[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 1.41

1.41    ! sommerfe    1: /*     $NetBSD: subr_pool.c,v 1.40 2000/08/12 16:28:30 sommerfeld 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>
1.3       pk         53:
                     54: #include <uvm/uvm.h>
                     55:
1.1       pk         56: /*
                     57:  * Pool resource management utility.
1.3       pk         58:  *
                     59:  * Memory is allocated in pages which are split into pieces according
                     60:  * to the pool item size. Each page is kept on a list headed by `pr_pagelist'
                     61:  * in the pool structure and the individual pool items are on a linked list
                     62:  * headed by `ph_itemlist' in each page header. The memory for building
                     63:  * the page list is either taken from the allocated pages themselves (for
                     64:  * small pool items) or taken from an internal pool of page headers (`phpool').
1.1       pk         65:  */
                     66:
1.3       pk         67: /* List of all pools */
1.5       thorpej    68: TAILQ_HEAD(,pool) pool_head = TAILQ_HEAD_INITIALIZER(pool_head);
1.3       pk         69:
                     70: /* Private pool for page header structures */
                     71: static struct pool phpool;
                     72:
                     73: /* # of seconds to retain page after last use */
                     74: int pool_inactive_time = 10;
                     75:
                     76: /* Next candidate for drainage (see pool_drain()) */
1.23      thorpej    77: static struct pool     *drainpp;
                     78:
                     79: /* This spin lock protects both pool_head and drainpp. */
                     80: struct simplelock pool_head_slock = SIMPLELOCK_INITIALIZER;
1.3       pk         81:
                     82: struct pool_item_header {
                     83:        /* Page headers */
                     84:        TAILQ_ENTRY(pool_item_header)
                     85:                                ph_pagelist;    /* pool page list */
                     86:        TAILQ_HEAD(,pool_item)  ph_itemlist;    /* chunk list for this page */
                     87:        LIST_ENTRY(pool_item_header)
                     88:                                ph_hashlist;    /* Off-page page headers */
                     89:        int                     ph_nmissing;    /* # of chunks in use */
                     90:        caddr_t                 ph_page;        /* this page's address */
                     91:        struct timeval          ph_time;        /* last referenced */
                     92: };
                     93:
1.1       pk         94: struct pool_item {
1.3       pk         95: #ifdef DIAGNOSTIC
                     96:        int pi_magic;
1.33      chs        97: #endif
1.25      thorpej    98: #define        PI_MAGIC 0xdeadbeef
1.3       pk         99:        /* Other entries use only this list entry */
                    100:        TAILQ_ENTRY(pool_item)  pi_list;
                    101: };
                    102:
                    103:
1.25      thorpej   104: #define        PR_HASH_INDEX(pp,addr) \
1.3       pk        105:        (((u_long)(addr) >> (pp)->pr_pageshift) & (PR_HASHTABSIZE - 1))
                    106:
                    107:
                    108:
                    109: static struct pool_item_header
                    110:                *pr_find_pagehead __P((struct pool *, caddr_t));
                    111: static void    pr_rmpage __P((struct pool *, struct pool_item_header *));
1.20      thorpej   112: static int     pool_catchup __P((struct pool *));
1.21      thorpej   113: static void    pool_prime_page __P((struct pool *, caddr_t));
1.3       pk        114: static void    *pool_page_alloc __P((unsigned long, int, int));
                    115: static void    pool_page_free __P((void *, unsigned long, int));
                    116:
1.25      thorpej   117: static void pool_print1 __P((struct pool *, const char *,
                    118:        void (*)(const char *, ...)));
1.3       pk        119:
                    120: /*
                    121:  * Pool log entry. An array of these is allocated in pool_create().
                    122:  */
                    123: struct pool_log {
                    124:        const char      *pl_file;
                    125:        long            pl_line;
                    126:        int             pl_action;
1.25      thorpej   127: #define        PRLOG_GET       1
                    128: #define        PRLOG_PUT       2
1.3       pk        129:        void            *pl_addr;
1.1       pk        130: };
                    131:
1.3       pk        132: /* Number of entries in pool log buffers */
1.17      thorpej   133: #ifndef POOL_LOGSIZE
                    134: #define        POOL_LOGSIZE    10
                    135: #endif
                    136:
                    137: int pool_logsize = POOL_LOGSIZE;
1.1       pk        138:
1.25      thorpej   139: #ifdef DIAGNOSTIC
1.3       pk        140: static void    pr_log __P((struct pool *, void *, int, const char *, long));
1.25      thorpej   141: static void    pr_printlog __P((struct pool *, struct pool_item *,
                    142:                    void (*)(const char *, ...)));
                    143: static void    pr_enter __P((struct pool *, const char *, long));
                    144: static void    pr_leave __P((struct pool *));
                    145: static void    pr_enter_check __P((struct pool *,
                    146:                    void (*)(const char *, ...)));
1.3       pk        147:
                    148: static __inline__ void
                    149: pr_log(pp, v, action, file, line)
                    150:        struct pool     *pp;
                    151:        void            *v;
                    152:        int             action;
                    153:        const char      *file;
                    154:        long            line;
                    155: {
                    156:        int n = pp->pr_curlogentry;
                    157:        struct pool_log *pl;
                    158:
1.20      thorpej   159:        if ((pp->pr_roflags & PR_LOGGING) == 0)
1.3       pk        160:                return;
                    161:
                    162:        /*
                    163:         * Fill in the current entry. Wrap around and overwrite
                    164:         * the oldest entry if necessary.
                    165:         */
                    166:        pl = &pp->pr_log[n];
                    167:        pl->pl_file = file;
                    168:        pl->pl_line = line;
                    169:        pl->pl_action = action;
                    170:        pl->pl_addr = v;
                    171:        if (++n >= pp->pr_logsize)
                    172:                n = 0;
                    173:        pp->pr_curlogentry = n;
                    174: }
                    175:
                    176: static void
1.25      thorpej   177: pr_printlog(pp, pi, pr)
1.3       pk        178:        struct pool *pp;
1.25      thorpej   179:        struct pool_item *pi;
                    180:        void (*pr) __P((const char *, ...));
1.3       pk        181: {
                    182:        int i = pp->pr_logsize;
                    183:        int n = pp->pr_curlogentry;
                    184:
1.20      thorpej   185:        if ((pp->pr_roflags & PR_LOGGING) == 0)
1.3       pk        186:                return;
                    187:
                    188:        /*
                    189:         * Print all entries in this pool's log.
                    190:         */
                    191:        while (i-- > 0) {
                    192:                struct pool_log *pl = &pp->pr_log[n];
                    193:                if (pl->pl_action != 0) {
1.25      thorpej   194:                        if (pi == NULL || pi == pl->pl_addr) {
                    195:                                (*pr)("\tlog entry %d:\n", i);
                    196:                                (*pr)("\t\taction = %s, addr = %p\n",
                    197:                                    pl->pl_action == PRLOG_GET ? "get" : "put",
                    198:                                    pl->pl_addr);
                    199:                                (*pr)("\t\tfile: %s at line %lu\n",
                    200:                                    pl->pl_file, pl->pl_line);
                    201:                        }
1.3       pk        202:                }
                    203:                if (++n >= pp->pr_logsize)
                    204:                        n = 0;
                    205:        }
                    206: }
1.25      thorpej   207:
                    208: static __inline__ void
                    209: pr_enter(pp, file, line)
                    210:        struct pool *pp;
                    211:        const char *file;
                    212:        long line;
                    213: {
                    214:
1.34      thorpej   215:        if (__predict_false(pp->pr_entered_file != NULL)) {
1.25      thorpej   216:                printf("pool %s: reentrancy at file %s line %ld\n",
                    217:                    pp->pr_wchan, file, line);
                    218:                printf("         previous entry at file %s line %ld\n",
                    219:                    pp->pr_entered_file, pp->pr_entered_line);
                    220:                panic("pr_enter");
                    221:        }
                    222:
                    223:        pp->pr_entered_file = file;
                    224:        pp->pr_entered_line = line;
                    225: }
                    226:
                    227: static __inline__ void
                    228: pr_leave(pp)
                    229:        struct pool *pp;
                    230: {
                    231:
1.34      thorpej   232:        if (__predict_false(pp->pr_entered_file == NULL)) {
1.25      thorpej   233:                printf("pool %s not entered?\n", pp->pr_wchan);
                    234:                panic("pr_leave");
                    235:        }
                    236:
                    237:        pp->pr_entered_file = NULL;
                    238:        pp->pr_entered_line = 0;
                    239: }
                    240:
                    241: static __inline__ void
                    242: pr_enter_check(pp, pr)
                    243:        struct pool *pp;
                    244:        void (*pr) __P((const char *, ...));
                    245: {
                    246:
                    247:        if (pp->pr_entered_file != NULL)
                    248:                (*pr)("\n\tcurrently entered from file %s line %ld\n",
                    249:                    pp->pr_entered_file, pp->pr_entered_line);
                    250: }
1.3       pk        251: #else
1.25      thorpej   252: #define        pr_log(pp, v, action, file, line)
                    253: #define        pr_printlog(pp, pi, pr)
                    254: #define        pr_enter(pp, file, line)
                    255: #define        pr_leave(pp)
                    256: #define        pr_enter_check(pp, pr)
                    257: #endif /* DIAGNOSTIC */
1.3       pk        258:
                    259: /*
                    260:  * Return the pool page header based on page address.
