1 /* $NetBSD: uvm_map.c,v 1.254 2008/04/27 11:39:47 ad 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.c 8.3 (Berkeley) 1/12/94 42 * from: Id: uvm_map.c,v 1.1.2.27 1998/02/07 01:16:54 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 /* 70 * uvm_map.c: uvm map operations 71 */ 72 73 #include <sys/cdefs.h> 74 __KERNEL_RCSID(0, "$NetBSD: uvm_map.c,v 1.254 2008/04/27 11:39:47 ad Exp $"); 75 76 #include "opt_ddb.h" 77 #include "opt_uvmhist.h" 78 #include "opt_uvm.h" 79 #include "opt_sysv.h" 80 81 #include <sys/param.h> 82 #include <sys/systm.h> 83 #include <sys/mman.h> 84 #include <sys/proc.h> 85 #include <sys/malloc.h> 86 #include <sys/pool.h> 87 #include <sys/kernel.h> 88 #include <sys/mount.h> 89 #include <sys/vnode.h> 90 #include <sys/lockdebug.h> 91 #include <sys/atomic.h> 92 93 #ifdef SYSVSHM 94 #include <sys/shm.h> 95 #endif 96 97 #include <uvm/uvm.h> 98 #undef RB_AUGMENT 99 #define RB_AUGMENT(x) uvm_rb_augment(x) 100 101 #ifdef DDB 102 #include <uvm/uvm_ddb.h> 103 #endif 104 105 #if defined(UVMMAP_NOCOUNTERS) 106 107 #define UVMMAP_EVCNT_DEFINE(name) /* nothing */ 108 #define UVMMAP_EVCNT_INCR(ev) /* nothing */ 109 #define UVMMAP_EVCNT_DECR(ev) /* nothing */ 110 111 #else /* defined(UVMMAP_NOCOUNTERS) */ 112 113 #include <sys/evcnt.h> 114 #define UVMMAP_EVCNT_DEFINE(name) \ 115 struct evcnt uvmmap_evcnt_##name = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, \ 116 "uvmmap", #name); \ 117 EVCNT_ATTACH_STATIC(uvmmap_evcnt_##name); 118 #define UVMMAP_EVCNT_INCR(ev) uvmmap_evcnt_##ev.ev_count++ 119 #define UVMMAP_EVCNT_DECR(ev) uvmmap_evcnt_##ev.ev_count-- 120 121 #endif /* defined(UVMMAP_NOCOUNTERS) */ 122 123 UVMMAP_EVCNT_DEFINE(ubackmerge) 124 UVMMAP_EVCNT_DEFINE(uforwmerge) 125 UVMMAP_EVCNT_DEFINE(ubimerge) 126 UVMMAP_EVCNT_DEFINE(unomerge) 127 UVMMAP_EVCNT_DEFINE(kbackmerge) 128 UVMMAP_EVCNT_DEFINE(kforwmerge) 129 UVMMAP_EVCNT_DEFINE(kbimerge) 130 UVMMAP_EVCNT_DEFINE(knomerge) 131 UVMMAP_EVCNT_DEFINE(map_call) 132 UVMMAP_EVCNT_DEFINE(mlk_call) 133 UVMMAP_EVCNT_DEFINE(mlk_hint) 134 135 UVMMAP_EVCNT_DEFINE(uke_alloc) 136 UVMMAP_EVCNT_DEFINE(uke_free) 137 UVMMAP_EVCNT_DEFINE(ukh_alloc) 138 UVMMAP_EVCNT_DEFINE(ukh_free) 139 140 const char vmmapbsy[] = "vmmapbsy"; 141 142 /* 143 * cache for vmspace structures. 144 */ 145 146 static struct pool_cache uvm_vmspace_cache; 147 148 /* 149 * cache for dynamically-allocated map entries. 150 */ 151 152 static struct pool_cache uvm_map_entry_cache; 153 154 MALLOC_DEFINE(M_VMMAP, "VM map", "VM map structures"); 155 MALLOC_DEFINE(M_VMPMAP, "VM pmap", "VM pmap"); 156 157 #ifdef PMAP_GROWKERNEL 158 /* 159 * This global represents the end of the kernel virtual address 160 * space. If we want to exceed this, we must grow the kernel 161 * virtual address space dynamically. 162 * 163 * Note, this variable is locked by kernel_map's lock. 164 */ 165 vaddr_t uvm_maxkaddr; 166 #endif 167 168 /* 169 * macros 170 */ 171 172 /* 173 * VM_MAP_USE_KMAPENT: determine if uvm_kmapent_alloc/free is used 174 * for the vm_map. 175 */ 176 extern struct vm_map *pager_map; /* XXX */ 177 #define VM_MAP_USE_KMAPENT_FLAGS(flags) \ 178 (((flags) & VM_MAP_INTRSAFE) != 0) 179 #define VM_MAP_USE_KMAPENT(map) \ 180 (VM_MAP_USE_KMAPENT_FLAGS((map)->flags) || (map) == kernel_map) 181 182 /* 183 * UVM_ET_ISCOMPATIBLE: check some requirements for map entry merging 184 */ 185 186 #define UVM_ET_ISCOMPATIBLE(ent, type, uobj, meflags, \ 187 prot, maxprot, inh, adv, wire) \ 188 ((ent)->etype == (type) && \ 189 (((ent)->flags ^ (meflags)) & (UVM_MAP_NOMERGE | UVM_MAP_QUANTUM)) \ 190 == 0 && \ 191 (ent)->object.uvm_obj == (uobj) && \ 192 (ent)->protection == (prot) && \ 193 (ent)->max_protection == (maxprot) && \ 194 (ent)->inheritance == (inh) && \ 195 (ent)->advice == (adv) && \ 196 (ent)->wired_count == (wire)) 197 198 /* 199 * uvm_map_entry_link: insert entry into a map 200 * 201 * => map must be locked 202 */ 203 #define uvm_map_entry_link(map, after_where, entry) do { \ 204 uvm_mapent_check(entry); \ 205 (map)->nentries++; \ 206 (entry)->prev = (after_where); \ 207 (entry)->next = (after_where)->next; \ 208 (entry)->prev->next = (entry); \ 209 (entry)->next->prev = (entry); \ 210 uvm_rb_insert((map), (entry)); \ 211 } while (/*CONSTCOND*/ 0) 212 213 /* 214 * uvm_map_entry_unlink: remove entry from a map 215 * 216 * => map must be locked 217 */ 218 #define uvm_map_entry_unlink(map, entry) do { \ 219 KASSERT((entry) != (map)->first_free); \ 220 KASSERT((entry) != (map)->hint); \ 221 uvm_mapent_check(entry); \ 222 (map)->nentries--; \ 223 (entry)->next->prev = (entry)->prev; \ 224 (entry)->prev->next = (entry)->next; \ 225 uvm_rb_remove((map), (entry)); \ 226 } while (/*CONSTCOND*/ 0) 227 228 /* 229 * SAVE_HINT: saves the specified entry as the hint for future lookups. 230 * 231 * => map need not be locked. 232 */ 233 #define SAVE_HINT(map, check, value) do { \ 234 atomic_cas_ptr(&(map)->hint, (check), (value)); \ 235 } while (/*CONSTCOND*/ 0) 236 237 /* 238 * clear_hints: ensure that hints don't point to the entry. 239 * 240 * => map must be write-locked. 241 */ 242 static void 243 clear_hints(struct vm_map *map, struct vm_map_entry *ent) 244 { 245 246 SAVE_HINT(map, ent, ent->prev); 247 if (map->first_free == ent) { 248 map->first_free = ent->prev; 249 } 250 } 251 252 /* 253 * VM_MAP_RANGE_CHECK: check and correct range 254 * 255 * => map must at least be read locked 256 */ 257 258 #define VM_MAP_RANGE_CHECK(map, start, end) do { \ 259 if (start < vm_map_min(map)) \ 260 start = vm_map_min(map); \ 261 if (end > vm_map_max(map)) \ 262 end = vm_map_max(map); \ 263 if (start > end) \ 264 start = end; \ 265 } while (/*CONSTCOND*/ 0) 266 267 /* 268 * local prototypes 269 */ 270 271 static struct vm_map_entry * 272 uvm_mapent_alloc(struct vm_map *, int); 273 static struct vm_map_entry * 274 uvm_mapent_alloc_split(struct vm_map *, 275 const struct vm_map_entry *, int, 276 struct uvm_mapent_reservation *); 277 static void uvm_mapent_copy(struct vm_map_entry *, struct vm_map_entry *); 278 static void uvm_mapent_free(struct vm_map_entry *); 279 #if defined(DEBUG) 280 static void _uvm_mapent_check(const struct vm_map_entry *, const char *, 281 int); 282 #define uvm_mapent_check(map) _uvm_mapent_check(map, __FILE__, __LINE__) 283 #else /* defined(DEBUG) */ 284 #define uvm_mapent_check(e) /* nothing */ 285 #endif /* defined(DEBUG) */ 286 static struct vm_map_entry * 287 uvm_kmapent_alloc(struct vm_map *, int); 288 static void uvm_kmapent_free(struct vm_map_entry *); 289 static vsize_t uvm_kmapent_overhead(vsize_t); 290 291 static void uvm_map_entry_unwire(struct vm_map *, struct vm_map_entry *); 292 static void uvm_map_reference_amap(struct vm_map_entry *, int); 293 static int uvm_map_space_avail(vaddr_t *, vsize_t, voff_t, vsize_t, int, 294 struct vm_map_entry *); 295 static void uvm_map_unreference_amap(struct vm_map_entry *, int); 296 297 int _uvm_map_sanity(struct vm_map *); 298 int _uvm_tree_sanity(struct vm_map *); 299 static vsize_t uvm_rb_subtree_space(const struct vm_map_entry *); 300 301 static inline int 302 uvm_compare(const struct vm_map_entry *a, const struct vm_map_entry *b) 303 { 304 305 if (a->start < b->start) 306 return (-1); 307 else if (a->start > b->start) 308 return (1); 309 310 return (0); 311 } 312 313 static inline void 314 uvm_rb_augment(struct vm_map_entry *entry) 315 { 316 317 entry->space = uvm_rb_subtree_space(entry); 318 } 319 320 RB_PROTOTYPE(uvm_tree, vm_map_entry, rb_entry, uvm_compare); 321 322 RB_GENERATE(uvm_tree, vm_map_entry, rb_entry, uvm_compare); 323 324 static inline vsize_t 325 uvm_rb_space(const struct vm_map *map, const struct vm_map_entry *entry) 326 { 327 /* XXX map is not used */ 328 329 KASSERT(entry->next != NULL); 330 return entry->next->start - entry->end; 331 } 332 333 static vsize_t 334 uvm_rb_subtree_space(const struct vm_map_entry *entry) 335 { 336 vaddr_t space, tmp; 337 338 space = entry->ownspace; 339 if (RB_LEFT(entry, rb_entry)) { 340 tmp = RB_LEFT(entry, rb_entry)->space; 341 if (tmp > space) 342 space = tmp; 343 } 344 345 if (RB_RIGHT(entry, rb_entry)) { 346 tmp = RB_RIGHT(entry, rb_entry)->space; 347 if (tmp > space) 348 space = tmp; 349 } 350 351 return (space); 352 } 353 354 static inline void 355 uvm_rb_fixup(struct vm_map *map, struct vm_map_entry *entry) 356 { 357 /* We need to traverse to the very top */ 358 do { 359 entry->ownspace = uvm_rb_space(map, entry); 360 entry->space = uvm_rb_subtree_space(entry); 361 } while ((entry = RB_PARENT(entry, rb_entry)) != NULL); 362 } 363 364 static void 365 uvm_rb_insert(struct vm_map *map, struct vm_map_entry *entry) 366 { 367 vaddr_t space = uvm_rb_space(map, entry); 368 struct vm_map_entry *tmp; 369 370 entry->ownspace = entry->space = space; 371 tmp = RB_INSERT(uvm_tree, &(map)->rbhead, entry); 372 #ifdef DIAGNOSTIC 373 if (tmp != NULL) 374 panic("uvm_rb_insert: duplicate entry?"); 375 #endif 376 uvm_rb_fixup(map, entry); 377 if (entry->prev != &map->header) 378 uvm_rb_fixup(map, entry->prev); 379 } 380 381 static void 382 uvm_rb_remove(struct vm_map *map, struct vm_map_entry *entry) 383 { 384 struct vm_map_entry *parent; 385 386 parent = RB_PARENT(entry, rb_entry); 387 RB_REMOVE(uvm_tree, &(map)->rbhead, entry); 388 if (entry->prev != &map->header) 389 uvm_rb_fixup(map, entry->prev); 390 if (parent) 391 uvm_rb_fixup(map, parent); 392 } 393 394 #if defined(DEBUG) 395 int uvm_debug_check_map = 0; 396 int uvm_debug_check_rbtree = 0; 397 #define uvm_map_check(map, name) \ 398 _uvm_map_check((map), (name), __FILE__, __LINE__) 399 static void 400 _uvm_map_check(struct vm_map *map, const char *name, 401 const char *file, int line) 402 { 403 404 if ((uvm_debug_check_map && _uvm_map_sanity(map)) || 405 (uvm_debug_check_rbtree && _uvm_tree_sanity(map))) { 406 panic("uvm_map_check failed: \"%s\" map=%p (%s:%d)", 407 name, map, file, line); 408 } 409 } 410 #else /* defined(DEBUG) */ 411 #define uvm_map_check(map, name) /* nothing */ 412 #endif /* defined(DEBUG) */ 413 414 #if defined(DEBUG) || defined(DDB) 415 int 416 _uvm_map_sanity(struct vm_map *map) 417 { 418 bool first_free_found = false; 419 bool hint_found = false; 420 const struct vm_map_entry *e; 421 422 e = &map->header; 423 for (;;) { 424 if (map->first_free == e) { 425 first_free_found = true; 426 } else if (!first_free_found && e->next->start > e->end) { 427 printf("first_free %p should be %p\n", 428 map->first_free, e); 429 return -1; 430 } 431 if (map->hint == e) { 432 hint_found = true; 433 } 434 435 e = e->next; 436 if (e == &map->header) { 437 break; 438 } 439 } 440 if (!first_free_found) { 441 printf("stale first_free\n"); 442 return -1; 443 } 444 if (!hint_found) { 445 printf("stale hint\n"); 446 return -1; 447 } 448 return 0; 449 } 450 451 int 452 _uvm_tree_sanity(struct vm_map *map) 453 { 454 struct vm_map_entry *tmp, *trtmp; 455 int n = 0, i = 1; 456 457 RB_FOREACH(tmp, uvm_tree, &map->rbhead) { 458 if (tmp->ownspace != uvm_rb_space(map, tmp)) { 459 printf("%d/%d ownspace %lx != %lx %s\n", 460 n + 1, map->nentries, 461 (ulong)tmp->ownspace, (ulong)uvm_rb_space(map, tmp), 462 tmp->next == &map->header ? "(last)" : ""); 463 goto error; 464 } 465 } 466 trtmp = NULL; 467 RB_FOREACH(tmp, uvm_tree, &map->rbhead) { 468 if (tmp->space != uvm_rb_subtree_space(tmp)) { 469 printf("space %lx != %lx\n", 470 (ulong)tmp->space, 471 (ulong)uvm_rb_subtree_space(tmp)); 472 goto error; 473 } 474 if (trtmp != NULL && trtmp->start >= tmp->start) { 475 printf("corrupt: 0x%lx >= 0x%lx\n", 476 trtmp->start, tmp->start); 477 goto error; 478 } 479 n++; 480 481 trtmp = tmp; 482 } 483 484 if (n != map->nentries) { 485 printf("nentries: %d vs %d\n", n, map->nentries); 486 goto error; 487 } 488 489 for (tmp = map->header.next; tmp && tmp != &map->header; 490 tmp = tmp->next, i++) { 491 trtmp = RB_FIND(uvm_tree, &map->rbhead, tmp); 492 if (trtmp != tmp) { 493 printf("lookup: %d: %p - %p: %p\n", i, tmp, trtmp, 494 RB_PARENT(tmp, rb_entry)); 495 goto error; 496 } 497 } 498 499 return (0); 500 error: 501 return (-1); 502 } 503 #endif /* defined(DEBUG) || defined(DDB) */ 504 505 #ifdef DIAGNOSTIC 506 static struct vm_map *uvm_kmapent_map(struct vm_map_entry *); 507 #endif 508 509 /* 510 * vm_map_lock: acquire an exclusive (write) lock on a map. 511 * 512 * => Note that "intrsafe" maps use only exclusive, spin locks. 513 * 514 * => The locking protocol provides for guaranteed upgrade from shared -> 515 * exclusive by whichever thread currently has the map marked busy. 516 * See "LOCKING PROTOCOL NOTES" in uvm_map.h. This is horrible; among 517 * other problems, it defeats any fairness guarantees provided by RW 518 * locks. 519 */ 520 521 void 522 vm_map_lock(struct vm_map *map) 523 { 524 525 if ((map->flags & VM_MAP_INTRSAFE) != 0) { 526 mutex_spin_enter(&map->mutex); 527 return; 528 } 529 530 for (;;) { 531 rw_enter(&map->lock, RW_WRITER); 532 if (map->busy == NULL) 533 break; 534 if (map->busy == curlwp) 535 break; 536 mutex_enter(&map->misc_lock); 537 rw_exit(&map->lock); 538 if (map->busy != NULL) 539 cv_wait(&map->cv, &map->misc_lock); 540 mutex_exit(&map->misc_lock); 541 } 542 543 map->timestamp++; 544 } 545 546 /* 547 * vm_map_lock_try: try to lock a map, failing if it is already locked. 548 */ 549 550 bool 551 vm_map_lock_try(struct vm_map *map) 552 { 553 554 if ((map->flags & VM_MAP_INTRSAFE) != 0) 555 return mutex_tryenter(&map->mutex); 556 if (!rw_tryenter(&map->lock, RW_WRITER)) 557 return false; 558 if (map->busy != NULL) { 559 rw_exit(&map->lock); 560 return false; 561 } 562 563 map->timestamp++; 564 return true; 565 } 566 567 /* 568 * vm_map_unlock: release an exclusive lock on a map. 569 */ 570 571 void 572 vm_map_unlock(struct vm_map *map) 573 { 574 575 if ((map->flags & VM_MAP_INTRSAFE) != 0) 576 mutex_spin_exit(&map->mutex); 577 else { 578 KASSERT(rw_write_held(&map->lock)); 579 KASSERT(map->busy == NULL || map->busy == curlwp); 580 rw_exit(&map->lock); 581 } 582 } 583 584 /* 585 * vm_map_unbusy: mark the map as unbusy, and wake any waiters that 586 * want an exclusive lock. 587 */ 588 589 void 590 vm_map_unbusy(struct vm_map *map) 591 { 592 593 KASSERT(map->busy == curlwp); 594 595 /* 596 * Safe to clear 'busy' and 'waiters' with only a read lock held: 597 * 598 * o they can only be set with a write lock held 599 * o writers are blocked out with a read or write hold 600 * o at any time, only one thread owns the set of values 601 */ 602 mutex_enter(&map->misc_lock); 603 map->busy = NULL; 604 cv_broadcast(&map->cv); 605 mutex_exit(&map->misc_lock); 606 } 607 608 /* 609 * vm_map_lock_read: acquire a shared (read) lock on a map. 610 */ 611 612 void 613 vm_map_lock_read(struct vm_map *map) 614 { 615 616 KASSERT((map->flags & VM_MAP_INTRSAFE) == 0); 617 618 rw_enter(&map->lock, RW_READER); 619 } 620 621 /* 622 * vm_map_unlock_read: release a shared lock on a map. 623 */ 624 625 void 626 vm_map_unlock_read(struct vm_map *map) 627 { 628 629 KASSERT((map->flags & VM_MAP_INTRSAFE) == 0); 630 631 rw_exit(&map->lock); 632 } 633 634 /* 635 * vm_map_busy: mark a map as busy. 636 * 637 * => the caller must hold the map write locked 638 */ 639 640 void 641 vm_map_busy(struct vm_map *map) 642 { 643 644 KASSERT(rw_write_held(&map->lock)); 645 KASSERT(map->busy == NULL); 646 647 map->busy = curlwp; 648 } 649 650 /* 651 * vm_map_locked_p: return true if the map is write locked. 652 */ 653 654 bool 655 vm_map_locked_p(struct vm_map *map) 656 { 657 658 if ((map->flags & VM_MAP_INTRSAFE) != 0) { 659 return mutex_owned(&map->mutex); 660 } else { 661 return rw_write_held(&map->lock); 662 } 663 } 664 665 /* 666 * uvm_mapent_alloc: allocate a map entry 667 */ 668 669 static struct vm_map_entry * 670 uvm_mapent_alloc(struct vm_map *map, int flags) 671 { 672 struct vm_map_entry *me; 673 int pflags = (flags & UVM_FLAG_NOWAIT) ? PR_NOWAIT : PR_WAITOK; 674 UVMHIST_FUNC("uvm_mapent_alloc"); UVMHIST_CALLED(maphist); 675 676 if (VM_MAP_USE_KMAPENT(map)) { 677 me = uvm_kmapent_alloc(map, flags); 678 } else { 679 me = pool_cache_get(&uvm_map_entry_cache, pflags); 680 if (__predict_false(me == NULL)) 681 return NULL; 682 me->flags = 0; 683 } 684 685 UVMHIST_LOG(maphist, "<- new entry=0x%x [kentry=%d]", me, 686 ((map->flags & VM_MAP_INTRSAFE) != 0 || map == kernel_map), 0, 0); 687 return (me); 688 } 689 690 /* 691 * uvm_mapent_alloc_split: allocate a map entry for clipping. 692 * 693 * => map must be locked by caller if UVM_MAP_QUANTUM is set. 694 */ 695 696 static struct vm_map_entry * 697 uvm_mapent_alloc_split(struct vm_map *map, 698 const struct vm_map_entry *old_entry, int flags, 699 struct uvm_mapent_reservation *umr) 700 { 701 struct vm_map_entry *me; 702 703 KASSERT(!VM_MAP_USE_KMAPENT(map) || 704 (old_entry->flags & UVM_MAP_QUANTUM) || !UMR_EMPTY(umr)); 705 706 if (old_entry->flags & UVM_MAP_QUANTUM) { 707 struct vm_map_kernel *vmk = vm_map_to_kernel(map); 708 709 KASSERT(vm_map_locked_p(map)); 710 me = vmk->vmk_merged_entries; 711 KASSERT(me); 712 vmk->vmk_merged_entries = me->next; 713 KASSERT(me->flags & UVM_MAP_QUANTUM); 714 } else { 715 me = uvm_mapent_alloc(map, flags); 716 } 717 718 return me; 719 } 720 721 /* 722 * uvm_mapent_free: free map entry 723 */ 724 725 static void 726 uvm_mapent_free(struct vm_map_entry *me) 727 { 728 UVMHIST_FUNC("uvm_mapent_free"); UVMHIST_CALLED(maphist); 729 730 UVMHIST_LOG(maphist,"<- freeing map entry=0x%x [flags=%d]", 731 me, me->flags, 0, 0); 732 if (me->flags & UVM_MAP_KERNEL) { 733 uvm_kmapent_free(me); 734 } else { 735 pool_cache_put(&uvm_map_entry_cache, me); 736 } 737 } 738 739 /* 740 * uvm_mapent_free_merged: free merged map entry 741 * 742 * => keep the entry if needed. 