[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 1.21.2.2.2.2

1.21.2.2.2.2! chs         1: /*     $NetBSD: subr_pool.c,v 1.21.2.2.2.1 1999/06/21 01:24:03 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:  */
                     39:
1.21.2.2.2.1  thorpej    40: #include "opt_pool.h"
                     41: #include "opt_poollog.h"
                     42:
1.1       pk         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()) */
1.21.2.2  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;
1.21.2.2.2.1  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:
1.21.2.2.2.1  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:
1.21.2.2.2.1  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;
1.21.2.2.2.1  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:
1.21.2.2.2.1  thorpej   141: #ifdef DIAGNOSTIC
1.3       pk        142: static void    pr_log __P((struct pool *, void *, int, const char *, long));
1.21.2.2.2.1  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
1.21.2.2.2.1  thorpej   179: pr_printlog(pp, pi, pr)
1.3       pk        180:        struct pool *pp;
1.21.2.2.2.1  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) {
1.21.2.2.2.1  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: }
                    209:
1.21.2.2.2.1  thorpej   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: }
                    253: #else
                    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:
1.21.2.1  chs       315:        if ((pp->pr_roflags & PR_PHINPAGE) == 0) {
1.21.2.2.2.1  thorpej   316:                int s;
1.21.2.1  chs       317:                LIST_REMOVE(ph, ph_hashlist);
1.21.2.2.2.1  thorpej   318:                s = splhigh();
1.21.2.1  chs       319:                pool_put(&phpool, ph);
1.21.2.2.2.1  thorpej   320:                splx(s);
1.21.2.1  chs       321:        }
                    322:
1.3       pk        323:        if (pp->pr_curpage == ph) {
                    324:                /*
                    325:                 * Find a new non-empty page header, if any.
                    326:                 * Start search from the page head, to increase the
                    327:                 * chance for "high water" pages to be freed.
                    328:                 */
                    329:                for (ph = TAILQ_FIRST(&pp->pr_pagelist); ph != NULL;
                    330:                     ph = TAILQ_NEXT(ph, ph_pagelist))
                    331:                        if (TAILQ_FIRST(&ph->ph_itemlist) != NULL)
                    332:                                break;
                    333:
                    334:                pp->pr_curpage = ph;
1.21      thorpej   335:        }
1.3       pk        336: }
                    337:
                    338: /*
                    339:  * Allocate and initialize a pool.
                    340:  */
1.1       pk        341: struct pool *
1.3       pk        342: pool_create(size, align, ioff, nitems, wchan, pagesz, alloc, release, mtype)
1.1       pk        343:        size_t  size;
1.3       pk        344:        u_int   align;
                    345:        u_int   ioff;
1.1       pk        346:        int     nitems;
1.21      thorpej   347:        const char *wchan;
1.3       pk        348:        size_t  pagesz;
                    349:        void    *(*alloc) __P((unsigned long, int, int));
                    350:        void    (*release) __P((void *, unsigned long, int));
1.1       pk        351:        int     mtype;
                    352: {
                    353:        struct pool *pp;
1.3       pk        354:        int flags;
1.1       pk        355:
1.3       pk        356:        pp = (struct pool *)malloc(sizeof(*pp), M_POOL, M_NOWAIT);
                    357:        if (pp == NULL)
1.1       pk        358:                return (NULL);
1.3       pk        359:
                    360:        flags = PR_FREEHEADER;
                    361:        pool_init(pp, size, align, ioff, flags, wchan, pagesz,
                    362:                  alloc, release, mtype);
                    363:
                    364:        if (nitems != 0) {
                    365:                if (pool_prime(pp, nitems, NULL) != 0) {
                    366:                        pool_destroy(pp);
                    367:                        return (NULL);
                    368:                }
1.1       pk        369:        }
                    370:
1.3       pk        371:        return (pp);
                    372: }
                    373:
                    374: /*
                    375:  * Initialize the given pool resource structure.
                    376:  *
                    377:  * We export this routine to allow other kernel parts to declare
                    378:  * static pools that must be initialized before malloc() is available.
                    379:  */
                    380: void
                    381: pool_init(pp, size, align, ioff, flags, wchan, pagesz, alloc, release, mtype)
                    382:        struct pool     *pp;
                    383:        size_t          size;
                    384:        u_int           align;
                    385:        u_int           ioff;
                    386:        int             flags;
1.21      thorpej   387:        const char      *wchan;
1.3       pk        388:        size_t          pagesz;
                    389:        void            *(*alloc) __P((unsigned long, int, int));
                    390:        void            (*release) __P((void *, unsigned long, int));
                    391:        int             mtype;
                    392: {
1.16      briggs    393:        int off, slack, i;
1.3       pk        394:
1.21.2.2.2.1  thorpej   395: #ifdef POOL_DIAGNOSTIC
                    396:        /*
                    397:         * Always log if POOL_DIAGNOSTIC is defined.
                    398:         */
                    399:        if (pool_logsize != 0)
                    400:                flags |= PR_LOGGING;
                    401: #endif
                    402:
1.3       pk        403:        /*
                    404:         * Check arguments and construct default values.
                    405:         */
                    406:        if (!powerof2(pagesz) || pagesz > PAGE_SIZE)
                    407:                panic("pool_init: page size invalid (%lx)\n", (u_long)pagesz);
                    408:
1.4       thorpej   409:        if (alloc == NULL && release == NULL) {
1.3       pk        410:                alloc = pool_page_alloc;
                    411:                release = pool_page_free;
1.4       thorpej   412:                pagesz = PAGE_SIZE;     /* Rounds to PAGE_SIZE anyhow. */
                    413:        } else if ((alloc != NULL && release != NULL) == 0) {
                    414:                /* If you specifiy one, must specify both. */
                    415:                panic("pool_init: must specify alloc and release together");
                    416:        }
                    417:
1.3       pk        418:        if (pagesz == 0)
                    419:                pagesz = PAGE_SIZE;
                    420:
                    421:        if (align == 0)
                    422:                align = ALIGN(1);
1.14      thorpej   423:
                    424:        if (size < sizeof(struct pool_item))
                    425:                size = sizeof(struct pool_item);
1.3       pk        426:
                    427:        /*
                    428:         * Initialize the pool structure.
                    429:         */
                    430:        TAILQ_INIT(&pp->pr_pagelist);
                    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;
1.3       pk        438:        pp->pr_size = ALIGN(size);
                    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;
                    451:        pp->pr_hardlimit_ratecap = 0;
                    452:        memset(&pp->pr_hardlimit_warning_last, 0,
                    453:            sizeof(pp->pr_hardlimit_warning_last));
1.3       pk        454:
                    455:        /*
                    456:         * Decide whether to put the page header off page to avoid
                    457:         * wasting too large a part of the page. Off-page page headers
                    458:         * go on a hash table, so we can match a returned item
                    459:         * with its header based on the page address.