                    261:  */
                    262: static __inline__ struct pool_item_header *
                    263: pr_find_pagehead(pp, page)
                    264:        struct pool *pp;
                    265:        caddr_t page;
                    266: {
                    267:        struct pool_item_header *ph;
                    268:
1.20      thorpej   269:        if ((pp->pr_roflags & PR_PHINPAGE) != 0)
1.3       pk        270:                return ((struct pool_item_header *)(page + pp->pr_phoffset));
                    271:
                    272:        for (ph = LIST_FIRST(&pp->pr_hashtab[PR_HASH_INDEX(pp, page)]);
                    273:             ph != NULL;
                    274:             ph = LIST_NEXT(ph, ph_hashlist)) {
                    275:                if (ph->ph_page == page)
                    276:                        return (ph);
                    277:        }
                    278:        return (NULL);
                    279: }
                    280:
                    281: /*
                    282:  * Remove a page from the pool.
                    283:  */
                    284: static __inline__ void
                    285: pr_rmpage(pp, ph)
                    286:        struct pool *pp;
                    287:        struct pool_item_header *ph;
                    288: {
                    289:
                    290:        /*
1.7       thorpej   291:         * If the page was idle, decrement the idle page count.
1.3       pk        292:         */
1.6       thorpej   293:        if (ph->ph_nmissing == 0) {
                    294: #ifdef DIAGNOSTIC
                    295:                if (pp->pr_nidle == 0)
                    296:                        panic("pr_rmpage: nidle inconsistent");
1.20      thorpej   297:                if (pp->pr_nitems < pp->pr_itemsperpage)
                    298:                        panic("pr_rmpage: nitems inconsistent");
1.6       thorpej   299: #endif
                    300:                pp->pr_nidle--;
                    301:        }
1.7       thorpej   302:
1.20      thorpej   303:        pp->pr_nitems -= pp->pr_itemsperpage;
                    304:
1.7       thorpej   305:        /*
                    306:         * Unlink a page from the pool and release it.
                    307:         */
                    308:        TAILQ_REMOVE(&pp->pr_pagelist, ph, ph_pagelist);
                    309:        (*pp->pr_free)(ph->ph_page, pp->pr_pagesz, pp->pr_mtype);
                    310:        pp->pr_npages--;
                    311:        pp->pr_npagefree++;
1.6       thorpej   312:
1.22      chs       313:        if ((pp->pr_roflags & PR_PHINPAGE) == 0) {
1.27      pk        314:                int s;
1.22      chs       315:                LIST_REMOVE(ph, ph_hashlist);
1.27      pk        316:                s = splhigh();
1.22      chs       317:                pool_put(&phpool, ph);
1.27      pk        318:                splx(s);
1.22      chs       319:        }
                    320:
1.3       pk        321:        if (pp->pr_curpage == ph) {
                    322:                /*
                    323:                 * Find a new non-empty page header, if any.
                    324:                 * Start search from the page head, to increase the
                    325:                 * chance for "high water" pages to be freed.
                    326:                 */
                    327:                for (ph = TAILQ_FIRST(&pp->pr_pagelist); ph != NULL;
                    328:                     ph = TAILQ_NEXT(ph, ph_pagelist))
                    329:                        if (TAILQ_FIRST(&ph->ph_itemlist) != NULL)
                    330:                                break;
                    331:
                    332:                pp->pr_curpage = ph;
1.21      thorpej   333:        }
1.3       pk        334: }
                    335:
                    336: /*
                    337:  * Allocate and initialize a pool.
                    338:  */
1.1       pk        339: struct pool *
1.3       pk        340: pool_create(size, align, ioff, nitems, wchan, pagesz, alloc, release, mtype)
1.1       pk        341:        size_t  size;
1.3       pk        342:        u_int   align;
                    343:        u_int   ioff;
1.1       pk        344:        int     nitems;
1.21      thorpej   345:        const char *wchan;
1.3       pk        346:        size_t  pagesz;
                    347:        void    *(*alloc) __P((unsigned long, int, int));
                    348:        void    (*release) __P((void *, unsigned long, int));
1.1       pk        349:        int     mtype;
                    350: {
                    351:        struct pool *pp;
1.3       pk        352:        int flags;
1.1       pk        353:
1.3       pk        354:        pp = (struct pool *)malloc(sizeof(*pp), M_POOL, M_NOWAIT);
                    355:        if (pp == NULL)
1.1       pk        356:                return (NULL);
1.3       pk        357:
                    358:        flags = PR_FREEHEADER;
                    359:        pool_init(pp, size, align, ioff, flags, wchan, pagesz,
                    360:                  alloc, release, mtype);
                    361:
                    362:        if (nitems != 0) {
                    363:                if (pool_prime(pp, nitems, NULL) != 0) {
                    364:                        pool_destroy(pp);
                    365:                        return (NULL);
                    366:                }
1.1       pk        367:        }
                    368:
1.3       pk        369:        return (pp);
                    370: }
                    371:
                    372: /*
                    373:  * Initialize the given pool resource structure.
                    374:  *
                    375:  * We export this routine to allow other kernel parts to declare
                    376:  * static pools that must be initialized before malloc() is available.
                    377:  */
                    378: void
                    379: pool_init(pp, size, align, ioff, flags, wchan, pagesz, alloc, release, mtype)
                    380:        struct pool     *pp;
                    381:        size_t          size;
                    382:        u_int           align;
                    383:        u_int           ioff;
                    384:        int             flags;
1.21      thorpej   385:        const char      *wchan;
1.3       pk        386:        size_t          pagesz;
                    387:        void            *(*alloc) __P((unsigned long, int, int));
                    388:        void            (*release) __P((void *, unsigned long, int));
                    389:        int             mtype;
                    390: {
1.16      briggs    391:        int off, slack, i;
1.3       pk        392:
1.25      thorpej   393: #ifdef POOL_DIAGNOSTIC
                    394:        /*
                    395:         * Always log if POOL_DIAGNOSTIC is defined.
                    396:         */
                    397:        if (pool_logsize != 0)
                    398:                flags |= PR_LOGGING;
                    399: #endif
                    400:
1.3       pk        401:        /*
                    402:         * Check arguments and construct default values.
                    403:         */
1.36      pk        404:        if (!powerof2(pagesz))
1.3       pk        405:                panic("pool_init: page size invalid (%lx)\n", (u_long)pagesz);
                    406:
1.4       thorpej   407:        if (alloc == NULL && release == NULL) {
1.3       pk        408:                alloc = pool_page_alloc;
                    409:                release = pool_page_free;
1.4       thorpej   410:                pagesz = PAGE_SIZE;     /* Rounds to PAGE_SIZE anyhow. */
                    411:        } else if ((alloc != NULL && release != NULL) == 0) {
                    412:                /* If you specifiy one, must specify both. */
                    413:                panic("pool_init: must specify alloc and release together");
                    414:        }
                    415:
1.3       pk        416:        if (pagesz == 0)
                    417:                pagesz = PAGE_SIZE;
                    418:
                    419:        if (align == 0)
                    420:                align = ALIGN(1);
1.14      thorpej   421:
                    422:        if (size < sizeof(struct pool_item))
                    423:                size = sizeof(struct pool_item);
1.3       pk        424:
1.35      pk        425:        size = ALIGN(size);
                    426:        if (size >= pagesz)
                    427:                panic("pool_init: pool item size (%lu) too large",
                    428:                      (u_long)size);
                    429:
1.3       pk        430:        /*
                    431:         * Initialize the pool structure.
