[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 1.34

1.34    ! thorpej     1: /*     $NetBSD: subr_pool.c,v 1.33 2000/04/13 00:44:19 chs 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"
1.28      thorpej    42: #include "opt_lockdebug.h"
1.1       pk         43:
                     44: #include <sys/param.h>
                     45: #include <sys/systm.h>
                     46: #include <sys/proc.h>
                     47: #include <sys/errno.h>
                     48: #include <sys/kernel.h>
                     49: #include <sys/malloc.h>
                     50: #include <sys/lock.h>
                     51: #include <sys/pool.h>
1.20      thorpej    52: #include <sys/syslog.h>
1.1       pk         53:
1.3       pk         54: #include <vm/vm.h>
                     55: #include <vm/vm_kern.h>
                     56:
                     57: #include <uvm/uvm.h>
                     58:
1.1       pk         59: /*
                     60:  * Pool resource management utility.
1.3       pk         61:  *
                     62:  * Memory is allocated in pages which are split into pieces according
                     63:  * to the pool item size. Each page is kept on a list headed by `pr_pagelist'
                     64:  * in the pool structure and the individual pool items are on a linked list
                     65:  * headed by `ph_itemlist' in each page header. The memory for building
                     66:  * the page list is either taken from the allocated pages themselves (for
                     67:  * small pool items) or taken from an internal pool of page headers (`phpool').
1.1       pk         68:  */
                     69:
1.3       pk         70: /* List of all pools */
1.5       thorpej    71: TAILQ_HEAD(,pool) pool_head = TAILQ_HEAD_INITIALIZER(pool_head);
1.3       pk         72:
                     73: /* Private pool for page header structures */
                     74: static struct pool phpool;
                     75:
                     76: /* # of seconds to retain page after last use */
                     77: int pool_inactive_time = 10;
                     78:
                     79: /* Next candidate for drainage (see pool_drain()) */
1.23      thorpej    80: static struct pool     *drainpp;
                     81:
                     82: /* This spin lock protects both pool_head and drainpp. */
                     83: struct simplelock pool_head_slock = SIMPLELOCK_INITIALIZER;
1.3       pk         84:
                     85: struct pool_item_header {
                     86:        /* Page headers */
                     87:        TAILQ_ENTRY(pool_item_header)
                     88:                                ph_pagelist;    /* pool page list */
                     89:        TAILQ_HEAD(,pool_item)  ph_itemlist;    /* chunk list for this page */
                     90:        LIST_ENTRY(pool_item_header)
                     91:                                ph_hashlist;    /* Off-page page headers */
                     92:        int                     ph_nmissing;    /* # of chunks in use */
                     93:        caddr_t                 ph_page;        /* this page's address */
                     94:        struct timeval          ph_time;        /* last referenced */
                     95: };
                     96:
1.1       pk         97: struct pool_item {
1.3       pk         98: #ifdef DIAGNOSTIC
                     99:        int pi_magic;
1.33      chs       100: #endif
1.25      thorpej   101: #define        PI_MAGIC 0xdeadbeef
1.3       pk        102:        /* Other entries use only this list entry */
                    103:        TAILQ_ENTRY(pool_item)  pi_list;
                    104: };
                    105:
                    106:
1.25      thorpej   107: #define        PR_HASH_INDEX(pp,addr) \
1.3       pk        108:        (((u_long)(addr) >> (pp)->pr_pageshift) & (PR_HASHTABSIZE - 1))
                    109:
                    110:
                    111:
                    112: static struct pool_item_header
                    113:                *pr_find_pagehead __P((struct pool *, caddr_t));
                    114: static void    pr_rmpage __P((struct pool *, struct pool_item_header *));
1.20      thorpej   115: static int     pool_catchup __P((struct pool *));
1.21      thorpej   116: static void    pool_prime_page __P((struct pool *, caddr_t));
1.3       pk        117: static void    *pool_page_alloc __P((unsigned long, int, int));
                    118: static void    pool_page_free __P((void *, unsigned long, int));
                    119:
1.25      thorpej   120: static void pool_print1 __P((struct pool *, const char *,
                    121:        void (*)(const char *, ...)));
1.3       pk        122:
                    123: /*
                    124:  * Pool log entry. An array of these is allocated in pool_create().
                    125:  */
                    126: struct pool_log {
                    127:        const char      *pl_file;
                    128:        long            pl_line;
                    129:        int             pl_action;
1.25      thorpej   130: #define        PRLOG_GET       1
                    131: #define        PRLOG_PUT       2
1.3       pk        132:        void            *pl_addr;
1.1       pk        133: };
                    134:
1.3       pk        135: /* Number of entries in pool log buffers */
1.17      thorpej   136: #ifndef POOL_LOGSIZE
                    137: #define        POOL_LOGSIZE    10
                    138: #endif
                    139:
                    140: int pool_logsize = POOL_LOGSIZE;
1.1       pk        141:
1.25      thorpej   142: #ifdef DIAGNOSTIC
1.3       pk        143: static void    pr_log __P((struct pool *, void *, int, const char *, long));
1.25      thorpej   144: static void    pr_printlog __P((struct pool *, struct pool_item *,
                    145:                    void (*)(const char *, ...)));
                    146: static void    pr_enter __P((struct pool *, const char *, long));
                    147: static void    pr_leave __P((struct pool *));
                    148: static void    pr_enter_check __P((struct pool *,
                    149:                    void (*)(const char *, ...)));
1.3       pk        150:
                    151: static __inline__ void
                    152: pr_log(pp, v, action, file, line)
                    153:        struct pool     *pp;
                    154:        void            *v;
                    155:        int             action;
                    156:        const char      *file;
                    157:        long            line;
                    158: {
                    159:        int n = pp->pr_curlogentry;
                    160:        struct pool_log *pl;
                    161:
1.20      thorpej   162:        if ((pp->pr_roflags & PR_LOGGING) == 0)
1.3       pk        163:                return;
                    164:
                    165:        /*
                    166:         * Fill in the current entry. Wrap around and overwrite
                    167:         * the oldest entry if necessary.
                    168:         */
                    169:        pl = &pp->pr_log[n];
                    170:        pl->pl_file = file;
                    171:        pl->pl_line = line;
                    172:        pl->pl_action = action;
                    173:        pl->pl_addr = v;
                    174:        if (++n >= pp->pr_logsize)
                    175:                n = 0;
                    176:        pp->pr_curlogentry = n;
                    177: }
                    178:
                    179: static void
1.25      thorpej   180: pr_printlog(pp, pi, pr)
1.3       pk        181:        struct pool *pp;
1.25      thorpej   182:        struct pool_item *pi;
                    183:        void (*pr) __P((const char *, ...));
1.3       pk        184: {
                    185:        int i = pp->pr_logsize;
                    186:        int n = pp->pr_curlogentry;
                    187:
1.20      thorpej   188:        if ((pp->pr_roflags & PR_LOGGING) == 0)
1.3       pk        189:                return;
                    190:
                    191:        /*
                    192:         * Print all entries in this pool's log.
                    193:         */
                    194:        while (i-- > 0) {
                    195:                struct pool_log *pl = &pp->pr_log[n];
                    196:                if (pl->pl_action != 0) {
1.25      thorpej   197:                        if (pi == NULL || pi == pl->pl_addr) {
                    198:                                (*pr)("\tlog entry %d:\n", i);
                    199:                                (*pr)("\t\taction = %s, addr = %p\n",
                    200:                                    pl->pl_action == PRLOG_GET ? "get" : "put",
                    201:                                    pl->pl_addr);
                    202:                                (*pr)("\t\tfile: %s at line %lu\n",
                    203:                                    pl->pl_file, pl->pl_line);
                    204:                        }
1.3       pk        205:                }
                    206:                if (++n >= pp->pr_logsize)
                    207:                        n = 0;
                    208:        }
                    209: }
1.25      thorpej   210:
                    211: static __inline__ void
                    212: pr_enter(pp, file, line)
                    213:        struct pool *pp;
                    214:        const char *file;
                    215:        long line;
                    216: {
                    217:
1.34    ! thorpej   218:        if (__predict_false(pp->pr_entered_file != NULL)) {
1.25      thorpej   219:                printf("pool %s: reentrancy at file %s line %ld\n",
                    220:                    pp->pr_wchan, file, line);
                    221:                printf("         previous entry at file %s line %ld\n",
                    222:                    pp->pr_entered_file, pp->pr_entered_line);
                    223:                panic("pr_enter");
                    224:        }
                    225:
                    226:        pp->pr_entered_file = file;
                    227:        pp->pr_entered_line = line;
                    228: }
                    229:
                    230: static __inline__ void
                    231: pr_leave(pp)
                    232:        struct pool *pp;
                    233: {
                    234:
1.34    ! thorpej   235:        if (__predict_false(pp->pr_entered_file == NULL)) {
1.25      thorpej   236:                printf("pool %s not entered?\n", pp->pr_wchan);
                    237:                panic("pr_leave");
                    238:        }
                    239:
                    240:        pp->pr_entered_file = NULL;
                    241:        pp->pr_entered_line = 0;
                    242: }
                    243:
                    244: static __inline__ void
                    245: pr_enter_check(pp, pr)
                    246:        struct pool *pp;
                    247:        void (*pr) __P((const char *, ...));
                    248: {
                    249:
                    250:        if (pp->pr_entered_file != NULL)
                    251:                (*pr)("\n\tcurrently entered from file %s line %ld\n",
                    252:                    pp->pr_entered_file, pp->pr_entered_line);
                    253: }
1.3       pk        254: #else
1.25      thorpej   255: #define        pr_log(pp, v, action, file, line)
                    256: #define        pr_printlog(pp, pi, pr)
                    257: #define        pr_enter(pp, file, line)
                    258: #define        pr_leave(pp)
                    259: #define        pr_enter_check(pp, pr)
                    260: #endif /* DIAGNOSTIC */
1.3       pk        261:
                    262: /*
                    263:  * Return the pool page header based on page address.
