xref: /netbsd-src/sys/uvm/uvm_page.c (revision 946379e7b37692fc43f68eb0d1c10daa0a7f3b6c)
1 /*	$NetBSD: uvm_page.c,v 1.187 2015/04/11 19:24:13 joerg 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 1.187 2015/04/11 19:24:13 joerg Exp $");
70 
71 #include "opt_ddb.h"
72 #include "opt_uvm.h"
73 #include "opt_uvmhist.h"
74 #include "opt_readahead.h"
75 
76 #include <sys/param.h>
77 #include <sys/systm.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 #ifndef	UVM_RESERVED_PAGES_PER_CPU
112 #define	UVM_RESERVED_PAGES_PER_CPU	5
113 #endif
114 int vm_page_reserve_kernel = UVM_RESERVED_PAGES_PER_CPU;
115 
116 /*
117  * physical memory size;
118  */
119 int physmem;
120 
121 /*
122  * local variables
123  */
124 
125 /*
126  * these variables record the values returned by vm_page_bootstrap,
127  * for debugging purposes.  The implementation of uvm_pageboot_alloc
128  * and pmap_startup here also uses them internally.
129  */
130 
131 static vaddr_t      virtual_space_start;
132 static vaddr_t      virtual_space_end;
133 
134 /*
135  * we allocate an initial number of page colors in uvm_page_init(),
136  * and remember them.  We may re-color pages as cache sizes are
137  * discovered during the autoconfiguration phase.  But we can never
138  * free the initial set of buckets, since they are allocated using
139  * uvm_pageboot_alloc().
140  */
141 
142 static size_t recolored_pages_memsize /* = 0 */;
143 
144 #ifdef DEBUG
145 vaddr_t uvm_zerocheckkva;
146 #endif /* DEBUG */
147 
148 /*
149  * local prototypes
150  */
151 
152 static void uvm_pageinsert(struct uvm_object *, struct vm_page *);
153 static void uvm_pageremove(struct uvm_object *, struct vm_page *);
154 
155 /*
156  * per-object tree of pages
157  */
158 
159 static signed int
160 uvm_page_compare_nodes(void *ctx, const void *n1, const void *n2)
161 {
162 	const struct vm_page *pg1 = n1;
163 	const struct vm_page *pg2 = n2;
164 	const voff_t a = pg1->offset;
165 	const voff_t b = pg2->offset;
166 
167 	if (a < b)
168 		return -1;
169 	if (a > b)
170 		return 1;
171 	return 0;
172 }
173 
174 static signed int
175 uvm_page_compare_key(void *ctx, const void *n, const void *key)
176 {
177 	const struct vm_page *pg = n;
178 	const voff_t a = pg->offset;
179 	const voff_t b = *(const voff_t *)key;
180 
181 	if (a < b)
182 		return -1;
183 	if (a > b)
184 		return 1;
185 	return 0;
186 }
187 
188 const rb_tree_ops_t uvm_page_tree_ops = {
189 	.rbto_compare_nodes = uvm_page_compare_nodes,
190 	.rbto_compare_key = uvm_page_compare_key,
191 	.rbto_node_offset = offsetof(struct vm_page, rb_node),
192 	.rbto_context = NULL
193 };
194 
195 /*
196  * inline functions
197  */
198 
199 /*
200  * uvm_pageinsert: insert a page in the object.
201  *
202  * => caller must lock object
203  * => caller must lock page queues
204  * => call should have already set pg's object and offset pointers
205  *    and bumped the version counter
206  */
207 
208 static inline void
209 uvm_pageinsert_list(struct uvm_object *uobj, struct vm_page *pg,
210     struct vm_page *where)
211 {
212 
213 	KASSERT(uobj == pg->uobject);
214 	KASSERT(mutex_owned(uobj->vmobjlock));
215 	KASSERT((pg->flags & PG_TABLED) == 0);
216 	KASSERT(where == NULL || (where->flags & PG_TABLED));
217 	KASSERT(where == NULL || (where->uobject == uobj));
218 
219 	if (UVM_OBJ_IS_VNODE(uobj)) {
220 		if (uobj->uo_npages == 0) {
221 			struct vnode *vp = (struct vnode *)uobj;
222 
223 			vholdl(vp);
224 		}
225 		if (UVM_OBJ_IS_VTEXT(uobj)) {
226 			atomic_inc_uint(&uvmexp.execpages);
227 		} else {
228 			atomic_inc_uint(&uvmexp.filepages);
229 		}
230 	} else if (UVM_OBJ_IS_AOBJ(uobj)) {
231 		atomic_inc_uint(&uvmexp.anonpages);
232 	}
233 
234 	if (where)
235 		TAILQ_INSERT_AFTER(&uobj->memq, where, pg, listq.queue);
236 	else
237 		TAILQ_INSERT_TAIL(&uobj->memq, pg, listq.queue);
238 	pg->flags |= PG_TABLED;
239 	uobj->uo_npages++;
240 }
241 
242 
243 static inline void
244 uvm_pageinsert_tree(struct uvm_object *uobj, struct vm_page *pg)
245 {
246 	struct vm_page *ret __diagused;
247 
248 	KASSERT(uobj == pg->uobject);
249 	ret = rb_tree_insert_node(&uobj->rb_tree, pg);
250 	KASSERT(ret == pg);
251 }
252 
253 static inline void
254 uvm_pageinsert(struct uvm_object *uobj, struct vm_page *pg)
255 {
256 
257 	KDASSERT(uobj != NULL);
258 	uvm_pageinsert_tree(uobj, pg);
259 	uvm_pageinsert_list(uobj, pg, NULL);
260 }
261 
262 /*
263  * uvm_page_remove: remove page from object.
264  *
265  * => caller must lock object
266  * => caller must lock page queues
267  */
268 
269 static inline void
270 uvm_pageremove_list(struct uvm_object *uobj, struct vm_page *pg)
271 {
272 
273 	KASSERT(uobj == pg->uobject);
274 	KASSERT(mutex_owned(uobj->vmobjlock));
275 	KASSERT(pg->flags & PG_TABLED);
276 
277 	if (UVM_OBJ_IS_VNODE(uobj)) {
278 		if (uobj->uo_npages == 1) {
279 			struct vnode *vp = (struct vnode *)uobj;
280 
281 			holdrelel(vp);
282 		}
283 		if (UVM_OBJ_IS_VTEXT(uobj)) {
284 			atomic_dec_uint(&uvmexp.execpages);
285 		} else {
286 			atomic_dec_uint(&uvmexp.filepages);
287 		}
288 	} else if (UVM_OBJ_IS_AOBJ(uobj)) {
289 		atomic_dec_uint(&uvmexp.anonpages);
290 	}
291 
292 	/* object should be locked */
293 	uobj->uo_npages--;
294 	TAILQ_REMOVE(&uobj->memq, pg, listq.queue);
295 	pg->flags &= ~PG_TABLED;
296 	pg->uobject = NULL;
297 }
298 
299 static inline void
300 uvm_pageremove_tree(struct uvm_object *uobj, struct vm_page *pg)
301 {
302 
303 	KASSERT(uobj == pg->uobject);
304 	rb_tree_remove_node(&uobj->rb_tree, pg);
305 }
306 
307 static inline void
308 uvm_pageremove(struct uvm_object *uobj, struct vm_page *pg)
309 {
310 
311 	KDASSERT(uobj != NULL);
312 	uvm_pageremove_tree(uobj, pg);
313 	uvm_pageremove_list(uobj, pg);
314 }
315 
316 static void
317 uvm_page_init_buckets(struct pgfreelist *pgfl)
318 {
319 	int color, i;
320 
321 	for (color = 0; color < uvmexp.ncolors; color++) {
322 		for (i = 0; i < PGFL_NQUEUES; i++) {
323 			LIST_INIT(&pgfl->pgfl_buckets[color].pgfl_queues[i]);
324 		}
325 	}
326 }
327 
328 /*
329  * uvm_page_init: init the page system.   called from uvm_init().
330  *
331  * => we return the range of kernel virtual memory in kvm_startp/kvm_endp
332  */
333 
334 void
335 uvm_page_init(vaddr_t *kvm_startp, vaddr_t *kvm_endp)
336 {
337 	static struct uvm_cpu boot_cpu;
338 	psize_t freepages, pagecount, bucketcount, n;
339 	struct pgflbucket *bucketarray, *cpuarray;
340 	struct vm_physseg *seg;
341 	struct vm_page *pagearray;
342 	int lcv;
343 	u_int i;
344 	paddr_t paddr;
345 
346 	KASSERT(ncpu <= 1);
347 	CTASSERT(sizeof(pagearray->offset) >= sizeof(struct uvm_cpu *));
348 
349 	/*
350 	 * init the page queues and page queue locks, except the free
351 	 * list; we allocate that later (with the initial vm_page
352 	 * structures).
353 	 */
354 
355 	uvm.cpus[0] = &boot_cpu;
356 	curcpu()->ci_data.cpu_uvm = &boot_cpu;
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 	KASSERT((uvmexp.colormask & uvmexp.ncolors) == 0);
397 
398 	/*
399 	 * we now know we have (PAGE_SIZE * freepages) bytes of memory we can
400 	 * use.   for each page of memory we use we need a vm_page structure.
401 	 * thus, the total number of pages we can use is the total size of
402 	 * the memory divided by the PAGE_SIZE plus the size of the vm_page
403 	 * structure.   we add one to freepages as a fudge factor to avoid
404 	 * truncation errors (since we can only allocate in terms of whole
405 	 * pages).