743 * => caller shouldn't hold map locked if VM_MAP_USE_KMAPENT(map) is true. 744 * => map should be locked if UVM_MAP_QUANTUM is set. 745 */ 746 747 static void 748 uvm_mapent_free_merged(struct vm_map *map, struct vm_map_entry *me) 749 { 750 751 KASSERT(!(me->flags & UVM_MAP_KERNEL) || uvm_kmapent_map(me) == map); 752 753 if (me->flags & UVM_MAP_QUANTUM) { 754 /* 755 * keep this entry for later splitting. 756 */ 757 struct vm_map_kernel *vmk; 758 759 KASSERT(vm_map_locked_p(map)); 760 KASSERT(VM_MAP_IS_KERNEL(map)); 761 KASSERT(!VM_MAP_USE_KMAPENT(map) || 762 (me->flags & UVM_MAP_KERNEL)); 763 764 vmk = vm_map_to_kernel(map); 765 me->next = vmk->vmk_merged_entries; 766 vmk->vmk_merged_entries = me; 767 } else { 768 uvm_mapent_free(me); 769 } 770 } 771 772 /* 773 * uvm_mapent_copy: copy a map entry, preserving flags 774 */ 775 776 static inline void 777 uvm_mapent_copy(struct vm_map_entry *src, struct vm_map_entry *dst) 778 { 779 780 memcpy(dst, src, ((char *)&src->uvm_map_entry_stop_copy) - 781 ((char *)src)); 782 } 783 784 /* 785 * uvm_mapent_overhead: calculate maximum kva overhead necessary for 786 * map entries. 787 * 788 * => size and flags are the same as uvm_km_suballoc's ones. 789 */ 790 791 vsize_t 792 uvm_mapent_overhead(vsize_t size, int flags) 793 { 794 795 if (VM_MAP_USE_KMAPENT_FLAGS(flags)) { 796 return uvm_kmapent_overhead(size); 797 } 798 return 0; 799 } 800 801 #if defined(DEBUG) 802 static void 803 _uvm_mapent_check(const struct vm_map_entry *entry, const char *file, int line) 804 { 805 806 if (entry->start >= entry->end) { 807 goto bad; 808 } 809 if (UVM_ET_ISOBJ(entry)) { 810 if (entry->object.uvm_obj == NULL) { 811 goto bad; 812 } 813 } else if (UVM_ET_ISSUBMAP(entry)) { 814 if (entry->object.sub_map == NULL) { 815 goto bad; 816 } 817 } else { 818 if (entry->object.uvm_obj != NULL || 819 entry->object.sub_map != NULL) { 820 goto bad; 821 } 822 } 823 if (!UVM_ET_ISOBJ(entry)) { 824 if (entry->offset != 0) { 825 goto bad; 826 } 827 } 828 829 return; 830 831 bad: 832 panic("%s: bad entry %p (%s:%d)", __func__, entry, file, line); 833 } 834 #endif /* defined(DEBUG) */ 835 836 /* 837 * uvm_map_entry_unwire: unwire a map entry 838 * 839 * => map should be locked by caller 840 */ 841 842 static inline void 843 uvm_map_entry_unwire(struct vm_map *map, struct vm_map_entry *entry) 844 { 845 846 entry->wired_count = 0; 847 uvm_fault_unwire_locked(map, entry->start, entry->end); 848 } 849 850 851 /* 852 * wrapper for calling amap_ref() 853 */ 854 static inline void 855 uvm_map_reference_amap(struct vm_map_entry *entry, int flags) 856 { 857 858 amap_ref(entry->aref.ar_amap, entry->aref.ar_pageoff, 859 (entry->end - entry->start) >> PAGE_SHIFT, flags); 860 } 861 862 863 /* 864 * wrapper for calling amap_unref() 865 */ 866 static inline void 867 uvm_map_unreference_amap(struct vm_map_entry *entry, int flags) 868 { 869 870 amap_unref(entry->aref.ar_amap, entry->aref.ar_pageoff, 871 (entry->end - entry->start) >> PAGE_SHIFT, flags); 872 } 873 874 875 /* 876 * uvm_map_init: init mapping system at boot time. 877 */ 878 879 void 880 uvm_map_init(void) 881 { 882 #if defined(UVMHIST) 883 static struct uvm_history_ent maphistbuf[100]; 884 static struct uvm_history_ent pdhistbuf[100]; 885 #endif 886 887 /* 888 * first, init logging system. 889 */ 890 891 UVMHIST_FUNC("uvm_map_init"); 892 UVMHIST_INIT_STATIC(maphist, maphistbuf); 893 UVMHIST_INIT_STATIC(pdhist, pdhistbuf); 894 UVMHIST_CALLED(maphist); 895 UVMHIST_LOG(maphist,"<starting uvm map system>", 0, 0, 0, 0); 896 897 /* 898 * initialize the global lock for kernel map entry. 899 */ 900 901 mutex_init(&uvm_kentry_lock, MUTEX_DRIVER, IPL_VM); 902 903 /* 904 * initialize caches. 905 */ 906 907 pool_cache_bootstrap(&uvm_map_entry_cache, sizeof(struct vm_map_entry), 908 0, 0, 0, "vmmpepl", NULL, IPL_NONE, NULL, NULL, NULL); 909 pool_cache_bootstrap(&uvm_vmspace_cache, sizeof(struct vmspace), 910 0, 0, 0, "vmsppl", NULL, IPL_NONE, NULL, NULL, NULL); 911 } 912 913 /* 914 * clippers 915 */ 916 917 /* 918 * uvm_mapent_splitadj: adjust map entries for splitting, after uvm_mapent_copy. 919 */ 920 921 static void 922 uvm_mapent_splitadj(struct vm_map_entry *entry1, struct vm_map_entry *entry2, 923 vaddr_t splitat) 924 { 925 vaddr_t adj; 926 927 KASSERT(entry1->start < splitat); 928 KASSERT(splitat < entry1->end); 929 930 adj = splitat - entry1->start; 931 entry1->end = entry2->start = splitat; 932 933 if (entry1->aref.ar_amap) { 934 amap_splitref(&entry1->aref, &entry2->aref, adj); 935 } 936 if (UVM_ET_ISSUBMAP(entry1)) { 937 /* ... unlikely to happen, but play it safe */ 938 uvm_map_reference(entry1->object.sub_map); 939 } else if (UVM_ET_ISOBJ(entry1)) { 940 KASSERT(entry1->object.uvm_obj != NULL); /* suppress coverity */ 941 entry2->offset += adj; 942 if (entry1->object.uvm_obj->pgops && 943 entry1->object.uvm_obj->pgops->pgo_reference) 944 entry1->object.uvm_obj->pgops->pgo_reference( 945 entry1->object.uvm_obj); 946 } 947 } 948 949 /* 950 * uvm_map_clip_start: ensure that the entry begins at or after 951 * the starting address, if it doesn't we split the entry. 952 * 953 * => caller should use UVM_MAP_CLIP_START macro rather than calling 954 * this directly 955 * => map must be locked by caller 956 */ 957 958 void 959 uvm_map_clip_start(struct vm_map *map, struct vm_map_entry *entry, 960 vaddr_t start, struct uvm_mapent_reservation *umr) 961 { 962 struct vm_map_entry *new_entry; 963 964 /* uvm_map_simplify_entry(map, entry); */ /* XXX */ 965 966 uvm_map_check(map, "clip_start entry"); 967 uvm_mapent_check(entry); 968 969 /* 970 * Split off the front portion. note that we must insert the new 971 * entry BEFORE this one, so that this entry has the specified 972 * starting address. 973 */ 974 new_entry = uvm_mapent_alloc_split(map, entry, 0, umr); 975 uvm_mapent_copy(entry, new_entry); /* entry -> new_entry */ 976 uvm_mapent_splitadj(new_entry, entry, start); 977 uvm_map_entry_link(map, entry->prev, new_entry); 978 979 uvm_map_check(map, "clip_start leave"); 980 } 981 982 /* 983 * uvm_map_clip_end: ensure that the entry ends at or before 984 * the ending address, if it does't we split the reference 985 * 986 * => caller should use UVM_MAP_CLIP_END macro rather than calling 987 * this directly 988 * => map must be locked by caller 989 */ 990 991 void 992 uvm_map_clip_end(struct vm_map *map, struct vm_map_entry *entry, vaddr_t end, 993 struct uvm_mapent_reservation *umr) 994 { 995 struct vm_map_entry *new_entry; 996 997 uvm_map_check(map, "clip_end entry"); 998 uvm_mapent_check(entry); 999 1000 /* 1001 * Create a new entry and insert it 1002 * AFTER the specified entry 1003 */ 1004 new_entry = uvm_mapent_alloc_split(map, entry, 0, umr); 1005 uvm_mapent_copy(entry, new_entry); /* entry -> new_entry */ 1006 uvm_mapent_splitadj(entry, new_entry, end); 1007 uvm_map_entry_link(map, entry, new_entry); 1008 1009 uvm_map_check(map, "clip_end leave"); 1010 } 1011 1012 static void 1013 vm_map_drain(struct vm_map *map, uvm_flag_t flags) 1014 { 1015 1016 if (!VM_MAP_IS_KERNEL(map)) { 1017 return; 1018 } 1019 1020 uvm_km_va_drain(map, flags); 1021 } 1022 1023 /* 1024 * M A P - m a i n e n t r y p o i n t 1025 */ 1026 /* 1027 * uvm_map: establish a valid mapping in a map 1028 * 1029 * => assume startp is page aligned. 1030 * => assume size is a multiple of PAGE_SIZE. 1031 * => assume sys_mmap provides enough of a "hint" to have us skip 1032 * over text/data/bss area. 1033 * => map must be unlocked (we will lock it) 1034 * => <uobj,uoffset> value meanings (4 cases): 1035 * [1] <NULL,uoffset> == uoffset is a hint for PMAP_PREFER 1036 * [2] <NULL,UVM_UNKNOWN_OFFSET> == don't PMAP_PREFER 1037 * [3] <uobj,uoffset> == normal mapping 1038 * [4] <uobj,UVM_UNKNOWN_OFFSET> == uvm_map finds offset based on VA 1039 * 1040 * case [4] is for kernel mappings where we don't know the offset until 1041 * we've found a virtual address. note that kernel object offsets are 1042 * always relative to vm_map_min(kernel_map). 1043 * 1044 * => if `align' is non-zero, we align the virtual address to the specified 1045 * alignment. 1046 * this is provided as a mechanism for large pages. 1047 * 1048 * => XXXCDC: need way to map in external amap? 1049 */ 1050 1051 int 1052 uvm_map(struct vm_map *map, vaddr_t *startp /* IN/OUT */, vsize_t size, 1053 struct uvm_object *uobj, voff_t uoffset, vsize_t align, uvm_flag_t flags) 1054 { 1055 struct uvm_map_args args; 1056 struct vm_map_entry *new_entry; 1057 int error; 1058 1059 KASSERT((flags & UVM_FLAG_QUANTUM) == 0 || VM_MAP_IS_KERNEL(map)); 1060 KASSERT((size & PAGE_MASK) == 0); 1061 1062 /* 1063 * for pager_map, allocate the new entry first to avoid sleeping 1064 * for memory while we have the map locked. 1065 * 1066 * besides, because we allocates entries for in-kernel maps 1067 * a bit differently (cf. uvm_kmapent_alloc/free), we need to 1068 * allocate them before locking the map. 1069 */ 1070 1071 new_entry = NULL; 1072 if (VM_MAP_USE_KMAPENT(map) || (flags & UVM_FLAG_QUANTUM) || 1073 map == pager_map) { 1074 new_entry = uvm_mapent_alloc(map, (flags & UVM_FLAG_NOWAIT)); 1075 if (__predict_false(new_entry == NULL)) 1076 return ENOMEM; 1077 if (flags & UVM_FLAG_QUANTUM) 1078 new_entry->flags |= UVM_MAP_QUANTUM; 1079 } 1080 if (map == pager_map) 1081 flags |= UVM_FLAG_NOMERGE; 1082 1083 error = uvm_map_prepare(map, *startp, size, uobj, uoffset, align, 1084 flags, &args); 1085 if (!error) { 1086 error = uvm_map_enter(map, &args, new_entry); 1087 *startp = args.uma_start; 1088 } else if (new_entry) { 1089 uvm_mapent_free(new_entry); 1090 } 1091 1092 #if defined(DEBUG) 1093 if (!error && VM_MAP_IS_KERNEL(map)) { 1094 uvm_km_check_empty(*startp, *startp + size, 1095 (map->flags & VM_MAP_INTRSAFE) != 0); 1096 } 1097 #endif /* defined(DEBUG) */ 1098 1099 return error; 1100 } 1101 1102 int 1103 uvm_map_prepare(struct vm_map *map, vaddr_t start, vsize_t size, 1104 struct uvm_object *uobj, voff_t uoffset, vsize_t align, uvm_flag_t flags, 1105 struct uvm_map_args *args) 1106 { 1107 struct vm_map_entry *prev_entry; 1108 vm_prot_t prot = UVM_PROTECTION(flags); 1109 vm_prot_t maxprot = UVM_MAXPROTECTION(flags); 1110 1111 UVMHIST_FUNC("uvm_map_prepare"); 1112 UVMHIST_CALLED(maphist); 1113 1114 UVMHIST_LOG(maphist, "(map=0x%x, start=0x%x, size=%d, flags=0x%x)", 1115 map, start, size, flags); 1116 UVMHIST_LOG(maphist, " uobj/offset 0x%x/%d", uobj, uoffset,0,0); 1117 1118 /* 1119 * detect a popular device driver bug. 1120 */ 1121 1122 KASSERT(doing_shutdown || curlwp != NULL || 1123 (map->flags & VM_MAP_INTRSAFE)); 1124 1125 /* 1126 * zero-sized mapping doesn't make any sense. 1127 */ 1128 KASSERT(size > 0); 1129 1130 KASSERT((~flags & (UVM_FLAG_NOWAIT | UVM_FLAG_WAITVA)) != 0); 1131 1132 uvm_map_check(map, "map entry"); 1133 1134 /* 1135 * check sanity of protection code 1136 */ 1137 1138 if ((prot & maxprot) != prot) { 1139 UVMHIST_LOG(maphist, "<- prot. failure: prot=0x%x, max=0x%x", 1140 prot, maxprot,0,0); 1141 return EACCES; 1142 } 1143 1144 /* 1145 * figure out where to put new VM range 1146 */ 1147 1148 retry: 1149 if (vm_map_lock_try(map) == false) { 1150 if ((flags & UVM_FLAG_TRYLOCK) != 0 && 1151 (map->flags & VM_MAP_INTRSAFE) == 0) { 1152 return EAGAIN; 1153 } 1154 vm_map_lock(map); /* could sleep here */ 1155 } 1156 prev_entry = uvm_map_findspace(map, start, size, &start, 1157 uobj, uoffset, align, flags); 1158 if (prev_entry == NULL) { 1159 unsigned int timestamp; 1160 1161 timestamp = map->timestamp; 1162 UVMHIST_LOG(maphist,"waiting va timestamp=0x%x", 1163 timestamp,0,0,0); 1164 map->flags |= VM_MAP_WANTVA; 1165 vm_map_unlock(map); 1166 1167 /* 1168 * try to reclaim kva and wait until someone does unmap. 1169 * fragile locking here, so we awaken every second to 1170 * recheck the condition. 1171 */ 1172 1173 vm_map_drain(map, flags); 1174 1175 mutex_enter(&map->misc_lock); 1176 while ((map->flags & VM_MAP_WANTVA) != 0 && 1177 map->timestamp == timestamp) { 1178 if ((flags & UVM_FLAG_WAITVA) == 0) { 1179 mutex_exit(&map->misc_lock); 1180 UVMHIST_LOG(maphist, 1181 "<- uvm_map_findspace failed!", 0,0,0,0); 1182 return ENOMEM; 1183 } else { 1184 cv_timedwait(&map->cv, &map->misc_lock, hz); 1185 } 1186 } 1187 mutex_exit(&map->misc_lock); 1188 goto retry; 1189 } 1190 1191 #ifdef PMAP_GROWKERNEL 1192 /* 1193 * If the kernel pmap can't map the requested space, 1194 * then allocate more resources for it. 1195 */ 1196 if (map == kernel_map && uvm_maxkaddr < (start + size)) 1197 uvm_maxkaddr = pmap_growkernel(start + size); 1198 #endif 1199 1200 UVMMAP_EVCNT_INCR(map_call); 1201 1202 /* 1203 * if uobj is null, then uoffset is either a VAC hint for PMAP_PREFER 1204 * [typically from uvm_map_reserve] or it is UVM_UNKNOWN_OFFSET. in 1205 * either case we want to zero it before storing it in the map entry 1206 * (because it looks strange and confusing when debugging...) 1207 * 1208 * if uobj is not null 1209 * if uoffset is not UVM_UNKNOWN_OFFSET then we have a normal mapping 1210 * and we do not need to change uoffset. 1211 * if uoffset is UVM_UNKNOWN_OFFSET then we need to find the offset 1212 * now (based on the starting address of the map). this case is 1213 * for kernel object mappings where we don't know the offset until 1214 * the virtual address is found (with uvm_map_findspace). the 1215 * offset is the distance we are from the start of the map. 1216 */ 1217 1218 if (uobj == NULL) { 1219 uoffset = 0; 1220 } else { 1221 if (uoffset == UVM_UNKNOWN_OFFSET) { 1222 KASSERT(UVM_OBJ_IS_KERN_OBJECT(uobj)); 1223 uoffset = start - vm_map_min(kernel_map); 1224 } 1225 } 1226 1227 args->uma_flags = flags; 1228 args->uma_prev = prev_entry; 1229 args->uma_start = start; 1230 args->uma_size = size; 1231 args->uma_uobj = uobj; 1232 args->uma_uoffset = uoffset; 1233 1234 return 0; 1235 } 1236 1237 int 1238 uvm_map_enter(struct vm_map *map, const struct uvm_map_args *args, 1239 struct vm_map_entry *new_entry) 1240 { 1241 struct vm_map_entry *prev_entry = args->uma_prev; 1242 struct vm_map_entry *dead = NULL; 1243 1244 const uvm_flag_t flags = args->uma_flags; 1245 const vm_prot_t prot = UVM_PROTECTION(flags); 1246 const vm_prot_t maxprot = UVM_MAXPROTECTION(flags); 1247 const vm_inherit_t inherit = UVM_INHERIT(flags); 1248 const int amapwaitflag = (flags & UVM_FLAG_NOWAIT) ? 1249 AMAP_EXTEND_NOWAIT : 0; 1250 const int advice = UVM_ADVICE(flags); 1251 const int meflagval = (flags & UVM_FLAG_QUANTUM) ? 1252 UVM_MAP_QUANTUM : 0; 1253 1254 vaddr_t start = args->uma_start; 1255 vsize_t size = args->uma_size; 1256 struct uvm_object *uobj = args->uma_uobj; 1257 voff_t uoffset = args->uma_uoffset; 1258 1259 const int kmap = (vm_map_pmap(map) == pmap_kernel()); 1260 int merged = 0; 1261 int error; 1262 int newetype; 1263 1264 UVMHIST_FUNC("uvm_map_enter"); 1265 UVMHIST_CALLED(maphist); 1266 1267 UVMHIST_LOG(maphist, "(map=0x%x, start=0x%x, size=%d, flags=0x%x)", 1268 map, start, size, flags); 1269 UVMHIST_LOG(maphist, " uobj/offset 0x%x/%d", uobj, uoffset,0,0); 1270 1271 KASSERT(map->hint == prev_entry); /* bimerge case assumes this */ 1272 1273 if (flags & UVM_FLAG_QUANTUM) { 1274 KASSERT(new_entry); 1275 KASSERT(new_entry->flags & UVM_MAP_QUANTUM); 1276 } 1277 1278 if (uobj) 1279 newetype = UVM_ET_OBJ; 1280 else 1281 newetype = 0; 1282 1283 if (flags & UVM_FLAG_COPYONW) { 1284 newetype |= UVM_ET_COPYONWRITE; 1285 if ((flags & UVM_FLAG_OVERLAY) == 0) 1286 newetype |= UVM_ET_NEEDSCOPY; 1287 } 1288 1289 /* 1290 * try and insert in map by extending previous entry, if possible. 1291 * XXX: we don't try and pull back the next entry. might be useful 1292 * for a stack, but we are currently allocating our stack in advance. 1293 */ 1294 1295 if (flags & UVM_FLAG_NOMERGE) 1296 goto nomerge; 1297 1298 if (prev_entry->end == start && 1299 prev_entry != &map->header && 1300 UVM_ET_ISCOMPATIBLE(prev_entry, newetype, uobj, meflagval, 1301 prot, maxprot, inherit, advice, 0)) { 1302 1303 if (uobj && prev_entry->offset + 1304 (prev_entry->end - prev_entry->start) != uoffset) 1305 goto forwardmerge; 1306 1307 /* 1308 * can't extend a shared amap. note: no need to lock amap to 1309 * look at refs since we don't care about its exact value. 1310 * if it is one (i.e. we have only reference) it will stay there 1311 */ 1312 1313 if (prev_entry->aref.ar_amap && 1314 amap_refs(prev_entry->aref.ar_amap) != 1) { 1315 goto forwardmerge; 1316 } 1317 1318 if (prev_entry->aref.ar_amap) { 1319 error = amap_extend(prev_entry, size, 1320 amapwaitflag | AMAP_EXTEND_FORWARDS); 1321 if (error) 1322 goto nomerge; 1323 } 1324 1325 if (kmap) 1326 UVMMAP_EVCNT_INCR(kbackmerge); 1327 else 1328 UVMMAP_EVCNT_INCR(ubackmerge); 1329 UVMHIST_LOG(maphist," starting back merge", 0, 0, 0, 0); 1330 1331 /* 1332 * drop our reference to uobj since we are extending a reference 1333 * that we already have (the ref count can not drop to zero). 1334 */ 1335 1336 if (uobj && uobj->pgops->pgo_detach) 1337 uobj->pgops->pgo_detach(uobj); 1338 1339 prev_entry->end += size; 1340 uvm_rb_fixup(map, prev_entry); 1341 1342 uvm_map_check(map, "map backmerged"); 1343 1344 UVMHIST_LOG(maphist,"<- done (via backmerge)!", 0, 0, 0, 0); 1345 merged++; 1346 } 1347 1348 forwardmerge: 1349 if (prev_entry->next->start == (start + size) && 1350 prev_entry->next != &map->header && 1351 UVM_ET_ISCOMPATIBLE(prev_entry->next, newetype, uobj, meflagval, 1352 prot, maxprot, inherit, advice, 0)) { 1353 1354 if (uobj && prev_entry->next->offset != uoffset + size) 1355 goto nomerge; 1356 1357 /* 1358 * can't extend a shared amap. note: no need to lock amap to 1359 * look at refs since we don't care about its exact value. 