                    264:  */
                    265: static __inline__ struct pool_item_header *
                    266: pr_find_pagehead(pp, page)
                    267:        struct pool *pp;
                    268:        caddr_t page;
                    269: {
                    270:        struct pool_item_header *ph;
                    271:
1.20      thorpej   272:        if ((pp->pr_roflags & PR_PHINPAGE) != 0)
1.3       pk        273:                return ((struct pool_item_header *)(page + pp->pr_phoffset));
                    274:
                    275:        for (ph = LIST_FIRST(&pp->pr_hashtab[PR_HASH_INDEX(pp, page)]);
                    276:             ph != NULL;
                    277:             ph = LIST_NEXT(ph, ph_hashlist)) {
                    278:                if (ph->ph_page == page)
                    279:                        return (ph);
                    280:        }
                    281:        return (NULL);
                    282: }
                    283:
                    284: /*
                    285:  * Remove a page from the pool.
                    286:  */
                    287: static __inline__ void
                    288: pr_rmpage(pp, ph)
                    289:        struct pool *pp;
                    290:        struct pool_item_header *ph;
                    291: {
                    292:
                    293:        /*
1.7       thorpej   294:         * If the page was idle, decrement the idle page count.
1.3       pk        295:         */
1.6       thorpej   296:        if (ph->ph_nmissing == 0) {
                    297: #ifdef DIAGNOSTIC
                    298:                if (pp->pr_nidle == 0)
                    299:                        panic("pr_rmpage: nidle inconsistent");
1.20      thorpej   300:                if (pp->pr_nitems < pp->pr_itemsperpage)
                    301:                        panic("pr_rmpage: nitems inconsistent");
1.6       thorpej   302: #endif
                    303:                pp->pr_nidle--;
                    304:        }
1.7       thorpej   305:
1.20      thorpej   306:        pp->pr_nitems -= pp->pr_itemsperpage;
                    307:
1.7       thorpej   308:        /*
                    309:         * Unlink a page from the pool and release it.
                    310:         */
                    311:        TAILQ_REMOVE(&pp->pr_pagelist, ph, ph_pagelist);
                    312:        (*pp->pr_free)(ph->ph_page, pp->pr_pagesz, pp->pr_mtype);
                    313:        pp->pr_npages--;
                    314:        pp->pr_npagefree++;
1.6       thorpej   315:
1.22      chs       316:        if ((pp->pr_roflags & PR_PHINPAGE) == 0) {
1.27      pk        317:                int s;
1.22      chs       318:                LIST_REMOVE(ph, ph_hashlist);
1.27      pk        319:                s = splhigh();
1.22      chs       320:                pool_put(&phpool, ph);
1.27      pk        321:                splx(s);
1.22      chs       322:        }
                    323:
1.3       pk        324:        if (pp->pr_curpage == ph) {
                    325:                /*
                    326:                 * Find a new non-empty page header, if any.
                    327:                 * Start search from the page head, to increase the
                    328:                 * chance for "high water" pages to be freed.
                    329:                 */
                    330:                for (ph = TAILQ_FIRST(&pp->pr_pagelist); ph != NULL;
                    331:                     ph = TAILQ_NEXT(ph, ph_pagelist))
                    332:                        if (TAILQ_FIRST(&ph->ph_itemlist) != NULL)
                    333:                                break;
                    334:
                    335:                pp->pr_curpage = ph;
1.21      thorpej   336:        }
1.3       pk        337: }
                    338:
                    339: /*
                    340:  * Allocate and initialize a pool.
                    341:  */
1.1       pk        342: struct pool *
1.3       pk        343: pool_create(size, align, ioff, nitems, wchan, pagesz, alloc, release, mtype)
1.1       pk        344:        size_t  size;
1.3       pk        345:        u_int   align;
                    346:        u_int   ioff;
1.1       pk        347:        int     nitems;
1.21      thorpej   348:        const char *wchan;
1.3       pk        349:        size_t  pagesz;
                    350:        void    *(*alloc) __P((unsigned long, int, int));
                    351:        void    (*release) __P((void *, unsigned long, int));
1.1       pk        352:        int     mtype;
                    353: {
                    354:        struct pool *pp;
1.3       pk        355:        int flags;
1.1       pk        356:
1.3       pk        357:        pp = (struct pool *)malloc(sizeof(*pp), M_POOL, M_NOWAIT);
                    358:        if (pp == NULL)
1.1       pk        359:                return (NULL);
1.3       pk        360:
                    361:        flags = PR_FREEHEADER;
                    362:        pool_init(pp, size, align, ioff, flags, wchan, pagesz,
                    363:                  alloc, release, mtype);
                    364:
                    365:        if (nitems != 0) {
                    366:                if (pool_prime(pp, nitems, NULL) != 0) {
                    367:                        pool_destroy(pp);
                    368:                        return (NULL);
                    369:                }
1.1       pk        370:        }
                    371:
1.3       pk        372:        return (pp);
                    373: }
                    374:
                    375: /*
                    376:  * Initialize the given pool resource structure.
                    377:  *
                    378:  * We export this routine to allow other kernel parts to declare
                    379:  * static pools that must be initialized before malloc() is available.
                    380:  */
                    381: void
                    382: pool_init(pp, size, align, ioff, flags, wchan, pagesz, alloc, release, mtype)
                    383:        struct pool     *pp;
                    384:        size_t          size;
                    385:        u_int           align;
                    386:        u_int           ioff;
                    387:        int             flags;
1.21      thorpej   388:        const char      *wchan;
1.3       pk        389:        size_t          pagesz;
                    390:        void            *(*alloc) __P((unsigned long, int, int));
                    391:        void            (*release) __P((void *, unsigned long, int));
                    392:        int             mtype;
                    393: {
1.16      briggs    394:        int off, slack, i;
1.3       pk        395:
1.25      thorpej   396: #ifdef POOL_DIAGNOSTIC
                    397:        /*
                    398:         * Always log if POOL_DIAGNOSTIC is defined.
                    399:         */
                    400:        if (pool_logsize != 0)
                    401:                flags |= PR_LOGGING;
                    402: #endif
                    403:
1.3       pk        404:        /*
                    405:         * Check arguments and construct default values.
                    406:         */
                    407:        if (!powerof2(pagesz) || pagesz > PAGE_SIZE)
                    408:                panic("pool_init: page size invalid (%lx)\n", (u_long)pagesz);
                    409:
1.4       thorpej   410:        if (alloc == NULL && release == NULL) {
1.3       pk        411:                alloc = pool_page_alloc;
                    412:                release = pool_page_free;
1.4       thorpej   413:                pagesz = PAGE_SIZE;     /* Rounds to PAGE_SIZE anyhow. */
                    414:        } else if ((alloc != NULL && release != NULL) == 0) {
                    415:                /* If you specifiy one, must specify both. */
                    416:                panic("pool_init: must specify alloc and release together");
                    417:        }
                    418:
1.3       pk        419:        if (pagesz == 0)
                    420:                pagesz = PAGE_SIZE;
                    421:
                    422:        if (align == 0)
                    423:                align = ALIGN(1);
1.14      thorpej   424:
                    425:        if (size < sizeof(struct pool_item))
                    426:                size = sizeof(struct pool_item);
1.3       pk        427:
                    428:        /*
                    429:         * Initialize the pool structure.