406 	 */
407 
408 	bucketcount = uvmexp.ncolors * VM_NFREELIST;
409 	pagecount = ((freepages + 1) << PAGE_SHIFT) /
410 	    (PAGE_SIZE + sizeof(struct vm_page));
411 
412 	bucketarray = (void *)uvm_pageboot_alloc((bucketcount *
413 	    sizeof(struct pgflbucket) * 2) + (pagecount *
414 	    sizeof(struct vm_page)));
415 	cpuarray = bucketarray + bucketcount;
416 	pagearray = (struct vm_page *)(bucketarray + bucketcount * 2);
417 
418 	for (lcv = 0; lcv < VM_NFREELIST; lcv++) {
419 		uvm.page_free[lcv].pgfl_buckets =
420 		    (bucketarray + (lcv * uvmexp.ncolors));
421 		uvm_page_init_buckets(&uvm.page_free[lcv]);
422 		uvm.cpus[0]->page_free[lcv].pgfl_buckets =
423 		    (cpuarray + (lcv * uvmexp.ncolors));
424 		uvm_page_init_buckets(&uvm.cpus[0]->page_free[lcv]);
425 	}
426 	memset(pagearray, 0, pagecount * sizeof(struct vm_page));
427 
428 	/*
429 	 * init the vm_page structures and put them in the correct place.
430 	 */
431 
432 	for (lcv = 0 ; lcv < vm_nphysmem ; lcv++) {
433 		seg = VM_PHYSMEM_PTR(lcv);
434 		n = seg->end - seg->start;
435 
436 		/* set up page array pointers */
437 		seg->pgs = pagearray;
438 		pagearray += n;
439 		pagecount -= n;
440 		seg->lastpg = seg->pgs + n;
441 
442 		/* init and free vm_pages (we've already zeroed them) */
443 		paddr = ctob(seg->start);
444 		for (i = 0 ; i < n ; i++, paddr += PAGE_SIZE) {
445 			seg->pgs[i].phys_addr = paddr;
446 #ifdef __HAVE_VM_PAGE_MD
447 			VM_MDPAGE_INIT(&seg->pgs[i]);
448 #endif
449 			if (atop(paddr) >= seg->avail_start &&
450 			    atop(paddr) < seg->avail_end) {
451 				uvmexp.npages++;
452 				/* add page to free pool */
453 				uvm_pagefree(&seg->pgs[i]);
454 			}
455 		}
456 	}
457 
458 	/*
459 	 * pass up the values of virtual_space_start and
460 	 * virtual_space_end (obtained by uvm_pageboot_alloc) to the upper
461 	 * layers of the VM.
462 	 */
463 
464 	*kvm_startp = round_page(virtual_space_start);
465 	*kvm_endp = trunc_page(virtual_space_end);
466 #ifdef DEBUG
467 	/*
468 	 * steal kva for uvm_pagezerocheck().
469 	 */
470 	uvm_zerocheckkva = *kvm_startp;
471 	*kvm_startp += PAGE_SIZE;
472 #endif /* DEBUG */
473 
474 	/*
475 	 * init various thresholds.
476 	 */
477 
478 	uvmexp.reserve_pagedaemon = 1;
479 	uvmexp.reserve_kernel = vm_page_reserve_kernel;
480 
481 	/*
482 	 * determine if we should zero pages in the idle loop.
483 	 */
484 
485 	uvm.cpus[0]->page_idle_zero = vm_page_zero_enable;
486 
487 	/*
488 	 * done!
489 	 */
490 
491 	uvm.page_init_done = true;
492 }
493 
494 /*
495  * uvm_setpagesize: set the page size
496  *
497  * => sets page_shift and page_mask from uvmexp.pagesize.
498  */
499 
500 void
501 uvm_setpagesize(void)
502 {
503 
504 	/*
505 	 * If uvmexp.pagesize is 0 at this point, we expect PAGE_SIZE
506 	 * to be a constant (indicated by being a non-zero value).
507 	 */
508 	if (uvmexp.pagesize == 0) {
509 		if (PAGE_SIZE == 0)
510 			panic("uvm_setpagesize: uvmexp.pagesize not set");
511 		uvmexp.pagesize = PAGE_SIZE;
512 	}
513 	uvmexp.pagemask = uvmexp.pagesize - 1;
514 	if ((uvmexp.pagemask & uvmexp.pagesize) != 0)
515 		panic("uvm_setpagesize: page size %u (%#x) not a power of two",
516 		    uvmexp.pagesize, uvmexp.pagesize);
517 	for (uvmexp.pageshift = 0; ; uvmexp.pageshift++)
518 		if ((1 << uvmexp.pageshift) == uvmexp.pagesize)
519 			break;
520 }
521 
522 /*
523  * uvm_pageboot_alloc: steal memory from physmem for bootstrapping
524  */
525 
526 vaddr_t
527 uvm_pageboot_alloc(vsize_t size)
528 {
529 	static bool initialized = false;
530 	vaddr_t addr;
531 #if !defined(PMAP_STEAL_MEMORY)
532 	vaddr_t vaddr;
533 	paddr_t paddr;
534 #endif
535 
536 	/*
537 	 * on first call to this function, initialize ourselves.
538 	 */
539 	if (initialized == false) {
540 		pmap_virtual_space(&virtual_space_start, &virtual_space_end);
541 
542 		/* round it the way we like it */
543 		virtual_space_start = round_page(virtual_space_start);
544 		virtual_space_end = trunc_page(virtual_space_end);
545 
546 		initialized = true;
547 	}
548 
549 	/* round to page size */
550 	size = round_page(size);
551 
552 #if defined(PMAP_STEAL_MEMORY)
553 
554 	/*
555 	 * defer bootstrap allocation to MD code (it may want to allocate
556 	 * from a direct-mapped segment).  pmap_steal_memory should adjust
557 	 * virtual_space_start/virtual_space_end if necessary.
558 	 */
559 
560 	addr = pmap_steal_memory(size, &virtual_space_start,
561 	    &virtual_space_end);
562 
563 	return(addr);
564 
565 #else /* !PMAP_STEAL_MEMORY */
566 
567 	/*
568 	 * allocate virtual memory for this request
569 	 */
570 	if (virtual_space_start == virtual_space_end ||
571 	    (virtual_space_end - virtual_space_start) < size)
572 		panic("uvm_pageboot_alloc: out of virtual space");
573 
574 	addr = virtual_space_start;
575 
576 #ifdef PMAP_GROWKERNEL
577 	/*
578 	 * If the kernel pmap can't map the requested space,
579 	 * then allocate more resources for it.
580 	 */
581 	if (uvm_maxkaddr < (addr + size)) {
582 		uvm_maxkaddr = pmap_growkernel(addr + size);
583 		if (uvm_maxkaddr < (addr + size))
584 			panic("uvm_pageboot_alloc: pmap_growkernel() failed");
585 	}
586 #endif
587 
588 	virtual_space_start += size;
589 
590 	/*
591 	 * allocate and mapin physical pages to back new virtual pages
592 	 */
593 
594 	for (vaddr = round_page(addr) ; vaddr < addr + size ;
595 	    vaddr += PAGE_SIZE) {
596 
597 		if (!uvm_page_physget(&paddr))
598 			panic("uvm_pageboot_alloc: out of memory");
599 
600 		/*
601 		 * Note this memory is no longer managed, so using
602 		 * pmap_kenter is safe.
603 		 */
604 		pmap_kenter_pa(vaddr, paddr, VM_PROT_READ|VM_PROT_WRITE, 0);
605 	}
606 	pmap_update(pmap_kernel());
607 	return(addr);
608 #endif	/* PMAP_STEAL_MEMORY */
609 }
610 
611 #if !defined(PMAP_STEAL_MEMORY)
612 /*
613  * uvm_page_physget: "steal" one page from the vm_physmem structure.
614  *
615  * => attempt to allocate it off the end of a segment in which the "avail"
616  *    values match the start/end values.   if we can't do that, then we
617  *    will advance both values (making them equal, and removing some
618  *    vm_page structures from the non-avail area).
619  * => return false if out of memory.
