xref: /openbsd/sys/uvm/uvm_map.h (revision 404b540a)
1 /*	$OpenBSD: uvm_map.h,v 1.42 2009/08/28 00:40:03 ariane Exp $	*/
2 /*	$NetBSD: uvm_map.h,v 1.24 2001/02/18 21:19:08 chs Exp $	*/
3 
4 /*
5  * Copyright (c) 1997 Charles D. Cranor and Washington University.
6  * Copyright (c) 1991, 1993, The Regents of the University of California.
7  *
8  * All rights reserved.
9  *
10  * This code is derived from software contributed to Berkeley by
11  * The Mach Operating System project at Carnegie-Mellon University.
12  *
13  * Redistribution and use in source and binary forms, with or without
14  * modification, are permitted provided that the following conditions
15  * are met:
16  * 1. Redistributions of source code must retain the above copyright
17  *    notice, this list of conditions and the following disclaimer.
18  * 2. Redistributions in binary form must reproduce the above copyright
19  *    notice, this list of conditions and the following disclaimer in the
20  *    documentation and/or other materials provided with the distribution.
21  * 3. All advertising materials mentioning features or use of this software
22  *    must display the following acknowledgement:
23  *	This product includes software developed by Charles D. Cranor,
24  *      Washington University, the University of California, Berkeley and
25  *      its contributors.
26  * 4. Neither the name of the University nor the names of its contributors
27  *    may be used to endorse or promote products derived from this software
28  *    without specific prior written permission.
29  *
30  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
31  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
32  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
33  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
34  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
35  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
36  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
37  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
38  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
39  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
40  * SUCH DAMAGE.
41  *
42  *	@(#)vm_map.h    8.3 (Berkeley) 3/15/94
43  * from: Id: uvm_map.h,v 1.1.2.3 1998/02/07 01:16:55 chs Exp
44  *
45  *
46  * Copyright (c) 1987, 1990 Carnegie-Mellon University.
47  * All rights reserved.
48  *
49  * Permission to use, copy, modify and distribute this software and
50  * its documentation is hereby granted, provided that both the copyright
51  * notice and this permission notice appear in all copies of the
52  * software, derivative works or modified versions, and any portions
53  * thereof, and that both notices appear in supporting documentation.
54  *
55  * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
56  * CONDITION.  CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
57  * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
58  *
59  * Carnegie Mellon requests users of this software to return to
60  *
61  *  Software Distribution Coordinator  or  Software.Distribution@CS.CMU.EDU
62  *  School of Computer Science
63  *  Carnegie Mellon University
64  *  Pittsburgh PA 15213-3890
65  *
66  * any improvements or extensions that they make and grant Carnegie the
67  * rights to redistribute these changes.
68  */
69 
70 #ifndef _UVM_UVM_MAP_H_
71 #define _UVM_UVM_MAP_H_
72 
73 #include <sys/rwlock.h>
74 
75 #ifdef _KERNEL
76 
77 /*
78  * UVM_MAP_CLIP_START: ensure that the entry begins at or after
79  * the starting address, if it doesn't we split the entry.
80  *
81  * => map must be locked by caller
82  */
83 
84 #define UVM_MAP_CLIP_START(MAP,ENTRY,VA) { \
85 	if ((VA) > (ENTRY)->start) uvm_map_clip_start(MAP,ENTRY,VA); }
86 
87 /*
88  * UVM_MAP_CLIP_END: ensure that the entry ends at or before
89  *      the ending address, if it does't we split the entry.
90  *
91  * => map must be locked by caller
92  */
93 
94 #define UVM_MAP_CLIP_END(MAP,ENTRY,VA) { \
95 	if ((VA) < (ENTRY)->end) uvm_map_clip_end(MAP,ENTRY,VA); }
96 
97 /*
98  * extract flags
99  */
100 #define UVM_EXTRACT_REMOVE	0x1	/* remove mapping from old map */
101 #define UVM_EXTRACT_CONTIG	0x2	/* try to keep it contig */
102 #define UVM_EXTRACT_QREF	0x4	/* use quick refs */
103 #define UVM_EXTRACT_FIXPROT	0x8	/* set prot to maxprot as we go */
104 
105 #endif /* _KERNEL */
106 
107 #include <uvm/uvm_anon.h>
108 
109 /*
110  * types defined:
111  *
112  *	vm_map_t		the high-level address map data structure.
113  *	vm_map_entry_t		an entry in an address map.
114  *	vm_map_version_t	a timestamp of a map, for use with vm_map_lookup
115  */
116 
117 /*
118  * Objects which live in maps may be either VM objects, or another map
119  * (called a "sharing map") which denotes read-write sharing with other maps.
