[BACK]Return to uvm_aobj.c CVS log [TXT][DIR] Up to [cvs.NetBSD.org] / src / sys / uvm

Annotation of src/sys/uvm/uvm_aobj.c, Revision 1.31

1.31    ! thorpej     1: /*     $NetBSD: uvm_aobj.c,v 1.30 2000/04/10 00:32:46 thorpej Exp $    */
1.6       mrg         2:
1.7       chs         3: /*
                      4:  * Copyright (c) 1998 Chuck Silvers, Charles D. Cranor and
                      5:  *                    Washington University.
                      6:  * All rights reserved.
                      7:  *
                      8:  * Redistribution and use in source and binary forms, with or without
                      9:  * modification, are permitted provided that the following conditions
                     10:  * are met:
                     11:  * 1. Redistributions of source code must retain the above copyright
                     12:  *    notice, this list of conditions and the following disclaimer.
                     13:  * 2. Redistributions in binary form must reproduce the above copyright
                     14:  *    notice, this list of conditions and the following disclaimer in the
                     15:  *    documentation and/or other materials provided with the distribution.
                     16:  * 3. All advertising materials mentioning features or use of this software
                     17:  *    must display the following acknowledgement:
                     18:  *      This product includes software developed by Charles D. Cranor and
                     19:  *      Washington University.
                     20:  * 4. The name of the author may not be used to endorse or promote products
                     21:  *    derived from this software without specific prior written permission.
                     22:  *
                     23:  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
                     24:  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
                     25:  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
                     26:  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
                     27:  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
                     28:  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
                     29:  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
                     30:  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
                     31:  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
                     32:  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
                     33:  *
1.4       mrg        34:  * from: Id: uvm_aobj.c,v 1.1.2.5 1998/02/06 05:14:38 chs Exp
                     35:  */
1.7       chs        36: /*
                     37:  * uvm_aobj.c: anonymous memory uvm_object pager
                     38:  *
                     39:  * author: Chuck Silvers <chuq@chuq.com>
                     40:  * started: Jan-1998
                     41:  *
                     42:  * - design mostly from Chuck Cranor
                     43:  */
                     44:
                     45:
                     46:
                     47: #include "opt_uvmhist.h"
1.1       mrg        48:
                     49: #include <sys/param.h>
                     50: #include <sys/systm.h>
                     51: #include <sys/proc.h>
                     52: #include <sys/malloc.h>
1.12      thorpej    53: #include <sys/pool.h>
1.27      chs        54: #include <sys/kernel.h>
1.1       mrg        55:
                     56: #include <vm/vm.h>
                     57: #include <vm/vm_page.h>
                     58: #include <vm/vm_kern.h>
                     59:
                     60: #include <uvm/uvm.h>
                     61:
                     62: /*
                     63:  * an aobj manages anonymous-memory backed uvm_objects.   in addition
                     64:  * to keeping the list of resident pages, it also keeps a list of
                     65:  * allocated swap blocks.  depending on the size of the aobj this list
                     66:  * of allocated swap blocks is either stored in an array (small objects)
                     67:  * or in a hash table (large objects).
                     68:  */
                     69:
                     70: /*
                     71:  * local structures
                     72:  */
                     73:
                     74: /*
                     75:  * for hash tables, we break the address space of the aobj into blocks
                     76:  * of UAO_SWHASH_CLUSTER_SIZE pages.   we require the cluster size to
                     77:  * be a power of two.
                     78:  */
                     79:
                     80: #define UAO_SWHASH_CLUSTER_SHIFT 4
                     81: #define UAO_SWHASH_CLUSTER_SIZE (1 << UAO_SWHASH_CLUSTER_SHIFT)
                     82:
                     83: /* get the "tag" for this page index */
                     84: #define UAO_SWHASH_ELT_TAG(PAGEIDX) \
                     85:        ((PAGEIDX) >> UAO_SWHASH_CLUSTER_SHIFT)
                     86:
                     87: /* given an ELT and a page index, find the swap slot */
                     88: #define UAO_SWHASH_ELT_PAGESLOT(ELT, PAGEIDX) \
                     89:        ((ELT)->slots[(PAGEIDX) & (UAO_SWHASH_CLUSTER_SIZE - 1)])
                     90:
                     91: /* given an ELT, return its pageidx base */
                     92: #define UAO_SWHASH_ELT_PAGEIDX_BASE(ELT) \
                     93:        ((ELT)->tag << UAO_SWHASH_CLUSTER_SHIFT)
                     94:
                     95: /*
                     96:  * the swhash hash function
                     97:  */
                     98: #define UAO_SWHASH_HASH(AOBJ, PAGEIDX) \
                     99:        (&(AOBJ)->u_swhash[(((PAGEIDX) >> UAO_SWHASH_CLUSTER_SHIFT) \
                    100:                            & (AOBJ)->u_swhashmask)])
                    101:
                    102: /*
                    103:  * the swhash threshhold determines if we will use an array or a
                    104:  * hash table to store the list of allocated swap blocks.
                    105:  */
                    106:
                    107: #define UAO_SWHASH_THRESHOLD (UAO_SWHASH_CLUSTER_SIZE * 4)
                    108: #define UAO_USES_SWHASH(AOBJ) \
                    109:        ((AOBJ)->u_pages > UAO_SWHASH_THRESHOLD)        /* use hash? */
                    110:
                    111: /*
1.3       chs       112:  * the number of buckets in a swhash, with an upper bound
1.1       mrg       113:  */
                    114: #define UAO_SWHASH_MAXBUCKETS 256
                    115: #define UAO_SWHASH_BUCKETS(AOBJ) \
                    116:        (min((AOBJ)->u_pages >> UAO_SWHASH_CLUSTER_SHIFT, \
                    117:             UAO_SWHASH_MAXBUCKETS))
                    118:
                    119:
                    120: /*
                    121:  * uao_swhash_elt: when a hash table is being used, this structure defines
                    122:  * the format of an entry in the bucket list.
                    123:  */
                    124:
                    125: struct uao_swhash_elt {
1.5       mrg       126:        LIST_ENTRY(uao_swhash_elt) list;        /* the hash list */
1.28      kleink    127:        voff_t tag;                             /* our 'tag' */
1.5       mrg       128:        int count;                              /* our number of active slots */
                    129:        int slots[UAO_SWHASH_CLUSTER_SIZE];     /* the slots */
1.1       mrg       130: };
                    131:
                    132: /*
                    133:  * uao_swhash: the swap hash table structure
                    134:  */
                    135:
                    136: LIST_HEAD(uao_swhash, uao_swhash_elt);
                    137:
1.12      thorpej   138: /*
                    139:  * uao_swhash_elt_pool: pool of uao_swhash_elt structures
                    140:  */
                    141:
                    142: struct pool uao_swhash_elt_pool;
1.1       mrg       143:
                    144: /*
                    145:  * uvm_aobj: the actual anon-backed uvm_object
                    146:  *
                    147:  * => the uvm_object is at the top of the structure, this allows
                    148:  *   (struct uvm_device *) == (struct uvm_object *)
                    149:  * => only one of u_swslots and u_swhash is used in any given aobj
                    150:  */
                    151:
                    152: struct uvm_aobj {
1.5       mrg       153:        struct uvm_object u_obj; /* has: lock, pgops, memq, #pages, #refs */
1.11      drochner  154:        int u_pages;             /* number of pages in entire object */
1.5       mrg       155:        int u_flags;             /* the flags (see uvm_aobj.h) */
                    156:        int *u_swslots;          /* array of offset->swapslot mappings */
                    157:                                 /*
                    158:                                  * hashtable of offset->swapslot mappings
                    159:                                  * (u_swhash is an array of bucket heads)
                    160:                                  */
                    161:        struct uao_swhash *u_swhash;
                    162:        u_long u_swhashmask;            /* mask for hashtable */
                    163:        LIST_ENTRY(uvm_aobj) u_list;    /* global list of aobjs */
1.1       mrg       164: };
                    165:
                    166: /*
1.12      thorpej   167:  * uvm_aobj_pool: pool of uvm_aobj structures
                    168:  */
                    169:
                    170: struct pool uvm_aobj_pool;
                    171:
                    172: /*
1.1       mrg       173:  * local functions
                    174:  */
                    175:
                    176: static struct uao_swhash_elt   *uao_find_swhash_elt __P((struct uvm_aobj *,
                    177:                                                          int, boolean_t));
1.28      kleink    178: static int                      uao_find_swslot __P((struct uvm_aobj *, int));
                    179: static boolean_t                uao_flush __P((struct uvm_object *,
                    180:                                                voff_t, voff_t, int));
1.1       mrg       181: static void                     uao_free __P((struct uvm_aobj *));
1.28      kleink    182: static int                      uao_get __P((struct uvm_object *, voff_t,
                    183:                                              vm_page_t *, int *, int,
1.1       mrg       184:                                              vm_prot_t, int, int));
1.28      kleink    185: static boolean_t                uao_releasepg __P((struct vm_page *,
1.1       mrg       186:                                                    struct vm_page **));
1.27      chs       187: static boolean_t                uao_pagein __P((struct uvm_aobj *, int, int));
                    188: static boolean_t                uao_pagein_page __P((struct uvm_aobj *, int));
1.1       mrg       189:
                    190:
                    191:
                    192: /*
                    193:  * aobj_pager
                    194:  *
                    195:  * note that some functions (e.g. put) are handled elsewhere
                    196:  */
                    197:
                    198: struct uvm_pagerops aobj_pager = {
1.27      chs       199:        NULL,                   /* init */
1.5       mrg       200:        uao_reference,          /* reference */
                    201:        uao_detach,             /* detach */
                    202:        NULL,                   /* fault */
                    203:        uao_flush,              /* flush */
                    204:        uao_get,                /* get */
                    205:        NULL,                   /* asyncget */
                    206:        NULL,                   /* put (done by pagedaemon) */
                    207:        NULL,                   /* cluster */
                    208:        NULL,                   /* mk_pcluster */
                    209:        NULL,                   /* aiodone */
                    210:        uao_releasepg           /* releasepg */
1.1       mrg       211: };
                    212:
                    213: /*
                    214:  * uao_list: global list of active aobjs, locked by uao_list_lock
                    215:  */
                    216:
                    217: static LIST_HEAD(aobjlist, uvm_aobj) uao_list;
                    218: static simple_lock_data_t uao_list_lock;
                    219:
                    220:
                    221: /*
                    222:  * functions
                    223:  */
                    224:
                    225: /*
                    226:  * hash table/array related functions
                    227:  */
                    228:
                    229: /*
                    230:  * uao_find_swhash_elt: find (or create) a hash table entry for a page
                    231:  * offset.
