[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 1.19

1.19    ! mrg         1: /*     $NetBSD: subr_pool.c,v 1.18 1999/03/23 02:49:03 thorpej Exp $   */
1.1       pk          2:
                      3: /*-
                      4:  * Copyright (c) 1997 The NetBSD Foundation, Inc.
                      5:  * All rights reserved.
                      6:  *
                      7:  * This code is derived from software contributed to The NetBSD Foundation
                      8:  * by Paul Kranenburg.
                      9:  *
                     10:  * Redistribution and use in source and binary forms, with or without
                     11:  * modification, are permitted provided that the following conditions
                     12:  * are met:
                     13:  * 1. Redistributions of source code must retain the above copyright
                     14:  *    notice, this list of conditions and the following disclaimer.
                     15:  * 2. Redistributions in binary form must reproduce the above copyright
                     16:  *    notice, this list of conditions and the following disclaimer in the
                     17:  *    documentation and/or other materials provided with the distribution.
                     18:  * 3. All advertising materials mentioning features or use of this software
                     19:  *    must display the following acknowledgement:
1.13      christos   20:  *     This product includes software developed by the NetBSD
                     21:  *     Foundation, Inc. and its contributors.
1.1       pk         22:  * 4. Neither the name of The NetBSD Foundation nor the names of its
                     23:  *    contributors may be used to endorse or promote products derived
                     24:  *    from this software without specific prior written permission.
                     25:  *
                     26:  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
                     27:  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
                     28:  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
                     29:  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
                     30:  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
                     31:  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
                     32:  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
                     33:  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
                     34:  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
                     35:  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
                     36:  * POSSIBILITY OF SUCH DAMAGE.
                     37:  */
                     38:
                     39: #include <sys/param.h>
                     40: #include <sys/systm.h>
                     41: #include <sys/proc.h>
                     42: #include <sys/errno.h>
                     43: #include <sys/kernel.h>
                     44: #include <sys/malloc.h>
                     45: #include <sys/lock.h>
                     46: #include <sys/pool.h>
                     47:
1.3       pk         48: #include <vm/vm.h>
                     49: #include <vm/vm_kern.h>
                     50:
                     51: #include <uvm/uvm.h>
                     52:
1.1       pk         53: /*
                     54:  * Pool resource management utility.
1.3       pk         55:  *
                     56:  * Memory is allocated in pages which are split into pieces according
                     57:  * to the pool item size. Each page is kept on a list headed by `pr_pagelist'
                     58:  * in the pool structure and the individual pool items are on a linked list
                     59:  * headed by `ph_itemlist' in each page header. The memory for building
                     60:  * the page list is either taken from the allocated pages themselves (for
                     61:  * small pool items) or taken from an internal pool of page headers (`phpool').
                     62:  *
1.1       pk         63:  */
                     64:
1.3       pk         65: /* List of all pools */
1.5       thorpej    66: TAILQ_HEAD(,pool) pool_head = TAILQ_HEAD_INITIALIZER(pool_head);
1.3       pk         67:
                     68: /* Private pool for page header structures */
                     69: static struct pool phpool;
                     70:
                     71: /* # of seconds to retain page after last use */
                     72: int pool_inactive_time = 10;
                     73:
                     74: /* Next candidate for drainage (see pool_drain()) */
                     75: static struct pool     *drainpp = NULL;
                     76:
                     77: struct pool_item_header {
                     78:        /* Page headers */
                     79:        TAILQ_ENTRY(pool_item_header)
                     80:                                ph_pagelist;    /* pool page list */
                     81:        TAILQ_HEAD(,pool_item)  ph_itemlist;    /* chunk list for this page */
                     82:        LIST_ENTRY(pool_item_header)
                     83:                                ph_hashlist;    /* Off-page page headers */
                     84:        int                     ph_nmissing;    /* # of chunks in use */
                     85:        caddr_t                 ph_page;        /* this page's address */
                     86:        struct timeval          ph_time;        /* last referenced */
                     87: };
                     88:
1.1       pk         89: struct pool_item {
1.3       pk         90: #ifdef DIAGNOSTIC
                     91:        int pi_magic;
                     92: #define PI_MAGIC 0xdeadbeef
                     93: #endif
                     94:        /* Other entries use only this list entry */
                     95:        TAILQ_ENTRY(pool_item)  pi_list;
                     96: };
                     97:
                     98:
                     99: #define PR_HASH_INDEX(pp,addr) \
                    100:        (((u_long)(addr) >> (pp)->pr_pageshift) & (PR_HASHTABSIZE - 1))
                    101:
                    102:
                    103:
                    104: static struct pool_item_header
                    105:                *pr_find_pagehead __P((struct pool *, caddr_t));
                    106: static void    pr_rmpage __P((struct pool *, struct pool_item_header *));
                    107: static int     pool_prime_page __P((struct pool *, caddr_t));
                    108: static void    *pool_page_alloc __P((unsigned long, int, int));
                    109: static void    pool_page_free __P((void *, unsigned long, int));
                    110:
                    111:
                    112: #ifdef POOL_DIAGNOSTIC
                    113: /*
                    114:  * Pool log entry. An array of these is allocated in pool_create().
