Annotation of OSKit-Mach/vm/vm_pageout.c, revision 1.1.1.1

1.1       root        1: /*
                      2:  * Mach Operating System
                      3:  * Copyright (c) 1991,1990,1989,1988,1987 Carnegie Mellon University.
                      4:  * Copyright (c) 1993,1994 The University of Utah and
                      5:  * the Computer Systems Laboratory (CSL).
                      6:  * All rights reserved.
                      7:  *
                      8:  * Permission to use, copy, modify and distribute this software and its
                      9:  * documentation is hereby granted, provided that both the copyright
                     10:  * notice and this permission notice appear in all copies of the
                     11:  * software, derivative works or modified versions, and any portions
                     12:  * thereof, and that both notices appear in supporting documentation.
                     13:  *
                     14:  * CARNEGIE MELLON, THE UNIVERSITY OF UTAH AND CSL ALLOW FREE USE OF
                     15:  * THIS SOFTWARE IN ITS "AS IS" CONDITION, AND DISCLAIM ANY LIABILITY
                     16:  * OF ANY KIND FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF
                     17:  * THIS SOFTWARE.
                     18:  *
                     19:  * Carnegie Mellon requests users of this software to return to
                     20:  *
                     21:  *  Software Distribution Coordinator  or  [email protected]
                     22:  *  School of Computer Science
                     23:  *  Carnegie Mellon University
                     24:  *  Pittsburgh PA 15213-3890
                     25:  *
                     26:  * any improvements or extensions that they make and grant Carnegie Mellon
                     27:  * the rights to redistribute these changes.
                     28:  */
                     29: /*
                     30:  *     File:   vm/vm_pageout.c
                     31:  *     Author: Avadis Tevanian, Jr., Michael Wayne Young
                     32:  *     Date:   1985
                     33:  *
                     34:  *     The proverbial page-out daemon.
                     35:  */
                     36: 
                     37: #include <mach_pagemap.h>
                     38: #include <norma_vm.h>
                     39: 
                     40: #include <mach/mach_types.h>
                     41: #include <mach/memory_object.h>
                     42: #include "memory_object_default.h"
                     43: #include "memory_object_user.h"
                     44: #include <mach/vm_param.h>
                     45: #include <mach/vm_statistics.h>
                     46: #include <kern/counters.h>
                     47: #include <kern/thread.h>
                     48: #include <vm/pmap.h>
                     49: #include <vm/vm_map.h>
                     50: #include <vm/vm_object.h>
                     51: #include <vm/vm_page.h>
                     52: #include <vm/vm_pageout.h>
                     53: #include <machine/vm_tuning.h>
                     54: 
                     55: 
                     56: 
                     57: #ifndef        VM_PAGEOUT_BURST_MAX
                     58: #define        VM_PAGEOUT_BURST_MAX    10              /* number of pages */
                     59: #endif /* VM_PAGEOUT_BURST_MAX */
                     60: 
                     61: #ifndef        VM_PAGEOUT_BURST_MIN
                     62: #define        VM_PAGEOUT_BURST_MIN    5               /* number of pages */
                     63: #endif /* VM_PAGEOUT_BURST_MIN */
                     64: 
                     65: #ifndef        VM_PAGEOUT_BURST_WAIT
                     66: #define        VM_PAGEOUT_BURST_WAIT   10              /* milliseconds per page */
                     67: #endif /* VM_PAGEOUT_BURST_WAIT */
                     68: 
                     69: #ifndef        VM_PAGEOUT_EMPTY_WAIT
                     70: #define VM_PAGEOUT_EMPTY_WAIT  75              /* milliseconds */
                     71: #endif /* VM_PAGEOUT_EMPTY_WAIT */
                     72: 
                     73: #ifndef        VM_PAGEOUT_PAUSE_MAX
                     74: #define        VM_PAGEOUT_PAUSE_MAX    10              /* number of pauses */
                     75: #endif /* VM_PAGEOUT_PAUSE_MAX */
                     76: 
                     77: /*
                     78:  *     To obtain a reasonable LRU approximation, the inactive queue
                     79:  *     needs to be large enough to give pages on it a chance to be
                     80:  *     referenced a second time.  This macro defines the fraction
                     81:  *     of active+inactive pages that should be inactive.
                     82:  *     The pageout daemon uses it to update vm_page_inactive_target.
                     83:  *
                     84:  *     If vm_page_free_count falls below vm_page_free_target and
                     85:  *     vm_page_inactive_count is below vm_page_inactive_target,
                     86:  *     then the pageout daemon starts running.
                     87:  */
                     88: 
                     89: #ifndef        VM_PAGE_INACTIVE_TARGET
                     90: #define        VM_PAGE_INACTIVE_TARGET(avail)  ((avail) * 2 / 3)
                     91: #endif /* VM_PAGE_INACTIVE_TARGET */
                     92: 
                     93: /*
                     94:  *     Once the pageout daemon starts running, it keeps going
                     95:  *     until vm_page_free_count meets or exceeds vm_page_free_target.
                     96:  */
                     97: 
                     98: #ifndef        VM_PAGE_FREE_TARGET
                     99: #define        VM_PAGE_FREE_TARGET(free)       (15 + (free) / 80)
                    100: #endif /* VM_PAGE_FREE_TARGET */
                    101: 
                    102: /*
                    103:  *     The pageout daemon always starts running once vm_page_free_count
                    104:  *     falls below vm_page_free_min.
