xref: /openbsd-src/sys/uvm/uvm_page.c (revision 2b0358df1d88d06ef4139321dd05bd5e05d91eaf)
1 /*	$OpenBSD: uvm_page.c,v 1.73 2009/04/06 17:03:51 oga Exp $	*/
2 /*	$NetBSD: uvm_page.c,v 1.44 2000/11/27 08:40:04 chs Exp $	*/
3 
4 /*
5  * Copyright (c) 1997 Charles D. Cranor and Washington University.
6  * Copyright (c) 1991, 1993, The Regents of the University of California.
7  *
8  * All rights reserved.
9  *
10  * This code is derived from software contributed to Berkeley by
11  * The Mach Operating System project at Carnegie-Mellon University.
12  *
13  * Redistribution and use in source and binary forms, with or without
14  * modification, are permitted provided that the following conditions
15  * are met:
16  * 1. Redistributions of source code must retain the above copyright
17  *    notice, this list of conditions and the following disclaimer.
18  * 2. Redistributions in binary form must reproduce the above copyright
19  *    notice, this list of conditions and the following disclaimer in the
20  *    documentation and/or other materials provided with the distribution.
21  * 3. All advertising materials mentioning features or use of this software
22  *    must display the following acknowledgement:
23  *	This product includes software developed by Charles D. Cranor,
24  *      Washington University, the University of California, Berkeley and
25  *      its contributors.
26  * 4. Neither the name of the University nor the names of its contributors
27  *    may be used to endorse or promote products derived from this software
28  *    without specific prior written permission.
29  *
30  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
31  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
32  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
33  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
34  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
35  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
36  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
37  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
38  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
39  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
40  * SUCH DAMAGE.
41  *
42  *	@(#)vm_page.c   8.3 (Berkeley) 3/21/94
43  * from: Id: uvm_page.c,v 1.1.2.18 1998/02/06 05:24:42 chs Exp
44  *
45  *
46  * Copyright (c) 1987, 1990 Carnegie-Mellon University.
47  * All rights reserved.
48  *
49  * Permission to use, copy, modify and distribute this software and
50  * its documentation is hereby granted, provided that both the copyright
51  * notice and this permission notice appear in all copies of the
52  * software, derivative works or modified versions, and any portions
53  * thereof, and that both notices appear in supporting documentation.
54  *
55  * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
56  * CONDITION.  CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
57  * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
58  *
59  * Carnegie Mellon requests users of this software to return to
60  *
61  *  Software Distribution Coordinator  or  Software.Distribution@CS.CMU.EDU
62  *  School of Computer Science
63  *  Carnegie Mellon University
64  *  Pittsburgh PA 15213-3890
65  *
66  * any improvements or extensions that they make and grant Carnegie the
67  * rights to redistribute these changes.
68  */
69 
70 /*
71  * uvm_page.c: page ops.
72  */
73 
74 #include <sys/param.h>
75 #include <sys/systm.h>
76 #include <sys/malloc.h>
77 #include <sys/sched.h>
78 #include <sys/kernel.h>
79 #include <sys/vnode.h>
80 #include <sys/mount.h>
81 
82 #include <uvm/uvm.h>
83 
84 /*
85  * global vars... XXXCDC: move to uvm. structure.
86  */
87 
88 /*
89  * physical memory config is stored in vm_physmem.
90  */
91 
92 struct vm_physseg vm_physmem[VM_PHYSSEG_MAX];	/* XXXCDC: uvm.physmem */
93 int vm_nphysseg = 0;				/* XXXCDC: uvm.nphysseg */
94 
95 /*
96  * Some supported CPUs in a given architecture don't support all
97  * of the things necessary to do idle page zero'ing efficiently.
98  * We therefore provide a way to disable it from machdep code here.
99  */
100 
101 /*
102  * XXX disabled until we can find a way to do this without causing
103  * problems for either cpu caches or DMA latency.
104  */
105 boolean_t vm_page_zero_enable = FALSE;
106 
107 /*
108  * local variables
109  */
110 
111 /*
112  * these variables record the values returned by vm_page_bootstrap,
113  * for debugging purposes.  The implementation of uvm_pageboot_alloc
114  * and pmap_startup here also uses them internally.
115  */
116 
117 static vaddr_t      virtual_space_start;
118 static vaddr_t      virtual_space_end;
119 
120 /*
121  * we use a hash table with only one bucket during bootup.  we will
122  * later rehash (resize) the hash table once the allocator is ready.
123  * we static allocate the one bootstrap bucket below...
124  */
125 
126 static struct pglist uvm_bootbucket;
127 
128 /*
129  * History
130  */
131 UVMHIST_DECL(pghist);
132 
133 /*
134  * local prototypes
135  */
136 
137 static void uvm_pageinsert(struct vm_page *);
138 static void uvm_pageremove(struct vm_page *);
139 
140 /*
141  * inline functions
142  */
143 
144 /*
145  * uvm_pageinsert: insert a page in the object and the hash table
146  *
147  * => caller must lock object
148  * => caller must lock page queues
149  * => call should have already set pg's object and offset pointers
150  *    and bumped the version counter
151  */
152 
153 __inline static void
154 uvm_pageinsert(struct vm_page *pg)
155 {
156 	struct pglist *buck;
157 	UVMHIST_FUNC("uvm_pageinsert"); UVMHIST_CALLED(pghist);
158 
159 	KASSERT((pg->pg_flags & PG_TABLED) == 0);
160 	mtx_enter(&uvm.hashlock);
161 	buck = &uvm.page_hash[uvm_pagehash(pg->uobject,pg->offset)];
162 	TAILQ_INSERT_TAIL(buck, pg, hashq);	/* put in hash */
163 	mtx_leave(&uvm.hashlock);
164 
165 	TAILQ_INSERT_TAIL(&pg->uobject->memq, pg, listq); /* put in object */
166 	atomic_setbits_int(&pg->pg_flags, PG_TABLED);
167 	pg->uobject->uo_npages++;
168 }
169 
170 /*
171  * uvm_page_remove: remove page from object and hash
172  *
173  * => caller must lock object
174  * => caller must lock page queues
175  */
176 
177 static __inline void
178 uvm_pageremove(struct vm_page *pg)
179 {
180 	struct pglist *buck;
181 	UVMHIST_FUNC("uvm_pageremove"); UVMHIST_CALLED(pghist);
182 
183 	KASSERT(pg->pg_flags & PG_TABLED);
184 	mtx_enter(&uvm.hashlock);
185 	buck = &uvm.page_hash[uvm_pagehash(pg->uobject,pg->offset)];
186 	TAILQ_REMOVE(buck, pg, hashq);
187 	mtx_leave(&uvm.hashlock);
188 
189 #ifdef UBC
190 	if (pg->uobject->pgops == &uvm_vnodeops) {
191 		uvm_pgcnt_vnode--;
192 	}
193 #endif
194 
195 	/* object should be locked */
196 	TAILQ_REMOVE(&pg->uobject->memq, pg, listq);
197 
198 	atomic_clearbits_int(&pg->pg_flags, PG_TABLED);
199 	pg->uobject->uo_npages--;
200 	pg->uobject = NULL;
201 	pg->pg_version++;
202 }
203 
204 /*
205  * uvm_page_init: init the page system.   called from uvm_init().