                    232:  *
                    233:  * => the object should be locked by the caller
                    234:  */
                    235:
1.5       mrg       236: static struct uao_swhash_elt *
                    237: uao_find_swhash_elt(aobj, pageidx, create)
                    238:        struct uvm_aobj *aobj;
                    239:        int pageidx;
                    240:        boolean_t create;
                    241: {
                    242:        struct uao_swhash *swhash;
                    243:        struct uao_swhash_elt *elt;
1.28      kleink    244:        voff_t page_tag;
1.1       mrg       245:
1.5       mrg       246:        swhash = UAO_SWHASH_HASH(aobj, pageidx); /* first hash to get bucket */
                    247:        page_tag = UAO_SWHASH_ELT_TAG(pageidx); /* tag to search for */
1.1       mrg       248:
1.5       mrg       249:        /*
                    250:         * now search the bucket for the requested tag
                    251:         */
                    252:        for (elt = swhash->lh_first; elt != NULL; elt = elt->list.le_next) {
                    253:                if (elt->tag == page_tag)
                    254:                        return(elt);
                    255:        }
                    256:
                    257:        /* fail now if we are not allowed to create a new entry in the bucket */
                    258:        if (!create)
                    259:                return NULL;
                    260:
                    261:
                    262:        /*
1.12      thorpej   263:         * allocate a new entry for the bucket and init/insert it in
1.5       mrg       264:         */
1.12      thorpej   265:        elt = pool_get(&uao_swhash_elt_pool, PR_WAITOK);
1.5       mrg       266:        LIST_INSERT_HEAD(swhash, elt, list);
                    267:        elt->tag = page_tag;
                    268:        elt->count = 0;
1.9       perry     269:        memset(elt->slots, 0, sizeof(elt->slots));
1.5       mrg       270:
                    271:        return(elt);
1.1       mrg       272: }
                    273:
                    274: /*
                    275:  * uao_find_swslot: find the swap slot number for an aobj/pageidx
                    276:  *
                    277:  * => object must be locked by caller
                    278:  */
1.5       mrg       279: __inline static int
                    280: uao_find_swslot(aobj, pageidx)
                    281:        struct uvm_aobj *aobj;
1.11      drochner  282:        int pageidx;
1.1       mrg       283: {
                    284:
1.5       mrg       285:        /*
                    286:         * if noswap flag is set, then we never return a slot
                    287:         */
1.1       mrg       288:
1.5       mrg       289:        if (aobj->u_flags & UAO_FLAG_NOSWAP)
                    290:                return(0);
1.1       mrg       291:
1.5       mrg       292:        /*
                    293:         * if hashing, look in hash table.
                    294:         */
1.1       mrg       295:
1.5       mrg       296:        if (UAO_USES_SWHASH(aobj)) {
                    297:                struct uao_swhash_elt *elt =
                    298:                    uao_find_swhash_elt(aobj, pageidx, FALSE);
                    299:
                    300:                if (elt)
                    301:                        return(UAO_SWHASH_ELT_PAGESLOT(elt, pageidx));
                    302:                else
1.31    ! thorpej   303:                        return(0);
1.5       mrg       304:        }
1.1       mrg       305:
1.5       mrg       306:        /*
                    307:         * otherwise, look in the array
                    308:         */
                    309:        return(aobj->u_swslots[pageidx]);
1.1       mrg       310: }
                    311:
                    312: /*
                    313:  * uao_set_swslot: set the swap slot for a page in an aobj.
                    314:  *
                    315:  * => setting a slot to zero frees the slot
                    316:  * => object must be locked by caller
                    317:  */
1.5       mrg       318: int
                    319: uao_set_swslot(uobj, pageidx, slot)
                    320:        struct uvm_object *uobj;
                    321:        int pageidx, slot;
                    322: {
                    323:        struct uvm_aobj *aobj = (struct uvm_aobj *)uobj;
                    324:        int oldslot;
                    325:        UVMHIST_FUNC("uao_set_swslot"); UVMHIST_CALLED(pdhist);
                    326:        UVMHIST_LOG(pdhist, "aobj %p pageidx %d slot %d",
                    327:            aobj, pageidx, slot, 0);
1.1       mrg       328:
1.5       mrg       329:        /*
                    330:         * if noswap flag is set, then we can't set a slot
                    331:         */
1.1       mrg       332:
1.5       mrg       333:        if (aobj->u_flags & UAO_FLAG_NOSWAP) {
1.1       mrg       334:
1.5       mrg       335:                if (slot == 0)
                    336:                        return(0);              /* a clear is ok */
1.1       mrg       337:
1.5       mrg       338:                /* but a set is not */
                    339:                printf("uao_set_swslot: uobj = %p\n", uobj);
                    340:            panic("uao_set_swslot: attempt to set a slot on a NOSWAP object");
                    341:        }
1.1       mrg       342:
1.5       mrg       343:        /*
                    344:         * are we using a hash table?  if so, add it in the hash.
                    345:         */
1.1       mrg       346:
1.5       mrg       347:        if (UAO_USES_SWHASH(aobj)) {
1.12      thorpej   348:                /*
                    349:                 * Avoid allocating an entry just to free it again if
                    350:                 * the page had not swap slot in the first place, and
                    351:                 * we are freeing.
                    352:                 */
1.5       mrg       353:                struct uao_swhash_elt *elt =
1.12      thorpej   354:                    uao_find_swhash_elt(aobj, pageidx, slot ? TRUE : FALSE);
                    355:                if (elt == NULL) {
                    356: #ifdef DIAGNOSTIC
                    357:                        if (slot)
                    358:                                panic("uao_set_swslot: didn't create elt");
                    359: #endif
                    360:                        return (0);
                    361:                }
1.5       mrg       362:
                    363:                oldslot = UAO_SWHASH_ELT_PAGESLOT(elt, pageidx);
                    364:                UAO_SWHASH_ELT_PAGESLOT(elt, pageidx) = slot;
                    365:
                    366:                /*
                    367:                 * now adjust the elt's reference counter and free it if we've
                    368:                 * dropped it to zero.
                    369:                 */
                    370:
                    371:                /* an allocation? */
                    372:                if (slot) {
                    373:                        if (oldslot == 0)
                    374:                                elt->count++;
                    375:                } else {                /* freeing slot ... */
                    376:                        if (oldslot)    /* to be safe */
                    377:                                elt->count--;
                    378:
                    379:                        if (elt->count == 0) {
                    380:                                LIST_REMOVE(elt, list);
1.12      thorpej   381:                                pool_put(&uao_swhash_elt_pool, elt);
1.5       mrg       382:                        }
                    383:                }
                    384:
                    385:        } else {
                    386:                /* we are using an array */
                    387:                oldslot = aobj->u_swslots[pageidx];
                    388:                aobj->u_swslots[pageidx] = slot;
                    389:        }
                    390:        return (oldslot);
1.1       mrg       391: }
                    392:
                    393: /*
                    394:  * end of hash/array functions
                    395:  */
                    396:
                    397: /*
                    398:  * uao_free: free all resources held by an aobj, and then free the aobj
                    399:  *
                    400:  * => the aobj should be dead
                    401:  */
                    402: static void
                    403: uao_free(aobj)
1.5       mrg       404:        struct uvm_aobj *aobj;
1.1       mrg       405: {
                    406:
1.27      chs       407:        simple_unlock(&aobj->u_obj.vmobjlock);
                    408:
1.5       mrg       409:        if (UAO_USES_SWHASH(aobj)) {
                    410:                int i, hashbuckets = aobj->u_swhashmask + 1;
1.1       mrg       411:
1.5       mrg       412:                /*
                    413:                 * free the swslots from each hash bucket,
                    414:                 * then the hash bucket, and finally the hash table itself.