                    432:         */
                    433:        TAILQ_INIT(&pp->pr_pagelist);
                    434:        pp->pr_curpage = NULL;
                    435:        pp->pr_npages = 0;
                    436:        pp->pr_minitems = 0;
                    437:        pp->pr_minpages = 0;
                    438:        pp->pr_maxpages = UINT_MAX;
1.20      thorpej   439:        pp->pr_roflags = flags;
                    440:        pp->pr_flags = 0;
1.35      pk        441:        pp->pr_size = size;
1.3       pk        442:        pp->pr_align = align;
                    443:        pp->pr_wchan = wchan;
                    444:        pp->pr_mtype = mtype;
                    445:        pp->pr_alloc = alloc;
                    446:        pp->pr_free = release;
                    447:        pp->pr_pagesz = pagesz;
                    448:        pp->pr_pagemask = ~(pagesz - 1);
                    449:        pp->pr_pageshift = ffs(pagesz) - 1;
1.20      thorpej   450:        pp->pr_nitems = 0;
                    451:        pp->pr_nout = 0;
                    452:        pp->pr_hardlimit = UINT_MAX;
                    453:        pp->pr_hardlimit_warning = NULL;
1.31      thorpej   454:        pp->pr_hardlimit_ratecap.tv_sec = 0;
                    455:        pp->pr_hardlimit_ratecap.tv_usec = 0;
                    456:        pp->pr_hardlimit_warning_last.tv_sec = 0;
                    457:        pp->pr_hardlimit_warning_last.tv_usec = 0;
1.3       pk        458:
                    459:        /*
                    460:         * Decide whether to put the page header off page to avoid
                    461:         * wasting too large a part of the page. Off-page page headers
                    462:         * go on a hash table, so we can match a returned item
                    463:         * with its header based on the page address.
                    464:         * We use 1/16 of the page size as the threshold (XXX: tune)
                    465:         */
                    466:        if (pp->pr_size < pagesz/16) {
                    467:                /* Use the end of the page for the page header */
1.20      thorpej   468:                pp->pr_roflags |= PR_PHINPAGE;
1.3       pk        469:                pp->pr_phoffset = off =
                    470:                        pagesz - ALIGN(sizeof(struct pool_item_header));
1.2       pk        471:        } else {
1.3       pk        472:                /* The page header will be taken from our page header pool */
                    473:                pp->pr_phoffset = 0;
                    474:                off = pagesz;
1.16      briggs    475:                for (i = 0; i < PR_HASHTABSIZE; i++) {
                    476:                        LIST_INIT(&pp->pr_hashtab[i]);
                    477:                }
1.2       pk        478:        }
1.1       pk        479:
1.3       pk        480:        /*
                    481:         * Alignment is to take place at `ioff' within the item. This means
                    482:         * we must reserve up to `align - 1' bytes on the page to allow
                    483:         * appropriate positioning of each item.
                    484:         *
                    485:         * Silently enforce `0 <= ioff < align'.
                    486:         */
                    487:        pp->pr_itemoffset = ioff = ioff % align;
                    488:        pp->pr_itemsperpage = (off - ((align - ioff) % align)) / pp->pr_size;
                    489:
                    490:        /*
                    491:         * Use the slack between the chunks and the page header
                    492:         * for "cache coloring".
                    493:         */
                    494:        slack = off - pp->pr_itemsperpage * pp->pr_size;
                    495:        pp->pr_maxcolor = (slack / align) * align;
                    496:        pp->pr_curcolor = 0;
                    497:
                    498:        pp->pr_nget = 0;
                    499:        pp->pr_nfail = 0;
                    500:        pp->pr_nput = 0;
                    501:        pp->pr_npagealloc = 0;
                    502:        pp->pr_npagefree = 0;
1.1       pk        503:        pp->pr_hiwat = 0;
1.8       thorpej   504:        pp->pr_nidle = 0;
1.3       pk        505:
1.25      thorpej   506:        if (flags & PR_LOGGING) {
                    507:                if (kmem_map == NULL ||
                    508:                    (pp->pr_log = malloc(pool_logsize * sizeof(struct pool_log),
                    509:                     M_TEMP, M_NOWAIT)) == NULL)
1.20      thorpej   510:                        pp->pr_roflags &= ~PR_LOGGING;
1.3       pk        511:                pp->pr_curlogentry = 0;
                    512:                pp->pr_logsize = pool_logsize;
                    513:        }
1.25      thorpej   514:
                    515:        pp->pr_entered_file = NULL;
                    516:        pp->pr_entered_line = 0;
1.3       pk        517:
1.21      thorpej   518:        simple_lock_init(&pp->pr_slock);
1.1       pk        519:
1.3       pk        520:        /*
                    521:         * Initialize private page header pool if we haven't done so yet.
1.23      thorpej   522:         * XXX LOCKING.
1.3       pk        523:         */
                    524:        if (phpool.pr_size == 0) {
                    525:                pool_init(&phpool, sizeof(struct pool_item_header), 0, 0,
                    526:                          0, "phpool", 0, 0, 0, 0);
1.1       pk        527:        }
                    528:
1.23      thorpej   529:        /* Insert into the list of all pools. */
                    530:        simple_lock(&pool_head_slock);
                    531:        TAILQ_INSERT_TAIL(&pool_head, pp, pr_poollist);
                    532:        simple_unlock(&pool_head_slock);
1.1       pk        533: }
                    534:
                    535: /*
                    536:  * De-commision a pool resource.
                    537:  */
                    538: void
                    539: pool_destroy(pp)
                    540:        struct pool *pp;
                    541: {
1.3       pk        542:        struct pool_item_header *ph;
                    543:
                    544: #ifdef DIAGNOSTIC
1.20      thorpej   545:        if (pp->pr_nout != 0) {
1.25      thorpej   546:                pr_printlog(pp, NULL, printf);
1.20      thorpej   547:                panic("pool_destroy: pool busy: still out: %u\n",
                    548:                    pp->pr_nout);
1.3       pk        549:        }
                    550: #endif
1.1       pk        551:
1.3       pk        552:        /* Remove all pages */
1.20      thorpej   553:        if ((pp->pr_roflags & PR_STATIC) == 0)
1.3       pk        554:                while ((ph = pp->pr_pagelist.tqh_first) != NULL)
                    555:                        pr_rmpage(pp, ph);
                    556:
                    557:        /* Remove from global pool list */
1.23      thorpej   558:        simple_lock(&pool_head_slock);
1.3       pk        559:        TAILQ_REMOVE(&pool_head, pp, pr_poollist);
1.23      thorpej   560:        /* XXX Only clear this if we were drainpp? */
1.3       pk        561:        drainpp = NULL;
1.23      thorpej   562:        simple_unlock(&pool_head_slock);
1.3       pk        563:
1.20      thorpej   564:        if ((pp->pr_roflags & PR_LOGGING) != 0)
1.3       pk        565:                free(pp->pr_log, M_TEMP);
1.2       pk        566:
1.20      thorpej   567:        if (pp->pr_roflags & PR_FREEHEADER)
1.3       pk        568:                free(pp, M_POOL);
1.1       pk        569: }
                    570:
                    571:
                    572: /*
1.3       pk        573:  * Grab an item from the pool; must be called at appropriate spl level
1.1       pk        574:  */
1.3       pk        575: void *
                    576: _pool_get(pp, flags, file, line)
                    577:        struct pool *pp;
                    578:        int flags;
                    579:        const char *file;
                    580:        long line;
1.1       pk        581: {
                    582:        void *v;
                    583:        struct pool_item *pi;
1.3       pk        584:        struct pool_item_header *ph;
1.1       pk        585:
1.2       pk        586: #ifdef DIAGNOSTIC
1.34      thorpej   587:        if (__predict_false((pp->pr_roflags & PR_STATIC) &&
                    588:                            (flags & PR_MALLOCOK))) {
1.25      thorpej   589:                pr_printlog(pp, NULL, printf);
1.2       pk        590:                panic("pool_get: static");
1.3       pk        591:        }
1.2       pk        592: #endif
                    593:
1.37      sommerfe  594:        if (__predict_false(curproc == NULL && doing_shutdown == 0 &&
                    595:                            (flags & PR_WAITOK) != 0))
1.3       pk        596:                panic("pool_get: must have NOWAIT");
1.1       pk        597:
1.21      thorpej   598:        simple_lock(&pp->pr_slock);
1.25      thorpej   599:        pr_enter(pp, file, line);
1.20      thorpej   600:
                    601:  startover:
                    602:        /*
                    603:         * Check to see if we've reached the hard limit.  If we have,
                    604:         * and we can wait, then wait until an item has been returned to
                    605:         * the pool.
                    606:         */
                    607: #ifdef DIAGNOSTIC
1.34      thorpej   608:        if (__predict_false(pp->pr_nout > pp->pr_hardlimit)) {
1.25      thorpej   609:                pr_leave(pp);
1.21      thorpej   610:                simple_unlock(&pp->pr_slock);
1.20      thorpej   611:                panic("pool_get: %s: crossed hard limit", pp->pr_wchan);
                    612:        }
                    613: #endif
1.34      thorpej   614:        if (__predict_false(pp->pr_nout == pp->pr_hardlimit)) {
1.29      sommerfe  615:                if ((flags & PR_WAITOK) && !(flags & PR_LIMITFAIL)) {
1.20      thorpej   616:                        /*
                    617:                         * XXX: A warning isn't logged in this case.  Should
                    618:                         * it be?
