xref: /netbsd-src/sys/uvm/uvm_page.c (revision 1b9578b8c2c1f848eeb16dabbfd7d1f0d9fdefbd)
1 /*	$NetBSD: uvm_page.c,v 1.175 2011/06/15 19:46:11 rmind Exp $	*/
2 
3 /*
4  * Copyright (c) 1997 Charles D. Cranor and Washington University.
5  * Copyright (c) 1991, 1993, The Regents of the University of California.
6  *
7  * All rights reserved.
8  *
9  * This code is derived from software contributed to Berkeley by
10  * The Mach Operating System project at Carnegie-Mellon University.
11  *
12  * Redistribution and use in source and binary forms, with or without
13  * modification, are permitted provided that the following conditions
14  * are met:
15  * 1. Redistributions of source code must retain the above copyright
16  *    notice, this list of conditions and the following disclaimer.
17  * 2. Redistributions in binary form must reproduce the above copyright
18  *    notice, this list of conditions and the following disclaimer in the
19  *    documentation and/or other materials provided with the distribution.
20  * 3. Neither the name of the University nor the names of its contributors
21  *    may be used to endorse or promote products derived from this software
22  *    without specific prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34  * SUCH DAMAGE.
35  *
36  *	@(#)vm_page.c   8.3 (Berkeley) 3/21/94
37  * from: Id: uvm_page.c,v 1.1.2.18 1998/02/06 05:24:42 chs Exp
38  *
39  *
40  * Copyright (c) 1987, 1990 Carnegie-Mellon University.
41  * All rights reserved.
42  *
43  * Permission to use, copy, modify and distribute this software and
44  * its documentation is hereby granted, provided that both the copyright
45  * notice and this permission notice appear in all copies of the
46  * software, derivative works or modified versions, and any portions
47  * thereof, and that both notices appear in supporting documentation.
48  *
49  * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
50  * CONDITION.  CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
51  * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
52  *
53  * Carnegie Mellon requests users of this software to return to
54  *
55  *  Software Distribution Coordinator  or  Software.Distribution@CS.CMU.EDU
56  *  School of Computer Science
57  *  Carnegie Mellon University
58  *  Pittsburgh PA 15213-3890
59  *
60  * any improvements or extensions that they make and grant Carnegie the
61  * rights to redistribute these changes.
62  */
63 
64 /*
65  * uvm_page.c: page ops.
66  */
67 
68 #include <sys/cdefs.h>
69 __KERNEL_RCSID(0, "$NetBSD: uvm_page.c,v 1.175 2011/06/15 19:46:11 rmind Exp $");
70 
71 #include "opt_ddb.h"
72 #include "opt_uvmhist.h"
73 #include "opt_readahead.h"
74 
75 #include <sys/param.h>
76 #include <sys/systm.h>
77 #include <sys/malloc.h>
78 #include <sys/sched.h>
79 #include <sys/kernel.h>
80 #include <sys/vnode.h>
81 #include <sys/proc.h>
82 #include <sys/atomic.h>
83 #include <sys/cpu.h>
84 
85 #include <uvm/uvm.h>
86 #include <uvm/uvm_ddb.h>
87 #include <uvm/uvm_pdpolicy.h>
88 
89 /*
90  * global vars... XXXCDC: move to uvm. structure.
91  */
92 
93 /*
94  * physical memory config is stored in vm_physmem.
95  */
96 
97 struct vm_physseg vm_physmem[VM_PHYSSEG_MAX];	/* XXXCDC: uvm.physmem */
98 int vm_nphysseg = 0;				/* XXXCDC: uvm.nphysseg */
99 #define	vm_nphysmem	vm_nphysseg
100 
101 /*
102  * Some supported CPUs in a given architecture don't support all
103  * of the things necessary to do idle page zero'ing efficiently.
104  * We therefore provide a way to enable it from machdep code here.
105  */
106 bool vm_page_zero_enable = false;
107 
108 /*
109  * number of pages per-CPU to reserve for the kernel.
110  */
111 int vm_page_reserve_kernel = 5;
112 
113 /*
114  * physical memory size;
115  */
116 int physmem;
117 
118 /*
119  * local variables
120  */
121 
122 /*
123  * these variables record the values returned by vm_page_bootstrap,
124  * for debugging purposes.  The implementation of uvm_pageboot_alloc
125  * and pmap_startup here also uses them internally.
126  */
127 
128 static vaddr_t      virtual_space_start;
129 static vaddr_t      virtual_space_end;
130 
131 /*
132  * we allocate an initial number of page colors in uvm_page_init(),
133  * and remember them.  We may re-color pages as cache sizes are
134  * discovered during the autoconfiguration phase.  But we can never
135  * free the initial set of buckets, since they are allocated using
136  * uvm_pageboot_alloc().
137  */
138 
139 static bool have_recolored_pages /* = false */;
140 
141 MALLOC_DEFINE(M_VMPAGE, "VM page", "VM page");
142 
143 #ifdef DEBUG
144 vaddr_t uvm_zerocheckkva;
145 #endif /* DEBUG */
146 
147 /*
148  * local prototypes
149  */
150 
151 static void uvm_pageinsert(struct uvm_object *, struct vm_page *);
152 static void uvm_pageremove(struct uvm_object *, struct vm_page *);
153 
154 /*
155  * per-object tree of pages
156  */
157 
158 static signed int
159 uvm_page_compare_nodes(void *ctx, const void *n1, const void *n2)
160 {
161 	const struct vm_page *pg1 = n1;
162 	const struct vm_page *pg2 = n2;
163 	const voff_t a = pg1->offset;
164 	const voff_t b = pg2->offset;
165 
166 	if (a < b)
167 		return -1;
168 	if (a > b)
169 		return 1;
170 	return 0;
171 }
172 
173 static signed int
174 uvm_page_compare_key(void *ctx, const void *n, const void *key)
175 {
176 	const struct vm_page *pg = n;
177 	const voff_t a = pg->offset;
178 	const voff_t b = *(const voff_t *)key;
179 
180 	if (a < b)
181 		return -1;
182 	if (a > b)
183 		return 1;
184 	return 0;
185 }
186 
187 const rb_tree_ops_t uvm_page_tree_ops = {
188 	.rbto_compare_nodes = uvm_page_compare_nodes,
189 	.rbto_compare_key = uvm_page_compare_key,
190 	.rbto_node_offset = offsetof(struct vm_page, rb_node),
191 	.rbto_context = NULL
192 };
193 
194 /*
195  * inline functions
196  */
197 
198 /*
199  * uvm_pageinsert: insert a page in the object.
200  *
201  * => caller must lock object
202  * => caller must lock page queues
203  * => call should have already set pg's object and offset pointers
204  *    and bumped the version counter
205  */
206 
207 static inline void
208 uvm_pageinsert_list(struct uvm_object *uobj, struct vm_page *pg,
209     struct vm_page *where)
210 {
211 
212 	KASSERT(uobj == pg->uobject);
213 	KASSERT(mutex_owned(uobj->vmobjlock));
214 	KASSERT((pg->flags & PG_TABLED) == 0);
215 	KASSERT(where == NULL || (where->flags & PG_TABLED));
216 	KASSERT(where == NULL || (where->uobject == uobj));
217 
218 	if (UVM_OBJ_IS_VNODE(uobj)) {
219 		if (uobj->uo_npages == 0) {
220 			struct vnode *vp = (struct vnode *)uobj;
221 
222 			vholdl(vp);
223 		}
224 		if (UVM_OBJ_IS_VTEXT(uobj)) {
225 			atomic_inc_uint(&uvmexp.execpages);
226 		} else {
227 			atomic_inc_uint(&uvmexp.filepages);
228 		}
229 	} else if (UVM_OBJ_IS_AOBJ(uobj)) {
230 		atomic_inc_uint(&uvmexp.anonpages);
231 	}
232 
233 	if (where)
234 		TAILQ_INSERT_AFTER(&uobj->memq, where, pg, listq.queue);
235 	else
236 		TAILQ_INSERT_TAIL(&uobj->memq, pg, listq.queue);
237 	pg->flags |= PG_TABLED;
238 	uobj->uo_npages++;
239 }
240 
241 
242 static inline void
243 uvm_pageinsert_tree(struct uvm_object *uobj, struct vm_page *pg)
244 {
245 	struct vm_page *ret;
246 
247 	KASSERT(uobj == pg->uobject);
248 	ret = rb_tree_insert_node(&uobj->rb_tree, pg);
249 	KASSERT(ret == pg);
250 }
251 
252 static inline void
253 uvm_pageinsert(struct uvm_object *uobj, struct vm_page *pg)
254 {
255 
256 	KDASSERT(uobj != NULL);
257 	uvm_pageinsert_tree(uobj, pg);
258 	uvm_pageinsert_list(uobj, pg, NULL);
259 }
260 
261 /*
262  * uvm_page_remove: remove page from object.
263  *
264  * => caller must lock object
265  * => caller must lock page queues
266  */
267 
268 static inline void
269 uvm_pageremove_list(struct uvm_object *uobj, struct vm_page *pg)
270 {
271 
272 	KASSERT(uobj == pg->uobject);
273 	KASSERT(mutex_owned(uobj->vmobjlock));
274 	KASSERT(pg->flags & PG_TABLED);
275 
276 	if (UVM_OBJ_IS_VNODE(uobj)) {
277 		if (uobj->uo_npages == 1) {
278 			struct vnode *vp = (struct vnode *)uobj;
279 
280 			holdrelel(vp);
281 		}
282 		if (UVM_OBJ_IS_VTEXT(uobj)) {
283 			atomic_dec_uint(&uvmexp.execpages);
284 		} else {
285 			atomic_dec_uint(&uvmexp.filepages);
286 		}
287 	} else if (UVM_OBJ_IS_AOBJ(uobj)) {
288 		atomic_dec_uint(&uvmexp.anonpages);
289 	}
290 
291 	/* object should be locked */
292 	uobj->uo_npages--;
293 	TAILQ_REMOVE(&uobj->memq, pg, listq.queue);
294 	pg->flags &= ~PG_TABLED;
295 	pg->uobject = NULL;
296 }
297 
298 static inline void
299 uvm_pageremove_tree(struct uvm_object *uobj, struct vm_page *pg)
300 {
301 
302 	KASSERT(uobj == pg->uobject);
303 	rb_tree_remove_node(&uobj->rb_tree, pg);
304 }
305 
306 static inline void
307 uvm_pageremove(struct uvm_object *uobj, struct vm_page *pg)
308 {
309 
310 	KDASSERT(uobj != NULL);
311 	uvm_pageremove_tree(uobj, pg);
312 	uvm_pageremove_list(uobj, pg);
313 }
314 
315 static void
316 uvm_page_init_buckets(struct pgfreelist *pgfl)
317 {
318 	int color, i;
319 
320 	for (color = 0; color < uvmexp.ncolors; color++) {
321 		for (i = 0; i < PGFL_NQUEUES; i++) {
322 			LIST_INIT(&pgfl->pgfl_buckets[color].pgfl_queues[i]);
323 		}
324 	}
325 }
326 
327 /*
328  * uvm_page_init: init the page system.   called from uvm_init().