                    415:                 */
                    416:                for (i = 0; i < hashbuckets; i++) {
                    417:                        struct uao_swhash_elt *elt, *next;
                    418:
1.27      chs       419:                        for (elt = LIST_FIRST(&aobj->u_swhash[i]);
                    420:                             elt != NULL;
                    421:                             elt = next) {
1.5       mrg       422:                                int j;
                    423:
1.27      chs       424:                                for (j = 0; j < UAO_SWHASH_CLUSTER_SIZE; j++) {
1.5       mrg       425:                                        int slot = elt->slots[j];
                    426:
1.18      chs       427:                                        if (slot) {
1.5       mrg       428:                                                uvm_swap_free(slot, 1);
1.18      chs       429:
                    430:                                                /*
                    431:                                                 * this page is no longer
                    432:                                                 * only in swap.
                    433:                                                 */
                    434:                                                simple_lock(&uvm.swap_data_lock);
                    435:                                                uvmexp.swpgonly--;
                    436:                                                simple_unlock(&uvm.swap_data_lock);
                    437:                                        }
1.5       mrg       438:                                }
                    439:
1.27      chs       440:                                next = LIST_NEXT(elt, list);
1.12      thorpej   441:                                pool_put(&uao_swhash_elt_pool, elt);
1.5       mrg       442:                        }
                    443:                }
                    444:                FREE(aobj->u_swhash, M_UVMAOBJ);
                    445:        } else {
                    446:                int i;
                    447:
                    448:                /*
                    449:                 * free the array
                    450:                 */
                    451:
1.27      chs       452:                for (i = 0; i < aobj->u_pages; i++) {
1.5       mrg       453:                        int slot = aobj->u_swslots[i];
                    454:
1.18      chs       455:                        if (slot) {
1.5       mrg       456:                                uvm_swap_free(slot, 1);
1.18      chs       457:
                    458:                                /* this page is no longer only in swap. */
                    459:                                simple_lock(&uvm.swap_data_lock);
                    460:                                uvmexp.swpgonly--;
                    461:                                simple_unlock(&uvm.swap_data_lock);
                    462:                        }
1.5       mrg       463:                }
                    464:                FREE(aobj->u_swslots, M_UVMAOBJ);
1.1       mrg       465:        }
                    466:
1.5       mrg       467:        /*
                    468:         * finally free the aobj itself
                    469:         */
1.12      thorpej   470:        pool_put(&uvm_aobj_pool, aobj);
1.1       mrg       471: }
                    472:
                    473: /*
                    474:  * pager functions
                    475:  */
                    476:
                    477: /*
                    478:  * uao_create: create an aobj of the given size and return its uvm_object.
                    479:  *
                    480:  * => for normal use, flags are always zero
                    481:  * => for the kernel object, the flags are:
                    482:  *     UAO_FLAG_KERNOBJ - allocate the kernel object (can only happen once)
                    483:  *     UAO_FLAG_KERNSWAP - enable swapping of kernel object ("           ")
                    484:  */
1.5       mrg       485: struct uvm_object *
                    486: uao_create(size, flags)
1.10      eeh       487:        vsize_t size;
1.5       mrg       488:        int flags;
                    489: {
1.27      chs       490:        static struct uvm_aobj kernel_object_store; /* home of kernel_object */
1.5       mrg       491:        static int kobj_alloced = 0;                    /* not allocated yet */
1.15      chs       492:        int pages = round_page(size) >> PAGE_SHIFT;
1.5       mrg       493:        struct uvm_aobj *aobj;
1.1       mrg       494:
1.5       mrg       495:        /*
1.27      chs       496:         * malloc a new aobj unless we are asked for the kernel object
                    497:         */
1.5       mrg       498:        if (flags & UAO_FLAG_KERNOBJ) {         /* want kernel object? */
                    499:                if (kobj_alloced)
                    500:                        panic("uao_create: kernel object already allocated");
                    501:
                    502:                aobj = &kernel_object_store;
                    503:                aobj->u_pages = pages;
                    504:                aobj->u_flags = UAO_FLAG_NOSWAP;        /* no swap to start */
                    505:                /* we are special, we never die */
                    506:                aobj->u_obj.uo_refs = UVM_OBJ_KERN;
                    507:                kobj_alloced = UAO_FLAG_KERNOBJ;
                    508:        } else if (flags & UAO_FLAG_KERNSWAP) {
                    509:                aobj = &kernel_object_store;
                    510:                if (kobj_alloced != UAO_FLAG_KERNOBJ)
                    511:                    panic("uao_create: asked to enable swap on kernel object");
                    512:                kobj_alloced = UAO_FLAG_KERNSWAP;
                    513:        } else {        /* normal object */
1.12      thorpej   514:                aobj = pool_get(&uvm_aobj_pool, PR_WAITOK);
1.5       mrg       515:                aobj->u_pages = pages;
                    516:                aobj->u_flags = 0;              /* normal object */
                    517:                aobj->u_obj.uo_refs = 1;        /* start with 1 reference */
                    518:        }
1.1       mrg       519:
1.5       mrg       520:        /*
                    521:         * allocate hash/array if necessary
                    522:         *
                    523:         * note: in the KERNSWAP case no need to worry about locking since
                    524:         * we are still booting we should be the only thread around.
                    525:         */
                    526:        if (flags == 0 || (flags & UAO_FLAG_KERNSWAP) != 0) {
                    527:                int mflags = (flags & UAO_FLAG_KERNSWAP) != 0 ?
                    528:                    M_NOWAIT : M_WAITOK;
                    529:
                    530:                /* allocate hash table or array depending on object size */
1.27      chs       531:                if (UAO_USES_SWHASH(aobj)) {
1.5       mrg       532:                        aobj->u_swhash = hashinit(UAO_SWHASH_BUCKETS(aobj),
                    533:                            M_UVMAOBJ, mflags, &aobj->u_swhashmask);
                    534:                        if (aobj->u_swhash == NULL)
                    535:                                panic("uao_create: hashinit swhash failed");
                    536:                } else {
                    537:                        MALLOC(aobj->u_swslots, int *, pages * sizeof(int),
                    538:                            M_UVMAOBJ, mflags);
                    539:                        if (aobj->u_swslots == NULL)
                    540:                                panic("uao_create: malloc swslots failed");
1.9       perry     541:                        memset(aobj->u_swslots, 0, pages * sizeof(int));
1.5       mrg       542:                }
                    543:
                    544:                if (flags) {
                    545:                        aobj->u_flags &= ~UAO_FLAG_NOSWAP; /* clear noswap */
                    546:                        return(&aobj->u_obj);
                    547:                        /* done! */
                    548:                }
                    549:        }
                    550:
                    551:        /*
                    552:         * init aobj fields
                    553:         */
                    554:        simple_lock_init(&aobj->u_obj.vmobjlock);
                    555:        aobj->u_obj.pgops = &aobj_pager;
                    556:        TAILQ_INIT(&aobj->u_obj.memq);
                    557:        aobj->u_obj.uo_npages = 0;
1.1       mrg       558:
1.5       mrg       559:        /*
                    560:         * now that aobj is ready, add it to the global list
                    561:         */
                    562:        simple_lock(&uao_list_lock);
                    563:        LIST_INSERT_HEAD(&uao_list, aobj, u_list);
                    564:        simple_unlock(&uao_list_lock);
                    565:
                    566:        /*
                    567:         * done!
                    568:         */
                    569:        return(&aobj->u_obj);
1.1       mrg       570: }
                    571:
                    572:
                    573:
                    574: /*
                    575:  * uao_init: set up aobj pager subsystem
                    576:  *
                    577:  * => called at boot time from uvm_pager_init()
                    578:  */
1.27      chs       579: void
1.5       mrg       580: uao_init()
                    581: {
1.12      thorpej   582:        static int uao_initialized;
                    583:
                    584:        if (uao_initialized)
                    585:                return;
                    586:        uao_initialized = TRUE;
1.1       mrg       587:
1.5       mrg       588:        LIST_INIT(&uao_list);
                    589:        simple_lock_init(&uao_list_lock);
1.12      thorpej   590:
1.14      thorpej   591:        /*
                    592:         * NOTE: Pages fror this pool must not come from a pageable
                    593:         * kernel map!
