING;
                    384: #endif
                    385:
1.3       pk        386:        /*
                    387:         * Check arguments and construct default values.
                    388:         */
1.36      pk        389:        if (!powerof2(pagesz))
1.3       pk        390:                panic("pool_init: page size invalid (%lx)\n", (u_long)pagesz);
                    391:
1.4       thorpej   392:        if (alloc == NULL && release == NULL) {
1.62      bjh21     393: #ifdef POOL_SUBPAGE
                    394:                alloc = pool_subpage_alloc;
                    395:                release = pool_subpage_free;
                    396:                pagesz = POOL_SUBPAGE;
                    397: #else
1.3       pk        398:                alloc = pool_page_alloc;
                    399:                release = pool_page_free;
1.4       thorpej   400:                pagesz = PAGE_SIZE;     /* Rounds to PAGE_SIZE anyhow. */
1.62      bjh21     401: #endif
1.4       thorpej   402:        } else if ((alloc != NULL && release != NULL) == 0) {
                    403:                /* If you specifiy one, must specify both. */
                    404:                panic("pool_init: must specify alloc and release together");
                    405:        }
1.62      bjh21     406: #ifdef POOL_SUBPAGE
                    407:        else if (alloc == pool_page_alloc_nointr &&
                    408:            release == pool_page_free_nointr)
                    409:                pagesz = POOL_SUBPAGE;
                    410: #endif
1.4       thorpej   411:
1.3       pk        412:        if (pagesz == 0)
                    413:                pagesz = PAGE_SIZE;
                    414:
                    415:        if (align == 0)
                    416:                align = ALIGN(1);
1.14      thorpej   417:
                    418:        if (size < sizeof(struct pool_item))
                    419:                size = sizeof(struct pool_item);
1.3       pk        420:
1.35      pk        421:        size = ALIGN(size);
1.43      thorpej   422:        if (size > pagesz)
1.35      pk        423:                panic("pool_init: pool item size (%lu) too large",
                    424:                      (u_long)size);
                    425:
1.3       pk        426:        /*
                    427:         * Initialize the pool structure.
                    428:         */
                    429:        TAILQ_INIT(&pp->pr_pagelist);
1.43      thorpej   430:        TAILQ_INIT(&pp->pr_cachelist);
1.3       pk        431:        pp->pr_curpage = NULL;
                    432:        pp->pr_npages = 0;
                    433:        pp->pr_minitems = 0;
                    434:        pp->pr_minpages = 0;
                    435:        pp->pr_maxpages = UINT_MAX;
1.20      thorpej   436:        pp->pr_roflags = flags;
                    437:        pp->pr_flags = 0;
1.35      pk        438:        pp->pr_size = size;
1.3       pk        439:        pp->pr_align = align;
                    440:        pp->pr_wchan = wchan;
                    441:        pp->pr_mtype = mtype;
                    442:        pp->pr_alloc = alloc;
                    443:        pp->pr_free = release;
                    444:        pp->pr_pagesz = pagesz;
                    445:        pp->pr_pagemask = ~(pagesz - 1);
                    446:        pp->pr_pageshift = ffs(pagesz) - 1;
1.20      thorpej   447:        pp->pr_nitems = 0;
                    448:        pp->pr_nout = 0;
                    449:        pp->pr_hardlimit = UINT_MAX;
                    450:        pp->pr_hardlimit_warning = NULL;
1.31      thorpej   451:        pp->pr_hardlimit_ratecap.tv_sec = 0;
                    452:        pp->pr_hardlimit_ratecap.tv_usec = 0;
                    453:        pp->pr_hardlimit_warning_last.tv_sec = 0;
                    454:        pp->pr_hardlimit_warning_last.tv_usec = 0;
1.3       pk        455:
                    456:        /*
                    457:         * Decide whether to put the page header off page to avoid
                    458:         * wasting too large a part of the page. Off-page page headers
                    459:         * go on a hash table, so we can match a returned item
                    460:         * with its header based on the page address.
                    461:         * We use 1/16 of the page size as the threshold (XXX: tune)
                    462:         */
                    463:        if (pp->pr_size < pagesz/16) {
                    464:                /* Use the end of the page for the page header */
1.20      thorpej   465:                pp->pr_roflags |= PR_PHINPAGE;
1.3       pk        466:                pp->pr_phoffset = off =
                    467:                        pagesz - ALIGN(sizeof(struct pool_item_header));
1.2       pk        468:        } else {
1.3       pk        469:                /* The page header will be taken from our page header pool */
                    470:                pp->pr_phoffset = 0;
                    471:                off = pagesz;
1.16      briggs    472:                for (i = 0; i < PR_HASHTABSIZE; i++) {
                    473:                        LIST_INIT(&pp->pr_hashtab[i]);
                    474:                }
1.2       pk        475:        }
1.1       pk        476:
1.3       pk        477:        /*
                    478:         * Alignment is to take place at `ioff' within the item. This means
                    479:         * we must reserve up to `align - 1' bytes on the page to allow
                    480:         * appropriate positioning of each item.
                    481:         *
                    482:         * Silently enforce `0 <= ioff < align'.
                    483:         */
                    484:        pp->pr_itemoffset = ioff = ioff % align;
                    485:        pp->pr_itemsperpage = (off - ((align - ioff) % align)) / pp->pr_size;
1.43      thorpej   486:        KASSERT(pp->pr_itemsperpage != 0);
1.3       pk        487:
                    488:        /*
                    489:         * Use the slack between the chunks and the page header
                    490:         * for "cache coloring".
                    491:         */
                    492:        slack = off - pp->pr_itemsperpage * pp->pr_size;
                    493:        pp->pr_maxcolor = (slack / align) * align;
                    494:        pp->pr_curcolor = 0;
                    495:
                    496:        pp->pr_nget = 0;
                    497:        pp->pr_nfail = 0;
                    498:        pp->pr_nput = 0;
                    499:        pp->pr_npagealloc = 0;
                    500:        pp->pr_npagefree = 0;
1.1       pk        501:        pp->pr_hiwat = 0;
1.8       thorpej   502:        pp->pr_nidle = 0;
1.3       pk        503:
1.59      thorpej   504: #ifdef POOL_DIAGNOSTIC
1.25      thorpej   505:        if (flags & PR_LOGGING) {
                    506:                if (kmem_map == NULL ||
                    507:                    (pp->pr_log = malloc(pool_logsize * sizeof(struct pool_log),
                    508:                     M_TEMP, M_NOWAIT)) == NULL)
1.20      thorpej   509:                        pp->pr_roflags &= ~PR_LOGGING;
1.3       pk        510:                pp->pr_curlogentry = 0;
                    511:                pp->pr_logsize = pool_logsize;
                    512:        }
1.59      thorpej   513: #endif
1.25      thorpej   514:
                    515:        pp->pr_entered_file = NULL;
                    516:        pp->pr_entered_line = 0;
1.3       pk        517:
1.21      thorpej   518:        simple_lock_init(&pp->pr_slock);
1.1       pk        519:
1.3       pk        520:        /*
1.43      thorpej   521:         * Initialize private page header pool and cache magazine pool if we
                    522:         * haven't done so yet.
1.23      thorpej   523:         * XXX LOCKING.
1.3       pk        524:         */
                    525:        if (phpool.pr_size == 0) {
1.62      bjh21     526: #ifdef POOL_SUBPAGE
                    527:                pool_init(&phpool, sizeof(struct pool_item_header), 0, 0, 0,
                    528:                    "phpool", PAGE_SIZE, pool_page_alloc, pool_page_free, 0);
                    529:                pool_init(&psppool, POOL_SUBPAGE, POOL_SUBPAGE, 0,
                    530:                    PR_RECURSIVE, "psppool", PAGE_SIZE,
                    531:                    pool_page_alloc, pool_page_free, 0);
                    532: #else
1.3       pk        533:                pool_init(&phpool, sizeof(struct pool_item_header), 0, 0,
1.43      thorpej   534:                    0, "phpool", 0, 0, 0, 0);
1.62      bjh21     535: #endif
1.43      thorpej   536:                pool_init(&pcgpool, sizeof(struct pool_cache_group), 0, 0,
                    537:                    0, "pcgpool", 0, 0, 0, 0);
1.1       pk        538:        }
                    539:
1.23      thorpej   540:        /* Insert into the list of all pools. */
                    541:        simple_lock(&pool_head_slock);
                    542:        TAILQ_INSERT_TAIL(&pool_head, pp, pr_poollist);
                    543:        simple_unlock(&pool_head_slock);
1.1       pk        544: }
                    545:
                    546: /*
                    547:  * De-commision a pool resource.