                    115:  */
                    116: struct pool_log {
                    117:        const char      *pl_file;
                    118:        long            pl_line;
                    119:        int             pl_action;
                    120: #define PRLOG_GET      1
                    121: #define PRLOG_PUT      2
                    122:        void            *pl_addr;
1.1       pk        123: };
                    124:
1.3       pk        125: /* Number of entries in pool log buffers */
1.17      thorpej   126: #ifndef POOL_LOGSIZE
                    127: #define        POOL_LOGSIZE    10
                    128: #endif
                    129:
                    130: int pool_logsize = POOL_LOGSIZE;
1.1       pk        131:
1.3       pk        132: static void    pr_log __P((struct pool *, void *, int, const char *, long));
                    133: static void    pr_printlog __P((struct pool *));
                    134:
                    135: static __inline__ void
                    136: pr_log(pp, v, action, file, line)
                    137:        struct pool     *pp;
                    138:        void            *v;
                    139:        int             action;
                    140:        const char      *file;
                    141:        long            line;
                    142: {
                    143:        int n = pp->pr_curlogentry;
                    144:        struct pool_log *pl;
                    145:
                    146:        if ((pp->pr_flags & PR_LOGGING) == 0)
                    147:                return;
                    148:
                    149:        /*
                    150:         * Fill in the current entry. Wrap around and overwrite
                    151:         * the oldest entry if necessary.
                    152:         */
                    153:        pl = &pp->pr_log[n];
                    154:        pl->pl_file = file;
                    155:        pl->pl_line = line;
                    156:        pl->pl_action = action;
                    157:        pl->pl_addr = v;
                    158:        if (++n >= pp->pr_logsize)
                    159:                n = 0;
                    160:        pp->pr_curlogentry = n;
                    161: }
                    162:
                    163: static void
                    164: pr_printlog(pp)
                    165:        struct pool *pp;
                    166: {
                    167:        int i = pp->pr_logsize;
                    168:        int n = pp->pr_curlogentry;
                    169:
                    170:        if ((pp->pr_flags & PR_LOGGING) == 0)
                    171:                return;
                    172:
                    173:        pool_print(pp, "printlog");
                    174:
                    175:        /*
                    176:         * Print all entries in this pool's log.
                    177:         */
                    178:        while (i-- > 0) {
                    179:                struct pool_log *pl = &pp->pr_log[n];
                    180:                if (pl->pl_action != 0) {
                    181:                        printf("log entry %d:\n", i);
                    182:                        printf("\taction = %s, addr = %p\n",
                    183:                                pl->pl_action == PRLOG_GET ? "get" : "put",
                    184:                                pl->pl_addr);
                    185:                        printf("\tfile: %s at line %lu\n",
                    186:                                pl->pl_file, pl->pl_line);
                    187:                }
                    188:                if (++n >= pp->pr_logsize)
                    189:                        n = 0;
                    190:        }
                    191: }
                    192: #else
                    193: #define pr_log(pp, v, action, file, line)
                    194: #define pr_printlog(pp)
                    195: #endif
                    196:
                    197:
                    198: /*
                    199:  * Return the pool page header based on page address.
                    200:  */
                    201: static __inline__ struct pool_item_header *
                    202: pr_find_pagehead(pp, page)
                    203:        struct pool *pp;
                    204:        caddr_t page;
                    205: {
                    206:        struct pool_item_header *ph;
                    207:
                    208:        if ((pp->pr_flags & PR_PHINPAGE) != 0)
                    209:                return ((struct pool_item_header *)(page + pp->pr_phoffset));
                    210:
                    211:        for (ph = LIST_FIRST(&pp->pr_hashtab[PR_HASH_INDEX(pp, page)]);
                    212:             ph != NULL;
                    213:             ph = LIST_NEXT(ph, ph_hashlist)) {
                    214:                if (ph->ph_page == page)
                    215:                        return (ph);
                    216:        }
                    217:        return (NULL);
                    218: }
                    219:
                    220: /*
                    221:  * Remove a page from the pool.
                    222:  */
                    223: static __inline__ void
                    224: pr_rmpage(pp, ph)
                    225:        struct pool *pp;
                    226:        struct pool_item_header *ph;
                    227: {
                    228:
                    229:        /*
1.7       thorpej   230:         * If the page was idle, decrement the idle page count.
1.3       pk        231:         */
1.6       thorpej   232:        if (ph->ph_nmissing == 0) {
                    233: #ifdef DIAGNOSTIC
                    234:                if (pp->pr_nidle == 0)
                    235:                        panic("pr_rmpage: nidle inconsistent");
                    236: #endif
                    237:                pp->pr_nidle--;
                    238:        }
1.7       thorpej   239:
                    240:        /*
                    241:         * Unlink a page from the pool and release it.
                    242:         */
                    243:        TAILQ_REMOVE(&pp->pr_pagelist, ph, ph_pagelist);
                    244:        (*pp->pr_free)(ph->ph_page, pp->pr_pagesz, pp->pr_mtype);
                    245:        pp->pr_npages--;
                    246:        pp->pr_npagefree++;
1.6       thorpej   247:
1.3       pk        248:        if ((pp->pr_flags & PR_PHINPAGE) == 0) {
                    249:                LIST_REMOVE(ph, ph_hashlist);
                    250:                pool_put(&phpool, ph);
                    251:        }
                    252:
                    253:        if (pp->pr_curpage == ph) {
                    254:                /*
                    255:                 * Find a new non-empty page header, if any.
                    256:                 * Start search from the page head, to increase the
                    257:                 * chance for "high water" pages to be freed.
                    258:                 */
                    259:                for (ph = TAILQ_FIRST(&pp->pr_pagelist); ph != NULL;
                    260:                     ph = TAILQ_NEXT(ph, ph_pagelist))
                    261:                        if (TAILQ_FIRST(&ph->ph_itemlist) != NULL)
                    262:                                break;
                    263:
                    264:                pp->pr_curpage = ph;
                    265:        }
                    266: }
                    267:
                    268: /*
                    269:  * Allocate and initialize a pool.