                    430:         */
                    431:        TAILQ_INIT(&pp->pr_pagelist);
                    432:        pp->pr_curpage = NULL;
                    433:        pp->pr_npages = 0;
                    434:        pp->pr_minitems = 0;
                    435:        pp->pr_minpages = 0;
                    436:        pp->pr_maxpages = UINT_MAX;
1.20      thorpej   437:        pp->pr_roflags = flags;
                    438:        pp->pr_flags = 0;
1.3       pk        439:        pp->pr_size = ALIGN(size);
                    440:        pp->pr_align = align;
                    441:        pp->pr_wchan = wchan;
                    442:        pp->pr_mtype = mtype;
                    443:        pp->pr_alloc = alloc;
                    444:        pp->pr_free = release;
                    445:        pp->pr_pagesz = pagesz;
                    446:        pp->pr_pagemask = ~(pagesz - 1);
                    447:        pp->pr_pageshift = ffs(pagesz) - 1;
1.20      thorpej   448:        pp->pr_nitems = 0;
                    449:        pp->pr_nout = 0;
                    450:        pp->pr_hardlimit = UINT_MAX;
                    451:        pp->pr_hardlimit_warning = NULL;
1.31      thorpej   452:        pp->pr_hardlimit_ratecap.tv_sec = 0;
                    453:        pp->pr_hardlimit_ratecap.tv_usec = 0;
                    454:        pp->pr_hardlimit_warning_last.tv_sec = 0;
                    455:        pp->pr_hardlimit_warning_last.tv_usec = 0;
1.3       pk        456:
                    457:        /*
                    458:         * Decide whether to put the page header off page to avoid
                    459:         * wasting too large a part of the page. Off-page page headers
                    460:         * go on a hash table, so we can match a returned item
                    461:         * with its header based on the page address.
                    462:         * We use 1/16 of the page size as the threshold (XXX: tune)
                    463:         */
                    464:        if (pp->pr_size < pagesz/16) {
                    465:                /* Use the end of the page for the page header */
1.20      thorpej   466:                pp->pr_roflags |= PR_PHINPAGE;
1.3       pk        467:                pp->pr_phoffset = off =
                    468:                        pagesz - ALIGN(sizeof(struct pool_item_header));
1.2       pk        469:        } else {
1.3       pk        470:                /* The page header will be taken from our page header pool */
                    471:                pp->pr_phoffset = 0;
                    472:                off = pagesz;
1.16      briggs    473:                for (i = 0; i < PR_HASHTABSIZE; i++) {
                    474:                        LIST_INIT(&pp->pr_hashtab[i]);
                    475:                }
1.2       pk        476:        }
1.1       pk        477:
1.3       pk        478:        /*
                    479:         * Alignment is to take place at `ioff' within the item. This means
                    480:         * we must reserve up to `align - 1' bytes on the page to allow
                    481:         * appropriate positioning of each item.
                    482:         *
                    483:         * Silently enforce `0 <= ioff < align'.
                    484:         */
                    485:        pp->pr_itemoffset = ioff = ioff % align;
                    486:        pp->pr_itemsperpage = (off - ((align - ioff) % align)) / pp->pr_size;
                    487:
                    488:        /*
                    489:         * Use the slack between the chunks and the page header
                    490:         * for "cache coloring".
                    491:         */
                    492:        slack = off - pp->pr_itemsperpage * pp->pr_size;
                    493:        pp->pr_maxcolor = (slack / align) * align;
                    494:        pp->pr_curcolor = 0;
                    495:
                    496:        pp->pr_nget = 0;
                    497:        pp->pr_nfail = 0;
                    498:        pp->pr_nput = 0;
                    499:        pp->pr_npagealloc = 0;
                    500:        pp->pr_npagefree = 0;
1.1       pk        501:        pp->pr_hiwat = 0;
1.8       thorpej   502:        pp->pr_nidle = 0;
1.3       pk        503:
1.25      thorpej   504:        if (flags & PR_LOGGING) {
                    505:                if (kmem_map == NULL ||
                    506:                    (pp->pr_log = malloc(pool_logsize * sizeof(struct pool_log),
                    507:                     M_TEMP, M_NOWAIT)) == NULL)
1.20      thorpej   508:                        pp->pr_roflags &= ~PR_LOGGING;
1.3       pk        509:                pp->pr_curlogentry = 0;
                    510:                pp->pr_logsize = pool_logsize;
                    511:        }
1.25      thorpej   512:
                    513:        pp->pr_entered_file = NULL;
                    514:        pp->pr_entered_line = 0;
1.3       pk        515:
1.21      thorpej   516:        simple_lock_init(&pp->pr_slock);
1.1       pk        517:
1.3       pk        518:        /*
                    519:         * Initialize private page header pool if we haven't done so yet.
1.23      thorpej   520:         * XXX LOCKING.
1.3       pk        521:         */
                    522:        if (phpool.pr_size == 0) {
                    523:                pool_init(&phpool, sizeof(struct pool_item_header), 0, 0,
                    524:                          0, "phpool", 0, 0, 0, 0);
1.1       pk        525:        }
                    526:
1.23      thorpej   527:        /* Insert into the list of all pools. */
                    528:        simple_lock(&pool_head_slock);
                    529:        TAILQ_INSERT_TAIL(&pool_head, pp, pr_poollist);
                    530:        simple_unlock(&pool_head_slock);
1.1       pk        531: }
                    532:
                    533: /*
                    534:  * De-commision a pool resource.
                    535:  */
                    536: void
                    537: pool_destroy(pp)
                    538:        struct pool *pp;
                    539: {
1.3       pk        540:        struct pool_item_header *ph;
                    541:
                    542: #ifdef DIAGNOSTIC
1.20      thorpej   543:        if (pp->pr_nout != 0) {
1.25      thorpej   544:                pr_printlog(pp, NULL, printf);
1.20      thorpej   545:                panic("pool_destroy: pool busy: still out: %u\n",
                    546:                    pp->pr_nout);
1.3       pk        547:        }
                    548: #endif
1.1       pk        549:
1.3       pk        550:        /* Remove all pages */
1.20      thorpej   551:        if ((pp->pr_roflags & PR_STATIC) == 0)
1.3       pk        552:                while ((ph = pp->pr_pagelist.tqh_first) != NULL)
                    553:                        pr_rmpage(pp, ph);
                    554:
                    555:        /* Remove from global pool list */
1.23      thorpej   556:        simple_lock(&pool_head_slock);
1.3       pk        557:        TAILQ_REMOVE(&pool_head, pp, pr_poollist);
1.23      thorpej   558:        /* XXX Only clear this if we were drainpp? */
1.3       pk        559:        drainpp = NULL;
1.23      thorpej   560:        simple_unlock(&pool_head_slock);
1.3       pk        561:
1.20      thorpej   562:        if ((pp->pr_roflags & PR_LOGGING) != 0)
1.3       pk        563:                free(pp->pr_log, M_TEMP);
1.2       pk        564:
1.20      thorpej   565:        if (pp->pr_roflags & PR_FREEHEADER)
1.3       pk        566:                free(pp, M_POOL);
1.1       pk        567: }
                    568:
                    569:
                    570: /*
1.3       pk        571:  * Grab an item from the pool; must be called at appropriate spl level
1.1       pk        572:  */
1.3       pk        573: void *
                    574: _pool_get(pp, flags, file, line)
                    575:        struct pool *pp;
                    576:        int flags;
                    577:        const char *file;
                    578:        long line;
1.1       pk        579: {
                    580:        void *v;
                    581:        struct pool_item *pi;
1.3       pk        582:        struct pool_item_header *ph;
1.1       pk        583:
1.2       pk        584: #ifdef DIAGNOSTIC
1.34    ! thorpej   585:        if (__predict_false((pp->pr_roflags & PR_STATIC) &&
        !           586:                            (flags & PR_MALLOCOK))) {
1.25      thorpej   587:                pr_printlog(pp, NULL, printf);
1.2       pk        588:                panic("pool_get: static");
1.3       pk        589:        }
1.2       pk        590: #endif
                    591:
1.34    ! thorpej   592:        if (__predict_false(curproc == NULL && (flags & PR_WAITOK) != 0))
1.3       pk        593:                panic("pool_get: must have NOWAIT");
1.1       pk        594:
1.21      thorpej   595:        simple_lock(&pp->pr_slock);
1.25      thorpej   596:        pr_enter(pp, file, line);
1.20      thorpej   597:
                    598:  startover:
                    599:        /*
                    600:         * Check to see if we've reached the hard limit.  If we have,
                    601:         * and we can wait, then wait until an item has been returned to
                    602:         * the pool.