                    594:         */
1.12      thorpej   595:        pool_init(&uao_swhash_elt_pool, sizeof(struct uao_swhash_elt),
1.13      thorpej   596:            0, 0, 0, "uaoeltpl", 0, NULL, NULL, M_UVMAOBJ);
1.12      thorpej   597:
                    598:        pool_init(&uvm_aobj_pool, sizeof(struct uvm_aobj), 0, 0, 0,
                    599:            "aobjpl", 0,
                    600:            pool_page_alloc_nointr, pool_page_free_nointr, M_UVMAOBJ);
1.1       mrg       601: }
                    602:
                    603: /*
                    604:  * uao_reference: add a ref to an aobj
                    605:  *
1.27      chs       606:  * => aobj must be unlocked
                    607:  * => just lock it and call the locked version
1.1       mrg       608:  */
1.5       mrg       609: void
                    610: uao_reference(uobj)
                    611:        struct uvm_object *uobj;
1.1       mrg       612: {
1.27      chs       613:        simple_lock(&uobj->vmobjlock);
                    614:        uao_reference_locked(uobj);
                    615:        simple_unlock(&uobj->vmobjlock);
                    616: }
                    617:
                    618: /*
                    619:  * uao_reference_locked: add a ref to an aobj that is already locked
                    620:  *
                    621:  * => aobj must be locked
                    622:  * this needs to be separate from the normal routine
                    623:  * since sometimes we need to add a reference to an aobj when
                    624:  * it's already locked.
                    625:  */
                    626: void
                    627: uao_reference_locked(uobj)
                    628:        struct uvm_object *uobj;
                    629: {
1.5       mrg       630:        UVMHIST_FUNC("uao_reference"); UVMHIST_CALLED(maphist);
1.1       mrg       631:
1.5       mrg       632:        /*
                    633:         * kernel_object already has plenty of references, leave it alone.
                    634:         */
1.1       mrg       635:
1.20      thorpej   636:        if (UVM_OBJ_IS_KERN_OBJECT(uobj))
1.5       mrg       637:                return;
1.1       mrg       638:
1.5       mrg       639:        uobj->uo_refs++;                /* bump! */
                    640:        UVMHIST_LOG(maphist, "<- done (uobj=0x%x, ref = %d)",
1.27      chs       641:                    uobj, uobj->uo_refs,0,0);
1.1       mrg       642: }
                    643:
1.27      chs       644:
1.1       mrg       645: /*
                    646:  * uao_detach: drop a reference to an aobj
                    647:  *
1.27      chs       648:  * => aobj must be unlocked
                    649:  * => just lock it and call the locked version
1.1       mrg       650:  */
1.5       mrg       651: void
                    652: uao_detach(uobj)
                    653:        struct uvm_object *uobj;
                    654: {
1.27      chs       655:        simple_lock(&uobj->vmobjlock);
                    656:        uao_detach_locked(uobj);
                    657: }
                    658:
                    659:
                    660: /*
                    661:  * uao_detach_locked: drop a reference to an aobj
                    662:  *
                    663:  * => aobj must be locked, and is unlocked (or freed) upon return.
                    664:  * this needs to be separate from the normal routine
                    665:  * since sometimes we need to detach from an aobj when
                    666:  * it's already locked.
                    667:  */
                    668: void
                    669: uao_detach_locked(uobj)
                    670:        struct uvm_object *uobj;
                    671: {
1.5       mrg       672:        struct uvm_aobj *aobj = (struct uvm_aobj *)uobj;
                    673:        struct vm_page *pg;
                    674:        boolean_t busybody;
                    675:        UVMHIST_FUNC("uao_detach"); UVMHIST_CALLED(maphist);
1.1       mrg       676:
1.5       mrg       677:        /*
                    678:         * detaching from kernel_object is a noop.
                    679:         */
1.27      chs       680:        if (UVM_OBJ_IS_KERN_OBJECT(uobj)) {
                    681:                simple_unlock(&uobj->vmobjlock);
1.5       mrg       682:                return;
1.27      chs       683:        }
1.5       mrg       684:
                    685:        UVMHIST_LOG(maphist,"  (uobj=0x%x)  ref=%d", uobj,uobj->uo_refs,0,0);
                    686:        uobj->uo_refs--;                                /* drop ref! */
                    687:        if (uobj->uo_refs) {                            /* still more refs? */
                    688:                simple_unlock(&uobj->vmobjlock);
                    689:                UVMHIST_LOG(maphist, "<- done (rc>0)", 0,0,0,0);
                    690:                return;
                    691:        }
                    692:
                    693:        /*
                    694:         * remove the aobj from the global list.
                    695:         */
                    696:        simple_lock(&uao_list_lock);
                    697:        LIST_REMOVE(aobj, u_list);
                    698:        simple_unlock(&uao_list_lock);
                    699:
                    700:        /*
1.27      chs       701:         * free all the pages that aren't PG_BUSY,
                    702:         * mark for release any that are.
1.5       mrg       703:         */
                    704:        busybody = FALSE;
1.27      chs       705:        for (pg = TAILQ_FIRST(&uobj->memq);
                    706:             pg != NULL;
                    707:             pg = TAILQ_NEXT(pg, listq)) {
1.5       mrg       708:                if (pg->flags & PG_BUSY) {
                    709:                        pg->flags |= PG_RELEASED;
                    710:                        busybody = TRUE;
                    711:                        continue;
                    712:                }
                    713:
                    714:                /* zap the mappings, free the swap slot, free the page */
1.26      chs       715:                pmap_page_protect(pg, VM_PROT_NONE);
1.18      chs       716:                uao_dropswap(&aobj->u_obj, pg->offset >> PAGE_SHIFT);
1.5       mrg       717:                uvm_lock_pageq();
                    718:                uvm_pagefree(pg);
                    719:                uvm_unlock_pageq();
                    720:        }
1.1       mrg       721:
1.5       mrg       722:        /*
                    723:         * if we found any busy pages, we're done for now.
                    724:         * mark the aobj for death, releasepg will finish up for us.
                    725:         */
                    726:        if (busybody) {
                    727:                aobj->u_flags |= UAO_FLAG_KILLME;
                    728:                simple_unlock(&aobj->u_obj.vmobjlock);
                    729:                return;
                    730:        }
1.1       mrg       731:
1.5       mrg       732:        /*
                    733:         * finally, free the rest.
                    734:         */
                    735:        uao_free(aobj);
                    736: }
1.1       mrg       737:
                    738: /*
1.22      thorpej   739:  * uao_flush: "flush" pages out of a uvm object
                    740:  *
                    741:  * => object should be locked by caller.  we may _unlock_ the object
                    742:  *     if (and only if) we need to clean a page (PGO_CLEANIT).
                    743:  *     XXXJRT Currently, however, we don't.  In the case of cleaning
                    744:  *     XXXJRT a page, we simply just deactivate it.  Should probably
                    745:  *     XXXJRT handle this better, in the future (although "flushing"
                    746:  *     XXXJRT anonymous memory isn't terribly important).
                    747:  * => if PGO_CLEANIT is not set, then we will neither unlock the object
                    748:  *     or block.
                    749:  * => if PGO_ALLPAGE is set, then all pages in the object are valid targets
                    750:  *     for flushing.
                    751:  * => NOTE: we rely on the fact that the object's memq is a TAILQ and
                    752:  *     that new pages are inserted on the tail end of the list.  thus,
                    753:  *     we can make a complete pass through the object in one go by starting
                    754:  *     at the head and working towards the tail (new pages are put in
                    755:  *     front of us).
                    756:  * => NOTE: we are allowed to lock the page queues, so the caller
                    757:  *     must not be holding the lock on them [e.g. pagedaemon had
                    758:  *     better not call us with the queues locked]
                    759:  * => we return TRUE unless we encountered some sort of I/O error
                    760:  *     XXXJRT currently never happens, as we never directly initiate
                    761:  *     XXXJRT I/O
                    762:  *
                    763:  * comment on "cleaning" object and PG_BUSY pages:
                    764:  *     this routine is holding the lock on the object.  the only time
                    765:  *     that is can run into a PG_BUSY page that it does not own is if
                    766:  *     some other process has started I/O on the page (e.g. either
                    767:  *     a pagein or a pageout).  if the PG_BUSY page is being paged
                    768:  *     in, then it can not be dirty (!PG_CLEAN) because no one has
                    769:  *     had a change to modify it yet.  if the PG_BUSY page is being
                    770:  *     paged out then it means that someone else has already started
                    771:  *     cleaning the page for us (how nice!).  in this case, if we
                    772:  *     have syncio specified, then after we make our pass through the
                    773:  *     object we need to wait for the other PG_BUSY pages to clear
                    774:  *     off (i.e. we need to do an iosync).  also note that once a
                    775:  *     page is PG_BUSY is must stary in its object until it is un-busyed.
