uvm_pdaemon.c revision 1.4 1 /* $NetBSD: uvm_pdaemon.c,v 1.4 1998/02/07 11:09:33 mrg Exp $ */
2
3 /*
4 * XXXCDC: "ROUGH DRAFT" QUALITY UVM PRE-RELEASE FILE!
5 * >>>USE AT YOUR OWN RISK, WORK IS NOT FINISHED<<<
6 */
7 /*
8 * Copyright (c) 1997 Charles D. Cranor and Washington University.
9 * Copyright (c) 1991, 1993, The Regents of the University of California.
10 *
11 * All rights reserved.
12 *
13 * This code is derived from software contributed to Berkeley by
14 * The Mach Operating System project at Carnegie-Mellon University.
15 *
16 * Redistribution and use in source and binary forms, with or without
17 * modification, are permitted provided that the following conditions
18 * are met:
19 * 1. Redistributions of source code must retain the above copyright
20 * notice, this list of conditions and the following disclaimer.
21 * 2. Redistributions in binary form must reproduce the above copyright
22 * notice, this list of conditions and the following disclaimer in the
23 * documentation and/or other materials provided with the distribution.
24 * 3. All advertising materials mentioning features or use of this software
25 * must display the following acknowledgement:
26 * This product includes software developed by Charles D. Cranor,
27 * Washington University, the University of California, Berkeley and
28 * its contributors.
29 * 4. Neither the name of the University nor the names of its contributors
30 * may be used to endorse or promote products derived from this software
31 * without specific prior written permission.
32 *
33 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
34 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
35 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
36 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
37 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
38 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
39 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
40 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
41 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
42 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
43 * SUCH DAMAGE.
44 *
45 * @(#)vm_pageout.c 8.5 (Berkeley) 2/14/94
46 * from: Id: uvm_pdaemon.c,v 1.1.2.32 1998/02/06 05:26:30 chs Exp
47 *
48 *
49 * Copyright (c) 1987, 1990 Carnegie-Mellon University.
50 * All rights reserved.
51 *
52 * Permission to use, copy, modify and distribute this software and
53 * its documentation is hereby granted, provided that both the copyright
54 * notice and this permission notice appear in all copies of the
55 * software, derivative works or modified versions, and any portions
56 * thereof, and that both notices appear in supporting documentation.
57 *
58 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
59 * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
60 * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
61 *
62 * Carnegie Mellon requests users of this software to return to
63 *
64 * Software Distribution Coordinator or Software.Distribution (at) CS.CMU.EDU
65 * School of Computer Science
66 * Carnegie Mellon University
67 * Pittsburgh PA 15213-3890
68 *
69 * any improvements or extensions that they make and grant Carnegie the
70 * rights to redistribute these changes.
71 */
72
73 /*
74 * uvm_pdaemon.c: the page daemon
75 */
76
77 #include <sys/param.h>
78 #include <sys/proc.h>
79 #include <sys/systm.h>
80 #include <sys/kernel.h>
81
82 #include <vm/vm.h>
83 #include <vm/vm_page.h>
84 #include <vm/vm_kern.h>
85
86 #include <uvm/uvm.h>
87
88 UVMHIST_DECL(pdhist);
89
90 /*
91 * local prototypes
92 */
93
94 static void uvmpd_scan __P((void));
95 static boolean_t uvmpd_scan_inactive __P((struct pglist *));
96 static void uvmpd_tune __P((void));
97
98
99 /*
100 * uvm_wait: wait (sleep) for the page daemon to free some pages
101 *
102 * => should be called with all locks released
103 * => should _not_ be called by the page daemon (to avoid deadlock)
104 */
105
106 void uvm_wait(wmsg)
107
108 char *wmsg;
109
110 {
111 int timo = 0;
112 int s = splbio();
113
114 /*
115 * check for page daemon going to sleep (waiting for itself)
116 */
117
118 if (curproc == uvm.pagedaemon_proc) {
119 /*
120 * now we have a problem: the pagedaemon wants to go to sleep until
121 * it frees more memory. but how can it free more memory if it is
122 * asleep? that is a deadlock. we have two options:
123 * [1] panic now
124 * [2] put a timeout on the sleep, thus causing the pagedaemon to
125 * only pause (rather than sleep forever)
126 *
127 * note that option [2] will only help us if we get lucky and some
128 * other process on the system breaks the deadlock by exiting or
129 * freeing memory (thus allowing the pagedaemon to continue).
