xref: /netbsd-src/sys/uvm/uvm_page.h (revision 1ad9454efb13a65cd7535ccf867508cb14d9d30e)
1 /*	$NetBSD: uvm_page.h,v 1.46 2006/09/15 15:51:13 yamt 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. All advertising materials mentioning features or use of this software
21  *    must display the following acknowledgement:
22  *	This product includes software developed by Charles D. Cranor,
23  *      Washington University, the University of California, Berkeley and
24  *      its contributors.
25  * 4. Neither the name of the University nor the names of its contributors
26  *    may be used to endorse or promote products derived from this software
27  *    without specific prior written permission.
28  *
29  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
30  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
31  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
32  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
33  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
34  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
35  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
36  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
37  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
38  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
39  * SUCH DAMAGE.
40  *
41  *	@(#)vm_page.h   7.3 (Berkeley) 4/21/91
42  * from: Id: uvm_page.h,v 1.1.2.6 1998/02/04 02:31:42 chuck Exp
43  *
44  *
45  * Copyright (c) 1987, 1990 Carnegie-Mellon University.
46  * All rights reserved.
47  *
48  * Permission to use, copy, modify and distribute this software and
49  * its documentation is hereby granted, provided that both the copyright
50  * notice and this permission notice appear in all copies of the
51  * software, derivative works or modified versions, and any portions
52  * thereof, and that both notices appear in supporting documentation.
53  *
54  * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
55  * CONDITION.  CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
56  * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
57  *
58  * Carnegie Mellon requests users of this software to return to
59  *
60  *  Software Distribution Coordinator  or  Software.Distribution@CS.CMU.EDU
61  *  School of Computer Science
62  *  Carnegie Mellon University
63  *  Pittsburgh PA 15213-3890
64  *
65  * any improvements or extensions that they make and grant Carnegie the
66  * rights to redistribute these changes.
67  */
68 
69 #ifndef _UVM_UVM_PAGE_H_
70 #define _UVM_UVM_PAGE_H_
71 
72 /*
73  * uvm_page.h
74  */
75 
76 /*
77  *	Resident memory system definitions.
78  */
79 
80 /*
81  *	Management of resident (logical) pages.
82  *
83  *	A small structure is kept for each resident
84  *	page, indexed by page number.  Each structure
85  *	is an element of several lists:
86  *
87  *		A hash table bucket used to quickly
88  *		perform object/offset lookups
89  *
90  *		A list of all pages for a given object,
91  *		so they can be quickly deactivated at
92  *		time of deallocation.
93  *
94  *		An ordered list of pages due for pageout.
95  *
96  *	In addition, the structure contains the object
97  *	and offset to which this page belongs (for pageout),
98  *	and sundry status bits.
99  *
100  *	Fields in this structure are locked either by the lock on the
101  *	object that the page belongs to (O) or by the lock on the page
102  *	queues (P) [or both].
103  */
104 
105 /*
106  * locking note: the mach version of this data structure had bit
107  * fields for the flags, and the bit fields were divided into two
108  * items (depending on who locked what).  some time, in BSD, the bit
109  * fields were dumped and all the flags were lumped into one short.
110  * that is fine for a single threaded uniprocessor OS, but bad if you
111  * want to actual make use of locking (simple_lock's).  so, we've
112  * separated things back out again.
113  *
114  * note the page structure has no lock of its own.