                    619:                         */
                    620:                        pp->pr_flags |= PR_WANTED;
1.25      thorpej   621:                        pr_leave(pp);
1.40      sommerfe  622:                        ltsleep(pp, PSWP, pp->pr_wchan, 0, &pp->pr_slock);
1.25      thorpej   623:                        pr_enter(pp, file, line);
1.20      thorpej   624:                        goto startover;
                    625:                }
1.31      thorpej   626:
                    627:                /*
                    628:                 * Log a message that the hard limit has been hit.
                    629:                 */
                    630:                if (pp->pr_hardlimit_warning != NULL &&
                    631:                    ratecheck(&pp->pr_hardlimit_warning_last,
                    632:                              &pp->pr_hardlimit_ratecap))
                    633:                        log(LOG_ERR, "%s\n", pp->pr_hardlimit_warning);
1.21      thorpej   634:
                    635:                if (flags & PR_URGENT)
                    636:                        panic("pool_get: urgent");
                    637:
                    638:                pp->pr_nfail++;
                    639:
1.25      thorpej   640:                pr_leave(pp);
1.21      thorpej   641:                simple_unlock(&pp->pr_slock);
1.20      thorpej   642:                return (NULL);
                    643:        }
                    644:
1.3       pk        645:        /*
                    646:         * The convention we use is that if `curpage' is not NULL, then
                    647:         * it points at a non-empty bucket. In particular, `curpage'
                    648:         * never points at a page header which has PR_PHINPAGE set and
                    649:         * has no items in its bucket.
                    650:         */
1.20      thorpej   651:        if ((ph = pp->pr_curpage) == NULL) {
1.15      pk        652:                void *v;
                    653:
1.20      thorpej   654: #ifdef DIAGNOSTIC
                    655:                if (pp->pr_nitems != 0) {
1.21      thorpej   656:                        simple_unlock(&pp->pr_slock);
1.20      thorpej   657:                        printf("pool_get: %s: curpage NULL, nitems %u\n",
                    658:                            pp->pr_wchan, pp->pr_nitems);
                    659:                        panic("pool_get: nitems inconsistent\n");
                    660:                }
                    661: #endif
                    662:
1.21      thorpej   663:                /*
                    664:                 * Call the back-end page allocator for more memory.
                    665:                 * Release the pool lock, as the back-end page allocator
                    666:                 * may block.
                    667:                 */
1.25      thorpej   668:                pr_leave(pp);
1.21      thorpej   669:                simple_unlock(&pp->pr_slock);
                    670:                v = (*pp->pr_alloc)(pp->pr_pagesz, flags, pp->pr_mtype);
                    671:                simple_lock(&pp->pr_slock);
1.25      thorpej   672:                pr_enter(pp, file, line);
1.15      pk        673:
1.21      thorpej   674:                if (v == NULL) {
                    675:                        /*
                    676:                         * We were unable to allocate a page, but
                    677:                         * we released the lock during allocation,
                    678:                         * so perhaps items were freed back to the
                    679:                         * pool.  Check for this case.
                    680:                         */
                    681:                        if (pp->pr_curpage != NULL)
                    682:                                goto startover;
1.15      pk        683:
1.3       pk        684:                        if (flags & PR_URGENT)
                    685:                                panic("pool_get: urgent");
1.21      thorpej   686:
1.3       pk        687:                        if ((flags & PR_WAITOK) == 0) {
                    688:                                pp->pr_nfail++;
1.25      thorpej   689:                                pr_leave(pp);
1.21      thorpej   690:                                simple_unlock(&pp->pr_slock);
1.1       pk        691:                                return (NULL);
1.3       pk        692:                        }
                    693:
1.15      pk        694:                        /*
                    695:                         * Wait for items to be returned to this pool.
1.21      thorpej   696:                         *
1.15      pk        697:                         * XXX: we actually want to wait just until
                    698:                         * the page allocator has memory again. Depending
                    699:                         * on this pool's usage, we might get stuck here
                    700:                         * for a long time.
1.20      thorpej   701:                         *
                    702:                         * XXX: maybe we should wake up once a second and
                    703:                         * try again?
1.15      pk        704:                         */
1.1       pk        705:                        pp->pr_flags |= PR_WANTED;
1.25      thorpej   706:                        pr_leave(pp);
1.40      sommerfe  707:                        ltsleep(pp, PSWP, pp->pr_wchan, 0, &pp->pr_slock);
1.25      thorpej   708:                        pr_enter(pp, file, line);
1.20      thorpej   709:                        goto startover;
1.1       pk        710:                }
1.3       pk        711:
1.15      pk        712:                /* We have more memory; add it to the pool */
                    713:                pp->pr_npagealloc++;
                    714:                pool_prime_page(pp, v);
                    715:
1.20      thorpej   716:                /* Start the allocation process over. */
                    717:                goto startover;
1.3       pk        718:        }
                    719:
1.34      thorpej   720:        if (__predict_false((v = pi = TAILQ_FIRST(&ph->ph_itemlist)) == NULL)) {
1.25      thorpej   721:                pr_leave(pp);
1.21      thorpej   722:                simple_unlock(&pp->pr_slock);
1.3       pk        723:                panic("pool_get: %s: page empty", pp->pr_wchan);
1.21      thorpej   724:        }
1.20      thorpej   725: #ifdef DIAGNOSTIC
1.34      thorpej   726:        if (__predict_false(pp->pr_nitems == 0)) {
1.25      thorpej   727:                pr_leave(pp);
1.21      thorpej   728:                simple_unlock(&pp->pr_slock);
1.20      thorpej   729:                printf("pool_get: %s: items on itemlist, nitems %u\n",
                    730:                    pp->pr_wchan, pp->pr_nitems);
                    731:                panic("pool_get: nitems inconsistent\n");
                    732:        }
                    733: #endif
1.3       pk        734:        pr_log(pp, v, PRLOG_GET, file, line);
                    735:
                    736: #ifdef DIAGNOSTIC
1.34      thorpej   737:        if (__predict_false(pi->pi_magic != PI_MAGIC)) {
1.25      thorpej   738:                pr_printlog(pp, pi, printf);
1.3       pk        739:                panic("pool_get(%s): free list modified: magic=%x; page %p;"
                    740:                       " item addr %p\n",
                    741:                        pp->pr_wchan, pi->pi_magic, ph->ph_page, pi);
                    742:        }
                    743: #endif
                    744:
                    745:        /*
                    746:         * Remove from item list.
                    747:         */
                    748:        TAILQ_REMOVE(&ph->ph_itemlist, pi, pi_list);
1.20      thorpej   749:        pp->pr_nitems--;
                    750:        pp->pr_nout++;
1.6       thorpej   751:        if (ph->ph_nmissing == 0) {
                    752: #ifdef DIAGNOSTIC
1.34      thorpej   753:                if (__predict_false(pp->pr_nidle == 0))
1.6       thorpej   754:                        panic("pool_get: nidle inconsistent");
                    755: #endif
                    756:                pp->pr_nidle--;
                    757:        }
1.3       pk        758:        ph->ph_nmissing++;
                    759:        if (TAILQ_FIRST(&ph->ph_itemlist) == NULL) {
1.21      thorpej   760: #ifdef DIAGNOSTIC
1.34      thorpej   761:                if (__predict_false(ph->ph_nmissing != pp->pr_itemsperpage)) {
1.25      thorpej   762:                        pr_leave(pp);
1.21      thorpej   763:                        simple_unlock(&pp->pr_slock);
                    764:                        panic("pool_get: %s: nmissing inconsistent",
                    765:                            pp->pr_wchan);
                    766:                }
                    767: #endif
1.3       pk        768:                /*
                    769:                 * Find a new non-empty page header, if any.
                    770:                 * Start search from the page head, to increase
                    771:                 * the chance for "high water" pages to be freed.
                    772:                 *
1.21      thorpej   773:                 * Migrate empty pages to the end of the list.  This
                    774:                 * will speed the update of curpage as pages become
                    775:                 * idle.  Empty pages intermingled with idle pages
                    776:                 * is no big deal.  As soon as a page becomes un-empty,
                    777:                 * it will move back to the head of the list.