130 * for now we panic if DEBUG is defined, otherwise we hope for the
131 * best with option [2] (better yet, this should never happen in
132 * the first place!).
133 */
134
135 printf("pagedaemon: deadlock detected!\n");
136 timo = hz >> 3; /* set timeout */
137 #if defined(DEBUG)
138 panic("pagedaemon deadlock"); /* DEBUG: panic so we can debug it */
139 #endif
140 }
141
142 simple_lock(&uvm.pagedaemon_lock);
143 thread_wakeup(&uvm.pagedaemon); /* wake the daemon! */
144 UVM_UNLOCK_AND_WAIT(&uvmexp.free, &uvm.pagedaemon_lock, FALSE, wmsg, timo);
145
146 splx(s);
147 }
148
149
150 /*
151 * uvmpd_tune: tune paging parameters
152 *
153 * => called when ever memory is added (or removed?) to the system
154 * => caller must call with page queues locked
155 */
156
157 static void uvmpd_tune()
158
159 {
160 UVMHIST_FUNC("uvmpd_tune"); UVMHIST_CALLED(pdhist);
161
162 uvmexp.freemin = uvmexp.npages / 20;
163 uvmexp.freemin = max(uvmexp.freemin, (16*1024)/PAGE_SIZE); /* at least 16K */
164 uvmexp.freemin = min(uvmexp.freemin, (256*1024)/PAGE_SIZE); /* at most 256K */
165
166 uvmexp.freetarg = (uvmexp.freemin * 4) / 3;
167 if (uvmexp.freetarg <= uvmexp.freemin)
168 uvmexp.freetarg = uvmexp.freemin + 1;
169
170 /* uvmexp.inactarg: computed in main daemon loop */
171
172 uvmexp.wiredmax = uvmexp.npages / 3;
173 UVMHIST_LOG(pdhist, "<- done, freemin=%d, freetarg=%d, wiredmax=%d",
174 uvmexp.freemin, uvmexp.freetarg, uvmexp.wiredmax, 0);
175 }
176
177 /*
178 * uvm_pageout: the main loop for the pagedaemon
179 */
180
181 void uvm_pageout()
182
183 {
184 int npages = 0;
185 int s;
186 struct uvm_aiodesc *aio, *nextaio;
187 UVMHIST_FUNC("uvm_pageout"); UVMHIST_CALLED(pdhist);
188
189 UVMHIST_LOG(pdhist,"<starting uvm pagedaemon>", 0, 0, 0, 0);
190
191 /*
192 * ensure correct priority and set paging parameters...
193 */
194
195 uvm.pagedaemon_proc = curproc;
196 (void) spl0();
197 uvm_lock_pageq();
198 npages = uvmexp.npages;
199 uvmpd_tune();
200 uvm_unlock_pageq();
201
202 /*
203 * main loop
204 */
205 while (TRUE) {
206
207 /*
208 * carefully attempt to go to sleep (without losing "wakeups"!).
209 * we need splbio because we want to make sure the aio_done list
210 * is totally empty before we go to sleep.
211 */
212
213 s = splbio();
214 simple_lock(&uvm.pagedaemon_lock);
215
216 /*
217 * if we've got done aio's, then bypass the sleep
218 */
219
220 if (uvm.aio_done.tqh_first == NULL) {
221 UVMHIST_LOG(maphist," <<SLEEPING>>",0,0,0,0);
222 UVM_UNLOCK_AND_WAIT(&uvm.pagedaemon, &uvm.pagedaemon_lock, FALSE,
223 "daemon_slp", 0);
224 uvmexp.pdwoke++;
225 UVMHIST_LOG(pdhist," <<WOKE UP>>",0,0,0,0);
226
227 /* relock pagedaemon_lock, still at splbio */
228 simple_lock(&uvm.pagedaemon_lock);
229 }
230
231 /*
232 * check for done aio structures
233 */
234
235 aio = uvm.aio_done.tqh_first; /* save current list (if any)*/
236 if (aio) {
237 TAILQ_INIT(&uvm.aio_done); /* zero global list */
238 }
239
240 simple_unlock(&uvm.pagedaemon_lock); /* unlock */
241 splx(s); /* drop splbio */
242
243 /*
244 * first clear out any pending aios (to free space in case we
245 * want to pageout more stuff).
