xref: /openbsd/sys/uvm/uvm_page.c (revision 82673a18)
1 /*	$OpenBSD: uvm_page.c,v 1.177 2024/05/01 12:54:27 mpi Exp $	*/
2 /*	$NetBSD: uvm_page.c,v 1.44 2000/11/27 08:40:04 chs Exp $	*/
3 
4 /*
5  * Copyright (c) 1997 Charles D. Cranor and Washington University.
6  * Copyright (c) 1991, 1993, The Regents of the University of California.
7  *
8  * All rights reserved.
9  *
10  * This code is derived from software contributed to Berkeley by
11  * The Mach Operating System project at Carnegie-Mellon University.
12  *
13  * Redistribution and use in source and binary forms, with or without
14  * modification, are permitted provided that the following conditions
15  * are met:
16  * 1. Redistributions of source code must retain the above copyright
17  *    notice, this list of conditions and the following disclaimer.
18  * 2. Redistributions in binary form must reproduce the above copyright
19  *    notice, this list of conditions and the following disclaimer in the
20  *    documentation and/or other materials provided with the distribution.
21  * 3. Neither the name of the University nor the names of its contributors
22  *    may be used to endorse or promote products derived from this software
23  *    without specific prior written permission.
24  *
25  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
26  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
27  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
28  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
29  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
30  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
31  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
32  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
33  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
34  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
35  * SUCH DAMAGE.
36  *
37  *	@(#)vm_page.c   8.3 (Berkeley) 3/21/94
38  * from: Id: uvm_page.c,v 1.1.2.18 1998/02/06 05:24:42 chs Exp
39  *
40  *
41  * Copyright (c) 1987, 1990 Carnegie-Mellon University.
42  * All rights reserved.
43  *
44  * Permission to use, copy, modify and distribute this software and
45  * its documentation is hereby granted, provided that both the copyright
46  * notice and this permission notice appear in all copies of the
47  * software, derivative works or modified versions, and any portions
48  * thereof, and that both notices appear in supporting documentation.
49  *
50  * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
51  * CONDITION.  CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
52  * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
53  *
54  * Carnegie Mellon requests users of this software to return to
55  *
56  *  Software Distribution Coordinator  or  Software.Distribution@CS.CMU.EDU
57  *  School of Computer Science
58  *  Carnegie Mellon University
59  *  Pittsburgh PA 15213-3890
60  *
61  * any improvements or extensions that they make and grant Carnegie the
62  * rights to redistribute these changes.
63  */
64 
65 /*
66  * uvm_page.c: page ops.
67  */
68 
69 #include <sys/param.h>
70 #include <sys/systm.h>
71 #include <sys/sched.h>
72 #include <sys/vnode.h>
73 #include <sys/mount.h>
74 #include <sys/proc.h>
75 #include <sys/smr.h>
76 
77 #include <uvm/uvm.h>
78 
79 /*
80  * for object trees
81  */
82 RBT_GENERATE(uvm_objtree, vm_page, objt, uvm_pagecmp);
83 
84 int
uvm_pagecmp(const struct vm_page * a,const struct vm_page * b)85 uvm_pagecmp(const struct vm_page *a, const struct vm_page *b)
86 {
87 	return a->offset < b->offset ? -1 : a->offset > b->offset;
88 }
89 
90 /*
91  * global vars... XXXCDC: move to uvm. structure.
92  */
93 /*
94  * physical memory config is stored in vm_physmem.
95  */
96 struct vm_physseg vm_physmem[VM_PHYSSEG_MAX];	/* XXXCDC: uvm.physmem */
97 int vm_nphysseg = 0;				/* XXXCDC: uvm.nphysseg */
98 
99 /*
100  * Some supported CPUs in a given architecture don't support all
101  * of the things necessary to do idle page zero'ing efficiently.
102  * We therefore provide a way to disable it from machdep code here.
103  */
104 
105 /*
106  * local variables
107  */
108 /*
109  * these variables record the values returned by vm_page_bootstrap,
110  * for debugging purposes.  The implementation of uvm_pageboot_alloc
111  * and pmap_startup here also uses them internally.
112  */
113 static vaddr_t      virtual_space_start;
114 static vaddr_t      virtual_space_end;
115 
116 /*
117  * local prototypes
118  */
119 static void uvm_pageinsert(struct vm_page *);
120 static void uvm_pageremove(struct vm_page *);
121 int uvm_page_owner_locked_p(struct vm_page *);
122 
123 /*
124  * inline functions
125  */
126 /*
127  * uvm_pageinsert: insert a page in the object
128  *
129  * => caller must lock object
130  * => call should have already set pg's object and offset pointers
131  *    and bumped the version counter
132  */
133 static inline void
uvm_pageinsert(struct vm_page * pg)134 uvm_pageinsert(struct vm_page *pg)
135 {
136 	struct vm_page	*dupe;
137 
138 	KASSERT(UVM_OBJ_IS_DUMMY(pg->uobject) ||
139 	    rw_write_held(pg->uobject->vmobjlock));
140 	KASSERT((pg->pg_flags & PG_TABLED) == 0);
141 
142 	dupe = RBT_INSERT(uvm_objtree, &pg->uobject->memt, pg);
143 	/* not allowed to insert over another page */
144 	KASSERT(dupe == NULL);
145 	atomic_setbits_int(&pg->pg_flags, PG_TABLED);
146 	pg->uobject->uo_npages++;
147 }
148 
149 /*
150  * uvm_page_remove: remove page from object
151  *
152  * => caller must lock object
153  */
154 static inline void
uvm_pageremove(struct vm_page * pg)155 uvm_pageremove(struct vm_page *pg)
156 {
157 	KASSERT(UVM_OBJ_IS_DUMMY(pg->uobject) ||
158 	    rw_write_held(pg->uobject->vmobjlock));
159 	KASSERT(pg->pg_flags & PG_TABLED);
160 
161 	RBT_REMOVE(uvm_objtree, &pg->uobject->memt, pg);
162 
163 	atomic_clearbits_int(&pg->pg_flags, PG_TABLED);
164 	pg->uobject->uo_npages--;
165 	pg->uobject = NULL;
166 	pg->pg_version++;
167 }
168 
169 /*
170  * uvm_page_init: init the page system.   called from uvm_init().
171  *
172  * => we return the range of kernel virtual memory in kvm_startp/kvm_endp
173  */
174 void
uvm_page_init(vaddr_t * kvm_startp,vaddr_t * kvm_endp)175 uvm_page_init(vaddr_t *kvm_startp, vaddr_t *kvm_endp)
176 {
177 	vsize_t freepages, pagecount, n;
178 	vm_page_t pagearray, curpg;
179 	int lcv, i;
180 	paddr_t paddr, pgno;
181 	struct vm_physseg *seg;
182 
183 	/*
184 	 * init the page queues and page queue locks
185 	 */
186 
187 	TAILQ_INIT(&uvm.page_active);
188 	TAILQ_INIT(&uvm.page_inactive);
189 	mtx_init(&uvm.pageqlock, IPL_VM);
190 	mtx_init(&uvm.fpageqlock, IPL_VM);
191 	uvm_pmr_init();
192 
193 	/*
194 	 * allocate vm_page structures.