                    548:  */
                    549: void
1.42      thorpej   550: pool_destroy(struct pool *pp)
1.1       pk        551: {
1.3       pk        552:        struct pool_item_header *ph;
1.43      thorpej   553:        struct pool_cache *pc;
                    554:
                    555:        /* Destroy all caches for this pool. */
                    556:        while ((pc = TAILQ_FIRST(&pp->pr_cachelist)) != NULL)
                    557:                pool_cache_destroy(pc);
1.3       pk        558:
                    559: #ifdef DIAGNOSTIC
1.20      thorpej   560:        if (pp->pr_nout != 0) {
1.25      thorpej   561:                pr_printlog(pp, NULL, printf);
1.20      thorpej   562:                panic("pool_destroy: pool busy: still out: %u\n",
                    563:                    pp->pr_nout);
1.3       pk        564:        }
                    565: #endif
1.1       pk        566:
1.3       pk        567:        /* Remove all pages */
1.20      thorpej   568:        if ((pp->pr_roflags & PR_STATIC) == 0)
1.61      chs       569:                while ((ph = TAILQ_FIRST(&pp->pr_pagelist)) != NULL)
                    570:                        pr_rmpage(pp, ph, NULL);
1.3       pk        571:
                    572:        /* Remove from global pool list */
1.23      thorpej   573:        simple_lock(&pool_head_slock);
1.3       pk        574:        TAILQ_REMOVE(&pool_head, pp, pr_poollist);
1.61      chs       575:        if (drainpp == pp) {
                    576:                drainpp = NULL;
                    577:        }
1.23      thorpej   578:        simple_unlock(&pool_head_slock);
1.3       pk        579:
1.59      thorpej   580: #ifdef POOL_DIAGNOSTIC
1.20      thorpej   581:        if ((pp->pr_roflags & PR_LOGGING) != 0)
1.3       pk        582:                free(pp->pr_log, M_TEMP);
1.59      thorpej   583: #endif
1.2       pk        584:
1.20      thorpej   585:        if (pp->pr_roflags & PR_FREEHEADER)
1.3       pk        586:                free(pp, M_POOL);
1.1       pk        587: }
                    588:
1.55      thorpej   589: static __inline struct pool_item_header *
                    590: pool_alloc_item_header(struct pool *pp, caddr_t storage, int flags)
                    591: {
                    592:        struct pool_item_header *ph;
                    593:        int s;
                    594:
                    595:        LOCK_ASSERT(simple_lock_held(&pp->pr_slock) == 0);
                    596:
                    597:        if ((pp->pr_roflags & PR_PHINPAGE) != 0)
                    598:                ph = (struct pool_item_header *) (storage + pp->pr_phoffset);
                    599:        else {
                    600:                s = splhigh();
                    601:                ph = pool_get(&phpool, flags);
                    602:                splx(s);
                    603:        }
                    604:
                    605:        return (ph);
                    606: }
1.1       pk        607:
                    608: /*
1.3       pk        609:  * Grab an item from the pool; must be called at appropriate spl level
1.1       pk        610:  */
1.3       pk        611: void *
1.59      thorpej   612: #ifdef POOL_DIAGNOSTIC
1.42      thorpej   613: _pool_get(struct pool *pp, int flags, const char *file, long line)
1.56      sommerfe  614: #else
                    615: pool_get(struct pool *pp, int flags)
                    616: #endif
1.1       pk        617: {
                    618:        struct pool_item *pi;
1.3       pk        619:        struct pool_item_header *ph;
1.55      thorpej   620:        void *v;
1.1       pk        621:
1.2       pk        622: #ifdef DIAGNOSTIC
1.34      thorpej   623:        if (__predict_false((pp->pr_roflags & PR_STATIC) &&
                    624:                            (flags & PR_MALLOCOK))) {
1.25      thorpej   625:                pr_printlog(pp, NULL, printf);
1.2       pk        626:                panic("pool_get: static");
1.3       pk        627:        }
1.2       pk        628:
1.37      sommerfe  629:        if (__predict_false(curproc == NULL && doing_shutdown == 0 &&
                    630:                            (flags & PR_WAITOK) != 0))
1.3       pk        631:                panic("pool_get: must have NOWAIT");
1.58      thorpej   632:
                    633: #ifdef LOCKDEBUG
                    634:        if (flags & PR_WAITOK)
                    635:                simple_lock_only_held(NULL, "pool_get(PR_WAITOK)");
1.56      sommerfe  636: #endif
1.58      thorpej   637: #endif /* DIAGNOSTIC */
1.1       pk        638:
1.21      thorpej   639:        simple_lock(&pp->pr_slock);
1.25      thorpej   640:        pr_enter(pp, file, line);
1.20      thorpej   641:
                    642:  startover:
                    643:        /*
                    644:         * Check to see if we've reached the hard limit.  If we have,
                    645:         * and we can wait, then wait until an item has been returned to
                    646:         * the pool.
                    647:         */
                    648: #ifdef DIAGNOSTIC
1.34      thorpej   649:        if (__predict_false(pp->pr_nout > pp->pr_hardlimit)) {
1.25      thorpej   650:                pr_leave(pp);
1.21      thorpej   651:                simple_unlock(&pp->pr_slock);
1.20      thorpej   652:                panic("pool_get: %s: crossed hard limit", pp->pr_wchan);
                    653:        }
                    654: #endif
1.34      thorpej   655:        if (__predict_false(pp->pr_nout == pp->pr_hardlimit)) {
1.29      sommerfe  656:                if ((flags & PR_WAITOK) && !(flags & PR_LIMITFAIL)) {
1.20      thorpej   657:                        /*
                    658:                         * XXX: A warning isn't logged in this case.  Should
                    659:                         * it be?
                    660:                         */
                    661:                        pp->pr_flags |= PR_WANTED;
1.25      thorpej   662:                        pr_leave(pp);
1.40      sommerfe  663:                        ltsleep(pp, PSWP, pp->pr_wchan, 0, &pp->pr_slock);
1.25      thorpej   664:                        pr_enter(pp, file, line);
1.20      thorpej   665:                        goto startover;
                    666:                }
1.31      thorpej   667:
                    668:                /*
                    669:                 * Log a message that the hard limit has been hit.
                    670:                 */
                    671:                if (pp->pr_hardlimit_warning != NULL &&
                    672:                    ratecheck(&pp->pr_hardlimit_warning_last,
                    673:                              &pp->pr_hardlimit_ratecap))
                    674:                        log(LOG_ERR, "%s\n", pp->pr_hardlimit_warning);
1.21      thorpej   675:
                    676:                if (flags & PR_URGENT)
                    677:                        panic("pool_get: urgent");
                    678:
                    679:                pp->pr_nfail++;
                    680:
1.25      thorpej   681:                pr_leave(pp);
1.21      thorpej   682:                simple_unlock(&pp->pr_slock);
1.20      thorpej   683:                return (NULL);
                    684:        }
                    685:
1.3       pk        686:        /*
                    687:         * The convention we use is that if `curpage' is not NULL, then
                    688:         * it points at a non-empty bucket. In particular, `curpage'
                    689:         * never points at a page header which has PR_PHINPAGE set and
                    690:         * has no items in its bucket.
                    691:         */
1.20      thorpej   692:        if ((ph = pp->pr_curpage) == NULL) {
                    693: #ifdef DIAGNOSTIC
                    694:                if (pp->pr_nitems != 0) {
1.21      thorpej   695:                        simple_unlock(&pp->pr_slock);
1.20      thorpej   696:                        printf("pool_get: %s: curpage NULL, nitems %u\n",
                    697:                            pp->pr_wchan, pp->pr_nitems);
                    698:                        panic("pool_get: nitems inconsistent\n");
                    699:                }
                    700: #endif
                    701:
1.21      thorpej   702:                /*
                    703:                 * Call the back-end page allocator for more memory.
                    704:                 * Release the pool lock, as the back-end page allocator
                    705:                 * may block.
                    706:                 */
1.25      thorpej   707:                pr_leave(pp);
1.21      thorpej   708:                simple_unlock(&pp->pr_slock);
                    709:                v = (*pp->pr_alloc)(pp->pr_pagesz, flags, pp->pr_mtype);
1.55      thorpej   710:                if (__predict_true(v != NULL))
                    711:                        ph = pool_alloc_item_header(pp, v, flags);
1.21      thorpej   712:                simple_lock(&pp->pr_slock);
1.25      thorpej   713:                pr_enter(pp, file, line);
1.15      pk        714:
1.55      thorpej   715:                if (__predict_false(v == NULL || ph == NULL)) {
                    716:                        if (v != NULL)
                    717:                                (*pp->pr_free)(v, pp->pr_pagesz, pp->pr_mtype);
                    718:
1.21      thorpej   719:                        /*
1.55      thorpej   720:                         * We were unable to allocate a page or item
                    721:                         * header, but we released the lock during
                    722:                         * allocation, so perhaps items were freed
                    723:                         * back to the pool.  Check for this case.
1.21      thorpej   724:                         */
                    725:                        if (pp->pr_curpage != NULL)
                    726:                                goto startover;
1.15      pk        727:
1.3       pk        728:                        if (flags & PR_URGENT)
                    729:                                panic("pool_get: urgent");
1.21      thorpej   730:
1.3       pk        731:                        if ((flags & PR_WAITOK) == 0) {
                    732:                                pp->pr_nfail++;
1.25      thorpej   733:                                pr_leave(pp);
1.21      thorpej   734:                                simple_unlock(&pp->pr_slock);
1.1       pk        735:                                return (NULL);
1.3       pk        736:                        }
                    737:
1.15      pk        738:                        /*
                    739:                         * Wait for items to be returned to this pool.
1.21      thorpej   740:                         *
1.15      pk        741:                         * XXX: we actually want to wait just until
                    742:                         * the page allocator has memory again. Depending
                    743:                         * on this pool's usage, we might get stuck here
                    744:                         * for a long time.