246 */
247
248 for (/*null*/; aio != NULL ; aio = nextaio) {
249
250 uvmexp.paging -= aio->npages;
251 nextaio = aio->aioq.tqe_next;
252 aio->aiodone(aio);
253
254 }
255
256 /*
257 * now lock page queues and recompute inactive count
258 */
259 uvm_lock_pageq();
260
261 if (npages != uvmexp.npages) { /* check for new pages? */
262 npages = uvmexp.npages;
263 uvmpd_tune();
264 }
265
266 uvmexp.inactarg = (uvmexp.active + uvmexp.inactive) / 3;
267 if (uvmexp.inactarg <= uvmexp.freetarg)
268 uvmexp.inactarg = uvmexp.freetarg + 1;
269
270 UVMHIST_LOG(pdhist," free/ftarg=%d/%d, inact/itarg=%d/%d",
271 uvmexp.free, uvmexp.freetarg, uvmexp.inactive, uvmexp.inactarg);
272
273 /*
274 * scan if needed
275 * [XXX: note we are reading uvm.free without locking]
276 */
277 if (uvmexp.free < uvmexp.freetarg || uvmexp.inactive < uvmexp.inactarg)
278 uvmpd_scan();
279
280 /*
281 * done scan. unlock page queues (the only lock we are holding).
282 */
283 uvm_unlock_pageq();
284
285 /*
286 * done! restart loop.
287 */
288 thread_wakeup(&uvmexp.free);
289 }
290 /*NOTREACHED*/
291 }
292
293 /*
294 * uvmpd_scan_inactive: the first loop of uvmpd_scan broken out into
295 * its own function for ease of reading.
296 *
297 * => called with page queues locked
298 * => we work on meeting our free target by converting inactive pages
299 * into free pages.
300 * => we handle the building of swap-backed clusters
301 * => we return TRUE if we are exiting because we met our target
302 */
303
304 static boolean_t uvmpd_scan_inactive(pglst)
305
306 struct pglist *pglst;
307
308 {
309 boolean_t retval = FALSE; /* assume we haven't hit target */
310 int s, free, result;
311 struct vm_page *p, *nextpg;
312 struct uvm_object *uobj;
313 struct vm_page *pps[MAXBSIZE/PAGE_SIZE], **ppsp;
314 int npages;
315 struct vm_page *swpps[MAXBSIZE/PAGE_SIZE]; /* XXX: see below */
316 int swnpages, swcpages; /* XXX: see below */
317 int swslot, oldslot;
318 struct vm_anon *anon;
319 boolean_t swap_backed;
320 vm_offset_t start;
321 UVMHIST_FUNC("uvmpd_scan_inactive"); UVMHIST_CALLED(pdhist);
322
323 /*
324 * note: we currently keep swap-backed pages on a seperate inactive
325 * list from object-backed pages. however, merging the two lists
326 * back together again hasn't been ruled out. thus, we keep our
327 * swap cluster in "swpps" rather than in pps (allows us to mix clustering
328 * types in the event of a mixed inactive queue).
329 */
330
331 /*
332 * swslot is non-zero if we are building a swap cluster. we want
333 * to stay in the loop while we have a page to scan or we have
334 * a swap-cluster to build.
335 */
336 swslot = 0;
337 swnpages = swcpages = 0;
338 free = 0;
339
340 for (p = pglst->tqh_first ; p != NULL || swslot != 0 ; p = nextpg) {
341
342 /*
343 * note that p can be NULL iff we have traversed the whole
344 * list and need to do one final swap-backed clustered pageout.