206  *
207  * => we return the range of kernel virtual memory in kvm_startp/kvm_endp
208  */
209 
210 void
211 uvm_page_init(vaddr_t *kvm_startp, vaddr_t *kvm_endp)
212 {
213 	vsize_t freepages, pagecount, n;
214 	vm_page_t pagearray;
215 	int lcv, i;
216 	paddr_t paddr;
217 #if defined(UVMHIST)
218 	static struct uvm_history_ent pghistbuf[100];
219 #endif
220 
221 	UVMHIST_FUNC("uvm_page_init");
222 	UVMHIST_INIT_STATIC(pghist, pghistbuf);
223 	UVMHIST_CALLED(pghist);
224 
225 	/*
226 	 * init the page queues and page queue locks
227 	 */
228 
229 	for (lcv = 0; lcv < VM_NFREELIST; lcv++) {
230 		for (i = 0; i < PGFL_NQUEUES; i++)
231 			TAILQ_INIT(&uvm.page_free[lcv].pgfl_queues[i]);
232 	}
233 	TAILQ_INIT(&uvm.page_active);
234 	TAILQ_INIT(&uvm.page_inactive_swp);
235 	TAILQ_INIT(&uvm.page_inactive_obj);
236 	simple_lock_init(&uvm.pageqlock);
237 	mtx_init(&uvm.fpageqlock, IPL_VM);
238 
239 	/*
240 	 * init the <obj,offset> => <page> hash table.  for now
241 	 * we just have one bucket (the bootstrap bucket).  later on we
242 	 * will allocate new buckets as we dynamically resize the hash table.
243 	 */
244 
245 	uvm.page_nhash = 1;			/* 1 bucket */
246 	uvm.page_hashmask = 0;			/* mask for hash function */
247 	uvm.page_hash = &uvm_bootbucket;	/* install bootstrap bucket */
248 	TAILQ_INIT(uvm.page_hash);		/* init hash table */
249 	mtx_init(&uvm.hashlock, IPL_VM);	/* init hash table lock */
250 
251 	/*
252 	 * allocate vm_page structures.
253 	 */
254 
255 	/*
256 	 * sanity check:
257 	 * before calling this function the MD code is expected to register
258 	 * some free RAM with the uvm_page_physload() function.   our job
259 	 * now is to allocate vm_page structures for this memory.
260 	 */
261 
262 	if (vm_nphysseg == 0)
263 		panic("uvm_page_bootstrap: no memory pre-allocated");
264 
265 	/*
266 	 * first calculate the number of free pages...
267 	 *
268 	 * note that we use start/end rather than avail_start/avail_end.
269 	 * this allows us to allocate extra vm_page structures in case we
270 	 * want to return some memory to the pool after booting.
271 	 */
272 
273 	freepages = 0;
274 	for (lcv = 0 ; lcv < vm_nphysseg ; lcv++)
275 		freepages += (vm_physmem[lcv].end - vm_physmem[lcv].start);
276 
277 	/*
278 	 * we now know we have (PAGE_SIZE * freepages) bytes of memory we can
279 	 * use.   for each page of memory we use we need a vm_page structure.
280 	 * thus, the total number of pages we can use is the total size of
281 	 * the memory divided by the PAGE_SIZE plus the size of the vm_page
282 	 * structure.   we add one to freepages as a fudge factor to avoid
283 	 * truncation errors (since we can only allocate in terms of whole
284 	 * pages).
285 	 */
286 
287 	pagecount = (((paddr_t)freepages + 1) << PAGE_SHIFT) /
288 	    (PAGE_SIZE + sizeof(struct vm_page));
289 	pagearray = (vm_page_t)uvm_pageboot_alloc(pagecount *
290 	    sizeof(struct vm_page));
291 	memset(pagearray, 0, pagecount * sizeof(struct vm_page));
292 
293 	/*
294 	 * init the vm_page structures and put them in the correct place.
295 	 */
296 
297 	for (lcv = 0 ; lcv < vm_nphysseg ; lcv++) {
298 		n = vm_physmem[lcv].end - vm_physmem[lcv].start;
299 		if (n > pagecount) {
300 			printf("uvm_page_init: lost %ld page(s) in init\n",
301 			    (long)(n - pagecount));
302 			panic("uvm_page_init");  /* XXXCDC: shouldn't happen? */
303 			/* n = pagecount; */
304 		}
305 
306 		/* set up page array pointers */
307 		vm_physmem[lcv].pgs = pagearray;
308 		pagearray += n;
309 		pagecount -= n;
310 		vm_physmem[lcv].lastpg = vm_physmem[lcv].pgs + (n - 1);
311 
312 		/* init and free vm_pages (we've already zeroed them) */
313 		paddr = ptoa(vm_physmem[lcv].start);
314 		for (i = 0 ; i < n ; i++, paddr += PAGE_SIZE) {
315 			vm_physmem[lcv].pgs[i].phys_addr = paddr;
316 #ifdef __HAVE_VM_PAGE_MD
317 			VM_MDPAGE_INIT(&vm_physmem[lcv].pgs[i]);
318 #endif
319 			if (atop(paddr) >= vm_physmem[lcv].avail_start &&
320 			    atop(paddr) <= vm_physmem[lcv].avail_end) {
321 				uvmexp.npages++;
322 				/* add page to free pool */
323 				uvm_pagefree(&vm_physmem[lcv].pgs[i]);
324 			}
325 		}
326 	}
327 
328 	/*
329 	 * pass up the values of virtual_space_start and
330 	 * virtual_space_end (obtained by uvm_pageboot_alloc) to the upper
331 	 * layers of the VM.
332 	 */
333 
334 	*kvm_startp = round_page(virtual_space_start);
335 	*kvm_endp = trunc_page(virtual_space_end);
336 
337 	/*
338 	 * init locks for kernel threads
339 	 */
340 
341 	simple_lock_init(&uvm.pagedaemon_lock);
342 	mtx_init(&uvm.aiodoned_lock, IPL_BIO);
343 
344 	/*
345 	 * init reserve thresholds
346 	 * XXXCDC - values may need adjusting
347 	 */
348 	uvmexp.reserve_pagedaemon = 4;
349 	uvmexp.reserve_kernel = 6;
350 	uvmexp.anonminpct = 10;
351 	uvmexp.vnodeminpct = 10;
352 	uvmexp.vtextminpct = 5;
353 	uvmexp.anonmin = uvmexp.anonminpct * 256 / 100;
354 	uvmexp.vnodemin = uvmexp.vnodeminpct * 256 / 100;
355 	uvmexp.vtextmin = uvmexp.vtextminpct * 256 / 100;
356 
357   	/*
358 	 * determine if we should zero pages in the idle loop.
359 	 */
360 
361 	uvm.page_idle_zero = vm_page_zero_enable;
362 
363 	/*
364 	 * done!
365 	 */
366 
367 	uvm.page_init_done = TRUE;
368 }
369 
370 /*
371  * uvm_setpagesize: set the page size
372  *
373  * => sets page_shift and page_mask from uvmexp.pagesize.
374  */
375 
376 void
377 uvm_setpagesize(void)
378 {
379 	if (uvmexp.pagesize == 0)
380 		uvmexp.pagesize = DEFAULT_PAGE_SIZE;
381 	uvmexp.pagemask = uvmexp.pagesize - 1;
382 	if ((uvmexp.pagemask & uvmexp.pagesize) != 0)
383 		panic("uvm_setpagesize: page size not a power of two");
384 	for (uvmexp.pageshift = 0; ; uvmexp.pageshift++)
385 		if ((1 << uvmexp.pageshift) == uvmexp.pagesize)
386 			break;
387 }
388 
389 /*
390  * uvm_pageboot_alloc: steal memory from physmem for bootstrapping
391  */
392 
393 vaddr_t
394 uvm_pageboot_alloc(vsize_t size)
395 {
396 #if defined(PMAP_STEAL_MEMORY)
397 	vaddr_t addr;
398 
399 	/*
400 	 * defer bootstrap allocation to MD code (it may want to allocate
401 	 * from a direct-mapped segment).  pmap_steal_memory should round
402 	 * off virtual_space_start/virtual_space_end.
403 	 */
404 
405 	addr = pmap_steal_memory(size, &virtual_space_start,
406 	    &virtual_space_end);
407 
408 	return(addr);
409 
410 #else /* !PMAP_STEAL_MEMORY */
411 
412 	static boolean_t initialized = FALSE;
413 	vaddr_t addr, vaddr;
414 	paddr_t paddr;
415 
416 	/* round to page size */
417 	size = round_page(size);
418 
419 	/*
420 	 * on first call to this function, initialize ourselves.