1360 * if it is one (i.e. we have only reference) it will stay there. 1361 * 1362 * note that we also can't merge two amaps, so if we 1363 * merged with the previous entry which has an amap, 1364 * and the next entry also has an amap, we give up. 1365 * 1366 * Interesting cases: 1367 * amap, new, amap -> give up second merge (single fwd extend) 1368 * amap, new, none -> double forward extend (extend again here) 1369 * none, new, amap -> double backward extend (done here) 1370 * uobj, new, amap -> single backward extend (done here) 1371 * 1372 * XXX should we attempt to deal with someone refilling 1373 * the deallocated region between two entries that are 1374 * backed by the same amap (ie, arefs is 2, "prev" and 1375 * "next" refer to it, and adding this allocation will 1376 * close the hole, thus restoring arefs to 1 and 1377 * deallocating the "next" vm_map_entry)? -- @@@ 1378 */ 1379 1380 if (prev_entry->next->aref.ar_amap && 1381 (amap_refs(prev_entry->next->aref.ar_amap) != 1 || 1382 (merged && prev_entry->aref.ar_amap))) { 1383 goto nomerge; 1384 } 1385 1386 if (merged) { 1387 /* 1388 * Try to extend the amap of the previous entry to 1389 * cover the next entry as well. If it doesn't work 1390 * just skip on, don't actually give up, since we've 1391 * already completed the back merge. 1392 */ 1393 if (prev_entry->aref.ar_amap) { 1394 if (amap_extend(prev_entry, 1395 prev_entry->next->end - 1396 prev_entry->next->start, 1397 amapwaitflag | AMAP_EXTEND_FORWARDS)) 1398 goto nomerge; 1399 } 1400 1401 /* 1402 * Try to extend the amap of the *next* entry 1403 * back to cover the new allocation *and* the 1404 * previous entry as well (the previous merge 1405 * didn't have an amap already otherwise we 1406 * wouldn't be checking here for an amap). If 1407 * it doesn't work just skip on, again, don't 1408 * actually give up, since we've already 1409 * completed the back merge. 1410 */ 1411 else if (prev_entry->next->aref.ar_amap) { 1412 if (amap_extend(prev_entry->next, 1413 prev_entry->end - 1414 prev_entry->start, 1415 amapwaitflag | AMAP_EXTEND_BACKWARDS)) 1416 goto nomerge; 1417 } 1418 } else { 1419 /* 1420 * Pull the next entry's amap backwards to cover this 1421 * new allocation. 1422 */ 1423 if (prev_entry->next->aref.ar_amap) { 1424 error = amap_extend(prev_entry->next, size, 1425 amapwaitflag | AMAP_EXTEND_BACKWARDS); 1426 if (error) 1427 goto nomerge; 1428 } 1429 } 1430 1431 if (merged) { 1432 if (kmap) { 1433 UVMMAP_EVCNT_DECR(kbackmerge); 1434 UVMMAP_EVCNT_INCR(kbimerge); 1435 } else { 1436 UVMMAP_EVCNT_DECR(ubackmerge); 1437 UVMMAP_EVCNT_INCR(ubimerge); 1438 } 1439 } else { 1440 if (kmap) 1441 UVMMAP_EVCNT_INCR(kforwmerge); 1442 else 1443 UVMMAP_EVCNT_INCR(uforwmerge); 1444 } 1445 UVMHIST_LOG(maphist," starting forward merge", 0, 0, 0, 0); 1446 1447 /* 1448 * drop our reference to uobj since we are extending a reference 1449 * that we already have (the ref count can not drop to zero). 1450 * (if merged, we've already detached) 1451 */ 1452 if (uobj && uobj->pgops->pgo_detach && !merged) 1453 uobj->pgops->pgo_detach(uobj); 1454 1455 if (merged) { 1456 dead = prev_entry->next; 1457 prev_entry->end = dead->end; 1458 uvm_map_entry_unlink(map, dead); 1459 if (dead->aref.ar_amap != NULL) { 1460 prev_entry->aref = dead->aref; 1461 dead->aref.ar_amap = NULL; 1462 } 1463 } else { 1464 prev_entry->next->start -= size; 1465 if (prev_entry != &map->header) 1466 uvm_rb_fixup(map, prev_entry); 1467 if (uobj) 1468 prev_entry->next->offset = uoffset; 1469 } 1470 1471 uvm_map_check(map, "map forwardmerged"); 1472 1473 UVMHIST_LOG(maphist,"<- done forwardmerge", 0, 0, 0, 0); 1474 merged++; 1475 } 1476 1477 nomerge: 1478 if (!merged) { 1479 UVMHIST_LOG(maphist," allocating new map entry", 0, 0, 0, 0); 1480 if (kmap) 1481 UVMMAP_EVCNT_INCR(knomerge); 1482 else 1483 UVMMAP_EVCNT_INCR(unomerge); 1484 1485 /* 1486 * allocate new entry and link it in. 1487 */ 1488 1489 if (new_entry == NULL) { 1490 new_entry = uvm_mapent_alloc(map, 1491 (flags & UVM_FLAG_NOWAIT)); 1492 if (__predict_false(new_entry == NULL)) { 1493 error = ENOMEM; 1494 goto done; 1495 } 1496 } 1497 new_entry->start = start; 1498 new_entry->end = new_entry->start + size; 1499 new_entry->object.uvm_obj = uobj; 1500 new_entry->offset = uoffset; 1501 1502 new_entry->etype = newetype; 1503 1504 if (flags & UVM_FLAG_NOMERGE) { 1505 new_entry->flags |= UVM_MAP_NOMERGE; 1506 } 1507 1508 new_entry->protection = prot; 1509 new_entry->max_protection = maxprot; 1510 new_entry->inheritance = inherit; 1511 new_entry->wired_count = 0; 1512 new_entry->advice = advice; 1513 if (flags & UVM_FLAG_OVERLAY) { 1514 1515 /* 1516 * to_add: for BSS we overallocate a little since we 1517 * are likely to extend 1518 */ 1519 1520 vaddr_t to_add = (flags & UVM_FLAG_AMAPPAD) ? 1521 UVM_AMAP_CHUNK << PAGE_SHIFT : 0; 1522 struct vm_amap *amap = amap_alloc(size, to_add, 1523 (flags & UVM_FLAG_NOWAIT)); 1524 if (__predict_false(amap == NULL)) { 1525 error = ENOMEM; 1526 goto done; 1527 } 1528 new_entry->aref.ar_pageoff = 0; 1529 new_entry->aref.ar_amap = amap; 1530 } else { 1531 new_entry->aref.ar_pageoff = 0; 1532 new_entry->aref.ar_amap = NULL; 1533 } 1534 uvm_map_entry_link(map, prev_entry, new_entry); 1535 1536 /* 1537 * Update the free space hint 1538 */ 1539 1540 if ((map->first_free == prev_entry) && 1541 (prev_entry->end >= new_entry->start)) 1542 map->first_free = new_entry; 1543 1544 new_entry = NULL; 1545 } 1546 1547 map->size += size; 1548 1549 UVMHIST_LOG(maphist,"<- done!", 0, 0, 0, 0); 1550 1551 error = 0; 1552 done: 1553 if ((flags & UVM_FLAG_QUANTUM) == 0) { 1554 /* 1555 * vmk_merged_entries is locked by the map's lock. 1556 */ 1557 vm_map_unlock(map); 1558 } 1559 if (new_entry && error == 0) { 1560 KDASSERT(merged); 1561 uvm_mapent_free_merged(map, new_entry); 1562 new_entry = NULL; 1563 } 1564 if (dead) { 1565 KDASSERT(merged); 1566 uvm_mapent_free_merged(map, dead); 1567 } 1568 if ((flags & UVM_FLAG_QUANTUM) != 0) { 1569 vm_map_unlock(map); 1570 } 1571 if (new_entry != NULL) { 1572 uvm_mapent_free(new_entry); 1573 } 1574 return error; 1575 } 1576 1577 /* 1578 * uvm_map_lookup_entry_bytree: lookup an entry in tree 1579 */ 1580 1581 static bool 1582 uvm_map_lookup_entry_bytree(struct vm_map *map, vaddr_t address, 1583 struct vm_map_entry **entry /* OUT */) 1584 { 1585 struct vm_map_entry *prev = &map->header; 1586 struct vm_map_entry *cur = RB_ROOT(&map->rbhead); 1587 1588 while (cur) { 1589 if (address >= cur->start) { 1590 if (address < cur->end) { 1591 *entry = cur; 1592 return true; 1593 } 1594 prev = cur; 1595 cur = RB_RIGHT(cur, rb_entry); 1596 } else 1597 cur = RB_LEFT(cur, rb_entry); 1598 } 1599 *entry = prev; 1600 return false; 1601 } 1602 1603 /* 1604 * uvm_map_lookup_entry: find map entry at or before an address 1605 * 1606 * => map must at least be read-locked by caller 1607 * => entry is returned in "entry" 1608 * => return value is true if address is in the returned entry 1609 */ 1610 1611 bool 1612 uvm_map_lookup_entry(struct vm_map *map, vaddr_t address, 1613 struct vm_map_entry **entry /* OUT */) 1614 { 1615 struct vm_map_entry *cur; 1616 bool use_tree = false; 1617 UVMHIST_FUNC("uvm_map_lookup_entry"); 1618 UVMHIST_CALLED(maphist); 1619 1620 UVMHIST_LOG(maphist,"(map=0x%x,addr=0x%x,ent=0x%x)", 1621 map, address, entry, 0); 1622 1623 /* 1624 * start looking either from the head of the 1625 * list, or from the hint. 1626 */ 1627 1628 cur = map->hint; 1629 1630 if (cur == &map->header) 1631 cur = cur->next; 1632 1633 UVMMAP_EVCNT_INCR(mlk_call); 1634 if (address >= cur->start) { 1635 1636 /* 1637 * go from hint to end of list. 1638 * 1639 * but first, make a quick check to see if 1640 * we are already looking at the entry we 1641 * want (which is usually the case). 1642 * note also that we don't need to save the hint 1643 * here... it is the same hint (unless we are 1644 * at the header, in which case the hint didn't 1645 * buy us anything anyway). 1646 */ 1647 1648 if (cur != &map->header && cur->end > address) { 1649 UVMMAP_EVCNT_INCR(mlk_hint); 1650 *entry = cur; 1651 UVMHIST_LOG(maphist,"<- got it via hint (0x%x)", 1652 cur, 0, 0, 0); 1653 uvm_mapent_check(*entry); 1654 return (true); 1655 } 1656 1657 if (map->nentries > 30) 1658 use_tree = true; 1659 } else { 1660 1661 /* 1662 * invalid hint. use tree. 1663 */ 1664 use_tree = true; 1665 } 1666 1667 uvm_map_check(map, __func__); 1668 1669 if (use_tree) { 1670 /* 1671 * Simple lookup in the tree. Happens when the hint is 1672 * invalid, or nentries reach a threshold. 1673 */ 1674 if (uvm_map_lookup_entry_bytree(map, address, entry)) { 1675 goto got; 1676 } else { 1677 goto failed; 1678 } 1679 } 1680 1681 /* 1682 * search linearly 1683 */ 1684 1685 while (cur != &map->header) { 1686 if (cur->end > address) { 1687 if (address >= cur->start) { 1688 /* 1689 * save this lookup for future 1690 * hints, and return 1691 */ 1692 1693 *entry = cur; 1694 got: 1695 SAVE_HINT(map, map->hint, *entry); 1696 UVMHIST_LOG(maphist,"<- search got it (0x%x)", 1697 cur, 0, 0, 0); 1698 KDASSERT((*entry)->start <= address); 1699 KDASSERT(address < (*entry)->end); 1700 uvm_mapent_check(*entry); 1701 return (true); 1702 } 1703 break; 1704 } 1705 cur = cur->next; 1706 } 1707 *entry = cur->prev; 1708 failed: 1709 SAVE_HINT(map, map->hint, *entry); 1710 UVMHIST_LOG(maphist,"<- failed!",0,0,0,0); 1711 KDASSERT((*entry) == &map->header || (*entry)->end <= address); 1712 KDASSERT((*entry)->next == &map->header || 1713 address < (*entry)->next->start); 1714 return (false); 1715 } 1716 1717 /* 1718 * See if the range between start and start + length fits in the gap 1719 * entry->next->start and entry->end. Returns 1 if fits, 0 if doesn't 1720 * fit, and -1 address wraps around. 1721 */ 1722 static int 1723 uvm_map_space_avail(vaddr_t *start, vsize_t length, voff_t uoffset, 1724 vsize_t align, int topdown, struct vm_map_entry *entry) 1725 { 1726 vaddr_t end; 1727 1728 #ifdef PMAP_PREFER 1729 /* 1730 * push start address forward as needed to avoid VAC alias problems. 1731 * we only do this if a valid offset is specified. 1732 */ 1733 1734 if (uoffset != UVM_UNKNOWN_OFFSET) 1735 PMAP_PREFER(uoffset, start, length, topdown); 1736 #endif 1737 if (align != 0) { 1738 if ((*start & (align - 1)) != 0) { 1739 if (topdown) 1740 *start &= ~(align - 1); 1741 else 1742 *start = roundup(*start, align); 1743 } 1744 /* 1745 * XXX Should we PMAP_PREFER() here again? 1746 * eh...i think we're okay 1747 */ 1748 } 1749 1750 /* 1751 * Find the end of the proposed new region. Be sure we didn't 1752 * wrap around the address; if so, we lose. Otherwise, if the 1753 * proposed new region fits before the next entry, we win. 1754 */ 1755 1756 end = *start + length; 1757 if (end < *start) 1758 return (-1); 1759 1760 if (entry->next->start >= end && *start >= entry->end) 1761 return (1); 1762 1763 return (0); 1764 } 1765 1766 /* 1767 * uvm_map_findspace: find "length" sized space in "map". 1768 * 1769 * => "hint" is a hint about where we want it, unless UVM_FLAG_FIXED is 1770 * set in "flags" (in which case we insist on using "hint"). 1771 * => "result" is VA returned 1772 * => uobj/uoffset are to be used to handle VAC alignment, if required 1773 * => if "align" is non-zero, we attempt to align to that value. 1774 * => caller must at least have read-locked map 1775 * => returns NULL on failure, or pointer to prev. map entry if success 1776 * => note this is a cross between the old vm_map_findspace and vm_map_find 1777 */ 1778 1779 struct vm_map_entry * 1780 uvm_map_findspace(struct vm_map *map, vaddr_t hint, vsize_t length, 1781 vaddr_t *result /* OUT */, struct uvm_object *uobj, voff_t uoffset, 1782 vsize_t align, int flags) 1783 { 1784 struct vm_map_entry *entry; 1785 struct vm_map_entry *child, *prev, *tmp; 1786 vaddr_t orig_hint; 1787 const int topdown = map->flags & VM_MAP_TOPDOWN; 1788 UVMHIST_FUNC("uvm_map_findspace"); 1789 UVMHIST_CALLED(maphist); 1790 1791 UVMHIST_LOG(maphist, "(map=0x%x, hint=0x%x, len=%d, flags=0x%x)", 1792 map, hint, length, flags); 1793 KASSERT((align & (align - 1)) == 0); 1794 KASSERT((flags & UVM_FLAG_FIXED) == 0 || align == 0); 1795 1796 uvm_map_check(map, "map_findspace entry"); 1797 1798 /* 1799 * remember the original hint. if we are aligning, then we 1800 * may have to try again with no alignment constraint if 1801 * we fail the first time. 1802 */ 1803 1804 orig_hint = hint; 1805 if (hint < vm_map_min(map)) { /* check ranges ... */ 1806 if (flags & UVM_FLAG_FIXED) { 1807 UVMHIST_LOG(maphist,"<- VA below map range",0,0,0,0); 1808 return (NULL); 1809 } 1810 hint = vm_map_min(map); 1811 } 1812 if (hint > vm_map_max(map)) { 1813 UVMHIST_LOG(maphist,"<- VA 0x%x > range [0x%x->0x%x]", 1814 hint, vm_map_min(map), vm_map_max(map), 0); 1815 return (NULL); 1816 } 1817 1818 /* 1819 * Look for the first possible address; if there's already 1820 * something at this address, we have to start after it. 1821 */ 1822 1823 /* 1824 * @@@: there are four, no, eight cases to consider. 1825 * 1826 * 0: found, fixed, bottom up -> fail 1827 * 1: found, fixed, top down -> fail 1828 * 2: found, not fixed, bottom up -> start after entry->end, 1829 * loop up 1830 * 3: found, not fixed, top down -> start before entry->start, 1831 * loop down 1832 * 4: not found, fixed, bottom up -> check entry->next->start, fail 1833 * 5: not found, fixed, top down -> check entry->next->start, fail 1834 * 6: not found, not fixed, bottom up -> check entry->next->start, 1835 * loop up 1836 * 7: not found, not fixed, top down -> check entry->next->start, 1837 * loop down 1838 * 1839 * as you can see, it reduces to roughly five cases, and that 1840 * adding top down mapping only adds one unique case (without 1841 * it, there would be four cases). 1842 */ 1843 1844 if ((flags & UVM_FLAG_FIXED) == 0 && hint == vm_map_min(map)) { 1845 entry = map->first_free; 1846 } else { 1847 if (uvm_map_lookup_entry(map, hint, &entry)) { 1848 /* "hint" address already in use ... */ 1849 if (flags & UVM_FLAG_FIXED) { 1850 UVMHIST_LOG(maphist, "<- fixed & VA in use", 1851 0, 0, 0, 0); 1852 return (NULL); 1853 } 1854 if (topdown) 1855 /* Start from lower gap. */ 1856 entry = entry->prev; 1857 } else if (flags & UVM_FLAG_FIXED) { 1858 if (entry->next->start >= hint + length && 1859 hint + length > hint) 1860 goto found; 1861 1862 /* "hint" address is gap but too small */ 1863 UVMHIST_LOG(maphist, "<- fixed mapping failed", 1864 0, 0, 0, 0); 1865 return (NULL); /* only one shot at it ... */ 1866 } else { 1867 /* 1868 * See if given hint fits in this gap. 1869 */ 1870 switch (uvm_map_space_avail(&hint, length, 1871 uoffset, align, topdown, entry)) { 1872 case 1: 1873 goto found; 1874 case -1: 1875 goto wraparound; 1876 } 1877 1878 if (topdown) { 1879 /* 1880 * Still there is a chance to fit 1881 * if hint > entry->end. 1882 */ 1883 } else { 1884 /* Start from higher gap. */ 1885 entry = entry->next; 1886 if (entry == &map->header) 1887 goto notfound; 1888 goto nextgap; 1889 } 1890 } 1891 } 1892 1893 /* 1894 * Note that all UVM_FLAGS_FIXED case is already handled. 1895 */ 1896 KDASSERT((flags & UVM_FLAG_FIXED) == 0); 1897 1898 /* Try to find the space in the red-black tree */ 1899 1900 /* Check slot before any entry */ 1901 hint = topdown ? entry->next->start - length : entry->end; 1902 switch (uvm_map_space_avail(&hint, length, uoffset, align, 1903 topdown, entry)) { 1904 case 1: 1905 goto found; 1906 case -1: 1907 goto wraparound; 1908 } 1909 1910 nextgap: 1911 KDASSERT((flags & UVM_FLAG_FIXED) == 0); 1912 /* If there is not enough space in the whole tree, we fail */ 1913 tmp = RB_ROOT(&map->rbhead); 1914 if (tmp == NULL || tmp->space < length) 1915 goto notfound; 1916 1917 prev = NULL; /* previous candidate */ 1918 1919 /* Find an entry close to hint that has enough space */ 1920 for (; tmp;) { 1921 KASSERT(tmp->next->start == tmp->end + tmp->ownspace); 1922 if (topdown) { 1923 if (tmp->next->start < hint + length && 1924 (prev == NULL || tmp->end > prev->end)) { 1925 if (tmp->ownspace >= length) 1926 prev = tmp; 1927 else if ((child = RB_LEFT(tmp, rb_entry)) 1928 != NULL && child->space >= length) 1929 prev = tmp; 1930 } 1931 } else { 1932 if (tmp->end >= hint && 1933 (prev == NULL || tmp->end < prev->end)) { 1934 if (tmp->ownspace >= length) 1935 prev = tmp; 1936 else if ((child = RB_RIGHT(tmp, rb_entry)) 1937 != NULL && child->space >= length) 1938 prev = tmp; 1939 } 1940 } 1941 if (tmp->next->start < hint + length) 1942 child = RB_RIGHT(tmp, rb_entry); 1943 else if (tmp->end > hint) 1944 child = RB_LEFT(tmp, rb_entry); 1945 else { 1946 if (tmp->ownspace >= length) 1947 break; 1948 if (topdown) 1949 child = RB_LEFT(tmp, rb_entry); 1950 else 1951 child = RB_RIGHT(tmp, rb_entry); 1952 } 1953 if (child == NULL || child->space < length) 1954 break; 1955 tmp = child; 1956 } 1957 1958 if (tmp != NULL && tmp->start < hint && hint < tmp->next->start) { 1959 /* 1960 * Check if the entry that we found satifies the 1961 * space requirement 1962 */ 1963 if (topdown) { 1964 if (hint > tmp->next->start - length) 1965 hint = tmp->next->start - length; 1966 } else { 1967 if (hint < tmp->end) 1968 hint = tmp->end; 1969 } 1970 switch (uvm_map_space_avail(&hint, length, uoffset, align, 1971 topdown, tmp)) { 1972 case 1: 1973 entry = tmp; 1974 goto found; 1975 case -1: 1976 goto wraparound; 1977 } 1978 if (tmp->ownspace >= length) 1979 goto listsearch; 1980 } 1981 if (prev == NULL) 1982 goto notfound; 1983 1984 if (topdown) { 1985 KASSERT(orig_hint >= prev->next->start - length || 1986 prev->next->start - length > prev->next->start); 1987 hint = prev->next->start - length; 1988 } else { 1989 KASSERT(orig_hint <= prev->end); 1990 hint = prev->end; 1991 } 1992 switch (uvm_map_space_avail(&hint, length, uoffset, align, 1993 topdown, prev)) { 1994 case 1: 1995 entry = prev; 1996 goto found; 1997 case -1: 1998 goto wraparound; 1999 } 2000 if (prev->ownspace >= length) 2001 goto listsearch; 2002 2003 if (topdown) 2004 tmp = RB_LEFT(prev, rb_entry); 2005 else 2006 tmp = RB_RIGHT(prev, rb_entry); 2007 for (;;) { 2008 KASSERT(tmp && tmp->space >= length); 2009 if (topdown) 2010 child = RB_RIGHT(tmp, rb_entry); 2011 else 2012 child = RB_LEFT(tmp, rb_entry); 2013 if (child && child->space >= length) { 2014 tmp = child; 2015 continue; 2016 } 2017 if (tmp->ownspace >= length) 2018 break; 2019 if (topdown) 2020 tmp = RB_LEFT(tmp, rb_entry); 2021 else 2022 tmp = RB_RIGHT(tmp, rb_entry); 2023 } 2024 2025 if (topdown) { 2026 KASSERT(orig_hint >= tmp->next->start - length || 2027 tmp->next->start - length > tmp->next->start); 2028 hint = tmp->next->start - length; 2029 } else { 2030 KASSERT(orig_hint <= tmp->end); 2031 hint = tmp->end; 2032 } 2033 switch (uvm_map_space_avail(&hint, length, uoffset, align, 2034 topdown, tmp)) { 2035 case 1: 2036 entry = tmp; 2037 goto found; 2038 case -1: 2039 goto wraparound; 2040 } 2041 2042 /* 2043 * The tree fails to find an entry because of offset or alignment 2044 * restrictions. Search the list instead. 