1 /* $NetBSD: uvm_map.h,v 1.39 2004/02/10 01:30:49 matt 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_map.h 8.3 (Berkeley) 3/15/94 42 * from: Id: uvm_map.h,v 1.1.2.3 1998/02/07 01:16:55 chs 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_MAP_H_ 70 #define _UVM_UVM_MAP_H_ 71 72 /* 73 * uvm_map.h 74 */ 75 76 #ifdef _KERNEL 77 78 /* 79 * macros 80 */ 81 82 /* 83 * UVM_MAP_CLIP_START: ensure that the entry begins at or after 84 * the starting address, if it doesn't we split the entry. 85 * 86 * => map must be locked by caller 87 */ 88 89 #define UVM_MAP_CLIP_START(MAP,ENTRY,VA) { \ 90 if ((VA) > (ENTRY)->start) uvm_map_clip_start(MAP,ENTRY,VA); } 91 92 /* 93 * UVM_MAP_CLIP_END: ensure that the entry ends at or before 94 * the ending address, if it does't we split the entry. 95 * 96 * => map must be locked by caller 97 */ 98 99 #define UVM_MAP_CLIP_END(MAP,ENTRY,VA) { \ 100 if ((VA) < (ENTRY)->end) uvm_map_clip_end(MAP,ENTRY,VA); } 101 102 /* 103 * extract flags 104 */ 105 #define UVM_EXTRACT_REMOVE 0x1 /* remove mapping from old map */ 106 #define UVM_EXTRACT_CONTIG 0x2 /* try to keep it contig */ 107 #define UVM_EXTRACT_QREF 0x4 /* use quick refs */ 108 #define UVM_EXTRACT_FIXPROT 0x8 /* set prot to maxprot as we go */ 109 110 #endif /* _KERNEL */ 111 112 #include <sys/tree.h> 113 114 #include <uvm/uvm_anon.h> 115 116 /* 117 * Address map entries consist of start and end addresses, 118 * a VM object (or sharing map) and offset into that object, 119 * and user-exported inheritance and protection information. 120 * Also included is control information for virtual copy operations. 121 */ 122 struct vm_map_entry { 123 RB_ENTRY(vm_map_entry) rb_entry; /* tree information */ 124 vaddr_t ownspace; /* free space after */ 125 vaddr_t space; /* space in subtree */ 126 struct vm_map_entry *prev; /* previous entry */ 127 struct vm_map_entry *next; /* next entry */ 128 vaddr_t start; /* start address */ 129 vaddr_t end; /* end address */ 130 union { 131 struct uvm_object *uvm_obj; /* uvm object */ 132 struct vm_map *sub_map; /* belongs to another map */ 133 } object; /* object I point to */ 134 voff_t offset; /* offset into object */ 135 int etype; /* entry type */ 136 vm_prot_t protection; /* protection code */ 137 vm_prot_t max_protection; /* maximum protection */ 138 vm_inherit_t inheritance; /* inheritance */ 139 int wired_count; /* can be paged if == 0 */ 140 struct vm_aref aref; /* anonymous overlay */ 141 int advice; /* madvise advice */ 142 #define uvm_map_entry_stop_copy flags 143 u_int8_t flags; /* flags */ 144 145 #define UVM_MAP_STATIC 0x01 /* static map entry */ 146 #define UVM_MAP_KMEM 0x02 /* from kmem entry pool */ 147 #define UVM_MAP_NOMERGE 0x10 /* this entry is not mergable */ 148 149 }; 150 151 #define VM_MAPENT_ISWIRED(entry) ((entry)->wired_count != 0) 152 153 /* 154 * Maps are doubly-linked lists of map entries, kept sorted 155 * by address. A single hint is provided to start 156 * searches again from the last successful search, 157 * insertion, or removal. 158 * 159 * LOCKING PROTOCOL NOTES: 160 * ----------------------- 161 * 162 * VM map locking is a little complicated. There are both shared 163 * and exclusive locks on maps. However, it is sometimes required 164 * to downgrade an exclusive lock to a shared lock, and upgrade to 165 * an exclusive lock again (to perform error recovery). However, 166 * another thread *must not* queue itself to receive an exclusive 167 * lock while before we upgrade back to exclusive, otherwise the 168 * error recovery becomes extremely difficult, if not impossible. 