                    270:  */
1.1       pk        271: struct pool *
1.3       pk        272: pool_create(size, align, ioff, nitems, wchan, pagesz, alloc, release, mtype)
1.1       pk        273:        size_t  size;
1.3       pk        274:        u_int   align;
                    275:        u_int   ioff;
1.1       pk        276:        int     nitems;
                    277:        char    *wchan;
1.3       pk        278:        size_t  pagesz;
                    279:        void    *(*alloc) __P((unsigned long, int, int));
                    280:        void    (*release) __P((void *, unsigned long, int));
1.1       pk        281:        int     mtype;
                    282: {
                    283:        struct pool *pp;
1.3       pk        284:        int flags;
1.1       pk        285:
1.3       pk        286:        pp = (struct pool *)malloc(sizeof(*pp), M_POOL, M_NOWAIT);
                    287:        if (pp == NULL)
1.1       pk        288:                return (NULL);
1.3       pk        289:
                    290:        flags = PR_FREEHEADER;
                    291: #ifdef POOL_DIAGNOSTIC
                    292:        if (pool_logsize != 0)
                    293:                flags |= PR_LOGGING;
                    294: #endif
                    295:
                    296:        pool_init(pp, size, align, ioff, flags, wchan, pagesz,
                    297:                  alloc, release, mtype);
                    298:
                    299:        if (nitems != 0) {
                    300:                if (pool_prime(pp, nitems, NULL) != 0) {
                    301:                        pool_destroy(pp);
                    302:                        return (NULL);
                    303:                }
1.1       pk        304:        }
                    305:
1.3       pk        306:        return (pp);
                    307: }
                    308:
                    309: /*
                    310:  * Initialize the given pool resource structure.
                    311:  *
                    312:  * We export this routine to allow other kernel parts to declare
                    313:  * static pools that must be initialized before malloc() is available.
                    314:  */
                    315: void
                    316: pool_init(pp, size, align, ioff, flags, wchan, pagesz, alloc, release, mtype)
                    317:        struct pool     *pp;
                    318:        size_t          size;
                    319:        u_int           align;
                    320:        u_int           ioff;
                    321:        int             flags;
                    322:        char            *wchan;
                    323:        size_t          pagesz;
                    324:        void            *(*alloc) __P((unsigned long, int, int));
                    325:        void            (*release) __P((void *, unsigned long, int));
                    326:        int             mtype;
                    327: {
1.16      briggs    328:        int off, slack, i;
1.3       pk        329:
                    330:        /*
                    331:         * Check arguments and construct default values.
                    332:         */
                    333:        if (!powerof2(pagesz) || pagesz > PAGE_SIZE)
                    334:                panic("pool_init: page size invalid (%lx)\n", (u_long)pagesz);
                    335:
1.4       thorpej   336:        if (alloc == NULL && release == NULL) {
1.3       pk        337:                alloc = pool_page_alloc;
                    338:                release = pool_page_free;
1.4       thorpej   339:                pagesz = PAGE_SIZE;     /* Rounds to PAGE_SIZE anyhow. */
                    340:        } else if ((alloc != NULL && release != NULL) == 0) {
                    341:                /* If you specifiy one, must specify both. */
                    342:                panic("pool_init: must specify alloc and release together");
                    343:        }
                    344:
1.3       pk        345:        if (pagesz == 0)
                    346:                pagesz = PAGE_SIZE;
                    347:
                    348:        if (align == 0)
                    349:                align = ALIGN(1);
1.14      thorpej   350:
                    351:        if (size < sizeof(struct pool_item))
                    352:                size = sizeof(struct pool_item);
1.3       pk        353:
                    354:        /*
                    355:         * Initialize the pool structure.
                    356:         */
                    357:        TAILQ_INSERT_TAIL(&pool_head, pp, pr_poollist);
                    358:        TAILQ_INIT(&pp->pr_pagelist);
                    359:        pp->pr_curpage = NULL;
                    360:        pp->pr_npages = 0;
                    361:        pp->pr_minitems = 0;
                    362:        pp->pr_minpages = 0;
                    363:        pp->pr_maxpages = UINT_MAX;
                    364:        pp->pr_flags = flags;
                    365:        pp->pr_size = ALIGN(size);
                    366:        pp->pr_align = align;
                    367:        pp->pr_wchan = wchan;
                    368:        pp->pr_mtype = mtype;
                    369:        pp->pr_alloc = alloc;
                    370:        pp->pr_free = release;
                    371:        pp->pr_pagesz = pagesz;
                    372:        pp->pr_pagemask = ~(pagesz - 1);
                    373:        pp->pr_pageshift = ffs(pagesz) - 1;
                    374:
                    375:        /*
                    376:         * Decide whether to put the page header off page to avoid
                    377:         * wasting too large a part of the page. Off-page page headers
                    378:         * go on a hash table, so we can match a returned item
                    379:         * with its header based on the page address.
                    380:         * We use 1/16 of the page size as the threshold (XXX: tune)
                    381:         */
                    382:        if (pp->pr_size < pagesz/16) {
                    383:                /* Use the end of the page for the page header */
                    384:                pp->pr_flags |= PR_PHINPAGE;
                    385:                pp->pr_phoffset = off =
                    386:                        pagesz - ALIGN(sizeof(struct pool_item_header));
1.2       pk        387:        } else {
1.3       pk        388:                /* The page header will be taken from our page header pool */
                    389:                pp->pr_phoffset = 0;
                    390:                off = pagesz;
1.16      briggs    391:                for (i = 0; i < PR_HASHTABSIZE; i++) {
                    392:                        LIST_INIT(&pp->pr_hashtab[i]);
                    393:                }
1.2       pk        394:        }
1.1       pk        395:
1.3       pk        396:        /*
                    397:         * Alignment is to take place at `ioff' within the item. This means
                    398:         * we must reserve up to `align - 1' bytes on the page to allow
                    399:         * appropriate positioning of each item.
                    400:         *
                    401:         * Silently enforce `0 <= ioff < align'.