1.20      thorpej   745:                         *
                    746:                         * XXX: maybe we should wake up once a second and
                    747:                         * try again?
1.15      pk        748:                         */
1.1       pk        749:                        pp->pr_flags |= PR_WANTED;
1.25      thorpej   750:                        pr_leave(pp);
1.40      sommerfe  751:                        ltsleep(pp, PSWP, pp->pr_wchan, 0, &pp->pr_slock);
1.25      thorpej   752:                        pr_enter(pp, file, line);
1.20      thorpej   753:                        goto startover;
1.1       pk        754:                }
1.3       pk        755:
1.15      pk        756:                /* We have more memory; add it to the pool */
1.55      thorpej   757:                pool_prime_page(pp, v, ph);
1.15      pk        758:                pp->pr_npagealloc++;
                    759:
1.20      thorpej   760:                /* Start the allocation process over. */
                    761:                goto startover;
1.3       pk        762:        }
                    763:
1.34      thorpej   764:        if (__predict_false((v = pi = TAILQ_FIRST(&ph->ph_itemlist)) == NULL)) {
1.25      thorpej   765:                pr_leave(pp);
1.21      thorpej   766:                simple_unlock(&pp->pr_slock);
1.3       pk        767:                panic("pool_get: %s: page empty", pp->pr_wchan);
1.21      thorpej   768:        }
1.20      thorpej   769: #ifdef DIAGNOSTIC
1.34      thorpej   770:        if (__predict_false(pp->pr_nitems == 0)) {
1.25      thorpej   771:                pr_leave(pp);
1.21      thorpej   772:                simple_unlock(&pp->pr_slock);
1.20      thorpej   773:                printf("pool_get: %s: items on itemlist, nitems %u\n",
                    774:                    pp->pr_wchan, pp->pr_nitems);
                    775:                panic("pool_get: nitems inconsistent\n");
                    776:        }
1.56      sommerfe  777:
1.3       pk        778:        pr_log(pp, v, PRLOG_GET, file, line);
                    779:
1.34      thorpej   780:        if (__predict_false(pi->pi_magic != PI_MAGIC)) {
1.25      thorpej   781:                pr_printlog(pp, pi, printf);
1.3       pk        782:                panic("pool_get(%s): free list modified: magic=%x; page %p;"
                    783:                       " item addr %p\n",
                    784:                        pp->pr_wchan, pi->pi_magic, ph->ph_page, pi);
                    785:        }
                    786: #endif
                    787:
                    788:        /*
                    789:         * Remove from item list.
                    790:         */
                    791:        TAILQ_REMOVE(&ph->ph_itemlist, pi, pi_list);
1.20      thorpej   792:        pp->pr_nitems--;
                    793:        pp->pr_nout++;
1.6       thorpej   794:        if (ph->ph_nmissing == 0) {
                    795: #ifdef DIAGNOSTIC
1.34      thorpej   796:                if (__predict_false(pp->pr_nidle == 0))
1.6       thorpej   797:                        panic("pool_get: nidle inconsistent");
                    798: #endif
                    799:                pp->pr_nidle--;
                    800:        }
1.3       pk        801:        ph->ph_nmissing++;
                    802:        if (TAILQ_FIRST(&ph->ph_itemlist) == NULL) {
1.21      thorpej   803: #ifdef DIAGNOSTIC
1.34      thorpej   804:                if (__predict_false(ph->ph_nmissing != pp->pr_itemsperpage)) {
1.25      thorpej   805:                        pr_leave(pp);
1.21      thorpej   806:                        simple_unlock(&pp->pr_slock);
                    807:                        panic("pool_get: %s: nmissing inconsistent",
                    808:                            pp->pr_wchan);
                    809:                }
                    810: #endif
1.3       pk        811:                /*
                    812:                 * Find a new non-empty page header, if any.
                    813:                 * Start search from the page head, to increase
                    814:                 * the chance for "high water" pages to be freed.
                    815:                 *
1.21      thorpej   816:                 * Migrate empty pages to the end of the list.  This
                    817:                 * will speed the update of curpage as pages become
                    818:                 * idle.  Empty pages intermingled with idle pages
                    819:                 * is no big deal.  As soon as a page becomes un-empty,
                    820:                 * it will move back to the head of the list.
1.3       pk        821:                 */
                    822:                TAILQ_REMOVE(&pp->pr_pagelist, ph, ph_pagelist);
1.21      thorpej   823:                TAILQ_INSERT_TAIL(&pp->pr_pagelist, ph, ph_pagelist);
1.61      chs       824:                TAILQ_FOREACH(ph, &pp->pr_pagelist, ph_pagelist)
1.3       pk        825:                        if (TAILQ_FIRST(&ph->ph_itemlist) != NULL)
                    826:                                break;
                    827:
                    828:                pp->pr_curpage = ph;
1.1       pk        829:        }
1.3       pk        830:
                    831:        pp->pr_nget++;
1.20      thorpej   832:
                    833:        /*
                    834:         * If we have a low water mark and we are now below that low
                    835:         * water mark, add more items to the pool.
                    836:         */
1.53      thorpej   837:        if (POOL_NEEDS_CATCHUP(pp) && pool_catchup(pp) != 0) {
1.20      thorpej   838:                /*
                    839:                 * XXX: Should we log a warning?  Should we set up a timeout
                    840:                 * to try again in a second or so?  The latter could break
                    841:                 * a caller's assumptions about interrupt protection, etc.
                    842:                 */
                    843:        }
                    844:
1.25      thorpej   845:        pr_leave(pp);
1.21      thorpej   846:        simple_unlock(&pp->pr_slock);
1.1       pk        847:        return (v);
                    848: }
                    849:
                    850: /*
1.43      thorpej   851:  * Internal version of pool_put().  Pool is already locked/entered.
1.1       pk        852:  */
1.43      thorpej   853: static void
1.56      sommerfe  854: pool_do_put(struct pool *pp, void *v)
1.1       pk        855: {
                    856:        struct pool_item *pi = v;
1.3       pk        857:        struct pool_item_header *ph;
                    858:        caddr_t page;
1.21      thorpej   859:        int s;
1.3       pk        860:
1.61      chs       861:        LOCK_ASSERT(simple_lock_held(&pp->pr_slock));
                    862:
1.3       pk        863:        page = (caddr_t)((u_long)v & pp->pr_pagemask);
1.1       pk        864:
1.30      thorpej   865: #ifdef DIAGNOSTIC
1.34      thorpej   866:        if (__predict_false(pp->pr_nout == 0)) {
1.30      thorpej   867:                printf("pool %s: putting with none out\n",
                    868:                    pp->pr_wchan);
                    869:                panic("pool_put");
                    870:        }
                    871: #endif
1.3       pk        872:
1.34      thorpej   873:        if (__predict_false((ph = pr_find_pagehead(pp, page)) == NULL)) {
1.25      thorpej   874:                pr_printlog(pp, NULL, printf);
1.3       pk        875:                panic("pool_put: %s: page header missing", pp->pr_wchan);
                    876:        }
1.28      thorpej   877:
                    878: #ifdef LOCKDEBUG
                    879:        /*
                    880:         * Check if we're freeing a locked simple lock.
                    881:         */
                    882:        simple_lock_freecheck((caddr_t)pi, ((caddr_t)pi) + pp->pr_size);
                    883: #endif
1.3       pk        884:
                    885:        /*
                    886:         * Return to item list.
                    887:         */
1.2       pk        888: #ifdef DIAGNOSTIC
1.3       pk        889:        pi->pi_magic = PI_MAGIC;
                    890: #endif
1.32      chs       891: #ifdef DEBUG
                    892:        {
                    893:                int i, *ip = v;
                    894:
                    895:                for (i = 0; i < pp->pr_size / sizeof(int); i++) {
                    896:                        *ip++ = PI_MAGIC;
                    897:                }
                    898:        }
                    899: #endif
                    900:
1.3       pk        901:        TAILQ_INSERT_HEAD(&ph->ph_itemlist, pi, pi_list);
                    902:        ph->ph_nmissing--;
                    903:        pp->pr_nput++;
1.20      thorpej   904:        pp->pr_nitems++;
                    905:        pp->pr_nout--;
1.3       pk        906:
                    907:        /* Cancel "pool empty" condition if it exists */
                    908:        if (pp->pr_curpage == NULL)
                    909:                pp->pr_curpage = ph;
                    910:
                    911:        if (pp->pr_flags & PR_WANTED) {
                    912:                pp->pr_flags &= ~PR_WANTED;
1.15      pk        913:                if (ph->ph_nmissing == 0)
                    914:                        pp->pr_nidle++;
1.3       pk        915:                wakeup((caddr_t)pp);
                    916:                return;
                    917:        }
                    918:
                    919:        /*
1.21      thorpej   920:         * If this page is now complete, do one of two things:
                    921:         *
                    922:         *      (1) If we have more pages than the page high water
                    923:         *          mark, free the page back to the system.
                    924:         *
                    925:         *      (2) Move it to the end of the page list, so that
                    926:         *          we minimize our chances of fragmenting the
                    927:         *          pool.  Idle pages migrate to the end (along with
                    928:         *          completely empty pages, so that we find un-empty
                    929:         *          pages more quickly when we update curpage) of the
                    930:         *          list so they can be more easily swept up by
                    931:         *          the pagedaemon when pages are scarce.
