1 /*	$OpenBSD: uvm_map.h,v 1.29 2003/04/14 04:53:51 art 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 /*
74  * uvm_map.h
75  */
76 
77 #ifdef _KERNEL
78 
79 /*
80  * macros
81  */
82 
83 /*
84  * UVM_MAP_CLIP_START: ensure that the entry begins at or after
85  * the starting address, if it doesn't we split the entry.
86  *
87  * => map must be locked by caller
88  */
89 
90 #define UVM_MAP_CLIP_START(MAP,ENTRY,VA) { \
91 	if ((VA) > (ENTRY)->start) uvm_map_clip_start(MAP,ENTRY,VA); }
92 
93 /*
94  * UVM_MAP_CLIP_END: ensure that the entry ends at or before
95  *      the ending address, if it does't we split the entry.
96  *
97  * => map must be locked by caller
98  */
99 
100 #define UVM_MAP_CLIP_END(MAP,ENTRY,VA) { \
101 	if ((VA) < (ENTRY)->end) uvm_map_clip_end(MAP,ENTRY,VA); }
102 
103 /*
104  * extract flags
105  */
106 #define UVM_EXTRACT_REMOVE	0x1	/* remove mapping from old map */
107 #define UVM_EXTRACT_CONTIG	0x2	/* try to keep it contig */
108 #define UVM_EXTRACT_QREF	0x4	/* use quick refs */
109 #define UVM_EXTRACT_FIXPROT	0x8	/* set prot to maxprot as we go */
110 
111 #endif /* _KERNEL */
112 
113 #include <uvm/uvm_anon.h>
114 
115 /*
116  * types defined:
117  *
118  *	vm_map_t		the high-level address map data structure.
119  *	vm_map_entry_t		an entry in an address map.
120  *	vm_map_version_t	a timestamp of a map, for use with vm_map_lookup
121  */
122 
123 /*
124  * Objects which live in maps may be either VM objects, or another map
125  * (called a "sharing map") which denotes read-write sharing with other maps.
126  *
127  * XXXCDC: private pager data goes here now
128  */
129 
130 union vm_map_object {
131 	struct uvm_object	*uvm_obj;	/* UVM OBJECT */
132 	struct vm_map		*sub_map;	/* belongs to another map */
133 };
134 
135 /*
136  * Address map entries consist of start and end addresses,
137  * a VM object (or sharing map) and offset into that object,
138  * and user-exported inheritance and protection information.
139  * Also included is control information for virtual copy operations.
140  */
141 struct vm_map_entry {
142 	RB_ENTRY(vm_map_entry)	rb_entry;	/* tree information */
143 	vaddr_t			ownspace;	/* free space after */
144 	vaddr_t			space;		/* space in subtree */
145 	struct vm_map_entry	*prev;		/* previous entry */
146 	struct vm_map_entry	*next;		/* next entry */
147 	vaddr_t			start;		/* start address */
148 	vaddr_t			end;		/* end address */
149 	union vm_map_object	object;		/* object I point to */
150 	voff_t			offset;		/* offset into object */
151 	int			etype;		/* entry type */
152 	vm_prot_t		protection;	/* protection code */
153 	vm_prot_t		max_protection;	/* maximum protection */
154 	vm_inherit_t		inheritance;	/* inheritance */
155 	int			wired_count;	/* can be paged if == 0 */
156 	struct vm_aref		aref;		/* anonymous overlay */
157 	int			advice;		/* madvise advice */
158 #define uvm_map_entry_stop_copy flags
159 	u_int8_t		flags;		/* flags */
160 
161 #define UVM_MAP_STATIC		0x01		/* static map entry */
162 #define UVM_MAP_KMEM		0x02		/* from kmem entry pool */
163 
164 };
165 
166 #define	VM_MAPENT_ISWIRED(entry)	((entry)->wired_count != 0)
167 
168 /*
169  *	Maps are doubly-linked lists of map entries, kept sorted
170  *	by address.  A single hint is provided to start
171  *	searches again from the last successful search,
172  *	insertion, or removal.
173  *
174  *	LOCKING PROTOCOL NOTES:
175  *	-----------------------
176  *
177  *	VM map locking is a little complicated.  There are both shared
178  *	and exclusive locks on maps.  However, it is sometimes required
179  *	to downgrade an exclusive lock to a shared lock, and upgrade to
180  *	an exclusive lock again (to perform error recovery).  However,
181  *	another thread *must not* queue itself to receive an exclusive
182  *	lock while before we upgrade back to exclusive, otherwise the
183  *	error recovery becomes extremely difficult, if not impossible.