115  */
116 
117 #include <uvm/uvm_extern.h>
118 #include <uvm/uvm_pglist.h>
119 
120 struct vm_page {
121 	TAILQ_ENTRY(vm_page)	pageq;		/* queue info for FIFO
122 						 * queue or free list (P) */
123 	TAILQ_ENTRY(vm_page)	hashq;		/* hash table links (O)*/
124 	TAILQ_ENTRY(vm_page)	listq;		/* pages in same object (O)*/
125 
126 	struct vm_anon		*uanon;		/* anon (O,P) */
127 	struct uvm_object	*uobject;	/* object (O,P) */
128 	voff_t			offset;		/* offset into object (O,P) */
129 	uint16_t		flags;		/* object flags [O] */
130 	uint16_t		loan_count;	/* number of active loans
131 						 * to read: [O or P]
132 						 * to modify: [O _and_ P] */
133 	uint16_t		wire_count;	/* wired down map refs [P] */
134 	uint16_t		pqflags;	/* page queue flags [P] */
135 	paddr_t			phys_addr;	/* physical address of page */
136 
137 #ifdef __HAVE_VM_PAGE_MD
138 	struct vm_page_md	mdpage;		/* pmap-specific data */
139 #endif
140 
141 #if defined(UVM_PAGE_TRKOWN)
142 	/* debugging fields to track page ownership */
143 	pid_t			owner;		/* proc that set PG_BUSY */
144 	const char		*owner_tag;	/* why it was set busy */
145 #endif
146 };
147 
148 /*
149  * These are the flags defined for vm_page.
150  */
151 
152 /*
153  * locking rules:
154  *   PG_ ==> locked by object lock
155  *   PQ_ ==> lock by page queue lock
156  *   PQ_FREE is locked by free queue lock and is mutex with all other PQs
157  *
158  * PG_ZERO is used to indicate that a page has been pre-zero'd.  This flag
159  * is only set when the page is on no queues, and is cleared when the page
160  * is placed on the free list.
161  */
162 
163 #define	PG_BUSY		0x0001		/* page is locked */
164 #define	PG_WANTED	0x0002		/* someone is waiting for page */
165 #define	PG_TABLED	0x0004		/* page is in VP table  */
166 #define	PG_CLEAN	0x0008		/* page has not been modified */
167 #define	PG_PAGEOUT	0x0010		/* page to be freed for pagedaemon */
168 #define PG_RELEASED	0x0020		/* page to be freed when unbusied */
169 #define	PG_FAKE		0x0040		/* page is not yet initialized */
170 #define	PG_RDONLY	0x0080		/* page must be mapped read-only */
171 #define	PG_ZERO		0x0100		/* page is pre-zero'd */
172 
173 #define PG_PAGER1	0x1000		/* pager-specific flag */
174 
175 #define	UVM_PGFLAGBITS \
176 	"\20\1BUSY\2WANTED\3TABLED\4CLEAN\5PAGEOUT\6RELEASED\7FAKE\10RDONLY" \
177 	"\11ZERO\15PAGER1"
178 
179 #define PQ_FREE		0x0001		/* page is on free list */
180 #define PQ_ANON		0x0002		/* page is part of an anon, rather
181 					   than an uvm_object */
182 #define PQ_AOBJ		0x0004		/* page is part of an anonymous
183 					   uvm_object */
184 #define PQ_SWAPBACKED	(PQ_ANON|PQ_AOBJ)
185 #define PQ_READAHEAD	0x0008	/* read-ahead but has not been "hit" yet */
186 
187 #define PQ_PRIVATE1	0x0100
188 #define PQ_PRIVATE2	0x0200
189 #define PQ_PRIVATE3	0x0400
190 #define PQ_PRIVATE4	0x0800
191 #define PQ_PRIVATE5	0x1000
192 #define PQ_PRIVATE6	0x2000
193 #define PQ_PRIVATE7	0x4000
194 #define PQ_PRIVATE8	0x8000
195 
196 #define	UVM_PQFLAGBITS \
197 	"\20\1FREE\2ANON\3AOBJ\4READAHEAD" \
198 	"\11PRIVATE1\12PRIVATE2\13PRIVATE3\14PRIVATE4" \
199 	"\15PRIVATE5\16PRIVATE6\17PRIVATE7\20PRIVATE8"
200 
201 /*
202  * physical memory layout structure
203  *
204  * MD vmparam.h must #define:
205  *   VM_PHYSEG_MAX = max number of physical memory segments we support
206  *		   (if this is "1" then we revert to a "contig" case)
207  *   VM_PHYSSEG_STRAT: memory sort/search options (for VM_PHYSEG_MAX > 1)
208  * 	- VM_PSTRAT_RANDOM:   linear search (random order)
209  *	- VM_PSTRAT_BSEARCH:  binary search (sorted by address)
210  *	- VM_PSTRAT_BIGFIRST: linear search (sorted by largest segment first)
211  *      - others?