                    460:         * We use 1/16 of the page size as the threshold (XXX: tune)
                    461:         */
                    462:        if (pp->pr_size < pagesz/16) {
                    463:                /* Use the end of the page for the page header */
1.20      thorpej   464:                pp->pr_roflags |= PR_PHINPAGE;
1.3       pk        465:                pp->pr_phoffset = off =
                    466:                        pagesz - ALIGN(sizeof(struct pool_item_header));
1.2       pk        467:        } else {
1.3       pk        468:                /* The page header will be taken from our page header pool */
                    469:                pp->pr_phoffset = 0;
                    470:                off = pagesz;
1.16      briggs    471:                for (i = 0; i < PR_HASHTABSIZE; i++) {
                    472:                        LIST_INIT(&pp->pr_hashtab[i]);
                    473:                }
1.2       pk        474:        }
1.1       pk        475:
1.3       pk        476:        /*
                    477:         * Alignment is to take place at `ioff' within the item. This means
                    478:         * we must reserve up to `align - 1' bytes on the page to allow
                    479:         * appropriate positioning of each item.
                    480:         *
                    481:         * Silently enforce `0 <= ioff < align'.
                    482:         */
                    483:        pp->pr_itemoffset = ioff = ioff % align;
                    484:        pp->pr_itemsperpage = (off - ((align - ioff) % align)) / pp->pr_size;
                    485:
                    486:        /*
                    487:         * Use the slack between the chunks and the page header
                    488:         * for "cache coloring".
                    489:         */
                    490:        slack = off - pp->pr_itemsperpage * pp->pr_size;
                    491:        pp->pr_maxcolor = (slack / align) * align;
                    492:        pp->pr_curcolor = 0;
                    493:
                    494:        pp->pr_nget = 0;
                    495:        pp->pr_nfail = 0;
                    496:        pp->pr_nput = 0;
                    497:        pp->pr_npagealloc = 0;
                    498:        pp->pr_npagefree = 0;
1.1       pk        499:        pp->pr_hiwat = 0;
1.8       thorpej   500:        pp->pr_nidle = 0;
1.3       pk        501:
1.21.2.2.2.1  thorpej   502:        if (flags & PR_LOGGING) {
                    503:                if (kmem_map == NULL ||
                    504:                    (pp->pr_log = malloc(pool_logsize * sizeof(struct pool_log),
                    505:                     M_TEMP, M_NOWAIT)) == NULL)
1.20      thorpej   506:                        pp->pr_roflags &= ~PR_LOGGING;
1.3       pk        507:                pp->pr_curlogentry = 0;
                    508:                pp->pr_logsize = pool_logsize;
                    509:        }
1.21.2.2.2.1  thorpej   510:
                    511:        pp->pr_entered_file = NULL;
                    512:        pp->pr_entered_line = 0;
1.3       pk        513:
1.21      thorpej   514:        simple_lock_init(&pp->pr_slock);
1.1       pk        515:
1.3       pk        516:        /*
                    517:         * Initialize private page header pool if we haven't done so yet.
1.21.2.2  thorpej   518:         * XXX LOCKING.
1.3       pk        519:         */
                    520:        if (phpool.pr_size == 0) {
                    521:                pool_init(&phpool, sizeof(struct pool_item_header), 0, 0,
                    522:                          0, "phpool", 0, 0, 0, 0);
1.1       pk        523:        }
                    524:
1.21.2.2  thorpej   525:        /* Insert into the list of all pools. */
                    526:        simple_lock(&pool_head_slock);
                    527:        TAILQ_INSERT_TAIL(&pool_head, pp, pr_poollist);
                    528:        simple_unlock(&pool_head_slock);
1.1       pk        529: }
                    530:
                    531: /*
                    532:  * De-commision a pool resource.
                    533:  */
                    534: void
                    535: pool_destroy(pp)
                    536:        struct pool *pp;
                    537: {
1.3       pk        538:        struct pool_item_header *ph;
                    539:
                    540: #ifdef DIAGNOSTIC
1.20      thorpej   541:        if (pp->pr_nout != 0) {
1.21.2.2.2.1  thorpej   542:                pr_printlog(pp, NULL, printf);
1.20      thorpej   543:                panic("pool_destroy: pool busy: still out: %u\n",
                    544:                    pp->pr_nout);
1.3       pk        545:        }
                    546: #endif
1.1       pk        547:
1.3       pk        548:        /* Remove all pages */
1.20      thorpej   549:        if ((pp->pr_roflags & PR_STATIC) == 0)
1.3       pk        550:                while ((ph = pp->pr_pagelist.tqh_first) != NULL)
                    551:                        pr_rmpage(pp, ph);
                    552:
                    553:        /* Remove from global pool list */
1.21.2.2  thorpej   554:        simple_lock(&pool_head_slock);
1.3       pk        555:        TAILQ_REMOVE(&pool_head, pp, pr_poollist);
1.21.2.2  thorpej   556:        /* XXX Only clear this if we were drainpp? */
1.3       pk        557:        drainpp = NULL;
1.21.2.2  thorpej   558:        simple_unlock(&pool_head_slock);
1.3       pk        559:
1.20      thorpej   560:        if ((pp->pr_roflags & PR_LOGGING) != 0)
1.3       pk        561:                free(pp->pr_log, M_TEMP);
1.2       pk        562:
1.20      thorpej   563:        if (pp->pr_roflags & PR_FREEHEADER)
1.3       pk        564:                free(pp, M_POOL);
1.1       pk        565: }
                    566:
                    567:
                    568: /*
1.3       pk        569:  * Grab an item from the pool; must be called at appropriate spl level
1.1       pk        570:  */
1.3       pk        571: void *
                    572: _pool_get(pp, flags, file, line)
                    573:        struct pool *pp;
                    574:        int flags;
                    575:        const char *file;
                    576:        long line;
1.1       pk        577: {
                    578:        void *v;
                    579:        struct pool_item *pi;
1.3       pk        580:        struct pool_item_header *ph;
1.1       pk        581:
1.2       pk        582: #ifdef DIAGNOSTIC
1.20      thorpej   583:        if ((pp->pr_roflags & PR_STATIC) && (flags & PR_MALLOCOK)) {
1.21.2.2.2.1  thorpej   584:                pr_printlog(pp, NULL, printf);
1.2       pk        585:                panic("pool_get: static");
1.3       pk        586:        }
1.2       pk        587: #endif
                    588:
1.3       pk        589:        if (curproc == NULL && (flags & PR_WAITOK) != 0)
                    590:                panic("pool_get: must have NOWAIT");
1.1       pk        591:
1.21      thorpej   592:        simple_lock(&pp->pr_slock);
1.21.2.2.2.1  thorpej   593:        pr_enter(pp, file, line);
1.20      thorpej   594:
                    595:  startover:
                    596:        /*
                    597:         * Check to see if we've reached the hard limit.  If we have,
                    598:         * and we can wait, then wait until an item has been returned to
                    599:         * the pool.
                    600:         */
                    601: #ifdef DIAGNOSTIC
                    602:        if (pp->pr_nout > pp->pr_hardlimit) {
1.21.2.2.2.1  thorpej   603:                pr_leave(pp);
1.21      thorpej   604:                simple_unlock(&pp->pr_slock);
1.20      thorpej   605:                panic("pool_get: %s: crossed hard limit", pp->pr_wchan);
                    606:        }
                    607: #endif
                    608:        if (pp->pr_nout == pp->pr_hardlimit) {
                    609:                if (flags & PR_WAITOK) {
                    610:                        /*
                    611:                         * XXX: A warning isn't logged in this case.  Should
                    612:                         * it be?