1.3       pk        932:         */
                    933:        if (ph->ph_nmissing == 0) {
1.6       thorpej   934:                pp->pr_nidle++;
1.3       pk        935:                if (pp->pr_npages > pp->pr_maxpages) {
1.61      chs       936:                        pr_rmpage(pp, ph, NULL);
1.3       pk        937:                } else {
                    938:                        TAILQ_REMOVE(&pp->pr_pagelist, ph, ph_pagelist);
                    939:                        TAILQ_INSERT_TAIL(&pp->pr_pagelist, ph, ph_pagelist);
                    940:
1.21      thorpej   941:                        /*
                    942:                         * Update the timestamp on the page.  A page must
                    943:                         * be idle for some period of time before it can
                    944:                         * be reclaimed by the pagedaemon.  This minimizes
                    945:                         * ping-pong'ing for memory.
                    946:                         */
                    947:                        s = splclock();
                    948:                        ph->ph_time = mono_time;
                    949:                        splx(s);
                    950:
                    951:                        /*
                    952:                         * Update the current page pointer.  Just look for
                    953:                         * the first page with any free items.
                    954:                         *
                    955:                         * XXX: Maybe we want an option to look for the
                    956:                         * page with the fewest available items, to minimize
                    957:                         * fragmentation?
                    958:                         */
1.61      chs       959:                        TAILQ_FOREACH(ph, &pp->pr_pagelist, ph_pagelist)
1.3       pk        960:                                if (TAILQ_FIRST(&ph->ph_itemlist) != NULL)
                    961:                                        break;
1.1       pk        962:
1.3       pk        963:                        pp->pr_curpage = ph;
1.1       pk        964:                }
                    965:        }
1.21      thorpej   966:        /*
                    967:         * If the page has just become un-empty, move it to the head of
                    968:         * the list, and make it the current page.  The next allocation
                    969:         * will get the item from this page, instead of further fragmenting
                    970:         * the pool.
                    971:         */
                    972:        else if (ph->ph_nmissing == (pp->pr_itemsperpage - 1)) {
                    973:                TAILQ_REMOVE(&pp->pr_pagelist, ph, ph_pagelist);
                    974:                TAILQ_INSERT_HEAD(&pp->pr_pagelist, ph, ph_pagelist);
                    975:                pp->pr_curpage = ph;
                    976:        }
1.43      thorpej   977: }
                    978:
                    979: /*
                    980:  * Return resource to the pool; must be called at appropriate spl level
                    981:  */
1.59      thorpej   982: #ifdef POOL_DIAGNOSTIC
1.43      thorpej   983: void
                    984: _pool_put(struct pool *pp, void *v, const char *file, long line)
                    985: {
                    986:
                    987:        simple_lock(&pp->pr_slock);
                    988:        pr_enter(pp, file, line);
                    989:
1.56      sommerfe  990:        pr_log(pp, v, PRLOG_PUT, file, line);
                    991:
                    992:        pool_do_put(pp, v);
1.21      thorpej   993:
1.25      thorpej   994:        pr_leave(pp);
1.21      thorpej   995:        simple_unlock(&pp->pr_slock);
1.1       pk        996: }
1.57      sommerfe  997: #undef pool_put
1.59      thorpej   998: #endif /* POOL_DIAGNOSTIC */
1.1       pk        999:
1.56      sommerfe 1000: void
                   1001: pool_put(struct pool *pp, void *v)
                   1002: {
                   1003:
                   1004:        simple_lock(&pp->pr_slock);
                   1005:
                   1006:        pool_do_put(pp, v);
                   1007:
                   1008:        simple_unlock(&pp->pr_slock);
                   1009: }
1.57      sommerfe 1010:
1.59      thorpej  1011: #ifdef POOL_DIAGNOSTIC
1.57      sommerfe 1012: #define                pool_put(h, v)  _pool_put((h), (v), __FILE__, __LINE__)
1.56      sommerfe 1013: #endif
                   1014:
1.1       pk       1015: /*
1.55      thorpej  1016:  * Add N items to the pool.
                   1017:  */
                   1018: int
                   1019: pool_prime(struct pool *pp, int n)
                   1020: {
                   1021:        struct pool_item_header *ph;
                   1022:        caddr_t cp;
                   1023:        int newpages, error = 0;
                   1024:
                   1025:        simple_lock(&pp->pr_slock);
                   1026:
                   1027:        newpages = roundup(n, pp->pr_itemsperpage) / pp->pr_itemsperpage;
                   1028:
                   1029:        while (newpages-- > 0) {
                   1030:                simple_unlock(&pp->pr_slock);
                   1031:                cp = (*pp->pr_alloc)(pp->pr_pagesz, PR_NOWAIT, pp->pr_mtype);
                   1032:                if (__predict_true(cp != NULL))
                   1033:                        ph = pool_alloc_item_header(pp, cp, PR_NOWAIT);
                   1034:                simple_lock(&pp->pr_slock);
                   1035:
                   1036:                if (__predict_false(cp == NULL || ph == NULL)) {
                   1037:                        error = ENOMEM;
                   1038:                        if (cp != NULL)
                   1039:                                (*pp->pr_free)(cp, pp->pr_pagesz, pp->pr_mtype);
                   1040:                        break;
                   1041:                }
                   1042:
                   1043:                pool_prime_page(pp, cp, ph);
                   1044:                pp->pr_npagealloc++;
                   1045:                pp->pr_minpages++;
                   1046:        }
                   1047:
                   1048:        if (pp->pr_minpages >= pp->pr_maxpages)
                   1049:                pp->pr_maxpages = pp->pr_minpages + 1;  /* XXX */
                   1050:
                   1051:        simple_unlock(&pp->pr_slock);
                   1052:        return (0);
                   1053: }
                   1054:
                   1055: /*
1.3       pk       1056:  * Add a page worth of items to the pool.
1.21      thorpej  1057:  *
                   1058:  * Note, we must be called with the pool descriptor LOCKED.
1.3       pk       1059:  */
1.55      thorpej  1060: static void
                   1061: pool_prime_page(struct pool *pp, caddr_t storage, struct pool_item_header *ph)
1.3       pk       1062: {
                   1063:        struct pool_item *pi;
                   1064:        caddr_t cp = storage;
                   1065:        unsigned int align = pp->pr_align;
                   1066:        unsigned int ioff = pp->pr_itemoffset;
1.55      thorpej  1067:        int n;
1.36      pk       1068:
                   1069:        if (((u_long)cp & (pp->pr_pagesz - 1)) != 0)
                   1070:                panic("pool_prime_page: %s: unaligned page", pp->pr_wchan);
1.3       pk       1071:
1.55      thorpej  1072:        if ((pp->pr_roflags & PR_PHINPAGE) == 0)
1.3       pk       1073:                LIST_INSERT_HEAD(&pp->pr_hashtab[PR_HASH_INDEX(pp, cp)],
1.55      thorpej  1074:                    ph, ph_hashlist);
1.3       pk       1075:
                   1076:        /*
                   1077:         * Insert page header.
                   1078:         */
                   1079:        TAILQ_INSERT_HEAD(&pp->pr_pagelist, ph, ph_pagelist);
                   1080:        TAILQ_INIT(&ph->ph_itemlist);
                   1081:        ph->ph_page = storage;
                   1082:        ph->ph_nmissing = 0;
1.21      thorpej  1083:        memset(&ph->ph_time, 0, sizeof(ph->ph_time));
1.3       pk       1084:
1.6       thorpej  1085:        pp->pr_nidle++;
                   1086:
1.3       pk       1087:        /*
                   1088:         * Color this page.
                   1089:         */
                   1090:        cp = (caddr_t)(cp + pp->pr_curcolor);
                   1091:        if ((pp->pr_curcolor += align) > pp->pr_maxcolor)
                   1092:                pp->pr_curcolor = 0;
                   1093:
                   1094:        /*
                   1095:         * Adjust storage to apply aligment to `pr_itemoffset' in each item.
                   1096:         */
                   1097:        if (ioff != 0)
                   1098:                cp = (caddr_t)(cp + (align - ioff));
                   1099:
                   1100:        /*
                   1101:         * Insert remaining chunks on the bucket list.
                   1102:         */
                   1103:        n = pp->pr_itemsperpage;
1.20      thorpej  1104:        pp->pr_nitems += n;
1.3       pk       1105:
                   1106:        while (n--) {
                   1107:                pi = (struct pool_item *)cp;
                   1108:
                   1109:                /* Insert on page list */
                   1110:                TAILQ_INSERT_TAIL(&ph->ph_itemlist, pi, pi_list);
                   1111: #ifdef DIAGNOSTIC
                   1112:                pi->pi_magic = PI_MAGIC;
                   1113: #endif
                   1114:                cp = (caddr_t)(cp + pp->pr_size);
                   1115:        }
                   1116:
                   1117:        /*
                   1118:         * If the pool was depleted, point at the new page.