2045 */ 2046 listsearch: 2047 /* 2048 * Look through the rest of the map, trying to fit a new region in 2049 * the gap between existing regions, or after the very last region. 2050 * note: entry->end = base VA of current gap, 2051 * entry->next->start = VA of end of current gap 2052 */ 2053 2054 for (;;) { 2055 /* Update hint for current gap. */ 2056 hint = topdown ? entry->next->start - length : entry->end; 2057 2058 /* See if it fits. */ 2059 switch (uvm_map_space_avail(&hint, length, uoffset, align, 2060 topdown, entry)) { 2061 case 1: 2062 goto found; 2063 case -1: 2064 goto wraparound; 2065 } 2066 2067 /* Advance to next/previous gap */ 2068 if (topdown) { 2069 if (entry == &map->header) { 2070 UVMHIST_LOG(maphist, "<- failed (off start)", 2071 0,0,0,0); 2072 goto notfound; 2073 } 2074 entry = entry->prev; 2075 } else { 2076 entry = entry->next; 2077 if (entry == &map->header) { 2078 UVMHIST_LOG(maphist, "<- failed (off end)", 2079 0,0,0,0); 2080 goto notfound; 2081 } 2082 } 2083 } 2084 2085 found: 2086 SAVE_HINT(map, map->hint, entry); 2087 *result = hint; 2088 UVMHIST_LOG(maphist,"<- got it! (result=0x%x)", hint, 0,0,0); 2089 KASSERT( topdown || hint >= orig_hint); 2090 KASSERT(!topdown || hint <= orig_hint); 2091 KASSERT(entry->end <= hint); 2092 KASSERT(hint + length <= entry->next->start); 2093 return (entry); 2094 2095 wraparound: 2096 UVMHIST_LOG(maphist, "<- failed (wrap around)", 0,0,0,0); 2097 2098 return (NULL); 2099 2100 notfound: 2101 UVMHIST_LOG(maphist, "<- failed (notfound)", 0,0,0,0); 2102 2103 return (NULL); 2104 } 2105 2106 /* 2107 * U N M A P - m a i n h e l p e r f u n c t i o n s 2108 */ 2109 2110 /* 2111 * uvm_unmap_remove: remove mappings from a vm_map (from "start" up to "stop") 2112 * 2113 * => caller must check alignment and size 2114 * => map must be locked by caller 2115 * => we return a list of map entries that we've remove from the map 2116 * in "entry_list" 2117 */ 2118 2119 void 2120 uvm_unmap_remove(struct vm_map *map, vaddr_t start, vaddr_t end, 2121 struct vm_map_entry **entry_list /* OUT */, 2122 struct uvm_mapent_reservation *umr, int flags) 2123 { 2124 struct vm_map_entry *entry, *first_entry, *next; 2125 vaddr_t len; 2126 UVMHIST_FUNC("uvm_unmap_remove"); UVMHIST_CALLED(maphist); 2127 2128 UVMHIST_LOG(maphist,"(map=0x%x, start=0x%x, end=0x%x)", 2129 map, start, end, 0); 2130 VM_MAP_RANGE_CHECK(map, start, end); 2131 2132 uvm_map_check(map, "unmap_remove entry"); 2133 2134 /* 2135 * find first entry 2136 */ 2137 2138 if (uvm_map_lookup_entry(map, start, &first_entry) == true) { 2139 /* clip and go... */ 2140 entry = first_entry; 2141 UVM_MAP_CLIP_START(map, entry, start, umr); 2142 /* critical! prevents stale hint */ 2143 SAVE_HINT(map, entry, entry->prev); 2144 } else { 2145 entry = first_entry->next; 2146 } 2147 2148 /* 2149 * Save the free space hint 2150 */ 2151 2152 if (map->first_free != &map->header && map->first_free->start >= start) 2153 map->first_free = entry->prev; 2154 2155 /* 2156 * note: we now re-use first_entry for a different task. we remove 2157 * a number of map entries from the map and save them in a linked 2158 * list headed by "first_entry". once we remove them from the map 2159 * the caller should unlock the map and drop the references to the 2160 * backing objects [c.f. uvm_unmap_detach]. the object is to 2161 * separate unmapping from reference dropping. why? 2162 * [1] the map has to be locked for unmapping 2163 * [2] the map need not be locked for reference dropping 2164 * [3] dropping references may trigger pager I/O, and if we hit 2165 * a pager that does synchronous I/O we may have to wait for it. 2166 * [4] we would like all waiting for I/O to occur with maps unlocked 2167 * so that we don't block other threads. 2168 */ 2169 2170 first_entry = NULL; 2171 *entry_list = NULL; 2172 2173 /* 2174 * break up the area into map entry sized regions and unmap. note 2175 * that all mappings have to be removed before we can even consider 2176 * dropping references to amaps or VM objects (otherwise we could end 2177 * up with a mapping to a page on the free list which would be very bad) 2178 */ 2179 2180 while ((entry != &map->header) && (entry->start < end)) { 2181 KASSERT((entry->flags & UVM_MAP_FIRST) == 0); 2182 2183 UVM_MAP_CLIP_END(map, entry, end, umr); 2184 next = entry->next; 2185 len = entry->end - entry->start; 2186 2187 /* 2188 * unwire before removing addresses from the pmap; otherwise 2189 * unwiring will put the entries back into the pmap (XXX). 2190 */ 2191 2192 if (VM_MAPENT_ISWIRED(entry)) { 2193 uvm_map_entry_unwire(map, entry); 2194 } 2195 if (flags & UVM_FLAG_VAONLY) { 2196 2197 /* nothing */ 2198 2199 } else if ((map->flags & VM_MAP_PAGEABLE) == 0) { 2200 2201 /* 2202 * if the map is non-pageable, any pages mapped there 2203 * must be wired and entered with pmap_kenter_pa(), 2204 * and we should free any such pages immediately. 2205 * this is mostly used for kmem_map and mb_map. 2206 */ 2207 2208 if ((entry->flags & UVM_MAP_KMAPENT) == 0) { 2209 uvm_km_pgremove_intrsafe(entry->start, 2210 entry->end); 2211 pmap_kremove(entry->start, len); 2212 } 2213 } else if (UVM_ET_ISOBJ(entry) && 2214 UVM_OBJ_IS_KERN_OBJECT(entry->object.uvm_obj)) { 2215 KASSERT(vm_map_pmap(map) == pmap_kernel()); 2216 2217 /* 2218 * note: kernel object mappings are currently used in 2219 * two ways: 2220 * [1] "normal" mappings of pages in the kernel object 2221 * [2] uvm_km_valloc'd allocations in which we 2222 * pmap_enter in some non-kernel-object page 2223 * (e.g. vmapbuf). 2224 * 2225 * for case [1], we need to remove the mapping from 2226 * the pmap and then remove the page from the kernel 2227 * object (because, once pages in a kernel object are 2228 * unmapped they are no longer needed, unlike, say, 2229 * a vnode where you might want the data to persist 2230 * until flushed out of a queue). 2231 * 2232 * for case [2], we need to remove the mapping from 2233 * the pmap. there shouldn't be any pages at the 2234 * specified offset in the kernel object [but it 2235 * doesn't hurt to call uvm_km_pgremove just to be 2236 * safe?] 2237 * 2238 * uvm_km_pgremove currently does the following: 2239 * for pages in the kernel object in range: 2240 * - drops the swap slot 2241 * - uvm_pagefree the page 2242 */ 2243 2244 /* 2245 * remove mappings from pmap and drop the pages 2246 * from the object. offsets are always relative 2247 * to vm_map_min(kernel_map). 2248 */ 2249 2250 pmap_remove(pmap_kernel(), entry->start, 2251 entry->start + len); 2252 uvm_km_pgremove(entry->start, entry->end); 2253 2254 /* 2255 * null out kernel_object reference, we've just 2256 * dropped it 2257 */ 2258 2259 entry->etype &= ~UVM_ET_OBJ; 2260 entry->object.uvm_obj = NULL; 2261 } else if (UVM_ET_ISOBJ(entry) || entry->aref.ar_amap) { 2262 2263 /* 2264 * remove mappings the standard way. 2265 */ 2266 2267 pmap_remove(map->pmap, entry->start, entry->end); 2268 } 2269 2270 #if defined(DEBUG) 2271 if ((entry->flags & UVM_MAP_KMAPENT) == 0) { 2272 2273 /* 2274 * check if there's remaining mapping, 2275 * which is a bug in caller. 2276 */ 2277 2278 vaddr_t va; 2279 for (va = entry->start; va < entry->end; 2280 va += PAGE_SIZE) { 2281 if (pmap_extract(vm_map_pmap(map), va, NULL)) { 2282 panic("uvm_unmap_remove: has mapping"); 2283 } 2284 } 2285 2286 if (VM_MAP_IS_KERNEL(map)) { 2287 uvm_km_check_empty(entry->start, entry->end, 2288 (map->flags & VM_MAP_INTRSAFE) != 0); 2289 } 2290 } 2291 #endif /* defined(DEBUG) */ 2292 2293 /* 2294 * remove entry from map and put it on our list of entries 2295 * that we've nuked. then go to next entry. 2296 */ 2297 2298 UVMHIST_LOG(maphist, " removed map entry 0x%x", entry, 0, 0,0); 2299 2300 /* critical! prevents stale hint */ 2301 SAVE_HINT(map, entry, entry->prev); 2302 2303 uvm_map_entry_unlink(map, entry); 2304 KASSERT(map->size >= len); 2305 map->size -= len; 2306 entry->prev = NULL; 2307 entry->next = first_entry; 2308 first_entry = entry; 2309 entry = next; 2310 } 2311 if ((map->flags & VM_MAP_DYING) == 0) { 2312 pmap_update(vm_map_pmap(map)); 2313 } 2314 2315 uvm_map_check(map, "unmap_remove leave"); 2316 2317 /* 2318 * now we've cleaned up the map and are ready for the caller to drop 2319 * references to the mapped objects. 2320 */ 2321 2322 *entry_list = first_entry; 2323 UVMHIST_LOG(maphist,"<- done!", 0, 0, 0, 0); 2324 2325 if (map->flags & VM_MAP_WANTVA) { 2326 mutex_enter(&map->misc_lock); 2327 map->flags &= ~VM_MAP_WANTVA; 2328 cv_broadcast(&map->cv); 2329 mutex_exit(&map->misc_lock); 2330 } 2331 } 2332 2333 /* 2334 * uvm_unmap_detach: drop references in a chain of map entries 2335 * 2336 * => we will free the map entries as we traverse the list. 2337 */ 2338 2339 void 2340 uvm_unmap_detach(struct vm_map_entry *first_entry, int flags) 2341 { 2342 struct vm_map_entry *next_entry; 2343 UVMHIST_FUNC("uvm_unmap_detach"); UVMHIST_CALLED(maphist); 2344 2345 while (first_entry) { 2346 KASSERT(!VM_MAPENT_ISWIRED(first_entry)); 2347 UVMHIST_LOG(maphist, 2348 " detach 0x%x: amap=0x%x, obj=0x%x, submap?=%d", 2349 first_entry, first_entry->aref.ar_amap, 2350 first_entry->object.uvm_obj, 2351 UVM_ET_ISSUBMAP(first_entry)); 2352 2353 /* 2354 * drop reference to amap, if we've got one 2355 */ 2356 2357 if (first_entry->aref.ar_amap) 2358 uvm_map_unreference_amap(first_entry, flags); 2359 2360 /* 2361 * drop reference to our backing object, if we've got one 2362 */ 2363 2364 KASSERT(!UVM_ET_ISSUBMAP(first_entry)); 2365 if (UVM_ET_ISOBJ(first_entry) && 2366 first_entry->object.uvm_obj->pgops->pgo_detach) { 2367 (*first_entry->object.uvm_obj->pgops->pgo_detach) 2368 (first_entry->object.uvm_obj); 2369 } 2370 next_entry = first_entry->next; 2371 uvm_mapent_free(first_entry); 2372 first_entry = next_entry; 2373 } 2374 UVMHIST_LOG(maphist, "<- done", 0,0,0,0); 2375 } 2376 2377 /* 2378 * E X T R A C T I O N F U N C T I O N S 2379 */ 2380 2381 /* 2382 * uvm_map_reserve: reserve space in a vm_map for future use. 2383 * 2384 * => we reserve space in a map by putting a dummy map entry in the 2385 * map (dummy means obj=NULL, amap=NULL, prot=VM_PROT_NONE) 2386 * => map should be unlocked (we will write lock it) 2387 * => we return true if we were able to reserve space 2388 * => XXXCDC: should be inline? 2389 */ 2390 2391 int 2392 uvm_map_reserve(struct vm_map *map, vsize_t size, 2393 vaddr_t offset /* hint for pmap_prefer */, 2394 vsize_t align /* alignment */, 2395 vaddr_t *raddr /* IN:hint, OUT: reserved VA */, 2396 uvm_flag_t flags /* UVM_FLAG_FIXED or 0 */) 2397 { 2398 UVMHIST_FUNC("uvm_map_reserve"); UVMHIST_CALLED(maphist); 2399 2400 UVMHIST_LOG(maphist, "(map=0x%x, size=0x%x, offset=0x%x,addr=0x%x)", 2401 map,size,offset,raddr); 2402 2403 size = round_page(size); 2404 2405 /* 2406 * reserve some virtual space. 2407 */ 2408 2409 if (uvm_map(map, raddr, size, NULL, offset, align, 2410 UVM_MAPFLAG(UVM_PROT_NONE, UVM_PROT_NONE, UVM_INH_NONE, 2411 UVM_ADV_RANDOM, UVM_FLAG_NOMERGE|flags)) != 0) { 2412 UVMHIST_LOG(maphist, "<- done (no VM)", 0,0,0,0); 2413 return (false); 2414 } 2415 2416 UVMHIST_LOG(maphist, "<- done (*raddr=0x%x)", *raddr,0,0,0); 2417 return (true); 2418 } 2419 2420 /* 2421 * uvm_map_replace: replace a reserved (blank) area of memory with 2422 * real mappings. 2423 * 2424 * => caller must WRITE-LOCK the map 2425 * => we return true if replacement was a success 2426 * => we expect the newents chain to have nnewents entrys on it and 2427 * we expect newents->prev to point to the last entry on the list 2428 * => note newents is allowed to be NULL 2429 */ 2430 2431 int 2432 uvm_map_replace(struct vm_map *map, vaddr_t start, vaddr_t end, 2433 struct vm_map_entry *newents, int nnewents, struct vm_map_entry **oldentryp) 2434 { 2435 struct vm_map_entry *oldent, *last; 2436 2437 uvm_map_check(map, "map_replace entry"); 2438 2439 /* 2440 * first find the blank map entry at the specified address 2441 */ 2442 2443 if (!uvm_map_lookup_entry(map, start, &oldent)) { 2444 return (false); 2445 } 2446 2447 /* 2448 * check to make sure we have a proper blank entry 2449 */ 2450 2451 if (end < oldent->end && !VM_MAP_USE_KMAPENT(map)) { 2452 UVM_MAP_CLIP_END(map, oldent, end, NULL); 2453 } 2454 if (oldent->start != start || oldent->end != end || 2455 oldent->object.uvm_obj != NULL || oldent->aref.ar_amap != NULL) { 2456 return (false); 2457 } 2458 2459 #ifdef DIAGNOSTIC 2460 2461 /* 2462 * sanity check the newents chain 2463 */ 2464 2465 { 2466 struct vm_map_entry *tmpent = newents; 2467 int nent = 0; 2468 vaddr_t cur = start; 2469 2470 while (tmpent) { 2471 nent++; 2472 if (tmpent->start < cur) 2473 panic("uvm_map_replace1"); 2474 if (tmpent->start > tmpent->end || tmpent->end > end) { 2475 printf("tmpent->start=0x%lx, tmpent->end=0x%lx, end=0x%lx\n", 2476 tmpent->start, tmpent->end, end); 2477 panic("uvm_map_replace2"); 2478 } 2479 cur = tmpent->end; 2480 if (tmpent->next) { 2481 if (tmpent->next->prev != tmpent) 2482 panic("uvm_map_replace3"); 2483 } else { 2484 if (newents->prev != tmpent) 2485 panic("uvm_map_replace4"); 2486 } 2487 tmpent = tmpent->next; 2488 } 2489 if (nent != nnewents) 2490 panic("uvm_map_replace5"); 2491 } 2492 #endif 2493 2494 /* 2495 * map entry is a valid blank! replace it. (this does all the 2496 * work of map entry link/unlink...). 2497 */ 2498 2499 if (newents) { 2500 last = newents->prev; 2501 2502 /* critical: flush stale hints out of map */ 2503 SAVE_HINT(map, map->hint, newents); 2504 if (map->first_free == oldent) 2505 map->first_free = last; 2506 2507 last->next = oldent->next; 2508 last->next->prev = last; 2509 2510 /* Fix RB tree */ 2511 uvm_rb_remove(map, oldent); 2512 2513 newents->prev = oldent->prev; 2514 newents->prev->next = newents; 2515 map->nentries = map->nentries + (nnewents - 1); 2516 2517 /* Fixup the RB tree */ 2518 { 2519 int i; 2520 struct vm_map_entry *tmp; 2521 2522 tmp = newents; 2523 for (i = 0; i < nnewents && tmp; i++) { 2524 uvm_rb_insert(map, tmp); 2525 tmp = tmp->next; 2526 } 2527 } 2528 } else { 2529 /* NULL list of new entries: just remove the old one */ 2530 clear_hints(map, oldent); 2531 uvm_map_entry_unlink(map, oldent); 2532 } 2533 2534 uvm_map_check(map, "map_replace leave"); 2535 2536 /* 2537 * now we can free the old blank entry and return. 2538 */ 2539 2540 *oldentryp = oldent; 2541 return (true); 2542 } 2543 2544 /* 2545 * uvm_map_extract: extract a mapping from a map and put it somewhere 2546 * (maybe removing the old mapping) 2547 * 2548 * => maps should be unlocked (we will write lock them) 2549 * => returns 0 on success, error code otherwise 2550 * => start must be page aligned 2551 * => len must be page sized 2552 * => flags: 2553 * UVM_EXTRACT_REMOVE: remove mappings from srcmap 2554 * UVM_EXTRACT_CONTIG: abort if unmapped area (advisory only) 2555 * UVM_EXTRACT_QREF: for a temporary extraction do quick obj refs 2556 * UVM_EXTRACT_FIXPROT: set prot to maxprot as we go 2557 * >>>NOTE: if you set REMOVE, you are not allowed to use CONTIG or QREF!<<< 2558 * >>>NOTE: QREF's must be unmapped via the QREF path, thus should only 2559 * be used from within the kernel in a kernel level map <<< 2560 */ 2561 2562 int 2563 uvm_map_extract(struct vm_map *srcmap, vaddr_t start, vsize_t len, 2564 struct vm_map *dstmap, vaddr_t *dstaddrp, int flags) 2565 { 2566 vaddr_t dstaddr, end, newend, oldoffset, fudge, orig_fudge; 2567 struct vm_map_entry *chain, *endchain, *entry, *orig_entry, *newentry, 2568 *deadentry, *oldentry; 2569 struct vm_map_entry *resentry = NULL; /* a dummy reservation entry */ 2570 vsize_t elen; 2571 int nchain, error, copy_ok; 2572 UVMHIST_FUNC("uvm_map_extract"); UVMHIST_CALLED(maphist); 2573 2574 UVMHIST_LOG(maphist,"(srcmap=0x%x,start=0x%x, len=0x%x", srcmap, start, 2575 len,0); 2576 UVMHIST_LOG(maphist," ...,dstmap=0x%x, flags=0x%x)", dstmap,flags,0,0); 2577 2578 uvm_map_check(srcmap, "map_extract src enter"); 2579 uvm_map_check(dstmap, "map_extract dst enter"); 2580 2581 /* 2582 * step 0: sanity check: start must be on a page boundary, length 2583 * must be page sized. can't ask for CONTIG/QREF if you asked for 2584 * REMOVE. 