[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 1.26

1.26    ! thorpej     1: /*     $NetBSD: subr_pool.c,v 1.25 1999/05/10 21:13:05 thorpej Exp $   */
1.1       pk          2:
                      3: /*-
1.20      thorpej     4:  * Copyright (c) 1997, 1999 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"
1.1       pk         42:
                     43: #include <sys/param.h>
                     44: #include <sys/systm.h>
                     45: #include <sys/proc.h>
                     46: #include <sys/errno.h>
                     47: #include <sys/kernel.h>
                     48: #include <sys/malloc.h>
                     49: #include <sys/lock.h>
                     50: #include <sys/pool.h>
1.20      thorpej    51: #include <sys/syslog.h>
1.1       pk         52:
1.3       pk         53: #include <vm/vm.h>
                     54: #include <vm/vm_kern.h>
                     55:
                     56: #include <uvm/uvm.h>
                     57:
1.1       pk         58: /*
                     59:  * Pool resource management utility.
1.3       pk         60:  *
                     61:  * Memory is allocated in pages which are split into pieces according
                     62:  * to the pool item size. Each page is kept on a list headed by `pr_pagelist'
                     63:  * in the pool structure and the individual pool items are on a linked list
                     64:  * headed by `ph_itemlist' in each page header. The memory for building
                     65:  * the page list is either taken from the allocated pages themselves (for
                     66:  * small pool items) or taken from an internal pool of page headers (`phpool').
1.1       pk         67:  */
                     68:
1.3       pk         69: /* List of all pools */
1.5       thorpej    70: TAILQ_HEAD(,pool) pool_head = TAILQ_HEAD_INITIALIZER(pool_head);
1.3       pk         71:
                     72: /* Private pool for page header structures */
                     73: static struct pool phpool;
                     74:
                     75: /* # of seconds to retain page after last use */
                     76: int pool_inactive_time = 10;
                     77:
                     78: /* Next candidate for drainage (see pool_drain()) */
1.23      thorpej    79: static struct pool     *drainpp;
                     80:
                     81: /* This spin lock protects both pool_head and drainpp. */
                     82: struct simplelock pool_head_slock = SIMPLELOCK_INITIALIZER;
1.3       pk         83:
                     84: struct pool_item_header {
                     85:        /* Page headers */
                     86:        TAILQ_ENTRY(pool_item_header)
                     87:                                ph_pagelist;    /* pool page list */
                     88:        TAILQ_HEAD(,pool_item)  ph_itemlist;    /* chunk list for this page */
                     89:        LIST_ENTRY(pool_item_header)
                     90:                                ph_hashlist;    /* Off-page page headers */
                     91:        int                     ph_nmissing;    /* # of chunks in use */
                     92:        caddr_t                 ph_page;        /* this page's address */
                     93:        struct timeval          ph_time;        /* last referenced */
                     94: };
                     95:
1.1       pk         96: struct pool_item {
1.3       pk         97: #ifdef DIAGNOSTIC
                     98:        int pi_magic;
1.25      thorpej    99: #define        PI_MAGIC 0xdeadbeef
1.3       pk        100: #endif
                    101:        /* Other entries use only this list entry */
                    102:        TAILQ_ENTRY(pool_item)  pi_list;
                    103: };
                    104:
                    105:
1.25      thorpej   106: #define        PR_HASH_INDEX(pp,addr) \
1.3       pk        107:        (((u_long)(addr) >> (pp)->pr_pageshift) & (PR_HASHTABSIZE - 1))
                    108:
                    109:
                    110:
                    111: static struct pool_item_header
                    112:                *pr_find_pagehead __P((struct pool *, caddr_t));
                    113: static void    pr_rmpage __P((struct pool *, struct pool_item_header *));
1.20      thorpej   114: static int     pool_catchup __P((struct pool *));
1.21      thorpej   115: static void    pool_prime_page __P((struct pool *, caddr_t));
1.3       pk        116: static void    *pool_page_alloc __P((unsigned long, int, int));
                    117: static void    pool_page_free __P((void *, unsigned long, int));
                    118:
1.25      thorpej   119: static void pool_print1 __P((struct pool *, const char *,
                    120:        void (*)(const char *, ...)));
1.3       pk        121:
                    122: /*
                    123:  * Pool log entry. An array of these is allocated in pool_create().
                    124:  */
                    125: struct pool_log {
                    126:        const char      *pl_file;
                    127:        long            pl_line;
                    128:        int             pl_action;
1.25      thorpej   129: #define        PRLOG_GET       1
                    130: #define        PRLOG_PUT       2
1.3       pk        131:        void            *pl_addr;
1.1       pk        132: };
                    133:
1.3       pk        134: /* Number of entries in pool log buffers */
1.17      thorpej   135: #ifndef POOL_LOGSIZE
                    136: #define        POOL_LOGSIZE    10
                    137: #endif
                    138:
                    139: int pool_logsize = POOL_LOGSIZE;
1.1       pk        140:
1.25      thorpej   141: #ifdef DIAGNOSTIC
1.3       pk        142: static void    pr_log __P((struct pool *, void *, int, const char *, long));
1.25      thorpej   143: static void    pr_printlog __P((struct pool *, struct pool_item *,
                    144:                    void (*)(const char *, ...)));
                    145: static void    pr_enter __P((struct pool *, const char *, long));
                    146: static void    pr_leave __P((struct pool *));
                    147: static void    pr_enter_check __P((struct pool *,
                    148:                    void (*)(const char *, ...)));
1.3       pk        149:
                    150: static __inline__ void
                    151: pr_log(pp, v, action, file, line)
                    152:        struct pool     *pp;
                    153:        void            *v;
                    154:        int             action;
                    155:        const char      *file;
                    156:        long            line;
                    157: {
                    158:        int n = pp->pr_curlogentry;
                    159:        struct pool_log *pl;
                    160:
1.20      thorpej   161:        if ((pp->pr_roflags & PR_LOGGING) == 0)
1.3       pk        162:                return;
                    163:
                    164:        /*
                    165:         * Fill in the current entry. Wrap around and overwrite
                    166:         * the oldest entry if necessary.
                    167:         */
                    168:        pl = &pp->pr_log[n];
                    169:        pl->pl_file = file;
                    170:        pl->pl_line = line;
                    171:        pl->pl_action = action;
                    172:        pl->pl_addr = v;
                    173:        if (++n >= pp->pr_logsize)
                    174:                n = 0;
                    175:        pp->pr_curlogentry = n;
                    176: }
                    177:
                    178: static void
1.25      thorpej   179: pr_printlog(pp, pi, pr)
1.3       pk        180:        struct pool *pp;
1.25      thorpej   181:        struct pool_item *pi;
                    182:        void (*pr) __P((const char *, ...));
1.3       pk        183: {
                    184:        int i = pp->pr_logsize;
                    185:        int n = pp->pr_curlogentry;
                    186:
1.20      thorpej   187:        if ((pp->pr_roflags & PR_LOGGING) == 0)
1.3       pk        188:                return;
                    189:
                    190:        /*
                    191:         * Print all entries in this pool's log.