1.3       pk        778:                 */
                    779:                TAILQ_REMOVE(&pp->pr_pagelist, ph, ph_pagelist);
1.21      thorpej   780:                TAILQ_INSERT_TAIL(&pp->pr_pagelist, ph, ph_pagelist);
                    781:                for (ph = TAILQ_FIRST(&pp->pr_pagelist); ph != NULL;
                    782:                     ph = TAILQ_NEXT(ph, ph_pagelist))
1.3       pk        783:                        if (TAILQ_FIRST(&ph->ph_itemlist) != NULL)
                    784:                                break;
                    785:
                    786:                pp->pr_curpage = ph;
1.1       pk        787:        }
1.3       pk        788:
                    789:        pp->pr_nget++;
1.20      thorpej   790:
                    791:        /*
                    792:         * If we have a low water mark and we are now below that low
                    793:         * water mark, add more items to the pool.
                    794:         */
                    795:        if (pp->pr_nitems < pp->pr_minitems && pool_catchup(pp) != 0) {
                    796:                /*
                    797:                 * XXX: Should we log a warning?  Should we set up a timeout
                    798:                 * to try again in a second or so?  The latter could break
                    799:                 * a caller's assumptions about interrupt protection, etc.
                    800:                 */
                    801:        }
                    802:
1.25      thorpej   803:        pr_leave(pp);
1.21      thorpej   804:        simple_unlock(&pp->pr_slock);
1.1       pk        805:        return (v);
                    806: }
                    807:
                    808: /*
1.3       pk        809:  * Return resource to the pool; must be called at appropriate spl level
1.1       pk        810:  */
1.3       pk        811: void
                    812: _pool_put(pp, v, file, line)
                    813:        struct pool *pp;
                    814:        void *v;
                    815:        const char *file;
                    816:        long line;
1.1       pk        817: {
                    818:        struct pool_item *pi = v;
1.3       pk        819:        struct pool_item_header *ph;
                    820:        caddr_t page;
1.21      thorpej   821:        int s;
1.3       pk        822:
                    823:        page = (caddr_t)((u_long)v & pp->pr_pagemask);
1.1       pk        824:
1.21      thorpej   825:        simple_lock(&pp->pr_slock);
1.25      thorpej   826:        pr_enter(pp, file, line);
1.30      thorpej   827:
                    828: #ifdef DIAGNOSTIC
1.34      thorpej   829:        if (__predict_false(pp->pr_nout == 0)) {
1.30      thorpej   830:                printf("pool %s: putting with none out\n",
                    831:                    pp->pr_wchan);
                    832:                panic("pool_put");
                    833:        }
                    834: #endif
1.3       pk        835:
                    836:        pr_log(pp, v, PRLOG_PUT, file, line);
                    837:
1.34      thorpej   838:        if (__predict_false((ph = pr_find_pagehead(pp, page)) == NULL)) {
1.25      thorpej   839:                pr_printlog(pp, NULL, printf);
1.3       pk        840:                panic("pool_put: %s: page header missing", pp->pr_wchan);
                    841:        }
1.28      thorpej   842:
                    843: #ifdef LOCKDEBUG
                    844:        /*
                    845:         * Check if we're freeing a locked simple lock.
                    846:         */
                    847:        simple_lock_freecheck((caddr_t)pi, ((caddr_t)pi) + pp->pr_size);
                    848: #endif
1.3       pk        849:
                    850:        /*
                    851:         * Return to item list.
                    852:         */
1.2       pk        853: #ifdef DIAGNOSTIC
1.3       pk        854:        pi->pi_magic = PI_MAGIC;
                    855: #endif
1.32      chs       856: #ifdef DEBUG
                    857:        {
                    858:                int i, *ip = v;
                    859:
                    860:                for (i = 0; i < pp->pr_size / sizeof(int); i++) {
                    861:                        *ip++ = PI_MAGIC;
                    862:                }
                    863:        }
                    864: #endif
                    865:
1.3       pk        866:        TAILQ_INSERT_HEAD(&ph->ph_itemlist, pi, pi_list);
                    867:        ph->ph_nmissing--;
                    868:        pp->pr_nput++;
1.20      thorpej   869:        pp->pr_nitems++;
                    870:        pp->pr_nout--;
1.3       pk        871:
                    872:        /* Cancel "pool empty" condition if it exists */
                    873:        if (pp->pr_curpage == NULL)
                    874:                pp->pr_curpage = ph;
                    875:
                    876:        if (pp->pr_flags & PR_WANTED) {
                    877:                pp->pr_flags &= ~PR_WANTED;
1.15      pk        878:                if (ph->ph_nmissing == 0)
                    879:                        pp->pr_nidle++;
1.25      thorpej   880:                pr_leave(pp);
1.21      thorpej   881:                simple_unlock(&pp->pr_slock);
1.3       pk        882:                wakeup((caddr_t)pp);
                    883:                return;
                    884:        }
                    885:
                    886:        /*
1.21      thorpej   887:         * If this page is now complete, do one of two things:
                    888:         *
                    889:         *      (1) If we have more pages than the page high water
                    890:         *          mark, free the page back to the system.
                    891:         *
                    892:         *      (2) Move it to the end of the page list, so that
                    893:         *          we minimize our chances of fragmenting the
                    894:         *          pool.  Idle pages migrate to the end (along with
                    895:         *          completely empty pages, so that we find un-empty
                    896:         *          pages more quickly when we update curpage) of the
                    897:         *          list so they can be more easily swept up by
                    898:         *          the pagedaemon when pages are scarce.
1.3       pk        899:         */
                    900:        if (ph->ph_nmissing == 0) {
1.6       thorpej   901:                pp->pr_nidle++;
1.3       pk        902:                if (pp->pr_npages > pp->pr_maxpages) {
                    903:                        pr_rmpage(pp, ph);
                    904:                } else {
                    905:                        TAILQ_REMOVE(&pp->pr_pagelist, ph, ph_pagelist);
                    906:                        TAILQ_INSERT_TAIL(&pp->pr_pagelist, ph, ph_pagelist);
                    907:
1.21      thorpej   908:                        /*
                    909:                         * Update the timestamp on the page.  A page must
                    910:                         * be idle for some period of time before it can
                    911:                         * be reclaimed by the pagedaemon.  This minimizes
                    912:                         * ping-pong'ing for memory.
                    913:                         */
                    914:                        s = splclock();
                    915:                        ph->ph_time = mono_time;
                    916:                        splx(s);
                    917:
                    918:                        /*
                    919:                         * Update the current page pointer.  Just look for
                    920:                         * the first page with any free items.
                    921:                         *
                    922:                         * XXX: Maybe we want an option to look for the
                    923:                         * page with the fewest available items, to minimize
                    924:                         * fragmentation?
                    925:                         */
1.3       pk        926:                        for (ph = TAILQ_FIRST(&pp->pr_pagelist); ph != NULL;
                    927:                             ph = TAILQ_NEXT(ph, ph_pagelist))
                    928:                                if (TAILQ_FIRST(&ph->ph_itemlist) != NULL)
                    929:                                        break;
1.1       pk        930:
1.3       pk        931:                        pp->pr_curpage = ph;
1.1       pk        932:                }
                    933:        }
1.21      thorpej   934:        /*
                    935:         * If the page has just become un-empty, move it to the head of
                    936:         * the list, and make it the current page.  The next allocation
                    937:         * will get the item from this page, instead of further fragmenting
                    938:         * the pool.
                    939:         */
                    940:        else if (ph->ph_nmissing == (pp->pr_itemsperpage - 1)) {
                    941:                TAILQ_REMOVE(&pp->pr_pagelist, ph, ph_pagelist);
                    942:                TAILQ_INSERT_HEAD(&pp->pr_pagelist, ph, ph_pagelist);
                    943:                pp->pr_curpage = ph;
                    944:        }
                    945:
1.25      thorpej   946:        pr_leave(pp);
1.21      thorpej   947:        simple_unlock(&pp->pr_slock);
1.3       pk        948:
1.1       pk        949: }
                    950:
                    951: /*
1.3       pk        952:  * Add N items to the pool.