                   1119:         */
                   1120:        if (pp->pr_curpage == NULL)
                   1121:                pp->pr_curpage = ph;
                   1122:
                   1123:        if (++pp->pr_npages > pp->pr_hiwat)
                   1124:                pp->pr_hiwat = pp->pr_npages;
                   1125: }
                   1126:
1.20      thorpej  1127: /*
1.52      thorpej  1128:  * Used by pool_get() when nitems drops below the low water mark.  This
                   1129:  * is used to catch up nitmes with the low water mark.
1.20      thorpej  1130:  *
1.21      thorpej  1131:  * Note 1, we never wait for memory here, we let the caller decide what to do.
1.20      thorpej  1132:  *
                   1133:  * Note 2, this doesn't work with static pools.
                   1134:  *
                   1135:  * Note 3, we must be called with the pool already locked, and we return
                   1136:  * with it locked.
                   1137:  */
                   1138: static int
1.42      thorpej  1139: pool_catchup(struct pool *pp)
1.20      thorpej  1140: {
1.55      thorpej  1141:        struct pool_item_header *ph;
1.20      thorpej  1142:        caddr_t cp;
                   1143:        int error = 0;
                   1144:
                   1145:        if (pp->pr_roflags & PR_STATIC) {
                   1146:                /*
                   1147:                 * We dropped below the low water mark, and this is not a
                   1148:                 * good thing.  Log a warning.
1.21      thorpej  1149:                 *
                   1150:                 * XXX: rate-limit this?
1.20      thorpej  1151:                 */
                   1152:                printf("WARNING: static pool `%s' dropped below low water "
                   1153:                    "mark\n", pp->pr_wchan);
                   1154:                return (0);
                   1155:        }
                   1156:
1.54      thorpej  1157:        while (POOL_NEEDS_CATCHUP(pp)) {
1.20      thorpej  1158:                /*
1.21      thorpej  1159:                 * Call the page back-end allocator for more memory.
                   1160:                 *
                   1161:                 * XXX: We never wait, so should we bother unlocking
                   1162:                 * the pool descriptor?
1.20      thorpej  1163:                 */
1.21      thorpej  1164:                simple_unlock(&pp->pr_slock);
1.55      thorpej  1165:                cp = (*pp->pr_alloc)(pp->pr_pagesz, PR_NOWAIT, pp->pr_mtype);
                   1166:                if (__predict_true(cp != NULL))
                   1167:                        ph = pool_alloc_item_header(pp, cp, PR_NOWAIT);
1.21      thorpej  1168:                simple_lock(&pp->pr_slock);
1.55      thorpej  1169:                if (__predict_false(cp == NULL || ph == NULL)) {
                   1170:                        if (cp != NULL)
                   1171:                                (*pp->pr_free)(cp, pp->pr_pagesz, pp->pr_mtype);
1.20      thorpej  1172:                        error = ENOMEM;
                   1173:                        break;
                   1174:                }
1.55      thorpej  1175:                pool_prime_page(pp, cp, ph);
1.26      thorpej  1176:                pp->pr_npagealloc++;
1.20      thorpej  1177:        }
                   1178:
                   1179:        return (error);
                   1180: }
                   1181:
1.3       pk       1182: void
1.42      thorpej  1183: pool_setlowat(struct pool *pp, int n)
1.3       pk       1184: {
1.20      thorpej  1185:        int error;
1.15      pk       1186:
1.21      thorpej  1187:        simple_lock(&pp->pr_slock);
                   1188:
1.3       pk       1189:        pp->pr_minitems = n;
1.15      pk       1190:        pp->pr_minpages = (n == 0)
                   1191:                ? 0
1.18      thorpej  1192:                : roundup(n, pp->pr_itemsperpage) / pp->pr_itemsperpage;
1.20      thorpej  1193:
                   1194:        /* Make sure we're caught up with the newly-set low water mark. */
1.53      thorpej  1195:        if (POOL_NEEDS_CATCHUP(pp) && (error = pool_catchup(pp) != 0)) {
1.20      thorpej  1196:                /*
                   1197:                 * XXX: Should we log a warning?  Should we set up a timeout
                   1198:                 * to try again in a second or so?  The latter could break
                   1199:                 * a caller's assumptions about interrupt protection, etc.
                   1200:                 */
                   1201:        }
1.21      thorpej  1202:
                   1203:        simple_unlock(&pp->pr_slock);
1.3       pk       1204: }
                   1205:
                   1206: void
1.42      thorpej  1207: pool_sethiwat(struct pool *pp, int n)
1.3       pk       1208: {
1.15      pk       1209:
1.21      thorpej  1210:        simple_lock(&pp->pr_slock);
                   1211:
1.15      pk       1212:        pp->pr_maxpages = (n == 0)
                   1213:                ? 0
1.18      thorpej  1214:                : roundup(n, pp->pr_itemsperpage) / pp->pr_itemsperpage;
1.21      thorpej  1215:
                   1216:        simple_unlock(&pp->pr_slock);
1.3       pk       1217: }
                   1218:
1.20      thorpej  1219: void
1.42      thorpej  1220: pool_sethardlimit(struct pool *pp, int n, const char *warnmess, int ratecap)
1.20      thorpej  1221: {
                   1222:
1.21      thorpej  1223:        simple_lock(&pp->pr_slock);
1.20      thorpej  1224:
                   1225:        pp->pr_hardlimit = n;
                   1226:        pp->pr_hardlimit_warning = warnmess;
1.31      thorpej  1227:        pp->pr_hardlimit_ratecap.tv_sec = ratecap;
                   1228:        pp->pr_hardlimit_warning_last.tv_sec = 0;
                   1229:        pp->pr_hardlimit_warning_last.tv_usec = 0;
1.20      thorpej  1230:
                   1231:        /*
1.21      thorpej  1232:         * In-line version of pool_sethiwat(), because we don't want to
                   1233:         * release the lock.
1.20      thorpej  1234:         */
                   1235:        pp->pr_maxpages = (n == 0)
                   1236:                ? 0
                   1237:                : roundup(n, pp->pr_itemsperpage) / pp->pr_itemsperpage;
1.21      thorpej  1238:
                   1239:        simple_unlock(&pp->pr_slock);
1.20      thorpej  1240: }
1.3       pk       1241:
                   1242: /*
                   1243:  * Default page allocator.
                   1244:  */
                   1245: static void *
1.42      thorpej  1246: pool_page_alloc(unsigned long sz, int flags, int mtype)
1.3       pk       1247: {
1.11      thorpej  1248:        boolean_t waitok = (flags & PR_WAITOK) ? TRUE : FALSE;
1.3       pk       1249:
1.11      thorpej  1250:        return ((void *)uvm_km_alloc_poolpage(waitok));
1.3       pk       1251: }
                   1252:
                   1253: static void
1.42      thorpej  1254: pool_page_free(void *v, unsigned long sz, int mtype)
1.3       pk       1255: {
                   1256:
1.10      eeh      1257:        uvm_km_free_poolpage((vaddr_t)v);
1.3       pk       1258: }
1.12      thorpej  1259:
1.62      bjh21    1260: #ifdef POOL_SUBPAGE
                   1261: /*
                   1262:  * Sub-page allocator, for machines with large hardware pages.
                   1263:  */
                   1264: static void *
                   1265: pool_subpage_alloc(unsigned long sz, int flags, int mtype)
                   1266: {
                   1267:
                   1268:        return pool_get(&psppool, flags);
                   1269: }
                   1270:
                   1271: static void
                   1272: pool_subpage_free(void *v, unsigned long sz, int mtype)
                   1273: {
                   1274:
                   1275:        pool_put(&psppool, v);
                   1276: }
                   1277: #endif
                   1278:
                   1279: #ifdef POOL_SUBPAGE
                   1280: /* We don't provide a real nointr allocator.  Maybe later. */
                   1281: void *
                   1282: pool_page_alloc_nointr(unsigned long sz, int flags, int mtype)
                   1283: {
                   1284:
                   1285:        return pool_subpage_alloc(sz, flags, mtype);
                   1286: }
                   1287:
                   1288: void
                   1289: pool_page_free_nointr(void *v, unsigned long sz, int mtype)
                   1290: {
                   1291:
                   1292:        pool_subpage_free(v, sz, mtype);
                   1293: }
                   1294: #else
1.12      thorpej  1295: /*
                   1296:  * Alternate pool page allocator for pools that know they will
                   1297:  * never be accessed in interrupt context.
                   1298:  */
                   1299: void *
1.42      thorpej  1300: pool_page_alloc_nointr(unsigned long sz, int flags, int mtype)
1.12      thorpej  1301: {
                   1302:        boolean_t waitok = (flags & PR_WAITOK) ? TRUE : FALSE;
                   1303:
                   1304:        return ((void *)uvm_km_alloc_poolpage1(kernel_map, uvm.kernel_object,
                   1305:            waitok));
                   1306: }
                   1307:
                   1308: void
1.42      thorpej  1309: pool_page_free_nointr(void *v, unsigned long sz, int mtype)
1.12      thorpej  1310: {
                   1311:
                   1312:        uvm_km_free_poolpage1(kernel_map, (vaddr_t)v);
                   1313: }
1.62      bjh21    1314: #endif
1.12      thorpej  1315:
1.3       pk       1316:
                   1317: /*
                   1318:  * Release all complete pages that have not been used recently.