195 	 */
196 
197 	/*
198 	 * sanity check:
199 	 * before calling this function the MD code is expected to register
200 	 * some free RAM with the uvm_page_physload() function.   our job
201 	 * now is to allocate vm_page structures for this memory.
202 	 */
203 
204 	if (vm_nphysseg == 0)
205 		panic("uvm_page_bootstrap: no memory pre-allocated");
206 
207 	/*
208 	 * first calculate the number of free pages...
209 	 *
210 	 * note that we use start/end rather than avail_start/avail_end.
211 	 * this allows us to allocate extra vm_page structures in case we
212 	 * want to return some memory to the pool after booting.
213 	 */
214 
215 	freepages = 0;
216 	for (lcv = 0, seg = vm_physmem; lcv < vm_nphysseg ; lcv++, seg++)
217 		freepages += (seg->end - seg->start);
218 
219 	/*
220 	 * we now know we have (PAGE_SIZE * freepages) bytes of memory we can
221 	 * use.   for each page of memory we use we need a vm_page structure.
222 	 * thus, the total number of pages we can use is the total size of
223 	 * the memory divided by the PAGE_SIZE plus the size of the vm_page
224 	 * structure.   we add one to freepages as a fudge factor to avoid
225 	 * truncation errors (since we can only allocate in terms of whole
226 	 * pages).
227 	 */
228 
229 	pagecount = (((paddr_t)freepages + 1) << PAGE_SHIFT) /
230 	    (PAGE_SIZE + sizeof(struct vm_page));
231 	pagearray = (vm_page_t)uvm_pageboot_alloc(pagecount *
232 	    sizeof(struct vm_page));
233 	memset(pagearray, 0, pagecount * sizeof(struct vm_page));
234 
235 	/* init the vm_page structures and put them in the correct place. */
236 	for (lcv = 0, seg = vm_physmem; lcv < vm_nphysseg ; lcv++, seg++) {
237 		n = seg->end - seg->start;
238 		if (n > pagecount) {
239 			panic("uvm_page_init: lost %ld page(s) in init",
240 			    (long)(n - pagecount));
241 			    /* XXXCDC: shouldn't happen? */
242 			/* n = pagecount; */
243 		}
244 
245 		/* set up page array pointers */
246 		seg->pgs = pagearray;
247 		pagearray += n;
248 		pagecount -= n;
249 		seg->lastpg = seg->pgs + (n - 1);
250 
251 		/* init and free vm_pages (we've already zeroed them) */
252 		pgno = seg->start;
253 		paddr = ptoa(pgno);
254 		for (i = 0, curpg = seg->pgs; i < n;
255 		    i++, curpg++, pgno++, paddr += PAGE_SIZE) {
256 			curpg->phys_addr = paddr;
257 			VM_MDPAGE_INIT(curpg);
258 			if (pgno >= seg->avail_start &&
259 			    pgno < seg->avail_end) {
260 				uvmexp.npages++;
261 			}
262 		}
263 
264 		/* Add pages to free pool. */
265 		uvm_pmr_freepages(&seg->pgs[seg->avail_start - seg->start],
266 		    seg->avail_end - seg->avail_start);
267 	}
268 
269 	/*
270 	 * pass up the values of virtual_space_start and
271 	 * virtual_space_end (obtained by uvm_pageboot_alloc) to the upper
272 	 * layers of the VM.
273 	 */
274 
275 	*kvm_startp = round_page(virtual_space_start);
276 	*kvm_endp = trunc_page(virtual_space_end);
277 
278 	/* init locks for kernel threads */
279 	mtx_init(&uvm.aiodoned_lock, IPL_BIO);
280 
281 	/*
282 	 * init reserve thresholds
283 	 * XXXCDC - values may need adjusting
284 	 */
285 	uvmexp.reserve_pagedaemon = 4;
286 	uvmexp.reserve_kernel = 8;
287 	uvmexp.anonminpct = 10;
288 	uvmexp.vnodeminpct = 10;
289 	uvmexp.vtextminpct = 5;
290 	uvmexp.anonmin = uvmexp.anonminpct * 256 / 100;
291 	uvmexp.vnodemin = uvmexp.vnodeminpct * 256 / 100;
292 	uvmexp.vtextmin = uvmexp.vtextminpct * 256 / 100;
293 
294 	uvm.page_init_done = TRUE;
295 }
296 
297 /*
298  * uvm_setpagesize: set the page size
299  *
300  * => sets page_shift and page_mask from uvmexp.pagesize.
301  */
302 void
uvm_setpagesize(void)303 uvm_setpagesize(void)
304 {
305 	if (uvmexp.pagesize == 0)
306 		uvmexp.pagesize = DEFAULT_PAGE_SIZE;
307 	uvmexp.pagemask = uvmexp.pagesize - 1;
308 	if ((uvmexp.pagemask & uvmexp.pagesize) != 0)
309 		panic("uvm_setpagesize: page size not a power of two");
310 	for (uvmexp.pageshift = 0; ; uvmexp.pageshift++)
311 		if ((1 << uvmexp.pageshift) == uvmexp.pagesize)
312 			break;
313 }
314 
315 /*
316  * uvm_pageboot_alloc: steal memory from physmem for bootstrapping
317  */
318 vaddr_t
uvm_pageboot_alloc(vsize_t size)319 uvm_pageboot_alloc(vsize_t size)
320 {
321 #if defined(PMAP_STEAL_MEMORY)
322 	vaddr_t addr;
323 
324 	/*
325 	 * defer bootstrap allocation to MD code (it may want to allocate
326 	 * from a direct-mapped segment).  pmap_steal_memory should round
327 	 * off virtual_space_start/virtual_space_end.
328 	 */
329 
330 	addr = pmap_steal_memory(size, &virtual_space_start,
331 	    &virtual_space_end);
332 
333 	return addr;
334 
335 #else /* !PMAP_STEAL_MEMORY */
336 
337 	static boolean_t initialized = FALSE;
338 	vaddr_t addr, vaddr;
339 	paddr_t paddr;
340 
341 	/* round to page size */
342 	size = round_page(size);
343 
344 	/* on first call to this function, initialize ourselves. */
345 	if (initialized == FALSE) {
346 		pmap_virtual_space(&virtual_space_start, &virtual_space_end);
347 
348 		/* round it the way we like it */
349 		virtual_space_start = round_page(virtual_space_start);
350 		virtual_space_end = trunc_page(virtual_space_end);
351 
352 		initialized = TRUE;
353 	}
354 
355 	/* allocate virtual memory for this request */
356 	if (virtual_space_start == virtual_space_end ||
357 	    (virtual_space_end - virtual_space_start) < size)
358 		panic("uvm_pageboot_alloc: out of virtual space");
359 
360 	addr = virtual_space_start;
361 
362 #ifdef PMAP_GROWKERNEL
363 	/*
364 	 * If the kernel pmap can't map the requested space,
365 	 * then allocate more resources for it.