329  *
330  * => we return the range of kernel virtual memory in kvm_startp/kvm_endp
331  */
332 
333 void
334 uvm_page_init(vaddr_t *kvm_startp, vaddr_t *kvm_endp)
335 {
336 	static struct uvm_cpu boot_cpu;
337 	psize_t freepages, pagecount, bucketcount, n;
338 	struct pgflbucket *bucketarray, *cpuarray;
339 	struct vm_physseg *seg;
340 	struct vm_page *pagearray;
341 	int lcv;
342 	u_int i;
343 	paddr_t paddr;
344 
345 	KASSERT(ncpu <= 1);
346 	CTASSERT(sizeof(pagearray->offset) >= sizeof(struct uvm_cpu *));
347 
348 	/*
349 	 * init the page queues and page queue locks, except the free
350 	 * list; we allocate that later (with the initial vm_page
351 	 * structures).
352 	 */
353 
354 	uvm.cpus[0] = &boot_cpu;
355 	curcpu()->ci_data.cpu_uvm = &boot_cpu;
356 	uvm_reclaim_init();
357 	uvmpdpol_init();
358 	mutex_init(&uvm_pageqlock, MUTEX_DRIVER, IPL_NONE);
359 	mutex_init(&uvm_fpageqlock, MUTEX_DRIVER, IPL_VM);
360 
361 	/*
362 	 * allocate vm_page structures.
363 	 */
364 
365 	/*
366 	 * sanity check:
367 	 * before calling this function the MD code is expected to register
368 	 * some free RAM with the uvm_page_physload() function.   our job
369 	 * now is to allocate vm_page structures for this memory.
370 	 */
371 
372 	if (vm_nphysmem == 0)
373 		panic("uvm_page_bootstrap: no memory pre-allocated");
374 
375 	/*
376 	 * first calculate the number of free pages...
377 	 *
378 	 * note that we use start/end rather than avail_start/avail_end.
379 	 * this allows us to allocate extra vm_page structures in case we
380 	 * want to return some memory to the pool after booting.
381 	 */
382 
383 	freepages = 0;
384 	for (lcv = 0 ; lcv < vm_nphysmem ; lcv++) {
385 		seg = VM_PHYSMEM_PTR(lcv);
386 		freepages += (seg->end - seg->start);
387 	}
388 
389 	/*
390 	 * Let MD code initialize the number of colors, or default
391 	 * to 1 color if MD code doesn't care.
392 	 */
393 	if (uvmexp.ncolors == 0)
394 		uvmexp.ncolors = 1;
395 	uvmexp.colormask = uvmexp.ncolors - 1;
396 
397 	/*
398 	 * we now know we have (PAGE_SIZE * freepages) bytes of memory we can
399 	 * use.   for each page of memory we use we need a vm_page structure.
400 	 * thus, the total number of pages we can use is the total size of
401 	 * the memory divided by the PAGE_SIZE plus the size of the vm_page
402 	 * structure.   we add one to freepages as a fudge factor to avoid
403 	 * truncation errors (since we can only allocate in terms of whole
404 	 * pages).
405 	 */
406 
407 	bucketcount = uvmexp.ncolors * VM_NFREELIST;
408 	pagecount = ((freepages + 1) << PAGE_SHIFT) /
409 	    (PAGE_SIZE + sizeof(struct vm_page));
410 
411 	bucketarray = (void *)uvm_pageboot_alloc((bucketcount *
412 	    sizeof(struct pgflbucket) * 2) + (pagecount *
413 	    sizeof(struct vm_page)));
414 	cpuarray = bucketarray + bucketcount;
415 	pagearray = (struct vm_page *)(bucketarray + bucketcount * 2);
416 
417 	for (lcv = 0; lcv < VM_NFREELIST; lcv++) {
418 		uvm.page_free[lcv].pgfl_buckets =
419 		    (bucketarray + (lcv * uvmexp.ncolors));
420 		uvm_page_init_buckets(&uvm.page_free[lcv]);
421 		uvm.cpus[0]->page_free[lcv].pgfl_buckets =
422 		    (cpuarray + (lcv * uvmexp.ncolors));
423 		uvm_page_init_buckets(&uvm.cpus[0]->page_free[lcv]);
424 	}
425 	memset(pagearray, 0, pagecount * sizeof(struct vm_page));
426 
427 	/*
428 	 * init the vm_page structures and put them in the correct place.
429 	 */
430 
431 	for (lcv = 0 ; lcv < vm_nphysmem ; lcv++) {
432 		seg = VM_PHYSMEM_PTR(lcv);
433 		n = seg->end - seg->start;
434 
435 		/* set up page array pointers */
436 		seg->pgs = pagearray;
437 		pagearray += n;
438 		pagecount -= n;
439 		seg->lastpg = seg->pgs + n;
440 
441 		/* init and free vm_pages (we've already zeroed them) */
442 		paddr = ctob(seg->start);
443 		for (i = 0 ; i < n ; i++, paddr += PAGE_SIZE) {
444 			seg->pgs[i].phys_addr = paddr;
445 #ifdef __HAVE_VM_PAGE_MD
446 			VM_MDPAGE_INIT(&seg->pgs[i]);
447 #endif
448 			if (atop(paddr) >= seg->avail_start &&
449 			    atop(paddr) < seg->avail_end) {
450 				uvmexp.npages++;
451 				/* add page to free pool */
452 				uvm_pagefree(&seg->pgs[i]);
453 			}
454 		}
455 	}
456 
457 	/*
458 	 * pass up the values of virtual_space_start and
459 	 * virtual_space_end (obtained by uvm_pageboot_alloc) to the upper
460 	 * layers of the VM.
461 	 */
462 
463 	*kvm_startp = round_page(virtual_space_start);
464 	*kvm_endp = trunc_page(virtual_space_end);
465 #ifdef DEBUG
466 	/*
467 	 * steal kva for uvm_pagezerocheck().
468 	 */
469 	uvm_zerocheckkva = *kvm_startp;
470 	*kvm_startp += PAGE_SIZE;
471 #endif /* DEBUG */
472 
473 	/*
474 	 * init various thresholds.
475 	 */
476 
477 	uvmexp.reserve_pagedaemon = 1;
478 	uvmexp.reserve_kernel = vm_page_reserve_kernel;
479 
480 	/*
481 	 * determine if we should zero pages in the idle loop.
482 	 */
483 
484 	uvm.cpus[0]->page_idle_zero = vm_page_zero_enable;
485 
486 	/*
487 	 * done!
488 	 */
489 
490 	uvm.page_init_done = true;
491 }
492 
493 /*
494  * uvm_setpagesize: set the page size
495  *
496  * => sets page_shift and page_mask from uvmexp.pagesize.
497  */
498 
499 void
500 uvm_setpagesize(void)
501 {
502 
503 	/*
504 	 * If uvmexp.pagesize is 0 at this point, we expect PAGE_SIZE
505 	 * to be a constant (indicated by being a non-zero value).
506 	 */
507 	if (uvmexp.pagesize == 0) {
508 		if (PAGE_SIZE == 0)
509 			panic("uvm_setpagesize: uvmexp.pagesize not set");
510 		uvmexp.pagesize = PAGE_SIZE;
511 	}
512 	uvmexp.pagemask = uvmexp.pagesize - 1;
513 	if ((uvmexp.pagemask & uvmexp.pagesize) != 0)
514 		panic("uvm_setpagesize: page size %u (%#x) not a power of two",
515 		    uvmexp.pagesize, uvmexp.pagesize);
516 	for (uvmexp.pageshift = 0; ; uvmexp.pageshift++)
517 		if ((1 << uvmexp.pageshift) == uvmexp.pagesize)
518 			break;
519 }
520 
521 /*
522  * uvm_pageboot_alloc: steal memory from physmem for bootstrapping
523  */
524 
525 vaddr_t
526 uvm_pageboot_alloc(vsize_t size)
527 {
528 	static bool initialized = false;
529 	vaddr_t addr;
530 #if !defined(PMAP_STEAL_MEMORY)
531 	vaddr_t vaddr;
532 	paddr_t paddr;
533 #endif
534 
535 	/*
536 	 * on first call to this function, initialize ourselves.
537 	 */
538 	if (initialized == false) {
539 		pmap_virtual_space(&virtual_space_start, &virtual_space_end);
540 
541 		/* round it the way we like it */
542 		virtual_space_start = round_page(virtual_space_start);
543 		virtual_space_end = trunc_page(virtual_space_end);
544 
545 		initialized = true;
546 	}
547 
548 	/* round to page size */
549 	size = round_page(size);
550 
551 #if defined(PMAP_STEAL_MEMORY)
552 
553 	/*
554 	 * defer bootstrap allocation to MD code (it may want to allocate
555 	 * from a direct-mapped segment).  pmap_steal_memory should adjust
556 	 * virtual_space_start/virtual_space_end if necessary.
557 	 */
558 
559 	addr = pmap_steal_memory(size, &virtual_space_start,
560 	    &virtual_space_end);
561 
562 	return(addr);
563 
564 #else /* !PMAP_STEAL_MEMORY */
565 
566 	/*
567 	 * allocate virtual memory for this request
568 	 */
569 	if (virtual_space_start == virtual_space_end ||
570 	    (virtual_space_end - virtual_space_start) < size)
571 		panic("uvm_pageboot_alloc: out of virtual space");
572 
573 	addr = virtual_space_start;
574 
575 #ifdef PMAP_GROWKERNEL
576 	/*
577 	 * If the kernel pmap can't map the requested space,
578 	 * then allocate more resources for it.