                    105:  */
                    106: 
                    107: #ifndef        VM_PAGE_FREE_MIN
                    108: #define        VM_PAGE_FREE_MIN(free)  (10 + (free) / 100)
                    109: #endif /* VM_PAGE_FREE_MIN */
                    110: 
                    111: /*      When vm_page_external_count exceeds vm_page_external_limit,
                    112:  *     allocations of externally paged pages stops.
                    113:  */
                    114: 
                    115: #ifndef VM_PAGE_EXTERNAL_LIMIT
                    116: #define VM_PAGE_EXTERNAL_LIMIT(free)           ((free) / 2)
                    117: #endif  /* VM_PAGE_EXTERNAL_LIMIT */
                    118: 
                    119: /*     Attempt to keep the number of externally paged pages less
                    120:  *     than vm_pages_external_target.
                    121:  */
                    122: #ifndef VM_PAGE_EXTERNAL_TARGET
                    123: #define VM_PAGE_EXTERNAL_TARGET(free)          ((free) / 4)
                    124: #endif  /* VM_PAGE_EXTERNAL_TARGET */
                    125: 
                    126: /*
                    127:  *     When vm_page_free_count falls below vm_page_free_reserved,
                    128:  *     only vm-privileged threads can allocate pages.  vm-privilege
                    129:  *     allows the pageout daemon and default pager (and any other
                    130:  *     associated threads needed for default pageout) to continue
                    131:  *     operation by dipping into the reserved pool of pages.  */
                    132: 
                    133: #ifndef        VM_PAGE_FREE_RESERVED
                    134: #define        VM_PAGE_FREE_RESERVED                   50
                    135: #endif /* VM_PAGE_FREE_RESERVED */
                    136: 
                    137: /*
                    138:  *     When vm_page_free_count falls below vm_pageout_reserved_internal,
                    139:  *     the pageout daemon no longer trusts external pagers to clean pages.
                    140:  *     External pagers are probably all wedged waiting for a free page.
                    141:  *     It forcibly double-pages dirty pages belonging to external objects,
                    142:  *     getting the pages to the default pager to clean.
                    143:  */
                    144: 
                    145: #ifndef        VM_PAGEOUT_RESERVED_INTERNAL
                    146: #define        VM_PAGEOUT_RESERVED_INTERNAL(reserve)   ((reserve) - 25)
                    147: #endif /* VM_PAGEOUT_RESERVED_INTERNAL */
                    148: 
                    149: /*
                    150:  *     When vm_page_free_count falls below vm_pageout_reserved_really,
                    151:  *     the pageout daemon stops work entirely to let the default pager
                    152:  *     catch up (assuming the default pager has pages to clean).
                    153:  *     Beyond this point, it is too dangerous to consume memory
                    154:  *     even for memory_object_data_write messages to the default pager.
                    155:  */
                    156: 
                    157: #ifndef        VM_PAGEOUT_RESERVED_REALLY
                    158: #define        VM_PAGEOUT_RESERVED_REALLY(reserve)     ((reserve) - 40)
                    159: #endif /* VM_PAGEOUT_RESERVED_REALLY */
                    160: 
                    161: extern void vm_pageout_continue();
                    162: extern void vm_pageout_scan_continue();
                    163: 
                    164: unsigned int vm_pageout_reserved_internal = 0;
                    165: unsigned int vm_pageout_reserved_really = 0;
                    166: 
                    167: unsigned int vm_page_external_target = 0;
                    168: 
                    169: unsigned int vm_pageout_burst_max = 0;
                    170: unsigned int vm_pageout_burst_min = 0;
                    171: unsigned int vm_pageout_burst_wait = 0;                /* milliseconds per page */
                    172: unsigned int vm_pageout_empty_wait = 0;                /* milliseconds */
                    173: unsigned int vm_pageout_pause_count = 0;
                    174: unsigned int vm_pageout_pause_max = 0;
                    175: 
                    176: /*
                    177:  *     These variables record the pageout daemon's actions:
                    178:  *     how many pages it looks at and what happens to those pages.
                    179:  *     No locking needed because only one thread modifies the variables.
                    180:  */
                    181: 
                    182: unsigned int vm_pageout_active = 0;            /* debugging */
                    183: unsigned int vm_pageout_inactive = 0;          /* debugging */
                    184: unsigned int vm_pageout_inactive_nolock = 0;   /* debugging */
                    185: unsigned int vm_pageout_inactive_busy = 0;     /* debugging */
                    186: unsigned int vm_pageout_inactive_absent = 0;   /* debugging */
                    187: unsigned int vm_pageout_inactive_used = 0;     /* debugging */
                    188: unsigned int vm_pageout_inactive_clean = 0;    /* debugging */
                    189: unsigned int vm_pageout_inactive_dirty = 0;    /* debugging */
                    190: unsigned int vm_pageout_inactive_double = 0;   /* debugging */
                    191: unsigned int vm_pageout_inactive_cleaned_external = 0;
                    192: 
                    193: #if    NORMA_VM
                    194: /*
                    195:  * Define them here, since they won't be defined by memory_object_user.h.
                    196:  */
                    197: extern kern_return_t memory_object_data_initialize();
                    198: extern kern_return_t memory_object_data_write();
                    199: #endif /* NORMA_VM */
                    200: 
                    201: /*
                    202:  *     Routine:        vm_pageout_setup
                    203:  *     Purpose:
                    204:  *             Set up a page for pageout.