                    402:         */
                    403:        pp->pr_itemoffset = ioff = ioff % align;
                    404:        pp->pr_itemsperpage = (off - ((align - ioff) % align)) / pp->pr_size;
                    405:
                    406:        /*
                    407:         * Use the slack between the chunks and the page header
                    408:         * for "cache coloring".
                    409:         */
                    410:        slack = off - pp->pr_itemsperpage * pp->pr_size;
                    411:        pp->pr_maxcolor = (slack / align) * align;
                    412:        pp->pr_curcolor = 0;
                    413:
                    414:        pp->pr_nget = 0;
                    415:        pp->pr_nfail = 0;
                    416:        pp->pr_nput = 0;
                    417:        pp->pr_npagealloc = 0;
                    418:        pp->pr_npagefree = 0;
1.1       pk        419:        pp->pr_hiwat = 0;
1.8       thorpej   420:        pp->pr_nidle = 0;
1.3       pk        421:
                    422: #ifdef POOL_DIAGNOSTIC
                    423:        if ((flags & PR_LOGGING) != 0) {
                    424:                pp->pr_log = malloc(pool_logsize * sizeof(struct pool_log),
                    425:                                    M_TEMP, M_NOWAIT);
                    426:                if (pp->pr_log == NULL)
                    427:                        pp->pr_flags &= ~PR_LOGGING;
                    428:                pp->pr_curlogentry = 0;
                    429:                pp->pr_logsize = pool_logsize;
                    430:        }
                    431: #endif
                    432:
1.1       pk        433:        simple_lock_init(&pp->pr_lock);
1.15      pk        434:        lockinit(&pp->pr_resourcelock, PSWP, wchan, 0, 0);
1.1       pk        435:
1.3       pk        436:        /*
                    437:         * Initialize private page header pool if we haven't done so yet.
                    438:         */
                    439:        if (phpool.pr_size == 0) {
                    440:                pool_init(&phpool, sizeof(struct pool_item_header), 0, 0,
                    441:                          0, "phpool", 0, 0, 0, 0);
1.1       pk        442:        }
                    443:
1.3       pk        444:        return;
1.1       pk        445: }
                    446:
                    447: /*
                    448:  * De-commision a pool resource.
                    449:  */
                    450: void
                    451: pool_destroy(pp)
                    452:        struct pool *pp;
                    453: {
1.3       pk        454:        struct pool_item_header *ph;
                    455:
                    456: #ifdef DIAGNOSTIC
                    457:        if (pp->pr_nget - pp->pr_nput != 0) {
                    458:                pr_printlog(pp);
                    459:                panic("pool_destroy: pool busy: still out: %lu\n",
                    460:                      pp->pr_nget - pp->pr_nput);
                    461:        }
                    462: #endif
1.1       pk        463:
1.3       pk        464:        /* Remove all pages */
                    465:        if ((pp->pr_flags & PR_STATIC) == 0)
                    466:                while ((ph = pp->pr_pagelist.tqh_first) != NULL)
                    467:                        pr_rmpage(pp, ph);
                    468:
                    469:        /* Remove from global pool list */
                    470:        TAILQ_REMOVE(&pool_head, pp, pr_poollist);
                    471:        drainpp = NULL;
                    472:
                    473: #ifdef POOL_DIAGNOSTIC
                    474:        if ((pp->pr_flags & PR_LOGGING) != 0)
                    475:                free(pp->pr_log, M_TEMP);
                    476: #endif
1.2       pk        477:
1.3       pk        478:        if (pp->pr_flags & PR_FREEHEADER)
                    479:                free(pp, M_POOL);
1.1       pk        480: }
                    481:
                    482:
                    483: /*
1.3       pk        484:  * Grab an item from the pool; must be called at appropriate spl level
1.1       pk        485:  */
1.3       pk        486: #ifdef POOL_DIAGNOSTIC
                    487: void *
                    488: _pool_get(pp, flags, file, line)
                    489:        struct pool *pp;
                    490:        int flags;
                    491:        const char *file;
                    492:        long line;
                    493: #else
1.1       pk        494: void *
                    495: pool_get(pp, flags)
                    496:        struct pool *pp;
                    497:        int flags;
1.3       pk        498: #endif
1.1       pk        499: {
                    500:        void *v;
                    501:        struct pool_item *pi;
1.3       pk        502:        struct pool_item_header *ph;
1.1       pk        503:
1.2       pk        504: #ifdef DIAGNOSTIC
1.3       pk        505:        if ((pp->pr_flags & PR_STATIC) && (flags & PR_MALLOCOK)) {
                    506:                pr_printlog(pp);
1.2       pk        507:                panic("pool_get: static");
1.3       pk        508:        }
1.2       pk        509: #endif
                    510:
1.1       pk        511:        simple_lock(&pp->pr_lock);
1.3       pk        512:        if (curproc == NULL && (flags & PR_WAITOK) != 0)
                    513:                panic("pool_get: must have NOWAIT");
1.1       pk        514:
1.3       pk        515:        /*
                    516:         * The convention we use is that if `curpage' is not NULL, then
                    517:         * it points at a non-empty bucket. In particular, `curpage'
                    518:         * never points at a page header which has PR_PHINPAGE set and
                    519:         * has no items in its bucket.