                    613:                         */
                    614:                        pp->pr_flags |= PR_WANTED;
1.21.2.2.2.1  thorpej   615:                        pr_leave(pp);
1.21      thorpej   616:                        simple_unlock(&pp->pr_slock);
1.20      thorpej   617:                        tsleep((caddr_t)pp, PSWP, pp->pr_wchan, 0);
1.21      thorpej   618:                        simple_lock(&pp->pr_slock);
1.21.2.2.2.1  thorpej   619:                        pr_enter(pp, file, line);
1.20      thorpej   620:                        goto startover;
                    621:                }
                    622:                if (pp->pr_hardlimit_warning != NULL) {
                    623:                        /*
                    624:                         * Log a message that the hard limit has been hit.
                    625:                         */
                    626:                        struct timeval curtime, logdiff;
                    627:                        int s = splclock();
                    628:                        curtime = mono_time;
                    629:                        splx(s);
                    630:                        timersub(&curtime, &pp->pr_hardlimit_warning_last,
                    631:                            &logdiff);
                    632:                        if (logdiff.tv_sec >= pp->pr_hardlimit_ratecap) {
                    633:                                pp->pr_hardlimit_warning_last = curtime;
                    634:                                log(LOG_ERR, "%s\n", pp->pr_hardlimit_warning);
                    635:                        }
                    636:                }
1.21      thorpej   637:
                    638:                if (flags & PR_URGENT)
                    639:                        panic("pool_get: urgent");
                    640:
                    641:                pp->pr_nfail++;
                    642:
1.21.2.2.2.1  thorpej   643:                pr_leave(pp);
1.21      thorpej   644:                simple_unlock(&pp->pr_slock);
1.20      thorpej   645:                return (NULL);
                    646:        }
                    647:
1.3       pk        648:        /*
                    649:         * The convention we use is that if `curpage' is not NULL, then
                    650:         * it points at a non-empty bucket. In particular, `curpage'
                    651:         * never points at a page header which has PR_PHINPAGE set and
                    652:         * has no items in its bucket.
                    653:         */
1.20      thorpej   654:        if ((ph = pp->pr_curpage) == NULL) {
1.15      pk        655:                void *v;
                    656:
1.20      thorpej   657: #ifdef DIAGNOSTIC
                    658:                if (pp->pr_nitems != 0) {
1.21      thorpej   659:                        simple_unlock(&pp->pr_slock);
1.20      thorpej   660:                        printf("pool_get: %s: curpage NULL, nitems %u\n",
                    661:                            pp->pr_wchan, pp->pr_nitems);
                    662:                        panic("pool_get: nitems inconsistent\n");
                    663:                }
                    664: #endif
                    665:
1.21      thorpej   666:                /*
                    667:                 * Call the back-end page allocator for more memory.
                    668:                 * Release the pool lock, as the back-end page allocator
                    669:                 * may block.
                    670:                 */
1.21.2.2.2.1  thorpej   671:                pr_leave(pp);
1.21      thorpej   672:                simple_unlock(&pp->pr_slock);
                    673:                v = (*pp->pr_alloc)(pp->pr_pagesz, flags, pp->pr_mtype);
                    674:                simple_lock(&pp->pr_slock);
1.21.2.2.2.1  thorpej   675:                pr_enter(pp, file, line);
1.15      pk        676:
1.21      thorpej   677:                if (v == NULL) {
                    678:                        /*
                    679:                         * We were unable to allocate a page, but
                    680:                         * we released the lock during allocation,
                    681:                         * so perhaps items were freed back to the
                    682:                         * pool.  Check for this case.
                    683:                         */
                    684:                        if (pp->pr_curpage != NULL)
                    685:                                goto startover;
1.15      pk        686:
1.3       pk        687:                        if (flags & PR_URGENT)
                    688:                                panic("pool_get: urgent");
1.21      thorpej   689:
1.3       pk        690:                        if ((flags & PR_WAITOK) == 0) {
                    691:                                pp->pr_nfail++;
1.21.2.2.2.1  thorpej   692:                                pr_leave(pp);
1.21      thorpej   693:                                simple_unlock(&pp->pr_slock);
1.1       pk        694:                                return (NULL);
1.3       pk        695:                        }
                    696:
1.15      pk        697:                        /*
                    698:                         * Wait for items to be returned to this pool.
1.21      thorpej   699:                         *
1.15      pk        700:                         * XXX: we actually want to wait just until
                    701:                         * the page allocator has memory again. Depending
                    702:                         * on this pool's usage, we might get stuck here
                    703:                         * for a long time.
1.20      thorpej   704:                         *
                    705:                         * XXX: maybe we should wake up once a second and
                    706:                         * try again?
1.15      pk        707:                         */
1.1       pk        708:                        pp->pr_flags |= PR_WANTED;
1.21.2.2.2.1  thorpej   709:                        pr_leave(pp);
1.21      thorpej   710:                        simple_unlock(&pp->pr_slock);
1.1       pk        711:                        tsleep((caddr_t)pp, PSWP, pp->pr_wchan, 0);
1.21      thorpej   712:                        simple_lock(&pp->pr_slock);
1.21.2.2.2.1  thorpej   713:                        pr_enter(pp, file, line);
1.20      thorpej   714:                        goto startover;
1.1       pk        715:                }
1.3       pk        716:
1.15      pk        717:                /* We have more memory; add it to the pool */
                    718:                pp->pr_npagealloc++;
                    719:                pool_prime_page(pp, v);
                    720:
1.20      thorpej   721:                /* Start the allocation process over. */
                    722:                goto startover;
1.3       pk        723:        }
                    724:
1.21      thorpej   725:        if ((v = pi = TAILQ_FIRST(&ph->ph_itemlist)) == NULL) {
1.21.2.2.2.1  thorpej   726:                pr_leave(pp);
1.21      thorpej   727:                simple_unlock(&pp->pr_slock);
1.3       pk        728:                panic("pool_get: %s: page empty", pp->pr_wchan);
1.21      thorpej   729:        }
1.20      thorpej   730: #ifdef DIAGNOSTIC
                    731:        if (pp->pr_nitems == 0) {
1.21.2.2.2.1  thorpej   732:                pr_leave(pp);
1.21      thorpej   733:                simple_unlock(&pp->pr_slock);
1.20      thorpej   734:                printf("pool_get: %s: items on itemlist, nitems %u\n",
                    735:                    pp->pr_wchan, pp->pr_nitems);
                    736:                panic("pool_get: nitems inconsistent\n");
                    737:        }
                    738: #endif
1.3       pk        739:        pr_log(pp, v, PRLOG_GET, file, line);
                    740:
                    741: #ifdef DIAGNOSTIC
                    742:        if (pi->pi_magic != PI_MAGIC) {
1.21.2.2.2.1  thorpej   743:                pr_printlog(pp, pi, printf);
1.3       pk        744:                panic("pool_get(%s): free list modified: magic=%x; page %p;"
                    745:                       " item addr %p\n",
                    746:                        pp->pr_wchan, pi->pi_magic, ph->ph_page, pi);
                    747:        }
                    748: #endif
                    749:
                    750:        /*
                    751:         * Remove from item list.