345 */
346 if (p) {
347 /*
348 * update our copy of "free" and see if we've met our target
349 */
350 s = splimp();
351 uvm_lock_fpageq();
352 free = uvmexp.free;
353 uvm_unlock_fpageq();
354 splx(s);
355
356 if (free >= uvmexp.freetarg) {
357 UVMHIST_LOG(pdhist," met free target: exit loop", 0, 0, 0, 0);
358 retval = TRUE; /* hit the target! */
359 if (swslot == 0)
360 break; /* exit now if no swap-i/o pending */
361 p = NULL; /* set p to null to signal final swap i/o */
362 }
363 }
364
365 uobj = NULL; /* be safe and shut gcc up */
366 anon = NULL; /* be safe and shut gcc up */
367
368 if (p) { /* if (we have a new page to consider) */
369 /*
370 * we are below target and have a new page to consider.
371 */
372 uvmexp.pdscans++;
373 nextpg = p->pageq.tqe_next;
374
375 /*
376 * move referenced pages back to active queue and skip to next page
377 * (unlikely to happen since inactive pages shouldn't have any
378 * valid mappings and we cleared reference before deactivating).
379 */
380 if (pmap_is_referenced(PMAP_PGARG(p))) {
381 uvm_pageactivate(p);
382 uvmexp.pdreact++;
383 continue;
384 }
385
386 /*
387 * first we attempt to lock the object that this page belongs to.
388 * if our attempt fails we skip on to the next page (no harm done).
389 * it is important to "try" locking the object as we are locking in the
390 * wrong order (pageq -> object) and we don't want to get deadlocked.
391 *
392 * the only time we exepct to see an ownerless page (i.e. a page
393 * with no uobject and !PQ_ANON) is if an anon has loaned a page
394 * from a uvm_object and the uvm_object has dropped the ownership.
395 * in that case, the anon can "take over" the loaned page and
396 * make it its own.
397 */
398
399 /* is page part of an anon or ownerless ? */
400 if ((p->pqflags & PQ_ANON) || p->uobject == NULL) {
401
402 anon = p->uanon;
403
404 #ifdef DIAGNOSTIC
405 /* to be on inactive q, page must be part of _something_ */
406 if (anon == NULL)
407 panic("pagedaemon: page with no anon or object detected - loop 1");
408 #endif
409
410 if (!simple_lock_try(&anon->an_lock))
411 continue; /* lock failed, skip this page */
412
413 /* if the page is ownerless, claim it in the name of "anon"! */
414 if ((p->pqflags & PQ_ANON) == 0) {
415 #ifdef DIAGNOSTIC
416 if (p->loan_count < 1)
417 panic("pagedaemon: non-loaned ownerless page detected - loop 1");
418 #endif
419 p->loan_count--;
420 p->pqflags |= PQ_ANON; /* anon now owns it */
421 }
422
423 if (p->flags & PG_BUSY) {
424 simple_unlock(&anon->an_lock);
425 uvmexp.pdbusy++;
426 continue; /* someone else owns page, skip it */
427 }
428
429 uvmexp.pdanscan++;
430
431 } else {
432
433 uobj = p->uobject;
434
435 if (!simple_lock_try(&uobj->vmobjlock))
436 continue; /* lock failed, skip this page */
437
438 if (p->flags & PG_BUSY) {
439 simple_unlock(&uobj->vmobjlock);
440 uvmexp.pdbusy++;
441 continue; /* someone else owns page, skip it */
442 }
443
444 uvmexp.pdobscan++;
445
446 }
447
448 /*
449 * we now have the object and the page queues locked. the page is
450 * not busy. if the page is clean we can free it now and continue.
451 */
452
453 if (p->flags & PG_CLEAN) {
454 /* zap all mappings with pmap_page_protect... */
455 pmap_page_protect(PMAP_PGARG(p), VM_PROT_NONE);
456 uvm_pagefree(p);
457 uvmexp.pdfreed++;
458
459 if (anon) {
460 #ifdef DIAGNOSTIC
461 /*
462 * an anonymous page can only be clean if it has valid
463 * backing store.