                    205:  *
                    206:  *             Move or copy the page to a new object, as part
                    207:  *             of which it will be sent to its memory manager
                    208:  *             in a memory_object_data_write or memory_object_initialize
                    209:  *             message.
                    210:  *
                    211:  *             The "paging_offset" argument specifies the offset
                    212:  *             of the page within its external memory object.
                    213:  *
                    214:  *             The "new_object" and "new_offset" arguments
                    215:  *             indicate where the page should be moved.
                    216:  *
                    217:  *             The "flush" argument specifies whether the page
                    218:  *             should be flushed from its object.  If not, a
                    219:  *             copy of the page is moved to the new object.
                    220:  *
                    221:  *     In/Out conditions:
                    222:  *             The page in question must not be on any pageout queues,
                    223:  *             and must be busy.  The object to which it belongs
                    224:  *             must be unlocked, and the caller must hold a paging
                    225:  *             reference to it.  The new_object must not be locked.
                    226:  *
                    227:  *             If the page is flushed from its original object,
                    228:  *             this routine returns a pointer to a place-holder page,
                    229:  *             inserted at the same offset, to block out-of-order
                    230:  *             requests for the page.  The place-holder page must
                    231:  *             be freed after the data_write or initialize message
                    232:  *             has been sent.  If the page is copied,
                    233:  *             the holding page is VM_PAGE_NULL.
                    234:  *
                    235:  *             The original page is put on a paging queue and marked
                    236:  *             not busy on exit.
                    237:  */
                    238: vm_page_t
                    239: vm_pageout_setup(m, paging_offset, new_object, new_offset, flush)
                    240:        register vm_page_t      m;
                    241:        vm_offset_t             paging_offset;
                    242:        register vm_object_t    new_object;
                    243:        vm_offset_t             new_offset;
                    244:        boolean_t               flush;
                    245: {
                    246:        register vm_object_t    old_object = m->object;
                    247:        register vm_page_t      holding_page = 0; /*'=0'to quiet gcc warnings*/
                    248:        register vm_page_t      new_m;
                    249: 
                    250:        assert(m->busy && !m->absent && !m->fictitious);
                    251: 
                    252:        /*
                    253:         *      If we are not flushing the page, allocate a
                    254:         *      page in the object.  If we cannot get the
                    255:         *      page, flush instead.
                    256:         */
                    257:        if (!flush) {
                    258:                vm_object_lock(new_object);
                    259:                new_m = vm_page_alloc(new_object, new_offset);
                    260:                if (new_m == VM_PAGE_NULL)
                    261:                        flush = TRUE;
                    262:                vm_object_unlock(new_object);
                    263:        }
                    264: 
                    265:        if (flush) {
                    266:                /*
                    267:                 *      Create a place-holder page where the old one was,
                    268:                 *      to prevent anyone from attempting to page in this
                    269:                 *      page while we`re unlocked.
                    270:                 */
                    271:                while ((holding_page = vm_page_grab_fictitious())
                    272:                                                        == VM_PAGE_NULL)
                    273:                        vm_page_more_fictitious();
                    274: 
                    275:                vm_object_lock(old_object);
                    276:                vm_page_lock_queues();
                    277:                vm_page_remove(m);
                    278:                vm_page_unlock_queues();
                    279:                PAGE_WAKEUP_DONE(m);
                    280: 
                    281:                vm_page_lock_queues();
                    282:                vm_page_insert(holding_page, old_object, m->offset);
                    283:                vm_page_unlock_queues();
                    284: 
                    285:                /*
                    286:                 *      Record that this page has been written out
                    287:                 */
                    288: #if    MACH_PAGEMAP
                    289:                vm_external_state_set(old_object->existence_info,
                    290:                                        paging_offset,
                    291:                                        VM_EXTERNAL_STATE_EXISTS);
                    292: #endif /* MACH_PAGEMAP */
                    293: 
                    294:                vm_object_unlock(old_object);
                    295: 
                    296:                vm_object_lock(new_object);
                    297: 
                    298:                /*
                    299:                 *      Move this page into the new object
                    300:                 */
                    301: 
                    302:                vm_page_lock_queues();
                    303:                vm_page_insert(m, new_object, new_offset);
                    304:                vm_page_unlock_queues();
                    305: 
                    306:                m->dirty = TRUE;
                    307:                m->precious = FALSE;
                    308:                m->page_lock = VM_PROT_NONE;
                    309:                m->unlock_request = VM_PROT_NONE;
                    310:        }
                    311:        else {
                    312:                /*
                    313:                 *      Copy the data into the new page,
                    314:                 *      and mark the new page as clean.
                    315:                 */
                    316:                vm_page_copy(m, new_m);
                    317: 
                    318:                vm_object_lock(old_object);
                    319:                m->dirty = FALSE;
                    320:                pmap_clear_modify(m->phys_addr);
                    321: 
                    322:                /*
                    323:                 *      Deactivate old page.
                    324:                 */
                    325:                vm_page_lock_queues();
                    326:                vm_page_deactivate(m);
                    327:                vm_page_unlock_queues();
                    328: 
                    329:                PAGE_WAKEUP_DONE(m);
                    330: 
                    331:                /*
                    332:                 *      Record that this page has been written out
                    333:                 */
                    334: 
                    335: #if    MACH_PAGEMAP
                    336:                vm_external_state_set(old_object->existence_info,
                    337:                                        paging_offset,
                    338:                                        VM_EXTERNAL_STATE_EXISTS);
                    339: #endif /* MACH_PAGEMAP */
                    340: 
                    341:                vm_object_unlock(old_object);
                    342: 
                    343:                vm_object_lock(new_object);
                    344: 
                    345:                /*
                    346:                 *      Use the new page below.