                    752:         */
                    753:        TAILQ_REMOVE(&ph->ph_itemlist, pi, pi_list);
1.20      thorpej   754:        pp->pr_nitems--;
                    755:        pp->pr_nout++;
1.6       thorpej   756:        if (ph->ph_nmissing == 0) {
                    757: #ifdef DIAGNOSTIC
                    758:                if (pp->pr_nidle == 0)
                    759:                        panic("pool_get: nidle inconsistent");
                    760: #endif
                    761:                pp->pr_nidle--;
                    762:        }
1.3       pk        763:        ph->ph_nmissing++;
                    764:        if (TAILQ_FIRST(&ph->ph_itemlist) == NULL) {
1.21      thorpej   765: #ifdef DIAGNOSTIC
                    766:                if (ph->ph_nmissing != pp->pr_itemsperpage) {
1.21.2.2.2.1  thorpej   767:                        pr_leave(pp);
1.21      thorpej   768:                        simple_unlock(&pp->pr_slock);
                    769:                        panic("pool_get: %s: nmissing inconsistent",
                    770:                            pp->pr_wchan);
                    771:                }
                    772: #endif
1.3       pk        773:                /*
                    774:                 * Find a new non-empty page header, if any.
                    775:                 * Start search from the page head, to increase
                    776:                 * the chance for "high water" pages to be freed.
                    777:                 *
1.21      thorpej   778:                 * Migrate empty pages to the end of the list.  This
                    779:                 * will speed the update of curpage as pages become
                    780:                 * idle.  Empty pages intermingled with idle pages
                    781:                 * is no big deal.  As soon as a page becomes un-empty,
                    782:                 * it will move back to the head of the list.
1.3       pk        783:                 */
                    784:                TAILQ_REMOVE(&pp->pr_pagelist, ph, ph_pagelist);
1.21      thorpej   785:                TAILQ_INSERT_TAIL(&pp->pr_pagelist, ph, ph_pagelist);
                    786:                for (ph = TAILQ_FIRST(&pp->pr_pagelist); ph != NULL;
                    787:                     ph = TAILQ_NEXT(ph, ph_pagelist))
1.3       pk        788:                        if (TAILQ_FIRST(&ph->ph_itemlist) != NULL)
                    789:                                break;
                    790:
                    791:                pp->pr_curpage = ph;
1.1       pk        792:        }
1.3       pk        793:
                    794:        pp->pr_nget++;
1.20      thorpej   795:
                    796:        /*
                    797:         * If we have a low water mark and we are now below that low
                    798:         * water mark, add more items to the pool.
                    799:         */
                    800:        if (pp->pr_nitems < pp->pr_minitems && pool_catchup(pp) != 0) {
                    801:                /*
                    802:                 * XXX: Should we log a warning?  Should we set up a timeout
                    803:                 * to try again in a second or so?  The latter could break
                    804:                 * a caller's assumptions about interrupt protection, etc.
                    805:                 */
                    806:        }
                    807:
1.21.2.2.2.1  thorpej   808:        pr_leave(pp);
1.21      thorpej   809:        simple_unlock(&pp->pr_slock);
1.1       pk        810:        return (v);
                    811: }
                    812:
                    813: /*
1.3       pk        814:  * Return resource to the pool; must be called at appropriate spl level
1.1       pk        815:  */
1.3       pk        816: void
                    817: _pool_put(pp, v, file, line)
                    818:        struct pool *pp;
                    819:        void *v;
                    820:        const char *file;
                    821:        long line;
1.1       pk        822: {
                    823:        struct pool_item *pi = v;
1.3       pk        824:        struct pool_item_header *ph;
                    825:        caddr_t page;
1.21      thorpej   826:        int s;
1.3       pk        827:
                    828:        page = (caddr_t)((u_long)v & pp->pr_pagemask);
1.1       pk        829:
1.21      thorpej   830:        simple_lock(&pp->pr_slock);
1.21.2.2.2.1  thorpej   831:        pr_enter(pp, file, line);
1.3       pk        832:
                    833:        pr_log(pp, v, PRLOG_PUT, file, line);
                    834:
                    835:        if ((ph = pr_find_pagehead(pp, page)) == NULL) {
1.21.2.2.2.1  thorpej   836:                pr_printlog(pp, NULL, printf);
1.3       pk        837:                panic("pool_put: %s: page header missing", pp->pr_wchan);
                    838:        }
                    839:
                    840:        /*
                    841:         * Return to item list.
                    842:         */
1.2       pk        843: #ifdef DIAGNOSTIC
1.3       pk        844:        pi->pi_magic = PI_MAGIC;
1.21.2.2.2.2! chs       845: #endif
        !           846: #ifdef DEBUG
        !           847:        {
        !           848:                int i, *ip = v;
        !           849:
        !           850:                for (i = 0; i < pp->pr_size / 4; i++) {
        !           851:                        *ip++ = PI_MAGIC;
        !           852:                }
        !           853:        }
1.3       pk        854: #endif
                    855:        TAILQ_INSERT_HEAD(&ph->ph_itemlist, pi, pi_list);
                    856:        ph->ph_nmissing--;
                    857:        pp->pr_nput++;
1.20      thorpej   858:        pp->pr_nitems++;
                    859:        pp->pr_nout--;
1.3       pk        860:
                    861:        /* Cancel "pool empty" condition if it exists */
                    862:        if (pp->pr_curpage == NULL)
                    863:                pp->pr_curpage = ph;
                    864:
                    865:        if (pp->pr_flags & PR_WANTED) {
                    866:                pp->pr_flags &= ~PR_WANTED;
1.15      pk        867:                if (ph->ph_nmissing == 0)
                    868:                        pp->pr_nidle++;
1.21.2.2.2.1  thorpej   869:                pr_leave(pp);
1.21      thorpej   870:                simple_unlock(&pp->pr_slock);
1.3       pk        871:                wakeup((caddr_t)pp);
                    872:                return;
                    873:        }
                    874:
                    875:        /*
1.21      thorpej   876:         * If this page is now complete, do one of two things:
                    877:         *
                    878:         *      (1) If we have more pages than the page high water
                    879:         *          mark, free the page back to the system.
                    880:         *
                    881:         *      (2) Move it to the end of the page list, so that
                    882:         *          we minimize our chances of fragmenting the
                    883:         *          pool.  Idle pages migrate to the end (along with
                    884:         *          completely empty pages, so that we find un-empty
                    885:         *          pages more quickly when we update curpage) of the
                    886:         *          list so they can be more easily swept up by
                    887:         *          the pagedaemon when pages are scarce.
1.3       pk        888:         */
                    889:        if (ph->ph_nmissing == 0) {
1.6       thorpej   890:                pp->pr_nidle++;
1.3       pk        891:                if (pp->pr_npages > pp->pr_maxpages) {
                    892:                        pr_rmpage(pp, ph);
                    893:                } else {
                    894:                        TAILQ_REMOVE(&pp->pr_pagelist, ph, ph_pagelist);
                    895:                        TAILQ_INSERT_TAIL(&pp->pr_pagelist, ph, ph_pagelist);
                    896:
1.21      thorpej   897:                        /*
                    898:                         * Update the timestamp on the page.  A page must
                    899:                         * be idle for some period of time before it can
                    900:                         * be reclaimed by the pagedaemon.  This minimizes
                    901:                         * ping-pong'ing for memory.