366 	 */
367 	if (uvm_maxkaddr < (addr + size)) {
368 		uvm_maxkaddr = pmap_growkernel(addr + size);
369 		if (uvm_maxkaddr < (addr + size))
370 			panic("uvm_pageboot_alloc: pmap_growkernel() failed");
371 	}
372 #endif
373 
374 	virtual_space_start += size;
375 
376 	/* allocate and mapin physical pages to back new virtual pages */
377 	for (vaddr = round_page(addr) ; vaddr < addr + size ;
378 	    vaddr += PAGE_SIZE) {
379 		if (!uvm_page_physget(&paddr))
380 			panic("uvm_pageboot_alloc: out of memory");
381 
382 		/*
383 		 * Note this memory is no longer managed, so using
384 		 * pmap_kenter is safe.
385 		 */
386 		pmap_kenter_pa(vaddr, paddr, PROT_READ | PROT_WRITE);
387 	}
388 	pmap_update(pmap_kernel());
389 	return addr;
390 #endif	/* PMAP_STEAL_MEMORY */
391 }
392 
393 #if !defined(PMAP_STEAL_MEMORY)
394 /*
395  * uvm_page_physget: "steal" one page from the vm_physmem structure.
396  *
397  * => attempt to allocate it off the end of a segment in which the "avail"
398  *    values match the start/end values.   if we can't do that, then we
399  *    will advance both values (making them equal, and removing some
400  *    vm_page structures from the non-avail area).
401  * => return false if out of memory.
402  */
403 
404 boolean_t
uvm_page_physget(paddr_t * paddrp)405 uvm_page_physget(paddr_t *paddrp)
406 {
407 	int lcv;
408 	struct vm_physseg *seg;
409 
410 	/* pass 1: try allocating from a matching end */
411 #if (VM_PHYSSEG_STRAT == VM_PSTRAT_BIGFIRST) || \
412 	(VM_PHYSSEG_STRAT == VM_PSTRAT_BSEARCH)
413 	for (lcv = vm_nphysseg - 1, seg = vm_physmem + lcv; lcv >= 0;
414 	    lcv--, seg--)
415 #else
416 	for (lcv = 0, seg = vm_physmem; lcv < vm_nphysseg ; lcv++, seg++)
417 #endif
418 	{
419 		if (uvm.page_init_done == TRUE)
420 			panic("uvm_page_physget: called _after_ bootstrap");
421 
422 		/* try from front */
423 		if (seg->avail_start == seg->start &&
424 		    seg->avail_start < seg->avail_end) {
425 			*paddrp = ptoa(seg->avail_start);
426 			seg->avail_start++;
427 			seg->start++;
428 			/* nothing left?   nuke it */
429 			if (seg->avail_start == seg->end) {
430 				if (vm_nphysseg == 1)
431 				    panic("uvm_page_physget: out of memory!");
432 				vm_nphysseg--;
433 				for (; lcv < vm_nphysseg; lcv++, seg++)
434 					/* structure copy */
435 					seg[0] = seg[1];
436 			}
437 			return TRUE;
438 		}
439 
440 		/* try from rear */
441 		if (seg->avail_end == seg->end &&
442 		    seg->avail_start < seg->avail_end) {
443 			*paddrp = ptoa(seg->avail_end - 1);
444 			seg->avail_end--;
445 			seg->end--;
446 			/* nothing left?   nuke it */
447 			if (seg->avail_end == seg->start) {
448 				if (vm_nphysseg == 1)
449 				    panic("uvm_page_physget: out of memory!");
450 				vm_nphysseg--;
451 				for (; lcv < vm_nphysseg ; lcv++, seg++)
452 					/* structure copy */
453 					seg[0] = seg[1];
454 			}
455 			return TRUE;
456 		}
457 	}
458 
459 	/* pass2: forget about matching ends, just allocate something */
460 #if (VM_PHYSSEG_STRAT == VM_PSTRAT_BIGFIRST) || \
461 	(VM_PHYSSEG_STRAT == VM_PSTRAT_BSEARCH)
462 	for (lcv = vm_nphysseg - 1, seg = vm_physmem + lcv; lcv >= 0;
463 	    lcv--, seg--)
464 #else
465 	for (lcv = 0, seg = vm_physmem; lcv < vm_nphysseg ; lcv++, seg++)
466 #endif
467 	{
468 
469 		/* any room in this bank? */
470 		if (seg->avail_start >= seg->avail_end)
471 			continue;  /* nope */
472 
473 		*paddrp = ptoa(seg->avail_start);
474 		seg->avail_start++;
475 		/* truncate! */
476 		seg->start = seg->avail_start;
477 
478 		/* nothing left?   nuke it */
479 		if (seg->avail_start == seg->end) {
480 			if (vm_nphysseg == 1)
481 				panic("uvm_page_physget: out of memory!");
482 			vm_nphysseg--;
483 			for (; lcv < vm_nphysseg ; lcv++, seg++)
484 				/* structure copy */
485 				seg[0] = seg[1];
486 		}
487 		return TRUE;
488 	}
489 
490 	return FALSE;        /* whoops! */
491 }
492 
493 #endif /* PMAP_STEAL_MEMORY */
494 
495 /*
496  * uvm_page_physload: load physical memory into VM system
497  *
498  * => all args are PFs
499  * => all pages in start/end get vm_page structures
500  * => areas marked by avail_start/avail_end get added to the free page pool
501  * => we are limited to VM_PHYSSEG_MAX physical memory segments
502  */
503 
504 void
uvm_page_physload(paddr_t start,paddr_t end,paddr_t avail_start,paddr_t avail_end,int flags)505 uvm_page_physload(paddr_t start, paddr_t end, paddr_t avail_start,
506     paddr_t avail_end, int flags)
507 {
508 	int preload, lcv;
509 	psize_t npages;
510 	struct vm_page *pgs;
511 	struct vm_physseg *ps, *seg;
512 
513 #ifdef DIAGNOSTIC
514 	if (uvmexp.pagesize == 0)
515 		panic("uvm_page_physload: page size not set!");
516 
517 	if (start >= end)
518 		panic("uvm_page_physload: start >= end");
519 #endif
520 
521 	/* do we have room? */
522 	if (vm_nphysseg == VM_PHYSSEG_MAX) {
523 		printf("uvm_page_physload: unable to load physical memory "
524 		    "segment\n");
525 		printf("\t%d segments allocated, ignoring 0x%llx -> 0x%llx\n",
526 		    VM_PHYSSEG_MAX, (long long)start, (long long)end);
527 		printf("\tincrease VM_PHYSSEG_MAX\n");
528 		return;
529 	}
530 
531 	/*
532 	 * check to see if this is a "preload" (i.e. uvm_mem_init hasn't been
533 	 * called yet, so malloc is not available).
534 	 */
535 	for (lcv = 0, seg = vm_physmem; lcv < vm_nphysseg; lcv++, seg++) {
536 		if (seg->pgs)
537 			break;
538 	}
539 	preload = (lcv == vm_nphysseg);
540 
541 	/* if VM is already running, attempt to malloc() vm_page structures */
542 	if (!preload) {
543 		/*
544 		 * XXXCDC: need some sort of lockout for this case
545 		 * right now it is only used by devices so it should be alright.