2585 */ 2586 2587 KASSERT((start & PAGE_MASK) == 0 && (len & PAGE_MASK) == 0); 2588 KASSERT((flags & UVM_EXTRACT_REMOVE) == 0 || 2589 (flags & (UVM_EXTRACT_CONTIG|UVM_EXTRACT_QREF)) == 0); 2590 2591 /* 2592 * step 1: reserve space in the target map for the extracted area 2593 */ 2594 2595 if ((flags & UVM_EXTRACT_RESERVED) == 0) { 2596 dstaddr = vm_map_min(dstmap); 2597 if (!uvm_map_reserve(dstmap, len, start, 0, &dstaddr, 0)) 2598 return (ENOMEM); 2599 *dstaddrp = dstaddr; /* pass address back to caller */ 2600 UVMHIST_LOG(maphist, " dstaddr=0x%x", dstaddr,0,0,0); 2601 } else { 2602 dstaddr = *dstaddrp; 2603 } 2604 2605 /* 2606 * step 2: setup for the extraction process loop by init'ing the 2607 * map entry chain, locking src map, and looking up the first useful 2608 * entry in the map. 2609 */ 2610 2611 end = start + len; 2612 newend = dstaddr + len; 2613 chain = endchain = NULL; 2614 nchain = 0; 2615 vm_map_lock(srcmap); 2616 2617 if (uvm_map_lookup_entry(srcmap, start, &entry)) { 2618 2619 /* "start" is within an entry */ 2620 if (flags & UVM_EXTRACT_QREF) { 2621 2622 /* 2623 * for quick references we don't clip the entry, so 2624 * the entry may map space "before" the starting 2625 * virtual address... this is the "fudge" factor 2626 * (which can be non-zero only the first time 2627 * through the "while" loop in step 3). 2628 */ 2629 2630 fudge = start - entry->start; 2631 } else { 2632 2633 /* 2634 * normal reference: we clip the map to fit (thus 2635 * fudge is zero) 2636 */ 2637 2638 UVM_MAP_CLIP_START(srcmap, entry, start, NULL); 2639 SAVE_HINT(srcmap, srcmap->hint, entry->prev); 2640 fudge = 0; 2641 } 2642 } else { 2643 2644 /* "start" is not within an entry ... skip to next entry */ 2645 if (flags & UVM_EXTRACT_CONTIG) { 2646 error = EINVAL; 2647 goto bad; /* definite hole here ... */ 2648 } 2649 2650 entry = entry->next; 2651 fudge = 0; 2652 } 2653 2654 /* save values from srcmap for step 6 */ 2655 orig_entry = entry; 2656 orig_fudge = fudge; 2657 2658 /* 2659 * step 3: now start looping through the map entries, extracting 2660 * as we go. 2661 */ 2662 2663 while (entry->start < end && entry != &srcmap->header) { 2664 2665 /* if we are not doing a quick reference, clip it */ 2666 if ((flags & UVM_EXTRACT_QREF) == 0) 2667 UVM_MAP_CLIP_END(srcmap, entry, end, NULL); 2668 2669 /* clear needs_copy (allow chunking) */ 2670 if (UVM_ET_ISNEEDSCOPY(entry)) { 2671 amap_copy(srcmap, entry, 2672 AMAP_COPY_NOWAIT|AMAP_COPY_NOMERGE, start, end); 2673 if (UVM_ET_ISNEEDSCOPY(entry)) { /* failed? */ 2674 error = ENOMEM; 2675 goto bad; 2676 } 2677 2678 /* amap_copy could clip (during chunk)! update fudge */ 2679 if (fudge) { 2680 fudge = start - entry->start; 2681 orig_fudge = fudge; 2682 } 2683 } 2684 2685 /* calculate the offset of this from "start" */ 2686 oldoffset = (entry->start + fudge) - start; 2687 2688 /* allocate a new map entry */ 2689 newentry = uvm_mapent_alloc(dstmap, 0); 2690 if (newentry == NULL) { 2691 error = ENOMEM; 2692 goto bad; 2693 } 2694 2695 /* set up new map entry */ 2696 newentry->next = NULL; 2697 newentry->prev = endchain; 2698 newentry->start = dstaddr + oldoffset; 2699 newentry->end = 2700 newentry->start + (entry->end - (entry->start + fudge)); 2701 if (newentry->end > newend || newentry->end < newentry->start) 2702 newentry->end = newend; 2703 newentry->object.uvm_obj = entry->object.uvm_obj; 2704 if (newentry->object.uvm_obj) { 2705 if (newentry->object.uvm_obj->pgops->pgo_reference) 2706 newentry->object.uvm_obj->pgops-> 2707 pgo_reference(newentry->object.uvm_obj); 2708 newentry->offset = entry->offset + fudge; 2709 } else { 2710 newentry->offset = 0; 2711 } 2712 newentry->etype = entry->etype; 2713 newentry->protection = (flags & UVM_EXTRACT_FIXPROT) ? 2714 entry->max_protection : entry->protection; 2715 newentry->max_protection = entry->max_protection; 2716 newentry->inheritance = entry->inheritance; 2717 newentry->wired_count = 0; 2718 newentry->aref.ar_amap = entry->aref.ar_amap; 2719 if (newentry->aref.ar_amap) { 2720 newentry->aref.ar_pageoff = 2721 entry->aref.ar_pageoff + (fudge >> PAGE_SHIFT); 2722 uvm_map_reference_amap(newentry, AMAP_SHARED | 2723 ((flags & UVM_EXTRACT_QREF) ? AMAP_REFALL : 0)); 2724 } else { 2725 newentry->aref.ar_pageoff = 0; 2726 } 2727 newentry->advice = entry->advice; 2728 if ((flags & UVM_EXTRACT_QREF) != 0) { 2729 newentry->flags |= UVM_MAP_NOMERGE; 2730 } 2731 2732 /* now link it on the chain */ 2733 nchain++; 2734 if (endchain == NULL) { 2735 chain = endchain = newentry; 2736 } else { 2737 endchain->next = newentry; 2738 endchain = newentry; 2739 } 2740 2741 /* end of 'while' loop! */ 2742 if ((flags & UVM_EXTRACT_CONTIG) && entry->end < end && 2743 (entry->next == &srcmap->header || 2744 entry->next->start != entry->end)) { 2745 error = EINVAL; 2746 goto bad; 2747 } 2748 entry = entry->next; 2749 fudge = 0; 2750 } 2751 2752 /* 2753 * step 4: close off chain (in format expected by uvm_map_replace) 2754 */ 2755 2756 if (chain) 2757 chain->prev = endchain; 2758 2759 /* 2760 * step 5: attempt to lock the dest map so we can pmap_copy. 2761 * note usage of copy_ok: 2762 * 1 => dstmap locked, pmap_copy ok, and we "replace" here (step 5) 2763 * 0 => dstmap unlocked, NO pmap_copy, and we will "replace" in step 7 2764 */ 2765 2766 if (srcmap == dstmap || vm_map_lock_try(dstmap) == true) { 2767 copy_ok = 1; 2768 if (!uvm_map_replace(dstmap, dstaddr, dstaddr+len, chain, 2769 nchain, &resentry)) { 2770 if (srcmap != dstmap) 2771 vm_map_unlock(dstmap); 2772 error = EIO; 2773 goto bad; 2774 } 2775 } else { 2776 copy_ok = 0; 2777 /* replace defered until step 7 */ 2778 } 2779 2780 /* 2781 * step 6: traverse the srcmap a second time to do the following: 2782 * - if we got a lock on the dstmap do pmap_copy 2783 * - if UVM_EXTRACT_REMOVE remove the entries 2784 * we make use of orig_entry and orig_fudge (saved in step 2) 2785 */ 2786 2787 if (copy_ok || (flags & UVM_EXTRACT_REMOVE)) { 2788 2789 /* purge possible stale hints from srcmap */ 2790 if (flags & UVM_EXTRACT_REMOVE) { 2791 SAVE_HINT(srcmap, srcmap->hint, orig_entry->prev); 2792 if (srcmap->first_free != &srcmap->header && 2793 srcmap->first_free->start >= start) 2794 srcmap->first_free = orig_entry->prev; 2795 } 2796 2797 entry = orig_entry; 2798 fudge = orig_fudge; 2799 deadentry = NULL; /* for UVM_EXTRACT_REMOVE */ 2800 2801 while (entry->start < end && entry != &srcmap->header) { 2802 if (copy_ok) { 2803 oldoffset = (entry->start + fudge) - start; 2804 elen = MIN(end, entry->end) - 2805 (entry->start + fudge); 2806 pmap_copy(dstmap->pmap, srcmap->pmap, 2807 dstaddr + oldoffset, elen, 2808 entry->start + fudge); 2809 } 2810 2811 /* we advance "entry" in the following if statement */ 2812 if (flags & UVM_EXTRACT_REMOVE) { 2813 pmap_remove(srcmap->pmap, entry->start, 2814 entry->end); 2815 oldentry = entry; /* save entry */ 2816 entry = entry->next; /* advance */ 2817 uvm_map_entry_unlink(srcmap, oldentry); 2818 /* add to dead list */ 2819 oldentry->next = deadentry; 2820 deadentry = oldentry; 2821 } else { 2822 entry = entry->next; /* advance */ 2823 } 2824 2825 /* end of 'while' loop */ 2826 fudge = 0; 2827 } 2828 pmap_update(srcmap->pmap); 2829 2830 /* 2831 * unlock dstmap. we will dispose of deadentry in 2832 * step 7 if needed 2833 */ 2834 2835 if (copy_ok && srcmap != dstmap) 2836 vm_map_unlock(dstmap); 2837 2838 } else { 2839 deadentry = NULL; 2840 } 2841 2842 /* 2843 * step 7: we are done with the source map, unlock. if copy_ok 2844 * is 0 then we have not replaced the dummy mapping in dstmap yet 2845 * and we need to do so now. 2846 */ 2847 2848 vm_map_unlock(srcmap); 2849 if ((flags & UVM_EXTRACT_REMOVE) && deadentry) 2850 uvm_unmap_detach(deadentry, 0); /* dispose of old entries */ 2851 2852 /* now do the replacement if we didn't do it in step 5 */ 2853 if (copy_ok == 0) { 2854 vm_map_lock(dstmap); 2855 error = uvm_map_replace(dstmap, dstaddr, dstaddr+len, chain, 2856 nchain, &resentry); 2857 vm_map_unlock(dstmap); 2858 2859 if (error == false) { 2860 error = EIO; 2861 goto bad2; 2862 } 2863 } 2864 2865 if (resentry != NULL) 2866 uvm_mapent_free(resentry); 2867 2868 uvm_map_check(srcmap, "map_extract src leave"); 2869 uvm_map_check(dstmap, "map_extract dst leave"); 2870 2871 return (0); 2872 2873 /* 2874 * bad: failure recovery 2875 */ 2876 bad: 2877 vm_map_unlock(srcmap); 2878 bad2: /* src already unlocked */ 2879 if (chain) 2880 uvm_unmap_detach(chain, 2881 (flags & UVM_EXTRACT_QREF) ? AMAP_REFALL : 0); 2882 2883 if (resentry != NULL) 2884 uvm_mapent_free(resentry); 2885 2886 uvm_map_check(srcmap, "map_extract src err leave"); 2887 uvm_map_check(dstmap, "map_extract dst err leave"); 2888 2889 if ((flags & UVM_EXTRACT_RESERVED) == 0) { 2890 uvm_unmap(dstmap, dstaddr, dstaddr+len); /* ??? */ 2891 } 2892 return (error); 2893 } 2894 2895 /* end of extraction functions */ 2896 2897 /* 2898 * uvm_map_submap: punch down part of a map into a submap 2899 * 2900 * => only the kernel_map is allowed to be submapped 2901 * => the purpose of submapping is to break up the locking granularity 2902 * of a larger map 2903 * => the range specified must have been mapped previously with a uvm_map() 2904 * call [with uobj==NULL] to create a blank map entry in the main map. 2905 * [And it had better still be blank!] 2906 * => maps which contain submaps should never be copied or forked. 2907 * => to remove a submap, use uvm_unmap() on the main map 2908 * and then uvm_map_deallocate() the submap. 2909 * => main map must be unlocked. 2910 * => submap must have been init'd and have a zero reference count. 2911 * [need not be locked as we don't actually reference it] 2912 */ 2913 2914 int 2915 uvm_map_submap(struct vm_map *map, vaddr_t start, vaddr_t end, 2916 struct vm_map *submap) 2917 { 2918 struct vm_map_entry *entry; 2919 struct uvm_mapent_reservation umr; 2920 int error; 2921 2922 uvm_mapent_reserve(map, &umr, 2, 0); 2923 2924 vm_map_lock(map); 2925 VM_MAP_RANGE_CHECK(map, start, end); 2926 2927 if (uvm_map_lookup_entry(map, start, &entry)) { 2928 UVM_MAP_CLIP_START(map, entry, start, &umr); 2929 UVM_MAP_CLIP_END(map, entry, end, &umr); /* to be safe */ 2930 } else { 2931 entry = NULL; 2932 } 2933 2934 if (entry != NULL && 2935 entry->start == start && entry->end == end && 2936 entry->object.uvm_obj == NULL && entry->aref.ar_amap == NULL && 2937 !UVM_ET_ISCOPYONWRITE(entry) && !UVM_ET_ISNEEDSCOPY(entry)) { 2938 entry->etype |= UVM_ET_SUBMAP; 2939 entry->object.sub_map = submap; 2940 entry->offset = 0; 2941 uvm_map_reference(submap); 2942 error = 0; 2943 } else { 2944 error = EINVAL; 2945 } 2946 vm_map_unlock(map); 2947 2948 uvm_mapent_unreserve(map, &umr); 2949 2950 return error; 2951 } 2952 2953 /* 2954 * uvm_map_setup_kernel: init in-kernel map 2955 * 2956 * => map must not be in service yet. 2957 */ 2958 2959 void 2960 uvm_map_setup_kernel(struct vm_map_kernel *map, 2961 vaddr_t vmin, vaddr_t vmax, int flags) 2962 { 2963 2964 uvm_map_setup(&map->vmk_map, vmin, vmax, flags); 2965 callback_head_init(&map->vmk_reclaim_callback, IPL_VM); 2966 LIST_INIT(&map->vmk_kentry_free); 2967 map->vmk_merged_entries = NULL; 2968 } 2969 2970 2971 /* 2972 * uvm_map_protect: change map protection 2973 * 2974 * => set_max means set max_protection. 2975 * => map must be unlocked. 2976 */ 2977 2978 #define MASK(entry) (UVM_ET_ISCOPYONWRITE(entry) ? \ 2979 ~VM_PROT_WRITE : VM_PROT_ALL) 2980 2981 int 2982 uvm_map_protect(struct vm_map *map, vaddr_t start, vaddr_t end, 2983 vm_prot_t new_prot, bool set_max) 2984 { 2985 struct vm_map_entry *current, *entry; 2986 int error = 0; 2987 UVMHIST_FUNC("uvm_map_protect"); UVMHIST_CALLED(maphist); 2988 UVMHIST_LOG(maphist,"(map=0x%x,start=0x%x,end=0x%x,new_prot=0x%x)", 2989 map, start, end, new_prot); 2990 2991 vm_map_lock(map); 2992 VM_MAP_RANGE_CHECK(map, start, end); 2993 if (uvm_map_lookup_entry(map, start, &entry)) { 2994 UVM_MAP_CLIP_START(map, entry, start, NULL); 2995 } else { 2996 entry = entry->next; 2997 } 2998 2999 /* 3000 * make a first pass to check for protection violations. 3001 */ 3002 3003 current = entry; 3004 while ((current != &map->header) && (current->start < end)) { 3005 if (UVM_ET_ISSUBMAP(current)) { 3006 error = EINVAL; 3007 goto out; 3008 } 3009 if ((new_prot & current->max_protection) != new_prot) { 3010 error = EACCES; 3011 goto out; 3012 } 3013 /* 3014 * Don't allow VM_PROT_EXECUTE to be set on entries that 3015 * point to vnodes that are associated with a NOEXEC file 3016 * system. 3017 */ 3018 if (UVM_ET_ISOBJ(current) && 3019 UVM_OBJ_IS_VNODE(current->object.uvm_obj)) { 3020 struct vnode *vp = 3021 (struct vnode *) current->object.uvm_obj; 3022 3023 if ((new_prot & VM_PROT_EXECUTE) != 0 && 3024 (vp->v_mount->mnt_flag & MNT_NOEXEC) != 0) { 3025 error = EACCES; 3026 goto out; 3027 } 3028 } 3029 3030 current = current->next; 3031 } 3032 3033 /* go back and fix up protections (no need to clip this time). */ 3034 3035 current = entry; 3036 while ((current != &map->header) && (current->start < end)) { 3037 vm_prot_t old_prot; 3038 3039 UVM_MAP_CLIP_END(map, current, end, NULL); 3040 old_prot = current->protection; 3041 if (set_max) 3042 current->protection = 3043 (current->max_protection = new_prot) & old_prot; 3044 else 3045 current->protection = new_prot; 3046 3047 /* 3048 * update physical map if necessary. worry about copy-on-write 3049 * here -- CHECK THIS XXX 3050 */ 3051 3052 if (current->protection != old_prot) { 3053 /* update pmap! */ 3054 pmap_protect(map->pmap, current->start, current->end, 3055 current->protection & MASK(entry)); 3056 3057 /* 3058 * If this entry points at a vnode, and the 3059 * protection includes VM_PROT_EXECUTE, mark 3060 * the vnode as VEXECMAP. 3061 */ 3062 if (UVM_ET_ISOBJ(current)) { 3063 struct uvm_object *uobj = 3064 current->object.uvm_obj; 3065 3066 if (UVM_OBJ_IS_VNODE(uobj) && 3067 (current->protection & VM_PROT_EXECUTE)) { 3068 mutex_enter(&uobj->vmobjlock); 3069 vn_markexec((struct vnode *) uobj); 3070 mutex_exit(&uobj->vmobjlock); 3071 } 3072 } 3073 } 3074 3075 /* 3076 * If the map is configured to lock any future mappings, 3077 * wire this entry now if the old protection was VM_PROT_NONE 3078 * and the new protection is not VM_PROT_NONE. 3079 */ 3080 3081 if ((map->flags & VM_MAP_WIREFUTURE) != 0 && 3082 VM_MAPENT_ISWIRED(entry) == 0 && 3083 old_prot == VM_PROT_NONE && 3084 new_prot != VM_PROT_NONE) { 3085 if (uvm_map_pageable(map, entry->start, 3086 entry->end, false, 3087 UVM_LK_ENTER|UVM_LK_EXIT) != 0) { 3088 3089 /* 3090 * If locking the entry fails, remember the 3091 * error if it's the first one. Note we 3092 * still continue setting the protection in 3093 * the map, but will return the error 3094 * condition regardless. 3095 * 3096 * XXX Ignore what the actual error is, 3097 * XXX just call it a resource shortage 3098 * XXX so that it doesn't get confused 3099 * XXX what uvm_map_protect() itself would 3100 * XXX normally return. 3101 */ 3102 3103 error = ENOMEM; 3104 } 3105 } 3106 current = current->next; 3107 } 3108 pmap_update(map->pmap); 3109 3110 out: 3111 vm_map_unlock(map); 3112 3113 UVMHIST_LOG(maphist, "<- done, error=%d",error,0,0,0); 3114 return error; 3115 } 3116 3117 #undef MASK 3118 3119 /* 3120 * uvm_map_inherit: set inheritance code for range of addrs in map. 3121 * 3122 * => map must be unlocked 3123 * => note that the inherit code is used during a "fork". see fork 3124 * code for details. 3125 */ 3126 3127 int 3128 uvm_map_inherit(struct vm_map *map, vaddr_t start, vaddr_t end, 3129 vm_inherit_t new_inheritance) 3130 { 3131 struct vm_map_entry *entry, *temp_entry; 3132 UVMHIST_FUNC("uvm_map_inherit"); UVMHIST_CALLED(maphist); 3133 UVMHIST_LOG(maphist,"(map=0x%x,start=0x%x,end=0x%x,new_inh=0x%x)", 3134 map, start, end, new_inheritance); 3135 3136 switch (new_inheritance) { 3137 case MAP_INHERIT_NONE: 3138 case MAP_INHERIT_COPY: 3139 case MAP_INHERIT_SHARE: 3140 break; 3141 default: 3142 UVMHIST_LOG(maphist,"<- done (INVALID ARG)",0,0,0,0); 3143 return EINVAL; 3144 } 3145 3146 vm_map_lock(map); 3147 VM_MAP_RANGE_CHECK(map, start, end); 3148 if (uvm_map_lookup_entry(map, start, &temp_entry)) { 3149 entry = temp_entry; 3150 UVM_MAP_CLIP_START(map, entry, start, NULL); 3151 } else { 3152 entry = temp_entry->next; 3153 } 3154 while ((entry != &map->header) && (entry->start < end)) { 3155 UVM_MAP_CLIP_END(map, entry, end, NULL); 3156 entry->inheritance = new_inheritance; 3157 entry = entry->next; 3158 } 3159 vm_map_unlock(map); 3160 UVMHIST_LOG(maphist,"<- done (OK)",0,0,0,0); 3161 return 0; 3162 } 3163 3164 /* 3165 * uvm_map_advice: set advice code for range of addrs in map. 3166 * 3167 * => map must be unlocked 3168 */ 3169 3170 int 3171 uvm_map_advice(struct vm_map *map, vaddr_t start, vaddr_t end, int new_advice) 3172 { 3173 struct vm_map_entry *entry, *temp_entry; 3174 UVMHIST_FUNC("uvm_map_advice"); UVMHIST_CALLED(maphist); 3175 UVMHIST_LOG(maphist,"(map=0x%x,start=0x%x,end=0x%x,new_adv=0x%x)", 3176 map, start, end, new_advice); 3177 3178 vm_map_lock(map); 3179 VM_MAP_RANGE_CHECK(map, start, end); 3180 if (uvm_map_lookup_entry(map, start, &temp_entry)) { 3181 entry = temp_entry; 3182 UVM_MAP_CLIP_START(map, entry, start, NULL); 3183 } else { 3184 entry = temp_entry->next; 3185 } 3186 3187 /* 3188 * XXXJRT: disallow holes? 