                    347:                 */
                    348:                m = new_m;
                    349:                m->dirty = TRUE;
                    350:                assert(!m->precious);
                    351:                PAGE_WAKEUP_DONE(m);
                    352:        }
                    353: 
                    354:        /*
                    355:         *      Make the old page eligible for replacement again; if a
                    356:         *      user-supplied memory manager fails to release the page,
                    357:         *      it will be paged out again to the default memory manager.
                    358:         *
                    359:         *      Note that pages written to the default memory manager
                    360:         *      must be wired down -- in return, it guarantees to free
                    361:         *      this page, rather than reusing it.
                    362:         */
                    363: 
                    364:        vm_page_lock_queues();
                    365:        vm_stat.pageouts++;
                    366:        if (m->laundry) {
                    367:                /*
                    368:                 *      vm_pageout_scan is telling us to put this page
                    369:                 *      at the front of the inactive queue, so it will
                    370:                 *      be immediately paged out to the default pager.
                    371:                 */
                    372: 
                    373:                assert(!old_object->internal);
                    374:                m->laundry = FALSE;
                    375: 
                    376:                queue_enter_first(&vm_page_queue_inactive, m,
                    377:                                  vm_page_t, pageq);
                    378:                m->inactive = TRUE;
                    379:                vm_page_inactive_count++;
                    380:        } else if (old_object->internal) {
                    381:                m->laundry = TRUE;
                    382:                vm_page_laundry_count++;
                    383: 
                    384:                vm_page_wire(m);
                    385:        } else
                    386:                vm_page_activate(m);
                    387:        vm_page_unlock_queues();
                    388: 
                    389:        /*
                    390:         *      Since IPC operations may block, we drop locks now.
                    391:         *      [The placeholder page is busy, and we still have
                    392:         *      paging_in_progress incremented.]
                    393:         */
                    394: 
                    395:        vm_object_unlock(new_object);
                    396: 
                    397:        /*
                    398:         *      Return the placeholder page to simplify cleanup.
                    399:         */
                    400:        return (flush ? holding_page : VM_PAGE_NULL);
                    401: }
                    402: 
                    403: /*
                    404:  *     Routine:        vm_pageout_page
                    405:  *     Purpose:
                    406:  *             Causes the specified page to be written back to
                    407:  *             the appropriate memory object.
                    408:  *
                    409:  *             The "initial" argument specifies whether this
                    410:  *             data is an initialization only, and should use
                    411:  *             memory_object_data_initialize instead of
                    412:  *             memory_object_data_write.
                    413:  *
                    414:  *             The "flush" argument specifies whether the page
                    415:  *             should be flushed from the object.  If not, a
                    416:  *             copy of the data is sent to the memory object.
                    417:  *
                    418:  *     In/out conditions:
                    419:  *             The page in question must not be on any pageout queues.
                    420:  *             The object to which it belongs must be locked.
                    421:  *     Implementation:
                    422:  *             Move this page to a completely new object, if flushing;
                    423:  *             copy to a new page in a new object, if not.
                    424:  */
                    425: void
                    426: vm_pageout_page(m, initial, flush)
                    427:        register vm_page_t      m;
                    428:        boolean_t               initial;
                    429:        boolean_t               flush;
                    430: {
                    431:        vm_map_copy_t           copy;
                    432:        register vm_object_t    old_object;
                    433:        register vm_object_t    new_object;
                    434:        register vm_page_t      holding_page;
                    435:        vm_offset_t             paging_offset;
                    436:        kern_return_t           rc;
                    437:        boolean_t               precious_clean;
                    438: 
                    439:        assert(m->busy);
                    440: 
                    441:        /*
                    442:         *      Cleaning but not flushing a clean precious page is a
                    443:         *      no-op.  Remember whether page is clean and precious now
                    444:         *      because vm_pageout_setup will mark it dirty and not precious.
                    445:         *
                    446:         * XXX Check if precious_clean && !flush can really happen.
                    447:         */
                    448:        precious_clean = (!m->dirty) && m->precious;
                    449:        if (precious_clean && !flush) {
                    450:                PAGE_WAKEUP_DONE(m);
                    451:                return;
                    452:        }
                    453: 
                    454:        /*
                    455:         *      Verify that we really want to clean this page.
                    456:         */
                    457:        if (m->absent || m->error || (!m->dirty && !m->precious)) {
                    458:                VM_PAGE_FREE(m);
                    459:                return;
                    460:        }
                    461: 
                    462:        /*
                    463:         *      Create a paging reference to let us play with the object.
                    464:         */
                    465:        old_object = m->object;
                    466:        paging_offset = m->offset + old_object->paging_offset;
                    467:        vm_object_paging_begin(old_object);
                    468:        vm_object_unlock(old_object);
                    469: 
                    470:        /*
                    471:         *      Allocate a new object into which we can put the page.
                    472:         */
                    473:        new_object = vm_object_allocate(PAGE_SIZE);
                    474: 
                    475:        /*
                    476:         *      Move the page into the new object.