546 		 */
547  		paddr_t paddr;
548 
549  		npages = end - start;  /* # of pages */
550 
551 		pgs = km_alloc(round_page(npages * sizeof(*pgs)),
552 		    &kv_any, &kp_zero, &kd_waitok);
553 		if (pgs == NULL) {
554 			printf("uvm_page_physload: can not malloc vm_page "
555 			    "structs for segment\n");
556 			printf("\tignoring 0x%lx -> 0x%lx\n", start, end);
557 			return;
558 		}
559 		/* init phys_addr and free pages, XXX uvmexp.npages */
560 		for (lcv = 0, paddr = ptoa(start); lcv < npages;
561 		    lcv++, paddr += PAGE_SIZE) {
562 			pgs[lcv].phys_addr = paddr;
563 			VM_MDPAGE_INIT(&pgs[lcv]);
564 			if (atop(paddr) >= avail_start &&
565 			    atop(paddr) < avail_end) {
566 				if (flags & PHYSLOAD_DEVICE) {
567 					atomic_setbits_int(&pgs[lcv].pg_flags,
568 					    PG_DEV);
569 					pgs[lcv].wire_count = 1;
570 				} else {
571 #if defined(VM_PHYSSEG_NOADD)
572 		panic("uvm_page_physload: tried to add RAM after vm_mem_init");
573 #endif
574 				}
575 			}
576 		}
577 
578 		/* Add pages to free pool. */
579 		if ((flags & PHYSLOAD_DEVICE) == 0) {
580 			uvm_pmr_freepages(&pgs[avail_start - start],
581 			    avail_end - avail_start);
582 		}
583 
584 		/* XXXCDC: need hook to tell pmap to rebuild pv_list, etc... */
585 	} else {
586 		/* gcc complains if these don't get init'd */
587 		pgs = NULL;
588 		npages = 0;
589 
590 	}
591 
592 	/* now insert us in the proper place in vm_physmem[] */
593 #if (VM_PHYSSEG_STRAT == VM_PSTRAT_RANDOM)
594 	/* random: put it at the end (easy!) */
595 	ps = &vm_physmem[vm_nphysseg];
596 #elif (VM_PHYSSEG_STRAT == VM_PSTRAT_BSEARCH)
597 	{
598 		int x;
599 		/* sort by address for binary search */
600 		for (lcv = 0, seg = vm_physmem; lcv < vm_nphysseg; lcv++, seg++)
601 			if (start < seg->start)
602 				break;
603 		ps = seg;
604 		/* move back other entries, if necessary ... */
605 		for (x = vm_nphysseg, seg = vm_physmem + x - 1; x > lcv;
606 		    x--, seg--)
607 			/* structure copy */
608 			seg[1] = seg[0];
609 	}
610 #elif (VM_PHYSSEG_STRAT == VM_PSTRAT_BIGFIRST)
611 	{
612 		int x;
613 		/* sort by largest segment first */
614 		for (lcv = 0, seg = vm_physmem; lcv < vm_nphysseg; lcv++, seg++)
615 			if ((end - start) >
616 			    (seg->end - seg->start))
617 				break;
618 		ps = &vm_physmem[lcv];
619 		/* move back other entries, if necessary ... */
620 		for (x = vm_nphysseg, seg = vm_physmem + x - 1; x > lcv;
621 		    x--, seg--)
622 			/* structure copy */
623 			seg[1] = seg[0];
624 	}
625 #else
626 	panic("uvm_page_physload: unknown physseg strategy selected!");
627 #endif
628 
629 	ps->start = start;
630 	ps->end = end;
631 	ps->avail_start = avail_start;
632 	ps->avail_end = avail_end;
633 	if (preload) {
634 		ps->pgs = NULL;
635 	} else {
636 		ps->pgs = pgs;
637 		ps->lastpg = pgs + npages - 1;
638 	}
639 	vm_nphysseg++;
640 
641 	return;
642 }
643 
644 #ifdef DDB /* XXXCDC: TMP TMP TMP DEBUG DEBUG DEBUG */
645 
646 void uvm_page_physdump(void); /* SHUT UP GCC */
647 
648 /* call from DDB */
649 void
uvm_page_physdump(void)650 uvm_page_physdump(void)
651 {
652 	int lcv;
653 	struct vm_physseg *seg;
654 
655 	printf("uvm_page_physdump: physical memory config [segs=%d of %d]:\n",
656 	    vm_nphysseg, VM_PHYSSEG_MAX);
657 	for (lcv = 0, seg = vm_physmem; lcv < vm_nphysseg ; lcv++, seg++)
658 		printf("0x%llx->0x%llx [0x%llx->0x%llx]\n",
659 		    (long long)seg->start,
660 		    (long long)seg->end,
661 		    (long long)seg->avail_start,
662 		    (long long)seg->avail_end);
663 	printf("STRATEGY = ");
664 	switch (VM_PHYSSEG_STRAT) {
665 	case VM_PSTRAT_RANDOM: printf("RANDOM\n"); break;
666 	case VM_PSTRAT_BSEARCH: printf("BSEARCH\n"); break;
667 	case VM_PSTRAT_BIGFIRST: printf("BIGFIRST\n"); break;
668 	default: printf("<<UNKNOWN>>!!!!\n");
669 	}
670 }
671 #endif
672 
673 void
uvm_shutdown(void)674 uvm_shutdown(void)
675 {
676 #ifdef UVM_SWAP_ENCRYPT
677 	uvm_swap_finicrypt_all();
678 #endif
679 	smr_flush();
680 }
681 
682 /*
683  * Perform insert of a given page in the specified anon of obj.
684  * This is basically, uvm_pagealloc, but with the page already given.
685  */
686 void
uvm_pagealloc_pg(struct vm_page * pg,struct uvm_object * obj,voff_t off,struct vm_anon * anon)687 uvm_pagealloc_pg(struct vm_page *pg, struct uvm_object *obj, voff_t off,
688     struct vm_anon *anon)
689 {
690 	int	flags;
691 
692 	KASSERT(obj == NULL || anon == NULL);
693 	KASSERT(anon == NULL || off == 0);
694 	KASSERT(off == trunc_page(off));
695 	KASSERT(obj == NULL || UVM_OBJ_IS_DUMMY(obj) ||
696 	    rw_write_held(obj->vmobjlock));
697 	KASSERT(anon == NULL || anon->an_lock == NULL ||
698 	    rw_write_held(anon->an_lock));
699 
700 	flags = PG_BUSY | PG_FAKE;
701 	pg->offset = off;
702 	pg->uobject = obj;
703 	pg->uanon = anon;
704 	KASSERT(uvm_page_owner_locked_p(pg));
705 	if (anon) {
706 		anon->an_page = pg;
707 		flags |= PQ_ANON;
708 	} else if (obj)
709 		uvm_pageinsert(pg);
710 	atomic_setbits_int(&pg->pg_flags, flags);
711 #if defined(UVM_PAGE_TRKOWN)
712 	pg->owner_tag = NULL;
713 #endif
714 	UVM_PAGE_OWN(pg, "new alloc");
715 }
716 
717 /*
718  * uvm_pglistalloc: allocate a list of pages
719  *
720  * => allocated pages are placed at the tail of rlist.  rlist is
721  *    assumed to be properly initialized by caller.