                    520:         */
1.15      pk        521:        while ((ph = pp->pr_curpage) == NULL) {
                    522:                void *v;
                    523:                int lkflags = LK_EXCLUSIVE | LK_INTERLOCK |
                    524:                              ((flags & PR_WAITOK) == 0 ? LK_NOWAIT : 0);
                    525:
                    526:                /* Get long-term lock on pool */
                    527:                if (lockmgr(&pp->pr_resourcelock, lkflags, &pp->pr_lock) != 0)
                    528:                        return (NULL);
                    529:
                    530:                /* Check if pool became non-empty while we slept */
                    531:                if ((ph = pp->pr_curpage) != NULL)
                    532:                        goto again;
                    533:
                    534:                /* Call the page back-end allocator for more memory */
                    535:                v = (*pp->pr_alloc)(pp->pr_pagesz, flags, pp->pr_mtype);
1.1       pk        536:                if (v == NULL) {
1.3       pk        537:                        if (flags & PR_URGENT)
                    538:                                panic("pool_get: urgent");
                    539:                        if ((flags & PR_WAITOK) == 0) {
                    540:                                pp->pr_nfail++;
1.15      pk        541:                                lockmgr(&pp->pr_resourcelock, LK_RELEASE, NULL);
1.1       pk        542:                                return (NULL);
1.3       pk        543:                        }
                    544:
1.15      pk        545:                        /*
                    546:                         * Wait for items to be returned to this pool.
                    547:                         * XXX: we actually want to wait just until
                    548:                         * the page allocator has memory again. Depending
                    549:                         * on this pool's usage, we might get stuck here
                    550:                         * for a long time.
                    551:                         */
1.1       pk        552:                        pp->pr_flags |= PR_WANTED;
1.15      pk        553:                        lockmgr(&pp->pr_resourcelock, LK_RELEASE, NULL);
1.1       pk        554:                        tsleep((caddr_t)pp, PSWP, pp->pr_wchan, 0);
1.3       pk        555:                        simple_lock(&pp->pr_lock);
1.15      pk        556:                        continue;
1.1       pk        557:                }
1.3       pk        558:
1.15      pk        559:                /* We have more memory; add it to the pool */
                    560:                pp->pr_npagealloc++;
                    561:                pool_prime_page(pp, v);
                    562:
                    563: again:
                    564:                /* Re-acquire pool interlock */
                    565:                simple_lock(&pp->pr_lock);
                    566:                lockmgr(&pp->pr_resourcelock, LK_RELEASE, NULL);
1.3       pk        567:        }
                    568:
                    569:        if ((v = pi = TAILQ_FIRST(&ph->ph_itemlist)) == NULL)
                    570:                panic("pool_get: %s: page empty", pp->pr_wchan);
                    571:
                    572:        pr_log(pp, v, PRLOG_GET, file, line);
                    573:
                    574: #ifdef DIAGNOSTIC
                    575:        if (pi->pi_magic != PI_MAGIC) {
                    576:                pr_printlog(pp);
                    577:                panic("pool_get(%s): free list modified: magic=%x; page %p;"
                    578:                       " item addr %p\n",
                    579:                        pp->pr_wchan, pi->pi_magic, ph->ph_page, pi);
                    580:        }
                    581: #endif
                    582:
                    583:        /*
                    584:         * Remove from item list.
                    585:         */
                    586:        TAILQ_REMOVE(&ph->ph_itemlist, pi, pi_list);
1.6       thorpej   587:        if (ph->ph_nmissing == 0) {
                    588: #ifdef DIAGNOSTIC
                    589:                if (pp->pr_nidle == 0)
                    590:                        panic("pool_get: nidle inconsistent");
                    591: #endif
                    592:                pp->pr_nidle--;
                    593:        }
1.3       pk        594:        ph->ph_nmissing++;
                    595:        if (TAILQ_FIRST(&ph->ph_itemlist) == NULL) {
                    596:                /*
                    597:                 * Find a new non-empty page header, if any.
                    598:                 * Start search from the page head, to increase
                    599:                 * the chance for "high water" pages to be freed.
                    600:                 *
                    601:                 * First, move the now empty page to the head of
                    602:                 * the page list.
                    603:                 */
                    604:                TAILQ_REMOVE(&pp->pr_pagelist, ph, ph_pagelist);
                    605:                TAILQ_INSERT_HEAD(&pp->pr_pagelist, ph, ph_pagelist);
                    606:                while ((ph = TAILQ_NEXT(ph, ph_pagelist)) != NULL)
                    607:                        if (TAILQ_FIRST(&ph->ph_itemlist) != NULL)
                    608:                                break;
                    609:
                    610:                pp->pr_curpage = ph;
1.1       pk        611:        }
1.3       pk        612:
                    613:        pp->pr_nget++;
1.1       pk        614:        simple_unlock(&pp->pr_lock);
                    615:        return (v);
                    616: }
                    617:
                    618: /*
1.3       pk        619:  * Return resource to the pool; must be called at appropriate spl level
1.1       pk        620:  */
1.3       pk        621: #ifdef POOL_DIAGNOSTIC
                    622: void
                    623: _pool_put(pp, v, file, line)
                    624:        struct pool *pp;
                    625:        void *v;
                    626:        const char *file;
                    627:        long line;
                    628: #else
1.1       pk        629: void
                    630: pool_put(pp, v)
                    631:        struct pool *pp;
                    632:        void *v;
1.3       pk        633: #endif
1.1       pk        634: {
                    635:        struct pool_item *pi = v;
1.3       pk        636:        struct pool_item_header *ph;
                    637:        caddr_t page;
                    638:
                    639:        page = (caddr_t)((u_long)v & pp->pr_pagemask);
1.1       pk        640:
                    641:        simple_lock(&pp->pr_lock);
1.3       pk        642:
                    643:        pr_log(pp, v, PRLOG_PUT, file, line);
                    644:
                    645:        if ((ph = pr_find_pagehead(pp, page)) == NULL) {
                    646:                pr_printlog(pp);
                    647:                panic("pool_put: %s: page header missing", pp->pr_wchan);
                    648:        }
                    649:
                    650:        /*
                    651:         * Return to item list.