                    477:         */
                    478:        holding_page = vm_pageout_setup(m,
                    479:                                paging_offset,
                    480:                                new_object,
                    481:                                0,              /* new offset */
                    482:                                flush);         /* flush */
                    483: 
                    484:        rc = vm_map_copyin_object(new_object, 0, PAGE_SIZE, &copy);
                    485:        assert(rc == KERN_SUCCESS);
                    486: 
                    487:        if (initial || old_object->use_old_pageout) {
                    488:                rc = (*(initial ? memory_object_data_initialize
                    489:                             : memory_object_data_write))
                    490:                        (old_object->pager,
                    491:                         old_object->pager_request,
                    492:                         paging_offset, (pointer_t) copy, PAGE_SIZE);
                    493:        }
                    494:        else {
                    495:                rc = memory_object_data_return(
                    496:                         old_object->pager,
                    497:                         old_object->pager_request,
                    498:                         paging_offset, (pointer_t) copy, PAGE_SIZE,
                    499:                         !precious_clean, !flush);
                    500:        }
                    501: 
                    502:        if (rc != KERN_SUCCESS)
                    503:                vm_map_copy_discard(copy);
                    504: 
                    505:        /*
                    506:         *      Clean up.
                    507:         */
                    508:        vm_object_lock(old_object);
                    509:        if (holding_page != VM_PAGE_NULL)
                    510:            VM_PAGE_FREE(holding_page);
                    511:        vm_object_paging_end(old_object);
                    512: }
                    513: 
                    514: /*
                    515:  *     vm_pageout_scan does the dirty work for the pageout daemon.
                    516:  *     It returns with vm_page_queue_free_lock held and
                    517:  *     vm_page_free_wanted == 0.
                    518:  */
                    519: 
                    520: void vm_pageout_scan()
                    521: {
                    522:        unsigned int burst_count;
                    523:        unsigned int want_pages;
                    524: 
                    525:        /*
                    526:         *      We want to gradually dribble pages from the active queue
                    527:         *      to the inactive queue.  If we let the inactive queue get
                    528:         *      very small, and then suddenly dump many pages into it,
                    529:         *      those pages won't get a sufficient chance to be referenced
                    530:         *      before we start taking them from the inactive queue.
                    531:         *
                    532:         *      We must limit the rate at which we send pages to the pagers.
                    533:         *      data_write messages consume memory, for message buffers and
                    534:         *      for map-copy objects.  If we get too far ahead of the pagers,
                    535:         *      we can potentially run out of memory.
                    536:         *
                    537:         *      We can use the laundry count to limit directly the number
                    538:         *      of pages outstanding to the default pager.  A similar
                    539:         *      strategy for external pagers doesn't work, because
                    540:         *      external pagers don't have to deallocate the pages sent them,
                    541:         *      and because we might have to send pages to external pagers
                    542:         *      even if they aren't processing writes.  So we also
                    543:         *      use a burst count to limit writes to external pagers.
                    544:         *
                    545:         *      When memory is very tight, we can't rely on external pagers to
                    546:         *      clean pages.  They probably aren't running, because they
                    547:         *      aren't vm-privileged.  If we kept sending dirty pages to them,
                    548:         *      we could exhaust the free list.  However, we can't just ignore
                    549:         *      pages belonging to external objects, because there might be no
                    550:         *      pages belonging to internal objects.  Hence, we get the page
                    551:         *      into an internal object and then immediately double-page it,
                    552:         *      sending it to the default pager.
                    553:         *
                    554:         *      consider_zone_gc should be last, because the other operations
                    555:         *      might return memory to zones.  When we pause we use
                    556:         *      vm_pageout_scan_continue as our continuation, so we will
                    557:         *      reenter vm_pageout_scan periodically and attempt to reclaim
                    558:         *      internal memory even if we never reach vm_page_free_target.
                    559:         */
                    560: 
                    561:     Restart:
                    562:        stack_collect();
                    563:        net_kmsg_collect();
                    564:        consider_lmm_collect();
                    565:        consider_task_collect();
                    566:        consider_thread_collect();
                    567:        consider_zone_gc();
                    568: 
                    569:        for (burst_count = 0;;) {
                    570:                register vm_page_t m;
                    571:                register vm_object_t object;
                    572:                unsigned int free_count;
                    573: 
                    574:                /*
                    575:                 *      Recalculate vm_page_inactivate_target.
                    576:                 */
                    577: 
                    578:                vm_page_lock_queues();
                    579:                vm_page_inactive_target =
                    580:                        VM_PAGE_INACTIVE_TARGET(vm_page_active_count +
                    581:                                                vm_page_inactive_count);
                    582: 
                    583:                /*
                    584:                 *      Move pages from active to inactive.
                    585:                 */
                    586: 
                    587:                while ((vm_page_inactive_count < vm_page_inactive_target) &&
                    588:                       !queue_empty(&vm_page_queue_active)) {
                    589:                        register vm_object_t obj;
                    590: 
                    591:                        vm_pageout_active++;
                    592:                        m = (vm_page_t) queue_first(&vm_page_queue_active);
                    593:                        assert(m->active && !m->inactive);
                    594: 
                    595:                        obj = m->object;
                    596:                        if (!vm_object_lock_try(obj)) {
                    597:                                /*
                    598:                                 *      Move page to end and continue.
                    599:                                 */
                    600: 
                    601:                                queue_remove(&vm_page_queue_active, m,
                    602:                                             vm_page_t, pageq);
                    603:                                queue_enter(&vm_page_queue_active, m,
                    604:                                            vm_page_t, pageq);
                    605:                                vm_page_unlock_queues();
                    606:                                vm_page_lock_queues();
                    607:                                continue;
                    608:                        }
                    609: 
                    610:                        /*
                    611:                         *      If the page is busy, then we pull it
                    612:                         *      off the active queue and leave it alone.