1.1       pk        953:  */
                    954: int
1.2       pk        955: pool_prime(pp, n, storage)
1.1       pk        956:        struct pool *pp;
                    957:        int n;
1.2       pk        958:        caddr_t storage;
1.1       pk        959: {
1.3       pk        960:        caddr_t cp;
                    961:        int newnitems, newpages;
1.2       pk        962:
                    963: #ifdef DIAGNOSTIC
1.34      thorpej   964:        if (__predict_false(storage && !(pp->pr_roflags & PR_STATIC)))
1.2       pk        965:                panic("pool_prime: static");
                    966:        /* !storage && static caught below */
                    967: #endif
1.1       pk        968:
1.21      thorpej   969:        simple_lock(&pp->pr_slock);
                    970:
1.3       pk        971:        newnitems = pp->pr_minitems + n;
                    972:        newpages =
1.18      thorpej   973:                roundup(newnitems, pp->pr_itemsperpage) / pp->pr_itemsperpage
1.3       pk        974:                - pp->pr_minpages;
                    975:
                    976:        while (newpages-- > 0) {
1.20      thorpej   977:                if (pp->pr_roflags & PR_STATIC) {
1.3       pk        978:                        cp = storage;
                    979:                        storage += pp->pr_pagesz;
                    980:                } else {
1.21      thorpej   981:                        simple_unlock(&pp->pr_slock);
1.3       pk        982:                        cp = (*pp->pr_alloc)(pp->pr_pagesz, 0, pp->pr_mtype);
1.21      thorpej   983:                        simple_lock(&pp->pr_slock);
1.3       pk        984:                }
1.2       pk        985:
1.3       pk        986:                if (cp == NULL) {
1.21      thorpej   987:                        simple_unlock(&pp->pr_slock);
1.1       pk        988:                        return (ENOMEM);
                    989:                }
                    990:
1.26      thorpej   991:                pp->pr_npagealloc++;
1.3       pk        992:                pool_prime_page(pp, cp);
                    993:                pp->pr_minpages++;
1.1       pk        994:        }
1.3       pk        995:
                    996:        pp->pr_minitems = newnitems;
                    997:
                    998:        if (pp->pr_minpages >= pp->pr_maxpages)
                    999:                pp->pr_maxpages = pp->pr_minpages + 1;  /* XXX */
                   1000:
1.21      thorpej  1001:        simple_unlock(&pp->pr_slock);
1.1       pk       1002:        return (0);
                   1003: }
1.3       pk       1004:
                   1005: /*
                   1006:  * Add a page worth of items to the pool.
1.21      thorpej  1007:  *
                   1008:  * Note, we must be called with the pool descriptor LOCKED.
1.3       pk       1009:  */
1.21      thorpej  1010: static void
1.3       pk       1011: pool_prime_page(pp, storage)
                   1012:        struct pool *pp;
                   1013:        caddr_t storage;
                   1014: {
                   1015:        struct pool_item *pi;
                   1016:        struct pool_item_header *ph;
                   1017:        caddr_t cp = storage;
                   1018:        unsigned int align = pp->pr_align;
                   1019:        unsigned int ioff = pp->pr_itemoffset;
1.27      pk       1020:        int s, n;
1.36      pk       1021:
                   1022:        if (((u_long)cp & (pp->pr_pagesz - 1)) != 0)
                   1023:                panic("pool_prime_page: %s: unaligned page", pp->pr_wchan);
1.3       pk       1024:
1.20      thorpej  1025:        if ((pp->pr_roflags & PR_PHINPAGE) != 0) {
1.3       pk       1026:                ph = (struct pool_item_header *)(cp + pp->pr_phoffset);
                   1027:        } else {
1.27      pk       1028:                s = splhigh();
1.3       pk       1029:                ph = pool_get(&phpool, PR_URGENT);
1.27      pk       1030:                splx(s);
1.3       pk       1031:                LIST_INSERT_HEAD(&pp->pr_hashtab[PR_HASH_INDEX(pp, cp)],
                   1032:                                 ph, ph_hashlist);
                   1033:        }
                   1034:
                   1035:        /*
                   1036:         * Insert page header.
                   1037:         */
                   1038:        TAILQ_INSERT_HEAD(&pp->pr_pagelist, ph, ph_pagelist);
                   1039:        TAILQ_INIT(&ph->ph_itemlist);
                   1040:        ph->ph_page = storage;
                   1041:        ph->ph_nmissing = 0;
1.21      thorpej  1042:        memset(&ph->ph_time, 0, sizeof(ph->ph_time));
1.3       pk       1043:
1.6       thorpej  1044:        pp->pr_nidle++;
                   1045:
1.3       pk       1046:        /*
                   1047:         * Color this page.
                   1048:         */
                   1049:        cp = (caddr_t)(cp + pp->pr_curcolor);
                   1050:        if ((pp->pr_curcolor += align) > pp->pr_maxcolor)
                   1051:                pp->pr_curcolor = 0;
                   1052:
                   1053:        /*
                   1054:         * Adjust storage to apply aligment to `pr_itemoffset' in each item.
                   1055:         */
                   1056:        if (ioff != 0)
                   1057:                cp = (caddr_t)(cp + (align - ioff));
                   1058:
                   1059:        /*
                   1060:         * Insert remaining chunks on the bucket list.
                   1061:         */
                   1062:        n = pp->pr_itemsperpage;
1.20      thorpej  1063:        pp->pr_nitems += n;
1.3       pk       1064:
                   1065:        while (n--) {
                   1066:                pi = (struct pool_item *)cp;
                   1067:
                   1068:                /* Insert on page list */
                   1069:                TAILQ_INSERT_TAIL(&ph->ph_itemlist, pi, pi_list);
                   1070: #ifdef DIAGNOSTIC
                   1071:                pi->pi_magic = PI_MAGIC;
                   1072: #endif
                   1073:                cp = (caddr_t)(cp + pp->pr_size);
                   1074:        }
                   1075:
                   1076:        /*
                   1077:         * If the pool was depleted, point at the new page.
                   1078:         */
                   1079:        if (pp->pr_curpage == NULL)
                   1080:                pp->pr_curpage = ph;
                   1081:
                   1082:        if (++pp->pr_npages > pp->pr_hiwat)
                   1083:                pp->pr_hiwat = pp->pr_npages;
                   1084: }
                   1085:
1.20      thorpej  1086: /*
                   1087:  * Like pool_prime(), except this is used by pool_get() when nitems
                   1088:  * drops below the low water mark.  This is used to catch up nitmes
                   1089:  * with the low water mark.
                   1090:  *
1.21      thorpej  1091:  * Note 1, we never wait for memory here, we let the caller decide what to do.
1.20      thorpej  1092:  *
                   1093:  * Note 2, this doesn't work with static pools.
                   1094:  *
                   1095:  * Note 3, we must be called with the pool already locked, and we return
                   1096:  * with it locked.
                   1097:  */
                   1098: static int
                   1099: pool_catchup(pp)
                   1100:        struct pool *pp;
                   1101: {
                   1102:        caddr_t cp;
                   1103:        int error = 0;
                   1104:
                   1105:        if (pp->pr_roflags & PR_STATIC) {
                   1106:                /*
                   1107:                 * We dropped below the low water mark, and this is not a
                   1108:                 * good thing.  Log a warning.
1.21      thorpej  1109:                 *
                   1110:                 * XXX: rate-limit this?
1.20      thorpej  1111:                 */
                   1112:                printf("WARNING: static pool `%s' dropped below low water "
                   1113:                    "mark\n", pp->pr_wchan);
                   1114:                return (0);
                   1115:        }
                   1116:
1.21      thorpej  1117:        while (pp->pr_nitems < pp->pr_minitems) {
1.20      thorpej  1118:                /*
1.21      thorpej  1119:                 * Call the page back-end allocator for more memory.
                   1120:                 *
                   1121:                 * XXX: We never wait, so should we bother unlocking
                   1122:                 * the pool descriptor?
1.20      thorpej  1123:                 */
1.21      thorpej  1124:                simple_unlock(&pp->pr_slock);
1.20      thorpej  1125:                cp = (*pp->pr_alloc)(pp->pr_pagesz, 0, pp->pr_mtype);
1.21      thorpej  1126:                simple_lock(&pp->pr_slock);
1.34      thorpej  1127:                if (__predict_false(cp == NULL)) {
1.20      thorpej  1128:                        error = ENOMEM;
                   1129:                        break;
                   1130:                }
1.26      thorpej  1131:                pp->pr_npagealloc++;
1.20      thorpej  1132:                pool_prime_page(pp, cp);
                   1133:        }
                   1134:
                   1135:        return (error);
                   1136: }
                   1137:
1.3       pk       1138: void
                   1139: pool_setlowat(pp, n)
                   1140:        pool_handle_t   pp;
                   1141:        int n;
                   1142: {
1.20      thorpej  1143:        int error;
1.15      pk       1144:
1.21      thorpej  1145:        simple_lock(&pp->pr_slock);
                   1146:
1.3       pk       1147:        pp->pr_minitems = n;
1.15      pk       1148:        pp->pr_minpages = (n == 0)
                   1149:                ? 0
1.18      thorpej  1150:                : roundup(n, pp->pr_itemsperpage) / pp->pr_itemsperpage;
1.20      thorpej  1151:
                   1152:        /* Make sure we're caught up with the newly-set low water mark. */
1.41    ! sommerfe 1153:        if ((pp->pr_nitems < pp->pr_minitems) &&
        !          1154:            (error = pool_catchup(pp)) != 0) {
1.20      thorpej  1155:                /*
                   1156:                 * XXX: Should we log a warning?  Should we set up a timeout
                   1157:                 * to try again in a second or so?  The latter could break
                   1158:                 * a caller's assumptions about interrupt protection, etc.