                    192:         */
                    193:        while (i-- > 0) {
                    194:                struct pool_log *pl = &pp->pr_log[n];
                    195:                if (pl->pl_action != 0) {
1.25      thorpej   196:                        if (pi == NULL || pi == pl->pl_addr) {
                    197:                                (*pr)("\tlog entry %d:\n", i);
                    198:                                (*pr)("\t\taction = %s, addr = %p\n",
                    199:                                    pl->pl_action == PRLOG_GET ? "get" : "put",
                    200:                                    pl->pl_addr);
                    201:                                (*pr)("\t\tfile: %s at line %lu\n",
                    202:                                    pl->pl_file, pl->pl_line);
                    203:                        }
1.3       pk        204:                }
                    205:                if (++n >= pp->pr_logsize)
                    206:                        n = 0;
                    207:        }
                    208: }
1.25      thorpej   209:
                    210: static __inline__ void
                    211: pr_enter(pp, file, line)
                    212:        struct pool *pp;
                    213:        const char *file;
                    214:        long line;
                    215: {
                    216:
                    217:        if (pp->pr_entered_file != NULL) {
                    218:                printf("pool %s: reentrancy at file %s line %ld\n",
                    219:                    pp->pr_wchan, file, line);
                    220:                printf("         previous entry at file %s line %ld\n",
                    221:                    pp->pr_entered_file, pp->pr_entered_line);
                    222:                panic("pr_enter");
                    223:        }
                    224:
                    225:        pp->pr_entered_file = file;
                    226:        pp->pr_entered_line = line;
                    227: }
                    228:
                    229: static __inline__ void
                    230: pr_leave(pp)
                    231:        struct pool *pp;
                    232: {
                    233:
                    234:        if (pp->pr_entered_file == NULL) {
                    235:                printf("pool %s not entered?\n", pp->pr_wchan);
                    236:                panic("pr_leave");
                    237:        }
                    238:
                    239:        pp->pr_entered_file = NULL;
                    240:        pp->pr_entered_line = 0;
                    241: }
                    242:
                    243: static __inline__ void
                    244: pr_enter_check(pp, pr)
                    245:        struct pool *pp;
                    246:        void (*pr) __P((const char *, ...));
                    247: {
                    248:
                    249:        if (pp->pr_entered_file != NULL)
                    250:                (*pr)("\n\tcurrently entered from file %s line %ld\n",
                    251:                    pp->pr_entered_file, pp->pr_entered_line);
                    252: }
1.3       pk        253: #else
1.25      thorpej   254: #define        pr_log(pp, v, action, file, line)
                    255: #define        pr_printlog(pp, pi, pr)
                    256: #define        pr_enter(pp, file, line)
                    257: #define        pr_leave(pp)
                    258: #define        pr_enter_check(pp, pr)
                    259: #endif /* DIAGNOSTIC */
1.3       pk        260:
                    261: /*
                    262:  * Return the pool page header based on page address.
                    263:  */
                    264: static __inline__ struct pool_item_header *
                    265: pr_find_pagehead(pp, page)
                    266:        struct pool *pp;
                    267:        caddr_t page;
                    268: {
                    269:        struct pool_item_header *ph;
                    270:
1.20      thorpej   271:        if ((pp->pr_roflags & PR_PHINPAGE) != 0)
1.3       pk        272:                return ((struct pool_item_header *)(page + pp->pr_phoffset));
                    273:
                    274:        for (ph = LIST_FIRST(&pp->pr_hashtab[PR_HASH_INDEX(pp, page)]);
                    275:             ph != NULL;
                    276:             ph = LIST_NEXT(ph, ph_hashlist)) {
                    277:                if (ph->ph_page == page)
                    278:                        return (ph);
                    279:        }
                    280:        return (NULL);
                    281: }
                    282:
                    283: /*
                    284:  * Remove a page from the pool.
                    285:  */
                    286: static __inline__ void
                    287: pr_rmpage(pp, ph)
                    288:        struct pool *pp;
                    289:        struct pool_item_header *ph;
                    290: {
                    291:
                    292:        /*
1.7       thorpej   293:         * If the page was idle, decrement the idle page count.
1.3       pk        294:         */
1.6       thorpej   295:        if (ph->ph_nmissing == 0) {
                    296: #ifdef DIAGNOSTIC
                    297:                if (pp->pr_nidle == 0)
                    298:                        panic("pr_rmpage: nidle inconsistent");
1.20      thorpej   299:                if (pp->pr_nitems < pp->pr_itemsperpage)
                    300:                        panic("pr_rmpage: nitems inconsistent");
1.6       thorpej   301: #endif
                    302:                pp->pr_nidle--;
                    303:        }
1.7       thorpej   304:
1.20      thorpej   305:        pp->pr_nitems -= pp->pr_itemsperpage;
                    306:
1.7       thorpej   307:        /*
                    308:         * Unlink a page from the pool and release it.
                    309:         */
                    310:        TAILQ_REMOVE(&pp->pr_pagelist, ph, ph_pagelist);
                    311:        (*pp->pr_free)(ph->ph_page, pp->pr_pagesz, pp->pr_mtype);
                    312:        pp->pr_npages--;
                    313:        pp->pr_npagefree++;
1.6       thorpej   314:
1.22      chs       315:        if ((pp->pr_roflags & PR_PHINPAGE) == 0) {
                    316:                LIST_REMOVE(ph, ph_hashlist);
                    317:                pool_put(&phpool, ph);
                    318:        }
                    319:
1.3       pk        320:        if (pp->pr_curpage == ph) {
                    321:                /*
                    322:                 * Find a new non-empty page header, if any.
                    323:                 * Start search from the page head, to increase the
                    324:                 * chance for "high water" pages to be freed.
                    325:                 */
                    326:                for (ph = TAILQ_FIRST(&pp->pr_pagelist); ph != NULL;
                    327:                     ph = TAILQ_NEXT(ph, ph_pagelist))
                    328:                        if (TAILQ_FIRST(&ph->ph_itemlist) != NULL)
                    329:                                break;
                    330:
                    331:                pp->pr_curpage = ph;
1.21      thorpej   332:        }
1.3       pk        333: }
                    334:
                    335: /*
                    336:  * Allocate and initialize a pool.
                    337:  */
1.1       pk        338: struct pool *
1.3       pk        339: pool_create(size, align, ioff, nitems, wchan, pagesz, alloc, release, mtype)
1.1       pk        340:        size_t  size;
1.3       pk        341:        u_int   align;
                    342:        u_int   ioff;
1.1       pk        343:        int     nitems;
1.21      thorpej   344:        const char *wchan;
1.3       pk        345:        size_t  pagesz;
                    346:        void    *(*alloc) __P((unsigned long, int, int));
                    347:        void    (*release) __P((void *, unsigned long, int));
1.1       pk        348:        int     mtype;
                    349: {
                    350:        struct pool *pp;
1.3       pk        351:        int flags;
1.1       pk        352:
1.3       pk        353:        pp = (struct pool *)malloc(sizeof(*pp), M_POOL, M_NOWAIT);
                    354:        if (pp == NULL)
1.1       pk        355:                return (NULL);
1.3       pk        356:
                    357:        flags = PR_FREEHEADER;
                    358:        pool_init(pp, size, align, ioff, flags, wchan, pagesz,
                    359:                  alloc, release, mtype);
                    360:
                    361:        if (nitems != 0) {
                    362:                if (pool_prime(pp, nitems, NULL) != 0) {
                    363:                        pool_destroy(pp);
                    364:                        return (NULL);
                    365:                }
1.1       pk        366:        }
                    367:
1.3       pk        368:        return (pp);
                    369: }
                    370:
                    371: /*
                    372:  * Initialize the given pool resource structure.
                    373:  *
                    374:  * We export this routine to allow other kernel parts to declare
                    375:  * static pools that must be initialized before malloc() is available.
                    376:  */
                    377: void
                    378: pool_init(pp, size, align, ioff, flags, wchan, pagesz, alloc, release, mtype)
                    379:        struct pool     *pp;
                    380:        size_t          size;
                    381:        u_int           align;
                    382:        u_int           ioff;
                    383:        int             flags;
1.21      thorpej   384:        const char      *wchan;
1.3       pk        385:        size_t          pagesz;
                    386:        void            *(*alloc) __P((unsigned long, int, int));
                    387:        void            (*release) __P((void *, unsigned long, int));
                    388:        int             mtype;
                    389: {
1.16      briggs    390:        int off, slack, i;
1.3       pk        391:
1.25      thorpej   392: #ifdef POOL_DIAGNOSTIC
                    393:        /*
                    394:         * Always log if POOL_DIAGNOSTIC is defined.
                    395:         */
                    396:        if (pool_logsize != 0)
                    397:                flags |= PR_LOGGING;
                    398: #endif
                    399:
1.3       pk        400:        /*
                    401:         * Check arguments and construct default values.
                    402:         */
                    403:        if (!powerof2(pagesz) || pagesz > PAGE_SIZE)
                    404:                panic("pool_init: page size invalid (%lx)\n", (u_long)pagesz);
                    405:
1.4       thorpej   406:        if (alloc == NULL && release == NULL) {
1.3       pk        407:                alloc = pool_page_alloc;
                    408:                release = pool_page_free;
1.4       thorpej   409:                pagesz = PAGE_SIZE;     /* Rounds to PAGE_SIZE anyhow. */
                    410:        } else if ((alloc != NULL && release != NULL) == 0) {
                    411:                /* If you specifiy one, must specify both. */
                    412:                panic("pool_init: must specify alloc and release together");
                    413:        }
                    414:
1.3       pk        415:        if (pagesz == 0)
                    416:                pagesz = PAGE_SIZE;
                    417:
                    418:        if (align == 0)
                    419:                align = ALIGN(1);
1.14      thorpej   420:
                    421:        if (size < sizeof(struct pool_item))
                    422:                size = sizeof(struct pool_item);
1.3       pk        423:
                    424:        /*
                    425:         * Initialize the pool structure.