3189 */ 3190 3191 while ((entry != &map->header) && (entry->start < end)) { 3192 UVM_MAP_CLIP_END(map, entry, end, NULL); 3193 3194 switch (new_advice) { 3195 case MADV_NORMAL: 3196 case MADV_RANDOM: 3197 case MADV_SEQUENTIAL: 3198 /* nothing special here */ 3199 break; 3200 3201 default: 3202 vm_map_unlock(map); 3203 UVMHIST_LOG(maphist,"<- done (INVALID ARG)",0,0,0,0); 3204 return EINVAL; 3205 } 3206 entry->advice = new_advice; 3207 entry = entry->next; 3208 } 3209 3210 vm_map_unlock(map); 3211 UVMHIST_LOG(maphist,"<- done (OK)",0,0,0,0); 3212 return 0; 3213 } 3214 3215 /* 3216 * uvm_map_pageable: sets the pageability of a range in a map. 3217 * 3218 * => wires map entries. should not be used for transient page locking. 3219 * for that, use uvm_fault_wire()/uvm_fault_unwire() (see uvm_vslock()). 3220 * => regions specified as not pageable require lock-down (wired) memory 3221 * and page tables. 3222 * => map must never be read-locked 3223 * => if islocked is true, map is already write-locked 3224 * => we always unlock the map, since we must downgrade to a read-lock 3225 * to call uvm_fault_wire() 3226 * => XXXCDC: check this and try and clean it up. 3227 */ 3228 3229 int 3230 uvm_map_pageable(struct vm_map *map, vaddr_t start, vaddr_t end, 3231 bool new_pageable, int lockflags) 3232 { 3233 struct vm_map_entry *entry, *start_entry, *failed_entry; 3234 int rv; 3235 #ifdef DIAGNOSTIC 3236 u_int timestamp_save; 3237 #endif 3238 UVMHIST_FUNC("uvm_map_pageable"); UVMHIST_CALLED(maphist); 3239 UVMHIST_LOG(maphist,"(map=0x%x,start=0x%x,end=0x%x,new_pageable=0x%x)", 3240 map, start, end, new_pageable); 3241 KASSERT(map->flags & VM_MAP_PAGEABLE); 3242 3243 if ((lockflags & UVM_LK_ENTER) == 0) 3244 vm_map_lock(map); 3245 VM_MAP_RANGE_CHECK(map, start, end); 3246 3247 /* 3248 * only one pageability change may take place at one time, since 3249 * uvm_fault_wire assumes it will be called only once for each 3250 * wiring/unwiring. therefore, we have to make sure we're actually 3251 * changing the pageability for the entire region. we do so before 3252 * making any changes. 3253 */ 3254 3255 if (uvm_map_lookup_entry(map, start, &start_entry) == false) { 3256 if ((lockflags & UVM_LK_EXIT) == 0) 3257 vm_map_unlock(map); 3258 3259 UVMHIST_LOG(maphist,"<- done (fault)",0,0,0,0); 3260 return EFAULT; 3261 } 3262 entry = start_entry; 3263 3264 /* 3265 * handle wiring and unwiring separately. 3266 */ 3267 3268 if (new_pageable) { /* unwire */ 3269 UVM_MAP_CLIP_START(map, entry, start, NULL); 3270 3271 /* 3272 * unwiring. first ensure that the range to be unwired is 3273 * really wired down and that there are no holes. 3274 */ 3275 3276 while ((entry != &map->header) && (entry->start < end)) { 3277 if (entry->wired_count == 0 || 3278 (entry->end < end && 3279 (entry->next == &map->header || 3280 entry->next->start > entry->end))) { 3281 if ((lockflags & UVM_LK_EXIT) == 0) 3282 vm_map_unlock(map); 3283 UVMHIST_LOG(maphist, "<- done (INVAL)",0,0,0,0); 3284 return EINVAL; 3285 } 3286 entry = entry->next; 3287 } 3288 3289 /* 3290 * POSIX 1003.1b - a single munlock call unlocks a region, 3291 * regardless of the number of mlock calls made on that 3292 * region. 3293 */ 3294 3295 entry = start_entry; 3296 while ((entry != &map->header) && (entry->start < end)) { 3297 UVM_MAP_CLIP_END(map, entry, end, NULL); 3298 if (VM_MAPENT_ISWIRED(entry)) 3299 uvm_map_entry_unwire(map, entry); 3300 entry = entry->next; 3301 } 3302 if ((lockflags & UVM_LK_EXIT) == 0) 3303 vm_map_unlock(map); 3304 UVMHIST_LOG(maphist,"<- done (OK UNWIRE)",0,0,0,0); 3305 return 0; 3306 } 3307 3308 /* 3309 * wire case: in two passes [XXXCDC: ugly block of code here] 3310 * 3311 * 1: holding the write lock, we create any anonymous maps that need 3312 * to be created. then we clip each map entry to the region to 3313 * be wired and increment its wiring count. 3314 * 3315 * 2: we downgrade to a read lock, and call uvm_fault_wire to fault 3316 * in the pages for any newly wired area (wired_count == 1). 3317 * 3318 * downgrading to a read lock for uvm_fault_wire avoids a possible 3319 * deadlock with another thread that may have faulted on one of 3320 * the pages to be wired (it would mark the page busy, blocking 3321 * us, then in turn block on the map lock that we hold). because 3322 * of problems in the recursive lock package, we cannot upgrade 3323 * to a write lock in vm_map_lookup. thus, any actions that 3324 * require the write lock must be done beforehand. because we 3325 * keep the read lock on the map, the copy-on-write status of the 3326 * entries we modify here cannot change. 3327 */ 3328 3329 while ((entry != &map->header) && (entry->start < end)) { 3330 if (VM_MAPENT_ISWIRED(entry) == 0) { /* not already wired? */ 3331 3332 /* 3333 * perform actions of vm_map_lookup that need the 3334 * write lock on the map: create an anonymous map 3335 * for a copy-on-write region, or an anonymous map 3336 * for a zero-fill region. (XXXCDC: submap case 3337 * ok?) 3338 */ 3339 3340 if (!UVM_ET_ISSUBMAP(entry)) { /* not submap */ 3341 if (UVM_ET_ISNEEDSCOPY(entry) && 3342 ((entry->max_protection & VM_PROT_WRITE) || 3343 (entry->object.uvm_obj == NULL))) { 3344 amap_copy(map, entry, 0, start, end); 3345 /* XXXCDC: wait OK? */ 3346 } 3347 } 3348 } 3349 UVM_MAP_CLIP_START(map, entry, start, NULL); 3350 UVM_MAP_CLIP_END(map, entry, end, NULL); 3351 entry->wired_count++; 3352 3353 /* 3354 * Check for holes 3355 */ 3356 3357 if (entry->protection == VM_PROT_NONE || 3358 (entry->end < end && 3359 (entry->next == &map->header || 3360 entry->next->start > entry->end))) { 3361 3362 /* 3363 * found one. amap creation actions do not need to 3364 * be undone, but the wired counts need to be restored. 3365 */ 3366 3367 while (entry != &map->header && entry->end > start) { 3368 entry->wired_count--; 3369 entry = entry->prev; 3370 } 3371 if ((lockflags & UVM_LK_EXIT) == 0) 3372 vm_map_unlock(map); 3373 UVMHIST_LOG(maphist,"<- done (INVALID WIRE)",0,0,0,0); 3374 return EINVAL; 3375 } 3376 entry = entry->next; 3377 } 3378 3379 /* 3380 * Pass 2. 3381 */ 3382 3383 #ifdef DIAGNOSTIC 3384 timestamp_save = map->timestamp; 3385 #endif 3386 vm_map_busy(map); 3387 vm_map_unlock(map); 3388 3389 rv = 0; 3390 entry = start_entry; 3391 while (entry != &map->header && entry->start < end) { 3392 if (entry->wired_count == 1) { 3393 rv = uvm_fault_wire(map, entry->start, entry->end, 3394 entry->max_protection, 1); 3395 if (rv) { 3396 3397 /* 3398 * wiring failed. break out of the loop. 3399 * we'll clean up the map below, once we 3400 * have a write lock again. 3401 */ 3402 3403 break; 3404 } 3405 } 3406 entry = entry->next; 3407 } 3408 3409 if (rv) { /* failed? */ 3410 3411 /* 3412 * Get back to an exclusive (write) lock. 3413 */ 3414 3415 vm_map_lock(map); 3416 vm_map_unbusy(map); 3417 3418 #ifdef DIAGNOSTIC 3419 if (timestamp_save + 1 != map->timestamp) 3420 panic("uvm_map_pageable: stale map"); 3421 #endif 3422 3423 /* 3424 * first drop the wiring count on all the entries 3425 * which haven't actually been wired yet. 3426 */ 3427 3428 failed_entry = entry; 3429 while (entry != &map->header && entry->start < end) { 3430 entry->wired_count--; 3431 entry = entry->next; 3432 } 3433 3434 /* 3435 * now, unwire all the entries that were successfully 3436 * wired above. 3437 */ 3438 3439 entry = start_entry; 3440 while (entry != failed_entry) { 3441 entry->wired_count--; 3442 if (VM_MAPENT_ISWIRED(entry) == 0) 3443 uvm_map_entry_unwire(map, entry); 3444 entry = entry->next; 3445 } 3446 if ((lockflags & UVM_LK_EXIT) == 0) 3447 vm_map_unlock(map); 3448 UVMHIST_LOG(maphist, "<- done (RV=%d)", rv,0,0,0); 3449 return (rv); 3450 } 3451 3452 if ((lockflags & UVM_LK_EXIT) == 0) { 3453 vm_map_unbusy(map); 3454 } else { 3455 3456 /* 3457 * Get back to an exclusive (write) lock. 3458 */ 3459 3460 vm_map_lock(map); 3461 vm_map_unbusy(map); 3462 } 3463 3464 UVMHIST_LOG(maphist,"<- done (OK WIRE)",0,0,0,0); 3465 return 0; 3466 } 3467 3468 /* 3469 * uvm_map_pageable_all: special case of uvm_map_pageable - affects 3470 * all mapped regions. 3471 * 3472 * => map must not be locked. 3473 * => if no flags are specified, all regions are unwired. 3474 * => XXXJRT: has some of the same problems as uvm_map_pageable() above. 3475 */ 3476 3477 int 3478 uvm_map_pageable_all(struct vm_map *map, int flags, vsize_t limit) 3479 { 3480 struct vm_map_entry *entry, *failed_entry; 3481 vsize_t size; 3482 int rv; 3483 #ifdef DIAGNOSTIC 3484 u_int timestamp_save; 3485 #endif 3486 UVMHIST_FUNC("uvm_map_pageable_all"); UVMHIST_CALLED(maphist); 3487 UVMHIST_LOG(maphist,"(map=0x%x,flags=0x%x)", map, flags, 0, 0); 3488 3489 KASSERT(map->flags & VM_MAP_PAGEABLE); 3490 3491 vm_map_lock(map); 3492 3493 /* 3494 * handle wiring and unwiring separately. 3495 */ 3496 3497 if (flags == 0) { /* unwire */ 3498 3499 /* 3500 * POSIX 1003.1b -- munlockall unlocks all regions, 3501 * regardless of how many times mlockall has been called. 3502 */ 3503 3504 for (entry = map->header.next; entry != &map->header; 3505 entry = entry->next) { 3506 if (VM_MAPENT_ISWIRED(entry)) 3507 uvm_map_entry_unwire(map, entry); 3508 } 3509 map->flags &= ~VM_MAP_WIREFUTURE; 3510 vm_map_unlock(map); 3511 UVMHIST_LOG(maphist,"<- done (OK UNWIRE)",0,0,0,0); 3512 return 0; 3513 } 3514 3515 if (flags & MCL_FUTURE) { 3516 3517 /* 3518 * must wire all future mappings; remember this. 3519 */ 3520 3521 map->flags |= VM_MAP_WIREFUTURE; 3522 } 3523 3524 if ((flags & MCL_CURRENT) == 0) { 3525 3526 /* 3527 * no more work to do! 3528 */ 3529 3530 UVMHIST_LOG(maphist,"<- done (OK no wire)",0,0,0,0); 3531 vm_map_unlock(map); 3532 return 0; 3533 } 3534 3535 /* 3536 * wire case: in three passes [XXXCDC: ugly block of code here] 3537 * 3538 * 1: holding the write lock, count all pages mapped by non-wired 3539 * entries. if this would cause us to go over our limit, we fail. 3540 * 3541 * 2: still holding the write lock, we create any anonymous maps that 3542 * need to be created. then we increment its wiring count. 3543 * 3544 * 3: we downgrade to a read lock, and call uvm_fault_wire to fault 3545 * in the pages for any newly wired area (wired_count == 1). 3546 * 3547 * downgrading to a read lock for uvm_fault_wire avoids a possible 3548 * deadlock with another thread that may have faulted on one of 3549 * the pages to be wired (it would mark the page busy, blocking 3550 * us, then in turn block on the map lock that we hold). because 3551 * of problems in the recursive lock package, we cannot upgrade 3552 * to a write lock in vm_map_lookup. thus, any actions that 3553 * require the write lock must be done beforehand. because we 3554 * keep the read lock on the map, the copy-on-write status of the 3555 * entries we modify here cannot change. 3556 */ 3557 3558 for (size = 0, entry = map->header.next; entry != &map->header; 3559 entry = entry->next) { 3560 if (entry->protection != VM_PROT_NONE && 3561 VM_MAPENT_ISWIRED(entry) == 0) { /* not already wired? */ 3562 size += entry->end - entry->start; 3563 } 3564 } 3565 3566 if (atop(size) + uvmexp.wired > uvmexp.wiredmax) { 3567 vm_map_unlock(map); 3568 return ENOMEM; 3569 } 3570 3571 if (limit != 0 && 3572 (size + ptoa(pmap_wired_count(vm_map_pmap(map))) > limit)) { 3573 vm_map_unlock(map); 3574 return ENOMEM; 3575 } 3576 3577 /* 3578 * Pass 2. 3579 */ 3580 3581 for (entry = map->header.next; entry != &map->header; 3582 entry = entry->next) { 3583 if (entry->protection == VM_PROT_NONE) 3584 continue; 3585 if (VM_MAPENT_ISWIRED(entry) == 0) { /* not already wired? */ 3586 3587 /* 3588 * perform actions of vm_map_lookup that need the 3589 * write lock on the map: create an anonymous map 3590 * for a copy-on-write region, or an anonymous map 3591 * for a zero-fill region. (XXXCDC: submap case 3592 * ok?) 3593 */ 3594 3595 if (!UVM_ET_ISSUBMAP(entry)) { /* not submap */ 3596 if (UVM_ET_ISNEEDSCOPY(entry) && 3597 ((entry->max_protection & VM_PROT_WRITE) || 3598 (entry->object.uvm_obj == NULL))) { 3599 amap_copy(map, entry, 0, entry->start, 3600 entry->end); 3601 /* XXXCDC: wait OK? */ 3602 } 3603 } 3604 } 3605 entry->wired_count++; 3606 } 3607 3608 /* 3609 * Pass 3. 3610 */ 3611 3612 #ifdef DIAGNOSTIC 3613 timestamp_save = map->timestamp; 3614 #endif 3615 vm_map_busy(map); 3616 vm_map_unlock(map); 3617 3618 rv = 0; 3619 for (entry = map->header.next; entry != &map->header; 3620 entry = entry->next) { 3621 if (entry->wired_count == 1) { 3622 rv = uvm_fault_wire(map, entry->start, entry->end, 3623 entry->max_protection, 1); 3624 if (rv) { 3625 3626 /* 3627 * wiring failed. break out of the loop. 3628 * we'll clean up the map below, once we 3629 * have a write lock again. 3630 */ 3631 3632 break; 3633 } 3634 } 3635 } 3636 3637 if (rv) { 3638 3639 /* 3640 * Get back an exclusive (write) lock. 3641 */ 3642 3643 vm_map_lock(map); 3644 vm_map_unbusy(map); 3645 3646 #ifdef DIAGNOSTIC 3647 if (timestamp_save + 1 != map->timestamp) 3648 panic("uvm_map_pageable_all: stale map"); 3649 #endif 3650 3651 /* 3652 * first drop the wiring count on all the entries 3653 * which haven't actually been wired yet. 3654 * 3655 * Skip VM_PROT_NONE entries like we did above. 3656 */ 3657 3658 failed_entry = entry; 3659 for (/* nothing */; entry != &map->header; 3660 entry = entry->next) { 3661 if (entry->protection == VM_PROT_NONE) 3662 continue; 3663 entry->wired_count--; 3664 } 3665 3666 /* 3667 * now, unwire all the entries that were successfully 3668 * wired above. 3669 * 3670 * Skip VM_PROT_NONE entries like we did above. 3671 */ 3672 3673 for (entry = map->header.next; entry != failed_entry; 3674 entry = entry->next) { 3675 if (entry->protection == VM_PROT_NONE) 3676 continue; 3677 entry->wired_count--; 3678 if (VM_MAPENT_ISWIRED(entry)) 3679 uvm_map_entry_unwire(map, entry); 3680 } 3681 vm_map_unlock(map); 3682 UVMHIST_LOG(maphist,"<- done (RV=%d)", rv,0,0,0); 3683 return (rv); 3684 } 3685 3686 vm_map_unbusy(map); 3687 3688 UVMHIST_LOG(maphist,"<- done (OK WIRE)",0,0,0,0); 3689 return 0; 3690 } 3691 3692 /* 3693 * uvm_map_clean: clean out a map range 3694 * 3695 * => valid flags: 3696 * if (flags & PGO_CLEANIT): dirty pages are cleaned first 3697 * if (flags & PGO_SYNCIO): dirty pages are written synchronously 3698 * if (flags & PGO_DEACTIVATE): any cached pages are deactivated after clean 3699 * if (flags & PGO_FREE): any cached pages are freed after clean 3700 * => returns an error if any part of the specified range isn't mapped 3701 * => never a need to flush amap layer since the anonymous memory has 3702 * no permanent home, but may deactivate pages there 3703 * => called from sys_msync() and sys_madvise() 3704 * => caller must not write-lock map (read OK). 3705 * => we may sleep while cleaning if SYNCIO [with map read-locked] 3706 */ 3707 3708 int 3709 uvm_map_clean(struct vm_map *map, vaddr_t start, vaddr_t end, int flags) 3710 { 3711 struct vm_map_entry *current, *entry; 3712 struct uvm_object *uobj; 3713 struct vm_amap *amap; 3714 struct vm_anon *anon; 3715 struct vm_page *pg; 3716 vaddr_t offset; 3717 vsize_t size; 3718 voff_t uoff; 3719 int error, refs; 3720 UVMHIST_FUNC("uvm_map_clean"); UVMHIST_CALLED(maphist); 3721 3722 UVMHIST_LOG(maphist,"(map=0x%x,start=0x%x,end=0x%x,flags=0x%x)", 3723 map, start, end, flags); 3724 KASSERT((flags & (PGO_FREE|PGO_DEACTIVATE)) != 3725 (PGO_FREE|PGO_DEACTIVATE)); 3726 3727 vm_map_lock_read(map); 3728 VM_MAP_RANGE_CHECK(map, start, end); 3729 if (uvm_map_lookup_entry(map, start, &entry) == false) { 3730 vm_map_unlock_read(map); 3731 return EFAULT; 3732 } 3733 3734 /* 3735 * Make a first pass to check for holes and wiring problems. 3736 */ 3737 3738 for (current = entry; current->start < end; current = current->next) { 3739 if (UVM_ET_ISSUBMAP(current)) { 3740 vm_map_unlock_read(map); 3741 return EINVAL; 3742 } 3743 if ((flags & PGO_FREE) != 0 && VM_MAPENT_ISWIRED(entry)) { 3744 vm_map_unlock_read(map); 3745 return EBUSY; 3746 } 3747 if (end <= current->end) { 3748 break; 3749 } 3750 if (current->end != current->next->start) { 3751 vm_map_unlock_read(map); 3752 return EFAULT; 3753 } 3754 } 3755 3756 error = 0; 3757 for (current = entry; start < end; current = current->next) { 3758 amap = current->aref.ar_amap; /* top layer */ 3759 uobj = current->object.uvm_obj; /* bottom layer */ 3760 KASSERT(start >= current->start); 3761 3762 /* 3763 * No amap cleaning necessary if: 3764 * 3765 * (1) There's no amap. 3766 * 3767 * (2) We're not deactivating or freeing pages. 3768 */ 3769 3770 if (amap == NULL || (flags & (PGO_DEACTIVATE|PGO_FREE)) == 0) 3771 goto flush_object; 3772 3773 amap_lock(amap); 3774 offset = start - current->start; 3775 size = MIN(end, current->end) - start; 3776 for ( ; size != 0; size -= PAGE_SIZE, offset += PAGE_SIZE) { 3777 anon = amap_lookup(¤t->aref, offset); 3778 if (anon == NULL) 3779 continue; 3780 3781 mutex_enter(&anon->an_lock); 3782 pg = anon->an_page; 3783 if (pg == NULL) { 3784 mutex_exit(&anon->an_lock); 3785 continue; 3786 } 3787 3788 switch (flags & (PGO_CLEANIT|PGO_FREE|PGO_DEACTIVATE)) { 3789 3790 /* 3791 * In these first 3 cases, we just deactivate the page. 3792 */ 3793 3794 case PGO_CLEANIT|PGO_FREE: 3795 case PGO_CLEANIT|PGO_DEACTIVATE: 3796 case PGO_DEACTIVATE: 3797 deactivate_it: 3798 /* 3799 * skip the page if it's loaned or wired, 3800 * since it shouldn't be on a paging queue 3801 * at all in these cases. 