464 */
465 if (anon->an_swslot == 0)
466 panic("pagedaemon: clean anon page without backing store?");
467 #endif
468 anon->u.an_page = NULL; /* remove from object */
469 simple_unlock(&anon->an_lock);
470 } else {
471 /* pagefree has already removed the page from the object */
472 simple_unlock(&uobj->vmobjlock);
473 }
474 continue;
475 }
476
477 /*
478 * this page is dirty, skip it if we'll have met
479 * our free target when all the current pageouts complete.
480 */
481 if (free + uvmexp.paging > uvmexp.freetarg)
482 {
483 if (anon) {
484 simple_unlock(&anon->an_lock);
485 } else {
486 simple_unlock(&uobj->vmobjlock);
487 }
488 continue;
489 }
490
491 /*
492 * the page we are looking at is dirty. we must clean it before
493 * it can be freed. to do this we first mark the page busy so that
494 * no one else will touch the page. we write protect all the mappings
495 * of the page so that no one touches it while it is in I/O.
496 */
497
498 swap_backed = ((p->pqflags & PQ_SWAPBACKED) != 0);
499 uvmexp.pdpageouts++;
500 p->flags |= PG_BUSY; /* now we own it */
501 UVM_PAGE_OWN(p, "scan_inactive");
502 pmap_page_protect(PMAP_PGARG(p), VM_PROT_READ);
503
504 /*
505 * for swap-backed pages we need to (re)allocate swap space.
506 */
507 if (swap_backed) {
508
509 /*
510 * free old swap slot (if any)
511 */
512 if (anon) {
513 if (anon->an_swslot) {
514 uvm_swap_free(anon->an_swslot, 1);
515 anon->an_swslot = 0;
516 }
517 } else {
518 oldslot = uao_set_swslot(uobj, p->offset/PAGE_SIZE, 0); /* remove */
519 if (oldslot)
520 uvm_swap_free(oldslot, 1); /* free */
521 }
522
523 /*
524 * start new cluster (if necessary)
525 */
526 if (swslot == 0) {
527 swnpages = MAXBSIZE/PAGE_SIZE; /* want this much */
528 swslot = uvm_swap_alloc(&swnpages, TRUE);
529
530 if (swslot == 0) {
531 /* no swap? give up! */
532 p->flags &= ~PG_BUSY;
533 UVM_PAGE_OWN(p, NULL);
534 if (anon)
535 simple_unlock(&anon->an_lock);
536 else
537 simple_unlock(&uobj->vmobjlock);
538 continue;
539 }
540 swcpages = 0; /* cluster is empty */
541 }
542
543 /*
544 * add block to cluster
545 */
546 swpps[swcpages] = p;
547 if (anon)
548 anon->an_swslot = swslot + swcpages;
549 else
550 uao_set_swslot(uobj, p->offset/PAGE_SIZE, swslot + swcpages);
551 swcpages++;
552
553 /* done (swap-backed) */
554 }
555
556 /* end: if (p) [end of "if we have new page to consider"] */
557 } else {
558
559 swap_backed = TRUE; /* if p == NULL we must be doing a last swap i/o */
560
561 }
562
563 /*
564 * now consider doing the pageout.
565 *
566 * for swap-backed pages, we do the pageout if we have either
567 * filled the cluster (in which case (swnpages == swcpages) or
568 * run out of pages (p == NULL).
569 *
570 * for object pages, we always do the pageout.
571 */
572 if (swap_backed) {
573
574 if (p) { /* if we just added a page to cluster */
575 if (anon)
576 simple_unlock(&anon->an_lock);
577 else
578 simple_unlock(&uobj->vmobjlock);
579 if (swcpages < swnpages) /* cluster not full yet? */
580 continue;
581 }
582
583 /* starting I/O now... set up for it */
584 npages = swcpages;
585 ppsp = swpps;
586 start = (vm_offset_t) swslot; /* for swap-backed pages only */
587
588 /* if this is final pageout we could have a few extra swap blocks */
589 if (swcpages < swnpages) {
590 uvm_swap_free(swslot + swcpages, (swnpages - swcpages));
591 }
592
593 } else {
594
595 /* normal object pageout */
596 ppsp = pps;
597 npages = sizeof(pps) / sizeof(struct vm_page *);
598 start = 0; /* not looked at because PGO_ALLPAGES is set */
599
600 }
601
602 /*
603 * now do the pageout.