                    603:         */
                    604: #ifdef DIAGNOSTIC
1.34    ! thorpej   605:        if (__predict_false(pp->pr_nout > pp->pr_hardlimit)) {
1.25      thorpej   606:                pr_leave(pp);
1.21      thorpej   607:                simple_unlock(&pp->pr_slock);
1.20      thorpej   608:                panic("pool_get: %s: crossed hard limit", pp->pr_wchan);
                    609:        }
                    610: #endif
1.34    ! thorpej   611:        if (__predict_false(pp->pr_nout == pp->pr_hardlimit)) {
1.29      sommerfe  612:                if ((flags & PR_WAITOK) && !(flags & PR_LIMITFAIL)) {
1.20      thorpej   613:                        /*
                    614:                         * XXX: A warning isn't logged in this case.  Should
                    615:                         * it be?
                    616:                         */
                    617:                        pp->pr_flags |= PR_WANTED;
1.25      thorpej   618:                        pr_leave(pp);
1.21      thorpej   619:                        simple_unlock(&pp->pr_slock);
1.20      thorpej   620:                        tsleep((caddr_t)pp, PSWP, pp->pr_wchan, 0);
1.21      thorpej   621:                        simple_lock(&pp->pr_slock);
1.25      thorpej   622:                        pr_enter(pp, file, line);
1.20      thorpej   623:                        goto startover;
                    624:                }
1.31      thorpej   625:
                    626:                /*
                    627:                 * Log a message that the hard limit has been hit.
                    628:                 */
                    629:                if (pp->pr_hardlimit_warning != NULL &&
                    630:                    ratecheck(&pp->pr_hardlimit_warning_last,
                    631:                              &pp->pr_hardlimit_ratecap))
                    632:                        log(LOG_ERR, "%s\n", pp->pr_hardlimit_warning);
1.21      thorpej   633:
                    634:                if (flags & PR_URGENT)
                    635:                        panic("pool_get: urgent");
                    636:
                    637:                pp->pr_nfail++;
                    638:
1.25      thorpej   639:                pr_leave(pp);
1.21      thorpej   640:                simple_unlock(&pp->pr_slock);
1.20      thorpej   641:                return (NULL);
                    642:        }
                    643:
1.3       pk        644:        /*
                    645:         * The convention we use is that if `curpage' is not NULL, then
                    646:         * it points at a non-empty bucket. In particular, `curpage'
                    647:         * never points at a page header which has PR_PHINPAGE set and
                    648:         * has no items in its bucket.
                    649:         */
1.20      thorpej   650:        if ((ph = pp->pr_curpage) == NULL) {
1.15      pk        651:                void *v;
                    652:
1.20      thorpej   653: #ifdef DIAGNOSTIC
                    654:                if (pp->pr_nitems != 0) {
1.21      thorpej   655:                        simple_unlock(&pp->pr_slock);
1.20      thorpej   656:                        printf("pool_get: %s: curpage NULL, nitems %u\n",
                    657:                            pp->pr_wchan, pp->pr_nitems);
                    658:                        panic("pool_get: nitems inconsistent\n");
                    659:                }
                    660: #endif
                    661:
1.21      thorpej   662:                /*
                    663:                 * Call the back-end page allocator for more memory.
                    664:                 * Release the pool lock, as the back-end page allocator
                    665:                 * may block.
                    666:                 */
1.25      thorpej   667:                pr_leave(pp);
1.21      thorpej   668:                simple_unlock(&pp->pr_slock);
                    669:                v = (*pp->pr_alloc)(pp->pr_pagesz, flags, pp->pr_mtype);
                    670:                simple_lock(&pp->pr_slock);
1.25      thorpej   671:                pr_enter(pp, file, line);
1.15      pk        672:
1.21      thorpej   673:                if (v == NULL) {
                    674:                        /*
                    675:                         * We were unable to allocate a page, but
                    676:                         * we released the lock during allocation,
                    677:                         * so perhaps items were freed back to the
                    678:                         * pool.  Check for this case.
                    679:                         */
                    680:                        if (pp->pr_curpage != NULL)
                    681:                                goto startover;
1.15      pk        682:
1.3       pk        683:                        if (flags & PR_URGENT)
                    684:                                panic("pool_get: urgent");
1.21      thorpej   685:
1.3       pk        686:                        if ((flags & PR_WAITOK) == 0) {
                    687:                                pp->pr_nfail++;
1.25      thorpej   688:                                pr_leave(pp);
1.21      thorpej   689:                                simple_unlock(&pp->pr_slock);
1.1       pk        690:                                return (NULL);
1.3       pk        691:                        }
                    692:
1.15      pk        693:                        /*
                    694:                         * Wait for items to be returned to this pool.
1.21      thorpej   695:                         *
1.15      pk        696:                         * XXX: we actually want to wait just until
                    697:                         * the page allocator has memory again. Depending
                    698:                         * on this pool's usage, we might get stuck here
                    699:                         * for a long time.
1.20      thorpej   700:                         *
                    701:                         * XXX: maybe we should wake up once a second and
                    702:                         * try again?
1.15      pk        703:                         */
1.1       pk        704:                        pp->pr_flags |= PR_WANTED;
1.25      thorpej   705:                        pr_leave(pp);
1.21      thorpej   706:                        simple_unlock(&pp->pr_slock);
1.1       pk        707:                        tsleep((caddr_t)pp, PSWP, pp->pr_wchan, 0);
1.21      thorpej   708:                        simple_lock(&pp->pr_slock);
1.25      thorpej   709:                        pr_enter(pp, file, line);
1.20      thorpej   710:                        goto startover;
1.1       pk        711:                }
1.3       pk        712:
1.15      pk        713:                /* We have more memory; add it to the pool */
                    714:                pp->pr_npagealloc++;
                    715:                pool_prime_page(pp, v);
                    716:
1.20      thorpej   717:                /* Start the allocation process over. */
                    718:                goto startover;
1.3       pk        719:        }
                    720:
1.34    ! thorpej   721:        if (__predict_false((v = pi = TAILQ_FIRST(&ph->ph_itemlist)) == NULL)) {
1.25      thorpej   722:                pr_leave(pp);
1.21      thorpej   723:                simple_unlock(&pp->pr_slock);
1.3       pk        724:                panic("pool_get: %s: page empty", pp->pr_wchan);
1.21      thorpej   725:        }
1.20      thorpej   726: #ifdef DIAGNOSTIC
1.34    ! thorpej   727:        if (__predict_false(pp->pr_nitems == 0)) {
1.25      thorpej   728:                pr_leave(pp);
1.21      thorpej   729:                simple_unlock(&pp->pr_slock);
1.20      thorpej   730:                printf("pool_get: %s: items on itemlist, nitems %u\n",
                    731:                    pp->pr_wchan, pp->pr_nitems);
                    732:                panic("pool_get: nitems inconsistent\n");
                    733:        }
                    734: #endif
1.3       pk        735:        pr_log(pp, v, PRLOG_GET, file, line);
                    736:
                    737: #ifdef DIAGNOSTIC
1.34    ! thorpej   738:        if (__predict_false(pi->pi_magic != PI_MAGIC)) {
1.25      thorpej   739:                pr_printlog(pp, pi, printf);
1.3       pk        740:                panic("pool_get(%s): free list modified: magic=%x; page %p;"
                    741:                       " item addr %p\n",
                    742:                        pp->pr_wchan, pi->pi_magic, ph->ph_page, pi);
                    743:        }
                    744: #endif
                    745:
                    746:        /*
                    747:         * Remove from item list.
                    748:         */
                    749:        TAILQ_REMOVE(&ph->ph_itemlist, pi, pi_list);
1.20      thorpej   750:        pp->pr_nitems--;
                    751:        pp->pr_nout++;
1.6       thorpej   752:        if (ph->ph_nmissing == 0) {
                    753: #ifdef DIAGNOSTIC
1.34    ! thorpej   754:                if (__predict_false(pp->pr_nidle == 0))
1.6       thorpej   755:                        panic("pool_get: nidle inconsistent");
                    756: #endif
                    757:                pp->pr_nidle--;
                    758:        }
1.3       pk        759:        ph->ph_nmissing++;
                    760:        if (TAILQ_FIRST(&ph->ph_itemlist) == NULL) {
1.21      thorpej   761: #ifdef DIAGNOSTIC
1.34    ! thorpej   762:                if (__predict_false(ph->ph_nmissing != pp->pr_itemsperpage)) {
1.25      thorpej   763:                        pr_leave(pp);
1.21      thorpej   764:                        simple_unlock(&pp->pr_slock);
                    765:                        panic("pool_get: %s: nmissing inconsistent",
                    766:                            pp->pr_wchan);
                    767:                }
                    768: #endif
1.3       pk        769:                /*
                    770:                 * Find a new non-empty page header, if any.
                    771:                 * Start search from the page head, to increase
                    772:                 * the chance for "high water" pages to be freed.