                    426:         */
                    427:        TAILQ_INIT(&pp->pr_pagelist);
                    428:        pp->pr_curpage = NULL;
                    429:        pp->pr_npages = 0;
                    430:        pp->pr_minitems = 0;
                    431:        pp->pr_minpages = 0;
                    432:        pp->pr_maxpages = UINT_MAX;
1.20      thorpej   433:        pp->pr_roflags = flags;
                    434:        pp->pr_flags = 0;
1.3       pk        435:        pp->pr_size = ALIGN(size);
                    436:        pp->pr_align = align;
                    437:        pp->pr_wchan = wchan;
                    438:        pp->pr_mtype = mtype;
                    439:        pp->pr_alloc = alloc;
                    440:        pp->pr_free = release;
                    441:        pp->pr_pagesz = pagesz;
                    442:        pp->pr_pagemask = ~(pagesz - 1);
                    443:        pp->pr_pageshift = ffs(pagesz) - 1;
1.20      thorpej   444:        pp->pr_nitems = 0;
                    445:        pp->pr_nout = 0;
                    446:        pp->pr_hardlimit = UINT_MAX;
                    447:        pp->pr_hardlimit_warning = NULL;
                    448:        pp->pr_hardlimit_ratecap = 0;
                    449:        memset(&pp->pr_hardlimit_warning_last, 0,
                    450:            sizeof(pp->pr_hardlimit_warning_last));
1.3       pk        451:
                    452:        /*
                    453:         * Decide whether to put the page header off page to avoid
                    454:         * wasting too large a part of the page. Off-page page headers
                    455:         * go on a hash table, so we can match a returned item
                    456:         * with its header based on the page address.
                    457:         * We use 1/16 of the page size as the threshold (XXX: tune)
                    458:         */
                    459:        if (pp->pr_size < pagesz/16) {
                    460:                /* Use the end of the page for the page header */
1.20      thorpej   461:                pp->pr_roflags |= PR_PHINPAGE;
1.3       pk        462:                pp->pr_phoffset = off =
                    463:                        pagesz - ALIGN(sizeof(struct pool_item_header));
1.2       pk        464:        } else {
1.3       pk        465:                /* The page header will be taken from our page header pool */
                    466:                pp->pr_phoffset = 0;
                    467:                off = pagesz;
1.16      briggs    468:                for (i = 0; i < PR_HASHTABSIZE; i++) {
                    469:                        LIST_INIT(&pp->pr_hashtab[i]);
                    470:                }
1.2       pk        471:        }
1.1       pk        472:
1.3       pk        473:        /*
                    474:         * Alignment is to take place at `ioff' within the item. This means
                    475:         * we must reserve up to `align - 1' bytes on the page to allow
                    476:         * appropriate positioning of each item.
                    477:         *
                    478:         * Silently enforce `0 <= ioff < align'.
                    479:         */
                    480:        pp->pr_itemoffset = ioff = ioff % align;
                    481:        pp->pr_itemsperpage = (off - ((align - ioff) % align)) / pp->pr_size;
                    482:
                    483:        /*
                    484:         * Use the slack between the chunks and the page header
                    485:         * for "cache coloring".
                    486:         */
                    487:        slack = off - pp->pr_itemsperpage * pp->pr_size;
                    488:        pp->pr_maxcolor = (slack / align) * align;
                    489:        pp->pr_curcolor = 0;
                    490:
                    491:        pp->pr_nget = 0;
                    492:        pp->pr_nfail = 0;
                    493:        pp->pr_nput = 0;
                    494:        pp->pr_npagealloc = 0;
                    495:        pp->pr_npagefree = 0;
1.1       pk        496:        pp->pr_hiwat = 0;
1.8       thorpej   497:        pp->pr_nidle = 0;
1.3       pk        498:
1.25      thorpej   499:        if (flags & PR_LOGGING) {
                    500:                if (kmem_map == NULL ||
                    501:                    (pp->pr_log = malloc(pool_logsize * sizeof(struct pool_log),
                    502:                     M_TEMP, M_NOWAIT)) == NULL)
1.20      thorpej   503:                        pp->pr_roflags &= ~PR_LOGGING;
1.3       pk        504:                pp->pr_curlogentry = 0;
                    505:                pp->pr_logsize = pool_logsize;
                    506:        }
1.25      thorpej   507:
                    508:        pp->pr_entered_file = NULL;
                    509:        pp->pr_entered_line = 0;
1.3       pk        510:
1.21      thorpej   511:        simple_lock_init(&pp->pr_slock);
1.1       pk        512:
1.3       pk        513:        /*
                    514:         * Initialize private page header pool if we haven't done so yet.
1.23      thorpej   515:         * XXX LOCKING.
1.3       pk        516:         */
                    517:        if (phpool.pr_size == 0) {
                    518:                pool_init(&phpool, sizeof(struct pool_item_header), 0, 0,
                    519:                          0, "phpool", 0, 0, 0, 0);
1.1       pk        520:        }
                    521:
1.23      thorpej   522:        /* Insert into the list of all pools. */
                    523:        simple_lock(&pool_head_slock);
                    524:        TAILQ_INSERT_TAIL(&pool_head, pp, pr_poollist);
                    525:        simple_unlock(&pool_head_slock);
1.1       pk        526: }
                    527:
                    528: /*
                    529:  * De-commision a pool resource.
                    530:  */
                    531: void
                    532: pool_destroy(pp)
                    533:        struct pool *pp;
                    534: {
1.3       pk        535:        struct pool_item_header *ph;
                    536:
                    537: #ifdef DIAGNOSTIC
1.20      thorpej   538:        if (pp->pr_nout != 0) {
1.25      thorpej   539:                pr_printlog(pp, NULL, printf);
1.20      thorpej   540:                panic("pool_destroy: pool busy: still out: %u\n",
                    541:                    pp->pr_nout);
1.3       pk        542:        }
                    543: #endif
1.1       pk        544:
1.3       pk        545:        /* Remove all pages */
1.20      thorpej   546:        if ((pp->pr_roflags & PR_STATIC) == 0)
1.3       pk        547:                while ((ph = pp->pr_pagelist.tqh_first) != NULL)
                    548:                        pr_rmpage(pp, ph);
                    549:
                    550:        /* Remove from global pool list */
1.23      thorpej   551:        simple_lock(&pool_head_slock);
1.3       pk        552:        TAILQ_REMOVE(&pool_head, pp, pr_poollist);
1.23      thorpej   553:        /* XXX Only clear this if we were drainpp? */
1.3       pk        554:        drainpp = NULL;
1.23      thorpej   555:        simple_unlock(&pool_head_slock);
1.3       pk        556:
1.20      thorpej   557:        if ((pp->pr_roflags & PR_LOGGING) != 0)
1.3       pk        558:                free(pp->pr_log, M_TEMP);
1.2       pk        559:
1.20      thorpej   560:        if (pp->pr_roflags & PR_FREEHEADER)
1.3       pk        561:                free(pp, M_POOL);
1.1       pk        562: }
                    563:
                    564:
                    565: /*
1.3       pk        566:  * Grab an item from the pool; must be called at appropriate spl level
1.1       pk        567:  */
1.3       pk        568: void *
                    569: _pool_get(pp, flags, file, line)
                    570:        struct pool *pp;
                    571:        int flags;
                    572:        const char *file;
                    573:        long line;
1.1       pk        574: {
                    575:        void *v;
                    576:        struct pool_item *pi;
1.3       pk        577:        struct pool_item_header *ph;
1.1       pk        578:
1.2       pk        579: #ifdef DIAGNOSTIC
1.20      thorpej   580:        if ((pp->pr_roflags & PR_STATIC) && (flags & PR_MALLOCOK)) {
1.25      thorpej   581:                pr_printlog(pp, NULL, printf);
1.2       pk        582:                panic("pool_get: static");
1.3       pk        583:        }
1.2       pk        584: #endif
                    585:
1.3       pk        586:        if (curproc == NULL && (flags & PR_WAITOK) != 0)
                    587:                panic("pool_get: must have NOWAIT");
1.1       pk        588:
1.21      thorpej   589:        simple_lock(&pp->pr_slock);
1.25      thorpej   590:        pr_enter(pp, file, line);
1.20      thorpej   591:
                    592:  startover:
                    593:        /*
                    594:         * Check to see if we've reached the hard limit.  If we have,
                    595:         * and we can wait, then wait until an item has been returned to
                    596:         * the pool.