3802 */ 3803 3804 mutex_enter(&uvm_pageqlock); 3805 if (pg->loan_count != 0 || 3806 pg->wire_count != 0) { 3807 mutex_exit(&uvm_pageqlock); 3808 mutex_exit(&anon->an_lock); 3809 continue; 3810 } 3811 KASSERT(pg->uanon == anon); 3812 uvm_pagedeactivate(pg); 3813 mutex_exit(&uvm_pageqlock); 3814 mutex_exit(&anon->an_lock); 3815 continue; 3816 3817 case PGO_FREE: 3818 3819 /* 3820 * If there are multiple references to 3821 * the amap, just deactivate the page. 3822 */ 3823 3824 if (amap_refs(amap) > 1) 3825 goto deactivate_it; 3826 3827 /* skip the page if it's wired */ 3828 if (pg->wire_count != 0) { 3829 mutex_exit(&anon->an_lock); 3830 continue; 3831 } 3832 amap_unadd(¤t->aref, offset); 3833 refs = --anon->an_ref; 3834 mutex_exit(&anon->an_lock); 3835 if (refs == 0) 3836 uvm_anfree(anon); 3837 continue; 3838 } 3839 } 3840 amap_unlock(amap); 3841 3842 flush_object: 3843 /* 3844 * flush pages if we've got a valid backing object. 3845 * note that we must always clean object pages before 3846 * freeing them since otherwise we could reveal stale 3847 * data from files. 3848 */ 3849 3850 uoff = current->offset + (start - current->start); 3851 size = MIN(end, current->end) - start; 3852 if (uobj != NULL) { 3853 mutex_enter(&uobj->vmobjlock); 3854 if (uobj->pgops->pgo_put != NULL) 3855 error = (uobj->pgops->pgo_put)(uobj, uoff, 3856 uoff + size, flags | PGO_CLEANIT); 3857 else 3858 error = 0; 3859 } 3860 start += size; 3861 } 3862 vm_map_unlock_read(map); 3863 return (error); 3864 } 3865 3866 3867 /* 3868 * uvm_map_checkprot: check protection in map 3869 * 3870 * => must allow specified protection in a fully allocated region. 3871 * => map must be read or write locked by caller. 3872 */ 3873 3874 bool 3875 uvm_map_checkprot(struct vm_map *map, vaddr_t start, vaddr_t end, 3876 vm_prot_t protection) 3877 { 3878 struct vm_map_entry *entry; 3879 struct vm_map_entry *tmp_entry; 3880 3881 if (!uvm_map_lookup_entry(map, start, &tmp_entry)) { 3882 return (false); 3883 } 3884 entry = tmp_entry; 3885 while (start < end) { 3886 if (entry == &map->header) { 3887 return (false); 3888 } 3889 3890 /* 3891 * no holes allowed 3892 */ 3893 3894 if (start < entry->start) { 3895 return (false); 3896 } 3897 3898 /* 3899 * check protection associated with entry 3900 */ 3901 3902 if ((entry->protection & protection) != protection) { 3903 return (false); 3904 } 3905 start = entry->end; 3906 entry = entry->next; 3907 } 3908 return (true); 3909 } 3910 3911 /* 3912 * uvmspace_alloc: allocate a vmspace structure. 3913 * 3914 * - structure includes vm_map and pmap 3915 * - XXX: no locking on this structure 3916 * - refcnt set to 1, rest must be init'd by caller 3917 */ 3918 struct vmspace * 3919 uvmspace_alloc(vaddr_t vmin, vaddr_t vmax) 3920 { 3921 struct vmspace *vm; 3922 UVMHIST_FUNC("uvmspace_alloc"); UVMHIST_CALLED(maphist); 3923 3924 vm = pool_cache_get(&uvm_vmspace_cache, PR_WAITOK); 3925 uvmspace_init(vm, NULL, vmin, vmax); 3926 UVMHIST_LOG(maphist,"<- done (vm=0x%x)", vm,0,0,0); 3927 return (vm); 3928 } 3929 3930 /* 3931 * uvmspace_init: initialize a vmspace structure. 3932 * 3933 * - XXX: no locking on this structure 3934 * - refcnt set to 1, rest must be init'd by caller 3935 */ 3936 void 3937 uvmspace_init(struct vmspace *vm, struct pmap *pmap, vaddr_t vmin, vaddr_t vmax) 3938 { 3939 UVMHIST_FUNC("uvmspace_init"); UVMHIST_CALLED(maphist); 3940 3941 memset(vm, 0, sizeof(*vm)); 3942 uvm_map_setup(&vm->vm_map, vmin, vmax, VM_MAP_PAGEABLE 3943 #ifdef __USING_TOPDOWN_VM 3944 | VM_MAP_TOPDOWN 3945 #endif 3946 ); 3947 if (pmap) 3948 pmap_reference(pmap); 3949 else 3950 pmap = pmap_create(); 3951 vm->vm_map.pmap = pmap; 3952 vm->vm_refcnt = 1; 3953 UVMHIST_LOG(maphist,"<- done",0,0,0,0); 3954 } 3955 3956 /* 3957 * uvmspace_share: share a vmspace between two processes 3958 * 3959 * - used for vfork, threads(?) 3960 */ 3961 3962 void 3963 uvmspace_share(struct proc *p1, struct proc *p2) 3964 { 3965 3966 uvmspace_addref(p1->p_vmspace); 3967 p2->p_vmspace = p1->p_vmspace; 3968 } 3969 3970 /* 3971 * uvmspace_unshare: ensure that process "p" has its own, unshared, vmspace 3972 * 3973 * - XXX: no locking on vmspace 3974 */ 3975 3976 void 3977 uvmspace_unshare(struct lwp *l) 3978 { 3979 struct proc *p = l->l_proc; 3980 struct vmspace *nvm, *ovm = p->p_vmspace; 3981 3982 if (ovm->vm_refcnt == 1) 3983 /* nothing to do: vmspace isn't shared in the first place */ 3984 return; 3985 3986 /* make a new vmspace, still holding old one */ 3987 nvm = uvmspace_fork(ovm); 3988 3989 kpreempt_disable(); 3990 pmap_deactivate(l); /* unbind old vmspace */ 3991 p->p_vmspace = nvm; 3992 pmap_activate(l); /* switch to new vmspace */ 3993 kpreempt_enable(); 3994 3995 uvmspace_free(ovm); /* drop reference to old vmspace */ 3996 } 3997 3998 /* 3999 * uvmspace_exec: the process wants to exec a new program 4000 */ 4001 4002 void 4003 uvmspace_exec(struct lwp *l, vaddr_t start, vaddr_t end) 4004 { 4005 struct proc *p = l->l_proc; 4006 struct vmspace *nvm, *ovm = p->p_vmspace; 4007 struct vm_map *map = &ovm->vm_map; 4008 4009 #ifdef __sparc__ 4010 /* XXX cgd 960926: the sparc #ifdef should be a MD hook */ 4011 kill_user_windows(l); /* before stack addresses go away */ 4012 #endif 4013 4014 /* 4015 * see if more than one process is using this vmspace... 4016 */ 4017 4018 if (ovm->vm_refcnt == 1) { 4019 4020 /* 4021 * if p is the only process using its vmspace then we can safely 4022 * recycle that vmspace for the program that is being exec'd. 4023 */ 4024 4025 #ifdef SYSVSHM 4026 /* 4027 * SYSV SHM semantics require us to kill all segments on an exec 4028 */ 4029 4030 if (ovm->vm_shm) 4031 shmexit(ovm); 4032 #endif 4033 4034 /* 4035 * POSIX 1003.1b -- "lock future mappings" is revoked 4036 * when a process execs another program image. 4037 */ 4038 4039 map->flags &= ~VM_MAP_WIREFUTURE; 4040 4041 /* 4042 * now unmap the old program 4043 */ 4044 4045 pmap_remove_all(map->pmap); 4046 uvm_unmap(map, vm_map_min(map), vm_map_max(map)); 4047 KASSERT(map->header.prev == &map->header); 4048 KASSERT(map->nentries == 0); 4049 4050 /* 4051 * resize the map 4052 */ 4053 4054 vm_map_setmin(map, start); 4055 vm_map_setmax(map, end); 4056 } else { 4057 4058 /* 4059 * p's vmspace is being shared, so we can't reuse it for p since 4060 * it is still being used for others. allocate a new vmspace 4061 * for p 4062 */ 4063 4064 nvm = uvmspace_alloc(start, end); 4065 4066 /* 4067 * install new vmspace and drop our ref to the old one. 4068 */ 4069 4070 kpreempt_disable(); 4071 pmap_deactivate(l); 4072 p->p_vmspace = nvm; 4073 pmap_activate(l); 4074 kpreempt_enable(); 4075 4076 uvmspace_free(ovm); 4077 } 4078 } 4079 4080 /* 4081 * uvmspace_addref: add a referece to a vmspace. 4082 */ 4083 4084 void 4085 uvmspace_addref(struct vmspace *vm) 4086 { 4087 struct vm_map *map = &vm->vm_map; 4088 4089 KASSERT((map->flags & VM_MAP_DYING) == 0); 4090 4091 mutex_enter(&map->misc_lock); 4092 KASSERT(vm->vm_refcnt > 0); 4093 vm->vm_refcnt++; 4094 mutex_exit(&map->misc_lock); 4095 } 4096 4097 /* 4098 * uvmspace_free: free a vmspace data structure 4099 */ 4100 4101 void 4102 uvmspace_free(struct vmspace *vm) 4103 { 4104 struct vm_map_entry *dead_entries; 4105 struct vm_map *map = &vm->vm_map; 4106 int n; 4107 4108 UVMHIST_FUNC("uvmspace_free"); UVMHIST_CALLED(maphist); 4109 4110 UVMHIST_LOG(maphist,"(vm=0x%x) ref=%d", vm, vm->vm_refcnt,0,0); 4111 mutex_enter(&map->misc_lock); 4112 n = --vm->vm_refcnt; 4113 mutex_exit(&map->misc_lock); 4114 if (n > 0) 4115 return; 4116 4117 /* 4118 * at this point, there should be no other references to the map. 4119 * delete all of the mappings, then destroy the pmap. 4120 */ 4121 4122 map->flags |= VM_MAP_DYING; 4123 pmap_remove_all(map->pmap); 4124 #ifdef SYSVSHM 4125 /* Get rid of any SYSV shared memory segments. */ 4126 if (vm->vm_shm != NULL) 4127 shmexit(vm); 4128 #endif 4129 if (map->nentries) { 4130 uvm_unmap_remove(map, vm_map_min(map), vm_map_max(map), 4131 &dead_entries, NULL, 0); 4132 if (dead_entries != NULL) 4133 uvm_unmap_detach(dead_entries, 0); 4134 } 4135 KASSERT(map->nentries == 0); 4136 KASSERT(map->size == 0); 4137 mutex_destroy(&map->misc_lock); 4138 mutex_destroy(&map->mutex); 4139 rw_destroy(&map->lock); 4140 pmap_destroy(map->pmap); 4141 pool_cache_put(&uvm_vmspace_cache, vm); 4142 } 4143 4144 /* 4145 * F O R K - m a i n e n t r y p o i n t 4146 */ 4147 /* 4148 * uvmspace_fork: fork a process' main map 4149 * 4150 * => create a new vmspace for child process from parent. 4151 * => parent's map must not be locked. 4152 */ 4153 4154 struct vmspace * 4155 uvmspace_fork(struct vmspace *vm1) 4156 { 4157 struct vmspace *vm2; 4158 struct vm_map *old_map = &vm1->vm_map; 4159 struct vm_map *new_map; 4160 struct vm_map_entry *old_entry; 4161 struct vm_map_entry *new_entry; 4162 UVMHIST_FUNC("uvmspace_fork"); UVMHIST_CALLED(maphist); 4163 4164 vm_map_lock(old_map); 4165 4166 vm2 = uvmspace_alloc(vm_map_min(old_map), vm_map_max(old_map)); 4167 memcpy(&vm2->vm_startcopy, &vm1->vm_startcopy, 4168 (char *) (vm1 + 1) - (char *) &vm1->vm_startcopy); 4169 new_map = &vm2->vm_map; /* XXX */ 4170 4171 old_entry = old_map->header.next; 4172 new_map->size = old_map->size; 4173 4174 /* 4175 * go entry-by-entry 4176 */ 4177 4178 while (old_entry != &old_map->header) { 4179 4180 /* 4181 * first, some sanity checks on the old entry 4182 */ 4183 4184 KASSERT(!UVM_ET_ISSUBMAP(old_entry)); 4185 KASSERT(UVM_ET_ISCOPYONWRITE(old_entry) || 4186 !UVM_ET_ISNEEDSCOPY(old_entry)); 4187 4188 switch (old_entry->inheritance) { 4189 case MAP_INHERIT_NONE: 4190 4191 /* 4192 * drop the mapping, modify size 4193 */ 4194 new_map->size -= old_entry->end - old_entry->start; 4195 break; 4196 4197 case MAP_INHERIT_SHARE: 4198 4199 /* 4200 * share the mapping: this means we want the old and 4201 * new entries to share amaps and backing objects. 4202 */ 4203 /* 4204 * if the old_entry needs a new amap (due to prev fork) 4205 * then we need to allocate it now so that we have 4206 * something we own to share with the new_entry. [in 4207 * other words, we need to clear needs_copy] 4208 */ 4209 4210 if (UVM_ET_ISNEEDSCOPY(old_entry)) { 4211 /* get our own amap, clears needs_copy */ 4212 amap_copy(old_map, old_entry, AMAP_COPY_NOCHUNK, 4213 0, 0); 4214 /* XXXCDC: WAITOK??? */ 4215 } 4216 4217 new_entry = uvm_mapent_alloc(new_map, 0); 4218 /* old_entry -> new_entry */ 4219 uvm_mapent_copy(old_entry, new_entry); 4220 4221 /* new pmap has nothing wired in it */ 4222 new_entry->wired_count = 0; 4223 4224 /* 4225 * gain reference to object backing the map (can't 4226 * be a submap, already checked this case). 4227 */ 4228 4229 if (new_entry->aref.ar_amap) 4230 uvm_map_reference_amap(new_entry, AMAP_SHARED); 4231 4232 if (new_entry->object.uvm_obj && 4233 new_entry->object.uvm_obj->pgops->pgo_reference) 4234 new_entry->object.uvm_obj-> 4235 pgops->pgo_reference( 4236 new_entry->object.uvm_obj); 4237 4238 /* insert entry at end of new_map's entry list */ 4239 uvm_map_entry_link(new_map, new_map->header.prev, 4240 new_entry); 4241 4242 break; 4243 4244 case MAP_INHERIT_COPY: 4245 4246 /* 4247 * copy-on-write the mapping (using mmap's 4248 * MAP_PRIVATE semantics) 4249 * 4250 * allocate new_entry, adjust reference counts. 4251 * (note that new references are read-only). 4252 */ 4253 4254 new_entry = uvm_mapent_alloc(new_map, 0); 4255 /* old_entry -> new_entry */ 4256 uvm_mapent_copy(old_entry, new_entry); 4257 4258 if (new_entry->aref.ar_amap) 4259 uvm_map_reference_amap(new_entry, 0); 4260 4261 if (new_entry->object.uvm_obj && 4262 new_entry->object.uvm_obj->pgops->pgo_reference) 4263 new_entry->object.uvm_obj->pgops->pgo_reference 4264 (new_entry->object.uvm_obj); 4265 4266 /* new pmap has nothing wired in it */ 4267 new_entry->wired_count = 0; 4268 4269 new_entry->etype |= 4270 (UVM_ET_COPYONWRITE|UVM_ET_NEEDSCOPY); 4271 uvm_map_entry_link(new_map, new_map->header.prev, 4272 new_entry); 4273 4274 /* 4275 * the new entry will need an amap. it will either 4276 * need to be copied from the old entry or created 4277 * from scratch (if the old entry does not have an 4278 * amap). can we defer this process until later 4279 * (by setting "needs_copy") or do we need to copy 4280 * the amap now? 4281 * 4282 * we must copy the amap now if any of the following 4283 * conditions hold: 4284 * 1. the old entry has an amap and that amap is 4285 * being shared. this means that the old (parent) 4286 * process is sharing the amap with another 4287 * process. if we do not clear needs_copy here 4288 * we will end up in a situation where both the 4289 * parent and child process are refering to the 4290 * same amap with "needs_copy" set. if the 4291 * parent write-faults, the fault routine will 4292 * clear "needs_copy" in the parent by allocating 4293 * a new amap. this is wrong because the 4294 * parent is supposed to be sharing the old amap 4295 * and the new amap will break that. 4296 * 4297 * 2. if the old entry has an amap and a non-zero 4298 * wire count then we are going to have to call 4299 * amap_cow_now to avoid page faults in the 4300 * parent process. since amap_cow_now requires 4301 * "needs_copy" to be clear we might as well 4302 * clear it here as well. 4303 * 4304 */ 4305 4306 if (old_entry->aref.ar_amap != NULL) { 4307 if ((amap_flags(old_entry->aref.ar_amap) & 4308 AMAP_SHARED) != 0 || 4309 VM_MAPENT_ISWIRED(old_entry)) { 4310 4311 amap_copy(new_map, new_entry, 4312 AMAP_COPY_NOCHUNK, 0, 0); 4313 /* XXXCDC: M_WAITOK ... ok? */ 4314 } 4315 } 4316 4317 /* 4318 * if the parent's entry is wired down, then the 4319 * parent process does not want page faults on 4320 * access to that memory. this means that we 4321 * cannot do copy-on-write because we can't write 4322 * protect the old entry. in this case we 4323 * resolve all copy-on-write faults now, using 4324 * amap_cow_now. note that we have already 4325 * allocated any needed amap (above). 4326 */ 4327 4328 if (VM_MAPENT_ISWIRED(old_entry)) { 4329 4330 /* 4331 * resolve all copy-on-write faults now 4332 * (note that there is nothing to do if 4333 * the old mapping does not have an amap). 4334 */ 4335 if (old_entry->aref.ar_amap) 4336 amap_cow_now(new_map, new_entry); 4337 4338 } else { 4339 4340 /* 4341 * setup mappings to trigger copy-on-write faults 4342 * we must write-protect the parent if it has 4343 * an amap and it is not already "needs_copy"... 4344 * if it is already "needs_copy" then the parent 4345 * has already been write-protected by a previous 4346 * fork operation. 4347 */ 4348 4349 if (old_entry->aref.ar_amap && 4350 !UVM_ET_ISNEEDSCOPY(old_entry)) { 4351 if (old_entry->max_protection & VM_PROT_WRITE) { 4352 pmap_protect(old_map->pmap, 4353 old_entry->start, 4354 old_entry->end, 4355 old_entry->protection & 4356 ~VM_PROT_WRITE); 4357 pmap_update(old_map->pmap); 4358 } 4359 old_entry->etype |= UVM_ET_NEEDSCOPY; 4360 } 4361 } 4362 break; 4363 } /* end of switch statement */ 4364 old_entry = old_entry->next; 4365 } 4366 4367 vm_map_unlock(old_map); 4368 4369 #ifdef SYSVSHM 4370 if (vm1->vm_shm) 4371 shmfork(vm1, vm2); 4372 #endif 4373 4374 #ifdef PMAP_FORK 4375 pmap_fork(vm1->vm_map.pmap, vm2->vm_map.pmap); 4376 #endif 4377 4378 UVMHIST_LOG(maphist,"<- done",0,0,0,0); 4379 return (vm2); 4380 } 4381 4382 4383 /* 4384 * in-kernel map entry allocation. 4385 */ 4386 4387 struct uvm_kmapent_hdr { 4388 LIST_ENTRY(uvm_kmapent_hdr) ukh_listq; 4389 int ukh_nused; 4390 struct vm_map_entry *ukh_freelist; 4391 struct vm_map *ukh_map; 4392 struct vm_map_entry ukh_entries[0]; 4393 }; 4394 4395 #define UVM_KMAPENT_CHUNK \ 4396 ((PAGE_SIZE - sizeof(struct uvm_kmapent_hdr)) \ 4397 / sizeof(struct vm_map_entry)) 4398 4399 #define UVM_KHDR_FIND(entry) \ 4400 ((struct uvm_kmapent_hdr *)(((vaddr_t)entry) & ~PAGE_MASK)) 4401 4402 4403 #ifdef DIAGNOSTIC 4404 static struct vm_map * 4405 uvm_kmapent_map(struct vm_map_entry *entry) 4406 { 4407 const struct uvm_kmapent_hdr *ukh; 4408 4409 ukh = UVM_KHDR_FIND(entry); 4410 return ukh->ukh_map; 4411 } 4412 #endif 4413 4414 static inline struct vm_map_entry * 4415 uvm_kmapent_get(struct uvm_kmapent_hdr *ukh) 4416 { 4417 struct vm_map_entry *entry; 4418 4419 KASSERT(ukh->ukh_nused <= UVM_KMAPENT_CHUNK); 4420 KASSERT(ukh->ukh_nused >= 0); 4421 4422 entry = ukh->ukh_freelist; 4423 if (entry) { 4424 KASSERT((entry->flags & (UVM_MAP_KERNEL | UVM_MAP_KMAPENT)) 4425 == UVM_MAP_KERNEL); 4426 ukh->ukh_freelist = entry->next; 4427 ukh->ukh_nused++; 4428 KASSERT(ukh->ukh_nused <= UVM_KMAPENT_CHUNK); 4429 } else { 4430 KASSERT(ukh->ukh_nused == UVM_KMAPENT_CHUNK); 4431 } 4432 4433 return entry; 4434 } 4435 4436 static inline void 4437 uvm_kmapent_put(struct uvm_kmapent_hdr *ukh, struct vm_map_entry *entry) 4438 { 4439 4440 KASSERT((entry->flags & (UVM_MAP_KERNEL | UVM_MAP_KMAPENT)) 4441 == UVM_MAP_KERNEL); 4442 KASSERT(ukh->ukh_nused <= UVM_KMAPENT_CHUNK); 4443 KASSERT(ukh->ukh_nused > 0); 4444 KASSERT(ukh->ukh_freelist != NULL || 4445 ukh->ukh_nused == UVM_KMAPENT_CHUNK); 4446 KASSERT(ukh->ukh_freelist == NULL || 4447 ukh->ukh_nused < UVM_KMAPENT_CHUNK); 4448 4449 ukh->ukh_nused--; 4450 entry->next = ukh->ukh_freelist; 4451 ukh->ukh_freelist = entry; 4452 } 4453 4454 /* 4455 * uvm_kmapent_alloc: allocate a map entry for in-kernel map 4456 */ 4457 4458 static struct vm_map_entry * 4459 uvm_kmapent_alloc(struct vm_map *map, int flags) 4460 { 4461 struct vm_page *pg; 4462 struct uvm_map_args args; 4463 struct uvm_kmapent_hdr *ukh; 4464 struct vm_map_entry *entry; 4465 uvm_flag_t mapflags = UVM_MAPFLAG(UVM_PROT_ALL, UVM_PROT_ALL, 4466 UVM_INH_NONE, UVM_ADV_RANDOM, flags | UVM_FLAG_NOMERGE); 4467 vaddr_t va; 4468 int error; 4469 int i; 4470 4471 KDASSERT(UVM_KMAPENT_CHUNK > 2); 4472 KDASSERT(kernel_map != NULL); 4473 KASSERT(vm_map_pmap(map) == pmap_kernel()); 4474 4475 UVMMAP_EVCNT_INCR(uke_alloc); 4476 entry = NULL; 4477 again: 4478 /* 4479 * try to grab an entry from freelist. 