                    652:         */
1.2       pk        653: #ifdef DIAGNOSTIC
1.3       pk        654:        pi->pi_magic = PI_MAGIC;
                    655: #endif
                    656:        TAILQ_INSERT_HEAD(&ph->ph_itemlist, pi, pi_list);
                    657:        ph->ph_nmissing--;
                    658:        pp->pr_nput++;
                    659:
                    660:        /* Cancel "pool empty" condition if it exists */
                    661:        if (pp->pr_curpage == NULL)
                    662:                pp->pr_curpage = ph;
                    663:
                    664:        if (pp->pr_flags & PR_WANTED) {
                    665:                pp->pr_flags &= ~PR_WANTED;
1.15      pk        666:                if (ph->ph_nmissing == 0)
                    667:                        pp->pr_nidle++;
1.3       pk        668:                wakeup((caddr_t)pp);
                    669:                simple_unlock(&pp->pr_lock);
                    670:                return;
                    671:        }
                    672:
                    673:        /*
                    674:         * If this page is now complete, move it to the end of the pagelist.
                    675:         * If this page has just become un-empty, move it the head.
                    676:         */
                    677:        if (ph->ph_nmissing == 0) {
1.6       thorpej   678:                pp->pr_nidle++;
1.3       pk        679:                if (pp->pr_npages > pp->pr_maxpages) {
                    680: #if 0
                    681:                        timeout(pool_drain, 0, pool_inactive_time*hz);
                    682: #else
                    683:                        pr_rmpage(pp, ph);
1.2       pk        684: #endif
1.3       pk        685:                } else {
                    686:                        TAILQ_REMOVE(&pp->pr_pagelist, ph, ph_pagelist);
                    687:                        TAILQ_INSERT_TAIL(&pp->pr_pagelist, ph, ph_pagelist);
                    688:                        ph->ph_time = time;
                    689:
                    690:                        /* XXX - update curpage */
                    691:                        for (ph = TAILQ_FIRST(&pp->pr_pagelist); ph != NULL;
                    692:                             ph = TAILQ_NEXT(ph, ph_pagelist))
                    693:                                if (TAILQ_FIRST(&ph->ph_itemlist) != NULL)
                    694:                                        break;
1.1       pk        695:
1.3       pk        696:                        pp->pr_curpage = ph;
1.1       pk        697:                }
                    698:        }
1.3       pk        699:
1.1       pk        700:        simple_unlock(&pp->pr_lock);
                    701: }
                    702:
                    703: /*
1.3       pk        704:  * Add N items to the pool.
1.1       pk        705:  */
                    706: int
1.2       pk        707: pool_prime(pp, n, storage)
1.1       pk        708:        struct pool *pp;
                    709:        int n;
1.2       pk        710:        caddr_t storage;
1.1       pk        711: {
1.3       pk        712:        caddr_t cp;
                    713:        int newnitems, newpages;
1.2       pk        714:
                    715: #ifdef DIAGNOSTIC
                    716:        if (storage && !(pp->pr_flags & PR_STATIC))
                    717:                panic("pool_prime: static");
                    718:        /* !storage && static caught below */
                    719: #endif
1.1       pk        720:
1.15      pk        721:        (void)lockmgr(&pp->pr_resourcelock, LK_EXCLUSIVE, NULL);
1.3       pk        722:        newnitems = pp->pr_minitems + n;
                    723:        newpages =
1.18      thorpej   724:                roundup(newnitems, pp->pr_itemsperpage) / pp->pr_itemsperpage
1.3       pk        725:                - pp->pr_minpages;
                    726:
                    727:        while (newpages-- > 0) {
                    728:
1.2       pk        729:                if (pp->pr_flags & PR_STATIC) {
1.3       pk        730:                        cp = storage;
                    731:                        storage += pp->pr_pagesz;
                    732:                } else {
                    733:                        cp = (*pp->pr_alloc)(pp->pr_pagesz, 0, pp->pr_mtype);
                    734:                }
1.2       pk        735:
1.3       pk        736:                if (cp == NULL) {
1.15      pk        737:                        (void)lockmgr(&pp->pr_resourcelock, LK_RELEASE, NULL);
1.1       pk        738:                        return (ENOMEM);
                    739:                }
                    740:
1.3       pk        741:                pool_prime_page(pp, cp);
                    742:                pp->pr_minpages++;
1.1       pk        743:        }
1.3       pk        744:
                    745:        pp->pr_minitems = newnitems;
                    746:
                    747:        if (pp->pr_minpages >= pp->pr_maxpages)
                    748:                pp->pr_maxpages = pp->pr_minpages + 1;  /* XXX */
                    749:
1.15      pk        750:        (void)lockmgr(&pp->pr_resourcelock, LK_RELEASE, NULL);
1.1       pk        751:        return (0);
                    752: }
1.3       pk        753:
                    754: /*
                    755:  * Add a page worth of items to the pool.
                    756:  */
                    757: int
                    758: pool_prime_page(pp, storage)
                    759:        struct pool *pp;
                    760:        caddr_t storage;
                    761: {
                    762:        struct pool_item *pi;
                    763:        struct pool_item_header *ph;
                    764:        caddr_t cp = storage;
                    765:        unsigned int align = pp->pr_align;
                    766:        unsigned int ioff = pp->pr_itemoffset;
                    767:        int n;
                    768:
1.15      pk        769:        simple_lock(&pp->pr_lock);
                    770:
1.3       pk        771:        if ((pp->pr_flags & PR_PHINPAGE) != 0) {
                    772:                ph = (struct pool_item_header *)(cp + pp->pr_phoffset);
                    773:        } else {
                    774:                ph = pool_get(&phpool, PR_URGENT);
                    775:                LIST_INSERT_HEAD(&pp->pr_hashtab[PR_HASH_INDEX(pp, cp)],
                    776:                                 ph, ph_hashlist);
                    777:        }
                    778:
                    779:        /*
                    780:         * Insert page header.