579 	 */
580 	if (uvm_maxkaddr < (addr + size)) {
581 		uvm_maxkaddr = pmap_growkernel(addr + size);
582 		if (uvm_maxkaddr < (addr + size))
583 			panic("uvm_pageboot_alloc: pmap_growkernel() failed");
584 	}
585 #endif
586 
587 	virtual_space_start += size;
588 
589 	/*
590 	 * allocate and mapin physical pages to back new virtual pages
591 	 */
592 
593 	for (vaddr = round_page(addr) ; vaddr < addr + size ;
594 	    vaddr += PAGE_SIZE) {
595 
596 		if (!uvm_page_physget(&paddr))
597 			panic("uvm_pageboot_alloc: out of memory");
598 
599 		/*
600 		 * Note this memory is no longer managed, so using
601 		 * pmap_kenter is safe.
602 		 */
603 		pmap_kenter_pa(vaddr, paddr, VM_PROT_READ|VM_PROT_WRITE, 0);
604 	}
605 	pmap_update(pmap_kernel());
606 	return(addr);
607 #endif	/* PMAP_STEAL_MEMORY */
608 }
609 
610 #if !defined(PMAP_STEAL_MEMORY)
611 /*
612  * uvm_page_physget: "steal" one page from the vm_physmem structure.
613  *
614  * => attempt to allocate it off the end of a segment in which the "avail"
615  *    values match the start/end values.   if we can't do that, then we
616  *    will advance both values (making them equal, and removing some
617  *    vm_page structures from the non-avail area).
618  * => return false if out of memory.
619  */
620 
621 /* subroutine: try to allocate from memory chunks on the specified freelist */
622 static bool uvm_page_physget_freelist(paddr_t *, int);
623 
624 static bool
625 uvm_page_physget_freelist(paddr_t *paddrp, int freelist)
626 {
627 	struct vm_physseg *seg;
628 	int lcv, x;
629 
630 	/* pass 1: try allocating from a matching end */
631 #if (VM_PHYSSEG_STRAT == VM_PSTRAT_BIGFIRST)
632 	for (lcv = vm_nphysmem - 1 ; lcv >= 0 ; lcv--)
633 #else
634 	for (lcv = 0 ; lcv < vm_nphysmem ; lcv++)
635 #endif
636 	{
637 		seg = VM_PHYSMEM_PTR(lcv);
638 
639 		if (uvm.page_init_done == true)
640 			panic("uvm_page_physget: called _after_ bootstrap");
641 
642 		if (seg->free_list != freelist)
643 			continue;
644 
645 		/* try from front */
646 		if (seg->avail_start == seg->start &&
647 		    seg->avail_start < seg->avail_end) {
648 			*paddrp = ctob(seg->avail_start);
649 			seg->avail_start++;
650 			seg->start++;
651 			/* nothing left?   nuke it */
652 			if (seg->avail_start == seg->end) {
653 				if (vm_nphysmem == 1)
654 				    panic("uvm_page_physget: out of memory!");
655 				vm_nphysmem--;
656 				for (x = lcv ; x < vm_nphysmem ; x++)
657 					/* structure copy */
658 					VM_PHYSMEM_PTR_SWAP(x, x + 1);
659 			}
660 			return (true);
661 		}
662 
663 		/* try from rear */
664 		if (seg->avail_end == seg->end &&
665 		    seg->avail_start < seg->avail_end) {
666 			*paddrp = ctob(seg->avail_end - 1);
667 			seg->avail_end--;
668 			seg->end--;
669 			/* nothing left?   nuke it */
670 			if (seg->avail_end == seg->start) {
671 				if (vm_nphysmem == 1)
672 				    panic("uvm_page_physget: out of memory!");
673 				vm_nphysmem--;
674 				for (x = lcv ; x < vm_nphysmem ; x++)
675 					/* structure copy */
676 					VM_PHYSMEM_PTR_SWAP(x, x + 1);
677 			}
678 			return (true);
679 		}
680 	}
681 
682 	/* pass2: forget about matching ends, just allocate something */
683 #if (VM_PHYSSEG_STRAT == VM_PSTRAT_BIGFIRST)
684 	for (lcv = vm_nphysmem - 1 ; lcv >= 0 ; lcv--)
685 #else
686 	for (lcv = 0 ; lcv < vm_nphysmem ; lcv++)
687 #endif
688 	{
689 		seg = VM_PHYSMEM_PTR(lcv);
690 
691 		/* any room in this bank? */
692 		if (seg->avail_start >= seg->avail_end)
693 			continue;  /* nope */
694 
695 		*paddrp = ctob(seg->avail_start);
696 		seg->avail_start++;
697 		/* truncate! */
698 		seg->start = seg->avail_start;
699 
700 		/* nothing left?   nuke it */
701 		if (seg->avail_start == seg->end) {
702 			if (vm_nphysmem == 1)
703 				panic("uvm_page_physget: out of memory!");
704 			vm_nphysmem--;
705 			for (x = lcv ; x < vm_nphysmem ; x++)
706 				/* structure copy */
707 				VM_PHYSMEM_PTR_SWAP(x, x + 1);
708 		}
709 		return (true);
710 	}
711 
712 	return (false);        /* whoops! */
713 }
714 
715 bool
716 uvm_page_physget(paddr_t *paddrp)
717 {
718 	int i;
719 
720 	/* try in the order of freelist preference */
721 	for (i = 0; i < VM_NFREELIST; i++)
722 		if (uvm_page_physget_freelist(paddrp, i) == true)
723 			return (true);
724 	return (false);
725 }
726 #endif /* PMAP_STEAL_MEMORY */
727 
728 /*
729  * uvm_page_physload: load physical memory into VM system
730  *
731  * => all args are PFs
732  * => all pages in start/end get vm_page structures
733  * => areas marked by avail_start/avail_end get added to the free page pool
734  * => we are limited to VM_PHYSSEG_MAX physical memory segments
735  */
736 
737 void
738 uvm_page_physload(paddr_t start, paddr_t end, paddr_t avail_start,
739     paddr_t avail_end, int free_list)
740 {
741 	int preload, lcv;
742 	psize_t npages;
743 	struct vm_page *pgs;
744 	struct vm_physseg *ps;
745 
746 	if (uvmexp.pagesize == 0)
747 		panic("uvm_page_physload: page size not set!");
748 	if (free_list >= VM_NFREELIST || free_list < VM_FREELIST_DEFAULT)
749 		panic("uvm_page_physload: bad free list %d", free_list);
750 	if (start >= end)
751 		panic("uvm_page_physload: start >= end");
752 
753 	/*
754 	 * do we have room?
755 	 */
756 
757 	if (vm_nphysmem == VM_PHYSSEG_MAX) {
758 		printf("uvm_page_physload: unable to load physical memory "
759 		    "segment\n");
760 		printf("\t%d segments allocated, ignoring 0x%llx -> 0x%llx\n",
761 		    VM_PHYSSEG_MAX, (long long)start, (long long)end);
762 		printf("\tincrease VM_PHYSSEG_MAX\n");
763 		return;
764 	}
765 
766 	/*
767 	 * check to see if this is a "preload" (i.e. uvm_page_init hasn't been
768 	 * called yet, so malloc is not available).
769 	 */
770 
771 	for (lcv = 0 ; lcv < vm_nphysmem ; lcv++) {
772 		if (VM_PHYSMEM_PTR(lcv)->pgs)
773 			break;
774 	}
775 	preload = (lcv == vm_nphysmem);
776 
777 	/*
778 	 * if VM is already running, attempt to malloc() vm_page structures
779 	 */
780 
781 	if (!preload) {
782 		panic("uvm_page_physload: tried to add RAM after vm_mem_init");
783 	} else {
784 		pgs = NULL;
785 		npages = 0;
786 	}
787 
788 	/*
789 	 * now insert us in the proper place in vm_physmem[]
790 	 */
791 
792 #if (VM_PHYSSEG_STRAT == VM_PSTRAT_RANDOM)
793 	/* random: put it at the end (easy!) */
794 	ps = VM_PHYSMEM_PTR(vm_nphysmem);
795 #elif (VM_PHYSSEG_STRAT == VM_PSTRAT_BSEARCH)
796 	{
797 		int x;
798 		/* sort by address for binary search */
799 		for (lcv = 0 ; lcv < vm_nphysmem ; lcv++)
800 			if (start < VM_PHYSMEM_PTR(lcv)->start)
801 				break;
802 		ps = VM_PHYSMEM_PTR(lcv);
803 		/* move back other entries, if necessary ... */
804 		for (x = vm_nphysmem ; x > lcv ; x--)
805 			/* structure copy */
806 			VM_PHYSMEM_PTR_SWAP(x, x - 1);
807 	}
808 #elif (VM_PHYSSEG_STRAT == VM_PSTRAT_BIGFIRST)
809 	{
810 		int x;
811 		/* sort by largest segment first */
812 		for (lcv = 0 ; lcv < vm_nphysmem ; lcv++)
813 			if ((end - start) >
814 			    (VM_PHYSMEM_PTR(lcv)->end - VM_PHYSMEM_PTR(lcv)->start))
815 				break;
816 		ps = VM_PHYSMEM_PTR(lcv);
817 		/* move back other entries, if necessary ... */
818 		for (x = vm_nphysmem ; x > lcv ; x--)
819 			/* structure copy */
820 			VM_PHYSMEM_PTR_SWAP(x, x - 1);
821 	}
822 #else
823 	panic("uvm_page_physload: unknown physseg strategy selected!");
824 #endif
825 
826 	ps->start = start;
827 	ps->end = end;
828 	ps->avail_start = avail_start;
829 	ps->avail_end = avail_end;
830 	if (preload) {
831 		ps->pgs = NULL;
832 	} else {
833 		ps->pgs = pgs;
834 		ps->lastpg = pgs + npages;
835 	}
836 	ps->free_list = free_list;
837 	vm_nphysmem++;
838 
839 	if (!preload) {
840 		uvmpdpol_reinit();
841 	}
842 }
843 
844 /*
845  * when VM_PHYSSEG_MAX is 1, we can simplify these functions
846  */
847 
848 #if VM_PHYSSEG_MAX == 1
849 static inline int vm_physseg_find_contig(struct vm_physseg *, int, paddr_t, int *);
850 #elif (VM_PHYSSEG_STRAT == VM_PSTRAT_BSEARCH)
851 static inline int vm_physseg_find_bsearch(struct vm_physseg *, int, paddr_t, int *);
852 #else
853 static inline int vm_physseg_find_linear(struct vm_physseg *, int, paddr_t, int *);
854 #endif
855 
856 /*
857  * vm_physseg_find: find vm_physseg structure that belongs to a PA
858  */
859 int
860 vm_physseg_find(paddr_t pframe, int *offp)
861 {
862 
863 #if VM_PHYSSEG_MAX == 1
864 	return vm_physseg_find_contig(vm_physmem, vm_nphysseg, pframe, offp);
865 #elif (VM_PHYSSEG_STRAT == VM_PSTRAT_BSEARCH)
866 	return vm_physseg_find_bsearch(vm_physmem, vm_nphysseg, pframe, offp);
867 #else
868 	return vm_physseg_find_linear(vm_physmem, vm_nphysseg, pframe, offp);
869 #endif
870 }
871 
872 #if VM_PHYSSEG_MAX == 1
873 static inline int
874 vm_physseg_find_contig(struct vm_physseg *segs, int nsegs, paddr_t pframe, int *offp)
875 {
876 
877 	/* 'contig' case */
878 	if (pframe >= segs[0].start && pframe < segs[0].end) {
879 		if (offp)
880 			*offp = pframe - segs[0].start;
881 		return(0);
882 	}
883 	return(-1);
884 }
885 
886 #elif (VM_PHYSSEG_STRAT == VM_PSTRAT_BSEARCH)
887 
888 static inline int
889 vm_physseg_find_bsearch(struct vm_physseg *segs, int nsegs, paddr_t pframe, int *offp)
890 {
891 	/* binary search for it */
892 	u_int	start, len, try;
893 
894 	/*
895 	 * if try is too large (thus target is less than try) we reduce
896 	 * the length to trunc(len/2) [i.e. everything smaller than "try"]
897 	 *
898 	 * if the try is too small (thus target is greater than try) then
899 	 * we set the new start to be (try + 1).   this means we need to
900 	 * reduce the length to (round(len/2) - 1).