604 *
605 * for swap_backed pages we have already built the cluster.
606 * for !swap_backed pages, uvm_pager_put will call the object's
607 * "make put cluster" function to build a cluster on our behalf.
608 *
609 * we pass the PGO_PDFREECLUST flag to uvm_pager_put to instruct
610 * it to free the cluster pages for us on a successful I/O (it always
611 * does this for un-successful I/O requests). this allows us to
612 * do clustered pageout without having to deal with cluster pages
613 * at this level.
614 *
615 * note locking semantics of uvm_pager_put with PGO_PDFREECLUST:
616 * IN: locked: uobj (if !swap_backed), page queues
617 * OUT: locked: uobj (if !swap_backed && result != VM_PAGER_PEND)
618 * !locked: page queues, uobj (if swap_backed || VM_PAGER_PEND)
619 *
620 * [the bit about VM_PAGER_PEND saves us one lock-unlock pair]
621 */
622
623 /* locked: uobj (if !swap_backed), page queues */
624 result = uvm_pager_put((swap_backed) ? NULL : uobj, p, &ppsp, &npages,
625 PGO_ALLPAGES|PGO_PDFREECLUST, start, 0);
626 /* locked: uobj (if !swap_backed && result != PEND) */
627 /* unlocked: page queues, object (if swap_backed || result == PEND) */
628
629 /*
630 * if we did i/o to swap, zero swslot to indicate that we are
631 * no longer building a swap-backed cluster.
632 */
633
634 if (swap_backed)
635 swslot = 0; /* done with this cluster */
636
637 /*
638 * first, we check for VM_PAGER_PEND which means that the async I/O
639 * is in progress and the async I/O done routine will clean up
640 * after us. in this case we move on to the next page.
641 *
642 * there is a very remote chance that the pending async i/o can
643 * finish _before_ we get here. if that happens, our page "p"
644 * may no longer be on the inactive queue. so we verify this
645 * when determining the next page (starting over at the head if
646 * we've lost our inactive page).
647 */
648
649 if (result == VM_PAGER_PEND) {
650 uvmexp.paging += npages;
651 uvm_lock_pageq(); /* relock page queues */
652 uvmexp.pdpending++;
653 if (p) {
654 if (p->pqflags & PQ_INACTIVE)
655 nextpg = p->pageq.tqe_next; /* reload! */
656 else
657 nextpg = pglst->tqh_first; /* reload! */
658 } else {
659 nextpg = NULL; /* done list */
660 }
661 continue;
662 }
663
664 /*
665 * clean up "p" if we have one
666 */
667
668 if (p) {
669 /*
670 * the I/O request to "p" is done and uvm_pager_put has freed
671 * any cluster pages it may have allocated during I/O. all
672 * that is left for us to do is clean up page "p" (which is
673 * still PG_BUSY).
674 *
675 * our result could be one of the following:
676 * VM_PAGER_OK: successful pageout
677 *
678 * VM_PAGER_AGAIN: tmp resource shortage, we skip to next page
679 * VM_PAGER_{FAIL,ERROR,BAD}: an error. we "reactivate"
680 * page to get it out of the way (it will eventually
681 * drift back into the inactive queue for a retry).