                    781:         */
                    782:        TAILQ_INSERT_HEAD(&pp->pr_pagelist, ph, ph_pagelist);
                    783:        TAILQ_INIT(&ph->ph_itemlist);
                    784:        ph->ph_page = storage;
                    785:        ph->ph_nmissing = 0;
                    786:        ph->ph_time.tv_sec = ph->ph_time.tv_usec = 0;
                    787:
1.6       thorpej   788:        pp->pr_nidle++;
                    789:
1.3       pk        790:        /*
                    791:         * Color this page.
                    792:         */
                    793:        cp = (caddr_t)(cp + pp->pr_curcolor);
                    794:        if ((pp->pr_curcolor += align) > pp->pr_maxcolor)
                    795:                pp->pr_curcolor = 0;
                    796:
                    797:        /*
                    798:         * Adjust storage to apply aligment to `pr_itemoffset' in each item.
                    799:         */
                    800:        if (ioff != 0)
                    801:                cp = (caddr_t)(cp + (align - ioff));
                    802:
                    803:        /*
                    804:         * Insert remaining chunks on the bucket list.
                    805:         */
                    806:        n = pp->pr_itemsperpage;
                    807:
                    808:        while (n--) {
                    809:                pi = (struct pool_item *)cp;
                    810:
                    811:                /* Insert on page list */
                    812:                TAILQ_INSERT_TAIL(&ph->ph_itemlist, pi, pi_list);
                    813: #ifdef DIAGNOSTIC
                    814:                pi->pi_magic = PI_MAGIC;
                    815: #endif
                    816:                cp = (caddr_t)(cp + pp->pr_size);
                    817:        }
                    818:
                    819:        /*
                    820:         * If the pool was depleted, point at the new page.
                    821:         */
                    822:        if (pp->pr_curpage == NULL)
                    823:                pp->pr_curpage = ph;
                    824:
                    825:        if (++pp->pr_npages > pp->pr_hiwat)
                    826:                pp->pr_hiwat = pp->pr_npages;
                    827:
1.15      pk        828:        simple_unlock(&pp->pr_lock);
1.3       pk        829:        return (0);
                    830: }
                    831:
                    832: void
                    833: pool_setlowat(pp, n)
                    834:        pool_handle_t   pp;
                    835:        int n;
                    836: {
1.15      pk        837:
                    838:        (void)lockmgr(&pp->pr_resourcelock, LK_EXCLUSIVE, NULL);
1.3       pk        839:        pp->pr_minitems = n;
1.15      pk        840:        pp->pr_minpages = (n == 0)
                    841:                ? 0
1.18      thorpej   842:                : roundup(n, pp->pr_itemsperpage) / pp->pr_itemsperpage;
1.15      pk        843:        (void)lockmgr(&pp->pr_resourcelock, LK_RELEASE, NULL);
1.3       pk        844: }
                    845:
                    846: void
                    847: pool_sethiwat(pp, n)
                    848:        pool_handle_t   pp;
                    849:        int n;
                    850: {
1.15      pk        851:
                    852:        (void)lockmgr(&pp->pr_resourcelock, LK_EXCLUSIVE, NULL);
                    853:        pp->pr_maxpages = (n == 0)
                    854:                ? 0
1.18      thorpej   855:                : roundup(n, pp->pr_itemsperpage) / pp->pr_itemsperpage;
1.15      pk        856:        (void)lockmgr(&pp->pr_resourcelock, LK_RELEASE, NULL);
1.3       pk        857: }
                    858:
                    859:
                    860: /*
                    861:  * Default page allocator.
                    862:  */
                    863: static void *
                    864: pool_page_alloc(sz, flags, mtype)
                    865:        unsigned long sz;
                    866:        int flags;
                    867:        int mtype;
                    868: {
1.11      thorpej   869:        boolean_t waitok = (flags & PR_WAITOK) ? TRUE : FALSE;
1.3       pk        870:
1.11      thorpej   871:        return ((void *)uvm_km_alloc_poolpage(waitok));
1.3       pk        872: }
                    873:
                    874: static void
                    875: pool_page_free(v, sz, mtype)
                    876:        void *v;
                    877:        unsigned long sz;
                    878:        int mtype;
                    879: {
                    880:
1.10      eeh       881:        uvm_km_free_poolpage((vaddr_t)v);
1.3       pk        882: }
1.12      thorpej   883:
                    884: /*
                    885:  * Alternate pool page allocator for pools that know they will
                    886:  * never be accessed in interrupt context.
                    887:  */
                    888: void *
                    889: pool_page_alloc_nointr(sz, flags, mtype)
                    890:        unsigned long sz;
                    891:        int flags;
                    892:        int mtype;
                    893: {
                    894:        boolean_t waitok = (flags & PR_WAITOK) ? TRUE : FALSE;
                    895:
                    896:        /*
                    897:         * With UVM, we can use the kernel_map.
                    898:         */
                    899:        return ((void *)uvm_km_alloc_poolpage1(kernel_map, uvm.kernel_object,
                    900:            waitok));
                    901: }
                    902:
                    903: void
                    904: pool_page_free_nointr(v, sz, mtype)
                    905:        void *v;
                    906:        unsigned long sz;
                    907:        int mtype;
                    908: {
                    909:
                    910:        uvm_km_free_poolpage1(kernel_map, (vaddr_t)v);
                    911: }
                    912:
1.3       pk        913:
                    914: /*
                    915:  * Release all complete pages that have not been used recently.