                   1319:  */
                   1320: void
1.59      thorpej  1321: #ifdef POOL_DIAGNOSTIC
1.42      thorpej  1322: _pool_reclaim(struct pool *pp, const char *file, long line)
1.56      sommerfe 1323: #else
                   1324: pool_reclaim(struct pool *pp)
                   1325: #endif
1.3       pk       1326: {
                   1327:        struct pool_item_header *ph, *phnext;
1.43      thorpej  1328:        struct pool_cache *pc;
1.21      thorpej  1329:        struct timeval curtime;
1.61      chs      1330:        struct pool_pagelist pq;
1.21      thorpej  1331:        int s;
1.3       pk       1332:
1.20      thorpej  1333:        if (pp->pr_roflags & PR_STATIC)
1.3       pk       1334:                return;
                   1335:
1.21      thorpej  1336:        if (simple_lock_try(&pp->pr_slock) == 0)
1.3       pk       1337:                return;
1.25      thorpej  1338:        pr_enter(pp, file, line);
1.61      chs      1339:        TAILQ_INIT(&pq);
1.3       pk       1340:
1.43      thorpej  1341:        /*
                   1342:         * Reclaim items from the pool's caches.
                   1343:         */
1.61      chs      1344:        TAILQ_FOREACH(pc, &pp->pr_cachelist, pc_poollist)
1.43      thorpej  1345:                pool_cache_reclaim(pc);
                   1346:
1.21      thorpej  1347:        s = splclock();
                   1348:        curtime = mono_time;
                   1349:        splx(s);
                   1350:
1.3       pk       1351:        for (ph = TAILQ_FIRST(&pp->pr_pagelist); ph != NULL; ph = phnext) {
                   1352:                phnext = TAILQ_NEXT(ph, ph_pagelist);
                   1353:
                   1354:                /* Check our minimum page claim */
                   1355:                if (pp->pr_npages <= pp->pr_minpages)
                   1356:                        break;
                   1357:
                   1358:                if (ph->ph_nmissing == 0) {
                   1359:                        struct timeval diff;
                   1360:                        timersub(&curtime, &ph->ph_time, &diff);
                   1361:                        if (diff.tv_sec < pool_inactive_time)
                   1362:                                continue;
1.21      thorpej  1363:
                   1364:                        /*
                   1365:                         * If freeing this page would put us below
                   1366:                         * the low water mark, stop now.
                   1367:                         */
                   1368:                        if ((pp->pr_nitems - pp->pr_itemsperpage) <
                   1369:                            pp->pr_minitems)
                   1370:                                break;
                   1371:
1.61      chs      1372:                        pr_rmpage(pp, ph, &pq);
1.3       pk       1373:                }
                   1374:        }
                   1375:
1.25      thorpej  1376:        pr_leave(pp);
1.21      thorpej  1377:        simple_unlock(&pp->pr_slock);
1.61      chs      1378:        if (TAILQ_EMPTY(&pq)) {
                   1379:                return;
                   1380:        }
                   1381:        while ((ph = TAILQ_FIRST(&pq)) != NULL) {
                   1382:                TAILQ_REMOVE(&pq, ph, ph_pagelist);
                   1383:                (*pp->pr_free)(ph->ph_page, pp->pr_pagesz, pp->pr_mtype);
                   1384:                if (pp->pr_roflags & PR_PHINPAGE) {
                   1385:                        continue;
                   1386:                }
                   1387:                LIST_REMOVE(ph, ph_hashlist);
                   1388:                s = splhigh();
                   1389:                pool_put(&phpool, ph);
                   1390:                splx(s);
                   1391:        }
1.3       pk       1392: }
                   1393:
                   1394:
                   1395: /*
                   1396:  * Drain pools, one at a time.
1.21      thorpej  1397:  *
                   1398:  * Note, we must never be called from an interrupt context.
1.3       pk       1399:  */
                   1400: void
1.42      thorpej  1401: pool_drain(void *arg)
1.3       pk       1402: {
                   1403:        struct pool *pp;
1.23      thorpej  1404:        int s;
1.3       pk       1405:
1.61      chs      1406:        pp = NULL;
1.49      thorpej  1407:        s = splvm();
1.23      thorpej  1408:        simple_lock(&pool_head_slock);
1.61      chs      1409:        if (drainpp == NULL) {
                   1410:                drainpp = TAILQ_FIRST(&pool_head);
                   1411:        }
                   1412:        if (drainpp) {
                   1413:                pp = drainpp;
                   1414:                drainpp = TAILQ_NEXT(pp, pr_poollist);
                   1415:        }
                   1416:        simple_unlock(&pool_head_slock);
1.63    ! chs      1417:        pool_reclaim(pp);
1.61      chs      1418:        splx(s);
1.3       pk       1419: }
                   1420:
                   1421:
                   1422: /*
                   1423:  * Diagnostic helpers.
                   1424:  */
                   1425: void
1.42      thorpej  1426: pool_print(struct pool *pp, const char *modif)
1.21      thorpej  1427: {
                   1428:        int s;
                   1429:
1.49      thorpej  1430:        s = splvm();
1.25      thorpej  1431:        if (simple_lock_try(&pp->pr_slock) == 0) {
                   1432:                printf("pool %s is locked; try again later\n",
                   1433:                    pp->pr_wchan);
                   1434:                splx(s);
                   1435:                return;
                   1436:        }
                   1437:        pool_print1(pp, modif, printf);
1.21      thorpej  1438:        simple_unlock(&pp->pr_slock);
                   1439:        splx(s);
                   1440: }
                   1441:
1.25      thorpej  1442: void
1.42      thorpej  1443: pool_printit(struct pool *pp, const char *modif, void (*pr)(const char *, ...))
1.25      thorpej  1444: {
                   1445:        int didlock = 0;
                   1446:
                   1447:        if (pp == NULL) {
                   1448:                (*pr)("Must specify a pool to print.\n");
                   1449:                return;
                   1450:        }
                   1451:
                   1452:        /*
                   1453:         * Called from DDB; interrupts should be blocked, and all
                   1454:         * other processors should be paused.  We can skip locking
                   1455:         * the pool in this case.
                   1456:         *
                   1457:         * We do a simple_lock_try() just to print the lock
                   1458:         * status, however.
                   1459:         */
                   1460:
                   1461:        if (simple_lock_try(&pp->pr_slock) == 0)
                   1462:                (*pr)("WARNING: pool %s is locked\n", pp->pr_wchan);
                   1463:        else
                   1464:                didlock = 1;
                   1465:
                   1466:        pool_print1(pp, modif, pr);
                   1467:
                   1468:        if (didlock)
                   1469:                simple_unlock(&pp->pr_slock);
                   1470: }
                   1471:
1.21      thorpej  1472: static void
1.42      thorpej  1473: pool_print1(struct pool *pp, const char *modif, void (*pr)(const char *, ...))