620  */
621 
622 /* subroutine: try to allocate from memory chunks on the specified freelist */
623 static bool uvm_page_physget_freelist(paddr_t *, int);
624 
625 static bool
626 uvm_page_physget_freelist(paddr_t *paddrp, int freelist)
627 {
628 	struct vm_physseg *seg;
629 	int lcv, x;
630 
631 	/* pass 1: try allocating from a matching end */
632 #if (VM_PHYSSEG_STRAT == VM_PSTRAT_BIGFIRST)
633 	for (lcv = vm_nphysmem - 1 ; lcv >= 0 ; lcv--)
634 #else
635 	for (lcv = 0 ; lcv < vm_nphysmem ; lcv++)
636 #endif
637 	{
638 		seg = VM_PHYSMEM_PTR(lcv);
639 
640 		if (uvm.page_init_done == true)
641 			panic("uvm_page_physget: called _after_ bootstrap");
642 
643 		if (seg->free_list != freelist)
644 			continue;
645 
646 		/* try from front */
647 		if (seg->avail_start == seg->start &&
648 		    seg->avail_start < seg->avail_end) {
649 			*paddrp = ctob(seg->avail_start);
650 			seg->avail_start++;
651 			seg->start++;
652 			/* nothing left?   nuke it */
653 			if (seg->avail_start == seg->end) {
654 				if (vm_nphysmem == 1)
655 				    panic("uvm_page_physget: out of memory!");
656 				vm_nphysmem--;
657 				for (x = lcv ; x < vm_nphysmem ; x++)
658 					/* structure copy */
659 					VM_PHYSMEM_PTR_SWAP(x, x + 1);
660 			}
661 			return (true);
662 		}
663 
664 		/* try from rear */
665 		if (seg->avail_end == seg->end &&
666 		    seg->avail_start < seg->avail_end) {
667 			*paddrp = ctob(seg->avail_end - 1);
668 			seg->avail_end--;
669 			seg->end--;
670 			/* nothing left?   nuke it */
671 			if (seg->avail_end == seg->start) {
672 				if (vm_nphysmem == 1)
673 				    panic("uvm_page_physget: out of memory!");
674 				vm_nphysmem--;
675 				for (x = lcv ; x < vm_nphysmem ; x++)
676 					/* structure copy */
677 					VM_PHYSMEM_PTR_SWAP(x, x + 1);
678 			}
679 			return (true);
680 		}
681 	}
682 
683 	/* pass2: forget about matching ends, just allocate something */
684 #if (VM_PHYSSEG_STRAT == VM_PSTRAT_BIGFIRST)
685 	for (lcv = vm_nphysmem - 1 ; lcv >= 0 ; lcv--)
686 #else
687 	for (lcv = 0 ; lcv < vm_nphysmem ; lcv++)
688 #endif
689 	{
690 		seg = VM_PHYSMEM_PTR(lcv);
691 
692 		/* any room in this bank? */
693 		if (seg->avail_start >= seg->avail_end)
694 			continue;  /* nope */
695 
696 		*paddrp = ctob(seg->avail_start);
697 		seg->avail_start++;
698 		/* truncate! */
699 		seg->start = seg->avail_start;
700 
701 		/* nothing left?   nuke it */
702 		if (seg->avail_start == seg->end) {
703 			if (vm_nphysmem == 1)
704 				panic("uvm_page_physget: out of memory!");
705 			vm_nphysmem--;
706 			for (x = lcv ; x < vm_nphysmem ; x++)
707 				/* structure copy */
708 				VM_PHYSMEM_PTR_SWAP(x, x + 1);
709 		}
710 		return (true);
711 	}
712 
713 	return (false);        /* whoops! */
714 }
715 
716 bool
717 uvm_page_physget(paddr_t *paddrp)
718 {
719 	int i;
720 
721 	/* try in the order of freelist preference */
722 	for (i = 0; i < VM_NFREELIST; i++)
723 		if (uvm_page_physget_freelist(paddrp, i) == true)
724 			return (true);
725 	return (false);
726 }
727 #endif /* PMAP_STEAL_MEMORY */
728 
729 /*
730  * uvm_page_physload: load physical memory into VM system
731  *
732  * => all args are PFs
733  * => all pages in start/end get vm_page structures
734  * => areas marked by avail_start/avail_end get added to the free page pool
735  * => we are limited to VM_PHYSSEG_MAX physical memory segments
736  */
737 
738 void
739 uvm_page_physload(paddr_t start, paddr_t end, paddr_t avail_start,
740     paddr_t avail_end, int free_list)
741 {
742 	int preload, lcv;
743 	psize_t npages;
744 	struct vm_page *pgs;
745 	struct vm_physseg *ps;
746 
747 	if (uvmexp.pagesize == 0)
748 		panic("uvm_page_physload: page size not set!");
749 	if (free_list >= VM_NFREELIST || free_list < VM_FREELIST_DEFAULT)
750 		panic("uvm_page_physload: bad free list %d", free_list);
751 	if (start >= end)
752 		panic("uvm_page_physload: start >= end");
753 
754 	/*
755 	 * do we have room?
756 	 */
757 
758 	if (vm_nphysmem == VM_PHYSSEG_MAX) {
759 		printf("uvm_page_physload: unable to load physical memory "
760 		    "segment\n");
761 		printf("\t%d segments allocated, ignoring 0x%llx -> 0x%llx\n",
762 		    VM_PHYSSEG_MAX, (long long)start, (long long)end);
763 		printf("\tincrease VM_PHYSSEG_MAX\n");
764 		return;
765 	}
766 
767 	/*
768 	 * check to see if this is a "preload" (i.e. uvm_page_init hasn't been
769 	 * called yet, so kmem is not available).
770 	 */
771 
772 	for (lcv = 0 ; lcv < vm_nphysmem ; lcv++) {
773 		if (VM_PHYSMEM_PTR(lcv)->pgs)
774 			break;
775 	}
776 	preload = (lcv == vm_nphysmem);
777 
778 	/*
779 	 * if VM is already running, attempt to kmem_alloc vm_page structures
780 	 */
781 
782 	if (!preload) {
783 		panic("uvm_page_physload: tried to add RAM after vm_mem_init");
784 	} else {
785 		pgs = NULL;
786 		npages = 0;
787 	}
788 
789 	/*
790 	 * now insert us in the proper place in vm_physmem[]
791 	 */
792 
793 #if (VM_PHYSSEG_STRAT == VM_PSTRAT_RANDOM)
794 	/* random: put it at the end (easy!) */
795 	ps = VM_PHYSMEM_PTR(vm_nphysmem);
796 #elif (VM_PHYSSEG_STRAT == VM_PSTRAT_BSEARCH)
797 	{
798 		int x;
799 		/* sort by address for binary search */
800 		for (lcv = 0 ; lcv < vm_nphysmem ; lcv++)
801 			if (start < VM_PHYSMEM_PTR(lcv)->start)
802 				break;
803 		ps = VM_PHYSMEM_PTR(lcv);
804 		/* move back other entries, if necessary ... */
805 		for (x = vm_nphysmem ; x > lcv ; x--)
806 			/* structure copy */
807 			VM_PHYSMEM_PTR_SWAP(x, x - 1);
808 	}
809 #elif (VM_PHYSSEG_STRAT == VM_PSTRAT_BIGFIRST)
810 	{
811 		int x;
812 		/* sort by largest segment first */
813 		for (lcv = 0 ; lcv < vm_nphysmem ; lcv++)
814 			if ((end - start) >
815 			    (VM_PHYSMEM_PTR(lcv)->end - VM_PHYSMEM_PTR(lcv)->start))
816 				break;
817 		ps = VM_PHYSMEM_PTR(lcv);
818 		/* move back other entries, if necessary ... */
819 		for (x = vm_nphysmem ; x > lcv ; x--)
820 			/* structure copy */
821 			VM_PHYSMEM_PTR_SWAP(x, x - 1);
822 	}
823 #else
824 	panic("uvm_page_physload: unknown physseg strategy selected!");
825 #endif
826 
827 	ps->start = start;
828 	ps->end = end;
829 	ps->avail_start = avail_start;
830 	ps->avail_end = avail_end;
831 	if (preload) {
832 		ps->pgs = NULL;
833 	} else {
834 		ps->pgs = pgs;
835 		ps->lastpg = pgs + npages;
836 	}
837 	ps->free_list = free_list;
838 	vm_nphysmem++;
839 
840 	if (!preload) {
841 		uvmpdpol_reinit();
842 	}
843 }
844 
845 /*
846  * when VM_PHYSSEG_MAX is 1, we can simplify these functions
847  */
848 
849 #if VM_PHYSSEG_MAX == 1
850 static inline int vm_physseg_find_contig(struct vm_physseg *, int, paddr_t, int *);
851 #elif (VM_PHYSSEG_STRAT == VM_PSTRAT_BSEARCH)
852 static inline int vm_physseg_find_bsearch(struct vm_physseg *, int, paddr_t, int *);
853 #else
854 static inline int vm_physseg_find_linear(struct vm_physseg *, int, paddr_t, int *);
855 #endif
856 
857 /*
858  * vm_physseg_find: find vm_physseg structure that belongs to a PA
859  */
860 int
861 vm_physseg_find(paddr_t pframe, int *offp)
862 {
863 
864 #if VM_PHYSSEG_MAX == 1
865 	return vm_physseg_find_contig(vm_physmem, vm_nphysseg, pframe, offp);
866 #elif (VM_PHYSSEG_STRAT == VM_PSTRAT_BSEARCH)
867 	return vm_physseg_find_bsearch(vm_physmem, vm_nphysseg, pframe, offp);
868 #else
869 	return vm_physseg_find_linear(vm_physmem, vm_nphysseg, pframe, offp);
870 #endif
871 }
872 
873 #if VM_PHYSSEG_MAX == 1
874 static inline int
875 vm_physseg_find_contig(struct vm_physseg *segs, int nsegs, paddr_t pframe, int *offp)
876 {
877 
878 	/* 'contig' case */
879 	if (pframe >= segs[0].start && pframe < segs[0].end) {
880 		if (offp)
881 			*offp = pframe - segs[0].start;
882 		return(0);
883 	}
884 	return(-1);
885 }
886 
887 #elif (VM_PHYSSEG_STRAT == VM_PSTRAT_BSEARCH)
888 
889 static inline int
890 vm_physseg_find_bsearch(struct vm_physseg *segs, int nsegs, paddr_t pframe, int *offp)
891 {
892 	/* binary search for it */
893 	u_int	start, len, guess;
894 
895 	/*
896 	 * if try is too large (thus target is less than try) we reduce
897 	 * the length to trunc(len/2) [i.e. everything smaller than "try"]
898 	 *
899 	 * if the try is too small (thus target is greater than try) then
900 	 * we set the new start to be (try + 1).   this means we need to
901 	 * reduce the length to (round(len/2) - 1).
902 	 *
903 	 * note "adjust" below which takes advantage of the fact that
904 	 *  (round(len/2) - 1) == trunc((len - 1) / 2)
905 	 * for any value of len we may have
906 	 */
907 
908 	for (start = 0, len = nsegs ; len != 0 ; len = len / 2) {
909 		guess = start + (len / 2);	/* try in the middle */
910 
911 		/* start past our try? */
912 		if (pframe >= segs[guess].start) {
913 			/* was try correct? */
914 			if (pframe < segs[guess].end) {
915 				if (offp)
916 					*offp = pframe - segs[guess].start;
917 				return guess;            /* got it */
918 			}
919 			start = guess + 1;	/* next time, start here */
920 			len--;			/* "adjust" */
921 		} else {
922 			/*
923 			 * pframe before try, just reduce length of
924 			 * region, done in "for" loop
925 			 */
926 		}
927 	}
928 	return(-1);
929 }
930 
931 #else
932 
933 static inline int
934 vm_physseg_find_linear(struct vm_physseg *segs, int nsegs, paddr_t pframe, int *offp)
935 {
936 	/* linear search for it */
937 	int	lcv;
938 
939 	for (lcv = 0; lcv < nsegs; lcv++) {
940 		if (pframe >= segs[lcv].start &&
941 		    pframe < segs[lcv].end) {
942 			if (offp)
943 				*offp = pframe - segs[lcv].start;
944 			return(lcv);		   /* got it */
945 		}
946 	}
947 	return(-1);
948 }
949 #endif
950 
951 /*
952  * PHYS_TO_VM_PAGE: find vm_page for a PA.   used by MI code to get vm_pages
953  * back from an I/O mapping (ugh!).   used in some MD code as well.