                    773:                 *
1.21      thorpej   774:                 * Migrate empty pages to the end of the list.  This
                    775:                 * will speed the update of curpage as pages become
                    776:                 * idle.  Empty pages intermingled with idle pages
                    777:                 * is no big deal.  As soon as a page becomes un-empty,
                    778:                 * it will move back to the head of the list.
1.3       pk        779:                 */
                    780:                TAILQ_REMOVE(&pp->pr_pagelist, ph, ph_pagelist);
1.21      thorpej   781:                TAILQ_INSERT_TAIL(&pp->pr_pagelist, ph, ph_pagelist);
                    782:                for (ph = TAILQ_FIRST(&pp->pr_pagelist); ph != NULL;
                    783:                     ph = TAILQ_NEXT(ph, ph_pagelist))
1.3       pk        784:                        if (TAILQ_FIRST(&ph->ph_itemlist) != NULL)
                    785:                                break;
                    786:
                    787:                pp->pr_curpage = ph;
1.1       pk        788:        }
1.3       pk        789:
                    790:        pp->pr_nget++;
1.20      thorpej   791:
                    792:        /*
                    793:         * If we have a low water mark and we are now below that low
                    794:         * water mark, add more items to the pool.
                    795:         */
                    796:        if (pp->pr_nitems < pp->pr_minitems && pool_catchup(pp) != 0) {
                    797:                /*
                    798:                 * XXX: Should we log a warning?  Should we set up a timeout
                    799:                 * to try again in a second or so?  The latter could break
                    800:                 * a caller's assumptions about interrupt protection, etc.
                    801:                 */
                    802:        }
                    803:
1.25      thorpej   804:        pr_leave(pp);
1.21      thorpej   805:        simple_unlock(&pp->pr_slock);
1.1       pk        806:        return (v);
                    807: }
                    808:
                    809: /*
1.3       pk        810:  * Return resource to the pool; must be called at appropriate spl level
1.1       pk        811:  */
1.3       pk        812: void
                    813: _pool_put(pp, v, file, line)
                    814:        struct pool *pp;
                    815:        void *v;
                    816:        const char *file;
                    817:        long line;
1.1       pk        818: {
                    819:        struct pool_item *pi = v;
1.3       pk        820:        struct pool_item_header *ph;
                    821:        caddr_t page;
1.21      thorpej   822:        int s;
1.3       pk        823:
                    824:        page = (caddr_t)((u_long)v & pp->pr_pagemask);
1.1       pk        825:
1.21      thorpej   826:        simple_lock(&pp->pr_slock);
1.25      thorpej   827:        pr_enter(pp, file, line);
1.30      thorpej   828:
                    829: #ifdef DIAGNOSTIC
1.34    ! thorpej   830:        if (__predict_false(pp->pr_nout == 0)) {
1.30      thorpej   831:                printf("pool %s: putting with none out\n",
                    832:                    pp->pr_wchan);
                    833:                panic("pool_put");
                    834:        }
                    835: #endif
1.3       pk        836:
                    837:        pr_log(pp, v, PRLOG_PUT, file, line);
                    838:
1.34    ! thorpej   839:        if (__predict_false((ph = pr_find_pagehead(pp, page)) == NULL)) {
1.25      thorpej   840:                pr_printlog(pp, NULL, printf);
1.3       pk        841:                panic("pool_put: %s: page header missing", pp->pr_wchan);
                    842:        }
1.28      thorpej   843:
                    844: #ifdef LOCKDEBUG
                    845:        /*
                    846:         * Check if we're freeing a locked simple lock.
                    847:         */
                    848:        simple_lock_freecheck((caddr_t)pi, ((caddr_t)pi) + pp->pr_size);
                    849: #endif
1.3       pk        850:
                    851:        /*
                    852:         * Return to item list.
                    853:         */
1.2       pk        854: #ifdef DIAGNOSTIC
1.3       pk        855:        pi->pi_magic = PI_MAGIC;
                    856: #endif
1.32      chs       857: #ifdef DEBUG
                    858:        {
                    859:                int i, *ip = v;
                    860:
                    861:                for (i = 0; i < pp->pr_size / sizeof(int); i++) {
                    862:                        *ip++ = PI_MAGIC;
                    863:                }
                    864:        }
                    865: #endif
                    866:
1.3       pk        867:        TAILQ_INSERT_HEAD(&ph->ph_itemlist, pi, pi_list);
                    868:        ph->ph_nmissing--;
                    869:        pp->pr_nput++;
1.20      thorpej   870:        pp->pr_nitems++;
                    871:        pp->pr_nout--;
1.3       pk        872:
                    873:        /* Cancel "pool empty" condition if it exists */
                    874:        if (pp->pr_curpage == NULL)
                    875:                pp->pr_curpage = ph;
                    876:
                    877:        if (pp->pr_flags & PR_WANTED) {
                    878:                pp->pr_flags &= ~PR_WANTED;
1.15      pk        879:                if (ph->ph_nmissing == 0)
                    880:                        pp->pr_nidle++;
1.25      thorpej   881:                pr_leave(pp);
1.21      thorpej   882:                simple_unlock(&pp->pr_slock);
1.3       pk        883:                wakeup((caddr_t)pp);
                    884:                return;
                    885:        }
                    886:
                    887:        /*
1.21      thorpej   888:         * If this page is now complete, do one of two things:
                    889:         *
                    890:         *      (1) If we have more pages than the page high water
                    891:         *          mark, free the page back to the system.
                    892:         *
                    893:         *      (2) Move it to the end of the page list, so that
                    894:         *          we minimize our chances of fragmenting the
                    895:         *          pool.  Idle pages migrate to the end (along with
                    896:         *          completely empty pages, so that we find un-empty
                    897:         *          pages more quickly when we update curpage) of the
                    898:         *          list so they can be more easily swept up by
                    899:         *          the pagedaemon when pages are scarce.
1.3       pk        900:         */
                    901:        if (ph->ph_nmissing == 0) {
1.6       thorpej   902:                pp->pr_nidle++;
1.3       pk        903:                if (pp->pr_npages > pp->pr_maxpages) {
                    904:                        pr_rmpage(pp, ph);
                    905:                } else {
                    906:                        TAILQ_REMOVE(&pp->pr_pagelist, ph, ph_pagelist);
                    907:                        TAILQ_INSERT_TAIL(&pp->pr_pagelist, ph, ph_pagelist);
                    908:
1.21      thorpej   909:                        /*
                    910:                         * Update the timestamp on the page.  A page must
                    911:                         * be idle for some period of time before it can
                    912:                         * be reclaimed by the pagedaemon.  This minimizes
                    913:                         * ping-pong'ing for memory.
                    914:                         */
                    915:                        s = splclock();
                    916:                        ph->ph_time = mono_time;
                    917:                        splx(s);
                    918:
                    919:                        /*
                    920:                         * Update the current page pointer.  Just look for
                    921:                         * the first page with any free items.
                    922:                         *
                    923:                         * XXX: Maybe we want an option to look for the
                    924:                         * page with the fewest available items, to minimize
                    925:                         * fragmentation?
                    926:                         */
1.3       pk        927:                        for (ph = TAILQ_FIRST(&pp->pr_pagelist); ph != NULL;
                    928:                             ph = TAILQ_NEXT(ph, ph_pagelist))
                    929:                                if (TAILQ_FIRST(&ph->ph_itemlist) != NULL)
                    930:                                        break;
1.1       pk        931:
1.3       pk        932:                        pp->pr_curpage = ph;
1.1       pk        933:                }
                    934:        }
1.21      thorpej   935:        /*
                    936:         * If the page has just become un-empty, move it to the head of
                    937:         * the list, and make it the current page.  The next allocation
                    938:         * will get the item from this page, instead of further fragmenting
                    939:         * the pool.
                    940:         */
                    941:        else if (ph->ph_nmissing == (pp->pr_itemsperpage - 1)) {
                    942:                TAILQ_REMOVE(&pp->pr_pagelist, ph, ph_pagelist);
                    943:                TAILQ_INSERT_HEAD(&pp->pr_pagelist, ph, ph_pagelist);
                    944:                pp->pr_curpage = ph;
                    945:        }
                    946:
1.25      thorpej   947:        pr_leave(pp);
1.21      thorpej   948:        simple_unlock(&pp->pr_slock);
1.3       pk        949:
1.1       pk        950: }
                    951:
                    952: /*
1.3       pk        953:  * Add N items to the pool.