                    902:                         */
                    903:                        s = splclock();
                    904:                        ph->ph_time = mono_time;
                    905:                        splx(s);
                    906:
                    907:                        /*
                    908:                         * Update the current page pointer.  Just look for
                    909:                         * the first page with any free items.
                    910:                         *
                    911:                         * XXX: Maybe we want an option to look for the
                    912:                         * page with the fewest available items, to minimize
                    913:                         * fragmentation?
                    914:                         */
1.3       pk        915:                        for (ph = TAILQ_FIRST(&pp->pr_pagelist); ph != NULL;
                    916:                             ph = TAILQ_NEXT(ph, ph_pagelist))
                    917:                                if (TAILQ_FIRST(&ph->ph_itemlist) != NULL)
                    918:                                        break;
1.1       pk        919:
1.3       pk        920:                        pp->pr_curpage = ph;
1.1       pk        921:                }
                    922:        }
1.21      thorpej   923:        /*
                    924:         * If the page has just become un-empty, move it to the head of
                    925:         * the list, and make it the current page.  The next allocation
                    926:         * will get the item from this page, instead of further fragmenting
                    927:         * the pool.
                    928:         */
                    929:        else if (ph->ph_nmissing == (pp->pr_itemsperpage - 1)) {
                    930:                TAILQ_REMOVE(&pp->pr_pagelist, ph, ph_pagelist);
                    931:                TAILQ_INSERT_HEAD(&pp->pr_pagelist, ph, ph_pagelist);
                    932:                pp->pr_curpage = ph;
                    933:        }
                    934:
1.21.2.2.2.1  thorpej   935:        pr_leave(pp);
1.21      thorpej   936:        simple_unlock(&pp->pr_slock);
1.3       pk        937:
1.1       pk        938: }
                    939:
                    940: /*
1.3       pk        941:  * Add N items to the pool.
1.1       pk        942:  */
                    943: int
1.2       pk        944: pool_prime(pp, n, storage)
1.1       pk        945:        struct pool *pp;
                    946:        int n;
1.2       pk        947:        caddr_t storage;
1.1       pk        948: {
1.3       pk        949:        caddr_t cp;
                    950:        int newnitems, newpages;
1.2       pk        951:
                    952: #ifdef DIAGNOSTIC
1.20      thorpej   953:        if (storage && !(pp->pr_roflags & PR_STATIC))
1.2       pk        954:                panic("pool_prime: static");
                    955:        /* !storage && static caught below */
                    956: #endif
1.1       pk        957:
1.21      thorpej   958:        simple_lock(&pp->pr_slock);
                    959:
1.3       pk        960:        newnitems = pp->pr_minitems + n;
                    961:        newpages =
1.18      thorpej   962:                roundup(newnitems, pp->pr_itemsperpage) / pp->pr_itemsperpage
1.3       pk        963:                - pp->pr_minpages;
                    964:
                    965:        while (newpages-- > 0) {
1.20      thorpej   966:                if (pp->pr_roflags & PR_STATIC) {
1.3       pk        967:                        cp = storage;
                    968:                        storage += pp->pr_pagesz;
                    969:                } else {
1.21      thorpej   970:                        simple_unlock(&pp->pr_slock);
1.3       pk        971:                        cp = (*pp->pr_alloc)(pp->pr_pagesz, 0, pp->pr_mtype);
1.21      thorpej   972:                        simple_lock(&pp->pr_slock);
1.3       pk        973:                }
1.2       pk        974:
1.3       pk        975:                if (cp == NULL) {
1.21      thorpej   976:                        simple_unlock(&pp->pr_slock);
1.1       pk        977:                        return (ENOMEM);
                    978:                }
                    979:
1.21.2.2.2.1  thorpej   980:                pp->pr_npagealloc++;
1.3       pk        981:                pool_prime_page(pp, cp);
                    982:                pp->pr_minpages++;
1.1       pk        983:        }
1.3       pk        984:
                    985:        pp->pr_minitems = newnitems;
                    986:
                    987:        if (pp->pr_minpages >= pp->pr_maxpages)
                    988:                pp->pr_maxpages = pp->pr_minpages + 1;  /* XXX */
                    989:
1.21      thorpej   990:        simple_unlock(&pp->pr_slock);
1.1       pk        991:        return (0);
                    992: }
1.3       pk        993:
                    994: /*
                    995:  * Add a page worth of items to the pool.
1.21      thorpej   996:  *
                    997:  * Note, we must be called with the pool descriptor LOCKED.
1.3       pk        998:  */
1.21      thorpej   999: static void
1.3       pk       1000: pool_prime_page(pp, storage)
                   1001:        struct pool *pp;
                   1002:        caddr_t storage;
                   1003: {
                   1004:        struct pool_item *pi;
                   1005:        struct pool_item_header *ph;
                   1006:        caddr_t cp = storage;
                   1007:        unsigned int align = pp->pr_align;
                   1008:        unsigned int ioff = pp->pr_itemoffset;
1.21.2.2.2.1  thorpej  1009:        int s, n;
1.3       pk       1010:
1.20      thorpej  1011:        if ((pp->pr_roflags & PR_PHINPAGE) != 0) {
1.3       pk       1012:                ph = (struct pool_item_header *)(cp + pp->pr_phoffset);
                   1013:        } else {
1.21.2.2.2.1  thorpej  1014:                s = splhigh();
1.3       pk       1015:                ph = pool_get(&phpool, PR_URGENT);
1.21.2.2.2.1  thorpej  1016:                splx(s);
1.3       pk       1017:                LIST_INSERT_HEAD(&pp->pr_hashtab[PR_HASH_INDEX(pp, cp)],
                   1018:                                 ph, ph_hashlist);
                   1019:        }
                   1020:
                   1021:        /*
                   1022:         * Insert page header.
                   1023:         */
                   1024:        TAILQ_INSERT_HEAD(&pp->pr_pagelist, ph, ph_pagelist);
                   1025:        TAILQ_INIT(&ph->ph_itemlist);
                   1026:        ph->ph_page = storage;
                   1027:        ph->ph_nmissing = 0;
1.21      thorpej  1028:        memset(&ph->ph_time, 0, sizeof(ph->ph_time));
1.3       pk       1029:
1.6       thorpej  1030:        pp->pr_nidle++;
                   1031:
1.3       pk       1032:        /*
                   1033:         * Color this page.
                   1034:         */
                   1035:        cp = (caddr_t)(cp + pp->pr_curcolor);
                   1036:        if ((pp->pr_curcolor += align) > pp->pr_maxcolor)
                   1037:                pp->pr_curcolor = 0;
                   1038:
                   1039:        /*
                   1040:         * Adjust storage to apply aligment to `pr_itemoffset' in each item.
                   1041:         */
                   1042:        if (ioff != 0)
                   1043:                cp = (caddr_t)(cp + (align - ioff));
                   1044:
                   1045:        /*
                   1046:         * Insert remaining chunks on the bucket list.