722  * => returns 0 on success or errno on failure
723  * => doesn't take into account clean non-busy pages on inactive list
724  *	that could be used(?)
725  * => params:
726  *	size		the size of the allocation, rounded to page size.
727  *	low		the low address of the allowed allocation range.
728  *	high		the high address of the allowed allocation range.
729  *	alignment	memory must be aligned to this power-of-two boundary.
730  *	boundary	no segment in the allocation may cross this
731  *			power-of-two boundary (relative to zero).
732  * => flags:
733  *	UVM_PLA_NOWAIT	fail if allocation fails
734  *	UVM_PLA_WAITOK	wait for memory to become avail
735  *	UVM_PLA_ZERO	return zeroed memory
736  */
737 int
uvm_pglistalloc(psize_t size,paddr_t low,paddr_t high,paddr_t alignment,paddr_t boundary,struct pglist * rlist,int nsegs,int flags)738 uvm_pglistalloc(psize_t size, paddr_t low, paddr_t high, paddr_t alignment,
739     paddr_t boundary, struct pglist *rlist, int nsegs, int flags)
740 {
741 	KASSERT((alignment & (alignment - 1)) == 0);
742 	KASSERT((boundary & (boundary - 1)) == 0);
743 	KASSERT(!(flags & UVM_PLA_WAITOK) ^ !(flags & UVM_PLA_NOWAIT));
744 
745 	if (size == 0)
746 		return EINVAL;
747 	size = atop(round_page(size));
748 
749 	/*
750 	 * XXX uvm_pglistalloc is currently only used for kernel
751 	 * objects. Unlike the checks in uvm_pagealloc, below, here
752 	 * we are always allowed to use the kernel reserve.
753 	 */
754 	flags |= UVM_PLA_USERESERVE;
755 
756 	if ((high & PAGE_MASK) != PAGE_MASK) {
757 		printf("uvm_pglistalloc: Upper boundary 0x%lx "
758 		    "not on pagemask.\n", (unsigned long)high);
759 	}
760 
761 	/*
762 	 * Our allocations are always page granularity, so our alignment
763 	 * must be, too.
764 	 */
765 	if (alignment < PAGE_SIZE)
766 		alignment = PAGE_SIZE;
767 
768 	low = atop(roundup(low, alignment));
769 	/*
770 	 * high + 1 may result in overflow, in which case high becomes 0x0,
771 	 * which is the 'don't care' value.
772 	 * The only requirement in that case is that low is also 0x0, or the
773 	 * low<high assert will fail.
774 	 */
775 	high = atop(high + 1);
776 	alignment = atop(alignment);
777 	if (boundary < PAGE_SIZE && boundary != 0)
778 		boundary = PAGE_SIZE;
779 	boundary = atop(boundary);
780 
781 	return uvm_pmr_getpages(size, low, high, alignment, boundary, nsegs,
782 	    flags, rlist);
783 }
784 
785 /*
786  * uvm_pglistfree: free a list of pages
787  *
788  * => pages should already be unmapped
789  */
790 void
uvm_pglistfree(struct pglist * list)791 uvm_pglistfree(struct pglist *list)
792 {
793 	uvm_pmr_freepageq(list);
794 }
795 
796 /*
797  * interface used by the buffer cache to allocate a buffer at a time.
798  * The pages are allocated wired in DMA accessible memory
799  */
800 int
uvm_pagealloc_multi(struct uvm_object * obj,voff_t off,vsize_t size,int flags)801 uvm_pagealloc_multi(struct uvm_object *obj, voff_t off, vsize_t size,
802     int flags)
803 {
804 	struct pglist    plist;
805 	struct vm_page  *pg;
806 	int              i, r;
807 
808 	KASSERT(UVM_OBJ_IS_BUFCACHE(obj));
809 	KERNEL_ASSERT_LOCKED();
810 
811 	TAILQ_INIT(&plist);
812 	r = uvm_pglistalloc(size, dma_constraint.ucr_low,
813 	    dma_constraint.ucr_high, 0, 0, &plist, atop(round_page(size)),
814 	    flags);
815 	if (r == 0) {
816 		i = 0;
817 		while ((pg = TAILQ_FIRST(&plist)) != NULL) {
818 			pg->wire_count = 1;
819 			atomic_setbits_int(&pg->pg_flags, PG_CLEAN | PG_FAKE);
820 			KASSERT((pg->pg_flags & PG_DEV) == 0);
821 			TAILQ_REMOVE(&plist, pg, pageq);
822 			uvm_pagealloc_pg(pg, obj, off + ptoa(i++), NULL);
823 		}
824 	}
825 	return r;
826 }
827 
828 /*
829  * interface used by the buffer cache to reallocate a buffer at a time.
830  * The pages are reallocated wired outside the DMA accessible region.
831  *
832  */
833 int
uvm_pagerealloc_multi(struct uvm_object * obj,voff_t off,vsize_t size,int flags,struct uvm_constraint_range * where)834 uvm_pagerealloc_multi(struct uvm_object *obj, voff_t off, vsize_t size,
835     int flags, struct uvm_constraint_range *where)
836 {
837 	struct pglist    plist;
838 	struct vm_page  *pg, *tpg;
839 	int              i, r;
840 	voff_t		offset;
841 
842 	KASSERT(UVM_OBJ_IS_BUFCACHE(obj));
843 	KERNEL_ASSERT_LOCKED();
844 
845 	TAILQ_INIT(&plist);
846 	if (size == 0)
847 		panic("size 0 uvm_pagerealloc");
848 	r = uvm_pglistalloc(size, where->ucr_low, where->ucr_high, 0,
849 	    0, &plist, atop(round_page(size)), flags);
850 	if (r == 0) {
851 		i = 0;
852 		while((pg = TAILQ_FIRST(&plist)) != NULL) {
853 			offset = off + ptoa(i++);
854 			tpg = uvm_pagelookup(obj, offset);
855 			KASSERT(tpg != NULL);
856 			pg->wire_count = 1;
857 			atomic_setbits_int(&pg->pg_flags, PG_CLEAN | PG_FAKE);
858 			KASSERT((pg->pg_flags & PG_DEV) == 0);
859 			TAILQ_REMOVE(&plist, pg, pageq);
860 			uvm_pagecopy(tpg, pg);
861 			KASSERT(tpg->wire_count == 1);
862 			tpg->wire_count = 0;
863 			uvm_lock_pageq();
864 			uvm_pagefree(tpg);
865 			uvm_unlock_pageq();
866 			uvm_pagealloc_pg(pg, obj, offset, NULL);
867 		}
868 	}
869 	return r;
870 }
871 
872 /*
873  * uvm_pagealloc: allocate vm_page from a particular free list.