1.1       pk        954:  */
                    955: int
1.2       pk        956: pool_prime(pp, n, storage)
1.1       pk        957:        struct pool *pp;
                    958:        int n;
1.2       pk        959:        caddr_t storage;
1.1       pk        960: {
1.3       pk        961:        caddr_t cp;
                    962:        int newnitems, newpages;
1.2       pk        963:
                    964: #ifdef DIAGNOSTIC
1.34    ! thorpej   965:        if (__predict_false(storage && !(pp->pr_roflags & PR_STATIC)))
1.2       pk        966:                panic("pool_prime: static");
                    967:        /* !storage && static caught below */
                    968: #endif
1.1       pk        969:
1.21      thorpej   970:        simple_lock(&pp->pr_slock);
                    971:
1.3       pk        972:        newnitems = pp->pr_minitems + n;
                    973:        newpages =
1.18      thorpej   974:                roundup(newnitems, pp->pr_itemsperpage) / pp->pr_itemsperpage
1.3       pk        975:                - pp->pr_minpages;
                    976:
                    977:        while (newpages-- > 0) {
1.20      thorpej   978:                if (pp->pr_roflags & PR_STATIC) {
1.3       pk        979:                        cp = storage;
                    980:                        storage += pp->pr_pagesz;
                    981:                } else {
1.21      thorpej   982:                        simple_unlock(&pp->pr_slock);
1.3       pk        983:                        cp = (*pp->pr_alloc)(pp->pr_pagesz, 0, pp->pr_mtype);
1.21      thorpej   984:                        simple_lock(&pp->pr_slock);
1.3       pk        985:                }
1.2       pk        986:
1.3       pk        987:                if (cp == NULL) {
1.21      thorpej   988:                        simple_unlock(&pp->pr_slock);
1.1       pk        989:                        return (ENOMEM);
                    990:                }
                    991:
1.26      thorpej   992:                pp->pr_npagealloc++;
1.3       pk        993:                pool_prime_page(pp, cp);
                    994:                pp->pr_minpages++;
1.1       pk        995:        }
1.3       pk        996:
                    997:        pp->pr_minitems = newnitems;
                    998:
                    999:        if (pp->pr_minpages >= pp->pr_maxpages)
                   1000:                pp->pr_maxpages = pp->pr_minpages + 1;  /* XXX */
                   1001:
1.21      thorpej  1002:        simple_unlock(&pp->pr_slock);
1.1       pk       1003:        return (0);
                   1004: }
1.3       pk       1005:
                   1006: /*
                   1007:  * Add a page worth of items to the pool.
1.21      thorpej  1008:  *
                   1009:  * Note, we must be called with the pool descriptor LOCKED.
1.3       pk       1010:  */
1.21      thorpej  1011: static void
1.3       pk       1012: pool_prime_page(pp, storage)
                   1013:        struct pool *pp;
                   1014:        caddr_t storage;
                   1015: {
                   1016:        struct pool_item *pi;
                   1017:        struct pool_item_header *ph;
                   1018:        caddr_t cp = storage;
                   1019:        unsigned int align = pp->pr_align;
                   1020:        unsigned int ioff = pp->pr_itemoffset;
1.27      pk       1021:        int s, n;
1.3       pk       1022:
1.20      thorpej  1023:        if ((pp->pr_roflags & PR_PHINPAGE) != 0) {
1.3       pk       1024:                ph = (struct pool_item_header *)(cp + pp->pr_phoffset);
                   1025:        } else {
1.27      pk       1026:                s = splhigh();
1.3       pk       1027:                ph = pool_get(&phpool, PR_URGENT);
1.27      pk       1028:                splx(s);
1.3       pk       1029:                LIST_INSERT_HEAD(&pp->pr_hashtab[PR_HASH_INDEX(pp, cp)],
                   1030:                                 ph, ph_hashlist);
                   1031:        }
                   1032:
                   1033:        /*
                   1034:         * Insert page header.
                   1035:         */
                   1036:        TAILQ_INSERT_HEAD(&pp->pr_pagelist, ph, ph_pagelist);
                   1037:        TAILQ_INIT(&ph->ph_itemlist);
                   1038:        ph->ph_page = storage;
                   1039:        ph->ph_nmissing = 0;
1.21      thorpej  1040:        memset(&ph->ph_time, 0, sizeof(ph->ph_time));
1.3       pk       1041:
1.6       thorpej  1042:        pp->pr_nidle++;
                   1043:
1.3       pk       1044:        /*
                   1045:         * Color this page.
                   1046:         */
                   1047:        cp = (caddr_t)(cp + pp->pr_curcolor);
                   1048:        if ((pp->pr_curcolor += align) > pp->pr_maxcolor)
                   1049:                pp->pr_curcolor = 0;
                   1050:
                   1051:        /*
                   1052:         * Adjust storage to apply aligment to `pr_itemoffset' in each item.
                   1053:         */
                   1054:        if (ioff != 0)
                   1055:                cp = (caddr_t)(cp + (align - ioff));
                   1056:
                   1057:        /*
                   1058:         * Insert remaining chunks on the bucket list.
                   1059:         */
                   1060:        n = pp->pr_itemsperpage;
1.20      thorpej  1061:        pp->pr_nitems += n;
1.3       pk       1062:
                   1063:        while (n--) {
                   1064:                pi = (struct pool_item *)cp;
                   1065:
                   1066:                /* Insert on page list */
                   1067:                TAILQ_INSERT_TAIL(&ph->ph_itemlist, pi, pi_list);
                   1068: #ifdef DIAGNOSTIC
                   1069:                pi->pi_magic = PI_MAGIC;
                   1070: #endif
                   1071:                cp = (caddr_t)(cp + pp->pr_size);
                   1072:        }
                   1073:
                   1074:        /*
                   1075:         * If the pool was depleted, point at the new page.
                   1076:         */
                   1077:        if (pp->pr_curpage == NULL)
                   1078:                pp->pr_curpage = ph;
                   1079:
                   1080:        if (++pp->pr_npages > pp->pr_hiwat)
                   1081:                pp->pr_hiwat = pp->pr_npages;
                   1082: }
                   1083:
1.20      thorpej  1084: /*
                   1085:  * Like pool_prime(), except this is used by pool_get() when nitems
                   1086:  * drops below the low water mark.  This is used to catch up nitmes
                   1087:  * with the low water mark.
                   1088:  *
1.21      thorpej  1089:  * Note 1, we never wait for memory here, we let the caller decide what to do.
1.20      thorpej  1090:  *
                   1091:  * Note 2, this doesn't work with static pools.
                   1092:  *
                   1093:  * Note 3, we must be called with the pool already locked, and we return
                   1094:  * with it locked.
                   1095:  */
                   1096: static int
                   1097: pool_catchup(pp)
                   1098:        struct pool *pp;
                   1099: {
                   1100:        caddr_t cp;
                   1101:        int error = 0;
                   1102:
                   1103:        if (pp->pr_roflags & PR_STATIC) {
                   1104:                /*
                   1105:                 * We dropped below the low water mark, and this is not a
                   1106:                 * good thing.  Log a warning.
1.21      thorpej  1107:                 *
                   1108:                 * XXX: rate-limit this?
1.20      thorpej  1109:                 */
                   1110:                printf("WARNING: static pool `%s' dropped below low water "
                   1111:                    "mark\n", pp->pr_wchan);
                   1112:                return (0);
                   1113:        }
                   1114:
1.21      thorpej  1115:        while (pp->pr_nitems < pp->pr_minitems) {
1.20      thorpej  1116:                /*
1.21      thorpej  1117:                 * Call the page back-end allocator for more memory.
                   1118:                 *
                   1119:                 * XXX: We never wait, so should we bother unlocking
                   1120:                 * the pool descriptor?
1.20      thorpej  1121:                 */
1.21      thorpej  1122:                simple_unlock(&pp->pr_slock);
1.20      thorpej  1123:                cp = (*pp->pr_alloc)(pp->pr_pagesz, 0, pp->pr_mtype);
1.21      thorpej  1124:                simple_lock(&pp->pr_slock);
1.34    ! thorpej  1125:                if (__predict_false(cp == NULL)) {
1.20      thorpej  1126:                        error = ENOMEM;
                   1127:                        break;
                   1128:                }
1.26      thorpej  1129:                pp->pr_npagealloc++;
1.20      thorpej  1130:                pool_prime_page(pp, cp);
                   1131:        }
                   1132:
                   1133:        return (error);
                   1134: }
                   1135:
1.3       pk       1136: void
                   1137: pool_setlowat(pp, n)
                   1138:        pool_handle_t   pp;
                   1139:        int n;
                   1140: {
1.20      thorpej  1141:        int error;
1.15      pk       1142:
1.21      thorpej  1143:        simple_lock(&pp->pr_slock);
                   1144:
1.3       pk       1145:        pp->pr_minitems = n;
1.15      pk       1146:        pp->pr_minpages = (n == 0)
                   1147:                ? 0
1.18      thorpej  1148:                : roundup(n, pp->pr_itemsperpage) / pp->pr_itemsperpage;
1.20      thorpej  1149:
                   1150:        /* Make sure we're caught up with the newly-set low water mark. */
1.21      thorpej  1151:        if ((error = pool_catchup(pp)) != 0) {
1.20      thorpej  1152:                /*
                   1153:                 * XXX: Should we log a warning?  Should we set up a timeout
                   1154:                 * to try again in a second or so?  The latter could break
                   1155:                 * a caller's assumptions about interrupt protection, etc.