4480 */ 4481 mutex_spin_enter(&uvm_kentry_lock); 4482 ukh = LIST_FIRST(&vm_map_to_kernel(map)->vmk_kentry_free); 4483 if (ukh) { 4484 entry = uvm_kmapent_get(ukh); 4485 if (ukh->ukh_nused == UVM_KMAPENT_CHUNK) 4486 LIST_REMOVE(ukh, ukh_listq); 4487 } 4488 mutex_spin_exit(&uvm_kentry_lock); 4489 4490 if (entry) 4491 return entry; 4492 4493 /* 4494 * there's no free entry for this vm_map. 4495 * now we need to allocate some vm_map_entry. 4496 * for simplicity, always allocate one page chunk of them at once. 4497 */ 4498 4499 pg = uvm_pagealloc(NULL, 0, NULL, 0); 4500 if (__predict_false(pg == NULL)) { 4501 if (flags & UVM_FLAG_NOWAIT) 4502 return NULL; 4503 uvm_wait("kme_alloc"); 4504 goto again; 4505 } 4506 4507 error = uvm_map_prepare(map, 0, PAGE_SIZE, NULL, UVM_UNKNOWN_OFFSET, 4508 0, mapflags, &args); 4509 if (error) { 4510 uvm_pagefree(pg); 4511 return NULL; 4512 } 4513 4514 va = args.uma_start; 4515 4516 pmap_kenter_pa(va, VM_PAGE_TO_PHYS(pg), VM_PROT_READ|VM_PROT_WRITE); 4517 pmap_update(vm_map_pmap(map)); 4518 4519 ukh = (void *)va; 4520 4521 /* 4522 * use the first entry for ukh itsself. 4523 */ 4524 4525 entry = &ukh->ukh_entries[0]; 4526 entry->flags = UVM_MAP_KERNEL | UVM_MAP_KMAPENT; 4527 error = uvm_map_enter(map, &args, entry); 4528 KASSERT(error == 0); 4529 4530 ukh->ukh_nused = UVM_KMAPENT_CHUNK; 4531 ukh->ukh_map = map; 4532 ukh->ukh_freelist = NULL; 4533 for (i = UVM_KMAPENT_CHUNK - 1; i >= 2; i--) { 4534 struct vm_map_entry *xentry = &ukh->ukh_entries[i]; 4535 4536 xentry->flags = UVM_MAP_KERNEL; 4537 uvm_kmapent_put(ukh, xentry); 4538 } 4539 KASSERT(ukh->ukh_nused == 2); 4540 4541 mutex_spin_enter(&uvm_kentry_lock); 4542 LIST_INSERT_HEAD(&vm_map_to_kernel(map)->vmk_kentry_free, 4543 ukh, ukh_listq); 4544 mutex_spin_exit(&uvm_kentry_lock); 4545 4546 /* 4547 * return second entry. 4548 */ 4549 4550 entry = &ukh->ukh_entries[1]; 4551 entry->flags = UVM_MAP_KERNEL; 4552 UVMMAP_EVCNT_INCR(ukh_alloc); 4553 return entry; 4554 } 4555 4556 /* 4557 * uvm_mapent_free: free map entry for in-kernel map 4558 */ 4559 4560 static void 4561 uvm_kmapent_free(struct vm_map_entry *entry) 4562 { 4563 struct uvm_kmapent_hdr *ukh; 4564 struct vm_page *pg; 4565 struct vm_map *map; 4566 struct pmap *pmap; 4567 vaddr_t va; 4568 paddr_t pa; 4569 struct vm_map_entry *deadentry; 4570 4571 UVMMAP_EVCNT_INCR(uke_free); 4572 ukh = UVM_KHDR_FIND(entry); 4573 map = ukh->ukh_map; 4574 4575 mutex_spin_enter(&uvm_kentry_lock); 4576 uvm_kmapent_put(ukh, entry); 4577 if (ukh->ukh_nused > 1) { 4578 if (ukh->ukh_nused == UVM_KMAPENT_CHUNK - 1) 4579 LIST_INSERT_HEAD( 4580 &vm_map_to_kernel(map)->vmk_kentry_free, 4581 ukh, ukh_listq); 4582 mutex_spin_exit(&uvm_kentry_lock); 4583 return; 4584 } 4585 4586 /* 4587 * now we can free this ukh. 4588 * 4589 * however, keep an empty ukh to avoid ping-pong. 4590 */ 4591 4592 if (LIST_FIRST(&vm_map_to_kernel(map)->vmk_kentry_free) == ukh && 4593 LIST_NEXT(ukh, ukh_listq) == NULL) { 4594 mutex_spin_exit(&uvm_kentry_lock); 4595 return; 4596 } 4597 LIST_REMOVE(ukh, ukh_listq); 4598 mutex_spin_exit(&uvm_kentry_lock); 4599 4600 KASSERT(ukh->ukh_nused == 1); 4601 4602 /* 4603 * remove map entry for ukh itsself. 4604 */ 4605 4606 va = (vaddr_t)ukh; 4607 KASSERT((va & PAGE_MASK) == 0); 4608 vm_map_lock(map); 4609 uvm_unmap_remove(map, va, va + PAGE_SIZE, &deadentry, NULL, 0); 4610 KASSERT(deadentry->flags & UVM_MAP_KERNEL); 4611 KASSERT(deadentry->flags & UVM_MAP_KMAPENT); 4612 KASSERT(deadentry->next == NULL); 4613 KASSERT(deadentry == &ukh->ukh_entries[0]); 4614 4615 /* 4616 * unmap the page from pmap and free it. 4617 */ 4618 4619 pmap = vm_map_pmap(map); 4620 KASSERT(pmap == pmap_kernel()); 4621 if (!pmap_extract(pmap, va, &pa)) 4622 panic("%s: no mapping", __func__); 4623 pmap_kremove(va, PAGE_SIZE); 4624 pmap_update(vm_map_pmap(map)); 4625 vm_map_unlock(map); 4626 pg = PHYS_TO_VM_PAGE(pa); 4627 uvm_pagefree(pg); 4628 UVMMAP_EVCNT_INCR(ukh_free); 4629 } 4630 4631 static vsize_t 4632 uvm_kmapent_overhead(vsize_t size) 4633 { 4634 4635 /* 4636 * - the max number of unmerged entries is howmany(size, PAGE_SIZE) 4637 * as the min allocation unit is PAGE_SIZE. 4638 * - UVM_KMAPENT_CHUNK "kmapent"s are allocated from a page. 4639 * one of them are used to map the page itself. 4640 */ 4641 4642 return howmany(howmany(size, PAGE_SIZE), (UVM_KMAPENT_CHUNK - 1)) * 4643 PAGE_SIZE; 4644 } 4645 4646 /* 4647 * map entry reservation 4648 */ 4649 4650 /* 4651 * uvm_mapent_reserve: reserve map entries for clipping before locking map. 4652 * 4653 * => needed when unmapping entries allocated without UVM_FLAG_QUANTUM. 4654 * => caller shouldn't hold map locked. 4655 */ 4656 int 4657 uvm_mapent_reserve(struct vm_map *map, struct uvm_mapent_reservation *umr, 4658 int nentries, int flags) 4659 { 4660 4661 umr->umr_nentries = 0; 4662 4663 if ((flags & UVM_FLAG_QUANTUM) != 0) 4664 return 0; 4665 4666 if (!VM_MAP_USE_KMAPENT(map)) 4667 return 0; 4668 4669 while (nentries--) { 4670 struct vm_map_entry *ent; 4671 ent = uvm_kmapent_alloc(map, flags); 4672 if (!ent) { 4673 uvm_mapent_unreserve(map, umr); 4674 return ENOMEM; 4675 } 4676 UMR_PUTENTRY(umr, ent); 4677 } 4678 4679 return 0; 4680 } 4681 4682 /* 4683 * uvm_mapent_unreserve: 4684 * 4685 * => caller shouldn't hold map locked. 4686 * => never fail or sleep. 4687 */ 4688 void 4689 uvm_mapent_unreserve(struct vm_map *map, struct uvm_mapent_reservation *umr) 4690 { 4691 4692 while (!UMR_EMPTY(umr)) 4693 uvm_kmapent_free(UMR_GETENTRY(umr)); 4694 } 4695 4696 /* 4697 * uvm_mapent_trymerge: try to merge an entry with its neighbors. 4698 * 4699 * => called with map locked. 4700 * => return non zero if successfully merged. 4701 */ 4702 4703 int 4704 uvm_mapent_trymerge(struct vm_map *map, struct vm_map_entry *entry, int flags) 4705 { 4706 struct uvm_object *uobj; 4707 struct vm_map_entry *next; 4708 struct vm_map_entry *prev; 4709 vsize_t size; 4710 int merged = 0; 4711 bool copying; 4712 int newetype; 4713 4714 if (VM_MAP_USE_KMAPENT(map)) { 4715 return 0; 4716 } 4717 if (entry->aref.ar_amap != NULL) { 4718 return 0; 4719 } 4720 if ((entry->flags & UVM_MAP_NOMERGE) != 0) { 4721 return 0; 4722 } 4723 4724 uobj = entry->object.uvm_obj; 4725 size = entry->end - entry->start; 4726 copying = (flags & UVM_MERGE_COPYING) != 0; 4727 newetype = copying ? (entry->etype & ~UVM_ET_NEEDSCOPY) : entry->etype; 4728 4729 next = entry->next; 4730 if (next != &map->header && 4731 next->start == entry->end && 4732 ((copying && next->aref.ar_amap != NULL && 4733 amap_refs(next->aref.ar_amap) == 1) || 4734 (!copying && next->aref.ar_amap == NULL)) && 4735 UVM_ET_ISCOMPATIBLE(next, newetype, 4736 uobj, entry->flags, entry->protection, 4737 entry->max_protection, entry->inheritance, entry->advice, 4738 entry->wired_count) && 4739 (uobj == NULL || entry->offset + size == next->offset)) { 4740 int error; 4741 4742 if (copying) { 4743 error = amap_extend(next, size, 4744 AMAP_EXTEND_NOWAIT|AMAP_EXTEND_BACKWARDS); 4745 } else { 4746 error = 0; 4747 } 4748 if (error == 0) { 4749 if (uobj) { 4750 if (uobj->pgops->pgo_detach) { 4751 uobj->pgops->pgo_detach(uobj); 4752 } 4753 } 4754 4755 entry->end = next->end; 4756 clear_hints(map, next); 4757 uvm_map_entry_unlink(map, next); 4758 if (copying) { 4759 entry->aref = next->aref; 4760 entry->etype &= ~UVM_ET_NEEDSCOPY; 4761 } 4762 uvm_map_check(map, "trymerge forwardmerge"); 4763 uvm_mapent_free_merged(map, next); 4764 merged++; 4765 } 4766 } 4767 4768 prev = entry->prev; 4769 if (prev != &map->header && 4770 prev->end == entry->start && 4771 ((copying && !merged && prev->aref.ar_amap != NULL && 4772 amap_refs(prev->aref.ar_amap) == 1) || 4773 (!copying && prev->aref.ar_amap == NULL)) && 4774 UVM_ET_ISCOMPATIBLE(prev, newetype, 4775 uobj, entry->flags, entry->protection, 4776 entry->max_protection, entry->inheritance, entry->advice, 4777 entry->wired_count) && 4778 (uobj == NULL || 4779 prev->offset + prev->end - prev->start == entry->offset)) { 4780 int error; 4781 4782 if (copying) { 4783 error = amap_extend(prev, size, 4784 AMAP_EXTEND_NOWAIT|AMAP_EXTEND_FORWARDS); 4785 } else { 4786 error = 0; 4787 } 4788 if (error == 0) { 4789 if (uobj) { 4790 if (uobj->pgops->pgo_detach) { 4791 uobj->pgops->pgo_detach(uobj); 4792 } 4793 entry->offset = prev->offset; 4794 } 4795 4796 entry->start = prev->start; 4797 clear_hints(map, prev); 4798 uvm_map_entry_unlink(map, prev); 4799 if (copying) { 4800 entry->aref = prev->aref; 4801 entry->etype &= ~UVM_ET_NEEDSCOPY; 4802 } 4803 uvm_map_check(map, "trymerge backmerge"); 4804 uvm_mapent_free_merged(map, prev); 4805 merged++; 4806 } 4807 } 4808 4809 return merged; 4810 } 4811 4812 #if defined(DDB) 4813 4814 /* 4815 * DDB hooks 4816 */ 4817 4818 /* 4819 * uvm_map_printit: actually prints the map 4820 */ 4821 4822 void 4823 uvm_map_printit(struct vm_map *map, bool full, 4824 void (*pr)(const char *, ...)) 4825 { 4826 struct vm_map_entry *entry; 4827 4828 (*pr)("MAP %p: [0x%lx->0x%lx]\n", map, vm_map_min(map), 4829 vm_map_max(map)); 4830 (*pr)("\t#ent=%d, sz=%d, ref=%d, version=%d, flags=0x%x\n", 4831 map->nentries, map->size, map->ref_count, map->timestamp, 4832 map->flags); 4833 (*pr)("\tpmap=%p(resident=%ld, wired=%ld)\n", map->pmap, 4834 pmap_resident_count(map->pmap), pmap_wired_count(map->pmap)); 4835 if (!full) 4836 return; 4837 for (entry = map->header.next; entry != &map->header; 4838 entry = entry->next) { 4839 (*pr)(" - %p: 0x%lx->0x%lx: obj=%p/0x%llx, amap=%p/%d\n", 4840 entry, entry->start, entry->end, entry->object.uvm_obj, 4841 (long long)entry->offset, entry->aref.ar_amap, 4842 entry->aref.ar_pageoff); 4843 (*pr)( 4844 "\tsubmap=%c, cow=%c, nc=%c, prot(max)=%d/%d, inh=%d, " 4845 "wc=%d, adv=%d\n", 4846 (entry->etype & UVM_ET_SUBMAP) ? 'T' : 'F', 4847 (entry->etype & UVM_ET_COPYONWRITE) ? 'T' : 'F', 4848 (entry->etype & UVM_ET_NEEDSCOPY) ? 'T' : 'F', 4849 entry->protection, entry->max_protection, 4850 entry->inheritance, entry->wired_count, entry->advice); 4851 } 4852 } 4853 4854 /* 4855 * uvm_object_printit: actually prints the object 4856 */ 4857 4858 void 4859 uvm_object_printit(struct uvm_object *uobj, bool full, 4860 void (*pr)(const char *, ...)) 4861 { 4862 struct vm_page *pg; 4863 int cnt = 0; 4864 4865 (*pr)("OBJECT %p: locked=%d, pgops=%p, npages=%d, ", 4866 uobj, mutex_owned(&uobj->vmobjlock), uobj->pgops, uobj->uo_npages); 4867 if (UVM_OBJ_IS_KERN_OBJECT(uobj)) 4868 (*pr)("refs=<SYSTEM>\n"); 4869 else 4870 (*pr)("refs=%d\n", uobj->uo_refs); 4871 4872 if (!full) { 4873 return; 4874 } 4875 (*pr)(" PAGES <pg,offset>:\n "); 4876 TAILQ_FOREACH(pg, &uobj->memq, listq) { 4877 cnt++; 4878 (*pr)("<%p,0x%llx> ", pg, (long long)pg->offset); 4879 if ((cnt % 3) == 0) { 4880 (*pr)("\n "); 4881 } 4882 } 4883 if ((cnt % 3) != 0) { 4884 (*pr)("\n"); 4885 } 4886 } 4887 4888 /* 4889 * uvm_page_printit: actually print the page 4890 */ 4891 4892 static const char page_flagbits[] = UVM_PGFLAGBITS; 4893 static const char page_pqflagbits[] = UVM_PQFLAGBITS; 4894 4895 void 4896 uvm_page_printit(struct vm_page *pg, bool full, 4897 void (*pr)(const char *, ...)) 4898 { 4899 struct vm_page *tpg; 4900 struct uvm_object *uobj; 4901 struct pglist *pgl; 4902 char pgbuf[128]; 4903 char pqbuf[128]; 4904 4905 (*pr)("PAGE %p:\n", pg); 4906 bitmask_snprintf(pg->flags, page_flagbits, pgbuf, sizeof(pgbuf)); 4907 bitmask_snprintf(pg->pqflags, page_pqflagbits, pqbuf, sizeof(pqbuf)); 4908 (*pr)(" flags=%s, pqflags=%s, wire_count=%d, pa=0x%lx\n", 4909 pgbuf, pqbuf, pg->wire_count, (long)VM_PAGE_TO_PHYS(pg)); 4910 (*pr)(" uobject=%p, uanon=%p, offset=0x%llx loan_count=%d\n", 4911 pg->uobject, pg->uanon, (long long)pg->offset, pg->loan_count); 4912 #if defined(UVM_PAGE_TRKOWN) 4913 if (pg->flags & PG_BUSY) 4914 (*pr)(" owning process = %d, tag=%s\n", 4915 pg->owner, pg->owner_tag); 4916 else 4917 (*pr)(" page not busy, no owner\n"); 4918 #else 4919 (*pr)(" [page ownership tracking disabled]\n"); 4920 #endif 4921 4922 if (!full) 4923 return; 4924 4925 /* cross-verify object/anon */ 4926 if ((pg->pqflags & PQ_FREE) == 0) { 4927 if (pg->pqflags & PQ_ANON) { 4928 if (pg->uanon == NULL || pg->uanon->an_page != pg) 4929 (*pr)(" >>> ANON DOES NOT POINT HERE <<< (%p)\n", 4930 (pg->uanon) ? pg->uanon->an_page : NULL); 4931 else 4932 (*pr)(" anon backpointer is OK\n"); 4933 } else { 4934 uobj = pg->uobject; 4935 if (uobj) { 4936 (*pr)(" checking object list\n"); 4937 TAILQ_FOREACH(tpg, &uobj->memq, listq) { 4938 if (tpg == pg) { 4939 break; 4940 } 4941 } 4942 if (tpg) 4943 (*pr)(" page found on object list\n"); 4944 else 4945 (*pr)(" >>> PAGE NOT FOUND ON OBJECT LIST! <<<\n"); 4946 } 4947 } 4948 } 4949 4950 /* cross-verify page queue */ 4951 if (pg->pqflags & PQ_FREE) { 4952 int fl = uvm_page_lookup_freelist(pg); 4953 int color = VM_PGCOLOR_BUCKET(pg); 4954 pgl = &uvm.page_free[fl].pgfl_buckets[color].pgfl_queues[ 4955 ((pg)->flags & PG_ZERO) ? PGFL_ZEROS : PGFL_UNKNOWN]; 4956 } else { 4957 pgl = NULL; 4958 } 4959 4960 if (pgl) { 4961 (*pr)(" checking pageq list\n"); 4962 TAILQ_FOREACH(tpg, pgl, pageq) { 4963 if (tpg == pg) { 4964 break; 4965 } 4966 } 4967 if (tpg) 4968 (*pr)(" page found on pageq list\n"); 4969 else 4970 (*pr)(" >>> PAGE NOT FOUND ON PAGEQ LIST! <<<\n"); 4971 } 4972 } 4973 4974 /* 4975 * uvm_pages_printthem - print a summary of all managed pages 4976 */ 4977 4978 void 4979 uvm_page_printall(void (*pr)(const char *, ...)) 4980 { 4981 unsigned i; 4982 struct vm_page *pg; 4983 4984 (*pr)("%18s %4s %4s %18s %18s" 4985 #ifdef UVM_PAGE_TRKOWN 4986 " OWNER" 4987 #endif 4988 "\n", "PAGE", "FLAG", "PQ", "UOBJECT", "UANON"); 4989 for (i = 0; i < vm_nphysseg; i++) { 4990 for (pg = vm_physmem[i].pgs; pg <= vm_physmem[i].lastpg; pg++) { 4991 (*pr)("%18p %04x %04x %18p %18p", 4992 pg, pg->flags, pg->pqflags, pg->uobject, 4993 pg->uanon); 4994 #ifdef UVM_PAGE_TRKOWN 4995 if (pg->flags & PG_BUSY) 4996 (*pr)(" %d [%s]", pg->owner, pg->owner_tag); 4997 #endif 4998 (*pr)("\n"); 4999 } 5000 } 5001 } 5002 5003 #endif 5004 5005 /* 5006 * uvm_map_create: create map 5007 */ 5008 5009 struct vm_map * 5010 uvm_map_create(pmap_t pmap, vaddr_t vmin, vaddr_t vmax, int flags) 5011 { 5012 struct vm_map *result; 5013 5014 MALLOC(result, struct vm_map *, sizeof(struct vm_map), 5015 M_VMMAP, M_WAITOK); 5016 uvm_map_setup(result, vmin, vmax, flags); 5017 result->pmap = pmap; 5018 return(result); 5019 } 5020 5021 /* 5022 * uvm_map_setup: init map 5023 * 5024 * => map must not be in service yet. 5025 */ 5026 5027 void 5028 uvm_map_setup(struct vm_map *map, vaddr_t vmin, vaddr_t vmax, int flags) 5029 { 5030 int ipl; 5031 5032 RB_INIT(&map->rbhead); 5033 map->header.next = map->header.prev = &map->header; 5034 map->nentries = 0; 5035 map->size = 0; 5036 map->ref_count = 1; 5037 vm_map_setmin(map, vmin); 5038 vm_map_setmax(map, vmax); 5039 map->flags = flags; 5040 map->first_free = &map->header; 5041 map->hint = &map->header; 5042 map->timestamp = 0; 5043 map->busy = NULL; 5044 5045 if ((flags & VM_MAP_INTRSAFE) != 0) { 5046 ipl = IPL_VM; 5047 } else { 5048 ipl = IPL_NONE; 5049 } 5050 5051 rw_init(&map->lock); 5052 cv_init(&map->cv, "vm_map"); 5053 mutex_init(&map->misc_lock, MUTEX_DRIVER, ipl); 5054 mutex_init(&map->mutex, MUTEX_DRIVER, ipl); 5055 } 5056 5057 5058 /* 5059 * U N M A P - m a i n e n t r y p o i n t 5060 */ 5061 5062 /* 5063 * uvm_unmap1: remove mappings from a vm_map (from "start" up to "stop") 5064 * 5065 * => caller must check alignment and size 5066 * => map must be unlocked (we will lock it) 5067 * => flags is UVM_FLAG_QUANTUM or 0. 5068 */ 5069 5070 void 5071 uvm_unmap1(struct vm_map *map, vaddr_t start, vaddr_t end, int flags) 5072 { 5073 struct vm_map_entry *dead_entries; 5074 struct uvm_mapent_reservation umr; 5075 UVMHIST_FUNC("uvm_unmap"); UVMHIST_CALLED(maphist); 5076 5077 UVMHIST_LOG(maphist, " (map=0x%x, start=0x%x, end=0x%x)", 5078 map, start, end, 0); 5079 if (map == kernel_map) { 5080 LOCKDEBUG_MEM_CHECK((void *)start, end - start); 5081 } 5082 /* 5083 * work now done by helper functions. wipe the pmap's and then 5084 * detach from the dead entries... 5085 */ 5086 uvm_mapent_reserve(map, &umr, 2, flags); 5087 vm_map_lock(map); 5088 uvm_unmap_remove(map, start, end, &dead_entries, &umr, flags); 5089 vm_map_unlock(map); 5090 uvm_mapent_unreserve(map, &umr); 5091 5092 if (dead_entries != NULL) 5093 uvm_unmap_detach(dead_entries, 0); 5094 5095 UVMHIST_LOG(maphist, "<- done", 0,0,0,0); 5096 } 5097 5098 5099 /* 5100 * uvm_map_reference: add reference to a map 5101 * 5102 * => map need not be locked (we use misc_lock). 5103 */ 5104 5105 void 5106 uvm_map_reference(struct vm_map *map) 5107 { 5108 mutex_enter(&map->misc_lock); 5109 map->ref_count++; 5110 mutex_exit(&map->misc_lock); 5111 } 5112 5113 struct vm_map_kernel * 5114 vm_map_to_kernel(struct vm_map *map) 5115 { 5116 5117 KASSERT(VM_MAP_IS_KERNEL(map)); 5118 5119 return (struct vm_map_kernel *)map; 5120 } 5121 5122 bool 5123 vm_map_starved_p(struct vm_map *map) 5124 { 5125 5126 if ((map->flags & VM_MAP_WANTVA) != 0) { 5127 return true; 5128 } 5129 /* XXX */ 5130 if ((vm_map_max(map) - vm_map_min(map)) / 16 * 15 < map->size) { 5131 return true; 5132 } 5133 return false; 5134 } 5135 5136 #if defined(DDB) 5137 void 5138 uvm_whatis(uintptr_t addr, void (*pr)(const char *, ...)) 5139 { 5140 struct vm_map *map; 5141 5142 for (map = kernel_map;;) { 5143 struct vm_map_entry *entry; 5144 5145 if (!uvm_map_lookup_entry_bytree(map, (vaddr_t)addr, &entry)) { 5146 break; 5147 } 5148 (*pr)("%p is %p+%zu from VMMAP %p\n", 5149 (void *)addr, (void *)entry->start, 5150 (size_t)(addr - (uintptr_t)entry->start), map); 5151 if (!UVM_ET_ISSUBMAP(entry)) { 5152 break; 5153 } 5154 map = entry->object.sub_map; 5155 } 5156 } 5157 #endif /* defined(DDB) */ 5158