421 	 */
422 	if (initialized == FALSE) {
423 		pmap_virtual_space(&virtual_space_start, &virtual_space_end);
424 
425 		/* round it the way we like it */
426 		virtual_space_start = round_page(virtual_space_start);
427 		virtual_space_end = trunc_page(virtual_space_end);
428 
429 		initialized = TRUE;
430 	}
431 
432 	/*
433 	 * allocate virtual memory for this request
434 	 */
435 	if (virtual_space_start == virtual_space_end ||
436 	    (virtual_space_end - virtual_space_start) < size)
437 		panic("uvm_pageboot_alloc: out of virtual space");
438 
439 	addr = virtual_space_start;
440 
441 #ifdef PMAP_GROWKERNEL
442 	/*
443 	 * If the kernel pmap can't map the requested space,
444 	 * then allocate more resources for it.
445 	 */
446 	if (uvm_maxkaddr < (addr + size)) {
447 		uvm_maxkaddr = pmap_growkernel(addr + size);
448 		if (uvm_maxkaddr < (addr + size))
449 			panic("uvm_pageboot_alloc: pmap_growkernel() failed");
450 	}
451 #endif
452 
453 	virtual_space_start += size;
454 
455 	/*
456 	 * allocate and mapin physical pages to back new virtual pages
457 	 */
458 
459 	for (vaddr = round_page(addr) ; vaddr < addr + size ;
460 	    vaddr += PAGE_SIZE) {
461 
462 		if (!uvm_page_physget(&paddr))
463 			panic("uvm_pageboot_alloc: out of memory");
464 
465 		/*
466 		 * Note this memory is no longer managed, so using
467 		 * pmap_kenter is safe.
468 		 */
469 		pmap_kenter_pa(vaddr, paddr, VM_PROT_READ|VM_PROT_WRITE);
470 	}
471 	pmap_update(pmap_kernel());
472 	return(addr);
473 #endif	/* PMAP_STEAL_MEMORY */
474 }
475 
476 #if !defined(PMAP_STEAL_MEMORY)
477 /*
478  * uvm_page_physget: "steal" one page from the vm_physmem structure.
479  *
480  * => attempt to allocate it off the end of a segment in which the "avail"
481  *    values match the start/end values.   if we can't do that, then we
482  *    will advance both values (making them equal, and removing some
483  *    vm_page structures from the non-avail area).
484  * => return false if out of memory.
485  */
486 
487 /* subroutine: try to allocate from memory chunks on the specified freelist */
488 static boolean_t uvm_page_physget_freelist(paddr_t *, int);
489 
490 static boolean_t
491 uvm_page_physget_freelist(paddr_t *paddrp, int freelist)
492 {
493 	int lcv, x;
494 	UVMHIST_FUNC("uvm_page_physget_freelist"); UVMHIST_CALLED(pghist);
495 
496 	/* pass 1: try allocating from a matching end */
497 #if (VM_PHYSSEG_STRAT == VM_PSTRAT_BIGFIRST) || \
498 	(VM_PHYSSEG_STRAT == VM_PSTRAT_BSEARCH)
499 	for (lcv = vm_nphysseg - 1 ; lcv >= 0 ; lcv--)
500 #else
501 	for (lcv = 0 ; lcv < vm_nphysseg ; lcv++)
502 #endif
503 	{
504 
505 		if (uvm.page_init_done == TRUE)
506 			panic("uvm_page_physget: called _after_ bootstrap");
507 
508 		if (vm_physmem[lcv].free_list != freelist)
509 			continue;
510 
511 		/* try from front */
512 		if (vm_physmem[lcv].avail_start == vm_physmem[lcv].start &&
513 		    vm_physmem[lcv].avail_start < vm_physmem[lcv].avail_end) {
514 			*paddrp = ptoa(vm_physmem[lcv].avail_start);
515 			vm_physmem[lcv].avail_start++;
516 			vm_physmem[lcv].start++;
517 			/* nothing left?   nuke it */
518 			if (vm_physmem[lcv].avail_start ==
519 			    vm_physmem[lcv].end) {
520 				if (vm_nphysseg == 1)
521 				    panic("uvm_page_physget: out of memory!");
522 				vm_nphysseg--;
523 				for (x = lcv ; x < vm_nphysseg ; x++)
524 					/* structure copy */
525 					vm_physmem[x] = vm_physmem[x+1];
526 			}
527 			return (TRUE);
528 		}
529 
530 		/* try from rear */
531 		if (vm_physmem[lcv].avail_end == vm_physmem[lcv].end &&
532 		    vm_physmem[lcv].avail_start < vm_physmem[lcv].avail_end) {
533 			*paddrp = ptoa(vm_physmem[lcv].avail_end - 1);
534 			vm_physmem[lcv].avail_end--;
535 			vm_physmem[lcv].end--;
536 			/* nothing left?   nuke it */
537 			if (vm_physmem[lcv].avail_end ==
538 			    vm_physmem[lcv].start) {
539 				if (vm_nphysseg == 1)
540 				    panic("uvm_page_physget: out of memory!");
541 				vm_nphysseg--;
542 				for (x = lcv ; x < vm_nphysseg ; x++)
543 					/* structure copy */
544 					vm_physmem[x] = vm_physmem[x+1];
545 			}
546 			return (TRUE);
547 		}
548 	}
549 
550 	/* pass2: forget about matching ends, just allocate something */
551 #if (VM_PHYSSEG_STRAT == VM_PSTRAT_BIGFIRST) || \
552 	(VM_PHYSSEG_STRAT == VM_PSTRAT_BSEARCH)
553 	for (lcv = vm_nphysseg - 1 ; lcv >= 0 ; lcv--)
554 #else
555 	for (lcv = 0 ; lcv < vm_nphysseg ; lcv++)
556 #endif
557 	{
558 
559 		/* any room in this bank? */
560 		if (vm_physmem[lcv].avail_start >= vm_physmem[lcv].avail_end)
561 			continue;  /* nope */
562 
563 		*paddrp = ptoa(vm_physmem[lcv].avail_start);
564 		vm_physmem[lcv].avail_start++;
565 		/* truncate! */
566 		vm_physmem[lcv].start = vm_physmem[lcv].avail_start;
567 
568 		/* nothing left?   nuke it */
569 		if (vm_physmem[lcv].avail_start == vm_physmem[lcv].end) {
570 			if (vm_nphysseg == 1)
571 				panic("uvm_page_physget: out of memory!");
572 			vm_nphysseg--;
573 			for (x = lcv ; x < vm_nphysseg ; x++)
574 				/* structure copy */
575 				vm_physmem[x] = vm_physmem[x+1];
576 		}
577 		return (TRUE);
578 	}
579 
580 	return (FALSE);        /* whoops! */
581 }
582 
583 boolean_t
584 uvm_page_physget(paddr_t *paddrp)
585 {
586 	int i;
587 	UVMHIST_FUNC("uvm_page_physget"); UVMHIST_CALLED(pghist);
588 
589 	/* try in the order of freelist preference */
590 	for (i = 0; i < VM_NFREELIST; i++)
591 		if (uvm_page_physget_freelist(paddrp, i) == TRUE)
592 			return (TRUE);
593 	return (FALSE);
594 }
595 #endif /* PMAP_STEAL_MEMORY */
596 
597 /*
598  * uvm_page_physload: load physical memory into VM system
599  *
600  * => all args are PFs
601  * => all pages in start/end get vm_page structures
602  * => areas marked by avail_start/avail_end get added to the free page pool
603  * => we are limited to VM_PHYSSEG_MAX physical memory segments
604  */
605 
606 void
607 uvm_page_physload(paddr_t start, paddr_t end, paddr_t avail_start,
608     paddr_t avail_end, int free_list)
609 {
610 	int preload, lcv;
611 	psize_t npages;
612 	struct vm_page *pgs;
613 	struct vm_physseg *ps;
614 
615 	if (uvmexp.pagesize == 0)
616 		panic("uvm_page_physload: page size not set!");
617 
618 	if (free_list >= VM_NFREELIST || free_list < VM_FREELIST_DEFAULT)
619 		panic("uvm_page_physload: bad free list %d", free_list);
620 
621 	if (start >= end)
622 		panic("uvm_page_physload: start >= end");
623 
624 	/*
625 	 * do we have room?