874  *
875  * => return null if no pages free
876  * => wake up pagedaemon if number of free pages drops below low water mark
877  * => only one of obj or anon can be non-null
878  * => caller must activate/deactivate page if it is not wired.
879  */
880 struct vm_page *
uvm_pagealloc(struct uvm_object * obj,voff_t off,struct vm_anon * anon,int flags)881 uvm_pagealloc(struct uvm_object *obj, voff_t off, struct vm_anon *anon,
882     int flags)
883 {
884 	struct vm_page *pg = NULL;
885 	int pmr_flags;
886 
887 	KASSERT(obj == NULL || anon == NULL);
888 	KASSERT(anon == NULL || off == 0);
889 	KASSERT(off == trunc_page(off));
890 	KASSERT(obj == NULL || UVM_OBJ_IS_DUMMY(obj) ||
891 	    rw_write_held(obj->vmobjlock));
892 	KASSERT(anon == NULL || anon->an_lock == NULL ||
893 	    rw_write_held(anon->an_lock));
894 
895 	pmr_flags = UVM_PLA_NOWAIT;
896 
897 	/*
898 	 * We're allowed to use the kernel reserve if the page is
899 	 * being allocated to a kernel object.
900 	 */
901 	if ((flags & UVM_PGA_USERESERVE) ||
902 	    (obj != NULL && UVM_OBJ_IS_KERN_OBJECT(obj)))
903 	    	pmr_flags |= UVM_PLA_USERESERVE;
904 
905 	if (flags & UVM_PGA_ZERO)
906 		pmr_flags |= UVM_PLA_ZERO;
907 
908 	pg = uvm_pmr_cache_get(pmr_flags);
909 	if (pg == NULL)
910 		return NULL;
911 	uvm_pagealloc_pg(pg, obj, off, anon);
912 	KASSERT((pg->pg_flags & PG_DEV) == 0);
913 	if (flags & UVM_PGA_ZERO)
914 		atomic_clearbits_int(&pg->pg_flags, PG_CLEAN);
915 	else
916 		atomic_setbits_int(&pg->pg_flags, PG_CLEAN);
917 
918 	return pg;
919 }
920 
921 /*
922  * uvm_pagerealloc: reallocate a page from one object to another
923  */
924 
925 void
uvm_pagerealloc(struct vm_page * pg,struct uvm_object * newobj,voff_t newoff)926 uvm_pagerealloc(struct vm_page *pg, struct uvm_object *newobj, voff_t newoff)
927 {
928 
929 	/* remove it from the old object */
930 	if (pg->uobject) {
931 		uvm_pageremove(pg);
932 	}
933 
934 	/* put it in the new object */
935 	if (newobj) {
936 		pg->uobject = newobj;
937 		pg->offset = newoff;
938 		pg->pg_version++;
939 		uvm_pageinsert(pg);
940 	}
941 }
942 
943 /*
944  * uvm_pageclean: clean page
945  *
946  * => erase page's identity (i.e. remove from object)
947  * => caller must lock page queues if `pg' is managed
948  * => assumes all valid mappings of pg are gone
949  */
950 void
uvm_pageclean(struct vm_page * pg)951 uvm_pageclean(struct vm_page *pg)
952 {
953 	u_int flags_to_clear = 0;
954 
955 	if ((pg->pg_flags & (PG_TABLED|PQ_ACTIVE|PQ_INACTIVE)) &&
956 	    (pg->uobject == NULL || !UVM_OBJ_IS_PMAP(pg->uobject)))
957 		MUTEX_ASSERT_LOCKED(&uvm.pageqlock);
958 
959 #ifdef DEBUG
960 	if (pg->uobject == (void *)0xdeadbeef &&
961 	    pg->uanon == (void *)0xdeadbeef) {
962 		panic("uvm_pagefree: freeing free page %p", pg);
963 	}
964 #endif
965 
966 	KASSERT((pg->pg_flags & PG_DEV) == 0);
967 	KASSERT(pg->uobject == NULL || UVM_OBJ_IS_DUMMY(pg->uobject) ||
968 	    rw_write_held(pg->uobject->vmobjlock));
969 	KASSERT(pg->uobject != NULL || pg->uanon == NULL ||
970 	    rw_write_held(pg->uanon->an_lock));
971 
972 	/*
973 	 * if the page was an object page (and thus "TABLED"), remove it
974 	 * from the object.
975 	 */
976 	if (pg->pg_flags & PG_TABLED)
977 		uvm_pageremove(pg);
978 
979 	/*
980 	 * now remove the page from the queues
981 	 */
982 	uvm_pagedequeue(pg);
983 
984 	/*
985 	 * if the page was wired, unwire it now.
986 	 */
987 	if (pg->wire_count) {
988 		pg->wire_count = 0;
989 		uvmexp.wired--;
990 	}
991 	if (pg->uanon) {
992 		pg->uanon->an_page = NULL;
993 		pg->uanon = NULL;
994 	}
995 
996 	/* Clean page state bits. */
997 	flags_to_clear |= PQ_ANON|PQ_AOBJ|PQ_ENCRYPT|PG_ZERO|PG_FAKE|PG_BUSY|
998 	    PG_RELEASED|PG_CLEAN|PG_CLEANCHK;
999 	atomic_clearbits_int(&pg->pg_flags, flags_to_clear);
1000 
1001 #ifdef DEBUG
1002 	pg->uobject = (void *)0xdeadbeef;
1003 	pg->offset = 0xdeadbeef;
1004 	pg->uanon = (void *)0xdeadbeef;
1005 #endif
1006 }
1007 
1008 /*
1009  * uvm_pagefree: free page
1010  *
1011  * => erase page's identity (i.e. remove from object)
1012  * => put page on free list
1013  * => caller must lock page queues if `pg' is managed
1014  * => assumes all valid mappings of pg are gone
1015  */
1016 void
uvm_pagefree(struct vm_page * pg)1017 uvm_pagefree(struct vm_page *pg)
1018 {
1019 	uvm_pageclean(pg);
1020 	uvm_pmr_cache_put(pg);
1021 }
1022 
1023 /*
1024  * uvm_page_unbusy: unbusy an array of pages.
1025  *
1026  * => pages must either all belong to the same object, or all belong to anons.
1027  * => if pages are object-owned, object must be locked.
1028  * => if pages are anon-owned, anons must have 0 refcount.
1029  * => caller must make sure that anon-owned pages are not PG_RELEASED.