                    613:                         */
                    614: 
                    615:                        if (m->busy) {
                    616:                                vm_object_unlock(obj);
                    617:                                queue_remove(&vm_page_queue_active, m,
                    618:                                             vm_page_t, pageq);
                    619:                                m->active = FALSE;
                    620:                                vm_page_active_count--;
                    621:                                continue;
                    622:                        }
                    623: 
                    624:                        /*
                    625:                         *      Deactivate the page while holding the object
                    626:                         *      locked, so we know the page is still not busy.
                    627:                         *      This should prevent races between pmap_enter
                    628:                         *      and pmap_clear_reference.  The page might be
                    629:                         *      absent or fictitious, but vm_page_deactivate
                    630:                         *      can handle that.
                    631:                         */
                    632: 
                    633:                        vm_page_deactivate(m);
                    634:                        vm_object_unlock(obj);
                    635:                }
                    636: 
                    637:                /*
                    638:                 *      We are done if we have met our targets *and*
                    639:                 *      nobody is still waiting for a page.
                    640:                 */
                    641: 
                    642:                simple_lock(&vm_page_queue_free_lock);
                    643:                free_count = vm_page_free_count;
                    644:                if ((free_count >= vm_page_free_target) &&
                    645:                    (vm_page_external_count <= vm_page_external_target) &&
                    646:                    (vm_page_free_wanted == 0)) {
                    647:                        vm_page_unlock_queues();
                    648:                        break;
                    649:                }
                    650:                want_pages = ((free_count < vm_page_free_target) ||
                    651:                              vm_page_free_wanted);
                    652:                simple_unlock(&vm_page_queue_free_lock);
                    653: 
                    654:                /*
                    655:                 * Sometimes we have to pause:
                    656:                 *      1) No inactive pages - nothing to do.
                    657:                 *      2) Flow control - wait for pagers to catch up.
                    658:                 *      3) Extremely low memory - sending out dirty pages
                    659:                 *      consumes memory.  We don't take the risk of doing
                    660:                 *      this if the default pager already has work to do.
                    661:                 */
                    662:        pause:
                    663:                if (queue_empty(&vm_page_queue_inactive) ||
                    664:                    (burst_count >= vm_pageout_burst_max) ||
                    665:                    (vm_page_laundry_count >= vm_pageout_burst_max) ||
                    666:                    ((free_count < vm_pageout_reserved_really) &&
                    667:                     (vm_page_laundry_count > 0))) {
                    668:                        unsigned int pages, msecs;
                    669: 
                    670:                        /*
                    671:                         *      vm_pageout_burst_wait is msecs/page.
                    672:                         *      If there is nothing for us to do, we wait
                    673:                         *      at least vm_pageout_empty_wait msecs.
                    674:                         */
                    675: 
                    676:                        if (vm_page_laundry_count > burst_count)
                    677:                                pages = vm_page_laundry_count;
                    678:                        else
                    679:                                pages = burst_count;
                    680:                        msecs = pages * vm_pageout_burst_wait;
                    681: 
                    682:                        if (queue_empty(&vm_page_queue_inactive) &&
                    683:                            (msecs < vm_pageout_empty_wait))
                    684:                                msecs = vm_pageout_empty_wait;
                    685:                        vm_page_unlock_queues();
                    686: 
                    687:                        thread_will_wait_with_timeout(current_thread(), msecs);
                    688:                        counter(c_vm_pageout_scan_block++);
                    689:                        thread_block(vm_pageout_scan_continue);
                    690: #ifndef CONTINUATIONS
                    691:                        /*
                    692:                         *      Unfortunately, we don't have call_continuation
                    693:                         *      so we can't rely on tail-recursion.
                    694:                         */
                    695: 
                    696:                        vm_pageout_scan_continue();
                    697:                        goto Restart;
                    698: #else  /* CONTINUATIONS */
                    699:                        call_continuation(vm_pageout_scan_continue);
                    700:                        /*NOTREACHED*/
                    701: #endif /* CONTINUATIONS */
                    702:                }
                    703: 
                    704:                vm_pageout_inactive++;
                    705: 
                    706:                /* Find a page we are interested in paging out.  If we
                    707:                   need pages, then we'll page anything out; otherwise
                    708:                   we only page out external pages. */
                    709:                m = (vm_page_t) queue_first (&vm_page_queue_inactive);
                    710:                while (1)
                    711:                  {
                    712:                    assert (!m->active && m->inactive);
                    713:                    if (want_pages || m->external)
                    714:                      break;
                    715: 
                    716:                    m = (vm_page_t) queue_next (m);
                    717:                    if (!m)
                    718:                      goto pause;
                    719:                  }
                    720: 
                    721:                object = m->object;
                    722: 
                    723:                /*
                    724:                 *      Try to lock object; since we've got the
                    725:                 *      page queues lock, we can only try for this one.
                    726:                 */
                    727: 
                    728:                if (!vm_object_lock_try(object)) {
                    729:                        /*
                    730:                         *      Move page to end and continue.
                    731:                         */
                    732: 
                    733:                        queue_remove(&vm_page_queue_inactive, m,
                    734:                                     vm_page_t, pageq);
                    735:                        queue_enter(&vm_page_queue_inactive, m,
                    736:                                    vm_page_t, pageq);
                    737:                        vm_page_unlock_queues();
                    738:                        vm_pageout_inactive_nolock++;
                    739:                        continue;
                    740:                }
                    741: 
                    742:                /*
                    743:                 *      Remove the page from the inactive list.