954  */
955 struct vm_page *
956 uvm_phys_to_vm_page(paddr_t pa)
957 {
958 	paddr_t pf = atop(pa);
959 	int	off;
960 	int	psi;
961 
962 	psi = vm_physseg_find(pf, &off);
963 	if (psi != -1)
964 		return(&VM_PHYSMEM_PTR(psi)->pgs[off]);
965 	return(NULL);
966 }
967 
968 paddr_t
969 uvm_vm_page_to_phys(const struct vm_page *pg)
970 {
971 
972 	return pg->phys_addr;
973 }
974 
975 /*
976  * uvm_page_recolor: Recolor the pages if the new bucket count is
977  * larger than the old one.
978  */
979 
980 void
981 uvm_page_recolor(int newncolors)
982 {
983 	struct pgflbucket *bucketarray, *cpuarray, *oldbucketarray;
984 	struct pgfreelist gpgfl, pgfl;
985 	struct vm_page *pg;
986 	vsize_t bucketcount;
987 	size_t bucketmemsize, oldbucketmemsize;
988 	int lcv, color, i, ocolors;
989 	struct uvm_cpu *ucpu;
990 
991 	KASSERT(((newncolors - 1) & newncolors) == 0);
992 
993 	if (newncolors <= uvmexp.ncolors)
994 		return;
995 
996 	if (uvm.page_init_done == false) {
997 		uvmexp.ncolors = newncolors;
998 		return;
999 	}
1000 
1001 	bucketcount = newncolors * VM_NFREELIST;
1002 	bucketmemsize = bucketcount * sizeof(struct pgflbucket) * 2;
1003 	bucketarray = kmem_alloc(bucketmemsize, KM_SLEEP);
1004 	cpuarray = bucketarray + bucketcount;
1005 	if (bucketarray == NULL) {
1006 		printf("WARNING: unable to allocate %ld page color buckets\n",
1007 		    (long) bucketcount);
1008 		return;
1009 	}
1010 
1011 	mutex_spin_enter(&uvm_fpageqlock);
1012 
1013 	/* Make sure we should still do this. */
1014 	if (newncolors <= uvmexp.ncolors) {
1015 		mutex_spin_exit(&uvm_fpageqlock);
1016 		kmem_free(bucketarray, bucketmemsize);
1017 		return;
1018 	}
1019 
1020 	oldbucketarray = uvm.page_free[0].pgfl_buckets;
1021 	ocolors = uvmexp.ncolors;
1022 
1023 	uvmexp.ncolors = newncolors;
1024 	uvmexp.colormask = uvmexp.ncolors - 1;
1025 
1026 	ucpu = curcpu()->ci_data.cpu_uvm;
1027 	for (lcv = 0; lcv < VM_NFREELIST; lcv++) {
1028 		gpgfl.pgfl_buckets = (bucketarray + (lcv * newncolors));
1029 		pgfl.pgfl_buckets = (cpuarray + (lcv * uvmexp.ncolors));
1030 		uvm_page_init_buckets(&gpgfl);
1031 		uvm_page_init_buckets(&pgfl);
1032 		for (color = 0; color < ocolors; color++) {
1033 			for (i = 0; i < PGFL_NQUEUES; i++) {
1034 				while ((pg = LIST_FIRST(&uvm.page_free[
1035 				    lcv].pgfl_buckets[color].pgfl_queues[i]))
1036 				    != NULL) {
1037 					LIST_REMOVE(pg, pageq.list); /* global */
1038 					LIST_REMOVE(pg, listq.list); /* cpu */
1039 					LIST_INSERT_HEAD(&gpgfl.pgfl_buckets[
1040 					    VM_PGCOLOR_BUCKET(pg)].pgfl_queues[
1041 					    i], pg, pageq.list);
1042 					LIST_INSERT_HEAD(&pgfl.pgfl_buckets[
1043 					    VM_PGCOLOR_BUCKET(pg)].pgfl_queues[
1044 					    i], pg, listq.list);
1045 				}
1046 			}
1047 		}
1048 		uvm.page_free[lcv].pgfl_buckets = gpgfl.pgfl_buckets;
1049 		ucpu->page_free[lcv].pgfl_buckets = pgfl.pgfl_buckets;
1050 	}
1051 
1052 	oldbucketmemsize = recolored_pages_memsize;
1053 
1054 	recolored_pages_memsize = bucketmemsize;
1055 	mutex_spin_exit(&uvm_fpageqlock);
1056 
1057 	if (oldbucketmemsize) {
1058 		kmem_free(oldbucketarray, recolored_pages_memsize);
1059 	}
1060 
1061 	/*
1062 	 * this calls uvm_km_alloc() which may want to hold
1063 	 * uvm_fpageqlock.
1064 	 */
1065 	uvm_pager_realloc_emerg();
1066 }
1067 
1068 /*
1069  * uvm_cpu_attach: initialize per-CPU data structures.
1070  */
1071 
1072 void
1073 uvm_cpu_attach(struct cpu_info *ci)
1074 {
1075 	struct pgflbucket *bucketarray;
1076 	struct pgfreelist pgfl;
1077 	struct uvm_cpu *ucpu;
1078 	vsize_t bucketcount;
1079 	int lcv;
1080 
1081 	if (CPU_IS_PRIMARY(ci)) {
1082 		/* Already done in uvm_page_init(). */
1083 		goto attachrnd;
1084 	}
1085 
1086 	/* Add more reserve pages for this CPU. */
1087 	uvmexp.reserve_kernel += vm_page_reserve_kernel;
1088 
1089 	/* Configure this CPU's free lists. */
1090 	bucketcount = uvmexp.ncolors * VM_NFREELIST;
1091 	bucketarray = kmem_alloc(bucketcount * sizeof(struct pgflbucket),
1092 	    KM_SLEEP);
1093 	ucpu = kmem_zalloc(sizeof(*ucpu), KM_SLEEP);
1094 	uvm.cpus[cpu_index(ci)] = ucpu;
1095 	ci->ci_data.cpu_uvm = ucpu;
1096 	for (lcv = 0; lcv < VM_NFREELIST; lcv++) {
1097 		pgfl.pgfl_buckets = (bucketarray + (lcv * uvmexp.ncolors));
1098 		uvm_page_init_buckets(&pgfl);
1099 		ucpu->page_free[lcv].pgfl_buckets = pgfl.pgfl_buckets;
1100 	}
1101 
1102 attachrnd:
1103 	/*
1104 	 * Attach RNG source for this CPU's VM events
1105 	 */
1106         rnd_attach_source(&uvm.cpus[cpu_index(ci)]->rs,
1107 			  ci->ci_data.cpu_name, RND_TYPE_VM,
1108 			  RND_FLAG_COLLECT_TIME|RND_FLAG_COLLECT_VALUE|
1109 			  RND_FLAG_ESTIMATE_VALUE);
1110 
1111 }
1112 
1113 /*
1114  * uvm_pagealloc_pgfl: helper routine for uvm_pagealloc_strat
1115  */
1116 
1117 static struct vm_page *
1118 uvm_pagealloc_pgfl(struct uvm_cpu *ucpu, int flist, int try1, int try2,
1119     int *trycolorp)
1120 {
1121 	struct pgflist *freeq;
1122 	struct vm_page *pg;
1123 	int color, trycolor = *trycolorp;
1124 	struct pgfreelist *gpgfl, *pgfl;
1125 
1126 	KASSERT(mutex_owned(&uvm_fpageqlock));
1127 
1128 	color = trycolor;
1129 	pgfl = &ucpu->page_free[flist];
1130 	gpgfl = &uvm.page_free[flist];
1131 	do {
1132 		/* cpu, try1 */
1133 		if ((pg = LIST_FIRST((freeq =
1134 		    &pgfl->pgfl_buckets[color].pgfl_queues[try1]))) != NULL) {
1135 			KASSERT(pg->pqflags & PQ_FREE);
1136 			KASSERT(try1 == PGFL_ZEROS || !(pg->flags & PG_ZERO));
1137 			KASSERT(try1 == PGFL_UNKNOWN || (pg->flags & PG_ZERO));
1138 			KASSERT(ucpu == VM_FREE_PAGE_TO_CPU(pg));
1139 			VM_FREE_PAGE_TO_CPU(pg)->pages[try1]--;
1140 		    	uvmexp.cpuhit++;
1141 			goto gotit;
1142 		}
1143 		/* global, try1 */
1144 		if ((pg = LIST_FIRST((freeq =
1145 		    &gpgfl->pgfl_buckets[color].pgfl_queues[try1]))) != NULL) {
1146 			KASSERT(pg->pqflags & PQ_FREE);
1147 			KASSERT(try1 == PGFL_ZEROS || !(pg->flags & PG_ZERO));
1148 			KASSERT(try1 == PGFL_UNKNOWN || (pg->flags & PG_ZERO));
1149 			KASSERT(ucpu != VM_FREE_PAGE_TO_CPU(pg));
1150 			VM_FREE_PAGE_TO_CPU(pg)->pages[try1]--;
1151 		    	uvmexp.cpumiss++;
1152 			goto gotit;
1153 		}
1154 		/* cpu, try2 */
1155 		if ((pg = LIST_FIRST((freeq =
1156 		    &pgfl->pgfl_buckets[color].pgfl_queues[try2]))) != NULL) {
1157 			KASSERT(pg->pqflags & PQ_FREE);
1158 			KASSERT(try2 == PGFL_ZEROS || !(pg->flags & PG_ZERO));
1159 			KASSERT(try2 == PGFL_UNKNOWN || (pg->flags & PG_ZERO));
1160 			KASSERT(ucpu == VM_FREE_PAGE_TO_CPU(pg));
1161 			VM_FREE_PAGE_TO_CPU(pg)->pages[try2]--;
1162 		    	uvmexp.cpuhit++;
1163 			goto gotit;
1164 		}
1165 		/* global, try2 */
1166 		if ((pg = LIST_FIRST((freeq =
1167 		    &gpgfl->pgfl_buckets[color].pgfl_queues[try2]))) != NULL) {
1168 			KASSERT(pg->pqflags & PQ_FREE);
1169 			KASSERT(try2 == PGFL_ZEROS || !(pg->flags & PG_ZERO));
1170 			KASSERT(try2 == PGFL_UNKNOWN || (pg->flags & PG_ZERO));
1171 			KASSERT(ucpu != VM_FREE_PAGE_TO_CPU(pg));
1172 			VM_FREE_PAGE_TO_CPU(pg)->pages[try2]--;
1173 		    	uvmexp.cpumiss++;
1174 			goto gotit;
1175 		}
1176 		color = (color + 1) & uvmexp.colormask;
1177 	} while (color != trycolor);
1178 
1179 	return (NULL);
1180 
1181  gotit:
1182 	LIST_REMOVE(pg, pageq.list);	/* global list */
1183 	LIST_REMOVE(pg, listq.list);	/* per-cpu list */
1184 	uvmexp.free--;
1185 
1186 	/* update zero'd page count */
1187 	if (pg->flags & PG_ZERO)
1188 		uvmexp.zeropages--;
1189 
1190 	if (color == trycolor)
1191 		uvmexp.colorhit++;
1192 	else {
1193 		uvmexp.colormiss++;
1194 		*trycolorp = color;
1195 	}
1196 
1197 	return (pg);
1198 }
1199 
1200 /*
1201  * uvm_pagealloc_strat: allocate vm_page from a particular free list.