                   1047:         */
                   1048:        n = pp->pr_itemsperpage;
1.20      thorpej  1049:        pp->pr_nitems += n;
1.3       pk       1050:
                   1051:        while (n--) {
                   1052:                pi = (struct pool_item *)cp;
                   1053:
                   1054:                /* Insert on page list */
                   1055:                TAILQ_INSERT_TAIL(&ph->ph_itemlist, pi, pi_list);
                   1056: #ifdef DIAGNOSTIC
                   1057:                pi->pi_magic = PI_MAGIC;
                   1058: #endif
                   1059:                cp = (caddr_t)(cp + pp->pr_size);
                   1060:        }
                   1061:
                   1062:        /*
                   1063:         * If the pool was depleted, point at the new page.
                   1064:         */
                   1065:        if (pp->pr_curpage == NULL)
                   1066:                pp->pr_curpage = ph;
                   1067:
                   1068:        if (++pp->pr_npages > pp->pr_hiwat)
                   1069:                pp->pr_hiwat = pp->pr_npages;
                   1070: }
                   1071:
1.20      thorpej  1072: /*
                   1073:  * Like pool_prime(), except this is used by pool_get() when nitems
                   1074:  * drops below the low water mark.  This is used to catch up nitmes
                   1075:  * with the low water mark.
                   1076:  *
1.21      thorpej  1077:  * Note 1, we never wait for memory here, we let the caller decide what to do.
1.20      thorpej  1078:  *
                   1079:  * Note 2, this doesn't work with static pools.
                   1080:  *
                   1081:  * Note 3, we must be called with the pool already locked, and we return
                   1082:  * with it locked.
                   1083:  */
                   1084: static int
                   1085: pool_catchup(pp)
                   1086:        struct pool *pp;
                   1087: {
                   1088:        caddr_t cp;
                   1089:        int error = 0;
                   1090:
                   1091:        if (pp->pr_roflags & PR_STATIC) {
                   1092:                /*
                   1093:                 * We dropped below the low water mark, and this is not a
                   1094:                 * good thing.  Log a warning.
1.21      thorpej  1095:                 *
                   1096:                 * XXX: rate-limit this?
1.20      thorpej  1097:                 */
                   1098:                printf("WARNING: static pool `%s' dropped below low water "
                   1099:                    "mark\n", pp->pr_wchan);
                   1100:                return (0);
                   1101:        }
                   1102:
1.21      thorpej  1103:        while (pp->pr_nitems < pp->pr_minitems) {
1.20      thorpej  1104:                /*
1.21      thorpej  1105:                 * Call the page back-end allocator for more memory.
                   1106:                 *
                   1107:                 * XXX: We never wait, so should we bother unlocking
                   1108:                 * the pool descriptor?
1.20      thorpej  1109:                 */
1.21      thorpej  1110:                simple_unlock(&pp->pr_slock);
1.20      thorpej  1111:                cp = (*pp->pr_alloc)(pp->pr_pagesz, 0, pp->pr_mtype);
1.21      thorpej  1112:                simple_lock(&pp->pr_slock);
1.20      thorpej  1113:                if (cp == NULL) {
                   1114:                        error = ENOMEM;
                   1115:                        break;
                   1116:                }
1.21.2.2.2.1  thorpej  1117:                pp->pr_npagealloc++;
1.20      thorpej  1118:                pool_prime_page(pp, cp);
                   1119:        }
                   1120:
                   1121:        return (error);
                   1122: }
                   1123:
1.3       pk       1124: void
                   1125: pool_setlowat(pp, n)
                   1126:        pool_handle_t   pp;
                   1127:        int n;
                   1128: {
1.20      thorpej  1129:        int error;
1.15      pk       1130:
1.21      thorpej  1131:        simple_lock(&pp->pr_slock);
                   1132:
1.3       pk       1133:        pp->pr_minitems = n;
1.15      pk       1134:        pp->pr_minpages = (n == 0)
                   1135:                ? 0
1.18      thorpej  1136:                : roundup(n, pp->pr_itemsperpage) / pp->pr_itemsperpage;
1.20      thorpej  1137:
                   1138:        /* Make sure we're caught up with the newly-set low water mark. */
1.21      thorpej  1139:        if ((error = pool_catchup(pp)) != 0) {
1.20      thorpej  1140:                /*
                   1141:                 * XXX: Should we log a warning?  Should we set up a timeout
                   1142:                 * to try again in a second or so?  The latter could break
                   1143:                 * a caller's assumptions about interrupt protection, etc.
                   1144:                 */
                   1145:        }
1.21      thorpej  1146:
                   1147:        simple_unlock(&pp->pr_slock);
1.3       pk       1148: }
                   1149:
                   1150: void
                   1151: pool_sethiwat(pp, n)
                   1152:        pool_handle_t   pp;
                   1153:        int n;
                   1154: {
1.15      pk       1155:
1.21      thorpej  1156:        simple_lock(&pp->pr_slock);
                   1157:
1.15      pk       1158:        pp->pr_maxpages = (n == 0)
                   1159:                ? 0
1.18      thorpej  1160:                : roundup(n, pp->pr_itemsperpage) / pp->pr_itemsperpage;
1.21      thorpej  1161:
                   1162:        simple_unlock(&pp->pr_slock);
1.3       pk       1163: }
                   1164:
1.20      thorpej  1165: void
                   1166: pool_sethardlimit(pp, n, warnmess, ratecap)
                   1167:        pool_handle_t pp;
                   1168:        int n;
                   1169:        const char *warnmess;
                   1170:        int ratecap;
                   1171: {
                   1172:
1.21      thorpej  1173:        simple_lock(&pp->pr_slock);
1.20      thorpej  1174:
                   1175:        pp->pr_hardlimit = n;
                   1176:        pp->pr_hardlimit_warning = warnmess;
                   1177:        pp->pr_hardlimit_ratecap = ratecap;
                   1178:        memset(&pp->pr_hardlimit_warning_last, 0,
                   1179:            sizeof(pp->pr_hardlimit_warning_last));
                   1180:
                   1181:        /*
1.21      thorpej  1182:         * In-line version of pool_sethiwat(), because we don't want to
                   1183:         * release the lock.
1.20      thorpej  1184:         */
                   1185:        pp->pr_maxpages = (n == 0)
                   1186:                ? 0
                   1187:                : roundup(n, pp->pr_itemsperpage) / pp->pr_itemsperpage;
1.21      thorpej  1188:
                   1189:        simple_unlock(&pp->pr_slock);
1.20      thorpej  1190: }
1.3       pk       1191:
                   1192: /*
                   1193:  * Default page allocator.
                   1194:  */
                   1195: static void *
                   1196: pool_page_alloc(sz, flags, mtype)
                   1197:        unsigned long sz;
                   1198:        int flags;
                   1199:        int mtype;
                   1200: {
1.11      thorpej  1201:        boolean_t waitok = (flags & PR_WAITOK) ? TRUE : FALSE;
1.3       pk       1202:
1.11      thorpej  1203:        return ((void *)uvm_km_alloc_poolpage(waitok));
1.3       pk       1204: }
                   1205:
                   1206: static void
                   1207: pool_page_free(v, sz, mtype)
                   1208:        void *v;
                   1209:        unsigned long sz;
                   1210:        int mtype;
                   1211: {
                   1212:
1.10      eeh      1213:        uvm_km_free_poolpage((vaddr_t)v);
1.3       pk       1214: }
1.12      thorpej  1215:
                   1216: /*
                   1217:  * Alternate pool page allocator for pools that know they will
                   1218:  * never be accessed in interrupt context.