1030  */
1031 void
uvm_page_unbusy(struct vm_page ** pgs,int npgs)1032 uvm_page_unbusy(struct vm_page **pgs, int npgs)
1033 {
1034 	struct vm_page *pg;
1035 	int i;
1036 
1037 	for (i = 0; i < npgs; i++) {
1038 		pg = pgs[i];
1039 
1040 		if (pg == NULL || pg == PGO_DONTCARE) {
1041 			continue;
1042 		}
1043 
1044 		KASSERT(uvm_page_owner_locked_p(pg));
1045 		KASSERT(pg->pg_flags & PG_BUSY);
1046 
1047 		if (pg->pg_flags & PG_WANTED) {
1048 			wakeup(pg);
1049 		}
1050 		if (pg->pg_flags & PG_RELEASED) {
1051 			KASSERT(pg->uobject != NULL ||
1052 			    (pg->uanon != NULL && pg->uanon->an_ref > 0));
1053 			atomic_clearbits_int(&pg->pg_flags, PG_RELEASED);
1054 			pmap_page_protect(pg, PROT_NONE);
1055 			uvm_pagefree(pg);
1056 		} else {
1057 			KASSERT((pg->pg_flags & PG_FAKE) == 0);
1058 			atomic_clearbits_int(&pg->pg_flags, PG_WANTED|PG_BUSY);
1059 			UVM_PAGE_OWN(pg, NULL);
1060 		}
1061 	}
1062 }
1063 
1064 /*
1065  * uvm_pagewait: wait for a busy page
1066  *
1067  * => page must be known PG_BUSY
1068  * => object must be locked
1069  * => object will be unlocked on return
1070  */
1071 void
uvm_pagewait(struct vm_page * pg,struct rwlock * lock,const char * wmesg)1072 uvm_pagewait(struct vm_page *pg, struct rwlock *lock, const char *wmesg)
1073 {
1074 	KASSERT(rw_lock_held(lock));
1075 	KASSERT((pg->pg_flags & PG_BUSY) != 0);
1076 
1077 	atomic_setbits_int(&pg->pg_flags, PG_WANTED);
1078 	rwsleep_nsec(pg, lock, PVM | PNORELOCK, wmesg, INFSLP);
1079 }
1080 
1081 #if defined(UVM_PAGE_TRKOWN)
1082 /*
1083  * uvm_page_own: set or release page ownership
1084  *
1085  * => this is a debugging function that keeps track of who sets PG_BUSY
1086  *	and where they do it.   it can be used to track down problems
1087  *	such a thread setting "PG_BUSY" and never releasing it.
1088  * => if "tag" is NULL then we are releasing page ownership
1089  */
1090 void
uvm_page_own(struct vm_page * pg,char * tag)1091 uvm_page_own(struct vm_page *pg, char *tag)
1092 {
1093 	/* gain ownership? */
1094 	if (tag) {
1095 		if (pg->owner_tag) {
1096 			printf("uvm_page_own: page %p already owned "
1097 			    "by thread %d [%s]\n", pg,
1098 			     pg->owner, pg->owner_tag);
1099 			panic("uvm_page_own");
1100 		}
1101 		pg->owner = (curproc) ? curproc->p_tid :  (pid_t) -1;
1102 		pg->owner_tag = tag;
1103 		return;
1104 	}
1105 
1106 	/* drop ownership */
1107 	if (pg->owner_tag == NULL) {
1108 		printf("uvm_page_own: dropping ownership of an non-owned "
1109 		    "page (%p)\n", pg);
1110 		panic("uvm_page_own");
1111 	}
1112 	pg->owner_tag = NULL;
1113 	return;
1114 }
1115 #endif
1116 
1117 /*
1118  * when VM_PHYSSEG_MAX is 1, we can simplify these functions
1119  */
1120 
1121 #if VM_PHYSSEG_MAX > 1
1122 /*
1123  * vm_physseg_find: find vm_physseg structure that belongs to a PA
1124  */
1125 int
vm_physseg_find(paddr_t pframe,int * offp)1126 vm_physseg_find(paddr_t pframe, int *offp)
1127 {
1128 	struct vm_physseg *seg;
1129 
1130 #if (VM_PHYSSEG_STRAT == VM_PSTRAT_BSEARCH)
1131 	/* binary search for it */
1132 	int	start, len, try;
1133 
1134 	/*
1135 	 * if try is too large (thus target is less than try) we reduce
1136 	 * the length to trunc(len/2) [i.e. everything smaller than "try"]
1137 	 *
1138 	 * if the try is too small (thus target is greater than try) then
1139 	 * we set the new start to be (try + 1).   this means we need to
1140 	 * reduce the length to (round(len/2) - 1).
1141 	 *
1142 	 * note "adjust" below which takes advantage of the fact that
1143 	 *  (round(len/2) - 1) == trunc((len - 1) / 2)
1144 	 * for any value of len we may have
1145 	 */
1146 
1147 	for (start = 0, len = vm_nphysseg ; len != 0 ; len = len / 2) {
1148 		try = start + (len / 2);	/* try in the middle */
1149 		seg = vm_physmem + try;
1150 
1151 		/* start past our try? */
1152 		if (pframe >= seg->start) {
1153 			/* was try correct? */
1154 			if (pframe < seg->end) {
1155 				if (offp)
1156 					*offp = pframe - seg->start;
1157 				return try;            /* got it */
1158 			}
1159 			start = try + 1;	/* next time, start here */
1160 			len--;			/* "adjust" */
1161 		} else {
1162 			/*
1163 			 * pframe before try, just reduce length of
1164 			 * region, done in "for" loop
1165 			 */
1166 		}
1167 	}
1168 	return -1;
1169 
1170 #else
1171 	/* linear search for it */
1172 	int	lcv;
1173 
1174 	for (lcv = 0, seg = vm_physmem; lcv < vm_nphysseg ; lcv++, seg++) {
1175 		if (pframe >= seg->start && pframe < seg->end) {
1176 			if (offp)
1177 				*offp = pframe - seg->start;
1178 			return lcv;		   /* got it */
1179 		}
1180 	}
1181 	return -1;
1182 
1183 #endif
1184 }
1185 
1186 /*
1187  * PHYS_TO_VM_PAGE: find vm_page for a PA.   used by MI code to get vm_pages
1188  * back from an I/O mapping (ugh!).   used in some MD code as well.
1189  */
1190 struct vm_page *
PHYS_TO_VM_PAGE(paddr_t pa)1191 PHYS_TO_VM_PAGE(paddr_t pa)
1192 {
1193 	paddr_t pf = atop(pa);
1194 	int	off;
1195 	int	psi;
1196 
1197 	psi = vm_physseg_find(pf, &off);
1198 
1199 	return (psi == -1) ? NULL : &vm_physmem[psi].pgs[off];
1200 }
1201 #endif /* VM_PHYSSEG_MAX > 1 */
1202 
1203 /*
1204  * uvm_pagelookup: look up a page
1205  */
1206 struct vm_page *
uvm_pagelookup(struct uvm_object * obj,voff_t off)1207 uvm_pagelookup(struct uvm_object *obj, voff_t off)
1208 {
1209 	/* XXX if stack is too much, handroll */
1210 	struct vm_page p, *pg;
1211 
1212 	p.offset = off;
1213 	pg = RBT_FIND(uvm_objtree, &obj->memt, &p);
1214 
1215 	KASSERT(pg == NULL || obj->uo_npages != 0);
1216 	KASSERT(pg == NULL || (pg->pg_flags & PG_RELEASED) == 0 ||
1217 	    (pg->pg_flags & PG_BUSY) != 0);
1218 	return (pg);
1219 }
1220 
1221 /*
1222  * uvm_pagewire: wire the page, thus removing it from the daemon's grasp
1223  *
1224  * => caller must lock page queues
1225  */
1226 void
uvm_pagewire(struct vm_page * pg)1227 uvm_pagewire(struct vm_page *pg)
1228 {
1229 	KASSERT(uvm_page_owner_locked_p(pg));
1230 	MUTEX_ASSERT_LOCKED(&uvm.pageqlock);
1231 
1232 	if (pg->wire_count == 0) {
1233 		uvm_pagedequeue(pg);
1234 		uvmexp.wired++;
1235 	}
1236 	pg->wire_count++;
1237 }
1238 
1239 /*
1240  * uvm_pageunwire: unwire the page.