1202  *
1203  * => return null if no pages free
1204  * => wake up pagedaemon if number of free pages drops below low water mark
1205  * => if obj != NULL, obj must be locked (to put in obj's tree)
1206  * => if anon != NULL, anon must be locked (to put in anon)
1207  * => only one of obj or anon can be non-null
1208  * => caller must activate/deactivate page if it is not wired.
1209  * => free_list is ignored if strat == UVM_PGA_STRAT_NORMAL.
1210  * => policy decision: it is more important to pull a page off of the
1211  *	appropriate priority free list than it is to get a zero'd or
1212  *	unknown contents page.  This is because we live with the
1213  *	consequences of a bad free list decision for the entire
1214  *	lifetime of the page, e.g. if the page comes from memory that
1215  *	is slower to access.
1216  */
1217 
1218 struct vm_page *
1219 uvm_pagealloc_strat(struct uvm_object *obj, voff_t off, struct vm_anon *anon,
1220     int flags, int strat, int free_list)
1221 {
1222 	int lcv, try1, try2, zeroit = 0, color;
1223 	struct uvm_cpu *ucpu;
1224 	struct vm_page *pg;
1225 	lwp_t *l;
1226 
1227 	KASSERT(obj == NULL || anon == NULL);
1228 	KASSERT(anon == NULL || (flags & UVM_FLAG_COLORMATCH) || off == 0);
1229 	KASSERT(off == trunc_page(off));
1230 	KASSERT(obj == NULL || mutex_owned(obj->vmobjlock));
1231 	KASSERT(anon == NULL || anon->an_lock == NULL ||
1232 	    mutex_owned(anon->an_lock));
1233 
1234 	mutex_spin_enter(&uvm_fpageqlock);
1235 
1236 	/*
1237 	 * This implements a global round-robin page coloring
1238 	 * algorithm.
1239 	 */
1240 
1241 	ucpu = curcpu()->ci_data.cpu_uvm;
1242 	if (flags & UVM_FLAG_COLORMATCH) {
1243 		color = atop(off) & uvmexp.colormask;
1244 	} else {
1245 		color = ucpu->page_free_nextcolor;
1246 	}
1247 
1248 	/*
1249 	 * check to see if we need to generate some free pages waking
1250 	 * the pagedaemon.
1251 	 */
1252 
1253 	uvm_kick_pdaemon();
1254 
1255 	/*
1256 	 * fail if any of these conditions is true:
1257 	 * [1]  there really are no free pages, or
1258 	 * [2]  only kernel "reserved" pages remain and
1259 	 *        reserved pages have not been requested.
1260 	 * [3]  only pagedaemon "reserved" pages remain and
1261 	 *        the requestor isn't the pagedaemon.
1262 	 * we make kernel reserve pages available if called by a
1263 	 * kernel thread or a realtime thread.
1264 	 */
1265 	l = curlwp;
1266 	if (__predict_true(l != NULL) && lwp_eprio(l) >= PRI_KTHREAD) {
1267 		flags |= UVM_PGA_USERESERVE;
1268 	}
1269 	if ((uvmexp.free <= uvmexp.reserve_kernel &&
1270 	    (flags & UVM_PGA_USERESERVE) == 0) ||
1271 	    (uvmexp.free <= uvmexp.reserve_pagedaemon &&
1272 	     curlwp != uvm.pagedaemon_lwp))
1273 		goto fail;
1274 
1275 #if PGFL_NQUEUES != 2
1276 #error uvm_pagealloc_strat needs to be updated
1277 #endif
1278 
1279 	/*
1280 	 * If we want a zero'd page, try the ZEROS queue first, otherwise
1281 	 * we try the UNKNOWN queue first.
1282 	 */
1283 	if (flags & UVM_PGA_ZERO) {
1284 		try1 = PGFL_ZEROS;
1285 		try2 = PGFL_UNKNOWN;
1286 	} else {
1287 		try1 = PGFL_UNKNOWN;
1288 		try2 = PGFL_ZEROS;
1289 	}
1290 
1291  again:
1292 	switch (strat) {
1293 	case UVM_PGA_STRAT_NORMAL:
1294 		/* Check freelists: descending priority (ascending id) order */
1295 		for (lcv = 0; lcv < VM_NFREELIST; lcv++) {
1296 			pg = uvm_pagealloc_pgfl(ucpu, lcv,
1297 			    try1, try2, &color);
1298 			if (pg != NULL)
1299 				goto gotit;
1300 		}
1301 
1302 		/* No pages free! */
1303 		goto fail;
1304 
1305 	case UVM_PGA_STRAT_ONLY:
1306 	case UVM_PGA_STRAT_FALLBACK:
1307 		/* Attempt to allocate from the specified free list. */
1308 		KASSERT(free_list >= 0 && free_list < VM_NFREELIST);
1309 		pg = uvm_pagealloc_pgfl(ucpu, free_list,
1310 		    try1, try2, &color);
1311 		if (pg != NULL)
1312 			goto gotit;
1313 
1314 		/* Fall back, if possible. */
1315 		if (strat == UVM_PGA_STRAT_FALLBACK) {
1316 			strat = UVM_PGA_STRAT_NORMAL;
1317 			goto again;
1318 		}
1319 
1320 		/* No pages free! */
1321 		goto fail;
1322 
1323 	default:
1324 		panic("uvm_pagealloc_strat: bad strat %d", strat);
1325 		/* NOTREACHED */
1326 	}
1327 
1328  gotit:
1329 	/*
1330 	 * We now know which color we actually allocated from; set
1331 	 * the next color accordingly.
1332 	 */
1333 
1334 	ucpu->page_free_nextcolor = (color + 1) & uvmexp.colormask;
1335 
1336 	/*
1337 	 * update allocation statistics and remember if we have to
1338 	 * zero the page
1339 	 */
1340 
1341 	if (flags & UVM_PGA_ZERO) {
1342 		if (pg->flags & PG_ZERO) {
1343 			uvmexp.pga_zerohit++;
1344 			zeroit = 0;
1345 		} else {
1346 			uvmexp.pga_zeromiss++;
1347 			zeroit = 1;
1348 		}
1349 		if (ucpu->pages[PGFL_ZEROS] < ucpu->pages[PGFL_UNKNOWN]) {
1350 			ucpu->page_idle_zero = vm_page_zero_enable;
1351 		}
1352 	}
1353 	KASSERT(pg->pqflags == PQ_FREE);
1354 
1355 	pg->offset = off;
1356 	pg->uobject = obj;
1357 	pg->uanon = anon;
1358 	pg->flags = PG_BUSY|PG_CLEAN|PG_FAKE;
1359 	if (anon) {
1360 		anon->an_page = pg;
1361 		pg->pqflags = PQ_ANON;
1362 		atomic_inc_uint(&uvmexp.anonpages);
1363 	} else {
1364 		if (obj) {
1365 			uvm_pageinsert(obj, pg);
1366 		}
1367 		pg->pqflags = 0;
1368 	}
1369 	mutex_spin_exit(&uvm_fpageqlock);
1370 
1371 #if defined(UVM_PAGE_TRKOWN)
1372 	pg->owner_tag = NULL;
1373 #endif
1374 	UVM_PAGE_OWN(pg, "new alloc");
1375 
1376 	if (flags & UVM_PGA_ZERO) {
1377 		/*
1378 		 * A zero'd page is not clean.  If we got a page not already
1379 		 * zero'd, then we have to zero it ourselves.