682 * VM_PAGER_UNLOCK: should never see this as it is only
683 * valid for "get" operations
684 */
685
686 /* relock p's object: page queues not lock yet, so no need for "try" */
687 if (swap_backed) { /* !swap_backed case: already locked... */
688 if (anon)
689 simple_lock(&anon->an_lock);
690 else
691 simple_lock(&uobj->vmobjlock);
692 }
693
694 #ifdef DIAGNOSTIC
695 if (result == VM_PAGER_UNLOCK)
696 panic("pagedaemon: pageout returned invalid 'unlock' code");
697 #endif
698
699 /* handle PG_WANTED now */
700 if (p->flags & PG_WANTED)
701 thread_wakeup(p); /* still holding object lock */
702 p->flags &= ~(PG_BUSY|PG_WANTED);
703 UVM_PAGE_OWN(p, NULL);
704
705 /* released during I/O? */
706 if (p->flags & PG_RELEASED) {
707 if (anon) {
708 anon->u.an_page = NULL; /* remove page so we can get nextpg */
709 simple_unlock(&anon->an_lock);/* XXX needed? */
710 uvm_anfree(anon); /* kills anon */
711 pmap_page_protect(PMAP_PGARG(p), VM_PROT_NONE);
712 anon = NULL;
713 uvm_lock_pageq();
714 nextpg = p->pageq.tqe_next;
715 uvm_pagefree(p); /* free released page */
716
717 } else {
718
719 #ifdef DIAGNOSTIC
720 if (uobj->pgops->pgo_releasepg == NULL)
721 panic("pagedaemon: no pgo_releasepg function");
722 #endif
723
724 /*
725 * pgo_releasepg nukes the page and gets "nextpg" for us.
726 * it returns with the page queues locked (when given nextpg ptr).
727 */
728 if (!uobj->pgops->pgo_releasepg(p, &nextpg))
729 uobj = NULL; /* uobj died after release */
730
731 /*
732 * lock page queues here so that they're always locked
733 * at the end of the loop.
734 */
735 uvm_lock_pageq();
736 }
737
738 } else { /* page was not released during I/O */
739
740 uvm_lock_pageq();
741 nextpg = p->pageq.tqe_next;
742
743 if (result != VM_PAGER_OK) {
744
745 /* pageout was a failure... */
746 if (result != VM_PAGER_AGAIN)
747 uvm_pageactivate(p);
748 pmap_clear_reference(PMAP_PGARG(p));
749 /* XXXCDC: if (swap_backed) FREE p's swap block? */
750
751 } else {
752
753 /* pageout was a success... */
754 pmap_clear_reference(PMAP_PGARG(p));
755 pmap_clear_modify(PMAP_PGARG(p));
756 p->flags |= PG_CLEAN;
757 /* XXX: could free page here, but old pagedaemon does not */
758
759 }
760 }
761
762 /*
763 * drop object lock (if there is an object left). do a safety
764 * check of nextpg to make sure it is on the inactive queue
765 * (it should be since PG_BUSY pages on the inactive queue can't
766 * be re-queued [note: not true for active queue]).
767 */
768
769 if (anon)
770 simple_unlock(&anon->an_lock);
771 else if (uobj)
772 simple_unlock(&uobj->vmobjlock);
773
774 } /* if (p) */ else {
775
776 /* if p is null in this loop, make sure it stays null in next loop */
777 nextpg = NULL;
778
779 /*
780 * lock page queues here just so they're always locked
781 * at the end of the loop.
782 */
783 uvm_lock_pageq();
784 }
785
786 if (nextpg && (nextpg->pqflags & PQ_INACTIVE) == 0) {
787 printf("pagedaemon: invalid nextpg! reverting to queue head\n");
788 nextpg = pglst->tqh_first; /* reload! */
789 }
790
791 } /* end of "inactive" 'for' loop */
792 return(retval);
793 }
794
795 /*
796 * uvmpd_scan: scan the page queues and attempt to meet our targets.
797 *
798 * => called with pageq's locked
799 */
800
801 void uvmpd_scan()
802
803 {
804 int s, free, pages_freed, page_shortage;
805 struct vm_page *p, *nextpg;
806 struct uvm_object *uobj;
807 boolean_t got_it;
808 UVMHIST_FUNC("uvmpd_scan"); UVMHIST_CALLED(pdhist);
809
810 uvmexp.pdrevs++; /* counter */
811
812 #ifdef __GNUC__
813 uobj = NULL; /* XXX gcc */
814 #endif
815 /*
816 * get current "free" page count
817 */
818 s = splimp();
819 uvm_lock_fpageq();
820 free = uvmexp.free;
821 uvm_unlock_fpageq();
822 splx(s);
823
824 #ifndef __SWAP_BROKEN
825 /*
826 * swap out some processes if we are below our free target.