626 	 */
627 	if (vm_nphysseg == VM_PHYSSEG_MAX) {
628 		printf("uvm_page_physload: unable to load physical memory "
629 		    "segment\n");
630 		printf("\t%d segments allocated, ignoring 0x%llx -> 0x%llx\n",
631 		    VM_PHYSSEG_MAX, (long long)start, (long long)end);
632 		printf("\tincrease VM_PHYSSEG_MAX\n");
633 		return;
634 	}
635 
636 	/*
637 	 * check to see if this is a "preload" (i.e. uvm_mem_init hasn't been
638 	 * called yet, so malloc is not available).
639 	 */
640 	for (lcv = 0 ; lcv < vm_nphysseg ; lcv++) {
641 		if (vm_physmem[lcv].pgs)
642 			break;
643 	}
644 	preload = (lcv == vm_nphysseg);
645 
646 	/*
647 	 * if VM is already running, attempt to malloc() vm_page structures
648 	 */
649 	if (!preload) {
650 #if defined(VM_PHYSSEG_NOADD)
651 		panic("uvm_page_physload: tried to add RAM after vm_mem_init");
652 #else
653 		/* XXXCDC: need some sort of lockout for this case */
654 		paddr_t paddr;
655 		npages = end - start;  /* # of pages */
656 		pgs = (vm_page *)uvm_km_alloc(kernel_map,
657 		    sizeof(struct vm_page) * npages);
658 		if (pgs == NULL) {
659 			printf("uvm_page_physload: can not malloc vm_page "
660 			    "structs for segment\n");
661 			printf("\tignoring 0x%lx -> 0x%lx\n", start, end);
662 			return;
663 		}
664 		/* zero data, init phys_addr and free_list, and free pages */
665 		memset(pgs, 0, sizeof(struct vm_page) * npages);
666 		for (lcv = 0, paddr = ptoa(start) ;
667 				 lcv < npages ; lcv++, paddr += PAGE_SIZE) {
668 			pgs[lcv].phys_addr = paddr;
669 			pgs[lcv].free_list = free_list;
670 			if (atop(paddr) >= avail_start &&
671 			    atop(paddr) <= avail_end)
672 				uvm_pagefree(&pgs[lcv]);
673 		}
674 		/* XXXCDC: incomplete: need to update uvmexp.free, what else? */
675 		/* XXXCDC: need hook to tell pmap to rebuild pv_list, etc... */
676 #endif
677 	} else {
678 
679 		/* gcc complains if these don't get init'd */
680 		pgs = NULL;
681 		npages = 0;
682 
683 	}
684 
685 	/*
686 	 * now insert us in the proper place in vm_physmem[]
687 	 */
688 
689 #if (VM_PHYSSEG_STRAT == VM_PSTRAT_RANDOM)
690 
691 	/* random: put it at the end (easy!) */
692 	ps = &vm_physmem[vm_nphysseg];
693 
694 #elif (VM_PHYSSEG_STRAT == VM_PSTRAT_BSEARCH)
695 
696 	{
697 		int x;
698 		/* sort by address for binary search */
699 		for (lcv = 0 ; lcv < vm_nphysseg ; lcv++)
700 			if (start < vm_physmem[lcv].start)
701 				break;
702 		ps = &vm_physmem[lcv];
703 		/* move back other entries, if necessary ... */
704 		for (x = vm_nphysseg ; x > lcv ; x--)
705 			/* structure copy */
706 			vm_physmem[x] = vm_physmem[x - 1];
707 	}
708 
709 #elif (VM_PHYSSEG_STRAT == VM_PSTRAT_BIGFIRST)
710 
711 	{
712 		int x;
713 		/* sort by largest segment first */
714 		for (lcv = 0 ; lcv < vm_nphysseg ; lcv++)
715 			if ((end - start) >
716 			    (vm_physmem[lcv].end - vm_physmem[lcv].start))
717 				break;
718 		ps = &vm_physmem[lcv];
719 		/* move back other entries, if necessary ... */
720 		for (x = vm_nphysseg ; x > lcv ; x--)
721 			/* structure copy */
722 			vm_physmem[x] = vm_physmem[x - 1];
723 	}
724 
725 #else
726 
727 	panic("uvm_page_physload: unknown physseg strategy selected!");
728 
729 #endif
730 
731 	ps->start = start;
732 	ps->end = end;
733 	ps->avail_start = avail_start;
734 	ps->avail_end = avail_end;
735 	if (preload) {
736 		ps->pgs = NULL;
737 	} else {
738 		ps->pgs = pgs;
739 		ps->lastpg = pgs + npages - 1;
740 	}
741 	ps->free_list = free_list;
742 	vm_nphysseg++;
743 
744 	/*
745 	 * done!
746 	 */
747 
748 	if (!preload)
749 		uvm_page_rehash();
750 
751 	return;
752 }
753 
754 /*
755  * uvm_page_rehash: reallocate hash table based on number of free pages.
756  */
757 
758 void
759 uvm_page_rehash(void)
760 {
761 	int freepages, lcv, bucketcount, oldcount;
762 	struct pglist *newbuckets, *oldbuckets;
763 	struct vm_page *pg;
764 	size_t newsize, oldsize;
765 
766 	/*
767 	 * compute number of pages that can go in the free pool
768 	 */
769 
770 	freepages = 0;
771 	for (lcv = 0 ; lcv < vm_nphysseg ; lcv++)
772 		freepages +=
773 		    (vm_physmem[lcv].avail_end - vm_physmem[lcv].avail_start);
774 
775 	/*
776 	 * compute number of buckets needed for this number of pages
777 	 */
778 
779 	bucketcount = 1;
780 	while (bucketcount < freepages)
781 		bucketcount = bucketcount * 2;
782 
783 	/*
784 	 * compute the size of the current table and new table.
785 	 */
786 
787 	oldbuckets = uvm.page_hash;
788 	oldcount = uvm.page_nhash;
789 	oldsize = round_page(sizeof(struct pglist) * oldcount);
790 	newsize = round_page(sizeof(struct pglist) * bucketcount);
791 
792 	/*
793 	 * allocate the new buckets
794 	 */
795 
796 	newbuckets = (struct pglist *) uvm_km_alloc(kernel_map, newsize);
797 	if (newbuckets == NULL) {
798 		printf("uvm_page_physrehash: WARNING: could not grow page "
799 		    "hash table\n");
800 		return;
801 	}
802 	for (lcv = 0 ; lcv < bucketcount ; lcv++)
803 		TAILQ_INIT(&newbuckets[lcv]);
804 
805 	/*
806 	 * now replace the old buckets with the new ones and rehash everything
807 	 */
808 
809 	mtx_enter(&uvm.hashlock);
810 	uvm.page_hash = newbuckets;
811 	uvm.page_nhash = bucketcount;
812 	uvm.page_hashmask = bucketcount - 1;  /* power of 2 */
813 
814 	/* ... and rehash */
815 	for (lcv = 0 ; lcv < oldcount ; lcv++) {
816 		while ((pg = TAILQ_FIRST(&oldbuckets[lcv])) != NULL) {
817 			TAILQ_REMOVE(&oldbuckets[lcv], pg, hashq);
818 			TAILQ_INSERT_TAIL(
819 			  &uvm.page_hash[uvm_pagehash(pg->uobject, pg->offset)],
820 			  pg, hashq);
821 		}
822 	}
823 	mtx_leave(&uvm.hashlock);
824 
825 	/*
826 	 * free old bucket array if is not the boot-time table
827 	 */
828 
829 	if (oldbuckets != &uvm_bootbucket)
830 		uvm_km_free(kernel_map, (vaddr_t) oldbuckets, oldsize);
831 
832 	/*
833 	 * done
834 	 */
835 	return;
836 }
837 
838 
839 #ifdef DDB /* XXXCDC: TMP TMP TMP DEBUG DEBUG DEBUG */
840 
841 void uvm_page_physdump(void); /* SHUT UP GCC */
842 
843 /* call from DDB */
844 void
845 uvm_page_physdump(void)
846 {
847 	int lcv;
848 
849 	printf("rehash: physical memory config [segs=%d of %d]:\n",
850 				 vm_nphysseg, VM_PHYSSEG_MAX);
851 	for (lcv = 0 ; lcv < vm_nphysseg ; lcv++)
852 		printf("0x%llx->0x%llx [0x%llx->0x%llx]\n",
853 		    (long long)vm_physmem[lcv].start,
854 		    (long long)vm_physmem[lcv].end,
855 		    (long long)vm_physmem[lcv].avail_start,
856 		    (long long)vm_physmem[lcv].avail_end);
857 	printf("STRATEGY = ");
858 	switch (VM_PHYSSEG_STRAT) {
859 	case VM_PSTRAT_RANDOM: printf("RANDOM\n"); break;
860 	case VM_PSTRAT_BSEARCH: printf("BSEARCH\n"); break;
861 	case VM_PSTRAT_BIGFIRST: printf("BIGFIRST\n"); break;
862 	default: printf("<<UNKNOWN>>!!!!\n");
863 	}
864 	printf("number of buckets = %d\n", uvm.page_nhash);
865 }
866 #endif
867 
868 void
869 uvm_shutdown(void)
870 {
871 #ifdef UVM_SWAP_ENCRYPT
872 	uvm_swap_finicrypt_all();
873 #endif
874 }
875 
876 /*
877  * uvm_pagealloc_strat: allocate vm_page from a particular free list.