1380 		 */
1381 		pg->flags &= ~PG_CLEAN;
1382 		if (zeroit)
1383 			pmap_zero_page(VM_PAGE_TO_PHYS(pg));
1384 	}
1385 
1386 	return(pg);
1387 
1388  fail:
1389 	mutex_spin_exit(&uvm_fpageqlock);
1390 	return (NULL);
1391 }
1392 
1393 /*
1394  * uvm_pagereplace: replace a page with another
1395  *
1396  * => object must be locked
1397  */
1398 
1399 void
1400 uvm_pagereplace(struct vm_page *oldpg, struct vm_page *newpg)
1401 {
1402 	struct uvm_object *uobj = oldpg->uobject;
1403 
1404 	KASSERT((oldpg->flags & PG_TABLED) != 0);
1405 	KASSERT(uobj != NULL);
1406 	KASSERT((newpg->flags & PG_TABLED) == 0);
1407 	KASSERT(newpg->uobject == NULL);
1408 	KASSERT(mutex_owned(uobj->vmobjlock));
1409 
1410 	newpg->uobject = uobj;
1411 	newpg->offset = oldpg->offset;
1412 
1413 	uvm_pageremove_tree(uobj, oldpg);
1414 	uvm_pageinsert_tree(uobj, newpg);
1415 	uvm_pageinsert_list(uobj, newpg, oldpg);
1416 	uvm_pageremove_list(uobj, oldpg);
1417 }
1418 
1419 /*
1420  * uvm_pagerealloc: reallocate a page from one object to another
1421  *
1422  * => both objects must be locked
1423  */
1424 
1425 void
1426 uvm_pagerealloc(struct vm_page *pg, struct uvm_object *newobj, voff_t newoff)
1427 {
1428 	/*
1429 	 * remove it from the old object
1430 	 */
1431 
1432 	if (pg->uobject) {
1433 		uvm_pageremove(pg->uobject, pg);
1434 	}
1435 
1436 	/*
1437 	 * put it in the new object
1438 	 */
1439 
1440 	if (newobj) {
1441 		pg->uobject = newobj;
1442 		pg->offset = newoff;
1443 		uvm_pageinsert(newobj, pg);
1444 	}
1445 }
1446 
1447 #ifdef DEBUG
1448 /*
1449  * check if page is zero-filled
1450  *
1451  *  - called with free page queue lock held.
1452  */
1453 void
1454 uvm_pagezerocheck(struct vm_page *pg)
1455 {
1456 	int *p, *ep;
1457 
1458 	KASSERT(uvm_zerocheckkva != 0);
1459 	KASSERT(mutex_owned(&uvm_fpageqlock));
1460 
1461 	/*
1462 	 * XXX assuming pmap_kenter_pa and pmap_kremove never call
1463 	 * uvm page allocator.
1464 	 *
1465 	 * it might be better to have "CPU-local temporary map" pmap interface.
1466 	 */
1467 	pmap_kenter_pa(uvm_zerocheckkva, VM_PAGE_TO_PHYS(pg), VM_PROT_READ, 0);
1468 	p = (int *)uvm_zerocheckkva;
1469 	ep = (int *)((char *)p + PAGE_SIZE);
1470 	pmap_update(pmap_kernel());
1471 	while (p < ep) {
1472 		if (*p != 0)
1473 			panic("PG_ZERO page isn't zero-filled");
1474 		p++;
1475 	}
1476 	pmap_kremove(uvm_zerocheckkva, PAGE_SIZE);
1477 	/*
1478 	 * pmap_update() is not necessary here because no one except us
1479 	 * uses this VA.
1480 	 */
1481 }
1482 #endif /* DEBUG */
1483 
1484 /*
1485  * uvm_pagefree: free page
1486  *
1487  * => erase page's identity (i.e. remove from object)
1488  * => put page on free list
1489  * => caller must lock owning object (either anon or uvm_object)
1490  * => caller must lock page queues
1491  * => assumes all valid mappings of pg are gone
1492  */
1493 
1494 void
1495 uvm_pagefree(struct vm_page *pg)
1496 {
1497 	struct pgflist *pgfl;
1498 	struct uvm_cpu *ucpu;
1499 	int index, color, queue;
1500 	bool iszero;
1501 
1502 #ifdef DEBUG
1503 	if (pg->uobject == (void *)0xdeadbeef &&
1504 	    pg->uanon == (void *)0xdeadbeef) {
1505 		panic("uvm_pagefree: freeing free page %p", pg);
1506 	}
1507 #endif /* DEBUG */
1508 
1509 	KASSERT((pg->flags & PG_PAGEOUT) == 0);
1510 	KASSERT(!(pg->pqflags & PQ_FREE));
1511 	//KASSERT(mutex_owned(&uvm_pageqlock) || !uvmpdpol_pageisqueued_p(pg));
1512 	KASSERT(pg->uobject == NULL || mutex_owned(pg->uobject->vmobjlock));
1513 	KASSERT(pg->uobject != NULL || pg->uanon == NULL ||
1514 		mutex_owned(pg->uanon->an_lock));
1515 
1516 	/*
1517 	 * if the page is loaned, resolve the loan instead of freeing.
1518 	 */
1519 
1520 	if (pg->loan_count) {
1521 		KASSERT(pg->wire_count == 0);
1522 
1523 		/*
1524 		 * if the page is owned by an anon then we just want to
1525 		 * drop anon ownership.  the kernel will free the page when
1526 		 * it is done with it.  if the page is owned by an object,
1527 		 * remove it from the object and mark it dirty for the benefit
1528 		 * of possible anon owners.
1529 		 *
1530 		 * regardless of previous ownership, wakeup any waiters,
1531 		 * unbusy the page, and we're done.
1532 		 */
1533 
1534 		if (pg->uobject != NULL) {
1535 			uvm_pageremove(pg->uobject, pg);
1536 			pg->flags &= ~PG_CLEAN;
1537 		} else if (pg->uanon != NULL) {
1538 			if ((pg->pqflags & PQ_ANON) == 0) {
1539 				pg->loan_count--;
1540 			} else {
1541 				pg->pqflags &= ~PQ_ANON;
1542 				atomic_dec_uint(&uvmexp.anonpages);
1543 			}
1544 			pg->uanon->an_page = NULL;
1545 			pg->uanon = NULL;
1546 		}
1547 		if (pg->flags & PG_WANTED) {
1548 			wakeup(pg);
1549 		}
1550 		pg->flags &= ~(PG_WANTED|PG_BUSY|PG_RELEASED|PG_PAGER1);
1551 #ifdef UVM_PAGE_TRKOWN
1552 		pg->owner_tag = NULL;
1553 #endif
1554 		if (pg->loan_count) {
1555 			KASSERT(pg->uobject == NULL);
1556 			if (pg->uanon == NULL) {
1557 				KASSERT(mutex_owned(&uvm_pageqlock));
1558 				uvm_pagedequeue(pg);
1559 			}
1560 			return;
1561 		}
1562 	}
1563 
1564 	/*
1565 	 * remove page from its object or anon.
1566 	 */
1567 
1568 	if (pg->uobject != NULL) {
1569 		uvm_pageremove(pg->uobject, pg);
1570 	} else if (pg->uanon != NULL) {
1571 		pg->uanon->an_page = NULL;
1572 		atomic_dec_uint(&uvmexp.anonpages);
1573 	}
1574 
1575 	/*
1576 	 * now remove the page from the queues.
1577 	 */
1578 	if (uvmpdpol_pageisqueued_p(pg)) {
1579 		KASSERT(mutex_owned(&uvm_pageqlock));
1580 		uvm_pagedequeue(pg);
1581 	}
1582 
1583 	/*
1584 	 * if the page was wired, unwire it now.
1585 	 */
1586 
1587 	if (pg->wire_count) {
1588 		pg->wire_count = 0;
1589 		uvmexp.wired--;
1590 	}
1591 
1592 	/*
1593 	 * and put on free queue
1594 	 */
1595 
1596 	iszero = (pg->flags & PG_ZERO);
1597 	index = uvm_page_lookup_freelist(pg);
1598 	color = VM_PGCOLOR_BUCKET(pg);
1599 	queue = (iszero ? PGFL_ZEROS : PGFL_UNKNOWN);
1600 
1601 #ifdef DEBUG
1602 	pg->uobject = (void *)0xdeadbeef;
1603 	pg->uanon = (void *)0xdeadbeef;
1604 #endif
1605 
1606 	mutex_spin_enter(&uvm_fpageqlock);
1607 	pg->pqflags = PQ_FREE;
1608 
1609 #ifdef DEBUG
1610 	if (iszero)
1611 		uvm_pagezerocheck(pg);
1612 #endif /* DEBUG */
1613 
1614 
1615 	/* global list */
1616 	pgfl = &uvm.page_free[index].pgfl_buckets[color].pgfl_queues[queue];
1617 	LIST_INSERT_HEAD(pgfl, pg, pageq.list);
1618 	uvmexp.free++;
1619 	if (iszero) {
1620 		uvmexp.zeropages++;
1621 	}
1622 
1623 	/* per-cpu list */
1624 	ucpu = curcpu()->ci_data.cpu_uvm;
1625 	pg->offset = (uintptr_t)ucpu;
1626 	pgfl = &ucpu->page_free[index].pgfl_buckets[color].pgfl_queues[queue];
1627 	LIST_INSERT_HEAD(pgfl, pg, listq.list);
1628 	ucpu->pages[queue]++;
1629 	if (ucpu->pages[PGFL_ZEROS] < ucpu->pages[PGFL_UNKNOWN]) {
1630 		ucpu->page_idle_zero = vm_page_zero_enable;
1631 	}
1632 
1633 	mutex_spin_exit(&uvm_fpageqlock);
1634 }
1635 
1636 /*
1637  * uvm_page_unbusy: unbusy an array of pages.
1638  *
1639  * => pages must either all belong to the same object, or all belong to anons.
1640  * => if pages are object-owned, object must be locked.
1641  * => if pages are anon-owned, anons must be locked.
1642  * => caller must lock page queues if pages may be released.
1643  * => caller must make sure that anon-owned pages are not PG_RELEASED.