                   1159:                 */
                   1160:        }
1.21      thorpej  1161:
                   1162:        simple_unlock(&pp->pr_slock);
1.3       pk       1163: }
                   1164:
                   1165: void
                   1166: pool_sethiwat(pp, n)
                   1167:        pool_handle_t   pp;
                   1168:        int n;
                   1169: {
1.15      pk       1170:
1.21      thorpej  1171:        simple_lock(&pp->pr_slock);
                   1172:
1.15      pk       1173:        pp->pr_maxpages = (n == 0)
                   1174:                ? 0
1.18      thorpej  1175:                : roundup(n, pp->pr_itemsperpage) / pp->pr_itemsperpage;
1.21      thorpej  1176:
                   1177:        simple_unlock(&pp->pr_slock);
1.3       pk       1178: }
                   1179:
1.20      thorpej  1180: void
                   1181: pool_sethardlimit(pp, n, warnmess, ratecap)
                   1182:        pool_handle_t pp;
                   1183:        int n;
                   1184:        const char *warnmess;
                   1185:        int ratecap;
                   1186: {
                   1187:
1.21      thorpej  1188:        simple_lock(&pp->pr_slock);
1.20      thorpej  1189:
                   1190:        pp->pr_hardlimit = n;
                   1191:        pp->pr_hardlimit_warning = warnmess;
1.31      thorpej  1192:        pp->pr_hardlimit_ratecap.tv_sec = ratecap;
                   1193:        pp->pr_hardlimit_warning_last.tv_sec = 0;
                   1194:        pp->pr_hardlimit_warning_last.tv_usec = 0;
1.20      thorpej  1195:
                   1196:        /*
1.21      thorpej  1197:         * In-line version of pool_sethiwat(), because we don't want to
                   1198:         * release the lock.
1.20      thorpej  1199:         */
                   1200:        pp->pr_maxpages = (n == 0)
                   1201:                ? 0
                   1202:                : roundup(n, pp->pr_itemsperpage) / pp->pr_itemsperpage;
1.21      thorpej  1203:
                   1204:        simple_unlock(&pp->pr_slock);
1.20      thorpej  1205: }
1.3       pk       1206:
                   1207: /*
                   1208:  * Default page allocator.
                   1209:  */
                   1210: static void *
                   1211: pool_page_alloc(sz, flags, mtype)
                   1212:        unsigned long sz;
                   1213:        int flags;
                   1214:        int mtype;
                   1215: {
1.11      thorpej  1216:        boolean_t waitok = (flags & PR_WAITOK) ? TRUE : FALSE;
1.3       pk       1217:
1.11      thorpej  1218:        return ((void *)uvm_km_alloc_poolpage(waitok));
1.3       pk       1219: }
                   1220:
                   1221: static void
                   1222: pool_page_free(v, sz, mtype)
                   1223:        void *v;
                   1224:        unsigned long sz;
                   1225:        int mtype;
                   1226: {
                   1227:
1.10      eeh      1228:        uvm_km_free_poolpage((vaddr_t)v);
1.3       pk       1229: }
1.12      thorpej  1230:
                   1231: /*
                   1232:  * Alternate pool page allocator for pools that know they will
                   1233:  * never be accessed in interrupt context.
                   1234:  */
                   1235: void *
                   1236: pool_page_alloc_nointr(sz, flags, mtype)
                   1237:        unsigned long sz;
                   1238:        int flags;
                   1239:        int mtype;
                   1240: {
                   1241:        boolean_t waitok = (flags & PR_WAITOK) ? TRUE : FALSE;
                   1242:
                   1243:        return ((void *)uvm_km_alloc_poolpage1(kernel_map, uvm.kernel_object,
                   1244:            waitok));
                   1245: }
                   1246:
                   1247: void
                   1248: pool_page_free_nointr(v, sz, mtype)
                   1249:        void *v;
                   1250:        unsigned long sz;
                   1251:        int mtype;
                   1252: {
                   1253:
                   1254:        uvm_km_free_poolpage1(kernel_map, (vaddr_t)v);
                   1255: }
                   1256:
1.3       pk       1257:
                   1258: /*
                   1259:  * Release all complete pages that have not been used recently.
                   1260:  */
                   1261: void
1.25      thorpej  1262: _pool_reclaim(pp, file, line)
1.3       pk       1263:        pool_handle_t pp;
1.25      thorpej  1264:        const char *file;
                   1265:        long line;
1.3       pk       1266: {
                   1267:        struct pool_item_header *ph, *phnext;
1.21      thorpej  1268:        struct timeval curtime;
                   1269:        int s;
1.3       pk       1270:
1.20      thorpej  1271:        if (pp->pr_roflags & PR_STATIC)
1.3       pk       1272:                return;
                   1273:
1.21      thorpej  1274:        if (simple_lock_try(&pp->pr_slock) == 0)
1.3       pk       1275:                return;
1.25      thorpej  1276:        pr_enter(pp, file, line);
1.3       pk       1277:
1.21      thorpej  1278:        s = splclock();
                   1279:        curtime = mono_time;
                   1280:        splx(s);
                   1281:
1.3       pk       1282:        for (ph = TAILQ_FIRST(&pp->pr_pagelist); ph != NULL; ph = phnext) {
                   1283:                phnext = TAILQ_NEXT(ph, ph_pagelist);
                   1284:
                   1285:                /* Check our minimum page claim */
                   1286:                if (pp->pr_npages <= pp->pr_minpages)
                   1287:                        break;
                   1288:
                   1289:                if (ph->ph_nmissing == 0) {
                   1290:                        struct timeval diff;
                   1291:                        timersub(&curtime, &ph->ph_time, &diff);
                   1292:                        if (diff.tv_sec < pool_inactive_time)
                   1293:                                continue;
1.21      thorpej  1294:
                   1295:                        /*
                   1296:                         * If freeing this page would put us below
                   1297:                         * the low water mark, stop now.
                   1298:                         */
                   1299:                        if ((pp->pr_nitems - pp->pr_itemsperpage) <
                   1300:                            pp->pr_minitems)
                   1301:                                break;
                   1302:
1.3       pk       1303:                        pr_rmpage(pp, ph);
                   1304:                }
                   1305:        }
                   1306:
1.25      thorpej  1307:        pr_leave(pp);
1.21      thorpej  1308:        simple_unlock(&pp->pr_slock);
1.3       pk       1309: }
                   1310:
                   1311:
                   1312: /*
                   1313:  * Drain pools, one at a time.
1.21      thorpej  1314:  *
                   1315:  * Note, we must never be called from an interrupt context.
1.3       pk       1316:  */
                   1317: void
                   1318: pool_drain(arg)
                   1319:        void *arg;
                   1320: {
                   1321:        struct pool *pp;
1.23      thorpej  1322:        int s;
1.3       pk       1323:
1.23      thorpej  1324:        s = splimp();
                   1325:        simple_lock(&pool_head_slock);
                   1326:
                   1327:        if (drainpp == NULL && (drainpp = TAILQ_FIRST(&pool_head)) == NULL)
                   1328:                goto out;
1.3       pk       1329:
                   1330:        pp = drainpp;
                   1331:        drainpp = TAILQ_NEXT(pp, pr_poollist);
                   1332:
                   1333:        pool_reclaim(pp);
1.23      thorpej  1334:
                   1335:  out:
                   1336:        simple_unlock(&pool_head_slock);
1.3       pk       1337:        splx(s);
                   1338: }
                   1339:
                   1340:
                   1341: /*
                   1342:  * Diagnostic helpers.