                   1156:                 */
                   1157:        }
1.21      thorpej  1158:
                   1159:        simple_unlock(&pp->pr_slock);
1.3       pk       1160: }
                   1161:
                   1162: void
                   1163: pool_sethiwat(pp, n)
                   1164:        pool_handle_t   pp;
                   1165:        int n;
                   1166: {
1.15      pk       1167:
1.21      thorpej  1168:        simple_lock(&pp->pr_slock);
                   1169:
1.15      pk       1170:        pp->pr_maxpages = (n == 0)
                   1171:                ? 0
1.18      thorpej  1172:                : roundup(n, pp->pr_itemsperpage) / pp->pr_itemsperpage;
1.21      thorpej  1173:
                   1174:        simple_unlock(&pp->pr_slock);
1.3       pk       1175: }
                   1176:
1.20      thorpej  1177: void
                   1178: pool_sethardlimit(pp, n, warnmess, ratecap)
                   1179:        pool_handle_t pp;
                   1180:        int n;
                   1181:        const char *warnmess;
                   1182:        int ratecap;
                   1183: {
                   1184:
1.21      thorpej  1185:        simple_lock(&pp->pr_slock);
1.20      thorpej  1186:
                   1187:        pp->pr_hardlimit = n;
                   1188:        pp->pr_hardlimit_warning = warnmess;
1.31      thorpej  1189:        pp->pr_hardlimit_ratecap.tv_sec = ratecap;
                   1190:        pp->pr_hardlimit_warning_last.tv_sec = 0;
                   1191:        pp->pr_hardlimit_warning_last.tv_usec = 0;
1.20      thorpej  1192:
                   1193:        /*
1.21      thorpej  1194:         * In-line version of pool_sethiwat(), because we don't want to
                   1195:         * release the lock.
1.20      thorpej  1196:         */
                   1197:        pp->pr_maxpages = (n == 0)
                   1198:                ? 0
                   1199:                : roundup(n, pp->pr_itemsperpage) / pp->pr_itemsperpage;
1.21      thorpej  1200:
                   1201:        simple_unlock(&pp->pr_slock);
1.20      thorpej  1202: }
1.3       pk       1203:
                   1204: /*
                   1205:  * Default page allocator.
                   1206:  */
                   1207: static void *
                   1208: pool_page_alloc(sz, flags, mtype)
                   1209:        unsigned long sz;
                   1210:        int flags;
                   1211:        int mtype;
                   1212: {
1.11      thorpej  1213:        boolean_t waitok = (flags & PR_WAITOK) ? TRUE : FALSE;
1.3       pk       1214:
1.11      thorpej  1215:        return ((void *)uvm_km_alloc_poolpage(waitok));
1.3       pk       1216: }
                   1217:
                   1218: static void
                   1219: pool_page_free(v, sz, mtype)
                   1220:        void *v;
                   1221:        unsigned long sz;
                   1222:        int mtype;
                   1223: {
                   1224:
1.10      eeh      1225:        uvm_km_free_poolpage((vaddr_t)v);
1.3       pk       1226: }
1.12      thorpej  1227:
                   1228: /*
                   1229:  * Alternate pool page allocator for pools that know they will
                   1230:  * never be accessed in interrupt context.
                   1231:  */
                   1232: void *
                   1233: pool_page_alloc_nointr(sz, flags, mtype)
                   1234:        unsigned long sz;
                   1235:        int flags;
                   1236:        int mtype;
                   1237: {
                   1238:        boolean_t waitok = (flags & PR_WAITOK) ? TRUE : FALSE;
                   1239:
                   1240:        return ((void *)uvm_km_alloc_poolpage1(kernel_map, uvm.kernel_object,
                   1241:            waitok));
                   1242: }
                   1243:
                   1244: void
                   1245: pool_page_free_nointr(v, sz, mtype)
                   1246:        void *v;
                   1247:        unsigned long sz;
                   1248:        int mtype;
                   1249: {
                   1250:
                   1251:        uvm_km_free_poolpage1(kernel_map, (vaddr_t)v);
                   1252: }
                   1253:
1.3       pk       1254:
                   1255: /*
                   1256:  * Release all complete pages that have not been used recently.
                   1257:  */
                   1258: void
1.25      thorpej  1259: _pool_reclaim(pp, file, line)
1.3       pk       1260:        pool_handle_t pp;
1.25      thorpej  1261:        const char *file;
                   1262:        long line;
1.3       pk       1263: {
                   1264:        struct pool_item_header *ph, *phnext;
1.21      thorpej  1265:        struct timeval curtime;
                   1266:        int s;
1.3       pk       1267:
1.20      thorpej  1268:        if (pp->pr_roflags & PR_STATIC)
1.3       pk       1269:                return;
                   1270:
1.21      thorpej  1271:        if (simple_lock_try(&pp->pr_slock) == 0)
1.3       pk       1272:                return;
1.25      thorpej  1273:        pr_enter(pp, file, line);
1.3       pk       1274:
1.21      thorpej  1275:        s = splclock();
                   1276:        curtime = mono_time;
                   1277:        splx(s);
                   1278:
1.3       pk       1279:        for (ph = TAILQ_FIRST(&pp->pr_pagelist); ph != NULL; ph = phnext) {
                   1280:                phnext = TAILQ_NEXT(ph, ph_pagelist);
                   1281:
                   1282:                /* Check our minimum page claim */
                   1283:                if (pp->pr_npages <= pp->pr_minpages)
                   1284:                        break;
                   1285:
                   1286:                if (ph->ph_nmissing == 0) {
                   1287:                        struct timeval diff;
                   1288:                        timersub(&curtime, &ph->ph_time, &diff);
                   1289:                        if (diff.tv_sec < pool_inactive_time)
                   1290:                                continue;
1.21      thorpej  1291:
                   1292:                        /*
                   1293:                         * If freeing this page would put us below
                   1294:                         * the low water mark, stop now.
                   1295:                         */
                   1296:                        if ((pp->pr_nitems - pp->pr_itemsperpage) <
                   1297:                            pp->pr_minitems)
                   1298:                                break;
                   1299:
1.3       pk       1300:                        pr_rmpage(pp, ph);
                   1301:                }
                   1302:        }
                   1303:
1.25      thorpej  1304:        pr_leave(pp);
1.21      thorpej  1305:        simple_unlock(&pp->pr_slock);
1.3       pk       1306: }
                   1307:
                   1308:
                   1309: /*
                   1310:  * Drain pools, one at a time.
1.21      thorpej  1311:  *
                   1312:  * Note, we must never be called from an interrupt context.
1.3       pk       1313:  */
                   1314: void
                   1315: pool_drain(arg)
                   1316:        void *arg;
                   1317: {
                   1318:        struct pool *pp;
1.23      thorpej  1319:        int s;
1.3       pk       1320:
1.23      thorpej  1321:        s = splimp();
                   1322:        simple_lock(&pool_head_slock);
                   1323:
                   1324:        if (drainpp == NULL && (drainpp = TAILQ_FIRST(&pool_head)) == NULL)
                   1325:                goto out;
1.3       pk       1326:
                   1327:        pp = drainpp;
                   1328:        drainpp = TAILQ_NEXT(pp, pr_poollist);
                   1329:
                   1330:        pool_reclaim(pp);
1.23      thorpej  1331:
                   1332:  out:
                   1333:        simple_unlock(&pool_head_slock);
1.3       pk       1334:        splx(s);
                   1335: }
                   1336:
                   1337:
                   1338: /*
                   1339:  * Diagnostic helpers.