1644  */
1645 
1646 void
1647 uvm_page_unbusy(struct vm_page **pgs, int npgs)
1648 {
1649 	struct vm_page *pg;
1650 	int i;
1651 	UVMHIST_FUNC("uvm_page_unbusy"); UVMHIST_CALLED(ubchist);
1652 
1653 	for (i = 0; i < npgs; i++) {
1654 		pg = pgs[i];
1655 		if (pg == NULL || pg == PGO_DONTCARE) {
1656 			continue;
1657 		}
1658 
1659 		KASSERT(uvm_page_locked_p(pg));
1660 		KASSERT(pg->flags & PG_BUSY);
1661 		KASSERT((pg->flags & PG_PAGEOUT) == 0);
1662 		if (pg->flags & PG_WANTED) {
1663 			wakeup(pg);
1664 		}
1665 		if (pg->flags & PG_RELEASED) {
1666 			UVMHIST_LOG(ubchist, "releasing pg %p", pg,0,0,0);
1667 			KASSERT(pg->uobject != NULL ||
1668 			    (pg->uanon != NULL && pg->uanon->an_ref > 0));
1669 			pg->flags &= ~PG_RELEASED;
1670 			uvm_pagefree(pg);
1671 		} else {
1672 			UVMHIST_LOG(ubchist, "unbusying pg %p", pg,0,0,0);
1673 			KASSERT((pg->flags & PG_FAKE) == 0);
1674 			pg->flags &= ~(PG_WANTED|PG_BUSY);
1675 			UVM_PAGE_OWN(pg, NULL);
1676 		}
1677 	}
1678 }
1679 
1680 #if defined(UVM_PAGE_TRKOWN)
1681 /*
1682  * uvm_page_own: set or release page ownership
1683  *
1684  * => this is a debugging function that keeps track of who sets PG_BUSY
1685  *	and where they do it.   it can be used to track down problems
1686  *	such a process setting "PG_BUSY" and never releasing it.
1687  * => page's object [if any] must be locked
1688  * => if "tag" is NULL then we are releasing page ownership
1689  */
1690 void
1691 uvm_page_own(struct vm_page *pg, const char *tag)
1692 {
1693 
1694 	KASSERT((pg->flags & (PG_PAGEOUT|PG_RELEASED)) == 0);
1695 	KASSERT((pg->flags & PG_WANTED) == 0);
1696 	KASSERT(uvm_page_locked_p(pg));
1697 
1698 	/* gain ownership? */
1699 	if (tag) {
1700 		KASSERT((pg->flags & PG_BUSY) != 0);
1701 		if (pg->owner_tag) {
1702 			printf("uvm_page_own: page %p already owned "
1703 			    "by proc %d [%s]\n", pg,
1704 			    pg->owner, pg->owner_tag);
1705 			panic("uvm_page_own");
1706 		}
1707 		pg->owner = curproc->p_pid;
1708 		pg->lowner = curlwp->l_lid;
1709 		pg->owner_tag = tag;
1710 		return;
1711 	}
1712 
1713 	/* drop ownership */
1714 	KASSERT((pg->flags & PG_BUSY) == 0);
1715 	if (pg->owner_tag == NULL) {
1716 		printf("uvm_page_own: dropping ownership of an non-owned "
1717 		    "page (%p)\n", pg);
1718 		panic("uvm_page_own");
1719 	}
1720 	if (!uvmpdpol_pageisqueued_p(pg)) {
1721 		KASSERT((pg->uanon == NULL && pg->uobject == NULL) ||
1722 		    pg->wire_count > 0);
1723 	} else {
1724 		KASSERT(pg->wire_count == 0);
1725 	}
1726 	pg->owner_tag = NULL;
1727 }
1728 #endif
1729 
1730 /*
1731  * uvm_pageidlezero: zero free pages while the system is idle.
1732  *
1733  * => try to complete one color bucket at a time, to reduce our impact
1734  *	on the CPU cache.
1735  * => we loop until we either reach the target or there is a lwp ready
1736  *      to run, or MD code detects a reason to break early.
1737  */
1738 void
1739 uvm_pageidlezero(void)
1740 {
1741 	struct vm_page *pg;
1742 	struct pgfreelist *pgfl, *gpgfl;
1743 	struct uvm_cpu *ucpu;
1744 	int free_list, firstbucket, nextbucket;
1745 	bool lcont = false;
1746 
1747 	ucpu = curcpu()->ci_data.cpu_uvm;
1748 	if (!ucpu->page_idle_zero ||
1749 	    ucpu->pages[PGFL_UNKNOWN] < uvmexp.ncolors) {
1750 	    	ucpu->page_idle_zero = false;
1751 		return;
1752 	}
1753 	if (!mutex_tryenter(&uvm_fpageqlock)) {
1754 		/* Contention: let other CPUs to use the lock. */
1755 		return;
1756 	}
1757 	firstbucket = ucpu->page_free_nextcolor;
1758 	nextbucket = firstbucket;
1759 	do {
1760 		for (free_list = 0; free_list < VM_NFREELIST; free_list++) {
1761 			if (sched_curcpu_runnable_p()) {
1762 				goto quit;
1763 			}
1764 			pgfl = &ucpu->page_free[free_list];
1765 			gpgfl = &uvm.page_free[free_list];
1766 			while ((pg = LIST_FIRST(&pgfl->pgfl_buckets[
1767 			    nextbucket].pgfl_queues[PGFL_UNKNOWN])) != NULL) {
1768 				if (lcont || sched_curcpu_runnable_p()) {
1769 					goto quit;
1770 				}
1771 				LIST_REMOVE(pg, pageq.list); /* global list */
1772 				LIST_REMOVE(pg, listq.list); /* per-cpu list */
1773 				ucpu->pages[PGFL_UNKNOWN]--;
1774 				uvmexp.free--;
1775 				KASSERT(pg->pqflags == PQ_FREE);
1776 				pg->pqflags = 0;
1777 				mutex_spin_exit(&uvm_fpageqlock);
1778 #ifdef PMAP_PAGEIDLEZERO
1779 				if (!PMAP_PAGEIDLEZERO(VM_PAGE_TO_PHYS(pg))) {
1780 
1781 					/*
1782 					 * The machine-dependent code detected
1783 					 * some reason for us to abort zeroing
1784 					 * pages, probably because there is a
1785 					 * process now ready to run.
1786 					 */
1787 
1788 					mutex_spin_enter(&uvm_fpageqlock);
1789 					pg->pqflags = PQ_FREE;
1790 					LIST_INSERT_HEAD(&gpgfl->pgfl_buckets[
1791 					    nextbucket].pgfl_queues[
1792 					    PGFL_UNKNOWN], pg, pageq.list);
1793 					LIST_INSERT_HEAD(&pgfl->pgfl_buckets[
1794 					    nextbucket].pgfl_queues[
1795 					    PGFL_UNKNOWN], pg, listq.list);
1796 					ucpu->pages[PGFL_UNKNOWN]++;
1797 					uvmexp.free++;
1798 					uvmexp.zeroaborts++;
1799 					goto quit;
1800 				}
1801 #else
1802 				pmap_zero_page(VM_PAGE_TO_PHYS(pg));
1803 #endif /* PMAP_PAGEIDLEZERO */
1804 				pg->flags |= PG_ZERO;
1805 
1806 				if (!mutex_tryenter(&uvm_fpageqlock)) {
1807 					lcont = true;
1808 					mutex_spin_enter(&uvm_fpageqlock);
1809 				} else {
1810 					lcont = false;
1811 				}
1812 				pg->pqflags = PQ_FREE;
1813 				LIST_INSERT_HEAD(&gpgfl->pgfl_buckets[
1814 				    nextbucket].pgfl_queues[PGFL_ZEROS],
1815 				    pg, pageq.list);
1816 				LIST_INSERT_HEAD(&pgfl->pgfl_buckets[
1817 				    nextbucket].pgfl_queues[PGFL_ZEROS],
1818 				    pg, listq.list);
1819 				ucpu->pages[PGFL_ZEROS]++;
1820 				uvmexp.free++;
1821 				uvmexp.zeropages++;
1822 			}
1823 		}
1824 		if (ucpu->pages[PGFL_UNKNOWN] < uvmexp.ncolors) {
1825 			break;
1826 		}
1827 		nextbucket = (nextbucket + 1) & uvmexp.colormask;
1828 	} while (nextbucket != firstbucket);
1829 	ucpu->page_idle_zero = false;
1830  quit:
1831 	mutex_spin_exit(&uvm_fpageqlock);
1832 }
1833 
1834 /*
1835  * uvm_pagelookup: look up a page
1836  *
1837  * => caller should lock object to keep someone from pulling the page
1838  *	out from under it
1839  */
1840 
1841 struct vm_page *
1842 uvm_pagelookup(struct uvm_object *obj, voff_t off)
1843 {
1844 	struct vm_page *pg;
1845 
1846 	KASSERT(mutex_owned(obj->vmobjlock));
1847 
1848 	pg = rb_tree_find_node(&obj->rb_tree, &off);
1849 
1850 	KASSERT(pg == NULL || obj->uo_npages != 0);
1851 	KASSERT(pg == NULL || (pg->flags & (PG_RELEASED|PG_PAGEOUT)) == 0 ||
1852 		(pg->flags & PG_BUSY) != 0);
1853 	return pg;
1854 }
1855 
1856 /*
1857  * uvm_pagewire: wire the page, thus removing it from the daemon's grasp
1858  *
1859  * => caller must lock page queues
1860  */
1861 
1862 void
1863 uvm_pagewire(struct vm_page *pg)
1864 {
1865 	KASSERT(mutex_owned(&uvm_pageqlock));
1866 #if defined(READAHEAD_STATS)
1867 	if ((pg->pqflags & PQ_READAHEAD) != 0) {
1868 		uvm_ra_hit.ev_count++;
1869 		pg->pqflags &= ~PQ_READAHEAD;
1870 	}
1871 #endif /* defined(READAHEAD_STATS) */
1872 	if (pg->wire_count == 0) {
1873 		uvm_pagedequeue(pg);
1874 		uvmexp.wired++;
1875 	}
1876 	pg->wire_count++;
1877 }
1878 
1879 /*
1880  * uvm_pageunwire: unwire the page.
1881  *
1882  * => activate if wire count goes to zero.