                    776:  *     XXXJRT We never actually do this, as we are "flushing" anonymous
                    777:  *     XXXJRT memory, which doesn't have persistent backing store.
                    778:  *
                    779:  * note on page traversal:
                    780:  *     we can traverse the pages in an object either by going down the
                    781:  *     linked list in "uobj->memq", or we can go over the address range
                    782:  *     by page doing hash table lookups for each address.  depending
                    783:  *     on how many pages are in the object it may be cheaper to do one
                    784:  *     or the other.  we set "by_list" to true if we are using memq.
                    785:  *     if the cost of a hash lookup was equal to the cost of the list
                    786:  *     traversal we could compare the number of pages in the start->stop
                    787:  *     range to the total number of pages in the object.  however, it
                    788:  *     seems that a hash table lookup is more expensive than the linked
                    789:  *     list traversal, so we multiply the number of pages in the
                    790:  *     start->stop range by a penalty which we define below.
1.1       mrg       791:  */
1.22      thorpej   792:
                    793: #define        UAO_HASH_PENALTY 4      /* XXX: a guess */
                    794:
1.5       mrg       795: boolean_t
1.22      thorpej   796: uao_flush(uobj, start, stop, flags)
1.5       mrg       797:        struct uvm_object *uobj;
1.28      kleink    798:        voff_t start, stop;
1.5       mrg       799:        int flags;
                    800: {
1.22      thorpej   801:        struct uvm_aobj *aobj = (struct uvm_aobj *) uobj;
                    802:        struct vm_page *pp, *ppnext;
                    803:        boolean_t retval, by_list;
1.28      kleink    804:        voff_t curoff;
1.22      thorpej   805:        UVMHIST_FUNC("uao_flush"); UVMHIST_CALLED(maphist);
                    806:
                    807:        curoff = 0;     /* XXX: shut up gcc */
                    808:
                    809:        retval = TRUE;  /* default to success */
                    810:
                    811:        if (flags & PGO_ALLPAGES) {
                    812:                start = 0;
                    813:                stop = aobj->u_pages << PAGE_SHIFT;
                    814:                by_list = TRUE;         /* always go by the list */
                    815:        } else {
                    816:                start = trunc_page(start);
                    817:                stop = round_page(stop);
                    818:                if (stop > (aobj->u_pages << PAGE_SHIFT)) {
                    819:                        printf("uao_flush: strange, got an out of range "
                    820:                            "flush (fixed)\n");
                    821:                        stop = aobj->u_pages << PAGE_SHIFT;
                    822:                }
                    823:                by_list = (uobj->uo_npages <=
                    824:                    ((stop - start) >> PAGE_SHIFT) * UAO_HASH_PENALTY);
                    825:        }
                    826:
                    827:        UVMHIST_LOG(maphist,
                    828:            " flush start=0x%lx, stop=0x%x, by_list=%d, flags=0x%x",
                    829:            start, stop, by_list, flags);
1.1       mrg       830:
1.5       mrg       831:        /*
1.22      thorpej   832:         * Don't need to do any work here if we're not freeing
                    833:         * or deactivating pages.
                    834:         */
                    835:        if ((flags & (PGO_DEACTIVATE|PGO_FREE)) == 0) {
                    836:                UVMHIST_LOG(maphist,
                    837:                    "<- done (no work to do)",0,0,0,0);
                    838:                return (retval);
                    839:        }
                    840:
1.5       mrg       841:        /*
1.22      thorpej   842:         * now do it.  note: we must update ppnext in the body of loop or we
                    843:         * will get stuck.  we need to use ppnext because we may free "pp"
                    844:         * before doing the next loop.
1.21      thorpej   845:         */
1.22      thorpej   846:
                    847:        if (by_list) {
                    848:                pp = uobj->memq.tqh_first;
                    849:        } else {
                    850:                curoff = start;
                    851:                pp = uvm_pagelookup(uobj, curoff);
                    852:        }
                    853:
                    854:        ppnext = NULL;  /* XXX: shut up gcc */
                    855:        uvm_lock_pageq();       /* page queues locked */
                    856:
                    857:        /* locked: both page queues and uobj */
                    858:        for ( ; (by_list && pp != NULL) ||
                    859:            (!by_list && curoff < stop) ; pp = ppnext) {
                    860:                if (by_list) {
                    861:                        ppnext = pp->listq.tqe_next;
                    862:
                    863:                        /* range check */
                    864:                        if (pp->offset < start || pp->offset >= stop)
                    865:                                continue;
                    866:                } else {
                    867:                        curoff += PAGE_SIZE;
                    868:                        if (curoff < stop)
                    869:                                ppnext = uvm_pagelookup(uobj, curoff);
                    870:
                    871:                        /* null check */
                    872:                        if (pp == NULL)
                    873:                                continue;
                    874:                }
                    875:
                    876:                switch (flags & (PGO_CLEANIT|PGO_FREE|PGO_DEACTIVATE)) {
                    877:                /*
                    878:                 * XXX In these first 3 cases, we always just
                    879:                 * XXX deactivate the page.  We may want to
                    880:                 * XXX handle the different cases more specifically
                    881:                 * XXX in the future.
                    882:                 */
                    883:                case PGO_CLEANIT|PGO_FREE:
                    884:                case PGO_CLEANIT|PGO_DEACTIVATE:
                    885:                case PGO_DEACTIVATE:
1.25      thorpej   886:  deactivate_it:
1.22      thorpej   887:                        /* skip the page if it's loaned or wired */
                    888:                        if (pp->loan_count != 0 ||
                    889:                            pp->wire_count != 0)
                    890:                                continue;
                    891:
                    892:                        /* zap all mappings for the page. */
1.26      chs       893:                        pmap_page_protect(pp, VM_PROT_NONE);
1.22      thorpej   894:
                    895:                        /* ...and deactivate the page. */
                    896:                        uvm_pagedeactivate(pp);
                    897:
                    898:                        continue;
                    899:
                    900:                case PGO_FREE:
1.25      thorpej   901:                        /*
                    902:                         * If there are multiple references to
                    903:                         * the object, just deactivate the page.
                    904:                         */
                    905:                        if (uobj->uo_refs > 1)
                    906:                                goto deactivate_it;
                    907:
1.22      thorpej   908:                        /* XXX skip the page if it's loaned or wired */
                    909:                        if (pp->loan_count != 0 ||
                    910:                            pp->wire_count != 0)
                    911:                                continue;
                    912:
                    913:                        /*
                    914:                         * mark the page as released if its busy.
                    915:                         */
                    916:                        if (pp->flags & PG_BUSY) {
                    917:                                pp->flags |= PG_RELEASED;
                    918:                                continue;
                    919:                        }
                    920:
                    921:                        /* zap all mappings for the page. */
1.26      chs       922:                        pmap_page_protect(pp, VM_PROT_NONE);
1.22      thorpej   923:
                    924:                        uao_dropswap(uobj, pp->offset >> PAGE_SHIFT);
                    925:                        uvm_pagefree(pp);
                    926:
                    927:                        continue;
                    928:
                    929:                default:
                    930:                        panic("uao_flush: weird flags");
                    931:                }
                    932: #ifdef DIAGNOSTIC
                    933:                panic("uao_flush: unreachable code");
                    934: #endif
                    935:        }
                    936:
                    937:        uvm_unlock_pageq();
                    938:
                    939:        UVMHIST_LOG(maphist,
                    940:            "<- done, rv=%d",retval,0,0,0);
                    941:        return (retval);
1.1       mrg       942: }
                    943:
                    944: /*
                    945:  * uao_get: fetch me a page
                    946:  *
                    947:  * we have three cases:
                    948:  * 1: page is resident     -> just return the page.
                    949:  * 2: page is zero-fill    -> allocate a new page and zero it.
                    950:  * 3: page is swapped out  -> fetch the page from swap.
                    951:  *
                    952:  * cases 1 and 2 can be handled with PGO_LOCKED, case 3 cannot.
                    953:  * so, if the "center" page hits case 3 (or any page, with PGO_ALLPAGES),
                    954:  * then we will need to return VM_PAGER_UNLOCK.
                    955:  *
                    956:  * => prefer map unlocked (not required)
                    957:  * => object must be locked!  we will _unlock_ it before starting any I/O.
                    958:  * => flags: PGO_ALLPAGES: get all of the pages
                    959:  *           PGO_LOCKED: fault data structures are locked
                    960:  * => NOTE: offset is the offset of pps[0], _NOT_ pps[centeridx]
                    961:  * => NOTE: caller must check for released pages!!