                   1343:  */
                   1344: void
1.25      thorpej  1345: pool_print(pp, modif)
1.3       pk       1346:        struct pool *pp;
1.25      thorpej  1347:        const char *modif;
1.21      thorpej  1348: {
                   1349:        int s;
                   1350:
                   1351:        s = splimp();
1.25      thorpej  1352:        if (simple_lock_try(&pp->pr_slock) == 0) {
                   1353:                printf("pool %s is locked; try again later\n",
                   1354:                    pp->pr_wchan);
                   1355:                splx(s);
                   1356:                return;
                   1357:        }
                   1358:        pool_print1(pp, modif, printf);
1.21      thorpej  1359:        simple_unlock(&pp->pr_slock);
                   1360:        splx(s);
                   1361: }
                   1362:
1.25      thorpej  1363: void
                   1364: pool_printit(pp, modif, pr)
                   1365:        struct pool *pp;
                   1366:        const char *modif;
                   1367:        void (*pr) __P((const char *, ...));
                   1368: {
                   1369:        int didlock = 0;
                   1370:
                   1371:        if (pp == NULL) {
                   1372:                (*pr)("Must specify a pool to print.\n");
                   1373:                return;
                   1374:        }
                   1375:
                   1376:        /*
                   1377:         * Called from DDB; interrupts should be blocked, and all
                   1378:         * other processors should be paused.  We can skip locking
                   1379:         * the pool in this case.
                   1380:         *
                   1381:         * We do a simple_lock_try() just to print the lock
                   1382:         * status, however.
                   1383:         */
                   1384:
                   1385:        if (simple_lock_try(&pp->pr_slock) == 0)
                   1386:                (*pr)("WARNING: pool %s is locked\n", pp->pr_wchan);
                   1387:        else
                   1388:                didlock = 1;
                   1389:
                   1390:        pool_print1(pp, modif, pr);
                   1391:
                   1392:        if (didlock)
                   1393:                simple_unlock(&pp->pr_slock);
                   1394: }
                   1395:
1.21      thorpej  1396: static void
1.25      thorpej  1397: pool_print1(pp, modif, pr)
1.21      thorpej  1398:        struct pool *pp;
1.25      thorpej  1399:        const char *modif;
                   1400:        void (*pr) __P((const char *, ...));
1.3       pk       1401: {
1.25      thorpej  1402:        struct pool_item_header *ph;
                   1403: #ifdef DIAGNOSTIC
                   1404:        struct pool_item *pi;
                   1405: #endif
                   1406:        int print_log = 0, print_pagelist = 0;
                   1407:        char c;
                   1408:
                   1409:        while ((c = *modif++) != '\0') {
                   1410:                if (c == 'l')
                   1411:                        print_log = 1;
                   1412:                if (c == 'p')
                   1413:                        print_pagelist = 1;
                   1414:                modif++;
                   1415:        }
                   1416:
                   1417:        (*pr)("POOL %s: size %u, align %u, ioff %u, roflags 0x%08x\n",
                   1418:            pp->pr_wchan, pp->pr_size, pp->pr_align, pp->pr_itemoffset,
                   1419:            pp->pr_roflags);
                   1420:        (*pr)("\tpagesz %u, mtype %d\n", pp->pr_pagesz, pp->pr_mtype);
                   1421:        (*pr)("\talloc %p, release %p\n", pp->pr_alloc, pp->pr_free);
                   1422:        (*pr)("\tminitems %u, minpages %u, maxpages %u, npages %u\n",
                   1423:            pp->pr_minitems, pp->pr_minpages, pp->pr_maxpages, pp->pr_npages);
                   1424:        (*pr)("\titemsperpage %u, nitems %u, nout %u, hardlimit %u\n",
                   1425:            pp->pr_itemsperpage, pp->pr_nitems, pp->pr_nout, pp->pr_hardlimit);
                   1426:
                   1427:        (*pr)("\n\tnget %lu, nfail %lu, nput %lu\n",
                   1428:            pp->pr_nget, pp->pr_nfail, pp->pr_nput);
                   1429:        (*pr)("\tnpagealloc %lu, npagefree %lu, hiwat %u, nidle %lu\n",
                   1430:            pp->pr_npagealloc, pp->pr_npagefree, pp->pr_hiwat, pp->pr_nidle);
                   1431:
                   1432:        if (print_pagelist == 0)
                   1433:                goto skip_pagelist;
                   1434:
                   1435:        if ((ph = TAILQ_FIRST(&pp->pr_pagelist)) != NULL)
                   1436:                (*pr)("\n\tpage list:\n");
                   1437:        for (; ph != NULL; ph = TAILQ_NEXT(ph, ph_pagelist)) {
                   1438:                (*pr)("\t\tpage %p, nmissing %d, time %lu,%lu\n",
                   1439:                    ph->ph_page, ph->ph_nmissing,
                   1440:                    (u_long)ph->ph_time.tv_sec,
                   1441:                    (u_long)ph->ph_time.tv_usec);
                   1442: #ifdef DIAGNOSTIC
                   1443:                for (pi = TAILQ_FIRST(&ph->ph_itemlist); pi != NULL;
                   1444:                     pi = TAILQ_NEXT(pi, pi_list)) {
                   1445:                        if (pi->pi_magic != PI_MAGIC) {
                   1446:                                (*pr)("\t\t\titem %p, magic 0x%x\n",
                   1447:                                    pi, pi->pi_magic);
                   1448:                        }
                   1449:                }
                   1450: #endif
                   1451:        }
                   1452:        if (pp->pr_curpage == NULL)
                   1453:                (*pr)("\tno current page\n");
                   1454:        else
                   1455:                (*pr)("\tcurpage %p\n", pp->pr_curpage->ph_page);
                   1456:
                   1457:  skip_pagelist:
                   1458:
                   1459:        if (print_log == 0)
                   1460:                goto skip_log;
                   1461:
                   1462:        (*pr)("\n");
                   1463:        if ((pp->pr_roflags & PR_LOGGING) == 0)
                   1464:                (*pr)("\tno log\n");
                   1465:        else
                   1466:                pr_printlog(pp, NULL, pr);
1.3       pk       1467:
1.25      thorpej  1468:  skip_log:
1.3       pk       1469:
1.25      thorpej  1470:        pr_enter_check(pp, pr);
1.3       pk       1471: }
                   1472:
                   1473: int
                   1474: pool_chk(pp, label)
                   1475:        struct pool *pp;
                   1476:        char *label;
                   1477: {
                   1478:        struct pool_item_header *ph;
                   1479:        int r = 0;
                   1480:
1.21      thorpej  1481:        simple_lock(&pp->pr_slock);
1.3       pk       1482:
                   1483:        for (ph = TAILQ_FIRST(&pp->pr_pagelist); ph != NULL;
                   1484:             ph = TAILQ_NEXT(ph, ph_pagelist)) {
                   1485:
                   1486:                struct pool_item *pi;
                   1487:                int n;
                   1488:                caddr_t page;
                   1489:
                   1490:                page = (caddr_t)((u_long)ph & pp->pr_pagemask);
1.20      thorpej  1491:                if (page != ph->ph_page &&
                   1492:                    (pp->pr_roflags & PR_PHINPAGE) != 0) {
1.3       pk       1493:                        if (label != NULL)
                   1494:                                printf("%s: ", label);
1.16      briggs   1495:                        printf("pool(%p:%s): page inconsistency: page %p;"
                   1496:                               " at page head addr %p (p %p)\n", pp,
1.3       pk       1497:                                pp->pr_wchan, ph->ph_page,
                   1498:                                ph, page);
                   1499:                        r++;
                   1500:                        goto out;
                   1501:                }
                   1502:
                   1503:                for (pi = TAILQ_FIRST(&ph->ph_itemlist), n = 0;
                   1504:                     pi != NULL;
                   1505:                     pi = TAILQ_NEXT(pi,pi_list), n++) {
                   1506:
                   1507: #ifdef DIAGNOSTIC
                   1508:                        if (pi->pi_magic != PI_MAGIC) {
                   1509:                                if (label != NULL)
                   1510:                                        printf("%s: ", label);
                   1511:                                printf("pool(%s): free list modified: magic=%x;"
                   1512:                                       " page %p; item ordinal %d;"
                   1513:                                       " addr %p (p %p)\n",
                   1514:                                        pp->pr_wchan, pi->pi_magic, ph->ph_page,
                   1515:                                        n, pi, page);
                   1516:                                panic("pool");
                   1517:                        }
                   1518: #endif
                   1519:                        page = (caddr_t)((u_long)pi & pp->pr_pagemask);
                   1520:                        if (page == ph->ph_page)
                   1521:                                continue;
                   1522:
                   1523:                        if (label != NULL)
                   1524:                                printf("%s: ", label);
1.16      briggs   1525:                        printf("pool(%p:%s): page inconsistency: page %p;"
                   1526:                               " item ordinal %d; addr %p (p %p)\n", pp,
1.3       pk       1527:                                pp->pr_wchan, ph->ph_page,
                   1528:                                n, pi, page);
                   1529:                        r++;
                   1530:                        goto out;
                   1531:                }
                   1532:        }
                   1533: out:
1.21      thorpej  1534:        simple_unlock(&pp->pr_slock);
1.3       pk       1535:        return (r);
                   1536: }

CVSweb <webmaster@jp.NetBSD.org>