1883  * => caller must lock page queues
1884  */
1885 
1886 void
1887 uvm_pageunwire(struct vm_page *pg)
1888 {
1889 	KASSERT(mutex_owned(&uvm_pageqlock));
1890 	pg->wire_count--;
1891 	if (pg->wire_count == 0) {
1892 		uvm_pageactivate(pg);
1893 		uvmexp.wired--;
1894 	}
1895 }
1896 
1897 /*
1898  * uvm_pagedeactivate: deactivate page
1899  *
1900  * => caller must lock page queues
1901  * => caller must check to make sure page is not wired
1902  * => object that page belongs to must be locked (so we can adjust pg->flags)
1903  * => caller must clear the reference on the page before calling
1904  */
1905 
1906 void
1907 uvm_pagedeactivate(struct vm_page *pg)
1908 {
1909 
1910 	KASSERT(mutex_owned(&uvm_pageqlock));
1911 	KASSERT(uvm_page_locked_p(pg));
1912 	KASSERT(pg->wire_count != 0 || uvmpdpol_pageisqueued_p(pg));
1913 	uvmpdpol_pagedeactivate(pg);
1914 }
1915 
1916 /*
1917  * uvm_pageactivate: activate page
1918  *
1919  * => caller must lock page queues
1920  */
1921 
1922 void
1923 uvm_pageactivate(struct vm_page *pg)
1924 {
1925 
1926 	KASSERT(mutex_owned(&uvm_pageqlock));
1927 	KASSERT(uvm_page_locked_p(pg));
1928 #if defined(READAHEAD_STATS)
1929 	if ((pg->pqflags & PQ_READAHEAD) != 0) {
1930 		uvm_ra_hit.ev_count++;
1931 		pg->pqflags &= ~PQ_READAHEAD;
1932 	}
1933 #endif /* defined(READAHEAD_STATS) */
1934 	if (pg->wire_count != 0) {
1935 		return;
1936 	}
1937 	uvmpdpol_pageactivate(pg);
1938 }
1939 
1940 /*
1941  * uvm_pagedequeue: remove a page from any paging queue
1942  */
1943 
1944 void
1945 uvm_pagedequeue(struct vm_page *pg)
1946 {
1947 
1948 	if (uvmpdpol_pageisqueued_p(pg)) {
1949 		KASSERT(mutex_owned(&uvm_pageqlock));
1950 	}
1951 
1952 	uvmpdpol_pagedequeue(pg);
1953 }
1954 
1955 /*
1956  * uvm_pageenqueue: add a page to a paging queue without activating.
1957  * used where a page is not really demanded (yet).  eg. read-ahead
1958  */
1959 
1960 void
1961 uvm_pageenqueue(struct vm_page *pg)
1962 {
1963 
1964 	KASSERT(mutex_owned(&uvm_pageqlock));
1965 	if (pg->wire_count != 0) {
1966 		return;
1967 	}
1968 	uvmpdpol_pageenqueue(pg);
1969 }
1970 
1971 /*
1972  * uvm_pagezero: zero fill a page
1973  *
1974  * => if page is part of an object then the object should be locked
1975  *	to protect pg->flags.
1976  */
1977 
1978 void
1979 uvm_pagezero(struct vm_page *pg)
1980 {
1981 	pg->flags &= ~PG_CLEAN;
1982 	pmap_zero_page(VM_PAGE_TO_PHYS(pg));
1983 }
1984 
1985 /*
1986  * uvm_pagecopy: copy a page
1987  *
1988  * => if page is part of an object then the object should be locked
1989  *	to protect pg->flags.
1990  */
1991 
1992 void
1993 uvm_pagecopy(struct vm_page *src, struct vm_page *dst)
1994 {
1995 
1996 	dst->flags &= ~PG_CLEAN;
1997 	pmap_copy_page(VM_PAGE_TO_PHYS(src), VM_PAGE_TO_PHYS(dst));
1998 }
1999 
2000 /*
2001  * uvm_pageismanaged: test it see that a page (specified by PA) is managed.
2002  */
2003 
2004 bool
2005 uvm_pageismanaged(paddr_t pa)
2006 {
2007 
2008 	return (vm_physseg_find(atop(pa), NULL) != -1);
2009 }
2010 
2011 /*
2012  * uvm_page_lookup_freelist: look up the free list for the specified page
2013  */
2014 
2015 int
2016 uvm_page_lookup_freelist(struct vm_page *pg)
2017 {
2018 	int lcv;
2019 
2020 	lcv = vm_physseg_find(atop(VM_PAGE_TO_PHYS(pg)), NULL);
2021 	KASSERT(lcv != -1);
2022 	return (VM_PHYSMEM_PTR(lcv)->free_list);
2023 }
2024 
2025 /*
2026  * uvm_page_locked_p: return true if object associated with page is
2027  * locked.  this is a weak check for runtime assertions only.
2028  */
2029 
2030 bool
2031 uvm_page_locked_p(struct vm_page *pg)
2032 {
2033 
2034 	if (pg->uobject != NULL) {
2035 		return mutex_owned(pg->uobject->vmobjlock);
2036 	}
2037 	if (pg->uanon != NULL) {
2038 		return mutex_owned(pg->uanon->an_lock);
2039 	}
2040 	return true;
2041 }
2042 
2043 #if defined(DDB) || defined(DEBUGPRINT)
2044 
2045 /*
2046  * uvm_page_printit: actually print the page
2047  */
2048 
2049 static const char page_flagbits[] = UVM_PGFLAGBITS;
2050 static const char page_pqflagbits[] = UVM_PQFLAGBITS;
2051 
2052 void
2053 uvm_page_printit(struct vm_page *pg, bool full,
2054     void (*pr)(const char *, ...))
2055 {
2056 	struct vm_page *tpg;
2057 	struct uvm_object *uobj;
2058 	struct pgflist *pgl;
2059 	char pgbuf[128];
2060 	char pqbuf[128];
2061 
2062 	(*pr)("PAGE %p:\n", pg);
2063 	snprintb(pgbuf, sizeof(pgbuf), page_flagbits, pg->flags);
2064 	snprintb(pqbuf, sizeof(pqbuf), page_pqflagbits, pg->pqflags);
2065 	(*pr)("  flags=%s, pqflags=%s, wire_count=%d, pa=0x%lx\n",
2066 	    pgbuf, pqbuf, pg->wire_count, (long)VM_PAGE_TO_PHYS(pg));
2067 	(*pr)("  uobject=%p, uanon=%p, offset=0x%llx loan_count=%d\n",
2068 	    pg->uobject, pg->uanon, (long long)pg->offset, pg->loan_count);
2069 #if defined(UVM_PAGE_TRKOWN)
2070 	if (pg->flags & PG_BUSY)
2071 		(*pr)("  owning process = %d, tag=%s\n",
2072 		    pg->owner, pg->owner_tag);
2073 	else
2074 		(*pr)("  page not busy, no owner\n");
2075 #else
2076 	(*pr)("  [page ownership tracking disabled]\n");
2077 #endif
2078 
2079 	if (!full)
2080 		return;
2081 
2082 	/* cross-verify object/anon */
2083 	if ((pg->pqflags & PQ_FREE) == 0) {
2084 		if (pg->pqflags & PQ_ANON) {
2085 			if (pg->uanon == NULL || pg->uanon->an_page != pg)
2086 			    (*pr)("  >>> ANON DOES NOT POINT HERE <<< (%p)\n",
2087 				(pg->uanon) ? pg->uanon->an_page : NULL);
2088 			else
2089 				(*pr)("  anon backpointer is OK\n");
2090 		} else {
2091 			uobj = pg->uobject;
2092 			if (uobj) {
2093 				(*pr)("  checking object list\n");
2094 				TAILQ_FOREACH(tpg, &uobj->memq, listq.queue) {
2095 					if (tpg == pg) {
2096 						break;
2097 					}
2098 				}
2099 				if (tpg)
2100 					(*pr)("  page found on object list\n");
2101 				else
2102 			(*pr)("  >>> PAGE NOT FOUND ON OBJECT LIST! <<<\n");
2103 			}
2104 		}
2105 	}
2106 
2107 	/* cross-verify page queue */
2108 	if (pg->pqflags & PQ_FREE) {
2109 		int fl = uvm_page_lookup_freelist(pg);
2110 		int color = VM_PGCOLOR_BUCKET(pg);
2111 		pgl = &uvm.page_free[fl].pgfl_buckets[color].pgfl_queues[
2112 		    ((pg)->flags & PG_ZERO) ? PGFL_ZEROS : PGFL_UNKNOWN];
2113 	} else {
2114 		pgl = NULL;
2115 	}
2116 
2117 	if (pgl) {
2118 		(*pr)("  checking pageq list\n");
2119 		LIST_FOREACH(tpg, pgl, pageq.list) {
2120 			if (tpg == pg) {
2121 				break;
2122 			}
2123 		}
2124 		if (tpg)
2125 			(*pr)("  page found on pageq list\n");
2126 		else
2127 			(*pr)("  >>> PAGE NOT FOUND ON PAGEQ LIST! <<<\n");
2128 	}
2129 }
2130 
2131 /*
2132  * uvm_pages_printthem - print a summary of all managed pages
2133  */
2134 
2135 void
2136 uvm_page_printall(void (*pr)(const char *, ...))
2137 {
2138 	unsigned i;
2139 	struct vm_page *pg;
2140 
2141 	(*pr)("%18s %4s %4s %18s %18s"
2142 #ifdef UVM_PAGE_TRKOWN
2143 	    " OWNER"
2144 #endif
2145 	    "\n", "PAGE", "FLAG", "PQ", "UOBJECT", "UANON");
2146 	for (i = 0; i < vm_nphysmem; i++) {
2147 		for (pg = VM_PHYSMEM_PTR(i)->pgs; pg < VM_PHYSMEM_PTR(i)->lastpg; pg++) {
2148 			(*pr)("%18p %04x %04x %18p %18p",
2149 			    pg, pg->flags, pg->pqflags, pg->uobject,
2150 			    pg->uanon);
2151 #ifdef UVM_PAGE_TRKOWN
2152 			if (pg->flags & PG_BUSY)
2153 				(*pr)(" %d [%s]", pg->owner, pg->owner_tag);
2154 #endif
2155 			(*pr)("\n");
2156 		}
2157 	}
2158 }
2159 
2160 #endif /* DDB || DEBUGPRINT */
2161