1241  *
1242  * => activate if wire count goes to zero.
1243  * => caller must lock page queues
1244  */
1245 void
uvm_pageunwire(struct vm_page * pg)1246 uvm_pageunwire(struct vm_page *pg)
1247 {
1248 	KASSERT(uvm_page_owner_locked_p(pg));
1249 	MUTEX_ASSERT_LOCKED(&uvm.pageqlock);
1250 
1251 	pg->wire_count--;
1252 	if (pg->wire_count == 0) {
1253 		uvm_pageactivate(pg);
1254 		uvmexp.wired--;
1255 	}
1256 }
1257 
1258 /*
1259  * uvm_pagedeactivate: deactivate page -- no pmaps have access to page
1260  *
1261  * => caller must lock page queues
1262  * => caller must check to make sure page is not wired
1263  * => object that page belongs to must be locked (so we can adjust pg->flags)
1264  */
1265 void
uvm_pagedeactivate(struct vm_page * pg)1266 uvm_pagedeactivate(struct vm_page *pg)
1267 {
1268 	KASSERT(uvm_page_owner_locked_p(pg));
1269 	MUTEX_ASSERT_LOCKED(&uvm.pageqlock);
1270 
1271 	if (pg->pg_flags & PQ_ACTIVE) {
1272 		TAILQ_REMOVE(&uvm.page_active, pg, pageq);
1273 		atomic_clearbits_int(&pg->pg_flags, PQ_ACTIVE);
1274 		uvmexp.active--;
1275 	}
1276 	if ((pg->pg_flags & PQ_INACTIVE) == 0) {
1277 		KASSERT(pg->wire_count == 0);
1278 		TAILQ_INSERT_TAIL(&uvm.page_inactive, pg, pageq);
1279 		atomic_setbits_int(&pg->pg_flags, PQ_INACTIVE);
1280 		uvmexp.inactive++;
1281 		pmap_clear_reference(pg);
1282 		/*
1283 		 * update the "clean" bit.  this isn't 100%
1284 		 * accurate, and doesn't have to be.  we'll
1285 		 * re-sync it after we zap all mappings when
1286 		 * scanning the inactive list.
1287 		 */
1288 		if ((pg->pg_flags & PG_CLEAN) != 0 &&
1289 		    pmap_is_modified(pg))
1290 			atomic_clearbits_int(&pg->pg_flags, PG_CLEAN);
1291 	}
1292 }
1293 
1294 /*
1295  * uvm_pageactivate: activate page
1296  *
1297  * => caller must lock page queues
1298  */
1299 void
uvm_pageactivate(struct vm_page * pg)1300 uvm_pageactivate(struct vm_page *pg)
1301 {
1302 	KASSERT(uvm_page_owner_locked_p(pg));
1303 	MUTEX_ASSERT_LOCKED(&uvm.pageqlock);
1304 
1305 	uvm_pagedequeue(pg);
1306 	if (pg->wire_count == 0) {
1307 		TAILQ_INSERT_TAIL(&uvm.page_active, pg, pageq);
1308 		atomic_setbits_int(&pg->pg_flags, PQ_ACTIVE);
1309 		uvmexp.active++;
1310 
1311 	}
1312 }
1313 
1314 /*
1315  * uvm_pagedequeue: remove a page from any paging queue
1316  */
1317 void
uvm_pagedequeue(struct vm_page * pg)1318 uvm_pagedequeue(struct vm_page *pg)
1319 {
1320 	if (pg->pg_flags & PQ_ACTIVE) {
1321 		TAILQ_REMOVE(&uvm.page_active, pg, pageq);
1322 		atomic_clearbits_int(&pg->pg_flags, PQ_ACTIVE);
1323 		uvmexp.active--;
1324 	}
1325 	if (pg->pg_flags & PQ_INACTIVE) {
1326 		TAILQ_REMOVE(&uvm.page_inactive, pg, pageq);
1327 		atomic_clearbits_int(&pg->pg_flags, PQ_INACTIVE);
1328 		uvmexp.inactive--;
1329 	}
1330 }
1331 /*
1332  * uvm_pagezero: zero fill a page
1333  */
1334 void
uvm_pagezero(struct vm_page * pg)1335 uvm_pagezero(struct vm_page *pg)
1336 {
1337 	atomic_clearbits_int(&pg->pg_flags, PG_CLEAN);
1338 	pmap_zero_page(pg);
1339 }
1340 
1341 /*
1342  * uvm_pagecopy: copy a page
1343  */
1344 void
uvm_pagecopy(struct vm_page * src,struct vm_page * dst)1345 uvm_pagecopy(struct vm_page *src, struct vm_page *dst)
1346 {
1347 	atomic_clearbits_int(&dst->pg_flags, PG_CLEAN);
1348 	pmap_copy_page(src, dst);
1349 }
1350 
1351 /*
1352  * uvm_page_owner_locked_p: return true if object associated with page is
1353  * locked.  this is a weak check for runtime assertions only.
1354  */
1355 int
uvm_page_owner_locked_p(struct vm_page * pg)1356 uvm_page_owner_locked_p(struct vm_page *pg)
1357 {
1358 	if (pg->uobject != NULL) {
1359 		if (UVM_OBJ_IS_DUMMY(pg->uobject))
1360 			return 1;
1361 		return rw_write_held(pg->uobject->vmobjlock);
1362 	}
1363 	if (pg->uanon != NULL) {
1364 		return rw_write_held(pg->uanon->an_lock);
1365 	}
1366 	return 1;
1367 }
1368 
1369 /*
1370  * uvm_pagecount: count the number of physical pages in the address range.
1371  */
1372 psize_t
uvm_pagecount(struct uvm_constraint_range * constraint)1373 uvm_pagecount(struct uvm_constraint_range* constraint)
1374 {
1375 	int lcv;
1376 	psize_t sz;
1377 	paddr_t low, high;
1378 	paddr_t ps_low, ps_high;
1379 
1380 	/* Algorithm uses page numbers. */
1381 	low = atop(constraint->ucr_low);
1382 	high = atop(constraint->ucr_high);
1383 
1384 	sz = 0;
1385 	for (lcv = 0; lcv < vm_nphysseg; lcv++) {
1386 		ps_low = MAX(low, vm_physmem[lcv].avail_start);
1387 		ps_high = MIN(high, vm_physmem[lcv].avail_end);
1388 		if (ps_low < ps_high)
1389 			sz += ps_high - ps_low;
1390 	}
1391 	return sz;
1392 }
1393