                   1219:  */
                   1220: void *
                   1221: pool_page_alloc_nointr(sz, flags, mtype)
                   1222:        unsigned long sz;
                   1223:        int flags;
                   1224:        int mtype;
                   1225: {
                   1226:        boolean_t waitok = (flags & PR_WAITOK) ? TRUE : FALSE;
                   1227:
                   1228:        return ((void *)uvm_km_alloc_poolpage1(kernel_map, uvm.kernel_object,
                   1229:            waitok));
                   1230: }
                   1231:
                   1232: void
                   1233: pool_page_free_nointr(v, sz, mtype)
                   1234:        void *v;
                   1235:        unsigned long sz;
                   1236:        int mtype;
                   1237: {
                   1238:
                   1239:        uvm_km_free_poolpage1(kernel_map, (vaddr_t)v);
                   1240: }
                   1241:
1.3       pk       1242:
                   1243: /*
                   1244:  * Release all complete pages that have not been used recently.
                   1245:  */
                   1246: void
1.21.2.2.2.1  thorpej  1247: _pool_reclaim(pp, file, line)
1.3       pk       1248:        pool_handle_t pp;
1.21.2.2.2.1  thorpej  1249:        const char *file;
                   1250:        long line;
1.3       pk       1251: {
                   1252:        struct pool_item_header *ph, *phnext;
1.21      thorpej  1253:        struct timeval curtime;
                   1254:        int s;
1.3       pk       1255:
1.20      thorpej  1256:        if (pp->pr_roflags & PR_STATIC)
1.3       pk       1257:                return;
                   1258:
1.21      thorpej  1259:        if (simple_lock_try(&pp->pr_slock) == 0)
1.3       pk       1260:                return;
1.21.2.2.2.1  thorpej  1261:        pr_enter(pp, file, line);
1.3       pk       1262:
1.21      thorpej  1263:        s = splclock();
                   1264:        curtime = mono_time;
                   1265:        splx(s);
                   1266:
1.3       pk       1267:        for (ph = TAILQ_FIRST(&pp->pr_pagelist); ph != NULL; ph = phnext) {
                   1268:                phnext = TAILQ_NEXT(ph, ph_pagelist);
                   1269:
                   1270:                /* Check our minimum page claim */
                   1271:                if (pp->pr_npages <= pp->pr_minpages)
                   1272:                        break;
                   1273:
                   1274:                if (ph->ph_nmissing == 0) {
                   1275:                        struct timeval diff;
                   1276:                        timersub(&curtime, &ph->ph_time, &diff);
                   1277:                        if (diff.tv_sec < pool_inactive_time)
                   1278:                                continue;
1.21      thorpej  1279:
                   1280:                        /*
                   1281:                         * If freeing this page would put us below
                   1282:                         * the low water mark, stop now.
                   1283:                         */
                   1284:                        if ((pp->pr_nitems - pp->pr_itemsperpage) <
                   1285:                            pp->pr_minitems)
                   1286:                                break;
                   1287:
1.3       pk       1288:                        pr_rmpage(pp, ph);
                   1289:                }
                   1290:        }
                   1291:
1.21.2.2.2.1  thorpej  1292:        pr_leave(pp);
1.21      thorpej  1293:        simple_unlock(&pp->pr_slock);
1.3       pk       1294: }
                   1295:
                   1296:
                   1297: /*
                   1298:  * Drain pools, one at a time.
1.21      thorpej  1299:  *
                   1300:  * Note, we must never be called from an interrupt context.
1.3       pk       1301:  */
                   1302: void
                   1303: pool_drain(arg)
                   1304:        void *arg;
                   1305: {
                   1306:        struct pool *pp;
1.21.2.2  thorpej  1307:        int s;
1.3       pk       1308:
1.21.2.2  thorpej  1309:        s = splimp();
                   1310:        simple_lock(&pool_head_slock);
                   1311:
                   1312:        if (drainpp == NULL && (drainpp = TAILQ_FIRST(&pool_head)) == NULL)
                   1313:                goto out;
1.3       pk       1314:
                   1315:        pp = drainpp;
                   1316:        drainpp = TAILQ_NEXT(pp, pr_poollist);
                   1317:
                   1318:        pool_reclaim(pp);
1.21.2.2  thorpej  1319:
                   1320:  out:
                   1321:        simple_unlock(&pool_head_slock);
1.3       pk       1322:        splx(s);
                   1323: }
                   1324:
                   1325:
                   1326: /*
                   1327:  * Diagnostic helpers.
                   1328:  */
                   1329: void
1.21.2.2.2.1  thorpej  1330: pool_print(pp, modif)
1.3       pk       1331:        struct pool *pp;
1.21.2.2.2.1  thorpej  1332:        const char *modif;
1.21      thorpej  1333: {
                   1334:        int s;
                   1335:
                   1336:        s = splimp();
1.21.2.2.2.1  thorpej  1337:        if (simple_lock_try(&pp->pr_slock) == 0) {
                   1338:                printf("pool %s is locked; try again later\n",
                   1339:                    pp->pr_wchan);
                   1340:                splx(s);
                   1341:                return;
                   1342:        }
                   1343:        pool_print1(pp, modif, printf);
1.21      thorpej  1344:        simple_unlock(&pp->pr_slock);
                   1345:        splx(s);
                   1346: }
                   1347:
1.21.2.2.2.1  thorpej  1348: void
                   1349: pool_printit(pp, modif, pr)
                   1350:        struct pool *pp;
                   1351:        const char *modif;
                   1352:        void (*pr) __P((const char *, ...));
                   1353: {
                   1354:        int didlock = 0;
                   1355:
                   1356:        if (pp == NULL) {
                   1357:                (*pr)("Must specify a pool to print.\n");
                   1358:                return;
                   1359:        }
                   1360:
                   1361:        /*
                   1362:         * Called from DDB; interrupts should be blocked, and all
                   1363:         * other processors should be paused.  We can skip locking
                   1364:         * the pool in this case.
                   1365:         *
                   1366:         * We do a simple_lock_try() just to print the lock
                   1367:         * status, however.