                   1340:  */
                   1341: void
1.25      thorpej  1342: pool_print(pp, modif)
1.3       pk       1343:        struct pool *pp;
1.25      thorpej  1344:        const char *modif;
1.21      thorpej  1345: {
                   1346:        int s;
                   1347:
                   1348:        s = splimp();
1.25      thorpej  1349:        if (simple_lock_try(&pp->pr_slock) == 0) {
                   1350:                printf("pool %s is locked; try again later\n",
                   1351:                    pp->pr_wchan);
                   1352:                splx(s);
                   1353:                return;
                   1354:        }
                   1355:        pool_print1(pp, modif, printf);
1.21      thorpej  1356:        simple_unlock(&pp->pr_slock);
                   1357:        splx(s);
                   1358: }
                   1359:
1.25      thorpej  1360: void
                   1361: pool_printit(pp, modif, pr)
                   1362:        struct pool *pp;
                   1363:        const char *modif;
                   1364:        void (*pr) __P((const char *, ...));
                   1365: {
                   1366:        int didlock = 0;
                   1367:
                   1368:        if (pp == NULL) {
                   1369:                (*pr)("Must specify a pool to print.\n");
                   1370:                return;
                   1371:        }
                   1372:
                   1373:        /*
                   1374:         * Called from DDB; interrupts should be blocked, and all
                   1375:         * other processors should be paused.  We can skip locking
                   1376:         * the pool in this case.
                   1377:         *
                   1378:         * We do a simple_lock_try() just to print the lock
                   1379:         * status, however.
                   1380:         */
                   1381:
                   1382:        if (simple_lock_try(&pp->pr_slock) == 0)
                   1383:                (*pr)("WARNING: pool %s is locked\n", pp->pr_wchan);
                   1384:        else
                   1385:                didlock = 1;
                   1386:
                   1387:        pool_print1(pp, modif, pr);
                   1388:
                   1389:        if (didlock)
                   1390:                simple_unlock(&pp->pr_slock);
                   1391: }
                   1392:
1.21      thorpej  1393: static void
1.25      thorpej  1394: pool_print1(pp, modif, pr)
1.21      thorpej  1395:        struct pool *pp;
1.25      thorpej  1396:        const char *modif;
                   1397:        void (*pr) __P((const char *, ...));
1.3       pk       1398: {
1.25      thorpej  1399:        struct pool_item_header *ph;
                   1400: #ifdef DIAGNOSTIC
                   1401:        struct pool_item *pi;
                   1402: #endif
                   1403:        int print_log = 0, print_pagelist = 0;
                   1404:        char c;
                   1405:
                   1406:        while ((c = *modif++) != '\0') {
                   1407:                if (c == 'l')
                   1408:                        print_log = 1;
                   1409:                if (c == 'p')
                   1410:                        print_pagelist = 1;
                   1411:                modif++;
                   1412:        }
                   1413:
                   1414:        (*pr)("POOL %s: size %u, align %u, ioff %u, roflags 0x%08x\n",
                   1415:            pp->pr_wchan, pp->pr_size, pp->pr_align, pp->pr_itemoffset,
                   1416:            pp->pr_roflags);
                   1417:        (*pr)("\tpagesz %u, mtype %d\n", pp->pr_pagesz, pp->pr_mtype);
                   1418:        (*pr)("\talloc %p, release %p\n", pp->pr_alloc, pp->pr_free);
                   1419:        (*pr)("\tminitems %u, minpages %u, maxpages %u, npages %u\n",
                   1420:            pp->pr_minitems, pp->pr_minpages, pp->pr_maxpages, pp->pr_npages);
                   1421:        (*pr)("\titemsperpage %u, nitems %u, nout %u, hardlimit %u\n",
                   1422:            pp->pr_itemsperpage, pp->pr_nitems, pp->pr_nout, pp->pr_hardlimit);
                   1423:
                   1424:        (*pr)("\n\tnget %lu, nfail %lu, nput %lu\n",
                   1425:            pp->pr_nget, pp->pr_nfail, pp->pr_nput);
                   1426:        (*pr)("\tnpagealloc %lu, npagefree %lu, hiwat %u, nidle %lu\n",
                   1427:            pp->pr_npagealloc, pp->pr_npagefree, pp->pr_hiwat, pp->pr_nidle);
                   1428:
                   1429:        if (print_pagelist == 0)
                   1430:                goto skip_pagelist;
                   1431:
                   1432:        if ((ph = TAILQ_FIRST(&pp->pr_pagelist)) != NULL)
                   1433:                (*pr)("\n\tpage list:\n");
                   1434:        for (; ph != NULL; ph = TAILQ_NEXT(ph, ph_pagelist)) {
                   1435:                (*pr)("\t\tpage %p, nmissing %d, time %lu,%lu\n",
                   1436:                    ph->ph_page, ph->ph_nmissing,
                   1437:                    (u_long)ph->ph_time.tv_sec,
                   1438:                    (u_long)ph->ph_time.tv_usec);
                   1439: #ifdef DIAGNOSTIC
                   1440:                for (pi = TAILQ_FIRST(&ph->ph_itemlist); pi != NULL;
                   1441:                     pi = TAILQ_NEXT(pi, pi_list)) {
                   1442:                        if (pi->pi_magic != PI_MAGIC) {
                   1443:                                (*pr)("\t\t\titem %p, magic 0x%x\n",
                   1444:                                    pi, pi->pi_magic);
                   1445:                        }
                   1446:                }
                   1447: #endif
                   1448:        }
                   1449:        if (pp->pr_curpage == NULL)
                   1450:                (*pr)("\tno current page\n");
                   1451:        else
                   1452:                (*pr)("\tcurpage %p\n", pp->pr_curpage->ph_page);
                   1453:
                   1454:  skip_pagelist:
                   1455:
                   1456:        if (print_log == 0)
                   1457:                goto skip_log;
                   1458:
                   1459:        (*pr)("\n");
                   1460:        if ((pp->pr_roflags & PR_LOGGING) == 0)
                   1461:                (*pr)("\tno log\n");
                   1462:        else
                   1463:                pr_printlog(pp, NULL, pr);
1.3       pk       1464:
1.25      thorpej  1465:  skip_log:
1.3       pk       1466:
1.25      thorpej  1467:        pr_enter_check(pp, pr);
1.3       pk       1468: }
                   1469:
                   1470: int
                   1471: pool_chk(pp, label)
                   1472:        struct pool *pp;
                   1473:        char *label;
                   1474: {
                   1475:        struct pool_item_header *ph;
                   1476:        int r = 0;
                   1477:
1.21      thorpej  1478:        simple_lock(&pp->pr_slock);
1.3       pk       1479:
                   1480:        for (ph = TAILQ_FIRST(&pp->pr_pagelist); ph != NULL;
                   1481:             ph = TAILQ_NEXT(ph, ph_pagelist)) {
                   1482:
                   1483:                struct pool_item *pi;
                   1484:                int n;
                   1485:                caddr_t page;
                   1486:
                   1487:                page = (caddr_t)((u_long)ph & pp->pr_pagemask);
1.20      thorpej  1488:                if (page != ph->ph_page &&
                   1489:                    (pp->pr_roflags & PR_PHINPAGE) != 0) {
1.3       pk       1490:                        if (label != NULL)
                   1491:                                printf("%s: ", label);
1.16      briggs   1492:                        printf("pool(%p:%s): page inconsistency: page %p;"
                   1493:                               " at page head addr %p (p %p)\n", pp,
1.3       pk       1494:                                pp->pr_wchan, ph->ph_page,
                   1495:                                ph, page);
                   1496:                        r++;
                   1497:                        goto out;
                   1498:                }
                   1499:
                   1500:                for (pi = TAILQ_FIRST(&ph->ph_itemlist), n = 0;
                   1501:                     pi != NULL;
                   1502:                     pi = TAILQ_NEXT(pi,pi_list), n++) {
                   1503:
                   1504: #ifdef DIAGNOSTIC
                   1505:                        if (pi->pi_magic != PI_MAGIC) {
                   1506:                                if (label != NULL)
                   1507:                                        printf("%s: ", label);
                   1508:                                printf("pool(%s): free list modified: magic=%x;"
                   1509:                                       " page %p; item ordinal %d;"
                   1510:                                       " addr %p (p %p)\n",
                   1511:                                        pp->pr_wchan, pi->pi_magic, ph->ph_page,
                   1512:                                        n, pi, page);
                   1513:                                panic("pool");
                   1514:                        }
                   1515: #endif
                   1516:                        page = (caddr_t)((u_long)pi & pp->pr_pagemask);
                   1517:                        if (page == ph->ph_page)
                   1518:                                continue;
                   1519:
                   1520:                        if (label != NULL)
                   1521:                                printf("%s: ", label);
1.16      briggs   1522:                        printf("pool(%p:%s): page inconsistency: page %p;"
                   1523:                               " item ordinal %d; addr %p (p %p)\n", pp,
1.3       pk       1524:                                pp->pr_wchan, ph->ph_page,
                   1525:                                n, pi, page);
                   1526:                        r++;
                   1527:                        goto out;
                   1528:                }
                   1529:        }
                   1530: out:
1.21      thorpej  1531:        simple_unlock(&pp->pr_slock);
1.3       pk       1532:        return (r);
                   1533: }

CVSweb <webmaster@jp.NetBSD.org>