120  *
121  * XXXCDC: private pager data goes here now
122  */
123 
124 union vm_map_object {
125 	struct uvm_object	*uvm_obj;	/* UVM OBJECT */
126 	struct vm_map		*sub_map;	/* belongs to another map */
127 };
128 
129 /*
130  * Address map entries consist of start and end addresses,
131  * a VM object (or sharing map) and offset into that object,
132  * and user-exported inheritance and protection information.
133  * Also included is control information for virtual copy operations.
134  */
135 struct vm_map_entry {
136 	RB_ENTRY(vm_map_entry)	rb_entry;	/* tree information */
137 	vaddr_t			ownspace;	/* free space after */
138 	vaddr_t			space;		/* space in subtree */
139 	struct vm_map_entry	*prev;		/* previous entry */
140 	struct vm_map_entry	*next;		/* next entry */
141 	vaddr_t			start;		/* start address */
142 	vaddr_t			end;		/* end address */
143 	union vm_map_object	object;		/* object I point to */
144 	voff_t			offset;		/* offset into object */
145 	int			etype;		/* entry type */
146 	vm_prot_t		protection;	/* protection code */
147 	vm_prot_t		max_protection;	/* maximum protection */
148 	vm_inherit_t		inheritance;	/* inheritance */
149 	int			wired_count;	/* can be paged if == 0 */
150 	struct vm_aref		aref;		/* anonymous overlay */
151 	int			advice;		/* madvise advice */
152 #define uvm_map_entry_stop_copy flags
153 	u_int8_t		flags;		/* flags */
154 
155 #define UVM_MAP_STATIC		0x01		/* static map entry */
156 #define UVM_MAP_KMEM		0x02		/* from kmem entry pool */
157 };
158 
159 /*
160  * Marks the map entry as a guard page, using vm_map_entry.etype.
161  */
162 #define MAP_ET_KVAGUARD		0x10		/* guard entry */
163 
164 #define	VM_MAPENT_ISWIRED(entry)	((entry)->wired_count != 0)
165 
166 /*
167  *	Maps are doubly-linked lists of map entries, kept sorted
168  *	by address.  A single hint is provided to start
169  *	searches again from the last successful search,
170  *	insertion, or removal.
171  *
172  *	LOCKING PROTOCOL NOTES:
173  *	-----------------------
174  *
175  *	VM map locking is a little complicated.  There are both shared
176  *	and exclusive locks on maps.  However, it is sometimes required
177  *	to downgrade an exclusive lock to a shared lock, and upgrade to
178  *	an exclusive lock again (to perform error recovery).  However,
179  *	another thread *must not* queue itself to receive an exclusive
180  *	lock while before we upgrade back to exclusive, otherwise the
181  *	error recovery becomes extremely difficult, if not impossible.
182  *
183  *	In order to prevent this scenario, we introduce the notion of
184  *	a `busy' map.  A `busy' map is read-locked, but other threads
185  *	attempting to write-lock wait for this flag to clear before
186  *	entering the lock manager.  A map may only be marked busy
187  *	when the map is write-locked (and then the map must be downgraded
188  *	to read-locked), and may only be marked unbusy by the thread
189  *	which marked it busy (holding *either* a read-lock or a
190  *	write-lock, the latter being gained by an upgrade).
191  *
192  *	Access to the map `flags' member is controlled by the `flags_lock'
193  *	simple lock.  Note that some flags are static (set once at map
194  *	creation time, and never changed), and thus require no locking
195  *	to check those flags.  All flags which are r/w must be set or
196  *	cleared while the `flags_lock' is asserted.  Additional locking
197  *	requirements are:
198  *
199  *		VM_MAP_PAGEABLE		r/o static flag; no locking required
200  *
201  *		VM_MAP_INTRSAFE		r/o static flag; no locking required
202  *
203  *		VM_MAP_WIREFUTURE	r/w; may only be set or cleared when
204  *					map is write-locked.  may be tested
205  *					without asserting `flags_lock'.
206  *
207  *		VM_MAP_BUSY		r/w; may only be set when map is
208  *					write-locked, may only be cleared by
209  *					thread which set it, map read-locked
210  *					or write-locked.  must be tested
211  *					while `flags_lock' is asserted.
212  *
213  *		VM_MAP_WANTLOCK		r/w; may only be set when the map
214  *					is busy, and thread is attempting
215  *					to write-lock.  must be tested
216  *					while `flags_lock' is asserted.