169 * 170 * In order to prevent this scenario, we introduce the notion of 171 * a `busy' map. A `busy' map is read-locked, but other threads 172 * attempting to write-lock wait for this flag to clear before 173 * entering the lock manager. A map may only be marked busy 174 * when the map is write-locked (and then the map must be downgraded 175 * to read-locked), and may only be marked unbusy by the thread 176 * which marked it busy (holding *either* a read-lock or a 177 * write-lock, the latter being gained by an upgrade). 178 * 179 * Access to the map `flags' member is controlled by the `flags_lock' 180 * simple lock. Note that some flags are static (set once at map 181 * creation time, and never changed), and thus require no locking 182 * to check those flags. All flags which are r/w must be set or 183 * cleared while the `flags_lock' is asserted. Additional locking 184 * requirements are: 185 * 186 * VM_MAP_PAGEABLE r/o static flag; no locking required 187 * 188 * VM_MAP_INTRSAFE r/o static flag; no locking required 189 * 190 * VM_MAP_WIREFUTURE r/w; may only be set or cleared when 191 * map is write-locked. may be tested 192 * without asserting `flags_lock'. 193 * 194 * VM_MAP_BUSY r/w; may only be set when map is 195 * write-locked, may only be cleared by 196 * thread which set it, map read-locked 197 * or write-locked. must be tested 198 * while `flags_lock' is asserted. 199 * 200 * VM_MAP_WANTLOCK r/w; may only be set when the map 201 * is busy, and thread is attempting 202 * to write-lock. must be tested 203 * while `flags_lock' is asserted. 204 * 205 * VM_MAP_DYING r/o; set when a vmspace is being 206 * destroyed to indicate that updates 207 * to the pmap can be skipped. 208 * 209 * VM_MAP_TOPDOWN r/o; set when the vmspace is 210 * created if the unspecified map 211 * allocations are to be arranged in 212 * a "top down" manner. 213 */ 214 struct vm_map { 215 struct pmap * pmap; /* Physical map */ 216 struct lock lock; /* Lock for map data */ 217 RB_HEAD(uvm_tree, vm_map_entry) rbhead; /* Tree for entries */ 218 struct vm_map_entry header; /* List of entries */ 219 int nentries; /* Number of entries */ 220 vsize_t size; /* virtual size */ 221 int ref_count; /* Reference count */ 222 struct simplelock ref_lock; /* Lock for ref_count field */ 223 struct vm_map_entry * hint; /* hint for quick lookups */ 224 struct simplelock hint_lock; /* lock for hint storage */ 225 struct vm_map_entry * first_free; /* First free space hint */ 226 int flags; /* flags */ 227 struct simplelock flags_lock; /* Lock for flags field */ 228 unsigned int timestamp; /* Version number */ 229 #define min_offset header.end 230 #define max_offset header.start 231 }; 232 233 /* vm_map flags */ 234 #define VM_MAP_PAGEABLE 0x01 /* ro: entries are pageable */ 235 #define VM_MAP_INTRSAFE 0x02 /* ro: interrupt safe map */ 236 #define VM_MAP_WIREFUTURE 0x04 /* rw: wire future mappings */ 237 #define VM_MAP_BUSY 0x08 /* rw: map is busy */ 238 #define VM_MAP_WANTLOCK 0x10 /* rw: want to write-lock */ 239 #define VM_MAP_DYING 0x20 /* rw: map is being destroyed */ 240 #define