                    744:                 */
                    745: 
                    746:                queue_remove(&vm_page_queue_inactive, m, vm_page_t, pageq);
                    747:                vm_page_inactive_count--;
                    748:                m->inactive = FALSE;
                    749: 
                    750:                if (m->busy || !object->alive) {
                    751:                        /*
                    752:                         *      Somebody is already playing with this page.
                    753:                         *      Leave it off the pageout queues.
                    754:                         */
                    755: 
                    756:                        vm_page_unlock_queues();
                    757:                        vm_object_unlock(object);
                    758:                        vm_pageout_inactive_busy++;
                    759:                        continue;
                    760:                }
                    761: 
                    762:                /*
                    763:                 *      If it's absent, we can reclaim the page.
                    764:                 */
                    765: 
                    766:                if (want_pages && m->absent) {
                    767:                        vm_pageout_inactive_absent++;
                    768:                    reclaim_page:
                    769:                        vm_page_free(m);
                    770:                        vm_page_unlock_queues();
                    771:                        vm_object_unlock(object);
                    772:                        continue;
                    773:                }
                    774: 
                    775:                /*
                    776:                 *      If it's being used, reactivate.
                    777:                 *      (Fictitious pages are either busy or absent.)
                    778:                 */
                    779: 
                    780:                assert(!m->fictitious);
                    781:                if (m->reference || pmap_is_referenced(m->phys_addr)) {
                    782:                        vm_object_unlock(object);
                    783:                        vm_page_activate(m);
                    784:                        vm_stat.reactivations++;
                    785:                        vm_page_unlock_queues();
                    786:                        vm_pageout_inactive_used++;
                    787:                        continue;
                    788:                }
                    789: 
                    790:                /*
                    791:                 *      Eliminate all mappings.
                    792:                 */
                    793: 
                    794:                m->busy = TRUE;
                    795:                pmap_page_protect(m->phys_addr, VM_PROT_NONE);
                    796:                if (!m->dirty)
                    797:                        m->dirty = pmap_is_modified(m->phys_addr);
                    798: 
                    799:                if (m->external) {
                    800:                        /* Figure out if we still care about this
                    801:                        page in the limit of externally managed pages.
                    802:                        Clean pages don't actually cause system hosage,
                    803:                        so it's ok to stop considering them as
                    804:                        "consumers" of memory. */
                    805:                        if (m->dirty && !m->extcounted) {
                    806:                                m->extcounted = TRUE;
                    807:                                vm_page_external_count++;
                    808:                        } else if (!m->dirty && m->extcounted) {
                    809:                                m->extcounted = FALSE;
                    810:                                vm_page_external_count--;
                    811:                        }
                    812:                }
                    813: 
                    814:                /* If we don't actually need more memory, and the page
                    815:                   is not dirty, put it on the tail of the inactive queue
                    816:                   and move on to the next page. */
                    817:                if (!want_pages && !m->dirty) {
                    818:                        queue_remove (&vm_page_queue_inactive, m,
                    819:                                      vm_page_t, pageq);
                    820:                        queue_enter (&vm_page_queue_inactive, m,
                    821:                                     vm_page_t, pageq);
                    822:                        vm_page_unlock_queues();
                    823:                        vm_pageout_inactive_cleaned_external++;
                    824:                        continue;
                    825:                }
                    826: 
                    827:                /*
                    828:                 *      If it's clean and not precious, we can free the page.
                    829:                 */
                    830: 
                    831:                if (!m->dirty && !m->precious) {
                    832:                        vm_pageout_inactive_clean++;
                    833:                        goto reclaim_page;
                    834:                }
                    835: 
                    836:                /*
                    837:                 *      If we are very low on memory, then we can't
                    838:                 *      rely on an external pager to clean a dirty page,
                    839:                 *      because external pagers are not vm-privileged.
                    840:                 *
                    841:                 *      The laundry bit tells vm_pageout_setup to
                    842:                 *      put the page back at the front of the inactive
                    843:                 *      queue instead of activating the page.  Hence,
                    844:                 *      we will pick the page up again immediately and
                    845:                 *      resend it to the default pager.
                    846:                 */
                    847: 
                    848:                assert(!m->laundry);
                    849:                if ((free_count < vm_pageout_reserved_internal) &&
                    850:                    !object->internal) {
                    851:                        m->laundry = TRUE;
                    852:                        vm_pageout_inactive_double++;
                    853:                }
                    854:                vm_page_unlock_queues();
                    855: 
                    856:                /*
                    857:                 *      If there is no memory object for the page, create
                    858:                 *      one and hand it to the default pager.
                    859:                 *      [First try to collapse, so we don't create
                    860:                 *      one unnecessarily.]
                    861:                 */
                    862: 
                    863:                if (!object->pager_initialized)
                    864:                        vm_object_collapse(object);
                    865:                if (!object->pager_initialized)
                    866:                        vm_object_pager_create(object);
                    867:                if (!object->pager_initialized)
                    868:                        panic("vm_pageout_scan");
                    869: 
                    870:                vm_pageout_inactive_dirty++;
                    871:                vm_pageout_page(m, FALSE, TRUE);        /* flush it */
                    872:                vm_object_unlock(object);
                    873:                burst_count++;
                    874:        }
                    875: }
                    876: 
                    877: void vm_pageout_scan_continue()
                    878: {
                    879:        /*
                    880:         *      We just paused to let the pagers catch up.