901 	 *
902 	 * note "adjust" below which takes advantage of the fact that
903 	 *  (round(len/2) - 1) == trunc((len - 1) / 2)
904 	 * for any value of len we may have
905 	 */
906 
907 	for (start = 0, len = nsegs ; len != 0 ; len = len / 2) {
908 		try = start + (len / 2);	/* try in the middle */
909 
910 		/* start past our try? */
911 		if (pframe >= segs[try].start) {
912 			/* was try correct? */
913 			if (pframe < segs[try].end) {
914 				if (offp)
915 					*offp = pframe - segs[try].start;
916 				return(try);            /* got it */
917 			}
918 			start = try + 1;	/* next time, start here */
919 			len--;			/* "adjust" */
920 		} else {
921 			/*
922 			 * pframe before try, just reduce length of
923 			 * region, done in "for" loop
924 			 */
925 		}
926 	}
927 	return(-1);
928 }
929 
930 #else
931 
932 static inline int
933 vm_physseg_find_linear(struct vm_physseg *segs, int nsegs, paddr_t pframe, int *offp)
934 {
935 	/* linear search for it */
936 	int	lcv;
937 
938 	for (lcv = 0; lcv < nsegs; lcv++) {
939 		if (pframe >= segs[lcv].start &&
940 		    pframe < segs[lcv].end) {
941 			if (offp)
942 				*offp = pframe - segs[lcv].start;
943 			return(lcv);		   /* got it */
944 		}
945 	}
946 	return(-1);
947 }
948 #endif
949 
950 /*
951  * PHYS_TO_VM_PAGE: find vm_page for a PA.   used by MI code to get vm_pages
952  * back from an I/O mapping (ugh!).   used in some MD code as well.
953  */
954 struct vm_page *
955 uvm_phys_to_vm_page(paddr_t pa)
956 {
957 	paddr_t pf = atop(pa);
958 	int	off;
959 	int	psi;
960 
961 	psi = vm_physseg_find(pf, &off);
962 	if (psi != -1)
963 		return(&VM_PHYSMEM_PTR(psi)->pgs[off]);
964 	return(NULL);
965 }
966 
967 paddr_t
968 uvm_vm_page_to_phys(const struct vm_page *pg)
969 {
970 
971 	return pg->phys_addr;
972 }
973 
974 /*
975  * uvm_page_recolor: Recolor the pages if the new bucket count is
976  * larger than the old one.
977  */
978 
979 void
980 uvm_page_recolor(int newncolors)
981 {
982 	struct pgflbucket *bucketarray, *cpuarray, *oldbucketarray;
983 	struct pgfreelist gpgfl, pgfl;
984 	struct vm_page *pg;
985 	vsize_t bucketcount;
986 	int lcv, color, i, ocolors;
987 	struct uvm_cpu *ucpu;
988 
989 	if (newncolors <= uvmexp.ncolors)
990 		return;
991 
992 	if (uvm.page_init_done == false) {
993 		uvmexp.ncolors = newncolors;
994 		return;
995 	}
996 
997 	bucketcount = newncolors * VM_NFREELIST;
998 	bucketarray = malloc(bucketcount * sizeof(struct pgflbucket) * 2,
999 	    M_VMPAGE, M_NOWAIT);
1000 	cpuarray = bucketarray + bucketcount;
1001 	if (bucketarray == NULL) {
1002 		printf("WARNING: unable to allocate %ld page color buckets\n",
1003 		    (long) bucketcount);
1004 		return;
1005 	}
1006 
1007 	mutex_spin_enter(&uvm_fpageqlock);
1008 
1009 	/* Make sure we should still do this. */
1010 	if (newncolors <= uvmexp.ncolors) {
1011 		mutex_spin_exit(&uvm_fpageqlock);
1012 		free(bucketarray, M_VMPAGE);
1013 		return;
1014 	}
1015 
1016 	oldbucketarray = uvm.page_free[0].pgfl_buckets;
1017 	ocolors = uvmexp.ncolors;
1018 
1019 	uvmexp.ncolors = newncolors;
1020 	uvmexp.colormask = uvmexp.ncolors - 1;
1021 
1022 	ucpu = curcpu()->ci_data.cpu_uvm;
1023 	for (lcv = 0; lcv < VM_NFREELIST; lcv++) {
1024 		gpgfl.pgfl_buckets = (bucketarray + (lcv * newncolors));
1025 		pgfl.pgfl_buckets = (cpuarray + (lcv * uvmexp.ncolors));
1026 		uvm_page_init_buckets(&gpgfl);
1027 		uvm_page_init_buckets(&pgfl);
1028 		for (color = 0; color < ocolors; color++) {
1029 			for (i = 0; i < PGFL_NQUEUES; i++) {
1030 				while ((pg = LIST_FIRST(&uvm.page_free[
1031 				    lcv].pgfl_buckets[color].pgfl_queues[i]))
1032 				    != NULL) {
1033 					LIST_REMOVE(pg, pageq.list); /* global */
1034 					LIST_REMOVE(pg, listq.list); /* cpu */
1035 					LIST_INSERT_HEAD(&gpgfl.pgfl_buckets[
1036 					    VM_PGCOLOR_BUCKET(pg)].pgfl_queues[
1037 					    i], pg, pageq.list);
1038 					LIST_INSERT_HEAD(&pgfl.pgfl_buckets[
1039 					    VM_PGCOLOR_BUCKET(pg)].pgfl_queues[
1040 					    i], pg, listq.list);
1041 				}
1042 			}
1043 		}
1044 		uvm.page_free[lcv].pgfl_buckets = gpgfl.pgfl_buckets;
1045 		ucpu->page_free[lcv].pgfl_buckets = pgfl.pgfl_buckets;
1046 	}
1047 
1048 	if (have_recolored_pages) {
1049 		mutex_spin_exit(&uvm_fpageqlock);
1050 		free(oldbucketarray, M_VMPAGE);
1051 		return;
1052 	}
1053 
1054 	have_recolored_pages = true;
1055 	mutex_spin_exit(&uvm_fpageqlock);
1056 }
1057 
1058 /*
1059  * uvm_cpu_attach: initialize per-CPU data structures.