184  *
185  *	In order to prevent this scenario, we introduce the notion of
186  *	a `busy' map.  A `busy' map is read-locked, but other threads
187  *	attempting to write-lock wait for this flag to clear before
188  *	entering the lock manager.  A map may only be marked busy
189  *	when the map is write-locked (and then the map must be downgraded
190  *	to read-locked), and may only be marked unbusy by the thread
191  *	which marked it busy (holding *either* a read-lock or a
192  *	write-lock, the latter being gained by an upgrade).
193  *
194  *	Access to the map `flags' member is controlled by the `flags_lock'
195  *	simple lock.  Note that some flags are static (set once at map
196  *	creation time, and never changed), and thus require no locking
197  *	to check those flags.  All flags which are r/w must be set or
198  *	cleared while the `flags_lock' is asserted.  Additional locking
199  *	requirements are:
200  *
201  *		VM_MAP_PAGEABLE		r/o static flag; no locking required
202  *
203  *		VM_MAP_INTRSAFE		r/o static flag; no locking required
204  *
205  *		VM_MAP_WIREFUTURE	r/w; may only be set or cleared when
206  *					map is write-locked.  may be tested
207  *					without asserting `flags_lock'.
208  *
209  *		VM_MAP_BUSY		r/w; may only be set when map is
210  *					write-locked, may only be cleared by
211  *					thread which set it, map read-locked
212  *					or write-locked.  must be tested
213  *					while `flags_lock' is asserted.
214  *
215  *		VM_MAP_WANTLOCK		r/w; may only be set when the map
216  *					is busy, and thread is attempting
217  *					to write-lock.  must be tested
218  *					while `flags_lock' is asserted.
219  */
220 struct vm_map {
221 	struct pmap *		pmap;		/* Physical map */
222 	lock_data_t		lock;		/* Lock for map data */
223 	RB_HEAD(uvm_tree, vm_map_entry) rbhead;	/* Tree for entries */
224 	struct vm_map_entry	header;		/* List of entries */
225 	int			nentries;	/* Number of entries */
226 	vsize_t			size;		/* virtual size */
227 	int			ref_count;	/* Reference count */
228 	simple_lock_data_t	ref_lock;	/* Lock for ref_count field */
229 	vm_map_entry_t		hint;		/* hint for quick lookups */
230 	simple_lock_data_t	hint_lock;	/* lock for hint storage */
231 	vm_map_entry_t		first_free;	/* First free space hint */
232 	int			flags;		/* flags */
233 	simple_lock_data_t	flags_lock;	/* Lock for flags field */
234 	unsigned int		timestamp;	/* Version number */
235 #define	min_offset		header.start
236 #define max_offset		header.end
237 };
238 
239 /* vm_map flags */
240 #define	VM_MAP_PAGEABLE		0x01		/* ro: entries are pageable */
241 #define	VM_MAP_INTRSAFE		0x02		/* ro: interrupt safe map */
242 #define	VM_MAP_WIREFUTURE	0x04		/* rw: wire future mappings */
243 #define	VM_MAP_BUSY		0x08		/* rw: map is busy */
244 #define	VM_MAP_WANTLOCK		0x10		/* rw: want to write-lock */
245 
246 /* XXX: number of kernel maps and entries to statically allocate */
247 
248 #if !defined(MAX_KMAPENT)
249 #if (50 + (2 * NPROC) > 1000)
250 #define MAX_KMAPENT (50 + (2 * NPROC))
251 #else
252 #define	MAX_KMAPENT	1000  /* XXXCDC: no crash */
253 #endif
254 #endif	/* !defined MAX_KMAPENT */
255 
256 #ifdef _KERNEL
257 #define	vm_map_modflags(map, set, clear)				\
258 do {									\
259 	simple_lock(&(map)->flags_lock);				\
260 	(map)->flags = ((map)->flags | (set)) & ~(clear);		\
261 	simple_unlock(&(map)->flags_lock);				\
262 } while (0)
263 #endif /* _KERNEL */
264 
265 /*
266  *	Interrupt-safe maps must also be kept on a special list,
267  *	to assist uvm_fault() in avoiding locking problems.