827 * we need to unlock the page queues for this.
828 */
829 if (free < uvmexp.freetarg) {
830
831 uvmexp.pdswout++;
832 UVMHIST_LOG(pdhist," free %d < target %d: swapout", free, uvmexp.freetarg,
833 0,0);
834 uvm_unlock_pageq();
835 uvm_swapout_threads();
836 pmap_update(); /* update so we can scan inactive q */
837 uvm_lock_pageq();
838
839 }
840 #endif
841
842 /*
843 * now we want to work on meeting our targets. first we work on our
844 * free target by converting inactive pages into free pages. then
845 * we work on meeting our inactive target by converting active pages
846 * to inactive ones.
847 */
848
849 UVMHIST_LOG(pdhist, " starting 'free' loop",0,0,0,0);
850 pages_freed = uvmexp.pdfreed; /* so far... */
851
852 /*
853 * do loop #1! alternate starting queue between swap and object based
854 * on the low bit of uvmexp.pdrevs (which we bump by one each call).
855 */
856
857 got_it = FALSE;
858 if ((uvmexp.pdrevs & 1) != 0 && uvmexp.nswapdev != 0)
859 got_it = uvmpd_scan_inactive(&uvm.page_inactive_swp);
860 if (!got_it)
861 got_it = uvmpd_scan_inactive(&uvm.page_inactive_obj);
862 if (!got_it && (uvmexp.pdrevs & 1) == 0 && uvmexp.nswapdev != 0)
863 (void) uvmpd_scan_inactive(&uvm.page_inactive_swp);
864
865 /*
866 * we have done the scan to get free pages. now we work on meeting
867 * our inactive target.
868 */
869
870 page_shortage = uvmexp.inactarg - uvmexp.inactive;
871 pages_freed = uvmexp.pdfreed - pages_freed; /* # pages freed in loop */
872 if (page_shortage <= 0 && pages_freed == 0)
873 page_shortage = 1;
874
875 UVMHIST_LOG(pdhist, " second loop: page_shortage=%d", page_shortage,0,0,0);
876 for (p = uvm.page_active.tqh_first ;
877 p != NULL && page_shortage > 0 ; p = nextpg) {
878
879 nextpg = p->pageq.tqe_next;
880 if (p->flags & PG_BUSY)
881 continue; /* quick check before trying to lock */
882
883 /*
884 * lock owner
885 */
886 /* is page anon owned or ownerless? */
887 if ((p->pqflags & PQ_ANON) || p->uobject == NULL) {
888
889 #ifdef DIAGNOSTIC
890 if (p->uanon == NULL)
891 panic("pagedaemon: page with no anon or object detected - loop 2");
892 #endif
893
894 if (!simple_lock_try(&p->uanon->an_lock))
895 continue;
896
897 /* take over the page? */
898 if ((p->pqflags & PQ_ANON) == 0) {
899
900 #ifdef DIAGNOSTIC
901 if (p->loan_count < 1)
902 panic("pagedaemon: non-loaned ownerless page detected - loop 2");
903 #endif
904
905 p->loan_count--;
906 p->pqflags |= PQ_ANON;
907 }
908
909 } else {
910
911 if (!simple_lock_try(&p->uobject->vmobjlock))
912 continue;
913
914 }
915
916 if ((p->flags & PG_BUSY) == 0) {
917 pmap_page_protect(PMAP_PGARG(p), VM_PROT_NONE);
918 /* no need to check wire_count as pg is "active" */
919 uvm_pagedeactivate(p);
920 uvmexp.pddeact++;
921 page_shortage--;
922 }
923
924 if (p->pqflags & PQ_ANON)
925 simple_unlock(&p->uanon->an_lock);
926 else
927 simple_unlock(&p->uobject->vmobjlock);
928 }
929
930 /*
931 * done scan
932 */
933 }
934