1060  */
1061 
1062 void
1063 uvm_cpu_attach(struct cpu_info *ci)
1064 {
1065 	struct pgflbucket *bucketarray;
1066 	struct pgfreelist pgfl;
1067 	struct uvm_cpu *ucpu;
1068 	vsize_t bucketcount;
1069 	int lcv;
1070 
1071 	if (CPU_IS_PRIMARY(ci)) {
1072 		/* Already done in uvm_page_init(). */
1073 		return;
1074 	}
1075 
1076 	/* Add more reserve pages for this CPU. */
1077 	uvmexp.reserve_kernel += vm_page_reserve_kernel;
1078 
1079 	/* Configure this CPU's free lists. */
1080 	bucketcount = uvmexp.ncolors * VM_NFREELIST;
1081 	bucketarray = malloc(bucketcount * sizeof(struct pgflbucket),
1082 	    M_VMPAGE, M_WAITOK);
1083 	ucpu = kmem_zalloc(sizeof(*ucpu), KM_SLEEP);
1084 	uvm.cpus[cpu_index(ci)] = ucpu;
1085 	ci->ci_data.cpu_uvm = ucpu;
1086 	for (lcv = 0; lcv < VM_NFREELIST; lcv++) {
1087 		pgfl.pgfl_buckets = (bucketarray + (lcv * uvmexp.ncolors));
1088 		uvm_page_init_buckets(&pgfl);
1089 		ucpu->page_free[lcv].pgfl_buckets = pgfl.pgfl_buckets;
1090 	}
1091 }
1092 
1093 /*
1094  * uvm_pagealloc_pgfl: helper routine for uvm_pagealloc_strat
1095  */
1096 
1097 static struct vm_page *
1098 uvm_pagealloc_pgfl(struct uvm_cpu *ucpu, int flist, int try1, int try2,
1099     int *trycolorp)
1100 {
1101 	struct pgflist *freeq;
1102 	struct vm_page *pg;
1103 	int color, trycolor = *trycolorp;
1104 	struct pgfreelist *gpgfl, *pgfl;
1105 
1106 	KASSERT(mutex_owned(&uvm_fpageqlock));
1107 
1108 	color = trycolor;
1109 	pgfl = &ucpu->page_free[flist];
1110 	gpgfl = &uvm.page_free[flist];
1111 	do {
1112 		/* cpu, try1 */
1113 		if ((pg = LIST_FIRST((freeq =
1114 		    &pgfl->pgfl_buckets[color].pgfl_queues[try1]))) != NULL) {
1115 			VM_FREE_PAGE_TO_CPU(pg)->pages[try1]--;
1116 		    	uvmexp.cpuhit++;
1117 			goto gotit;
1118 		}
1119 		/* global, try1 */
1120 		if ((pg = LIST_FIRST((freeq =
1121 		    &gpgfl->pgfl_buckets[color].pgfl_queues[try1]))) != NULL) {
1122 			VM_FREE_PAGE_TO_CPU(pg)->pages[try1]--;
1123 		    	uvmexp.cpumiss++;
1124 			goto gotit;
1125 		}
1126 		/* cpu, try2 */
1127 		if ((pg = LIST_FIRST((freeq =
1128 		    &pgfl->pgfl_buckets[color].pgfl_queues[try2]))) != NULL) {
1129 			VM_FREE_PAGE_TO_CPU(pg)->pages[try2]--;
1130 		    	uvmexp.cpuhit++;
1131 			goto gotit;
1132 		}
1133 		/* global, try2 */
1134 		if ((pg = LIST_FIRST((freeq =
1135 		    &gpgfl->pgfl_buckets[color].pgfl_queues[try2]))) != NULL) {
1136 			VM_FREE_PAGE_TO_CPU(pg)->pages[try2]--;
1137 		    	uvmexp.cpumiss++;
1138 			goto gotit;
1139 		}
1140 		color = (color + 1) & uvmexp.colormask;
1141 	} while (color != trycolor);
1142 
1143 	return (NULL);
1144 
1145  gotit:
1146 	LIST_REMOVE(pg, pageq.list);	/* global list */
1147 	LIST_REMOVE(pg, listq.list);	/* per-cpu list */
1148 	uvmexp.free--;
1149 
1150 	/* update zero'd page count */
1151 	if (pg->flags & PG_ZERO)
1152 		uvmexp.zeropages--;
1153 
1154 	if (color == trycolor)
1155 		uvmexp.colorhit++;
1156 	else {
1157 		uvmexp.colormiss++;
1158 		*trycolorp = color;
1159 	}
1160 
1161 	return (pg);
1162 }
1163 
1164 /*
1165  * uvm_pagealloc_strat: allocate vm_page from a particular free list.
1166  *
1167  * => return null if no pages free
1168  * => wake up pagedaemon if number of free pages drops below low water mark
1169  * => if obj != NULL, obj must be locked (to put in obj's tree)
1170  * => if anon != NULL, anon must be locked (to put in anon)
1171  * => only one of obj or anon can be non-null
1172  * => caller must activate/deactivate page if it is not wired.
1173  * => free_list is ignored if strat == UVM_PGA_STRAT_NORMAL.
1174  * => policy decision: it is more important to pull a page off of the
1175  *	appropriate priority free list than it is to get a zero'd or
1176  *	unknown contents page.  This is because we live with the
1177  *	consequences of a bad free list decision for the entire
1178  *	lifetime of the page, e.g. if the page comes from memory that
1179  *	is slower to access.
1180  */
1181 
1182 struct vm_page *
1183 uvm_pagealloc_strat(struct uvm_object *obj, voff_t off, struct vm_anon *anon,
1184     int flags, int strat, int free_list)
1185 {
1186 	int lcv, try1, try2, zeroit = 0, color;
1187 	struct uvm_cpu *ucpu;
1188 	struct vm_page *pg;
1189 	lwp_t *l;
1190 
1191 	KASSERT(obj == NULL || anon == NULL);
1192 	KASSERT(anon == NULL || (flags & UVM_FLAG_COLORMATCH) || off == 0);
1193 	KASSERT(off == trunc_page(off));
1194 	KASSERT(obj == NULL || mutex_owned(obj->vmobjlock));
1195 	KASSERT(anon == NULL || anon->an_lock == NULL ||
1196 	    mutex_owned(anon->an_lock));
1197 
1198 	mutex_spin_enter(&uvm_fpageqlock);
1199 
1200 	/*
1201 	 * This implements a global round-robin page coloring
1202 	 * algorithm.
1203 	 */
1204 
1205 	ucpu = curcpu()->ci_data.cpu_uvm;
1206 	if (flags & UVM_FLAG_COLORMATCH) {
1207 		color = atop(off) & uvmexp.colormask;
1208 	} else {
1209 		color = ucpu->page_free_nextcolor;
1210 	}
1211 
1212 	/*
1213 	 * check to see if we need to generate some free pages waking
1214 	 * the pagedaemon.
1215 	 */
1216 
1217 	uvm_kick_pdaemon();
1218 
1219 	/*
1220 	 * fail if any of these conditions is true:
1221 	 * [1]  there really are no free pages, or
1222 	 * [2]  only kernel "reserved" pages remain and
1223 	 *        reserved pages have not been requested.
1224 	 * [3]  only pagedaemon "reserved" pages remain and
1225 	 *        the requestor isn't the pagedaemon.
1226 	 * we make kernel reserve pages available if called by a
1227 	 * kernel thread or a realtime thread.
1228 	 */
1229 	l = curlwp;
1230 	if (__predict_true(l != NULL) && lwp_eprio(l) >= PRI_KTHREAD) {
1231 		flags |= UVM_PGA_USERESERVE;
1232 	}
1233 	if ((uvmexp.free <= uvmexp.reserve_kernel &&
1234 	    (flags & UVM_PGA_USERESERVE) == 0) ||
1235 	    (uvmexp.free <= uvmexp.reserve_pagedaemon &&
1236 	     curlwp != uvm.pagedaemon_lwp))
1237 		goto fail;
1238 
1239 #if PGFL_NQUEUES != 2
1240 #error uvm_pagealloc_strat needs to be updated
1241 #endif
1242 
1243 	/*
1244 	 * If we want a zero'd page, try the ZEROS queue first, otherwise
1245 	 * we try the UNKNOWN queue first.
1246 	 */
1247 	if (flags & UVM_PGA_ZERO) {
1248 		try1 = PGFL_ZEROS;
1249 		try2 = PGFL_UNKNOWN;
1250 	} else {
1251 		try1 = PGFL_UNKNOWN;
1252 		try2 = PGFL_ZEROS;
1253 	}
1254 
1255  again:
1256 	switch (strat) {
1257 	case UVM_PGA_STRAT_NORMAL:
1258 		/* Check freelists: descending priority (ascending id) order */
1259 		for (lcv = 0; lcv < VM_NFREELIST; lcv++) {
1260 			pg = uvm_pagealloc_pgfl(ucpu, lcv,
1261 			    try1, try2, &color);
1262 			if (pg != NULL)
1263 				goto gotit;
1264 		}
1265 
1266 		/* No pages free! */
1267 		goto fail;
1268 
1269 	case UVM_PGA_STRAT_ONLY:
1270 	case UVM_PGA_STRAT_FALLBACK:
1271 		/* Attempt to allocate from the specified free list. */
1272 		KASSERT(free_list >= 0 && free_list < VM_NFREELIST);
1273 		pg = uvm_pagealloc_pgfl(ucpu, free_list,
1274 		    try1, try2, &color);
1275 		if (pg != NULL)
1276 			goto gotit;
1277 
1278 		/* Fall back, if possible. */
1279 		if (strat == UVM_PGA_STRAT_FALLBACK) {
1280 			strat = UVM_PGA_STRAT_NORMAL;
1281 			goto again;
1282 		}
1283 
1284 		/* No pages free! */
1285 		goto fail;
1286 
1287 	default:
1288 		panic("uvm_pagealloc_strat: bad strat %d", strat);
1289 		/* NOTREACHED */
1290 	}
1291 
1292  gotit:
1293 	/*
1294 	 * We now know which color we actually allocated from; set
1295 	 * the next color accordingly.
1296 	 */
1297 
1298 	ucpu->page_free_nextcolor = (color + 1) & uvmexp.colormask;
1299 
1300 	/*
1301 	 * update allocation statistics and remember if we have to
1302 	 * zero the page
1303 	 */
1304 
1305 	if (flags & UVM_PGA_ZERO) {
1306 		if (pg->flags & PG_ZERO) {
1307 			uvmexp.pga_zerohit++;
1308 			zeroit = 0;
1309 		} else {
1310 			uvmexp.pga_zeromiss++;
1311 			zeroit = 1;
1312 		}
1313 		if (ucpu->pages[PGFL_ZEROS] < ucpu->pages[PGFL_UNKNOWN]) {
1314 			ucpu->page_idle_zero = vm_page_zero_enable;
1315 		}
1316 	}
1317 	KASSERT(pg->pqflags == PQ_FREE);
1318 
1319 	pg->offset = off;
1320 	pg->uobject = obj;
1321 	pg->uanon = anon;
1322 	pg->flags = PG_BUSY|PG_CLEAN|PG_FAKE;
1323 	if (anon) {
1324 		anon->an_page = pg;
1325 		pg->pqflags = PQ_ANON;
1326 		atomic_inc_uint(&uvmexp.anonpages);
1327 	} else {
1328 		if (obj) {
1329 			uvm_pageinsert(obj, pg);
1330 		}
1331 		pg->pqflags = 0;
1332 	}
1333 	mutex_spin_exit(&uvm_fpageqlock);
1334 
1335 #if defined(UVM_PAGE_TRKOWN)
1336 	pg->owner_tag = NULL;
1337 #endif
1338 	UVM_PAGE_OWN(pg, "new alloc");
1339 
1340 	if (flags & UVM_PGA_ZERO) {
1341 		/*
1342 		 * A zero'd page is not clean.  If we got a page not already
1343 		 * zero'd, then we have to zero it ourselves.