217  */
218 struct vm_map {
219 	struct pmap *		pmap;		/* Physical map */
220 	struct rwlock		lock;		/* Lock for map data */
221 	RB_HEAD(uvm_tree, vm_map_entry) rbhead;	/* Tree for entries */
222 	struct vm_map_entry	header;		/* List of entries */
223 	int			nentries;	/* Number of entries */
224 	vsize_t			size;		/* virtual size */
225 	int			ref_count;	/* Reference count */
226 	simple_lock_data_t	ref_lock;	/* Lock for ref_count field */
227 	vm_map_entry_t		hint;		/* hint for quick lookups */
228 	simple_lock_data_t	hint_lock;	/* lock for hint storage */
229 	vm_map_entry_t		first_free;	/* First free space hint */
230 	int			flags;		/* flags */
231 	unsigned int		timestamp;	/* Version number */
232 #define	min_offset		header.start
233 #define max_offset		header.end
234 };
235 
236 /* vm_map flags */
237 #define	VM_MAP_PAGEABLE		0x01		/* ro: entries are pageable */
238 #define	VM_MAP_INTRSAFE		0x02		/* ro: interrupt safe map */
239 #define	VM_MAP_WIREFUTURE	0x04		/* rw: wire future mappings */
240 #define	VM_MAP_BUSY		0x08		/* rw: map is busy */
241 #define	VM_MAP_WANTLOCK		0x10		/* rw: want to write-lock */
242 
243 /* XXX: number of kernel maps and entries to statically allocate */
244 
245 #if !defined(MAX_KMAPENT)
246 #define	MAX_KMAPENT	1024  /* XXXCDC: no crash */
247 #endif	/* !defined MAX_KMAPENT */
248 
249 #ifdef _KERNEL
250 #define	vm_map_modflags(map, set, clear)				\
251 do {									\
252 	(map)->flags = ((map)->flags | (set)) & ~(clear);		\
253 } while (0)
254 #endif /* _KERNEL */
255 
256 /*
257  *	Interrupt-safe maps must also be kept on a special list,
258  *	to assist uvm_fault() in avoiding locking problems.
259  */
260 struct vm_map_intrsafe {
261 	struct vm_map	vmi_map;
262 	LIST_ENTRY(vm_map_intrsafe) vmi_list;
263 };
264 
265 /*
266  * globals:
267  */
268 
269 #ifdef _KERNEL
270 
271 #ifdef PMAP_GROWKERNEL
272 extern vaddr_t	uvm_maxkaddr;
273 #endif
274 
275 /*
276  * protos: the following prototypes define the interface to vm_map
277  */
278 
279 void		uvm_map_deallocate(vm_map_t);
280 
281 int		uvm_map_clean(vm_map_t, vaddr_t, vaddr_t, int);
282 void		uvm_map_clip_start(vm_map_t, vm_map_entry_t, vaddr_t);
283 void		uvm_map_clip_end(vm_map_t, vm_map_entry_t, vaddr_t);
284 vm_map_t	uvm_map_create(pmap_t, vaddr_t, vaddr_t, int);
285 int		uvm_map_extract(vm_map_t, vaddr_t, vsize_t,
286 		    vm_map_t, vaddr_t *, int);
287 vm_map_entry_t	uvm_map_findspace(vm_map_t, vaddr_t, vsize_t, vaddr_t *,
288 		    struct uvm_object *, voff_t, vsize_t, int);
289 vaddr_t		uvm_map_pie(vaddr_t);
290 vaddr_t		uvm_map_hint(struct proc *, vm_prot_t);
291 int		uvm_map_inherit(vm_map_t, vaddr_t, vaddr_t, vm_inherit_t);
292 int		uvm_map_advice(vm_map_t, vaddr_t, vaddr_t, int);
293 void		uvm_map_init(void);
294 boolean_t	uvm_map_lookup_entry(vm_map_t, vaddr_t, vm_map_entry_t *);
295 void		uvm_map_reference(vm_map_t);
296 int		uvm_map_replace(vm_map_t, vaddr_t, vaddr_t,
297 		    vm_map_entry_t, int);
298 int		uvm_map_reserve(vm_map_t, vsize_t, vaddr_t, vsize_t,
299 		    vaddr_t *);
300 void		uvm_map_setup(vm_map_t, vaddr_t, vaddr_t, int);
301 int		uvm_map_submap(vm_map_t, vaddr_t, vaddr_t, vm_map_t);
302 #define		uvm_unmap(_m, _s, _e) uvm_unmap_p(_m, _s, _e, 0)
303 void		uvm_unmap_p(vm_map_t, vaddr_t, vaddr_t, struct proc *);
304 void		uvm_unmap_detach(vm_map_entry_t,int);
305 void		uvm_unmap_remove(vm_map_t, vaddr_t, vaddr_t, vm_map_entry_t *,
306 		    struct proc *, boolean_t);
307 
308 #endif /* _KERNEL */
309 
310 /*
311  * VM map locking operations:
312  *
313  *	These operations perform locking on the data portion of the
314  *	map.