878  *
879  * => return null if no pages free
880  * => wake up pagedaemon if number of free pages drops below low water mark
881  * => if obj != NULL, obj must be locked (to put in hash)
882  * => if anon != NULL, anon must be locked (to put in anon)
883  * => only one of obj or anon can be non-null
884  * => caller must activate/deactivate page if it is not wired.
885  * => free_list is ignored if strat == UVM_PGA_STRAT_NORMAL.
886  * => policy decision: it is more important to pull a page off of the
887  *	appropriate priority free list than it is to get a zero'd or
888  *	unknown contents page.  This is because we live with the
889  *	consequences of a bad free list decision for the entire
890  *	lifetime of the page, e.g. if the page comes from memory that
891  *	is slower to access.
892  */
893 
894 struct vm_page *
895 uvm_pagealloc_strat(struct uvm_object *obj, voff_t off, struct vm_anon *anon,
896     int flags, int strat, int free_list)
897 {
898 	int lcv, try1, try2, zeroit = 0;
899 	struct vm_page *pg;
900 	struct pglist *freeq;
901 	struct pgfreelist *pgfl;
902 	boolean_t use_reserve;
903 	UVMHIST_FUNC("uvm_pagealloc_strat"); UVMHIST_CALLED(pghist);
904 
905 	KASSERT(obj == NULL || anon == NULL);
906 	KASSERT(off == trunc_page(off));
907 
908 	uvm_lock_fpageq();
909 
910 	/*
911 	 * check to see if we need to generate some free pages waking
912 	 * the pagedaemon.
913 	 */
914 	if ((uvmexp.free - BUFPAGES_DEFICIT) < uvmexp.freemin ||
915 	    ((uvmexp.free - BUFPAGES_DEFICIT) < uvmexp.freetarg &&
916 	     uvmexp.inactive < uvmexp.inactarg))
917 		wakeup(&uvm.pagedaemon);
918 
919 	/*
920 	 * fail if any of these conditions is true:
921 	 * [1]  there really are no free pages, or
922 	 * [2]  only kernel "reserved" pages remain and
923 	 *        the page isn't being allocated to a kernel object.
924 	 * [3]  only pagedaemon "reserved" pages remain and
925 	 *        the requestor isn't the pagedaemon.
926 	 */
927 
928 	use_reserve = (flags & UVM_PGA_USERESERVE) ||
929 		(obj && UVM_OBJ_IS_KERN_OBJECT(obj));
930 	if ((uvmexp.free <= uvmexp.reserve_kernel && !use_reserve) ||
931 	    (uvmexp.free <= uvmexp.reserve_pagedaemon &&
932 	     !((curproc == uvm.pagedaemon_proc) ||
933 	      (curproc == syncerproc))))
934 		goto fail;
935 
936 #if PGFL_NQUEUES != 2
937 #error uvm_pagealloc_strat needs to be updated
938 #endif
939 
940 	/*
941 	 * If we want a zero'd page, try the ZEROS queue first, otherwise
942 	 * we try the UNKNOWN queue first.
943 	 */
944 	if (flags & UVM_PGA_ZERO) {
945 		try1 = PGFL_ZEROS;
946 		try2 = PGFL_UNKNOWN;
947 	} else {
948 		try1 = PGFL_UNKNOWN;
949 		try2 = PGFL_ZEROS;
950 	}
951 
952 	UVMHIST_LOG(pghist, "obj=%p off=%lx anon=%p flags=%lx",
953 	    obj, (u_long)off, anon, flags);
954 	UVMHIST_LOG(pghist, "strat=%ld free_list=%ld", strat, free_list, 0, 0);
955  again:
956 	switch (strat) {
957 	case UVM_PGA_STRAT_NORMAL:
958 		/* Check all freelists in descending priority order. */
959 		for (lcv = 0; lcv < VM_NFREELIST; lcv++) {
960 			pgfl = &uvm.page_free[lcv];
961 			if ((pg = TAILQ_FIRST((freeq =
962 			      &pgfl->pgfl_queues[try1]))) != NULL ||
963 			    (pg = TAILQ_FIRST((freeq =
964 			      &pgfl->pgfl_queues[try2]))) != NULL)
965 				goto gotit;
966 		}
967 
968 		/* No pages free! */
969 		goto fail;
970 
971 	case UVM_PGA_STRAT_ONLY:
972 	case UVM_PGA_STRAT_FALLBACK:
973 		/* Attempt to allocate from the specified free list. */
974 		KASSERT(free_list >= 0 && free_list < VM_NFREELIST);
975 		pgfl = &uvm.page_free[free_list];
976 		if ((pg = TAILQ_FIRST((freeq =
977 		      &pgfl->pgfl_queues[try1]))) != NULL ||
978 		    (pg = TAILQ_FIRST((freeq =
979 		      &pgfl->pgfl_queues[try2]))) != NULL)
980 			goto gotit;
981 
982 		/* Fall back, if possible. */
983 		if (strat == UVM_PGA_STRAT_FALLBACK) {
984 			strat = UVM_PGA_STRAT_NORMAL;
985 			goto again;
986 		}
987 
988 		/* No pages free! */
989 		goto fail;
990 
991 	default:
992 		panic("uvm_pagealloc_strat: bad strat %d", strat);
993 		/* NOTREACHED */
994 	}
995 
996  gotit:
997 	TAILQ_REMOVE(freeq, pg, pageq);
998 	uvmexp.free--;
999 
1000 	/* update zero'd page count */
1001 	if (pg->pg_flags & PG_ZERO)
1002 		uvmexp.zeropages--;
1003 
1004 	/*
1005 	 * update allocation statistics and remember if we have to
1006 	 * zero the page
1007 	 */
1008 	if (flags & UVM_PGA_ZERO) {
1009 		if (pg->pg_flags & PG_ZERO) {
1010 			uvmexp.pga_zerohit++;
1011 			zeroit = 0;
1012 		} else {
1013 			uvmexp.pga_zeromiss++;
1014 			zeroit = 1;
1015 		}
1016 	}
1017 
1018 	uvm_unlock_fpageq();		/* unlock free page queue */
1019 
1020 	pg->offset = off;
1021 	pg->uobject = obj;
1022 	pg->uanon = anon;
1023 	pg->pg_flags = PG_BUSY|PG_CLEAN|PG_FAKE;
1024 	pg->pg_version++;
1025 	if (anon) {
1026 		anon->an_page = pg;
1027 		atomic_setbits_int(&pg->pg_flags, PQ_ANON);
1028 #ifdef UBC
1029 		uvm_pgcnt_anon++;
1030 #endif
1031 	} else {
1032 		if (obj)
1033 			uvm_pageinsert(pg);
1034 	}
1035 #if defined(UVM_PAGE_TRKOWN)
1036 	pg->owner_tag = NULL;
1037 #endif
1038 	UVM_PAGE_OWN(pg, "new alloc");
1039 
1040 	if (flags & UVM_PGA_ZERO) {
1041 		/*
1042 		 * A zero'd page is not clean.  If we got a page not already
1043 		 * zero'd, then we have to zero it ourselves.