                    597:         */
                    598: #ifdef DIAGNOSTIC
                    599:        if (pp->pr_nout > pp->pr_hardlimit) {
1.25      thorpej   600:                pr_leave(pp);
1.21      thorpej   601:                simple_unlock(&pp->pr_slock);
1.20      thorpej   602:                panic("pool_get: %s: crossed hard limit", pp->pr_wchan);
                    603:        }
                    604: #endif
                    605:        if (pp->pr_nout == pp->pr_hardlimit) {
                    606:                if (flags & PR_WAITOK) {
                    607:                        /*
                    608:                         * XXX: A warning isn't logged in this case.  Should
                    609:                         * it be?
                    610:                         */
                    611:                        pp->pr_flags |= PR_WANTED;
1.25      thorpej   612:                        pr_leave(pp);
1.21      thorpej   613:                        simple_unlock(&pp->pr_slock);
1.20      thorpej   614:                        tsleep((caddr_t)pp, PSWP, pp->pr_wchan, 0);
1.21      thorpej   615:                        simple_lock(&pp->pr_slock);
1.25      thorpej   616:                        pr_enter(pp, file, line);
1.20      thorpej   617:                        goto startover;
                    618:                }
                    619:                if (pp->pr_hardlimit_warning != NULL) {
                    620:                        /*
                    621:                         * Log a message that the hard limit has been hit.
                    622:                         */
                    623:                        struct timeval curtime, logdiff;
                    624:                        int s = splclock();
                    625:                        curtime = mono_time;
                    626:                        splx(s);
                    627:                        timersub(&curtime, &pp->pr_hardlimit_warning_last,
                    628:                            &logdiff);
                    629:                        if (logdiff.tv_sec >= pp->pr_hardlimit_ratecap) {
                    630:                                pp->pr_hardlimit_warning_last = curtime;
                    631:                                log(LOG_ERR, "%s\n", pp->pr_hardlimit_warning);
                    632:                        }
                    633:                }
1.21      thorpej   634:
                    635:                if (flags & PR_URGENT)
                    636:                        panic("pool_get: urgent");
                    637:
                    638:                pp->pr_nfail++;
                    639:
1.25      thorpej   640:                pr_leave(pp);
1.21      thorpej   641:                simple_unlock(&pp->pr_slock);
1.20      thorpej   642:                return (NULL);
                    643:        }
                    644:
1.3       pk        645:        /*
                    646:         * The convention we use is that if `curpage' is not NULL, then
                    647:         * it points at a non-empty bucket. In particular, `curpage'
                    648:         * never points at a page header which has PR_PHINPAGE set and
                    649:         * has no items in its bucket.
                    650:         */
1.20      thorpej   651:        if ((ph = pp->pr_curpage) == NULL) {
1.15      pk        652:                void *v;
                    653:
1.20      thorpej   654: #ifdef DIAGNOSTIC
                    655:                if (pp->pr_nitems != 0) {
1.21      thorpej   656:                        simple_unlock(&pp->pr_slock);
1.20      thorpej   657:                        printf("pool_get: %s: curpage NULL, nitems %u\n",
                    658:                            pp->pr_wchan, pp->pr_nitems);
                    659:                        panic("pool_get: nitems inconsistent\n");
                    660:                }
                    661: #endif
                    662:
1.21      thorpej   663:                /*
                    664:                 * Call the back-end page allocator for more memory.
                    665:                 * Release the pool lock, as the back-end page allocator
                    666:                 * may block.
                    667:                 */
1.25      thorpej   668:                pr_leave(pp);
1.21      thorpej   669:                simple_unlock(&pp->pr_slock);
                    670:                v = (*pp->pr_alloc)(pp->pr_pagesz, flags, pp->pr_mtype);
                    671:                simple_lock(&pp->pr_slock);
1.25      thorpej   672:                pr_enter(pp, file, line);
1.15      pk        673:
1.21      thorpej   674:                if (v == NULL) {
                    675:                        /*
                    676:                         * We were unable to allocate a page, but
                    677:                         * we released the lock during allocation,
                    678:                         * so perhaps items were freed back to the
                    679:                         * pool.  Check for this case.
                    680:                         */
                    681:                        if (pp->pr_curpage != NULL)
                    682:                                goto startover;
1.15      pk        683:
1.3       pk        684:                        if (flags & PR_URGENT)
                    685:                                panic("pool_get: urgent");
1.21      thorpej   686:
1.3       pk        687:                        if ((flags & PR_WAITOK) == 0) {
                    688:                                pp->pr_nfail++;
1.25      thorpej   689:                                pr_leave(pp);
1.21      thorpej   690:                                simple_unlock(&pp->pr_slock);
1.1       pk        691:                                return (NULL);
1.3       pk        692:                        }
                    693:
1.15      pk        694:                        /*
                    695:                         * Wait for items to be returned to this pool.
1.21      thorpej   696:                         *
1.15      pk        697:                         * XXX: we actually want to wait just until
                    698:                         * the page allocator has memory again. Depending
                    699:                         * on this pool's usage, we might get stuck here
                    700:                         * for a long time.
1.20      thorpej   701:                         *
                    702:                         * XXX: maybe we should wake up once a second and
                    703:                         * try again?
1.15      pk        704:                         */
1.1       pk        705:                        pp->pr_flags |= PR_WANTED;
1.25      thorpej   706:                        pr_leave(pp);
1.21      thorpej   707:                        simple_unlock(&pp->pr_slock);
1.1       pk        708:                        tsleep((caddr_t)pp, PSWP, pp->pr_wchan, 0);
1.21      thorpej   709:                        simple_lock(&pp->pr_slock);
1.25      thorpej   710:                        pr_enter(pp, file, line);
1.20      thorpej   711:                        goto startover;
1.1       pk        712:                }
1.3       pk        713:
1.15      pk        714:                /* We have more memory; add it to the pool */
                    715:                pp->pr_npagealloc++;
                    716:                pool_prime_page(pp, v);
                    717:
1.20      thorpej   718:                /* Start the allocation process over. */
                    719:                goto startover;
1.3       pk        720:        }
                    721:
1.21      thorpej   722:        if ((v = pi = TAILQ_FIRST(&ph->ph_itemlist)) == NULL) {
1.25      thorpej   723:                pr_leave(pp);
1.21      thorpej   724:                simple_unlock(&pp->pr_slock);
1.3       pk        725:                panic("pool_get: %s: page empty", pp->pr_wchan);
1.21      thorpej   726:        }
1.20      thorpej   727: #ifdef DIAGNOSTIC
                    728:        if (pp->pr_nitems == 0) {
1.25      thorpej   729:                pr_leave(pp);
1.21      thorpej   730:                simple_unlock(&pp->pr_slock);
1.20      thorpej   731:                printf("pool_get: %s: items on itemlist, nitems %u\n",
                    732:                    pp->pr_wchan, pp->pr_nitems);
                    733:                panic("pool_get: nitems inconsistent\n");
                    734:        }
                    735: #endif
1.3       pk        736:        pr_log(pp, v, PRLOG_GET, file, line);
                    737:
                    738: #ifdef DIAGNOSTIC
                    739:        if (pi->pi_magic != PI_MAGIC) {
1.25      thorpej   740:                pr_printlog(pp, pi, printf);
1.3       pk        741:                panic("pool_get(%s): free list modified: magic=%x; page %p;"
                    742:                       " item addr %p\n",
                    743:                        pp->pr_wchan, pi->pi_magic, ph->ph_page, pi);
                    744:        }
                    745: #endif
                    746:
                    747:        /*
                    748:         * Remove from item list.
                    749:         */
                    750:        TAILQ_REMOVE(&ph->ph_itemlist, pi, pi_list);
1.20      thorpej   751:        pp->pr_nitems--;
                    752:        pp->pr_nout++;
1.6       thorpej   753:        if (ph->ph_nmissing == 0) {
                    754: #ifdef DIAGNOSTIC
                    755:                if (pp->pr_nidle == 0)
                    756:                        panic("pool_get: nidle inconsistent");
                    757: #endif
                    758:                pp->pr_nidle--;
                    759:        }
1.3       pk        760:        ph->ph_nmissing++;
                    761:        if (TAILQ_FIRST(&ph->ph_itemlist) == NULL) {
1.21      thorpej   762: #ifdef DIAGNOSTIC
                    763:                if (ph->ph_nmissing != pp->pr_itemsperpage) {
1.25      thorpej   764:                        pr_leave(pp);
1.21      thorpej   765:                        simple_unlock(&pp->pr_slock);
                    766:                        panic("pool_get: %s: nmissing inconsistent",
                    767:                            pp->pr_wchan);
                    768:                }
                    769: #endif
1.3       pk        770:                /*
                    771:                 * Find a new non-empty page header, if any.