268  */
269 struct vm_map_intrsafe {
270 	struct vm_map	vmi_map;
271 	LIST_ENTRY(vm_map_intrsafe) vmi_list;
272 };
273 
274 LIST_HEAD(vmi_list, vm_map_intrsafe);
275 #ifdef _KERNEL
276 extern simple_lock_data_t vmi_list_slock;
277 extern struct vmi_list vmi_list;
278 
279 static __inline int vmi_list_lock(void);
280 static __inline void vmi_list_unlock(int);
281 
282 static __inline int
vmi_list_lock()283 vmi_list_lock()
284 {
285 	int s;
286 
287 	s = splhigh();
288 	simple_lock(&vmi_list_slock);
289 	return (s);
290 }
291 
292 static __inline void
vmi_list_unlock(s)293 vmi_list_unlock(s)
294 	int s;
295 {
296 
297 	simple_unlock(&vmi_list_slock);
298 	splx(s);
299 }
300 #endif /* _KERNEL */
301 
302 /*
303  * handle inline options
304  */
305 
306 #ifdef UVM_MAP_INLINE
307 #define MAP_INLINE static __inline
308 #else
309 #define MAP_INLINE /* nothing */
310 #endif /* UVM_MAP_INLINE */
311 
312 /*
313  * globals:
314  */
315 
316 #ifdef _KERNEL
317 
318 #ifdef PMAP_GROWKERNEL
319 extern vaddr_t	uvm_maxkaddr;
320 #endif
321 
322 /*
323  * protos: the following prototypes define the interface to vm_map
324  */
325 
326 MAP_INLINE
327 void		uvm_map_deallocate(vm_map_t);
328 
329 int		uvm_map_clean(vm_map_t, vaddr_t, vaddr_t, int);
330 void		uvm_map_clip_start(vm_map_t, vm_map_entry_t, vaddr_t);
331 void		uvm_map_clip_end(vm_map_t, vm_map_entry_t, vaddr_t);
332 MAP_INLINE
333 vm_map_t	uvm_map_create(pmap_t, vaddr_t, vaddr_t, int);
334 int		uvm_map_extract(vm_map_t, vaddr_t, vsize_t,
335 			vm_map_t, vaddr_t *, int);
336 vm_map_entry_t	uvm_map_findspace(vm_map_t, vaddr_t, vsize_t, vaddr_t *,
337 			struct uvm_object *, voff_t, vsize_t, int);
338 vaddr_t		uvm_map_hint(struct proc *, vm_prot_t);
339 int		uvm_map_inherit(vm_map_t, vaddr_t, vaddr_t, vm_inherit_t);
340 int		uvm_map_advice(vm_map_t, vaddr_t, vaddr_t, int);
341 void		uvm_map_init(void);
342 boolean_t	uvm_map_lookup_entry(vm_map_t, vaddr_t, vm_map_entry_t *);
343 MAP_INLINE
344 void		uvm_map_reference(vm_map_t);
345 int		uvm_map_replace(vm_map_t, vaddr_t, vaddr_t,
346 			vm_map_entry_t, int);
347 int		uvm_map_reserve(vm_map_t, vsize_t, vaddr_t, vsize_t,
348 			vaddr_t *);
349 void		uvm_map_setup(vm_map_t, vaddr_t, vaddr_t, int);
350 int		uvm_map_submap(vm_map_t, vaddr_t, vaddr_t, vm_map_t);
351 MAP_INLINE
352 void		uvm_unmap(vm_map_t, vaddr_t, vaddr_t);
353 void		uvm_unmap_detach(vm_map_entry_t,int);
354 void		uvm_unmap_remove(vm_map_t, vaddr_t, vaddr_t,
355 				      vm_map_entry_t *);
356 
357 #endif /* _KERNEL */
358 
359 /*
360  * VM map locking operations:
361  *
362  *	These operations perform locking on the data portion of the
363  *	map.
364  *
365  *	vm_map_lock_try: try to lock a map, failing if it is already locked.
366  *
367  *	vm_map_lock: acquire an exclusive (write) lock on a map.
368  *
369  *	vm_map_lock_read: acquire a shared (read) lock on a map.
370  *
371  *	vm_map_unlock: release an exclusive lock on a map.
372  *
373  *	vm_map_unlock_read: release a shared lock on a map.
374  *
375  *	vm_map_downgrade: downgrade an exclusive lock to a shared lock.
376  *
377  *	vm_map_upgrade: upgrade a shared lock to an exclusive lock.
378  *
379  *	vm_map_busy: mark a map as busy.
380  *
381  *	vm_map_unbusy: clear busy status on a map.
382  *
383  * Note that "intrsafe" maps use only exclusive, spin locks.  We simply
384  * use the sleep lock's interlock for this.