                    881:         *      If vm_page_laundry_count is still high,
                    882:         *      then we aren't waiting long enough.
                    883:         *      If we have paused some vm_pageout_pause_max times without
                    884:         *      adjusting vm_pageout_burst_wait, it might be too big,
                    885:         *      so we decrease it.
                    886:         */
                    887: 
                    888:        vm_page_lock_queues();
                    889:        if (vm_page_laundry_count > vm_pageout_burst_min) {
                    890:                vm_pageout_burst_wait++;
                    891:                vm_pageout_pause_count = 0;
                    892:        } else if (++vm_pageout_pause_count > vm_pageout_pause_max) {
                    893:                vm_pageout_burst_wait = (vm_pageout_burst_wait * 3) / 4;
                    894:                if (vm_pageout_burst_wait < 1)
                    895:                        vm_pageout_burst_wait = 1;
                    896:                vm_pageout_pause_count = 0;
                    897:        }
                    898:        vm_page_unlock_queues();
                    899: 
                    900: #ifdef CONTINUATIONS
                    901:        vm_pageout_continue();
                    902:        /*NOTREACHED*/
                    903: #endif /* CONTINUATIONS */
                    904: }
                    905: 
                    906: /*
                    907:  *     vm_pageout is the high level pageout daemon.
                    908:  */
                    909: 
                    910: void vm_pageout_continue()
                    911: {
                    912:        /*
                    913:         *      The pageout daemon is never done, so loop forever.
                    914:         *      We should call vm_pageout_scan at least once each
                    915:         *      time we are woken, even if vm_page_free_wanted is
                    916:         *      zero, to check vm_page_free_target and
                    917:         *      vm_page_inactive_target.
                    918:         */
                    919: 
                    920:        for (;;) {
                    921:                vm_pageout_scan();
                    922:                /* we hold vm_page_queue_free_lock now */
                    923:                assert(vm_page_free_wanted == 0);
                    924: 
                    925:                assert_wait(&vm_page_free_wanted, FALSE);
                    926:                simple_unlock(&vm_page_queue_free_lock);
                    927:                counter(c_vm_pageout_block++);
                    928:                thread_block(vm_pageout_continue);
                    929:        }
                    930: }
                    931: 
                    932: void vm_pageout()
                    933: {
                    934:        int             free_after_reserve;
                    935: 
                    936:        current_thread()->vm_privilege = TRUE;
                    937:        stack_privilege(current_thread());
                    938: 
                    939:        /*
                    940:         *      Initialize some paging parameters.
                    941:         */
                    942: 
                    943:        if (vm_pageout_burst_max == 0)
                    944:                vm_pageout_burst_max = VM_PAGEOUT_BURST_MAX;
                    945: 
                    946:        if (vm_pageout_burst_min == 0)
                    947:                vm_pageout_burst_min = VM_PAGEOUT_BURST_MIN;
                    948: 
                    949:        if (vm_pageout_burst_wait == 0)
                    950:                vm_pageout_burst_wait = VM_PAGEOUT_BURST_WAIT;
                    951: 
                    952:        if (vm_pageout_empty_wait == 0)
                    953:                vm_pageout_empty_wait = VM_PAGEOUT_EMPTY_WAIT;
                    954: 
                    955:        if (vm_page_free_reserved == 0)
                    956:                vm_page_free_reserved = VM_PAGE_FREE_RESERVED;
                    957: 
                    958:        if (vm_pageout_pause_max == 0)
                    959:                vm_pageout_pause_max = VM_PAGEOUT_PAUSE_MAX;
                    960: 
                    961:        if (vm_pageout_reserved_internal == 0)
                    962:                vm_pageout_reserved_internal =
                    963:                        VM_PAGEOUT_RESERVED_INTERNAL(vm_page_free_reserved);
                    964: 
                    965:        if (vm_pageout_reserved_really == 0)
                    966:                vm_pageout_reserved_really =
                    967:                        VM_PAGEOUT_RESERVED_REALLY(vm_page_free_reserved);
                    968: 
                    969:        free_after_reserve = vm_page_free_count - vm_page_free_reserved;
                    970: 
                    971:        if (vm_page_external_limit == 0)
                    972:                vm_page_external_limit =
                    973:                        VM_PAGE_EXTERNAL_LIMIT (free_after_reserve);
                    974: 
                    975:        if (vm_page_external_target == 0)
                    976:                vm_page_external_target =
                    977:                        VM_PAGE_EXTERNAL_TARGET (free_after_reserve);
                    978: 
                    979:        if (vm_page_free_min == 0)
                    980:                vm_page_free_min = vm_page_free_reserved +
                    981:                        VM_PAGE_FREE_MIN(free_after_reserve);
                    982: 
                    983:        if (vm_page_free_target == 0)
                    984:                vm_page_free_target = vm_page_free_reserved +
                    985:                        VM_PAGE_FREE_TARGET(free_after_reserve);
                    986: 
                    987:        if (vm_page_free_target < vm_page_free_min + 5)
                    988:                vm_page_free_target = vm_page_free_min + 5;
                    989: 
                    990:        /*
                    991:         *      vm_pageout_scan will set vm_page_inactive_target.
                    992:         */
                    993: 
                    994:        vm_pageout_continue();
                    995:        /*NOTREACHED*/
                    996: }

unix.superglobalmegacorp.com

This archive runs on limited infrastructure. Preserving old code on modern bandwidth. Automated agents are requested to crawl responsibly.