212  *   XXXCDC: eventually we should purge all left-over global variables...
213  */
214 #define VM_PSTRAT_RANDOM	1
215 #define VM_PSTRAT_BSEARCH	2
216 #define VM_PSTRAT_BIGFIRST	3
217 
218 /*
219  * vm_physseg: describes one segment of physical memory
220  */
221 struct vm_physseg {
222 	paddr_t	start;			/* PF# of first page in segment */
223 	paddr_t	end;			/* (PF# of last page in segment) + 1 */
224 	paddr_t	avail_start;		/* PF# of first free page in segment */
225 	paddr_t	avail_end;		/* (PF# of last free page in segment) +1  */
226 	int	free_list;		/* which free list they belong on */
227 	struct	vm_page *pgs;		/* vm_page structures (from start) */
228 	struct	vm_page *lastpg;	/* vm_page structure for end */
229 #ifdef __HAVE_PMAP_PHYSSEG
230 	struct	pmap_physseg pmseg;	/* pmap specific (MD) data */
231 #endif
232 };
233 
234 #ifdef _KERNEL
235 
236 /*
237  * globals
238  */
239 
240 extern boolean_t vm_page_zero_enable;
241 
242 /*
243  * physical memory config is stored in vm_physmem.
244  */
245 
246 extern struct vm_physseg vm_physmem[VM_PHYSSEG_MAX];
247 extern int vm_nphysseg;
248 
249 /*
250  * prototypes: the following prototypes define the interface to pages
251  */
252 
253 void uvm_page_init(vaddr_t *, vaddr_t *);
254 #if defined(UVM_PAGE_TRKOWN)
255 void uvm_page_own(struct vm_page *, const char *);
256 #endif
257 #if !defined(PMAP_STEAL_MEMORY)
258 boolean_t uvm_page_physget(paddr_t *);
259 #endif
260 void uvm_page_rehash(void);
261 void uvm_page_recolor(int);
262 void uvm_pageidlezero(void);
263 
264 int uvm_lock_fpageq(void);
265 void uvm_unlock_fpageq(int);
266 
267 void uvm_pageactivate(struct vm_page *);
268 vaddr_t uvm_pageboot_alloc(vsize_t);
269 void uvm_pagecopy(struct vm_page *, struct vm_page *);
270 void uvm_pagedeactivate(struct vm_page *);
271 void uvm_pagedequeue(struct vm_page *);
272 void uvm_pageenqueue(struct vm_page *);
273 void uvm_pagefree(struct vm_page *);
274 void uvm_page_unbusy(struct vm_page **, int);
275 struct vm_page *uvm_pagelookup(struct uvm_object *, voff_t);
276 void uvm_pageunwire(struct vm_page *);
277 void uvm_pagewait(struct vm_page *, int);
278 void uvm_pagewake(struct vm_page *);
279 void uvm_pagewire(struct vm_page *);
280 void uvm_pagezero(struct vm_page *);
281 
282 int uvm_page_lookup_freelist(struct vm_page *);
283 
284 static struct vm_page *PHYS_TO_VM_PAGE(paddr_t);
285 static int vm_physseg_find(paddr_t, int *);
286 
287 /*
288  * macros
289  */
290 
291 #define UVM_PAGE_HASH_PENALTY	4	/* XXX: a guess */
292 
293 #define uvm_lock_pageq()	simple_lock(&uvm.pageqlock)
294 #define uvm_unlock_pageq()	simple_unlock(&uvm.pageqlock)
295 #define	UVM_LOCK_ASSERT_PAGEQ()	LOCK_ASSERT(simple_lock_held(&uvm.pageqlock))
296 
297 #define uvm_pagehash(obj,off) \
298 	(((unsigned long)obj+(unsigned long)atop(off)) & uvm.page_hashmask)
299 
300 #define	UVM_PAGEZERO_TARGET	(uvmexp.free)
301 
302 #define VM_PAGE_TO_PHYS(entry)	((entry)->phys_addr)
303 
304 /*
305  * Compute the page color bucket for a given page.