1.3       pk       1474: {
1.25      thorpej  1475:        struct pool_item_header *ph;
1.44      thorpej  1476:        struct pool_cache *pc;
                   1477:        struct pool_cache_group *pcg;
1.25      thorpej  1478: #ifdef DIAGNOSTIC
                   1479:        struct pool_item *pi;
                   1480: #endif
1.44      thorpej  1481:        int i, print_log = 0, print_pagelist = 0, print_cache = 0;
1.25      thorpej  1482:        char c;
                   1483:
                   1484:        while ((c = *modif++) != '\0') {
                   1485:                if (c == 'l')
                   1486:                        print_log = 1;
                   1487:                if (c == 'p')
                   1488:                        print_pagelist = 1;
1.44      thorpej  1489:                if (c == 'c')
                   1490:                        print_cache = 1;
1.25      thorpej  1491:                modif++;
                   1492:        }
                   1493:
                   1494:        (*pr)("POOL %s: size %u, align %u, ioff %u, roflags 0x%08x\n",
                   1495:            pp->pr_wchan, pp->pr_size, pp->pr_align, pp->pr_itemoffset,
                   1496:            pp->pr_roflags);
                   1497:        (*pr)("\tpagesz %u, mtype %d\n", pp->pr_pagesz, pp->pr_mtype);
                   1498:        (*pr)("\talloc %p, release %p\n", pp->pr_alloc, pp->pr_free);
                   1499:        (*pr)("\tminitems %u, minpages %u, maxpages %u, npages %u\n",
                   1500:            pp->pr_minitems, pp->pr_minpages, pp->pr_maxpages, pp->pr_npages);
                   1501:        (*pr)("\titemsperpage %u, nitems %u, nout %u, hardlimit %u\n",
                   1502:            pp->pr_itemsperpage, pp->pr_nitems, pp->pr_nout, pp->pr_hardlimit);
                   1503:
                   1504:        (*pr)("\n\tnget %lu, nfail %lu, nput %lu\n",
                   1505:            pp->pr_nget, pp->pr_nfail, pp->pr_nput);
                   1506:        (*pr)("\tnpagealloc %lu, npagefree %lu, hiwat %u, nidle %lu\n",
                   1507:            pp->pr_npagealloc, pp->pr_npagefree, pp->pr_hiwat, pp->pr_nidle);
                   1508:
                   1509:        if (print_pagelist == 0)
                   1510:                goto skip_pagelist;
                   1511:
                   1512:        if ((ph = TAILQ_FIRST(&pp->pr_pagelist)) != NULL)
                   1513:                (*pr)("\n\tpage list:\n");
                   1514:        for (; ph != NULL; ph = TAILQ_NEXT(ph, ph_pagelist)) {
                   1515:                (*pr)("\t\tpage %p, nmissing %d, time %lu,%lu\n",
                   1516:                    ph->ph_page, ph->ph_nmissing,
                   1517:                    (u_long)ph->ph_time.tv_sec,
                   1518:                    (u_long)ph->ph_time.tv_usec);
                   1519: #ifdef DIAGNOSTIC
1.61      chs      1520:                TAILQ_FOREACH(pi, &ph->ph_itemlist, pi_list) {
1.25      thorpej  1521:                        if (pi->pi_magic != PI_MAGIC) {
                   1522:                                (*pr)("\t\t\titem %p, magic 0x%x\n",
                   1523:                                    pi, pi->pi_magic);
                   1524:                        }
                   1525:                }
                   1526: #endif
                   1527:        }
                   1528:        if (pp->pr_curpage == NULL)
                   1529:                (*pr)("\tno current page\n");
                   1530:        else
                   1531:                (*pr)("\tcurpage %p\n", pp->pr_curpage->ph_page);
                   1532:
                   1533:  skip_pagelist:
                   1534:
                   1535:        if (print_log == 0)
                   1536:                goto skip_log;
                   1537:
                   1538:        (*pr)("\n");
                   1539:        if ((pp->pr_roflags & PR_LOGGING) == 0)
                   1540:                (*pr)("\tno log\n");
                   1541:        else
                   1542:                pr_printlog(pp, NULL, pr);
1.3       pk       1543:
1.25      thorpej  1544:  skip_log:
1.44      thorpej  1545:
                   1546:        if (print_cache == 0)
                   1547:                goto skip_cache;
                   1548:
1.61      chs      1549:        TAILQ_FOREACH(pc, &pp->pr_cachelist, pc_poollist) {
1.44      thorpej  1550:                (*pr)("\tcache %p: allocfrom %p freeto %p\n", pc,
                   1551:                    pc->pc_allocfrom, pc->pc_freeto);
1.48      thorpej  1552:                (*pr)("\t    hits %lu misses %lu ngroups %lu nitems %lu\n",
                   1553:                    pc->pc_hits, pc->pc_misses, pc->pc_ngroups, pc->pc_nitems);
1.61      chs      1554:                TAILQ_FOREACH(pcg, &pc->pc_grouplist, pcg_list) {
1.44      thorpej  1555:                        (*pr)("\t\tgroup %p: avail %d\n", pcg, pcg->pcg_avail);
                   1556:                        for (i = 0; i < PCG_NOBJECTS; i++)
                   1557:                                (*pr)("\t\t\t%p\n", pcg->pcg_objects[i]);
                   1558:                }
                   1559:        }
                   1560:
                   1561:  skip_cache:
1.3       pk       1562:
1.25      thorpej  1563:        pr_enter_check(pp, pr);
1.3       pk       1564: }
                   1565:
                   1566: int
1.42      thorpej  1567: pool_chk(struct pool *pp, const char *label)
1.3       pk       1568: {
                   1569:        struct pool_item_header *ph;
                   1570:        int r = 0;
                   1571:
1.21      thorpej  1572:        simple_lock(&pp->pr_slock);
1.3       pk       1573:
1.61      chs      1574:        TAILQ_FOREACH(ph, &pp->pr_pagelist, ph_pagelist) {
1.3       pk       1575:                struct pool_item *pi;
                   1576:                int n;
                   1577:                caddr_t page;
                   1578:
                   1579:                page = (caddr_t)((u_long)ph & pp->pr_pagemask);
1.20      thorpej  1580:                if (page != ph->ph_page &&
                   1581:                    (pp->pr_roflags & PR_PHINPAGE) != 0) {
1.3       pk       1582:                        if (label != NULL)
                   1583:                                printf("%s: ", label);
1.16      briggs   1584:                        printf("pool(%p:%s): page inconsistency: page %p;"
                   1585:                               " at page head addr %p (p %p)\n", pp,
1.3       pk       1586:                                pp->pr_wchan, ph->ph_page,
                   1587:                                ph, page);
                   1588:                        r++;
                   1589:                        goto out;
                   1590:                }
                   1591:
                   1592:                for (pi = TAILQ_FIRST(&ph->ph_itemlist), n = 0;
                   1593:                     pi != NULL;
                   1594:                     pi = TAILQ_NEXT(pi,pi_list), n++) {
                   1595:
                   1596: #ifdef DIAGNOSTIC
                   1597:                        if (pi->pi_magic != PI_MAGIC) {
                   1598:                                if (label != NULL)
                   1599:                                        printf("%s: ", label);
                   1600:                                printf("pool(%s): free list modified: magic=%x;"
                   1601:                                       " page %p; item ordinal %d;"
                   1602:                                       " addr %p (p %p)\n",
                   1603:                                        pp->pr_wchan, pi->pi_magic, ph->ph_page,
                   1604:                                        n, pi, page);
                   1605:                                panic("pool");
                   1606:                        }
                   1607: #endif
                   1608:                        page = (caddr_t)((u_long)pi & pp->pr_pagemask);
                   1609:                        if (page == ph->ph_page)
                   1610:                                continue;
                   1611:
                   1612:                        if (label != NULL)
                   1613:                                printf("%s: ", label);
1.16      briggs   1614:                        printf("pool(%p:%s): page inconsistency: page %p;"
                   1615:                               " item ordinal %d; addr %p (p %p)\n", pp,
1.3       pk       1616:                                pp->pr_wchan, ph->ph_page,
                   1617:                                n, pi, page);
                   1618:                        r++;
                   1619:                        goto out;
                   1620:                }
                   1621:        }
                   1622: out:
1.21      thorpej  1623:        simple_unlock(&pp->pr_slock);
1.3       pk       1624:        return (r);
1.43      thorpej  1625: }
                   1626:
                   1627: /*
                   1628:  * pool_cache_init:
                   1629:  *
                   1630:  *     Initialize a pool cache.
                   1631:  *
                   1632:  *     NOTE: If the pool must be protected from interrupts, we expect
                   1633:  *     to be called at the appropriate interrupt priority level.
                   1634:  */
                   1635: void
                   1636: pool_cache_init(struct pool_cache *pc, struct pool *pp,
                   1637:     int (*ctor)(void *, void *, int),
                   1638:     void (*dtor)(void *, void *),
                   1639:     void *arg)
                   1640: {
                   1641:
                   1642:        TAILQ_INIT(&pc->pc_grouplist);
                   1643:        simple_lock_init(&pc->pc_slock);
                   1644:
                   1645:        pc->pc_allocfrom = NULL;
                   1646:        pc->pc_freeto = NULL;
                   1647:        pc->pc_pool = pp;
                   1648:
                   1649:        pc->pc_ctor = ctor;
                   1650:        pc->pc_dtor = dtor;
                   1651:        pc->pc_arg  = arg;
                   1652:
1.48      thorpej  1653:        pc->pc_hits   = 0;
                   1654:        pc->pc_misses = 0;
                   1655:
                   1656:        pc->pc_ngroups = 0;
                   1657:
                   1658:        pc->pc_nitems = 0;
                   1659:
1.43      thorpej  1660:        simple_lock(&pp->pr_slock);
                   1661:        TAILQ_INSERT_TAIL(&pp->pr_cachelist, pc, pc_poollist);
                   1662:        simple_unlock(&pp->pr_slock);
                   1663: }
                   1664:
                   1665: /*
                   1666:  * pool_cache_destroy:
                   1667:  *
                   1668:  *     Destroy a pool cache.
                   1669:  */
                   1670: void
                   1671: pool_cache_destroy(struct pool_cache *pc)
                   1672: {
                   1673:        struct pool *pp = pc->pc_pool;
                   1674:
                   1675:        /* First, invalidate the entire cache. */
                   1676:        pool_cache_invalidate(pc);
                   1677:
                   1678:        /* ...and remove it from the pool's cache list. */
                   1679:        simple_lock(&pp->pr_slock);
                   1680:        TAILQ_REMOVE(&pp->pr_cachelist, pc, pc_poollist);
                   1681:        simple_unlock(&pp->pr_slock);
                   1682: }
                   1683:
                   1684: static __inline void *
                   1685: pcg_get(struct pool_cache_group *pcg)
                   1686: {
                   1687:        void *object;
                   1688:        u_int idx;
                   1689:
                   1690:        KASSERT(pcg->pcg_avail <= PCG_NOBJECTS);
1.45      thorpej  1691:        KASSERT(pcg->pcg_avail != 0);
1.43      thorpej  1692:        idx = --pcg->pcg_avail;
                   1693:
                   1694:        KASSERT(pcg->pcg_objects[idx] != NULL);
                   1695:        object = pcg->pcg_objects[idx];
                   1696:        pcg->pcg_objects[idx] = NULL;
                   1697:
                   1698:        return (object);
                   1699: }
                   1700:
                   1701: static __inline void
                   1702: pcg_put(struct pool_cache_group *pcg, void *object)
                   1703: {
                   1704:        u_int idx;
                   1705:
                   1706:        KASSERT(pcg->pcg_avail < PCG_NOBJECTS);
                   1707:        idx = pcg->pcg_avail++;
                   1708:
                   1709:        KASSERT(pcg->pcg_objects[idx] == NULL);
                   1710:        pcg->pcg_objects[idx] = object;
                   1711: }
                   1712:
                   1713: /*
                   1714:  * pool_cache_get:
                   1715:  *
                   1716:  *     Get an object from a pool cache.