1044 		 */
1045 		atomic_clearbits_int(&pg->pg_flags, PG_CLEAN);
1046 		if (zeroit)
1047 			pmap_zero_page(pg);
1048 	}
1049 
1050 	UVMHIST_LOG(pghist, "allocated pg %p/%lx", pg,
1051 	    (u_long)VM_PAGE_TO_PHYS(pg), 0, 0);
1052 	return(pg);
1053 
1054  fail:
1055 	uvm_unlock_fpageq();
1056 	UVMHIST_LOG(pghist, "failed!", 0, 0, 0, 0);
1057 	return (NULL);
1058 }
1059 
1060 /*
1061  * uvm_pagerealloc: reallocate a page from one object to another
1062  *
1063  * => both objects must be locked
1064  */
1065 
1066 void
1067 uvm_pagerealloc(struct vm_page *pg, struct uvm_object *newobj, voff_t newoff)
1068 {
1069 
1070 	UVMHIST_FUNC("uvm_pagerealloc"); UVMHIST_CALLED(pghist);
1071 
1072 	/*
1073 	 * remove it from the old object
1074 	 */
1075 
1076 	if (pg->uobject) {
1077 		uvm_pageremove(pg);
1078 	}
1079 
1080 	/*
1081 	 * put it in the new object
1082 	 */
1083 
1084 	if (newobj) {
1085 		pg->uobject = newobj;
1086 		pg->offset = newoff;
1087 		pg->pg_version++;
1088 		uvm_pageinsert(pg);
1089 	}
1090 }
1091 
1092 
1093 /*
1094  * uvm_pagefree: free page
1095  *
1096  * => erase page's identity (i.e. remove from hash/object)
1097  * => put page on free list
1098  * => caller must lock owning object (either anon or uvm_object)
1099  * => caller must lock page queues
1100  * => assumes all valid mappings of pg are gone
1101  */
1102 
1103 void
1104 uvm_pagefree(struct vm_page *pg)
1105 {
1106 	int saved_loan_count = pg->loan_count;
1107 	UVMHIST_FUNC("uvm_pagefree"); UVMHIST_CALLED(pghist);
1108 
1109 #ifdef DEBUG
1110 	if (pg->uobject == (void *)0xdeadbeef &&
1111 	    pg->uanon == (void *)0xdeadbeef) {
1112 		panic("uvm_pagefree: freeing free page %p", pg);
1113 	}
1114 #endif
1115 
1116 	UVMHIST_LOG(pghist, "freeing pg %p/%lx", pg,
1117 	    (u_long)VM_PAGE_TO_PHYS(pg), 0, 0);
1118 
1119 	/*
1120 	 * if the page was an object page (and thus "TABLED"), remove it
1121 	 * from the object.
1122 	 */
1123 
1124 	if (pg->pg_flags & PG_TABLED) {
1125 
1126 		/*
1127 		 * if the object page is on loan we are going to drop ownership.
1128 		 * it is possible that an anon will take over as owner for this
1129 		 * page later on.   the anon will want a !PG_CLEAN page so that
1130 		 * it knows it needs to allocate swap if it wants to page the
1131 		 * page out.
1132 		 */
1133 
1134 		/* in case an anon takes over */
1135 		if (saved_loan_count)
1136 			atomic_clearbits_int(&pg->pg_flags, PG_CLEAN);
1137 		uvm_pageremove(pg);
1138 
1139 		/*
1140 		 * if our page was on loan, then we just lost control over it
1141 		 * (in fact, if it was loaned to an anon, the anon may have
1142 		 * already taken over ownership of the page by now and thus
1143 		 * changed the loan_count [e.g. in uvmfault_anonget()]) we just
1144 		 * return (when the last loan is dropped, then the page can be
1145 		 * freed by whatever was holding the last loan).
1146 		 */
1147 
1148 		if (saved_loan_count)
1149 			return;
1150 	} else if (saved_loan_count && pg->uanon) {
1151 		/*
1152 		 * if our page is owned by an anon and is loaned out to the
1153 		 * kernel then we just want to drop ownership and return.
1154 		 * the kernel must free the page when all its loans clear ...
1155 		 * note that the kernel can't change the loan status of our
1156 		 * page as long as we are holding PQ lock.
1157 		 */
1158 		atomic_clearbits_int(&pg->pg_flags, PQ_ANON);
1159 		pg->uanon->an_page = NULL;
1160 		pg->uanon = NULL;
1161 		return;
1162 	}
1163 	KASSERT(saved_loan_count == 0);
1164 
1165 	/*
1166 	 * now remove the page from the queues
1167 	 */
1168 
1169 	if (pg->pg_flags & PQ_ACTIVE) {
1170 		TAILQ_REMOVE(&uvm.page_active, pg, pageq);
1171 		atomic_clearbits_int(&pg->pg_flags, PQ_ACTIVE);
1172 		uvmexp.active--;
1173 	}
1174 	if (pg->pg_flags & PQ_INACTIVE) {
1175 		if (pg->pg_flags & PQ_SWAPBACKED)
1176 			TAILQ_REMOVE(&uvm.page_inactive_swp, pg, pageq);
1177 		else
1178 			TAILQ_REMOVE(&uvm.page_inactive_obj, pg, pageq);
1179 		atomic_clearbits_int(&pg->pg_flags, PQ_INACTIVE);
1180 		uvmexp.inactive--;
1181 	}
1182 
1183 	/*
1184 	 * if the page was wired, unwire it now.
1185 	 */
1186 
1187 	if (pg->wire_count) {
1188 		pg->wire_count = 0;
1189 		uvmexp.wired--;
1190 	}
1191 	if (pg->uanon) {
1192 		pg->uanon->an_page = NULL;
1193 #ifdef UBC
1194 		uvm_pgcnt_anon--;
1195 #endif
1196 	}
1197 
1198 	/*
1199 	 * and put on free queue
1200 	 */
1201 
1202 	atomic_clearbits_int(&pg->pg_flags, PG_ZERO);
1203 
1204 	uvm_lock_fpageq();
1205 	TAILQ_INSERT_TAIL(&uvm.page_free[
1206 	    uvm_page_lookup_freelist(pg)].pgfl_queues[PGFL_UNKNOWN], pg, pageq);
1207 	atomic_clearbits_int(&pg->pg_flags, PQ_MASK);
1208 	atomic_setbits_int(&pg->pg_flags, PQ_FREE);
1209 #ifdef DEBUG
1210 	pg->uobject = (void *)0xdeadbeef;
1211 	pg->offset = 0xdeadbeef;
1212 	pg->uanon = (void *)0xdeadbeef;
1213 #endif
1214 	uvmexp.free++;
1215 
1216 	if (uvmexp.zeropages < UVM_PAGEZERO_TARGET)
1217 		uvm.page_idle_zero = vm_page_zero_enable;
1218 
1219 	uvm_unlock_fpageq();
1220 }
1221 
1222 /*
1223  * uvm_page_unbusy: unbusy an array of pages.
1224  *
1225  * => pages must either all belong to the same object, or all belong to anons.
1226  * => if pages are object-owned, object must be locked.
1227  * => if pages are anon-owned, anons must be unlockd and have 0 refcount.