1344 		 */
1345 		pg->flags &= ~PG_CLEAN;
1346 		if (zeroit)
1347 			pmap_zero_page(VM_PAGE_TO_PHYS(pg));
1348 	}
1349 
1350 	return(pg);
1351 
1352  fail:
1353 	mutex_spin_exit(&uvm_fpageqlock);
1354 	return (NULL);
1355 }
1356 
1357 /*
1358  * uvm_pagereplace: replace a page with another
1359  *
1360  * => object must be locked
1361  */
1362 
1363 void
1364 uvm_pagereplace(struct vm_page *oldpg, struct vm_page *newpg)
1365 {
1366 	struct uvm_object *uobj = oldpg->uobject;
1367 
1368 	KASSERT((oldpg->flags & PG_TABLED) != 0);
1369 	KASSERT(uobj != NULL);
1370 	KASSERT((newpg->flags & PG_TABLED) == 0);
1371 	KASSERT(newpg->uobject == NULL);
1372 	KASSERT(mutex_owned(uobj->vmobjlock));
1373 
1374 	newpg->uobject = uobj;
1375 	newpg->offset = oldpg->offset;
1376 
1377 	uvm_pageremove_tree(uobj, oldpg);
1378 	uvm_pageinsert_tree(uobj, newpg);
1379 	uvm_pageinsert_list(uobj, newpg, oldpg);
1380 	uvm_pageremove_list(uobj, oldpg);
1381 }
1382 
1383 /*
1384  * uvm_pagerealloc: reallocate a page from one object to another
1385  *
1386  * => both objects must be locked
1387  */
1388 
1389 void
1390 uvm_pagerealloc(struct vm_page *pg, struct uvm_object *newobj, voff_t newoff)
1391 {
1392 	/*
1393 	 * remove it from the old object
1394 	 */
1395 
1396 	if (pg->uobject) {
1397 		uvm_pageremove(pg->uobject, pg);
1398 	}
1399 
1400 	/*
1401 	 * put it in the new object
1402 	 */
1403 
1404 	if (newobj) {
1405 		pg->uobject = newobj;
1406 		pg->offset = newoff;
1407 		uvm_pageinsert(newobj, pg);
1408 	}
1409 }
1410 
1411 #ifdef DEBUG
1412 /*
1413  * check if page is zero-filled
1414  *
1415  *  - called with free page queue lock held.
1416  */
1417 void
1418 uvm_pagezerocheck(struct vm_page *pg)
1419 {
1420 	int *p, *ep;
1421 
1422 	KASSERT(uvm_zerocheckkva != 0);
1423 	KASSERT(mutex_owned(&uvm_fpageqlock));
1424 
1425 	/*
1426 	 * XXX assuming pmap_kenter_pa and pmap_kremove never call
1427 	 * uvm page allocator.
1428 	 *
1429 	 * it might be better to have "CPU-local temporary map" pmap interface.
1430 	 */
1431 	pmap_kenter_pa(uvm_zerocheckkva, VM_PAGE_TO_PHYS(pg), VM_PROT_READ, 0);
1432 	p = (int *)uvm_zerocheckkva;
1433 	ep = (int *)((char *)p + PAGE_SIZE);
1434 	pmap_update(pmap_kernel());
1435 	while (p < ep) {
1436 		if (*p != 0)
1437 			panic("PG_ZERO page isn't zero-filled");
1438 		p++;
1439 	}
1440 	pmap_kremove(uvm_zerocheckkva, PAGE_SIZE);
1441 	/*
1442 	 * pmap_update() is not necessary here because no one except us
1443 	 * uses this VA.
1444 	 */
1445 }
1446 #endif /* DEBUG */
1447 
1448 /*
1449  * uvm_pagefree: free page
1450  *
1451  * => erase page's identity (i.e. remove from object)
1452  * => put page on free list
1453  * => caller must lock owning object (either anon or uvm_object)
1454  * => caller must lock page queues
1455  * => assumes all valid mappings of pg are gone
1456  */
1457 
1458 void
1459 uvm_pagefree(struct vm_page *pg)
1460 {
1461 	struct pgflist *pgfl;
1462 	struct uvm_cpu *ucpu;
1463 	int index, color, queue;
1464 	bool iszero;
1465 
1466 #ifdef DEBUG
1467 	if (pg->uobject == (void *)0xdeadbeef &&
1468 	    pg->uanon == (void *)0xdeadbeef) {
1469 		panic("uvm_pagefree: freeing free page %p", pg);
1470 	}
1471 #endif /* DEBUG */
1472 
1473 	KASSERT((pg->flags & PG_PAGEOUT) == 0);
1474 	KASSERT(!(pg->pqflags & PQ_FREE));
1475 	KASSERT(mutex_owned(&uvm_pageqlock) || !uvmpdpol_pageisqueued_p(pg));
1476 	KASSERT(pg->uobject == NULL || mutex_owned(pg->uobject->vmobjlock));
1477 	KASSERT(pg->uobject != NULL || pg->uanon == NULL ||
1478 		mutex_owned(pg->uanon->an_lock));
1479 
1480 	/*
1481 	 * if the page is loaned, resolve the loan instead of freeing.
1482 	 */
1483 
1484 	if (pg->loan_count) {
1485 		KASSERT(pg->wire_count == 0);
1486 
1487 		/*
1488 		 * if the page is owned by an anon then we just want to
1489 		 * drop anon ownership.  the kernel will free the page when
1490 		 * it is done with it.  if the page is owned by an object,
1491 		 * remove it from the object and mark it dirty for the benefit
1492 		 * of possible anon owners.
1493 		 *
1494 		 * regardless of previous ownership, wakeup any waiters,
1495 		 * unbusy the page, and we're done.
1496 		 */
1497 
1498 		if (pg->uobject != NULL) {
1499 			uvm_pageremove(pg->uobject, pg);
1500 			pg->flags &= ~PG_CLEAN;
1501 		} else if (pg->uanon != NULL) {
1502 			if ((pg->pqflags & PQ_ANON) == 0) {
1503 				pg->loan_count--;
1504 			} else {
1505 				pg->pqflags &= ~PQ_ANON;
1506 				atomic_dec_uint(&uvmexp.anonpages);
1507 			}
1508 			pg->uanon->an_page = NULL;
1509 			pg->uanon = NULL;
1510 		}
1511 		if (pg->flags & PG_WANTED) {
1512 			wakeup(pg);
1513 		}
1514 		pg->flags &= ~(PG_WANTED|PG_BUSY|PG_RELEASED|PG_PAGER1);
1515 #ifdef UVM_PAGE_TRKOWN
1516 		pg->owner_tag = NULL;
1517 #endif
1518 		if (pg->loan_count) {
1519 			KASSERT(pg->uobject == NULL);
1520 			if (pg->uanon == NULL) {
1521 				uvm_pagedequeue(pg);
1522 			}
1523 			return;
1524 		}
1525 	}
1526 
1527 	/*
1528 	 * remove page from its object or anon.
1529 	 */
1530 
1531 	if (pg->uobject != NULL) {
1532 		uvm_pageremove(pg->uobject, pg);
1533 	} else if (pg->uanon != NULL) {
1534 		pg->uanon->an_page = NULL;
1535 		atomic_dec_uint(&uvmexp.anonpages);
1536 	}
1537 
1538 	/*
1539 	 * now remove the page from the queues.
1540 	 */
1541 
1542 	uvm_pagedequeue(pg);
1543 
1544 	/*
1545 	 * if the page was wired, unwire it now.
1546 	 */
1547 
1548 	if (pg->wire_count) {
1549 		pg->wire_count = 0;
1550 		uvmexp.wired--;
1551 	}
1552 
1553 	/*
1554 	 * and put on free queue
1555 	 */
1556 
1557 	iszero = (pg->flags & PG_ZERO);
1558 	index = uvm_page_lookup_freelist(pg);
1559 	color = VM_PGCOLOR_BUCKET(pg);
1560 	queue = (iszero ? PGFL_ZEROS : PGFL_UNKNOWN);
1561 
1562 #ifdef DEBUG
1563 	pg->uobject = (void *)0xdeadbeef;
1564 	pg->uanon = (void *)0xdeadbeef;
1565 #endif
1566 
1567 	mutex_spin_enter(&uvm_fpageqlock);
1568 	pg->pqflags = PQ_FREE;
1569 
1570 #ifdef DEBUG
1571 	if (iszero)
1572 		uvm_pagezerocheck(pg);
1573 #endif /* DEBUG */
1574 
1575 
1576 	/* global list */
1577 	pgfl = &uvm.page_free[index].pgfl_buckets[color].pgfl_queues[queue];
1578 	LIST_INSERT_HEAD(pgfl, pg, pageq.list);
1579 	uvmexp.free++;
1580 	if (iszero) {
1581 		uvmexp.zeropages++;
1582 	}
1583 
1584 	/* per-cpu list */
1585 	ucpu = curcpu()->ci_data.cpu_uvm;
1586 	pg->offset = (uintptr_t)ucpu;
1587 	pgfl = &ucpu->page_free[index].pgfl_buckets[color].pgfl_queues[queue];
1588 	LIST_INSERT_HEAD(pgfl, pg, listq.list);
1589 	ucpu->pages[queue]++;
1590 	if (ucpu->pages[PGFL_ZEROS] < ucpu->pages[PGFL_UNKNOWN]) {
1591 		ucpu->page_idle_zero = vm_page_zero_enable;
1592 	}
1593 
1594 	mutex_spin_exit(&uvm_fpageqlock);
1595 }
1596 
1597 /*
1598  * uvm_page_unbusy: unbusy an array of pages.
1599  *
1600  * => pages must either all belong to the same object, or all belong to anons.
1601  * => if pages are object-owned, object must be locked.
1602  * => if pages are anon-owned, anons must be locked.
1603  * => caller must lock page queues if pages may be released.
1604  * => caller must make sure that anon-owned pages are not PG_RELEASED.