                    962:  */
1.5       mrg       963: static int
                    964: uao_get(uobj, offset, pps, npagesp, centeridx, access_type, advice, flags)
                    965:        struct uvm_object *uobj;
1.28      kleink    966:        voff_t offset;
1.5       mrg       967:        struct vm_page **pps;
                    968:        int *npagesp;
                    969:        int centeridx, advice, flags;
                    970:        vm_prot_t access_type;
                    971: {
                    972:        struct uvm_aobj *aobj = (struct uvm_aobj *)uobj;
1.28      kleink    973:        voff_t current_offset;
1.5       mrg       974:        vm_page_t ptmp;
1.27      chs       975:        int lcv, gotpages, maxpages, swslot, rv, pageidx;
1.5       mrg       976:        boolean_t done;
                    977:        UVMHIST_FUNC("uao_get"); UVMHIST_CALLED(pdhist);
                    978:
1.27      chs       979:        UVMHIST_LOG(pdhist, "aobj=%p offset=%d, flags=%d",
                    980:                    aobj, offset, flags,0);
1.5       mrg       981:
                    982:        /*
                    983:         * get number of pages
                    984:         */
                    985:        maxpages = *npagesp;
                    986:
                    987:        /*
                    988:         * step 1: handled the case where fault data structures are locked.
                    989:         */
1.1       mrg       990:
1.5       mrg       991:        if (flags & PGO_LOCKED) {
                    992:                /*
                    993:                 * step 1a: get pages that are already resident.   only do
                    994:                 * this if the data structures are locked (i.e. the first
                    995:                 * time through).
                    996:                 */
                    997:
                    998:                done = TRUE;    /* be optimistic */
                    999:                gotpages = 0;   /* # of pages we got so far */
                   1000:
                   1001:                for (lcv = 0, current_offset = offset ; lcv < maxpages ;
                   1002:                    lcv++, current_offset += PAGE_SIZE) {
                   1003:                        /* do we care about this page?  if not, skip it */
                   1004:                        if (pps[lcv] == PGO_DONTCARE)
                   1005:                                continue;
                   1006:
                   1007:                        ptmp = uvm_pagelookup(uobj, current_offset);
                   1008:
                   1009:                        /*
1.30      thorpej  1010:                         * if page is new, attempt to allocate the page,
                   1011:                         * zero-fill'd.
1.5       mrg      1012:                         */
                   1013:                        if (ptmp == NULL && uao_find_swslot(aobj,
1.15      chs      1014:                            current_offset >> PAGE_SHIFT) == 0) {
1.5       mrg      1015:                                ptmp = uvm_pagealloc(uobj, current_offset,
1.30      thorpej  1016:                                    NULL, UVM_PGA_ZERO);
1.5       mrg      1017:                                if (ptmp) {
                   1018:                                        /* new page */
                   1019:                                        ptmp->flags &= ~(PG_BUSY|PG_FAKE);
                   1020:                                        ptmp->pqflags |= PQ_AOBJ;
                   1021:                                        UVM_PAGE_OWN(ptmp, NULL);
                   1022:                                }
                   1023:                        }
                   1024:
                   1025:                        /*
                   1026:                         * to be useful must get a non-busy, non-released page
                   1027:                         */
                   1028:                        if (ptmp == NULL ||
                   1029:                            (ptmp->flags & (PG_BUSY|PG_RELEASED)) != 0) {
                   1030:                                if (lcv == centeridx ||
                   1031:                                    (flags & PGO_ALLPAGES) != 0)
                   1032:                                        /* need to do a wait or I/O! */
                   1033:                                        done = FALSE;
                   1034:                                        continue;
                   1035:                        }
                   1036:
                   1037:                        /*
                   1038:                         * useful page: busy/lock it and plug it in our
                   1039:                         * result array
                   1040:                         */
                   1041:                        /* caller must un-busy this page */
                   1042:                        ptmp->flags |= PG_BUSY;
                   1043:                        UVM_PAGE_OWN(ptmp, "uao_get1");
                   1044:                        pps[lcv] = ptmp;
                   1045:                        gotpages++;
                   1046:
                   1047:                }       /* "for" lcv loop */
                   1048:
                   1049:                /*
                   1050:                 * step 1b: now we've either done everything needed or we
                   1051:                 * to unlock and do some waiting or I/O.
                   1052:                 */
                   1053:
                   1054:                UVMHIST_LOG(pdhist, "<- done (done=%d)", done, 0,0,0);
                   1055:
                   1056:                *npagesp = gotpages;
                   1057:                if (done)
                   1058:                        /* bingo! */
                   1059:                        return(VM_PAGER_OK);
                   1060:                else
                   1061:                        /* EEK!   Need to unlock and I/O */
                   1062:                        return(VM_PAGER_UNLOCK);
1.1       mrg      1063:        }
                   1064:
1.5       mrg      1065:        /*
                   1066:         * step 2: get non-resident or busy pages.
                   1067:         * object is locked.   data structures are unlocked.
                   1068:         */
                   1069:
                   1070:        for (lcv = 0, current_offset = offset ; lcv < maxpages ;
                   1071:            lcv++, current_offset += PAGE_SIZE) {
1.27      chs      1072:
1.5       mrg      1073:                /*
                   1074:                 * - skip over pages we've already gotten or don't want
                   1075:                 * - skip over pages we don't _have_ to get
                   1076:                 */
1.27      chs      1077:
1.5       mrg      1078:                if (pps[lcv] != NULL ||
                   1079:                    (lcv != centeridx && (flags & PGO_ALLPAGES) == 0))
                   1080:                        continue;
                   1081:
1.27      chs      1082:                pageidx = current_offset >> PAGE_SHIFT;
                   1083:
1.5       mrg      1084:                /*
                   1085:                 * we have yet to locate the current page (pps[lcv]).   we
                   1086:                 * first look for a page that is already at the current offset.
                   1087:                 * if we find a page, we check to see if it is busy or
                   1088:                 * released.  if that is the case, then we sleep on the page
                   1089:                 * until it is no longer busy or released and repeat the lookup.
                   1090:                 * if the page we found is neither busy nor released, then we
                   1091:                 * busy it (so we own it) and plug it into pps[lcv].   this
                   1092:                 * 'break's the following while loop and indicates we are
                   1093:                 * ready to move on to the next page in the "lcv" loop above.
                   1094:                 *
                   1095:                 * if we exit the while loop with pps[lcv] still set to NULL,
                   1096:                 * then it means that we allocated a new busy/fake/clean page
                   1097:                 * ptmp in the object and we need to do I/O to fill in the data.
                   1098:                 */
                   1099:
                   1100:                /* top of "pps" while loop */
                   1101:                while (pps[lcv] == NULL) {
                   1102:                        /* look for a resident page */
                   1103:                        ptmp = uvm_pagelookup(uobj, current_offset);
                   1104:
                   1105:                        /* not resident?   allocate one now (if we can) */
                   1106:                        if (ptmp == NULL) {
                   1107:
                   1108:                                ptmp = uvm_pagealloc(uobj, current_offset,
1.19      chs      1109:                                    NULL, 0);
1.5       mrg      1110:
                   1111:                                /* out of RAM? */
                   1112:                                if (ptmp == NULL) {
                   1113:                                        simple_unlock(&uobj->vmobjlock);
                   1114:                                        UVMHIST_LOG(pdhist,
                   1115:                                            "sleeping, ptmp == NULL\n",0,0,0,0);
                   1116:                                        uvm_wait("uao_getpage");
                   1117:                                        simple_lock(&uobj->vmobjlock);
                   1118:                                        /* goto top of pps while loop */
                   1119:                                        continue;
                   1120:                                }
                   1121:
                   1122:                                /*
                   1123:                                 * safe with PQ's unlocked: because we just
                   1124:                                 * alloc'd the page
                   1125:                                 */
                   1126:                                ptmp->pqflags |= PQ_AOBJ;
                   1127:
                   1128:                                /*
                   1129:                                 * got new page ready for I/O.  break pps while
                   1130:                                 * loop.  pps[lcv] is still NULL.
                   1131:                                 */
                   1132:                                break;
                   1133:                        }
                   1134:
                   1135:                        /* page is there, see if we need to wait on it */
                   1136:                        if ((ptmp->flags & (PG_BUSY|PG_RELEASED)) != 0) {
                   1137:                                ptmp->flags |= PG_WANTED;
                   1138:                                UVMHIST_LOG(pdhist,
                   1139:                                    "sleeping, ptmp->flags 0x%x\n",
                   1140:                                    ptmp->flags,0,0,0);
1.23      thorpej  1141:                                UVM_UNLOCK_AND_WAIT(ptmp, &uobj->vmobjlock,
                   1142:                                    FALSE, "uao_get", 0);
1.5       mrg      1143:                                simple_lock(&uobj->vmobjlock);
                   1144:                                continue;       /* goto top of pps while loop */
                   1145:                        }
                   1146:
                   1147:                        /*
                   1148:                         * if we get here then the page has become resident and
                   1149:                         * unbusy between steps 1 and 2.  we busy it now (so we
                   1150:                         * own it) and set pps[lcv] (so that we exit the while
                   1151:                         * loop).