1228  */
1229 
1230 void
1231 uvm_page_unbusy(struct vm_page **pgs, int npgs)
1232 {
1233 	struct vm_page *pg;
1234 	struct uvm_object *uobj;
1235 	int i;
1236 	UVMHIST_FUNC("uvm_page_unbusy"); UVMHIST_CALLED(pdhist);
1237 
1238 	for (i = 0; i < npgs; i++) {
1239 		pg = pgs[i];
1240 
1241 		if (pg == NULL || pg == PGO_DONTCARE) {
1242 			continue;
1243 		}
1244 		if (pg->pg_flags & PG_WANTED) {
1245 			wakeup(pg);
1246 		}
1247 		if (pg->pg_flags & PG_RELEASED) {
1248 			UVMHIST_LOG(pdhist, "releasing pg %p", pg,0,0,0);
1249 			uobj = pg->uobject;
1250 			if (uobj != NULL) {
1251 				uobj->pgops->pgo_releasepg(pg, NULL);
1252 			} else {
1253 				atomic_clearbits_int(&pg->pg_flags, PG_BUSY);
1254 				UVM_PAGE_OWN(pg, NULL);
1255 				uvm_anfree(pg->uanon);
1256 			}
1257 		} else {
1258 			UVMHIST_LOG(pdhist, "unbusying pg %p", pg,0,0,0);
1259 			atomic_clearbits_int(&pg->pg_flags, PG_WANTED|PG_BUSY);
1260 			UVM_PAGE_OWN(pg, NULL);
1261 		}
1262 	}
1263 }
1264 
1265 #if defined(UVM_PAGE_TRKOWN)
1266 /*
1267  * uvm_page_own: set or release page ownership
1268  *
1269  * => this is a debugging function that keeps track of who sets PG_BUSY
1270  *	and where they do it.   it can be used to track down problems
1271  *	such a process setting "PG_BUSY" and never releasing it.
1272  * => page's object [if any] must be locked
1273  * => if "tag" is NULL then we are releasing page ownership
1274  */
1275 void
1276 uvm_page_own(struct vm_page *pg, char *tag)
1277 {
1278 	/* gain ownership? */
1279 	if (tag) {
1280 		if (pg->owner_tag) {
1281 			printf("uvm_page_own: page %p already owned "
1282 			    "by proc %d [%s]\n", pg,
1283 			     pg->owner, pg->owner_tag);
1284 			panic("uvm_page_own");
1285 		}
1286 		pg->owner = (curproc) ? curproc->p_pid :  (pid_t) -1;
1287 		pg->owner_tag = tag;
1288 		return;
1289 	}
1290 
1291 	/* drop ownership */
1292 	if (pg->owner_tag == NULL) {
1293 		printf("uvm_page_own: dropping ownership of an non-owned "
1294 		    "page (%p)\n", pg);
1295 		panic("uvm_page_own");
1296 	}
1297 	pg->owner_tag = NULL;
1298 	return;
1299 }
1300 #endif
1301 
1302 /*
1303  * uvm_pageidlezero: zero free pages while the system is idle.
1304  *
1305  * => we do at least one iteration per call, if we are below the target.
1306  * => we loop until we either reach the target or whichqs indicates that
1307  *	there is a process ready to run.
1308  */
1309 void
1310 uvm_pageidlezero(void)
1311 {
1312 	struct vm_page *pg;
1313 	struct pgfreelist *pgfl;
1314 	int free_list;
1315 	UVMHIST_FUNC("uvm_pageidlezero"); UVMHIST_CALLED(pghist);
1316 
1317 	do {
1318 		uvm_lock_fpageq();
1319 
1320 		if (uvmexp.zeropages >= UVM_PAGEZERO_TARGET) {
1321 			uvm.page_idle_zero = FALSE;
1322 			uvm_unlock_fpageq();
1323 			return;
1324 		}
1325 
1326 		for (free_list = 0; free_list < VM_NFREELIST; free_list++) {
1327 			pgfl = &uvm.page_free[free_list];
1328 			if ((pg = TAILQ_FIRST(&pgfl->pgfl_queues[
1329 			    PGFL_UNKNOWN])) != NULL)
1330 				break;
1331 		}
1332 
1333 		if (pg == NULL) {
1334 			/*
1335 			 * No non-zero'd pages; don't bother trying again
1336 			 * until we know we have non-zero'd pages free.
1337 			 */
1338 			uvm.page_idle_zero = FALSE;
1339 			uvm_unlock_fpageq();
1340 			return;
1341 		}
1342 
1343 		TAILQ_REMOVE(&pgfl->pgfl_queues[PGFL_UNKNOWN], pg, pageq);
1344 		uvmexp.free--;
1345 		uvm_unlock_fpageq();
1346 
1347 #ifdef PMAP_PAGEIDLEZERO
1348 		if (PMAP_PAGEIDLEZERO(pg) == FALSE) {
1349 			/*
1350 			 * The machine-dependent code detected some
1351 			 * reason for us to abort zeroing pages,
1352 			 * probably because there is a process now
1353 			 * ready to run.
1354 			 */
1355 			uvm_lock_fpageq();
1356 			TAILQ_INSERT_HEAD(&pgfl->pgfl_queues[PGFL_UNKNOWN],
1357 			    pg, pageq);
1358 			uvmexp.free++;
1359 			uvmexp.zeroaborts++;
1360 			uvm_unlock_fpageq();
1361 			return;
1362 		}
1363 #else
1364 		/*
1365 		 * XXX This will toast the cache unless the pmap_zero_page()
1366 		 * XXX implementation does uncached access.
1367 		 */
1368 		pmap_zero_page(pg);
1369 #endif
1370 		atomic_setbits_int(&pg->pg_flags, PG_ZERO);
1371 
1372 		uvm_lock_fpageq();
1373 		TAILQ_INSERT_HEAD(&pgfl->pgfl_queues[PGFL_ZEROS], pg, pageq);
1374 		uvmexp.free++;
1375 		uvmexp.zeropages++;
1376 		uvm_unlock_fpageq();
1377 	} while (curcpu_is_idle());
1378 }
1379 
1380 /*
1381  * when VM_PHYSSEG_MAX is 1, we can simplify these functions
1382  */
1383 
1384 #if VM_PHYSSEG_MAX > 1
1385 /*
1386  * vm_physseg_find: find vm_physseg structure that belongs to a PA
1387  */
1388 int
1389 vm_physseg_find(paddr_t pframe, int *offp)
1390 {
1391 
1392 #if (VM_PHYSSEG_STRAT == VM_PSTRAT_BSEARCH)
1393 	/* binary search for it */
1394 	int	start, len, try;
1395 
1396 	/*
1397 	 * if try is too large (thus target is less than than try) we reduce
1398 	 * the length to trunc(len/2) [i.e. everything smaller than "try"]
1399 	 *
1400 	 * if the try is too small (thus target is greater than try) then
1401 	 * we set the new start to be (try + 1).   this means we need to
1402 	 * reduce the length to (round(len/2) - 1).
1403 	 *
1404 	 * note "adjust" below which takes advantage of the fact that
1405 	 *  (round(len/2) - 1) == trunc((len - 1) / 2)
1406 	 * for any value of len we may have
1407 	 */
1408 
1409 	for (start = 0, len = vm_nphysseg ; len != 0 ; len = len / 2) {
1410 		try = start + (len / 2);	/* try in the middle */
1411 
1412 		/* start past our try? */
1413 		if (pframe >= vm_physmem[try].start) {
1414 			/* was try correct? */
1415 			if (pframe < vm_physmem[try].end) {
1416 				if (offp)
1417 					*offp = pframe - vm_physmem[try].start;
1418 				return(try);            /* got it */
1419 			}
1420 			start = try + 1;	/* next time, start here */
1421 			len--;			/* "adjust" */
1422 		} else {
1423 			/*
1424 			 * pframe before try, just reduce length of
1425 			 * region, done in "for" loop
1426 			 */
1427 		}
1428 	}
1429 	return(-1);
1430 
1431 #else
1432 	/* linear search for it */
1433 	int	lcv;
1434 
1435 	for (lcv = 0; lcv < vm_nphysseg; lcv++) {
1436 		if (pframe >= vm_physmem[lcv].start &&
1437 		    pframe < vm_physmem[lcv].end) {
1438 			if (offp)
1439 				*offp = pframe - vm_physmem[lcv].start;
1440 			return(lcv);		   /* got it */
1441 		}
1442 	}
1443 	return(-1);
1444 
1445 #endif
1446 }
1447 
1448 /*
1449  * PHYS_TO_VM_PAGE: find vm_page for a PA.   used by MI code to get vm_pages
1450  * back from an I/O mapping (ugh!).   used in some MD code as well.