                   1717:  */
                   1718: void *
                   1719: pool_cache_get(struct pool_cache *pc, int flags)
                   1720: {
                   1721:        struct pool_cache_group *pcg;
                   1722:        void *object;
1.58      thorpej  1723:
                   1724: #ifdef LOCKDEBUG
                   1725:        if (flags & PR_WAITOK)
                   1726:                simple_lock_only_held(NULL, "pool_cache_get(PR_WAITOK)");
                   1727: #endif
1.43      thorpej  1728:
                   1729:        simple_lock(&pc->pc_slock);
                   1730:
                   1731:        if ((pcg = pc->pc_allocfrom) == NULL) {
1.61      chs      1732:                TAILQ_FOREACH(pcg, &pc->pc_grouplist, pcg_list) {
1.43      thorpej  1733:                        if (pcg->pcg_avail != 0) {
                   1734:                                pc->pc_allocfrom = pcg;
                   1735:                                goto have_group;
                   1736:                        }
                   1737:                }
                   1738:
                   1739:                /*
                   1740:                 * No groups with any available objects.  Allocate
                   1741:                 * a new object, construct it, and return it to
                   1742:                 * the caller.  We will allocate a group, if necessary,
                   1743:                 * when the object is freed back to the cache.
                   1744:                 */
1.48      thorpej  1745:                pc->pc_misses++;
1.43      thorpej  1746:                simple_unlock(&pc->pc_slock);
                   1747:                object = pool_get(pc->pc_pool, flags);
                   1748:                if (object != NULL && pc->pc_ctor != NULL) {
                   1749:                        if ((*pc->pc_ctor)(pc->pc_arg, object, flags) != 0) {
                   1750:                                pool_put(pc->pc_pool, object);
                   1751:                                return (NULL);
                   1752:                        }
                   1753:                }
                   1754:                return (object);
                   1755:        }
                   1756:
                   1757:  have_group:
1.48      thorpej  1758:        pc->pc_hits++;
                   1759:        pc->pc_nitems--;
1.43      thorpej  1760:        object = pcg_get(pcg);
                   1761:
                   1762:        if (pcg->pcg_avail == 0)
                   1763:                pc->pc_allocfrom = NULL;
1.45      thorpej  1764:
1.43      thorpej  1765:        simple_unlock(&pc->pc_slock);
                   1766:
                   1767:        return (object);
                   1768: }
                   1769:
                   1770: /*
                   1771:  * pool_cache_put:
                   1772:  *
                   1773:  *     Put an object back to the pool cache.
                   1774:  */
                   1775: void
                   1776: pool_cache_put(struct pool_cache *pc, void *object)
                   1777: {
                   1778:        struct pool_cache_group *pcg;
1.60      thorpej  1779:        int s;
1.43      thorpej  1780:
                   1781:        simple_lock(&pc->pc_slock);
                   1782:
                   1783:        if ((pcg = pc->pc_freeto) == NULL) {
1.61      chs      1784:                TAILQ_FOREACH(pcg, &pc->pc_grouplist, pcg_list) {
1.43      thorpej  1785:                        if (pcg->pcg_avail != PCG_NOBJECTS) {
                   1786:                                pc->pc_freeto = pcg;
                   1787:                                goto have_group;
                   1788:                        }
                   1789:                }
                   1790:
                   1791:                /*
                   1792:                 * No empty groups to free the object to.  Attempt to
1.47      thorpej  1793:                 * allocate one.
1.43      thorpej  1794:                 */
1.47      thorpej  1795:                simple_unlock(&pc->pc_slock);
1.60      thorpej  1796:                s = splvm();
1.43      thorpej  1797:                pcg = pool_get(&pcgpool, PR_NOWAIT);
1.60      thorpej  1798:                splx(s);
1.43      thorpej  1799:                if (pcg != NULL) {
                   1800:                        memset(pcg, 0, sizeof(*pcg));
1.47      thorpej  1801:                        simple_lock(&pc->pc_slock);
1.48      thorpej  1802:                        pc->pc_ngroups++;
1.43      thorpej  1803:                        TAILQ_INSERT_TAIL(&pc->pc_grouplist, pcg, pcg_list);
1.47      thorpej  1804:                        if (pc->pc_freeto == NULL)
                   1805:                                pc->pc_freeto = pcg;
1.43      thorpej  1806:                        goto have_group;
                   1807:                }
                   1808:
                   1809:                /*
                   1810:                 * Unable to allocate a cache group; destruct the object
                   1811:                 * and free it back to the pool.
                   1812:                 */
1.51      thorpej  1813:                pool_cache_destruct_object(pc, object);
1.43      thorpej  1814:                return;
                   1815:        }
                   1816:
                   1817:  have_group:
1.48      thorpej  1818:        pc->pc_nitems++;
1.43      thorpej  1819:        pcg_put(pcg, object);
                   1820:
                   1821:        if (pcg->pcg_avail == PCG_NOBJECTS)
                   1822:                pc->pc_freeto = NULL;
                   1823:
                   1824:        simple_unlock(&pc->pc_slock);
1.51      thorpej  1825: }
                   1826:
                   1827: /*
                   1828:  * pool_cache_destruct_object:
                   1829:  *
                   1830:  *     Force destruction of an object and its release back into
                   1831:  *     the pool.
                   1832:  */
                   1833: void
                   1834: pool_cache_destruct_object(struct pool_cache *pc, void *object)
                   1835: {
                   1836:
                   1837:        if (pc->pc_dtor != NULL)
                   1838:                (*pc->pc_dtor)(pc->pc_arg, object);
                   1839:        pool_put(pc->pc_pool, object);
1.43      thorpej  1840: }
                   1841:
                   1842: /*
                   1843:  * pool_cache_do_invalidate:
                   1844:  *
                   1845:  *     This internal function implements pool_cache_invalidate() and
                   1846:  *     pool_cache_reclaim().
                   1847:  */
                   1848: static void
                   1849: pool_cache_do_invalidate(struct pool_cache *pc, int free_groups,
1.56      sommerfe 1850:     void (*putit)(struct pool *, void *))
1.43      thorpej  1851: {
                   1852:        struct pool_cache_group *pcg, *npcg;
                   1853:        void *object;
1.60      thorpej  1854:        int s;
1.43      thorpej  1855:
                   1856:        for (pcg = TAILQ_FIRST(&pc->pc_grouplist); pcg != NULL;
                   1857:             pcg = npcg) {
                   1858:                npcg = TAILQ_NEXT(pcg, pcg_list);
                   1859:                while (pcg->pcg_avail != 0) {
1.48      thorpej  1860:                        pc->pc_nitems--;
1.43      thorpej  1861:                        object = pcg_get(pcg);
1.45      thorpej  1862:                        if (pcg->pcg_avail == 0 && pc->pc_allocfrom == pcg)
                   1863:                                pc->pc_allocfrom = NULL;
1.43      thorpej  1864:                        if (pc->pc_dtor != NULL)
                   1865:                                (*pc->pc_dtor)(pc->pc_arg, object);
1.56      sommerfe 1866:                        (*putit)(pc->pc_pool, object);
1.43      thorpej  1867:                }
                   1868:                if (free_groups) {
1.48      thorpej  1869:                        pc->pc_ngroups--;
1.43      thorpej  1870:                        TAILQ_REMOVE(&pc->pc_grouplist, pcg, pcg_list);
1.46      thorpej  1871:                        if (pc->pc_freeto == pcg)
                   1872:                                pc->pc_freeto = NULL;
1.60      thorpej  1873:                        s = splvm();
1.43      thorpej  1874:                        pool_put(&pcgpool, pcg);
1.60      thorpej  1875:                        splx(s);
1.43      thorpej  1876:                }
                   1877:        }
                   1878: }
                   1879:
                   1880: /*
                   1881:  * pool_cache_invalidate:
                   1882:  *
                   1883:  *     Invalidate a pool cache (destruct and release all of the
                   1884:  *     cached objects).
                   1885:  */
                   1886: void
                   1887: pool_cache_invalidate(struct pool_cache *pc)
                   1888: {
                   1889:
                   1890:        simple_lock(&pc->pc_slock);
1.56      sommerfe 1891:        pool_cache_do_invalidate(pc, 0, pool_put);
1.43      thorpej  1892:        simple_unlock(&pc->pc_slock);
                   1893: }
                   1894:
                   1895: /*
                   1896:  * pool_cache_reclaim:
                   1897:  *
                   1898:  *     Reclaim a pool cache for pool_reclaim().
                   1899:  */
                   1900: static void
                   1901: pool_cache_reclaim(struct pool_cache *pc)
                   1902: {
                   1903:
1.47      thorpej  1904:        simple_lock(&pc->pc_slock);
1.43      thorpej  1905:        pool_cache_do_invalidate(pc, 1, pool_do_put);
                   1906:        simple_unlock(&pc->pc_slock);
1.3       pk       1907: }

CVSweb <webmaster@jp.NetBSD.org>