                    772:                 * Start search from the page head, to increase
                    773:                 * the chance for "high water" pages to be freed.
                    774:                 *
1.21      thorpej   775:                 * Migrate empty pages to the end of the list.  This
                    776:                 * will speed the update of curpage as pages become
                    777:                 * idle.  Empty pages intermingled with idle pages
                    778:                 * is no big deal.  As soon as a page becomes un-empty,
                    779:                 * it will move back to the head of the list.
1.3       pk        780:                 */
                    781:                TAILQ_REMOVE(&pp->pr_pagelist, ph, ph_pagelist);
1.21      thorpej   782:                TAILQ_INSERT_TAIL(&pp->pr_pagelist, ph, ph_pagelist);
                    783:                for (ph = TAILQ_FIRST(&pp->pr_pagelist); ph != NULL;
                    784:                     ph = TAILQ_NEXT(ph, ph_pagelist))
1.3       pk        785:                        if (TAILQ_FIRST(&ph->ph_itemlist) != NULL)
                    786:                                break;
                    787:
                    788:                pp->pr_curpage = ph;
1.1       pk        789:        }
1.3       pk        790:
                    791:        pp->pr_nget++;
1.20      thorpej   792:
                    793:        /*
                    794:         * If we have a low water mark and we are now below that low
                    795:         * water mark, add more items to the pool.
                    796:         */
                    797:        if (pp->pr_nitems < pp->pr_minitems && pool_catchup(pp) != 0) {
                    798:                /*
                    799:                 * XXX: Should we log a warning?  Should we set up a timeout
                    800:                 * to try again in a second or so?  The latter could break
                    801:                 * a caller's assumptions about interrupt protection, etc.
                    802:                 */
                    803:        }
                    804:
1.25      thorpej   805:        pr_leave(pp);
1.21      thorpej   806:        simple_unlock(&pp->pr_slock);
1.1       pk        807:        return (v);
                    808: }
                    809:
                    810: /*
1.3       pk        811:  * Return resource to the pool; must be called at appropriate spl level
1.1       pk        812:  */
1.3       pk        813: void
                    814: _pool_put(pp, v, file, line)
                    815:        struct pool *pp;
                    816:        void *v;
                    817:        const char *file;
                    818:        long line;
1.1       pk        819: {
                    820:        struct pool_item *pi = v;
1.3       pk        821:        struct pool_item_header *ph;
                    822:        caddr_t page;
1.21      thorpej   823:        int s;
1.3       pk        824:
                    825:        page = (caddr_t)((u_long)v & pp->pr_pagemask);
1.1       pk        826:
1.21      thorpej   827:        simple_lock(&pp->pr_slock);
1.25      thorpej   828:        pr_enter(pp, file, line);
1.3       pk        829:
                    830:        pr_log(pp, v, PRLOG_PUT, file, line);
                    831:
                    832:        if ((ph = pr_find_pagehead(pp, page)) == NULL) {
1.25      thorpej   833:                pr_printlog(pp, NULL, printf);
1.3       pk        834:                panic("pool_put: %s: page header missing", pp->pr_wchan);
                    835:        }
                    836:
                    837:        /*
                    838:         * Return to item list.
                    839:         */
1.2       pk        840: #ifdef DIAGNOSTIC
1.25      thorpej   841:        /* XXX Should fill the item. */
1.3       pk        842:        pi->pi_magic = PI_MAGIC;
                    843: #endif
                    844:        TAILQ_INSERT_HEAD(&ph->ph_itemlist, pi, pi_list);
                    845:        ph->ph_nmissing--;
                    846:        pp->pr_nput++;
1.20      thorpej   847:        pp->pr_nitems++;
                    848:        pp->pr_nout--;
1.3       pk        849:
                    850:        /* Cancel "pool empty" condition if it exists */
                    851:        if (pp->pr_curpage == NULL)
                    852:                pp->pr_curpage = ph;
                    853:
                    854:        if (pp->pr_flags & PR_WANTED) {
                    855:                pp->pr_flags &= ~PR_WANTED;
1.15      pk        856:                if (ph->ph_nmissing == 0)
                    857:                        pp->pr_nidle++;
1.25      thorpej   858:                pr_leave(pp);
1.21      thorpej   859:                simple_unlock(&pp->pr_slock);
1.3       pk        860:                wakeup((caddr_t)pp);
                    861:                return;
                    862:        }
                    863:
                    864:        /*
1.21      thorpej   865:         * If this page is now complete, do one of two things:
                    866:         *
                    867:         *      (1) If we have more pages than the page high water
                    868:         *          mark, free the page back to the system.
                    869:         *
                    870:         *      (2) Move it to the end of the page list, so that
                    871:         *          we minimize our chances of fragmenting the
                    872:         *          pool.  Idle pages migrate to the end (along with
                    873:         *          completely empty pages, so that we find un-empty
                    874:         *          pages more quickly when we update curpage) of the
                    875:         *          list so they can be more easily swept up by
                    876:         *          the pagedaemon when pages are scarce.
1.3       pk        877:         */
                    878:        if (ph->ph_nmissing == 0) {
1.6       thorpej   879:                pp->pr_nidle++;
1.3       pk        880:                if (pp->pr_npages > pp->pr_maxpages) {
                    881:                        pr_rmpage(pp, ph);
                    882:                } else {
                    883:                        TAILQ_REMOVE(&pp->pr_pagelist, ph, ph_pagelist);
                    884:                        TAILQ_INSERT_TAIL(&pp->pr_pagelist, ph, ph_pagelist);
                    885:
1.21      thorpej   886:                        /*
                    887:                         * Update the timestamp on the page.  A page must
                    888:                         * be idle for some period of time before it can
                    889:                         * be reclaimed by the pagedaemon.  This minimizes
                    890:                         * ping-pong'ing for memory.
                    891:                         */
                    892:                        s = splclock();
                    893:                        ph->ph_time = mono_time;
                    894:                        splx(s);
                    895:
                    896:                        /*
                    897:                         * Update the current page pointer.  Just look for
                    898:                         * the first page with any free items.
                    899:                         *
                    900:                         * XXX: Maybe we want an option to look for the
                    901:                         * page with the fewest available items, to minimize
                    902:                         * fragmentation?
                    903:                         */
1.3       pk        904:                        for (ph = TAILQ_FIRST(&pp->pr_pagelist); ph != NULL;
                    905:                             ph = TAILQ_NEXT(ph, ph_pagelist))
                    906:                                if (TAILQ_FIRST(&ph->ph_itemlist) != NULL)
                    907:                                        break;
1.1       pk        908:
1.3       pk        909:                        pp->pr_curpage = ph;
1.1       pk        910:                }
                    911:        }
1.21      thorpej   912:        /*
                    913:         * If the page has just become un-empty, move it to the head of
                    914:         * the list, and make it the current page.  The next allocation
                    915:         * will get the item from this page, instead of further fragmenting
                    916:         * the pool.
                    917:         */
                    918:        else if (ph->ph_nmissing == (pp->pr_itemsperpage - 1)) {
                    919:                TAILQ_REMOVE(&pp->pr_pagelist, ph, ph_pagelist);
                    920:                TAILQ_INSERT_HEAD(&pp->pr_pagelist, ph, ph_pagelist);
                    921:                pp->pr_curpage = ph;
                    922:        }
                    923:
1.25      thorpej   924:        pr_leave(pp);
1.21      thorpej   925:        simple_unlock(&pp->pr_slock);
1.3       pk        926:
1.1       pk        927: }
                    928:
                    929: /*
1.3       pk        930:  * Add N items to the pool.