VM_MAP_TOPDOWN 0x40 /* ro: arrange map top-down */ 241 242 /* XXX: number of kernel maps and entries to statically allocate */ 243 244 #if !defined(MAX_KMAPENT) 245 #if (50 + (2 * NPROC) > 1000) 246 #define MAX_KMAPENT (50 + (2 * NPROC)) 247 #else 248 #define MAX_KMAPENT 1000 /* XXXCDC: no crash */ 249 #endif 250 #endif /* !defined MAX_KMAPENT */ 251 252 #ifdef _KERNEL 253 #define vm_map_modflags(map, set, clear) \ 254 do { \ 255 simple_lock(&(map)->flags_lock); \ 256 (map)->flags = ((map)->flags | (set)) & ~(clear); \ 257 simple_unlock(&(map)->flags_lock); \ 258 } while (/*CONSTCOND*/ 0) 259 #endif /* _KERNEL */ 260 261 /* 262 * handle inline options 263 */ 264 265 #ifdef UVM_MAP_INLINE 266 #define MAP_INLINE static __inline 267 #else 268 #define MAP_INLINE /* nothing */ 269 #endif /* UVM_MAP_INLINE */ 270 271 /* 272 * globals: 273 */ 274 275 #ifdef _KERNEL 276 277 #ifdef PMAP_GROWKERNEL 278 extern vaddr_t uvm_maxkaddr; 279 #endif 280 281 /* 282 * protos: the following prototypes define the interface to vm_map 283 */ 284 285 MAP_INLINE 286 void uvm_map_deallocate(struct vm_map *); 287 288 int uvm_map_clean(struct vm_map *, vaddr_t, vaddr_t, int); 289 void uvm_map_clip_start(struct vm_map *, struct vm_map_entry *, 290 vaddr_t); 291 void uvm_map_clip_end(struct vm_map *, struct vm_map_entry *, 292 vaddr_t); 293 MAP_INLINE 294 struct vm_map *uvm_map_create(pmap_t, vaddr_t, vaddr_t, int); 295 int uvm_map_extract(struct vm_map *, vaddr_t, vsize_t, 296 struct vm_map *, vaddr_t *, int); 297 struct vm_map_entry * 298 uvm_map_findspace(struct vm_map *, vaddr_t, vsize_t, 299 vaddr_t *, struct uvm_object *, voff_t, vsize_t, int); 300 int uvm_map_inherit(struct vm_map *, vaddr_t, vaddr_t, 301 vm_inherit_t); 302 int uvm_map_advice(struct vm_map *, vaddr_t, vaddr_t, int); 303 void uvm_map_init(void); 304 boolean_t uvm_map_lookup_entry(struct vm_map *, vaddr_t, 305 struct vm_map_entry **); 306 MAP_INLINE 307 void uvm_map_reference(struct vm_map *); 308 int uvm_map_replace(struct vm_map *, vaddr_t, vaddr_t, 309 struct vm_map_entry *, int); 310 int uvm_map_reserve(struct vm_map *, vsize_t, vaddr_t, vsize_t, 311 vaddr_t *); 312 void uvm_map_setup(struct vm_map *, vaddr_t, vaddr_t, int); 313 int uvm_map_submap(struct vm_map *, vaddr_t, vaddr_t, 314 struct vm_map *); 315 MAP_INLINE 316 void uvm_unmap(struct vm_map *, vaddr_t, vaddr_t); 317 void uvm_unmap_detach(struct vm_map_entry *,int); 318 void uvm_unmap_remove(struct vm_map *, vaddr_t, vaddr_t, 319 struct vm_map_entry **); 320 321 #endif /* _KERNEL */ 322 323 /* 324 * VM map locking operations: 325 * 326 * These operations perform locking on the data portion of the 327 * map. 328 * 329 * vm_map_lock_try: try to lock a map, failing if it is already locked. 330 * 331 * vm_map_lock: acquire an exclusive (write) lock on a map. 332 * 333 * vm_map_lock_read: acquire a shared (read) lock on a map. 334 * 335 * vm_map_unlock: release an exclusive lock on a map. 336 * 337 * vm_map_unlock_read: release a shared lock on a map. 338 * 339 * vm_map_downgrade: downgrade an exclusive lock to a shared lock. 340 * 341 * vm_map_upgrade: upgrade a shared lock to an exclusive lock. 342 * 343 * vm_map_busy: mark a map as busy. 344 * 345 * vm_map_unbusy: clear busy status on a map. 346 * 347 * Note that "intrsafe" maps use only exclusive, spin locks. We simply 348 * use the