315  *
316  *	vm_map_lock_try: try to lock a map, failing if it is already locked.
317  *
318  *	vm_map_lock: acquire an exclusive (write) lock on a map.
319  *
320  *	vm_map_lock_read: acquire a shared (read) lock on a map.
321  *
322  *	vm_map_unlock: release an exclusive lock on a map.
323  *
324  *	vm_map_unlock_read: release a shared lock on a map.
325  *
326  *	vm_map_downgrade: downgrade an exclusive lock to a shared lock.
327  *
328  *	vm_map_upgrade: upgrade a shared lock to an exclusive lock.
329  *
330  *	vm_map_busy: mark a map as busy.
331  *
332  *	vm_map_unbusy: clear busy status on a map.
333  *
334  * Note that "intrsafe" maps use only exclusive, spin locks.  We simply
335  * use the sleep lock's interlock for this.
336  */
337 
338 #ifdef _KERNEL
339 /* XXX: clean up later */
340 #include <sys/time.h>
341 #include <sys/proc.h>	/* for tsleep(), wakeup() */
342 #include <sys/systm.h>	/* for panic() */
343 
344 static __inline boolean_t vm_map_lock_try(vm_map_t);
345 static __inline void vm_map_lock(vm_map_t);
346 extern const char vmmapbsy[];
347 
348 static __inline boolean_t
349 vm_map_lock_try(struct vm_map *map)
350 {
351 	boolean_t rv;
352 
353 	if (map->flags & VM_MAP_INTRSAFE) {
354 		rv = TRUE;
355 	} else {
356 		if (map->flags & VM_MAP_BUSY) {
357 			return (FALSE);
358 		}
359 		rv = (rw_enter(&map->lock, RW_WRITE|RW_NOSLEEP) == 0);
360 	}
361 
362 	if (rv)
363 		map->timestamp++;
364 
365 	return (rv);
366 }
367 
368 static __inline void
369 vm_map_lock(struct vm_map *map)
370 {
371 	if (map->flags & VM_MAP_INTRSAFE)
372 		return;
373 
374 	do {
375 		while (map->flags & VM_MAP_BUSY) {
376 			map->flags |= VM_MAP_WANTLOCK;
377 			tsleep(&map->flags, PVM, (char *)vmmapbsy, 0);
378 		}
379 	} while (rw_enter(&map->lock, RW_WRITE|RW_SLEEPFAIL) != 0);
380 
381 	map->timestamp++;
382 }
383 
384 #define	vm_map_lock_read(map) rw_enter_read(&(map)->lock)
385 
386 #define	vm_map_unlock(map)						\
387 do {									\
388 	if (((map)->flags & VM_MAP_INTRSAFE) == 0)			\
389 		rw_exit(&(map)->lock);					\
390 } while (0)
391 
392 #define	vm_map_unlock_read(map)	rw_exit_read(&(map)->lock)
393 
394 #define	vm_map_downgrade(map) rw_enter(&(map)->lock, RW_DOWNGRADE)
395 
396 #define	vm_map_upgrade(map)						\
397 do {									\
398 	rw_exit_read(&(map)->lock);					\
399 	rw_enter_write(&(map)->lock);					\
400 } while (0)
401 
402 #define	vm_map_busy(map)						\
403 do {									\
404 	(map)->flags |= VM_MAP_BUSY;					\
405 } while (0)
406 
407 #define	vm_map_unbusy(map)						\
408 do {									\
409 	int oflags;							\
410 									\
411 	oflags = (map)->flags;						\
412 	(map)->flags &= ~(VM_MAP_BUSY|VM_MAP_WANTLOCK);			\
413 	if (oflags & VM_MAP_WANTLOCK)					\
414 		wakeup(&(map)->flags);					\
415 } while (0)
416 #endif /* _KERNEL */
417 
418 /*
419  *	Functions implemented as macros
420  */
421 #define		vm_map_min(map)		((map)->min_offset)
422 #define		vm_map_max(map)		((map)->max_offset)
423 #define		vm_map_pmap(map)	((map)->pmap)
424 
425 #endif /* _UVM_UVM_MAP_H_ */
426