                   1368:         */
                   1369:
                   1370:        if (simple_lock_try(&pp->pr_slock) == 0)
                   1371:                (*pr)("WARNING: pool %s is locked\n", pp->pr_wchan);
                   1372:        else
                   1373:                didlock = 1;
                   1374:
                   1375:        pool_print1(pp, modif, pr);
                   1376:
                   1377:        if (didlock)
                   1378:                simple_unlock(&pp->pr_slock);
                   1379: }
                   1380:
1.21      thorpej  1381: static void
1.21.2.2.2.1  thorpej  1382: pool_print1(pp, modif, pr)
1.21      thorpej  1383:        struct pool *pp;
1.21.2.2.2.1  thorpej  1384:        const char *modif;
                   1385:        void (*pr) __P((const char *, ...));
1.3       pk       1386: {
1.21.2.2.2.1  thorpej  1387:        struct pool_item_header *ph;
                   1388: #ifdef DIAGNOSTIC
                   1389:        struct pool_item *pi;
                   1390: #endif
                   1391:        int print_log = 0, print_pagelist = 0;
                   1392:        char c;
                   1393:
                   1394:        while ((c = *modif++) != '\0') {
                   1395:                if (c == 'l')
                   1396:                        print_log = 1;
                   1397:                if (c == 'p')
                   1398:                        print_pagelist = 1;
                   1399:                modif++;
                   1400:        }
                   1401:
                   1402:        (*pr)("POOL %s: size %u, align %u, ioff %u, roflags 0x%08x\n",
                   1403:            pp->pr_wchan, pp->pr_size, pp->pr_align, pp->pr_itemoffset,
                   1404:            pp->pr_roflags);
                   1405:        (*pr)("\tpagesz %u, mtype %d\n", pp->pr_pagesz, pp->pr_mtype);
                   1406:        (*pr)("\talloc %p, release %p\n", pp->pr_alloc, pp->pr_free);
                   1407:        (*pr)("\tminitems %u, minpages %u, maxpages %u, npages %u\n",
                   1408:            pp->pr_minitems, pp->pr_minpages, pp->pr_maxpages, pp->pr_npages);
                   1409:        (*pr)("\titemsperpage %u, nitems %u, nout %u, hardlimit %u\n",
                   1410:            pp->pr_itemsperpage, pp->pr_nitems, pp->pr_nout, pp->pr_hardlimit);
                   1411:
                   1412:        (*pr)("\n\tnget %lu, nfail %lu, nput %lu\n",
                   1413:            pp->pr_nget, pp->pr_nfail, pp->pr_nput);
                   1414:        (*pr)("\tnpagealloc %lu, npagefree %lu, hiwat %u, nidle %lu\n",
                   1415:            pp->pr_npagealloc, pp->pr_npagefree, pp->pr_hiwat, pp->pr_nidle);
                   1416:
                   1417:        if (print_pagelist == 0)
                   1418:                goto skip_pagelist;
                   1419:
                   1420:        if ((ph = TAILQ_FIRST(&pp->pr_pagelist)) != NULL)
                   1421:                (*pr)("\n\tpage list:\n");
                   1422:        for (; ph != NULL; ph = TAILQ_NEXT(ph, ph_pagelist)) {
                   1423:                (*pr)("\t\tpage %p, nmissing %d, time %lu,%lu\n",
                   1424:                    ph->ph_page, ph->ph_nmissing,
                   1425:                    (u_long)ph->ph_time.tv_sec,
                   1426:                    (u_long)ph->ph_time.tv_usec);
                   1427: #ifdef DIAGNOSTIC
                   1428:                for (pi = TAILQ_FIRST(&ph->ph_itemlist); pi != NULL;
                   1429:                     pi = TAILQ_NEXT(pi, pi_list)) {
                   1430:                        if (pi->pi_magic != PI_MAGIC) {
                   1431:                                (*pr)("\t\t\titem %p, magic 0x%x\n",
                   1432:                                    pi, pi->pi_magic);
                   1433:                        }
                   1434:                }
                   1435: #endif
                   1436:        }
                   1437:        if (pp->pr_curpage == NULL)
                   1438:                (*pr)("\tno current page\n");
                   1439:        else
                   1440:                (*pr)("\tcurpage %p\n", pp->pr_curpage->ph_page);
                   1441:
                   1442:  skip_pagelist:
                   1443:
                   1444:        if (print_log == 0)
                   1445:                goto skip_log;
                   1446:
                   1447:        (*pr)("\n");
                   1448:        if ((pp->pr_roflags & PR_LOGGING) == 0)
                   1449:                (*pr)("\tno log\n");
                   1450:        else
                   1451:                pr_printlog(pp, NULL, pr);
1.3       pk       1452:
1.21.2.2.2.1  thorpej  1453:  skip_log:
1.3       pk       1454:
1.21.2.2.2.1  thorpej  1455:        pr_enter_check(pp, pr);
1.3       pk       1456: }
                   1457:
                   1458: int
                   1459: pool_chk(pp, label)
                   1460:        struct pool *pp;
                   1461:        char *label;
                   1462: {
                   1463:        struct pool_item_header *ph;
                   1464:        int r = 0;
                   1465:
1.21      thorpej  1466:        simple_lock(&pp->pr_slock);
1.3       pk       1467:
                   1468:        for (ph = TAILQ_FIRST(&pp->pr_pagelist); ph != NULL;
                   1469:             ph = TAILQ_NEXT(ph, ph_pagelist)) {
                   1470:
                   1471:                struct pool_item *pi;
                   1472:                int n;
                   1473:                caddr_t page;
                   1474:
                   1475:                page = (caddr_t)((u_long)ph & pp->pr_pagemask);
1.20      thorpej  1476:                if (page != ph->ph_page &&
                   1477:                    (pp->pr_roflags & PR_PHINPAGE) != 0) {
1.3       pk       1478:                        if (label != NULL)
                   1479:                                printf("%s: ", label);
1.16      briggs   1480:                        printf("pool(%p:%s): page inconsistency: page %p;"
                   1481:                               " at page head addr %p (p %p)\n", pp,
1.3       pk       1482:                                pp->pr_wchan, ph->ph_page,
                   1483:                                ph, page);
                   1484:                        r++;
                   1485:                        goto out;
                   1486:                }
                   1487:
                   1488:                for (pi = TAILQ_FIRST(&ph->ph_itemlist), n = 0;
                   1489:                     pi != NULL;
                   1490:                     pi = TAILQ_NEXT(pi,pi_list), n++) {
                   1491:
                   1492: #ifdef DIAGNOSTIC
                   1493:                        if (pi->pi_magic != PI_MAGIC) {
                   1494:                                if (label != NULL)
                   1495:                                        printf("%s: ", label);
                   1496:                                printf("pool(%s): free list modified: magic=%x;"
                   1497:                                       " page %p; item ordinal %d;"
                   1498:                                       " addr %p (p %p)\n",
                   1499:                                        pp->pr_wchan, pi->pi_magic, ph->ph_page,
                   1500:                                        n, pi, page);
                   1501:                                panic("pool");
                   1502:                        }
                   1503: #endif
                   1504:                        page = (caddr_t)((u_long)pi & pp->pr_pagemask);
                   1505:                        if (page == ph->ph_page)
                   1506:                                continue;
                   1507:
                   1508:                        if (label != NULL)
                   1509:                                printf("%s: ", label);
1.16      briggs   1510:                        printf("pool(%p:%s): page inconsistency: page %p;"
                   1511:                               " item ordinal %d; addr %p (p %p)\n", pp,
1.3       pk       1512:                                pp->pr_wchan, ph->ph_page,
                   1513:                                n, pi, page);
                   1514:                        r++;
                   1515:                        goto out;
                   1516:                }
                   1517:        }
                   1518: out:
1.21      thorpej  1519:        simple_unlock(&pp->pr_slock);
1.3       pk       1520:        return (r);
                   1521: }

CVSweb <webmaster@jp.NetBSD.org>