1.1       pk        931:  */
                    932: int
1.2       pk        933: pool_prime(pp, n, storage)
1.1       pk        934:        struct pool *pp;
                    935:        int n;
1.2       pk        936:        caddr_t storage;
1.1       pk        937: {
1.3       pk        938:        caddr_t cp;
                    939:        int newnitems, newpages;
1.2       pk        940:
                    941: #ifdef DIAGNOSTIC
1.20      thorpej   942:        if (storage && !(pp->pr_roflags & PR_STATIC))
1.2       pk        943:                panic("pool_prime: static");
                    944:        /* !storage && static caught below */
                    945: #endif
1.1       pk        946:
1.21      thorpej   947:        simple_lock(&pp->pr_slock);
                    948:
1.3       pk        949:        newnitems = pp->pr_minitems + n;
                    950:        newpages =
1.18      thorpej   951:                roundup(newnitems, pp->pr_itemsperpage) / pp->pr_itemsperpage
1.3       pk        952:                - pp->pr_minpages;
                    953:
                    954:        while (newpages-- > 0) {
1.20      thorpej   955:                if (pp->pr_roflags & PR_STATIC) {
1.3       pk        956:                        cp = storage;
                    957:                        storage += pp->pr_pagesz;
                    958:                } else {
1.21      thorpej   959:                        simple_unlock(&pp->pr_slock);
1.3       pk        960:                        cp = (*pp->pr_alloc)(pp->pr_pagesz, 0, pp->pr_mtype);
1.21      thorpej   961:                        simple_lock(&pp->pr_slock);
1.3       pk        962:                }
1.2       pk        963:
1.3       pk        964:                if (cp == NULL) {
1.21      thorpej   965:                        simple_unlock(&pp->pr_slock);
1.1       pk        966:                        return (ENOMEM);
                    967:                }
                    968:
1.26    ! thorpej   969:                pp->pr_npagealloc++;
1.3       pk        970:                pool_prime_page(pp, cp);
                    971:                pp->pr_minpages++;
1.1       pk        972:        }
1.3       pk        973:
                    974:        pp->pr_minitems = newnitems;
                    975:
                    976:        if (pp->pr_minpages >= pp->pr_maxpages)
                    977:                pp->pr_maxpages = pp->pr_minpages + 1;  /* XXX */
                    978:
1.21      thorpej   979:        simple_unlock(&pp->pr_slock);
1.1       pk        980:        return (0);
                    981: }
1.3       pk        982:
                    983: /*
                    984:  * Add a page worth of items to the pool.
1.21      thorpej   985:  *
                    986:  * Note, we must be called with the pool descriptor LOCKED.
1.3       pk        987:  */
1.21      thorpej   988: static void
1.3       pk        989: pool_prime_page(pp, storage)
                    990:        struct pool *pp;
                    991:        caddr_t storage;
                    992: {
                    993:        struct pool_item *pi;
                    994:        struct pool_item_header *ph;
                    995:        caddr_t cp = storage;
                    996:        unsigned int align = pp->pr_align;
                    997:        unsigned int ioff = pp->pr_itemoffset;
                    998:        int n;
                    999:
1.20      thorpej  1000:        if ((pp->pr_roflags & PR_PHINPAGE) != 0) {
1.3       pk       1001:                ph = (struct pool_item_header *)(cp + pp->pr_phoffset);
                   1002:        } else {
                   1003:                ph = pool_get(&phpool, PR_URGENT);
                   1004:                LIST_INSERT_HEAD(&pp->pr_hashtab[PR_HASH_INDEX(pp, cp)],
                   1005:                                 ph, ph_hashlist);
                   1006:        }
                   1007:
                   1008:        /*
                   1009:         * Insert page header.
                   1010:         */
                   1011:        TAILQ_INSERT_HEAD(&pp->pr_pagelist, ph, ph_pagelist);
                   1012:        TAILQ_INIT(&ph->ph_itemlist);
                   1013:        ph->ph_page = storage;
                   1014:        ph->ph_nmissing = 0;
1.21      thorpej  1015:        memset(&ph->ph_time, 0, sizeof(ph->ph_time));
1.3       pk       1016:
1.6       thorpej  1017:        pp->pr_nidle++;
                   1018:
1.3       pk       1019:        /*
                   1020:         * Color this page.
                   1021:         */
                   1022:        cp = (caddr_t)(cp + pp->pr_curcolor);
                   1023:        if ((pp->pr_curcolor += align) > pp->pr_maxcolor)
                   1024:                pp->pr_curcolor = 0;
                   1025:
                   1026:        /*
                   1027:         * Adjust storage to apply aligment to `pr_itemoffset' in each item.
                   1028:         */
                   1029:        if (ioff != 0)
                   1030:                cp = (caddr_t)(cp + (align - ioff));
                   1031:
                   1032:        /*
                   1033:         * Insert remaining chunks on the bucket list.
                   1034:         */
                   1035:        n = pp->pr_itemsperpage;
1.20      thorpej  1036:        pp->pr_nitems += n;
1.3       pk       1037:
                   1038:        while (n--) {
                   1039:                pi = (struct pool_item *)cp;
                   1040:
                   1041:                /* Insert on page list */
                   1042:                TAILQ_INSERT_TAIL(&ph->ph_itemlist, pi, pi_list);
                   1043: #ifdef DIAGNOSTIC
                   1044:                pi->pi_magic = PI_MAGIC;
                   1045: #endif
                   1046:                cp = (caddr_t)(cp + pp->pr_size);
                   1047:        }
                   1048:
                   1049:        /*
                   1050:         * If the pool was depleted, point at the new page.
                   1051:         */
                   1052:        if (pp->pr_curpage == NULL)
                   1053:                pp->pr_curpage = ph;
                   1054:
                   1055:        if (++pp->pr_npages > pp->pr_hiwat)
                   1056:                pp->pr_hiwat = pp->pr_npages;
                   1057: }
                   1058:
1.20      thorpej  1059: /*
                   1060:  * Like pool_prime(), except this is used by pool_get() when nitems
                   1061:  * drops below the low water mark.  This is used to catch up nitmes
                   1062:  * with the low water mark.
                   1063:  *
1.21      thorpej  1064:  * Note 1, we never wait for memory here, we let the caller decide what to do.
1.20      thorpej  1065:  *
                   1066:  * Note 2, this doesn't work with static pools.
                   1067:  *
                   1068:  * Note 3, we must be called with the pool already locked, and we return
                   1069:  * with it locked.
                   1070:  */
                   1071: static int
                   1072: pool_catchup(pp)
                   1073:        struct pool *pp;
                   1074: {
                   1075:        caddr_t cp;
                   1076:        int error = 0;
                   1077:
                   1078:        if (pp->pr_roflags & PR_STATIC) {
                   1079:                /*
                   1080:                 * We dropped below the low water mark, and this is not a
                   1081:                 * good thing.  Log a warning.
1.21      thorpej  1082:                 *
                   1083:                 * XXX: rate-limit this?
1.20      thorpej  1084:                 */
                   1085:                printf("WARNING: static pool `%s' dropped below low water "
                   1086:                    "mark\n", pp->pr_wchan);
                   1087:                return (0);
                   1088:        }
                   1089:
1.21      thorpej  1090:        while (pp->pr_nitems < pp->pr_minitems) {
1.20      thorpej  1091:                /*
1.21      thorpej  1092:                 * Call the page back-end allocator for more memory.
                   1093:                 *
                   1094:                 * XXX: We never wait, so should we bother unlocking
                   1095:                 * the pool descriptor?
1.20      thorpej  1096:                 */
1.21      thorpej  1097:                simple_unlock(&pp->pr_slock);
1.20      thorpej  1098:                cp = (*pp->pr_alloc)(pp->pr_pagesz, 0, pp->pr_mtype);
1.21      thorpej  1099:                simple_lock(&pp->pr_slock);
1.20      thorpej  1100:                if (cp == NULL) {
                   1101:                        error = ENOMEM;
                   1102:                        break;
                   1103:                }
1.26    ! thorpej  1104:                pp->pr_npagealloc++;
1.20      thorpej  1105:                pool_prime_page(pp, cp);
                   1106:        }
                   1107:
                   1108:        return (error);
                   1109: }
                   1110:
1.3       pk       1111: void
                   1112: pool_setlowat(pp, n)
                   1113:        pool_handle_t   pp;
                   1114:        int n;
                   1115: {
1.20      thorpej  1116:        int error;
1.15      pk       1117:
1.21      thorpej  1118:        simple_lock(&pp->pr_slock);
                   1119:
1.3       pk       1120:        pp->pr_minitems = n;
1.15      pk       1121:        pp->pr_minpages = (n == 0)
                   1122:                ? 0
1.18      thorpej  1123:                : roundup(n, pp->pr_itemsperpage) / pp->pr_itemsperpage;
1.20      thorpej  1124:
                   1125:        /* Make sure we're caught up with the newly-set low water mark. */
1.21      thorpej  1126:        if ((error = pool_catchup(pp)) != 0) {
1.20      thorpej  1127:                /*
                   1128:                 * XXX: Should we log a warning?  Should we set up a timeout
                   1129:                 * to try again in a second or so?  The latter could break
                   1130:                 * a caller's assumptions about interrupt protection, etc.