                   1152:                         */
                   1153:                        /* we own it, caller must un-busy */
                   1154:                        ptmp->flags |= PG_BUSY;
                   1155:                        UVM_PAGE_OWN(ptmp, "uao_get2");
                   1156:                        pps[lcv] = ptmp;
                   1157:                }
                   1158:
                   1159:                /*
                   1160:                 * if we own the valid page at the correct offset, pps[lcv] will
                   1161:                 * point to it.   nothing more to do except go to the next page.
                   1162:                 */
                   1163:                if (pps[lcv])
                   1164:                        continue;                       /* next lcv */
                   1165:
                   1166:                /*
                   1167:                 * we have a "fake/busy/clean" page that we just allocated.
                   1168:                 * do the needed "i/o", either reading from swap or zeroing.
                   1169:                 */
1.27      chs      1170:                swslot = uao_find_swslot(aobj, pageidx);
1.5       mrg      1171:
                   1172:                /*
                   1173:                 * just zero the page if there's nothing in swap.
                   1174:                 */
                   1175:                if (swslot == 0)
                   1176:                {
                   1177:                        /*
                   1178:                         * page hasn't existed before, just zero it.
                   1179:                         */
                   1180:                        uvm_pagezero(ptmp);
1.27      chs      1181:                } else {
1.5       mrg      1182:                        UVMHIST_LOG(pdhist, "pagein from swslot %d",
                   1183:                             swslot, 0,0,0);
                   1184:
                   1185:                        /*
                   1186:                         * page in the swapped-out page.
                   1187:                         * unlock object for i/o, relock when done.
                   1188:                         */
                   1189:                        simple_unlock(&uobj->vmobjlock);
                   1190:                        rv = uvm_swap_get(ptmp, swslot, PGO_SYNCIO);
                   1191:                        simple_lock(&uobj->vmobjlock);
                   1192:
                   1193:                        /*
                   1194:                         * I/O done.  check for errors.
                   1195:                         */
                   1196:                        if (rv != VM_PAGER_OK)
                   1197:                        {
                   1198:                                UVMHIST_LOG(pdhist, "<- done (error=%d)",
                   1199:                                    rv,0,0,0);
                   1200:                                if (ptmp->flags & PG_WANTED)
1.24      thorpej  1201:                                        wakeup(ptmp);
1.27      chs      1202:
                   1203:                                /*
                   1204:                                 * remove the swap slot from the aobj
                   1205:                                 * and mark the aobj as having no real slot.
                   1206:                                 * don't free the swap slot, thus preventing
                   1207:                                 * it from being used again.
                   1208:                                 */
                   1209:                                swslot = uao_set_swslot(&aobj->u_obj, pageidx,
                   1210:                                                        SWSLOT_BAD);
                   1211:                                uvm_swap_markbad(swslot, 1);
                   1212:
1.5       mrg      1213:                                ptmp->flags &= ~(PG_WANTED|PG_BUSY);
                   1214:                                UVM_PAGE_OWN(ptmp, NULL);
                   1215:                                uvm_lock_pageq();
                   1216:                                uvm_pagefree(ptmp);
                   1217:                                uvm_unlock_pageq();
1.27      chs      1218:
1.5       mrg      1219:                                simple_unlock(&uobj->vmobjlock);
                   1220:                                return (rv);
                   1221:                        }
                   1222:                }
                   1223:
                   1224:                /*
                   1225:                 * we got the page!   clear the fake flag (indicates valid
                   1226:                 * data now in page) and plug into our result array.   note
                   1227:                 * that page is still busy.
                   1228:                 *
                   1229:                 * it is the callers job to:
                   1230:                 * => check if the page is released
                   1231:                 * => unbusy the page
                   1232:                 * => activate the page
                   1233:                 */
                   1234:
                   1235:                ptmp->flags &= ~PG_FAKE;                /* data is valid ... */
1.26      chs      1236:                pmap_clear_modify(ptmp);                /* ... and clean */
1.5       mrg      1237:                pps[lcv] = ptmp;
1.1       mrg      1238:
1.5       mrg      1239:        }       /* lcv loop */
1.1       mrg      1240:
                   1241:        /*
1.5       mrg      1242:         * finally, unlock object and return.
                   1243:         */
1.1       mrg      1244:
                   1245:        simple_unlock(&uobj->vmobjlock);
1.5       mrg      1246:        UVMHIST_LOG(pdhist, "<- done (OK)",0,0,0,0);
                   1247:        return(VM_PAGER_OK);
1.1       mrg      1248: }
                   1249:
                   1250: /*
                   1251:  * uao_releasepg: handle released page in an aobj
                   1252:  *
                   1253:  * => "pg" is a PG_BUSY [caller owns it], PG_RELEASED page that we need
                   1254:  *      to dispose of.
                   1255:  * => caller must handle PG_WANTED case
                   1256:  * => called with page's object locked, pageq's unlocked
                   1257:  * => returns TRUE if page's object is still alive, FALSE if we
                   1258:  *      killed the page's object.    if we return TRUE, then we
                   1259:  *      return with the object locked.
                   1260:  * => if (nextpgp != NULL) => we return pageq.tqe_next here, and return
                   1261:  *                              with the page queues locked [for pagedaemon]
                   1262:  * => if (nextpgp == NULL) => we return with page queues unlocked [normal case]
                   1263:  * => we kill the aobj if it is not referenced and we are suppose to
                   1264:  *      kill it ("KILLME").
                   1265:  */
1.27      chs      1266: static boolean_t
                   1267: uao_releasepg(pg, nextpgp)
1.5       mrg      1268:        struct vm_page *pg;
                   1269:        struct vm_page **nextpgp;       /* OUT */
1.1       mrg      1270: {
1.5       mrg      1271:        struct uvm_aobj *aobj = (struct uvm_aobj *) pg->uobject;
1.1       mrg      1272:
                   1273: #ifdef DIAGNOSTIC
1.5       mrg      1274:        if ((pg->flags & PG_RELEASED) == 0)
                   1275:                panic("uao_releasepg: page not released!");
1.1       mrg      1276: #endif
1.27      chs      1277:
1.5       mrg      1278:        /*
                   1279:         * dispose of the page [caller handles PG_WANTED] and swap slot.
                   1280:         */
1.26      chs      1281:        pmap_page_protect(pg, VM_PROT_NONE);
1.18      chs      1282:        uao_dropswap(&aobj->u_obj, pg->offset >> PAGE_SHIFT);
1.5       mrg      1283:        uvm_lock_pageq();
                   1284:        if (nextpgp)
                   1285:                *nextpgp = pg->pageq.tqe_next;  /* next page for daemon */
                   1286:        uvm_pagefree(pg);
                   1287:        if (!nextpgp)
1.27      chs      1288:                uvm_unlock_pageq();             /* keep locked for daemon */
1.5       mrg      1289:
                   1290:        /*
                   1291:         * if we're not killing the object, we're done.
                   1292:         */
                   1293:        if ((aobj->u_flags & UAO_FLAG_KILLME) == 0)
                   1294:                return TRUE;
1.1       mrg      1295:
                   1296: #ifdef DIAGNOSTIC
1.5       mrg      1297:        if (aobj->u_obj.uo_refs)
                   1298:                panic("uvm_km_releasepg: kill flag set on referenced object!");
1.1       mrg      1299: #endif
                   1300:
1.5       mrg      1301:        /*
                   1302:         * if there are still pages in the object, we're done for now.
                   1303:         */
                   1304:        if (aobj->u_obj.uo_npages != 0)
                   1305:                return TRUE;
1.1       mrg      1306:
                   1307: #ifdef DIAGNOSTIC
1.27      chs      1308:        if (TAILQ_FIRST(&aobj->u_obj.memq))
1.5       mrg      1309:                panic("uvn_releasepg: pages in object with npages == 0");
1.1       mrg      1310: #endif
                   1311:
1.5       mrg      1312:        /*
                   1313:         * finally, free the rest.
                   1314:         */
                   1315:        uao_free(aobj);
1.1       mrg      1316:
1.5       mrg      1317:        return FALSE;
1.18      chs      1318: }
                   1319:
1.27      chs      1320:
1.18      chs      1321: /*
                   1322:  * uao_dropswap:  release any swap resources from this aobj page.
                   1323:  *
                   1324:  * => aobj must be locked or have a reference count of 0.
                   1325:  */
                   1326:
                   1327: void
                   1328: uao_dropswap(uobj, pageidx)
                   1329:        struct uvm_object *uobj;
                   1330:        int pageidx;
                   1331: {
                   1332:        int slot;
                   1333:
                   1334:        slot = uao_set_swslot(uobj, pageidx, 0);
                   1335:        if (slot) {
                   1336:                uvm_swap_free(slot, 1);
                   1337:        }
1.27      chs      1338: }
                   1339:
                   1340:
                   1341: /*
                   1342:  * page in every page in every aobj that is paged-out to a range of swslots.