1605  */
1606 
1607 void
1608 uvm_page_unbusy(struct vm_page **pgs, int npgs)
1609 {
1610 	struct vm_page *pg;
1611 	int i;
1612 	UVMHIST_FUNC("uvm_page_unbusy"); UVMHIST_CALLED(ubchist);
1613 
1614 	for (i = 0; i < npgs; i++) {
1615 		pg = pgs[i];
1616 		if (pg == NULL || pg == PGO_DONTCARE) {
1617 			continue;
1618 		}
1619 
1620 		KASSERT(pg->uobject == NULL ||
1621 		    mutex_owned(pg->uobject->vmobjlock));
1622 		KASSERT(pg->uobject != NULL ||
1623 		    (pg->uanon != NULL && mutex_owned(pg->uanon->an_lock)));
1624 
1625 		KASSERT(pg->flags & PG_BUSY);
1626 		KASSERT((pg->flags & PG_PAGEOUT) == 0);
1627 		if (pg->flags & PG_WANTED) {
1628 			wakeup(pg);
1629 		}
1630 		if (pg->flags & PG_RELEASED) {
1631 			UVMHIST_LOG(ubchist, "releasing pg %p", pg,0,0,0);
1632 			KASSERT(pg->uobject != NULL ||
1633 			    (pg->uanon != NULL && pg->uanon->an_ref > 0));
1634 			pg->flags &= ~PG_RELEASED;
1635 			uvm_pagefree(pg);
1636 		} else {
1637 			UVMHIST_LOG(ubchist, "unbusying pg %p", pg,0,0,0);
1638 			KASSERT((pg->flags & PG_FAKE) == 0);
1639 			pg->flags &= ~(PG_WANTED|PG_BUSY);
1640 			UVM_PAGE_OWN(pg, NULL);
1641 		}
1642 	}
1643 }
1644 
1645 #if defined(UVM_PAGE_TRKOWN)
1646 /*
1647  * uvm_page_own: set or release page ownership
1648  *
1649  * => this is a debugging function that keeps track of who sets PG_BUSY
1650  *	and where they do it.   it can be used to track down problems
1651  *	such a process setting "PG_BUSY" and never releasing it.
1652  * => page's object [if any] must be locked
1653  * => if "tag" is NULL then we are releasing page ownership
1654  */
1655 void
1656 uvm_page_own(struct vm_page *pg, const char *tag)
1657 {
1658 	struct uvm_object *uobj;
1659 	struct vm_anon *anon;
1660 
1661 	KASSERT((pg->flags & (PG_PAGEOUT|PG_RELEASED)) == 0);
1662 
1663 	uobj = pg->uobject;
1664 	anon = pg->uanon;
1665 	if (uobj != NULL) {
1666 		KASSERT(mutex_owned(uobj->vmobjlock));
1667 	} else if (anon != NULL) {
1668 		KASSERT(mutex_owned(anon->an_lock));
1669 	}
1670 
1671 	KASSERT((pg->flags & PG_WANTED) == 0);
1672 
1673 	/* gain ownership? */
1674 	if (tag) {
1675 		KASSERT((pg->flags & PG_BUSY) != 0);
1676 		if (pg->owner_tag) {
1677 			printf("uvm_page_own: page %p already owned "
1678 			    "by proc %d [%s]\n", pg,
1679 			    pg->owner, pg->owner_tag);
1680 			panic("uvm_page_own");
1681 		}
1682 		pg->owner = (curproc) ? curproc->p_pid :  (pid_t) -1;
1683 		pg->lowner = (curlwp) ? curlwp->l_lid :  (lwpid_t) -1;
1684 		pg->owner_tag = tag;
1685 		return;
1686 	}
1687 
1688 	/* drop ownership */
1689 	KASSERT((pg->flags & PG_BUSY) == 0);
1690 	if (pg->owner_tag == NULL) {
1691 		printf("uvm_page_own: dropping ownership of an non-owned "
1692 		    "page (%p)\n", pg);
1693 		panic("uvm_page_own");
1694 	}
1695 	if (!uvmpdpol_pageisqueued_p(pg)) {
1696 		KASSERT((pg->uanon == NULL && pg->uobject == NULL) ||
1697 		    pg->wire_count > 0);
1698 	} else {
1699 		KASSERT(pg->wire_count == 0);
1700 	}
1701 	pg->owner_tag = NULL;
1702 }
1703 #endif
1704 
1705 /*
1706  * uvm_pageidlezero: zero free pages while the system is idle.
1707  *
1708  * => try to complete one color bucket at a time, to reduce our impact
1709  *	on the CPU cache.
1710  * => we loop until we either reach the target or there is a lwp ready
1711  *      to run, or MD code detects a reason to break early.
1712  */
1713 void
1714 uvm_pageidlezero(void)
1715 {
1716 	struct vm_page *pg;
1717 	struct pgfreelist *pgfl, *gpgfl;
1718 	struct uvm_cpu *ucpu;
1719 	int free_list, firstbucket, nextbucket;
1720 	bool lcont = false;
1721 
1722 	ucpu = curcpu()->ci_data.cpu_uvm;
1723 	if (!ucpu->page_idle_zero ||
1724 	    ucpu->pages[PGFL_UNKNOWN] < uvmexp.ncolors) {
1725 	    	ucpu->page_idle_zero = false;
1726 		return;
1727 	}
1728 	if (!mutex_tryenter(&uvm_fpageqlock)) {
1729 		/* Contention: let other CPUs to use the lock. */
1730 		return;
1731 	}
1732 	firstbucket = ucpu->page_free_nextcolor;
1733 	nextbucket = firstbucket;
1734 	do {
1735 		for (free_list = 0; free_list < VM_NFREELIST; free_list++) {
1736 			if (sched_curcpu_runnable_p()) {
1737 				goto quit;
1738 			}
1739 			pgfl = &ucpu->page_free[free_list];
1740 			gpgfl = &uvm.page_free[free_list];
1741 			while ((pg = LIST_FIRST(&pgfl->pgfl_buckets[
1742 			    nextbucket].pgfl_queues[PGFL_UNKNOWN])) != NULL) {
1743 				if (lcont || sched_curcpu_runnable_p()) {
1744 					goto quit;
1745 				}
1746 				LIST_REMOVE(pg, pageq.list); /* global list */
1747 				LIST_REMOVE(pg, listq.list); /* per-cpu list */
1748 				ucpu->pages[PGFL_UNKNOWN]--;
1749 				uvmexp.free--;
1750 				KASSERT(pg->pqflags == PQ_FREE);
1751 				pg->pqflags = 0;
1752 				mutex_spin_exit(&uvm_fpageqlock);
1753 #ifdef PMAP_PAGEIDLEZERO
1754 				if (!PMAP_PAGEIDLEZERO(VM_PAGE_TO_PHYS(pg))) {
1755 
1756 					/*
1757 					 * The machine-dependent code detected
1758 					 * some reason for us to abort zeroing
1759 					 * pages, probably because there is a
1760 					 * process now ready to run.
1761 					 */
1762 
1763 					mutex_spin_enter(&uvm_fpageqlock);
1764 					pg->pqflags = PQ_FREE;
1765 					LIST_INSERT_HEAD(&gpgfl->pgfl_buckets[
1766 					    nextbucket].pgfl_queues[
1767 					    PGFL_UNKNOWN], pg, pageq.list);
1768 					LIST_INSERT_HEAD(&pgfl->pgfl_buckets[
1769 					    nextbucket].pgfl_queues[
1770 					    PGFL_UNKNOWN], pg, listq.list);
1771 					ucpu->pages[PGFL_UNKNOWN]++;
1772 					uvmexp.free++;
1773 					uvmexp.zeroaborts++;
1774 					goto quit;
1775 				}
1776 #else
1777 				pmap_zero_page(VM_PAGE_TO_PHYS(pg));
1778 #endif /* PMAP_PAGEIDLEZERO */
1779 				pg->flags |= PG_ZERO;
1780 
1781 				if (!mutex_tryenter(&uvm_fpageqlock)) {
1782 					lcont = true;
1783 					mutex_spin_enter(&uvm_fpageqlock);
1784 				} else {
1785 					lcont = false;
1786 				}
1787 				pg->pqflags = PQ_FREE;
1788 				LIST_INSERT_HEAD(&gpgfl->pgfl_buckets[
1789 				    nextbucket].pgfl_queues[PGFL_ZEROS],
1790 				    pg, pageq.list);
1791 				LIST_INSERT_HEAD(&pgfl->pgfl_buckets[
1792 				    nextbucket].pgfl_queues[PGFL_ZEROS],
1793 				    pg, listq.list);
1794 				ucpu->pages[PGFL_ZEROS]++;
1795 				uvmexp.free++;
1796 				uvmexp.zeropages++;
1797 			}
1798 		}
1799 		if (ucpu->pages[PGFL_UNKNOWN] < uvmexp.ncolors) {
1800 			break;
1801 		}
1802 		nextbucket = (nextbucket + 1) & uvmexp.colormask;
1803 	} while (nextbucket != firstbucket);
1804 	ucpu->page_idle_zero = false;
1805  quit:
1806 	mutex_spin_exit(&uvm_fpageqlock);
1807 }
1808 
1809 /*
1810  * uvm_pagelookup: look up a page
1811  *
1812  * => caller should lock object to keep someone from pulling the page
1813  *	out from under it
1814  */
1815 
1816 struct vm_page *
1817 uvm_pagelookup(struct uvm_object *obj, voff_t off)
1818 {
1819 	struct vm_page *pg;
1820 
1821 	KASSERT(mutex_owned(obj->vmobjlock));
1822 
1823 	pg = rb_tree_find_node(&obj->rb_tree, &off);
1824 
1825 	KASSERT(pg == NULL || obj->uo_npages != 0);
1826 	KASSERT(pg == NULL || (pg->flags & (PG_RELEASED|PG_PAGEOUT)) == 0 ||
1827 		(pg->flags & PG_BUSY) != 0);
1828 	return pg;
1829 }
1830 
1831 /*
1832  * uvm_pagewire: wire the page, thus removing it from the daemon's grasp
1833  *
1834  * => caller must lock page queues
1835  */
1836 
1837 void
1838 uvm_pagewire(struct vm_page *pg)
1839 {
1840 	KASSERT(mutex_owned(&uvm_pageqlock));
1841 #if defined(READAHEAD_STATS)
1842 	if ((pg->pqflags & PQ_READAHEAD) != 0) {
1843 		uvm_ra_hit.ev_count++;
1844 		pg->pqflags &= ~PQ_READAHEAD;
1845 	}
1846 #endif /* defined(READAHEAD_STATS) */
1847 	if (pg->wire_count == 0) {
1848 		uvm_pagedequeue(pg);
1849 		uvmexp.wired++;
1850 	}
1851 	pg->wire_count++;
1852 }
1853 
1854 /*
1855  * uvm_pageunwire: unwire the page.
1856  *
1857  * => activate if wire count goes to zero.