                   1131:                 */
                   1132:        }
1.21      thorpej  1133:
                   1134:        simple_unlock(&pp->pr_slock);
1.3       pk       1135: }
                   1136:
                   1137: void
                   1138: pool_sethiwat(pp, n)
                   1139:        pool_handle_t   pp;
                   1140:        int n;
                   1141: {
1.15      pk       1142:
1.21      thorpej  1143:        simple_lock(&pp->pr_slock);
                   1144:
1.15      pk       1145:        pp->pr_maxpages = (n == 0)
                   1146:                ? 0
1.18      thorpej  1147:                : roundup(n, pp->pr_itemsperpage) / pp->pr_itemsperpage;
1.21      thorpej  1148:
                   1149:        simple_unlock(&pp->pr_slock);
1.3       pk       1150: }
                   1151:
1.20      thorpej  1152: void
                   1153: pool_sethardlimit(pp, n, warnmess, ratecap)
                   1154:        pool_handle_t pp;
                   1155:        int n;
                   1156:        const char *warnmess;
                   1157:        int ratecap;
                   1158: {
                   1159:
1.21      thorpej  1160:        simple_lock(&pp->pr_slock);
1.20      thorpej  1161:
                   1162:        pp->pr_hardlimit = n;
                   1163:        pp->pr_hardlimit_warning = warnmess;
                   1164:        pp->pr_hardlimit_ratecap = ratecap;
                   1165:        memset(&pp->pr_hardlimit_warning_last, 0,
                   1166:            sizeof(pp->pr_hardlimit_warning_last));
                   1167:
                   1168:        /*
1.21      thorpej  1169:         * In-line version of pool_sethiwat(), because we don't want to
                   1170:         * release the lock.
1.20      thorpej  1171:         */
                   1172:        pp->pr_maxpages = (n == 0)
                   1173:                ? 0
                   1174:                : roundup(n, pp->pr_itemsperpage) / pp->pr_itemsperpage;
1.21      thorpej  1175:
                   1176:        simple_unlock(&pp->pr_slock);
1.20      thorpej  1177: }
1.3       pk       1178:
                   1179: /*
                   1180:  * Default page allocator.
                   1181:  */
                   1182: static void *
                   1183: pool_page_alloc(sz, flags, mtype)
                   1184:        unsigned long sz;
                   1185:        int flags;
                   1186:        int mtype;
                   1187: {
1.11      thorpej  1188:        boolean_t waitok = (flags & PR_WAITOK) ? TRUE : FALSE;
1.3       pk       1189:
1.11      thorpej  1190:        return ((void *)uvm_km_alloc_poolpage(waitok));
1.3       pk       1191: }
                   1192:
                   1193: static void
                   1194: pool_page_free(v, sz, mtype)
                   1195:        void *v;
                   1196:        unsigned long sz;
                   1197:        int mtype;
                   1198: {
                   1199:
1.10      eeh      1200:        uvm_km_free_poolpage((vaddr_t)v);
1.3       pk       1201: }
1.12      thorpej  1202:
                   1203: /*
                   1204:  * Alternate pool page allocator for pools that know they will
                   1205:  * never be accessed in interrupt context.
                   1206:  */
                   1207: void *
                   1208: pool_page_alloc_nointr(sz, flags, mtype)
                   1209:        unsigned long sz;
                   1210:        int flags;
                   1211:        int mtype;
                   1212: {
                   1213:        boolean_t waitok = (flags & PR_WAITOK) ? TRUE : FALSE;
                   1214:
                   1215:        return ((void *)uvm_km_alloc_poolpage1(kernel_map, uvm.kernel_object,
                   1216:            waitok));
                   1217: }
                   1218:
                   1219: void
                   1220: pool_page_free_nointr(v, sz, mtype)
                   1221:        void *v;
                   1222:        unsigned long sz;
                   1223:        int mtype;
                   1224: {
                   1225:
                   1226:        uvm_km_free_poolpage1(kernel_map, (vaddr_t)v);
                   1227: }
                   1228:
1.3       pk       1229:
                   1230: /*
                   1231:  * Release all complete pages that have not been used recently.
                   1232:  */
                   1233: void
1.25      thorpej  1234: _pool_reclaim(pp, file, line)
1.3       pk       1235:        pool_handle_t pp;
1.25      thorpej  1236:        const char *file;
                   1237:        long line;
1.3       pk       1238: {
                   1239:        struct pool_item_header *ph, *phnext;
1.21      thorpej  1240:        struct timeval curtime;
                   1241:        int s;
1.3       pk       1242:
1.20      thorpej  1243:        if (pp->pr_roflags & PR_STATIC)
1.3       pk       1244:                return;
                   1245:
1.21      thorpej  1246:        if (simple_lock_try(&pp->pr_slock) == 0)
1.3       pk       1247:                return;
1.25      thorpej  1248:        pr_enter(pp, file, line);
1.3       pk       1249:
1.21      thorpej  1250:        s = splclock();
                   1251:        curtime = mono_time;
                   1252:        splx(s);
                   1253:
1.3       pk       1254:        for (ph = TAILQ_FIRST(&pp->pr_pagelist); ph != NULL; ph = phnext) {
                   1255:                phnext = TAILQ_NEXT(ph, ph_pagelist);
                   1256:
                   1257:                /* Check our minimum page claim */
                   1258:                if (pp->pr_npages <= pp->pr_minpages)
                   1259:                        break;
                   1260:
                   1261:                if (ph->ph_nmissing == 0) {
                   1262:                        struct timeval diff;
                   1263:                        timersub(&curtime, &ph->ph_time, &diff);
                   1264:                        if (diff.tv_sec < pool_inactive_time)
                   1265:                                continue;
1.21      thorpej  1266:
                   1267:                        /*
                   1268:                         * If freeing this page would put us below
                   1269:                         * the low water mark, stop now.
                   1270:                         */
                   1271:                        if ((pp->pr_nitems - pp->pr_itemsperpage) <
                   1272:                            pp->pr_minitems)
                   1273:                                break;
                   1274:
1.3       pk       1275:                        pr_rmpage(pp, ph);
                   1276:                }
                   1277:        }
                   1278:
1.25      thorpej  1279:        pr_leave(pp);
1.21      thorpej  1280:        simple_unlock(&pp->pr_slock);
1.3       pk       1281: }
                   1282:
                   1283:
                   1284: /*
                   1285:  * Drain pools, one at a time.
1.21      thorpej  1286:  *
                   1287:  * Note, we must never be called from an interrupt context.
1.3       pk       1288:  */
                   1289: void
                   1290: pool_drain(arg)
                   1291:        void *arg;
                   1292: {
                   1293:        struct pool *pp;
1.23      thorpej  1294:        int s;
1.3       pk       1295:
1.23      thorpej  1296:        s = splimp();
                   1297:        simple_lock(&pool_head_slock);
                   1298:
                   1299:        if (drainpp == NULL && (drainpp = TAILQ_FIRST(&pool_head)) == NULL)
                   1300:                goto out;
1.3       pk       1301:
                   1302:        pp = drainpp;
                   1303:        drainpp = TAILQ_NEXT(pp, pr_poollist);
                   1304:
                   1305:        pool_reclaim(pp);
1.23      thorpej  1306:
                   1307:  out:
                   1308:        simple_unlock(&pool_head_slock);
1.3       pk       1309:        splx(s);
                   1310: }
                   1311:
                   1312:
                   1313: /*
                   1314:  * Diagnostic helpers.
                   1315:  */
                   1316: void
1.25      thorpej  1317: pool_print(pp, modif)
1.3       pk       1318:        struct pool *pp;
1.25      thorpej  1319:        const char *modif;
1.21      thorpej  1320: {
                   1321:        int s;
                   1322:
                   1323:        s = splimp();
1.25      thorpej  1324:        if (simple_lock_try(&pp->pr_slock) == 0) {
                   1325:                printf("pool %s is locked; try again later\n",
                   1326:                    pp->pr_wchan);
                   1327:                splx(s);
                   1328:                return;
                   1329:        }
                   1330:        pool_print1(pp, modif, printf);
1.21      thorpej  1331:        simple_unlock(&pp->pr_slock);
                   1332:        splx(s);
                   1333: }
                   1334:
1.25      thorpej  1335: void
                   1336: pool_printit(pp, modif, pr)
                   1337:        struct pool *pp;
                   1338:        const char *modif;
                   1339:        void (*pr) __P((const char *, ...));
                   1340: {
                   1341:        int didlock = 0;
                   1342:
                   1343:        if (pp == NULL) {
                   1344:                (*pr)("Must specify a pool to print.\n");
                   1345:                return;
                   1346:        }
                   1347:
                   1348:        /*
                   1349:         * Called from DDB; interrupts should be blocked, and all
                   1350:         * other processors should be paused.  We can skip locking
                   1351:         * the pool in this case.