sleep lock's interlock for this. 349 */ 350 351 #ifdef _KERNEL 352 /* XXX: clean up later */ 353 #include <sys/time.h> 354 #include <sys/proc.h> /* for tsleep(), wakeup() */ 355 #include <sys/systm.h> /* for panic() */ 356 357 static __inline boolean_t vm_map_lock_try(struct vm_map *); 358 static __inline void vm_map_lock(struct vm_map *); 359 extern const char vmmapbsy[]; 360 361 static __inline boolean_t 362 vm_map_lock_try(struct vm_map *map) 363 { 364 boolean_t rv; 365 366 if (map->flags & VM_MAP_INTRSAFE) 367 rv = simple_lock_try(&map->lock.lk_interlock); 368 else { 369 simple_lock(&map->flags_lock); 370 if (map->flags & VM_MAP_BUSY) { 371 simple_unlock(&map->flags_lock); 372 return (FALSE); 373 } 374 rv = (lockmgr(&map->lock, LK_EXCLUSIVE|LK_NOWAIT|LK_INTERLOCK, 375 &map->flags_lock) == 0); 376 } 377 378 if (rv) 379 map->timestamp++; 380 381 return (rv); 382 } 383 384 static __inline void 385 vm_map_lock(struct vm_map *map) 386 { 387 int error; 388 389 if (map->flags & VM_MAP_INTRSAFE) { 390 simple_lock(&map->lock.lk_interlock); 391 return; 392 } 393 394 try_again: 395 simple_lock(&map->flags_lock); 396 while (map->flags & VM_MAP_BUSY) { 397 map->flags |= VM_MAP_WANTLOCK; 398 ltsleep(&map->flags, PVM, vmmapbsy, 0, &map->flags_lock); 399 } 400 401 error = lockmgr(&map->lock, LK_EXCLUSIVE|LK_SLEEPFAIL|LK_INTERLOCK, 402 &map->flags_lock); 403 404 if (error) { 405 KASSERT(error == ENOLCK); 406 goto try_again; 407 } 408 409 (map)->timestamp++; 410 } 411 412 #ifdef DIAGNOSTIC 413 #define vm_map_lock_read(map) \ 414 do { \ 415 if ((map)->flags & VM_MAP_INTRSAFE) \ 416 panic("vm_map_lock_read: intrsafe Map"); \ 417 (void) lockmgr(&(map)->lock, LK_SHARED, NULL); \ 418 } while (/*CONSTCOND*/ 0) 419 #else 420 #define vm_map_lock_read(map) \ 421 (void) lockmgr(&(map)->lock, LK_SHARED, NULL) 422 #endif 423 424 #define vm_map_unlock(map) \ 425 do { \ 426 if ((map)->flags & VM_MAP_INTRSAFE) \ 427 simple_unlock(&(map)->lock.lk_interlock); \ 428 else \ 429 (void) lockmgr(&(map)->lock, LK_RELEASE, NULL); \ 430 } while (/*CONSTCOND*/ 0) 431 432 #define vm_map_unlock_read(map) \ 433 (void) lockmgr(&(map)->lock, LK_RELEASE, NULL) 434 435 #define vm_map_downgrade(map) \ 436 (void) lockmgr(&(map)->lock, LK_DOWNGRADE, NULL) 437 438 #ifdef DIAGNOSTIC 439 #define vm_map_upgrade(map) \ 440 do { \ 441 if (lockmgr(&(map)->lock, LK_UPGRADE, NULL) != 0) \ 442 panic("vm_map_upgrade: failed to upgrade lock"); \ 443 } while (/*CONSTCOND*/ 0) 444 #else 445 #define vm_map_upgrade(map) \ 446 (void) lockmgr(&(map)->lock, LK_UPGRADE, NULL) 447 #endif 448 449 #define vm_map_busy(map) \ 450 do { \ 451 simple_lock(&(map)->flags_lock); \ 452 (map)->flags |= VM_MAP_BUSY; \ 453 simple_unlock(&(map)->flags_lock); \ 454 } while (/*CONSTCOND*/ 0) 455 456 #define vm_map_unbusy(map) \ 457 do { \ 458 int oflags; \ 459 \ 460 simple_lock(&(map)->flags_lock); \ 461 oflags = (map)->flags; \ 462 (map)->flags &= ~(VM_MAP_BUSY|VM_MAP_WANTLOCK); \ 463 simple_unlock(&(map)->flags_lock); \ 464 if (oflags & VM_MAP_WANTLOCK) \ 465 wakeup(&(map)->flags); \ 466 } while (/*CONSTCOND*/ 0) 467 #endif /* _KERNEL */ 468 469 /* 470 * Functions implemented as macros 471 */ 472 #define vm_map_min(map) ((map)->min_offset) 473 #define vm_map_max(map) ((map)->max_offset) 474 #define vm_map_pmap(map) ((map)->pmap) 475 476 #endif /* _UVM_UVM_MAP_H_ */ 477