385  */
386 
387 #ifdef _KERNEL
388 /* XXX: clean up later */
389 #include <sys/time.h>
390 #include <sys/proc.h>	/* for tsleep(), wakeup() */
391 #include <sys/systm.h>	/* for panic() */
392 
393 static __inline boolean_t vm_map_lock_try(vm_map_t);
394 static __inline void vm_map_lock(vm_map_t);
395 extern const char vmmapbsy[];
396 
397 static __inline boolean_t
vm_map_lock_try(map)398 vm_map_lock_try(map)
399 	vm_map_t map;
400 {
401 	boolean_t rv;
402 
403 	if (map->flags & VM_MAP_INTRSAFE)
404 		rv = simple_lock_try(&map->lock.lk_interlock);
405 	else {
406 		simple_lock(&map->flags_lock);
407 		if (map->flags & VM_MAP_BUSY) {
408 			simple_unlock(&map->flags_lock);
409 			return (FALSE);
410 		}
411 		rv = (lockmgr(&map->lock, LK_EXCLUSIVE|LK_NOWAIT|LK_INTERLOCK,
412 		    &map->flags_lock, curproc) == 0);
413 	}
414 
415 	if (rv)
416 		map->timestamp++;
417 
418 	return (rv);
419 }
420 
421 static __inline void
vm_map_lock(map)422 vm_map_lock(map)
423 	vm_map_t map;
424 {
425 	int error;
426 
427 	if (map->flags & VM_MAP_INTRSAFE) {
428 		simple_lock(&map->lock.lk_interlock);
429 		return;
430 	}
431 
432  try_again:
433 	simple_lock(&map->flags_lock);
434 	while (map->flags & VM_MAP_BUSY) {
435 		map->flags |= VM_MAP_WANTLOCK;
436 		ltsleep(&map->flags, PVM, (char *)vmmapbsy, 0, &map->flags_lock);
437 	}
438 
439 	error = lockmgr(&map->lock, LK_EXCLUSIVE|LK_SLEEPFAIL|LK_INTERLOCK,
440 	    &map->flags_lock, curproc);
441 
442 	if (error) {
443 		goto try_again;
444 	}
445 
446 	(map)->timestamp++;
447 }
448 
449 #ifdef DIAGNOSTIC
450 #define	vm_map_lock_read(map)						\
451 do {									\
452 	if (map->flags & VM_MAP_INTRSAFE)				\
453 		panic("vm_map_lock_read: intrsafe map");		\
454 	(void) lockmgr(&(map)->lock, LK_SHARED, NULL, curproc);		\
455 } while (0)
456 #else
457 #define	vm_map_lock_read(map)						\
458 	(void) lockmgr(&(map)->lock, LK_SHARED, NULL, curproc)
459 #endif
460 
461 #define	vm_map_unlock(map)						\
462 do {									\
463 	if ((map)->flags & VM_MAP_INTRSAFE)				\
464 		simple_unlock(&(map)->lock.lk_interlock);		\
465 	else								\
466 		(void) lockmgr(&(map)->lock, LK_RELEASE, NULL, curproc);\
467 } while (0)
468 
469 #define	vm_map_unlock_read(map)						\
470 	(void) lockmgr(&(map)->lock, LK_RELEASE, NULL, curproc)
471 
472 #define	vm_map_downgrade(map)						\
473 	(void) lockmgr(&(map)->lock, LK_DOWNGRADE, NULL, curproc)
474 
475 #ifdef DIAGNOSTIC
476 #define	vm_map_upgrade(map)						\
477 do {									\
478 	if (lockmgr(&(map)->lock, LK_UPGRADE, NULL, curproc) != 0)	\
479 		panic("vm_map_upgrade: failed to upgrade lock");	\
480 } while (0)
481 #else
482 #define	vm_map_upgrade(map)						\
483 	(void) lockmgr(&(map)->lock, LK_UPGRADE, NULL, curproc)
484 #endif
485 
486 #define	vm_map_busy(map)						\
487 do {									\
488 	simple_lock(&(map)->flags_lock);				\
489 	(map)->flags |= VM_MAP_BUSY;					\
490 	simple_unlock(&(map)->flags_lock);				\
491 } while (0)
492 
493 #define	vm_map_unbusy(map)						\
494 do {									\
495 	int oflags;							\
496 									\
497 	simple_lock(&(map)->flags_lock);				\
498 	oflags = (map)->flags;						\
499 	(map)->flags &= ~(VM_MAP_BUSY|VM_MAP_WANTLOCK);			\
500 	simple_unlock(&(map)->flags_lock);				\
501 	if (oflags & VM_MAP_WANTLOCK)					\
502 		wakeup(&(map)->flags);					\
503 } while (0)
504 #endif /* _KERNEL */
505 
506 /*
507  *	Functions implemented as macros
508  */
509 #define		vm_map_min(map)		((map)->min_offset)
510 #define		vm_map_max(map)		((map)->max_offset)
511 #define		vm_map_pmap(map)	((map)->pmap)
512 
513 #endif /* _UVM_UVM_MAP_H_ */
514