                   1352:         *
                   1353:         * We do a simple_lock_try() just to print the lock
                   1354:         * status, however.
                   1355:         */
                   1356:
                   1357:        if (simple_lock_try(&pp->pr_slock) == 0)
                   1358:                (*pr)("WARNING: pool %s is locked\n", pp->pr_wchan);
                   1359:        else
                   1360:                didlock = 1;
                   1361:
                   1362:        pool_print1(pp, modif, pr);
                   1363:
                   1364:        if (didlock)
                   1365:                simple_unlock(&pp->pr_slock);
                   1366: }
                   1367:
1.21      thorpej  1368: static void
1.25      thorpej  1369: pool_print1(pp, modif, pr)
1.21      thorpej  1370:        struct pool *pp;
1.25      thorpej  1371:        const char *modif;
                   1372:        void (*pr) __P((const char *, ...));
1.3       pk       1373: {
1.25      thorpej  1374:        struct pool_item_header *ph;
                   1375: #ifdef DIAGNOSTIC
                   1376:        struct pool_item *pi;
                   1377: #endif
                   1378:        int print_log = 0, print_pagelist = 0;
                   1379:        char c;
                   1380:
                   1381:        while ((c = *modif++) != '\0') {
                   1382:                if (c == 'l')
                   1383:                        print_log = 1;
                   1384:                if (c == 'p')
                   1385:                        print_pagelist = 1;
                   1386:                modif++;
                   1387:        }
                   1388:
                   1389:        (*pr)("POOL %s: size %u, align %u, ioff %u, roflags 0x%08x\n",
                   1390:            pp->pr_wchan, pp->pr_size, pp->pr_align, pp->pr_itemoffset,
                   1391:            pp->pr_roflags);
                   1392:        (*pr)("\tpagesz %u, mtype %d\n", pp->pr_pagesz, pp->pr_mtype);
                   1393:        (*pr)("\talloc %p, release %p\n", pp->pr_alloc, pp->pr_free);
                   1394:        (*pr)("\tminitems %u, minpages %u, maxpages %u, npages %u\n",
                   1395:            pp->pr_minitems, pp->pr_minpages, pp->pr_maxpages, pp->pr_npages);
                   1396:        (*pr)("\titemsperpage %u, nitems %u, nout %u, hardlimit %u\n",
                   1397:            pp->pr_itemsperpage, pp->pr_nitems, pp->pr_nout, pp->pr_hardlimit);
                   1398:
                   1399:        (*pr)("\n\tnget %lu, nfail %lu, nput %lu\n",
                   1400:            pp->pr_nget, pp->pr_nfail, pp->pr_nput);
                   1401:        (*pr)("\tnpagealloc %lu, npagefree %lu, hiwat %u, nidle %lu\n",
                   1402:            pp->pr_npagealloc, pp->pr_npagefree, pp->pr_hiwat, pp->pr_nidle);
                   1403:
                   1404:        if (print_pagelist == 0)
                   1405:                goto skip_pagelist;
                   1406:
                   1407:        if ((ph = TAILQ_FIRST(&pp->pr_pagelist)) != NULL)
                   1408:                (*pr)("\n\tpage list:\n");
                   1409:        for (; ph != NULL; ph = TAILQ_NEXT(ph, ph_pagelist)) {
                   1410:                (*pr)("\t\tpage %p, nmissing %d, time %lu,%lu\n",
                   1411:                    ph->ph_page, ph->ph_nmissing,
                   1412:                    (u_long)ph->ph_time.tv_sec,
                   1413:                    (u_long)ph->ph_time.tv_usec);
                   1414: #ifdef DIAGNOSTIC
                   1415:                for (pi = TAILQ_FIRST(&ph->ph_itemlist); pi != NULL;
                   1416:                     pi = TAILQ_NEXT(pi, pi_list)) {
                   1417:                        if (pi->pi_magic != PI_MAGIC) {
                   1418:                                (*pr)("\t\t\titem %p, magic 0x%x\n",
                   1419:                                    pi, pi->pi_magic);
                   1420:                        }
                   1421:                }
                   1422: #endif
                   1423:        }
                   1424:        if (pp->pr_curpage == NULL)
                   1425:                (*pr)("\tno current page\n");
                   1426:        else
                   1427:                (*pr)("\tcurpage %p\n", pp->pr_curpage->ph_page);
                   1428:
                   1429:  skip_pagelist:
                   1430:
                   1431:        if (print_log == 0)
                   1432:                goto skip_log;
                   1433:
                   1434:        (*pr)("\n");
                   1435:        if ((pp->pr_roflags & PR_LOGGING) == 0)
                   1436:                (*pr)("\tno log\n");
                   1437:        else
                   1438:                pr_printlog(pp, NULL, pr);
1.3       pk       1439:
1.25      thorpej  1440:  skip_log:
1.3       pk       1441:
1.25      thorpej  1442:        pr_enter_check(pp, pr);
1.3       pk       1443: }
                   1444:
                   1445: int
                   1446: pool_chk(pp, label)
                   1447:        struct pool *pp;
                   1448:        char *label;
                   1449: {
                   1450:        struct pool_item_header *ph;
                   1451:        int r = 0;
                   1452:
1.21      thorpej  1453:        simple_lock(&pp->pr_slock);
1.3       pk       1454:
                   1455:        for (ph = TAILQ_FIRST(&pp->pr_pagelist); ph != NULL;
                   1456:             ph = TAILQ_NEXT(ph, ph_pagelist)) {
                   1457:
                   1458:                struct pool_item *pi;
                   1459:                int n;
                   1460:                caddr_t page;
                   1461:
                   1462:                page = (caddr_t)((u_long)ph & pp->pr_pagemask);
1.20      thorpej  1463:                if (page != ph->ph_page &&
                   1464:                    (pp->pr_roflags & PR_PHINPAGE) != 0) {
1.3       pk       1465:                        if (label != NULL)
                   1466:                                printf("%s: ", label);
1.16      briggs   1467:                        printf("pool(%p:%s): page inconsistency: page %p;"
                   1468:                               " at page head addr %p (p %p)\n", pp,
1.3       pk       1469:                                pp->pr_wchan, ph->ph_page,
                   1470:                                ph, page);
                   1471:                        r++;
                   1472:                        goto out;
                   1473:                }
                   1474:
                   1475:                for (pi = TAILQ_FIRST(&ph->ph_itemlist), n = 0;
                   1476:                     pi != NULL;
                   1477:                     pi = TAILQ_NEXT(pi,pi_list), n++) {
                   1478:
                   1479: #ifdef DIAGNOSTIC
                   1480:                        if (pi->pi_magic != PI_MAGIC) {
                   1481:                                if (label != NULL)
                   1482:                                        printf("%s: ", label);
                   1483:                                printf("pool(%s): free list modified: magic=%x;"
                   1484:                                       " page %p; item ordinal %d;"
                   1485:                                       " addr %p (p %p)\n",
                   1486:                                        pp->pr_wchan, pi->pi_magic, ph->ph_page,
                   1487:                                        n, pi, page);
                   1488:                                panic("pool");
                   1489:                        }
                   1490: #endif
                   1491:                        page = (caddr_t)((u_long)pi & pp->pr_pagemask);
                   1492:                        if (page == ph->ph_page)
                   1493:                                continue;
                   1494:
                   1495:                        if (label != NULL)
                   1496:                                printf("%s: ", label);
1.16      briggs   1497:                        printf("pool(%p:%s): page inconsistency: page %p;"
                   1498:                               " item ordinal %d; addr %p (p %p)\n", pp,
1.3       pk       1499:                                pp->pr_wchan, ph->ph_page,
                   1500:                                n, pi, page);
                   1501:                        r++;
                   1502:                        goto out;
                   1503:                }
                   1504:        }
                   1505: out:
1.21      thorpej  1506:        simple_unlock(&pp->pr_slock);
1.3       pk       1507:        return (r);
                   1508: }

CVSweb <webmaster@jp.NetBSD.org>