                   1343:  *
                   1344:  * => nothing should be locked.
                   1345:  * => returns TRUE if pagein was aborted due to lack of memory.
                   1346:  */
                   1347: boolean_t
                   1348: uao_swap_off(startslot, endslot)
                   1349:        int startslot, endslot;
                   1350: {
                   1351:        struct uvm_aobj *aobj, *nextaobj;
                   1352:
                   1353:        /*
                   1354:         * walk the list of all aobjs.
                   1355:         */
                   1356:
                   1357: restart:
                   1358:        simple_lock(&uao_list_lock);
                   1359:
                   1360:        for (aobj = LIST_FIRST(&uao_list);
                   1361:             aobj != NULL;
                   1362:             aobj = nextaobj) {
                   1363:                boolean_t rv;
                   1364:
                   1365:                /*
                   1366:                 * try to get the object lock,
                   1367:                 * start all over if we fail.
                   1368:                 * most of the time we'll get the aobj lock,
                   1369:                 * so this should be a rare case.
                   1370:                 */
                   1371:                if (!simple_lock_try(&aobj->u_obj.vmobjlock)) {
                   1372:                        simple_unlock(&uao_list_lock);
                   1373:                        goto restart;
                   1374:                }
                   1375:
                   1376:                /*
                   1377:                 * add a ref to the aobj so it doesn't disappear
                   1378:                 * while we're working.
                   1379:                 */
                   1380:                uao_reference_locked(&aobj->u_obj);
                   1381:
                   1382:                /*
                   1383:                 * now it's safe to unlock the uao list.
                   1384:                 */
                   1385:                simple_unlock(&uao_list_lock);
                   1386:
                   1387:                /*
                   1388:                 * page in any pages in the swslot range.
                   1389:                 * if there's an error, abort and return the error.
                   1390:                 */
                   1391:                rv = uao_pagein(aobj, startslot, endslot);
                   1392:                if (rv) {
                   1393:                        uao_detach_locked(&aobj->u_obj);
                   1394:                        return rv;
                   1395:                }
                   1396:
                   1397:                /*
                   1398:                 * we're done with this aobj.
                   1399:                 * relock the list and drop our ref on the aobj.
                   1400:                 */
                   1401:                simple_lock(&uao_list_lock);
                   1402:                nextaobj = LIST_NEXT(aobj, u_list);
                   1403:                uao_detach_locked(&aobj->u_obj);
                   1404:        }
                   1405:
                   1406:        /*
                   1407:         * done with traversal, unlock the list
                   1408:         */
                   1409:        simple_unlock(&uao_list_lock);
                   1410:        return FALSE;
                   1411: }
                   1412:
                   1413:
                   1414: /*
                   1415:  * page in any pages from aobj in the given range.
                   1416:  *
                   1417:  * => aobj must be locked and is returned locked.
                   1418:  * => returns TRUE if pagein was aborted due to lack of memory.
                   1419:  */
                   1420: static boolean_t
                   1421: uao_pagein(aobj, startslot, endslot)
                   1422:        struct uvm_aobj *aobj;
                   1423:        int startslot, endslot;
                   1424: {
                   1425:        boolean_t rv;
                   1426:
                   1427:        if (UAO_USES_SWHASH(aobj)) {
                   1428:                struct uao_swhash_elt *elt;
                   1429:                int bucket;
                   1430:
                   1431: restart:
                   1432:                for (bucket = aobj->u_swhashmask; bucket >= 0; bucket--) {
                   1433:                        for (elt = LIST_FIRST(&aobj->u_swhash[bucket]);
                   1434:                             elt != NULL;
                   1435:                             elt = LIST_NEXT(elt, list)) {
                   1436:                                int i;
                   1437:
                   1438:                                for (i = 0; i < UAO_SWHASH_CLUSTER_SIZE; i++) {
                   1439:                                        int slot = elt->slots[i];
                   1440:
                   1441:                                        /*
                   1442:                                         * if the slot isn't in range, skip it.
                   1443:                                         */
                   1444:                                        if (slot < startslot ||
                   1445:                                            slot >= endslot) {
                   1446:                                                continue;
                   1447:                                        }
                   1448:
                   1449:                                        /*
                   1450:                                         * process the page,
                   1451:                                         * the start over on this object
                   1452:                                         * since the swhash elt
                   1453:                                         * may have been freed.
                   1454:                                         */
                   1455:                                        rv = uao_pagein_page(aobj,
                   1456:                                          UAO_SWHASH_ELT_PAGEIDX_BASE(elt) + i);
                   1457:                                        if (rv) {
                   1458:                                                return rv;
                   1459:                                        }
                   1460:                                        goto restart;
                   1461:                                }
                   1462:                        }
                   1463:                }
                   1464:        } else {
                   1465:                int i;
                   1466:
                   1467:                for (i = 0; i < aobj->u_pages; i++) {
                   1468:                        int slot = aobj->u_swslots[i];
                   1469:
                   1470:                        /*
                   1471:                         * if the slot isn't in range, skip it
                   1472:                         */
                   1473:                        if (slot < startslot || slot >= endslot) {
                   1474:                                continue;
                   1475:                        }
                   1476:
                   1477:                        /*
                   1478:                         * process the page.
                   1479:                         */
                   1480:                        rv = uao_pagein_page(aobj, i);
                   1481:                        if (rv) {
                   1482:                                return rv;
                   1483:                        }
                   1484:                }
                   1485:        }
                   1486:
                   1487:        return FALSE;
                   1488: }
                   1489:
                   1490: /*
                   1491:  * page in a page from an aobj.  used for swap_off.
                   1492:  * returns TRUE if pagein was aborted due to lack of memory.
                   1493:  *
                   1494:  * => aobj must be locked and is returned locked.
                   1495:  */
                   1496: static boolean_t
                   1497: uao_pagein_page(aobj, pageidx)
                   1498:        struct uvm_aobj *aobj;
                   1499:        int pageidx;
                   1500: {
                   1501:        struct vm_page *pg;
                   1502:        int rv, slot, npages;
                   1503:        UVMHIST_FUNC("uao_pagein_page");  UVMHIST_CALLED(pdhist);
                   1504:
                   1505:        pg = NULL;
                   1506:        npages = 1;
                   1507:        /* locked: aobj */
                   1508:        rv = uao_get(&aobj->u_obj, pageidx << PAGE_SHIFT,
                   1509:                     &pg, &npages, 0, VM_PROT_READ|VM_PROT_WRITE, 0, 0);
                   1510:        /* unlocked: aobj */
                   1511:
                   1512:        /*
                   1513:         * relock and finish up.
                   1514:         */
                   1515:        simple_lock(&aobj->u_obj.vmobjlock);
                   1516:
                   1517:        switch (rv) {
                   1518:        case VM_PAGER_OK:
                   1519:                break;
                   1520:
                   1521:        case VM_PAGER_ERROR:
                   1522:        case VM_PAGER_REFAULT:
                   1523:                /*
                   1524:                 * nothing more to do on errors.
                   1525:                 * VM_PAGER_REFAULT can only mean that the anon was freed,
                   1526:                 * so again there's nothing to do.
                   1527:                 */
                   1528:                return FALSE;
                   1529:
                   1530: #ifdef DIAGNOSTIC
                   1531:        default:
                   1532:                panic("uao_pagein_page: uao_get -> %d\n", rv);
                   1533: #endif
                   1534:        }
                   1535:
                   1536: #ifdef DIAGNOSTIC
                   1537:        /*
                   1538:         * this should never happen, since we have a reference on the aobj.
                   1539:         */
                   1540:        if (pg->flags & PG_RELEASED) {
                   1541:                panic("uao_pagein_page: found PG_RELEASED page?\n");
                   1542:        }
                   1543: #endif
                   1544:
                   1545:        /*
                   1546:         * ok, we've got the page now.
                   1547:         * mark it as dirty, clear its swslot and un-busy it.
                   1548:         */
                   1549:        slot = uao_set_swslot(&aobj->u_obj, pageidx, 0);
                   1550:        uvm_swap_free(slot, 1);
                   1551:        pg->flags &= ~(PG_BUSY|PG_CLEAN|PG_FAKE);
                   1552:        UVM_PAGE_OWN(pg, NULL);
                   1553:
                   1554:        /*
                   1555:         * deactivate the page (to put it on a page queue).
                   1556:         */
                   1557:        pmap_clear_reference(pg);
                   1558:        pmap_page_protect(pg, VM_PROT_NONE);
                   1559:        uvm_lock_pageq();
                   1560:        uvm_pagedeactivate(pg);
                   1561:        uvm_unlock_pageq();
                   1562:
                   1563:        return FALSE;
1.1       mrg      1564: }

CVSweb <webmaster@jp.NetBSD.org>