1451  */
1452 struct vm_page *
1453 PHYS_TO_VM_PAGE(paddr_t pa)
1454 {
1455 	paddr_t pf = atop(pa);
1456 	int	off;
1457 	int	psi;
1458 
1459 	psi = vm_physseg_find(pf, &off);
1460 
1461 	return ((psi == -1) ? NULL : &vm_physmem[psi].pgs[off]);
1462 }
1463 #endif /* VM_PHYSSEG_MAX > 1 */
1464 
1465 /*
1466  * uvm_pagelookup: look up a page
1467  *
1468  * => caller should lock object to keep someone from pulling the page
1469  *	out from under it
1470  */
1471 struct vm_page *
1472 uvm_pagelookup(struct uvm_object *obj, voff_t off)
1473 {
1474 	struct vm_page *pg;
1475 	struct pglist *buck;
1476 
1477 	mtx_enter(&uvm.hashlock);
1478 	buck = &uvm.page_hash[uvm_pagehash(obj,off)];
1479 
1480 	TAILQ_FOREACH(pg, buck, hashq) {
1481 		if (pg->uobject == obj && pg->offset == off) {
1482 			break;
1483 		}
1484 	}
1485 	mtx_leave(&uvm.hashlock);
1486 	return(pg);
1487 }
1488 
1489 /*
1490  * uvm_pagewire: wire the page, thus removing it from the daemon's grasp
1491  *
1492  * => caller must lock page queues
1493  */
1494 void
1495 uvm_pagewire(struct vm_page *pg)
1496 {
1497 	if (pg->wire_count == 0) {
1498 		if (pg->pg_flags & PQ_ACTIVE) {
1499 			TAILQ_REMOVE(&uvm.page_active, pg, pageq);
1500 			atomic_clearbits_int(&pg->pg_flags, PQ_ACTIVE);
1501 			uvmexp.active--;
1502 		}
1503 		if (pg->pg_flags & PQ_INACTIVE) {
1504 			if (pg->pg_flags & PQ_SWAPBACKED)
1505 				TAILQ_REMOVE(&uvm.page_inactive_swp, pg, pageq);
1506 			else
1507 				TAILQ_REMOVE(&uvm.page_inactive_obj, pg, pageq);
1508 			atomic_clearbits_int(&pg->pg_flags, PQ_INACTIVE);
1509 			uvmexp.inactive--;
1510 		}
1511 		uvmexp.wired++;
1512 	}
1513 	pg->wire_count++;
1514 }
1515 
1516 /*
1517  * uvm_pageunwire: unwire the page.
1518  *
1519  * => activate if wire count goes to zero.
1520  * => caller must lock page queues
1521  */
1522 void
1523 uvm_pageunwire(struct vm_page *pg)
1524 {
1525 	pg->wire_count--;
1526 	if (pg->wire_count == 0) {
1527 		TAILQ_INSERT_TAIL(&uvm.page_active, pg, pageq);
1528 		uvmexp.active++;
1529 		atomic_setbits_int(&pg->pg_flags, PQ_ACTIVE);
1530 		uvmexp.wired--;
1531 	}
1532 }
1533 
1534 /*
1535  * uvm_pagedeactivate: deactivate page -- no pmaps have access to page
1536  *
1537  * => caller must lock page queues
1538  * => caller must check to make sure page is not wired
1539  * => object that page belongs to must be locked (so we can adjust pg->flags)
1540  */
1541 void
1542 uvm_pagedeactivate(struct vm_page *pg)
1543 {
1544 	if (pg->pg_flags & PQ_ACTIVE) {
1545 		TAILQ_REMOVE(&uvm.page_active, pg, pageq);
1546 		atomic_clearbits_int(&pg->pg_flags, PQ_ACTIVE);
1547 		uvmexp.active--;
1548 	}
1549 	if ((pg->pg_flags & PQ_INACTIVE) == 0) {
1550 		KASSERT(pg->wire_count == 0);
1551 		if (pg->pg_flags & PQ_SWAPBACKED)
1552 			TAILQ_INSERT_TAIL(&uvm.page_inactive_swp, pg, pageq);
1553 		else
1554 			TAILQ_INSERT_TAIL(&uvm.page_inactive_obj, pg, pageq);
1555 		atomic_setbits_int(&pg->pg_flags, PQ_INACTIVE);
1556 		uvmexp.inactive++;
1557 		pmap_clear_reference(pg);
1558 		/*
1559 		 * update the "clean" bit.  this isn't 100%
1560 		 * accurate, and doesn't have to be.  we'll
1561 		 * re-sync it after we zap all mappings when
1562 		 * scanning the inactive list.
1563 		 */
1564 		if ((pg->pg_flags & PG_CLEAN) != 0 &&
1565 		    pmap_is_modified(pg))
1566 			atomic_clearbits_int(&pg->pg_flags, PG_CLEAN);
1567 	}
1568 }
1569 
1570 /*
1571  * uvm_pageactivate: activate page
1572  *
1573  * => caller must lock page queues
1574  */
1575 void
1576 uvm_pageactivate(struct vm_page *pg)
1577 {
1578 	if (pg->pg_flags & PQ_INACTIVE) {
1579 		if (pg->pg_flags & PQ_SWAPBACKED)
1580 			TAILQ_REMOVE(&uvm.page_inactive_swp, pg, pageq);
1581 		else
1582 			TAILQ_REMOVE(&uvm.page_inactive_obj, pg, pageq);
1583 		atomic_clearbits_int(&pg->pg_flags, PQ_INACTIVE);
1584 		uvmexp.inactive--;
1585 	}
1586 	if (pg->wire_count == 0) {
1587 
1588 		/*
1589 		 * if page is already active, remove it from list so we
1590 		 * can put it at tail.  if it wasn't active, then mark
1591 		 * it active and bump active count
1592 		 */
1593 		if (pg->pg_flags & PQ_ACTIVE)
1594 			TAILQ_REMOVE(&uvm.page_active, pg, pageq);
1595 		else {
1596 			atomic_setbits_int(&pg->pg_flags, PQ_ACTIVE);
1597 			uvmexp.active++;
1598 		}
1599 
1600 		TAILQ_INSERT_TAIL(&uvm.page_active, pg, pageq);
1601 	}
1602 }
1603 
1604 /*
1605  * uvm_pagezero: zero fill a page
1606  *
1607  * => if page is part of an object then the object should be locked
1608  *	to protect pg->flags.
1609  */
1610 void
1611 uvm_pagezero(struct vm_page *pg)
1612 {
1613 	atomic_clearbits_int(&pg->pg_flags, PG_CLEAN);
1614 	pmap_zero_page(pg);
1615 }
1616 
1617 /*
1618  * uvm_pagecopy: copy a page
1619  *
1620  * => if page is part of an object then the object should be locked
1621  *	to protect pg->flags.
1622  */
1623 void
1624 uvm_pagecopy(struct vm_page *src, struct vm_page *dst)
1625 {
1626 	atomic_clearbits_int(&dst->pg_flags, PG_CLEAN);
1627 	pmap_copy_page(src, dst);
1628 }
1629 
1630 /*
1631  * uvm_page_lookup_freelist: look up the free list for the specified page
1632  */
1633 int
1634 uvm_page_lookup_freelist(struct vm_page *pg)
1635 {
1636 #if VM_PHYSSEG_MAX == 1
1637 	return (vm_physmem[0].free_list);
1638 #else
1639 	int lcv;
1640 
1641 	lcv = vm_physseg_find(atop(VM_PAGE_TO_PHYS(pg)), NULL);
1642 	KASSERT(lcv != -1);
1643 	return (vm_physmem[lcv].free_list);
1644 #endif
1645 }
1646