                    916:  */
                    917: void
                    918: pool_reclaim (pp)
                    919:        pool_handle_t pp;
                    920: {
                    921:        struct pool_item_header *ph, *phnext;
                    922:        struct timeval curtime = time;
                    923:
                    924:        if (pp->pr_flags & PR_STATIC)
                    925:                return;
                    926:
                    927:        if (simple_lock_try(&pp->pr_lock) == 0)
                    928:                return;
                    929:
                    930:        for (ph = TAILQ_FIRST(&pp->pr_pagelist); ph != NULL; ph = phnext) {
                    931:                phnext = TAILQ_NEXT(ph, ph_pagelist);
                    932:
                    933:                /* Check our minimum page claim */
                    934:                if (pp->pr_npages <= pp->pr_minpages)
                    935:                        break;
                    936:
                    937:                if (ph->ph_nmissing == 0) {
                    938:                        struct timeval diff;
                    939:                        timersub(&curtime, &ph->ph_time, &diff);
                    940:                        if (diff.tv_sec < pool_inactive_time)
                    941:                                continue;
                    942:                        pr_rmpage(pp, ph);
                    943:                }
                    944:        }
                    945:
                    946:        simple_unlock(&pp->pr_lock);
                    947: }
                    948:
                    949:
                    950: /*
                    951:  * Drain pools, one at a time.
                    952:  */
                    953: void
                    954: pool_drain(arg)
                    955:        void *arg;
                    956: {
                    957:        struct pool *pp;
                    958:        int s = splimp();
                    959:
                    960:        /* XXX:lock pool head */
                    961:        if (drainpp == NULL && (drainpp = TAILQ_FIRST(&pool_head)) == NULL) {
                    962:                splx(s);
                    963:                return;
                    964:        }
                    965:
                    966:        pp = drainpp;
                    967:        drainpp = TAILQ_NEXT(pp, pr_poollist);
                    968:        /* XXX:unlock pool head */
                    969:
                    970:        pool_reclaim(pp);
                    971:        splx(s);
                    972: }
                    973:
                    974:
1.17      thorpej   975: #if defined(POOL_DIAGNOSTIC) || defined(DEBUG)
1.3       pk        976: /*
                    977:  * Diagnostic helpers.
                    978:  */
                    979: void
                    980: pool_print(pp, label)
                    981:        struct pool *pp;
                    982:        char *label;
                    983: {
                    984:
                    985:        if (label != NULL)
                    986:                printf("%s: ", label);
                    987:
                    988:        printf("pool %s: nalloc %lu nfree %lu npagealloc %lu npagefree %lu\n"
1.6       thorpej   989:               "         npages %u minitems %u itemsperpage %u itemoffset %u\n"
                    990:               "         nidle %lu\n",
1.3       pk        991:                pp->pr_wchan,
                    992:                pp->pr_nget,
                    993:                pp->pr_nput,
                    994:                pp->pr_npagealloc,
                    995:                pp->pr_npagefree,
                    996:                pp->pr_npages,
                    997:                pp->pr_minitems,
                    998:                pp->pr_itemsperpage,
1.6       thorpej   999:                pp->pr_itemoffset,
                   1000:                pp->pr_nidle);
1.3       pk       1001: }
                   1002:
                   1003: int
                   1004: pool_chk(pp, label)
                   1005:        struct pool *pp;
                   1006:        char *label;
                   1007: {
                   1008:        struct pool_item_header *ph;
                   1009:        int r = 0;
                   1010:
                   1011:        simple_lock(&pp->pr_lock);
                   1012:
                   1013:        for (ph = TAILQ_FIRST(&pp->pr_pagelist); ph != NULL;
                   1014:             ph = TAILQ_NEXT(ph, ph_pagelist)) {
                   1015:
                   1016:                struct pool_item *pi;
                   1017:                int n;
                   1018:                caddr_t page;
                   1019:
                   1020:                page = (caddr_t)((u_long)ph & pp->pr_pagemask);
1.16      briggs   1021:                if (page != ph->ph_page && (pp->pr_flags & PR_PHINPAGE) != 0) {
1.3       pk       1022:                        if (label != NULL)
                   1023:                                printf("%s: ", label);
1.16      briggs   1024:                        printf("pool(%p:%s): page inconsistency: page %p;"
                   1025:                               " at page head addr %p (p %p)\n", pp,
1.3       pk       1026:                                pp->pr_wchan, ph->ph_page,
                   1027:                                ph, page);
                   1028:                        r++;
                   1029:                        goto out;
                   1030:                }
                   1031:
                   1032:                for (pi = TAILQ_FIRST(&ph->ph_itemlist), n = 0;
                   1033:                     pi != NULL;
                   1034:                     pi = TAILQ_NEXT(pi,pi_list), n++) {
                   1035:
                   1036: #ifdef DIAGNOSTIC
                   1037:                        if (pi->pi_magic != PI_MAGIC) {
                   1038:                                if (label != NULL)
                   1039:                                        printf("%s: ", label);
                   1040:                                printf("pool(%s): free list modified: magic=%x;"
                   1041:                                       " page %p; item ordinal %d;"
                   1042:                                       " addr %p (p %p)\n",
                   1043:                                        pp->pr_wchan, pi->pi_magic, ph->ph_page,
                   1044:                                        n, pi, page);
                   1045:                                panic("pool");
                   1046:                        }
                   1047: #endif
                   1048:                        page = (caddr_t)((u_long)pi & pp->pr_pagemask);
                   1049:                        if (page == ph->ph_page)
                   1050:                                continue;
                   1051:
                   1052:                        if (label != NULL)
                   1053:                                printf("%s: ", label);
1.16      briggs   1054:                        printf("pool(%p:%s): page inconsistency: page %p;"
                   1055:                               " item ordinal %d; addr %p (p %p)\n", pp,
1.3       pk       1056:                                pp->pr_wchan, ph->ph_page,
                   1057:                                n, pi, page);
                   1058:                        r++;
                   1059:                        goto out;
                   1060:                }
                   1061:        }
                   1062: out:
                   1063:        simple_unlock(&pp->pr_lock);
                   1064:        return (r);
                   1065: }
1.17      thorpej  1066: #endif /* POOL_DIAGNOSTIC || DEBUG */

CVSweb <webmaster@jp.NetBSD.org>