306  */
307 #define	VM_PGCOLOR_BUCKET(pg) \
308 	(atop(VM_PAGE_TO_PHYS((pg))) & uvmexp.colormask)
309 
310 /*
311  * when VM_PHYSSEG_MAX is 1, we can simplify these functions
312  */
313 
314 /*
315  * vm_physseg_find: find vm_physseg structure that belongs to a PA
316  */
317 static __inline int
318 vm_physseg_find(pframe, offp)
319 	paddr_t pframe;
320 	int	*offp;
321 {
322 #if VM_PHYSSEG_MAX == 1
323 
324 	/* 'contig' case */
325 	if (pframe >= vm_physmem[0].start && pframe < vm_physmem[0].end) {
326 		if (offp)
327 			*offp = pframe - vm_physmem[0].start;
328 		return(0);
329 	}
330 	return(-1);
331 
332 #elif (VM_PHYSSEG_STRAT == VM_PSTRAT_BSEARCH)
333 	/* binary search for it */
334 	u_int	start, len, try;
335 
336 	/*
337 	 * if try is too large (thus target is less than try) we reduce
338 	 * the length to trunc(len/2) [i.e. everything smaller than "try"]
339 	 *
340 	 * if the try is too small (thus target is greater than try) then
341 	 * we set the new start to be (try + 1).   this means we need to
342 	 * reduce the length to (round(len/2) - 1).
343 	 *
344 	 * note "adjust" below which takes advantage of the fact that
345 	 *  (round(len/2) - 1) == trunc((len - 1) / 2)
346 	 * for any value of len we may have
347 	 */
348 
349 	for (start = 0, len = vm_nphysseg ; len != 0 ; len = len / 2) {
350 		try = start + (len / 2);	/* try in the middle */
351 
352 		/* start past our try? */
353 		if (pframe >= vm_physmem[try].start) {
354 			/* was try correct? */
355 			if (pframe < vm_physmem[try].end) {
356 				if (offp)
357 					*offp = pframe - vm_physmem[try].start;
358 				return(try);            /* got it */
359 			}
360 			start = try + 1;	/* next time, start here */
361 			len--;			/* "adjust" */
362 		} else {
363 			/*
364 			 * pframe before try, just reduce length of
365 			 * region, done in "for" loop
366 			 */
367 		}
368 	}
369 	return(-1);
370 
371 #else
372 	/* linear search for it */
373 	int	lcv;
374 
375 	for (lcv = 0; lcv < vm_nphysseg; lcv++) {
376 		if (pframe >= vm_physmem[lcv].start &&
377 		    pframe < vm_physmem[lcv].end) {
378 			if (offp)
379 				*offp = pframe - vm_physmem[lcv].start;
380 			return(lcv);		   /* got it */
381 		}
382 	}
383 	return(-1);
384 
385 #endif
386 }
387 
388 
389 /*
390  * IS_VM_PHYSADDR: only used my mips/pmax/pica trap/pmap.
391  */
392 
393 #define IS_VM_PHYSADDR(PA) (vm_physseg_find(atop(PA), NULL) != -1)
394 
395 /*
396  * PHYS_TO_VM_PAGE: find vm_page for a PA.   used by MI code to get vm_pages
397  * back from an I/O mapping (ugh!).   used in some MD code as well.
398  */
399 static __inline struct vm_page *
400 PHYS_TO_VM_PAGE(pa)
401 	paddr_t pa;
402 {
403 	paddr_t pf = atop(pa);
404 	int	off;
405 	int	psi;
406 
407 	psi = vm_physseg_find(pf, &off);
408 	if (psi != -1)
409 		return(&vm_physmem[psi].pgs[off]);
410 	return(NULL);
411 }
412 
413 #define VM_PAGE_IS_FREE(entry)  ((entry)->pqflags & PQ_FREE)
414 
415 #ifdef DEBUG
416 void uvm_pagezerocheck(struct vm_page *);
417 #endif /* DEBUG */
418 
419 #endif /* _KERNEL */
420 
421 #endif /* _UVM_UVM_PAGE_H_ */
422