1858  * => caller must lock page queues
1859  */
1860 
1861 void
1862 uvm_pageunwire(struct vm_page *pg)
1863 {
1864 	KASSERT(mutex_owned(&uvm_pageqlock));
1865 	pg->wire_count--;
1866 	if (pg->wire_count == 0) {
1867 		uvm_pageactivate(pg);
1868 		uvmexp.wired--;
1869 	}
1870 }
1871 
1872 /*
1873  * uvm_pagedeactivate: deactivate page
1874  *
1875  * => caller must lock page queues
1876  * => caller must check to make sure page is not wired
1877  * => object that page belongs to must be locked (so we can adjust pg->flags)
1878  * => caller must clear the reference on the page before calling
1879  */
1880 
1881 void
1882 uvm_pagedeactivate(struct vm_page *pg)
1883 {
1884 
1885 	KASSERT(mutex_owned(&uvm_pageqlock));
1886 	KASSERT(uvm_page_locked_p(pg));
1887 	KASSERT(pg->wire_count != 0 || uvmpdpol_pageisqueued_p(pg));
1888 	uvmpdpol_pagedeactivate(pg);
1889 }
1890 
1891 /*
1892  * uvm_pageactivate: activate page
1893  *
1894  * => caller must lock page queues
1895  */
1896 
1897 void
1898 uvm_pageactivate(struct vm_page *pg)
1899 {
1900 
1901 	KASSERT(mutex_owned(&uvm_pageqlock));
1902 	KASSERT(uvm_page_locked_p(pg));
1903 #if defined(READAHEAD_STATS)
1904 	if ((pg->pqflags & PQ_READAHEAD) != 0) {
1905 		uvm_ra_hit.ev_count++;
1906 		pg->pqflags &= ~PQ_READAHEAD;
1907 	}
1908 #endif /* defined(READAHEAD_STATS) */
1909 	if (pg->wire_count != 0) {
1910 		return;
1911 	}
1912 	uvmpdpol_pageactivate(pg);
1913 }
1914 
1915 /*
1916  * uvm_pagedequeue: remove a page from any paging queue
1917  */
1918 
1919 void
1920 uvm_pagedequeue(struct vm_page *pg)
1921 {
1922 
1923 	if (uvmpdpol_pageisqueued_p(pg)) {
1924 		KASSERT(mutex_owned(&uvm_pageqlock));
1925 	}
1926 
1927 	uvmpdpol_pagedequeue(pg);
1928 }
1929 
1930 /*
1931  * uvm_pageenqueue: add a page to a paging queue without activating.
1932  * used where a page is not really demanded (yet).  eg. read-ahead
1933  */
1934 
1935 void
1936 uvm_pageenqueue(struct vm_page *pg)
1937 {
1938 
1939 	KASSERT(mutex_owned(&uvm_pageqlock));
1940 	if (pg->wire_count != 0) {
1941 		return;
1942 	}
1943 	uvmpdpol_pageenqueue(pg);
1944 }
1945 
1946 /*
1947  * uvm_pagezero: zero fill a page
1948  *
1949  * => if page is part of an object then the object should be locked
1950  *	to protect pg->flags.
1951  */
1952 
1953 void
1954 uvm_pagezero(struct vm_page *pg)
1955 {
1956 	pg->flags &= ~PG_CLEAN;
1957 	pmap_zero_page(VM_PAGE_TO_PHYS(pg));
1958 }
1959 
1960 /*
1961  * uvm_pagecopy: copy a page
1962  *
1963  * => if page is part of an object then the object should be locked
1964  *	to protect pg->flags.
1965  */
1966 
1967 void
1968 uvm_pagecopy(struct vm_page *src, struct vm_page *dst)
1969 {
1970 
1971 	dst->flags &= ~PG_CLEAN;
1972 	pmap_copy_page(VM_PAGE_TO_PHYS(src), VM_PAGE_TO_PHYS(dst));
1973 }
1974 
1975 /*
1976  * uvm_pageismanaged: test it see that a page (specified by PA) is managed.
1977  */
1978 
1979 bool
1980 uvm_pageismanaged(paddr_t pa)
1981 {
1982 
1983 	return (vm_physseg_find(atop(pa), NULL) != -1);
1984 }
1985 
1986 /*
1987  * uvm_page_lookup_freelist: look up the free list for the specified page
1988  */
1989 
1990 int
1991 uvm_page_lookup_freelist(struct vm_page *pg)
1992 {
1993 	int lcv;
1994 
1995 	lcv = vm_physseg_find(atop(VM_PAGE_TO_PHYS(pg)), NULL);
1996 	KASSERT(lcv != -1);
1997 	return (VM_PHYSMEM_PTR(lcv)->free_list);
1998 }
1999 
2000 /*
2001  * uvm_page_locked_p: return true if object associated with page is
2002  * locked.  this is a weak check for runtime assertions only.
2003  */
2004 
2005 bool
2006 uvm_page_locked_p(struct vm_page *pg)
2007 {
2008 
2009 	if (pg->uobject != NULL) {
2010 		return mutex_owned(pg->uobject->vmobjlock);
2011 	}
2012 	if (pg->uanon != NULL) {
2013 		return mutex_owned(pg->uanon->an_lock);
2014 	}
2015 	return true;
2016 }
2017 
2018 #if defined(DDB) || defined(DEBUGPRINT)
2019 
2020 /*
2021  * uvm_page_printit: actually print the page
2022  */
2023 
2024 static const char page_flagbits[] = UVM_PGFLAGBITS;
2025 static const char page_pqflagbits[] = UVM_PQFLAGBITS;
2026 
2027 void
2028 uvm_page_printit(struct vm_page *pg, bool full,
2029     void (*pr)(const char *, ...))
2030 {
2031 	struct vm_page *tpg;
2032 	struct uvm_object *uobj;
2033 	struct pgflist *pgl;
2034 	char pgbuf[128];
2035 	char pqbuf[128];
2036 
2037 	(*pr)("PAGE %p:\n", pg);
2038 	snprintb(pgbuf, sizeof(pgbuf), page_flagbits, pg->flags);
2039 	snprintb(pqbuf, sizeof(pqbuf), page_pqflagbits, pg->pqflags);
2040 	(*pr)("  flags=%s, pqflags=%s, wire_count=%d, pa=0x%lx\n",
2041 	    pgbuf, pqbuf, pg->wire_count, (long)VM_PAGE_TO_PHYS(pg));
2042 	(*pr)("  uobject=%p, uanon=%p, offset=0x%llx loan_count=%d\n",
2043 	    pg->uobject, pg->uanon, (long long)pg->offset, pg->loan_count);
2044 #if defined(UVM_PAGE_TRKOWN)
2045 	if (pg->flags & PG_BUSY)
2046 		(*pr)("  owning process = %d, tag=%s\n",
2047 		    pg->owner, pg->owner_tag);
2048 	else
2049 		(*pr)("  page not busy, no owner\n");
2050 #else
2051 	(*pr)("  [page ownership tracking disabled]\n");
2052 #endif
2053 
2054 	if (!full)
2055 		return;
2056 
2057 	/* cross-verify object/anon */
2058 	if ((pg->pqflags & PQ_FREE) == 0) {
2059 		if (pg->pqflags & PQ_ANON) {
2060 			if (pg->uanon == NULL || pg->uanon->an_page != pg)
2061 			    (*pr)("  >>> ANON DOES NOT POINT HERE <<< (%p)\n",
2062 				(pg->uanon) ? pg->uanon->an_page : NULL);
2063 			else
2064 				(*pr)("  anon backpointer is OK\n");
2065 		} else {
2066 			uobj = pg->uobject;
2067 			if (uobj) {
2068 				(*pr)("  checking object list\n");
2069 				TAILQ_FOREACH(tpg, &uobj->memq, listq.queue) {
2070 					if (tpg == pg) {
2071 						break;
2072 					}
2073 				}
2074 				if (tpg)
2075 					(*pr)("  page found on object list\n");
2076 				else
2077 			(*pr)("  >>> PAGE NOT FOUND ON OBJECT LIST! <<<\n");
2078 			}
2079 		}
2080 	}
2081 
2082 	/* cross-verify page queue */
2083 	if (pg->pqflags & PQ_FREE) {
2084 		int fl = uvm_page_lookup_freelist(pg);
2085 		int color = VM_PGCOLOR_BUCKET(pg);
2086 		pgl = &uvm.page_free[fl].pgfl_buckets[color].pgfl_queues[
2087 		    ((pg)->flags & PG_ZERO) ? PGFL_ZEROS : PGFL_UNKNOWN];
2088 	} else {
2089 		pgl = NULL;
2090 	}
2091 
2092 	if (pgl) {
2093 		(*pr)("  checking pageq list\n");
2094 		LIST_FOREACH(tpg, pgl, pageq.list) {
2095 			if (tpg == pg) {
2096 				break;
2097 			}
2098 		}
2099 		if (tpg)
2100 			(*pr)("  page found on pageq list\n");
2101 		else
2102 			(*pr)("  >>> PAGE NOT FOUND ON PAGEQ LIST! <<<\n");
2103 	}
2104 }
2105 
2106 /*
2107  * uvm_pages_printthem - print a summary of all managed pages
2108  */
2109 
2110 void
2111 uvm_page_printall(void (*pr)(const char *, ...))
2112 {
2113 	unsigned i;
2114 	struct vm_page *pg;
2115 
2116 	(*pr)("%18s %4s %4s %18s %18s"
2117 #ifdef UVM_PAGE_TRKOWN
2118 	    " OWNER"
2119 #endif
2120 	    "\n", "PAGE", "FLAG", "PQ", "UOBJECT", "UANON");
2121 	for (i = 0; i < vm_nphysmem; i++) {
2122 		for (pg = VM_PHYSMEM_PTR(i)->pgs; pg < VM_PHYSMEM_PTR(i)->lastpg; pg++) {
2123 			(*pr)("%18p %04x %04x %18p %18p",
2124 			    pg, pg->flags, pg->pqflags, pg->uobject,
2125 			    pg->uanon);
2126 #ifdef UVM_PAGE_TRKOWN
2127 			if (pg->flags & PG_BUSY)
2128 				(*pr)(" %d [%s]", pg->owner, pg->owner_tag);
2129 #endif
2130 			(*pr)("\n");
2131 		}
2132 	}
2133 }
2134 
2135 #endif /* DDB || DEBUGPRINT */
2136