1 /* $NetBSD: uvm_map.c,v 1.183 2005/01/23 15:58:13 chs 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 1.183 2005/01/23 15:58:13 chs 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 91 #ifdef SYSVSHM 92 #include <sys/shm.h> 93 #endif 94 95 #define UVM_MAP 96 #include <uvm/uvm.h> 97 #undef RB_AUGMENT 98 #define RB_AUGMENT(x) uvm_rb_augment(x) 99 100 #ifdef DDB 101 #include <uvm/uvm_ddb.h> 102 #endif 103 104 #ifndef UVMMAP_NOCOUNTERS 105 #include <sys/device.h> 106 struct evcnt map_ubackmerge = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 107 "uvmmap", "ubackmerge"); 108 struct evcnt map_uforwmerge = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 109 "uvmmap", "uforwmerge"); 110 struct evcnt map_ubimerge = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 111 "uvmmap", "ubimerge"); 112 struct evcnt map_unomerge = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 113 "uvmmap", "unomerge"); 114 struct evcnt map_kbackmerge = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 115 "uvmmap", "kbackmerge"); 116 struct evcnt map_kforwmerge = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 117 "uvmmap", "kforwmerge"); 118 struct evcnt map_kbimerge = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 119 "uvmmap", "kbimerge"); 120 struct evcnt map_knomerge = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 121 "uvmmap", "knomerge"); 122 struct evcnt uvm_map_call = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 123 "uvmmap", "map_call"); 124 struct evcnt uvm_mlk_call = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 125 "uvmmap", "mlk_call"); 126 struct evcnt uvm_mlk_hint = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 127 "uvmmap", "mlk_hint"); 128 129 EVCNT_ATTACH_STATIC(map_ubackmerge); 130 EVCNT_ATTACH_STATIC(map_uforwmerge); 131 EVCNT_ATTACH_STATIC(map_ubimerge); 132 EVCNT_ATTACH_STATIC(map_unomerge); 133 EVCNT_ATTACH_STATIC(map_kbackmerge); 134 EVCNT_ATTACH_STATIC(map_kforwmerge); 135 EVCNT_ATTACH_STATIC(map_kbimerge); 136 EVCNT_ATTACH_STATIC(map_knomerge); 137 EVCNT_ATTACH_STATIC(uvm_map_call); 138 EVCNT_ATTACH_STATIC(uvm_mlk_call); 139 EVCNT_ATTACH_STATIC(uvm_mlk_hint); 140 141 #define UVMCNT_INCR(ev) ev.ev_count++ 142 #define UVMCNT_DECR(ev) ev.ev_count-- 143 #else 144 #define UVMCNT_INCR(ev) 145 #define UVMCNT_DECR(ev) 146 #endif 147 148 const char vmmapbsy[] = "vmmapbsy"; 149 150 /* 151 * pool for vmspace structures. 152 */ 153 154 POOL_INIT(uvm_vmspace_pool, sizeof(struct vmspace), 0, 0, 0, "vmsppl", 155 &pool_allocator_nointr); 156 157 /* 158 * pool for dynamically-allocated map entries. 159 */ 160 161 POOL_INIT(uvm_map_entry_pool, sizeof(struct vm_map_entry), 0, 0, 0, "vmmpepl", 162 &pool_allocator_nointr); 163 164 MALLOC_DEFINE(M_VMMAP, "VM map", "VM map structures"); 165 MALLOC_DEFINE(M_VMPMAP, "VM pmap", "VM pmap"); 166 167 #ifdef PMAP_GROWKERNEL 168 /* 169 * This global represents the end of the kernel virtual address 170 * space. If we want to exceed this, we must grow the kernel 171 * virtual address space dynamically. 172 * 173 * Note, this variable is locked by kernel_map's lock. 174 */ 175 vaddr_t uvm_maxkaddr; 176 #endif 177 178 /* 179 * macros 180 */ 181 182 /* 183 * VM_MAP_USE_KMAPENT: determine if uvm_kmapent_alloc/free is used 184 * for the vm_map. 185 */ 186 extern struct vm_map *pager_map; /* XXX */ 187 #define VM_MAP_USE_KMAPENT(map) \ 188 (((map)->flags & VM_MAP_INTRSAFE) || (map) == kernel_map) 189 190 /* 191 * uvm_map_entry_link: insert entry into a map 192 * 193 * => map must be locked 194 */ 195 #define uvm_map_entry_link(map, after_where, entry) do { \ 196 KASSERT(entry->start < entry->end); \ 197 (map)->nentries++; \ 198 (entry)->prev = (after_where); \ 199 (entry)->next = (after_where)->next; \ 200 (entry)->prev->next = (entry); \ 201 (entry)->next->prev = (entry); \ 202 uvm_rb_insert((map), (entry)); \ 203 } while (/*CONSTCOND*/ 0) 204 205 /* 206 * uvm_map_entry_unlink: remove entry from a map 207 * 208 * => map must be locked 209 */ 210 #define uvm_map_entry_unlink(map, entry) do { \ 211 (map)->nentries--; \ 212 (entry)->next->prev = (entry)->prev; \ 213 (entry)->prev->next = (entry)->next; \ 214 uvm_rb_remove((map), (entry)); \ 215 } while (/*CONSTCOND*/ 0) 216 217 /* 218 * SAVE_HINT: saves the specified entry as the hint for future lookups. 219 * 220 * => map need not be locked (protected by hint_lock). 221 */ 222 #define SAVE_HINT(map,check,value) do { \ 223 simple_lock(&(map)->hint_lock); \ 224 if ((map)->hint == (check)) \ 225 (map)->hint = (value); \ 226 simple_unlock(&(map)->hint_lock); \ 227 } while (/*CONSTCOND*/ 0) 228 229 /* 230 * VM_MAP_RANGE_CHECK: check and correct range 231 * 232 * => map must at least be read locked 233 */ 234 235 #define VM_MAP_RANGE_CHECK(map, start, end) do { \ 236 if (start < vm_map_min(map)) \ 237 start = vm_map_min(map); \ 238 if (end > vm_map_max(map)) \ 239 end = vm_map_max(map); \ 240 if (start > end) \ 241 start = end; \ 242 } while (/*CONSTCOND*/ 0) 243 244 /* 245 * local prototypes 246 */ 247 248 static struct vm_map_entry * 249 uvm_mapent_alloc(struct vm_map *, int); 250 static struct vm_map_entry * 251 uvm_mapent_alloc_split(struct vm_map *, 252 const struct vm_map_entry *, int, 253 struct uvm_mapent_reservation *); 254 static void uvm_mapent_copy(struct vm_map_entry *, struct vm_map_entry *); 255 static void uvm_mapent_free(struct vm_map_entry *); 256 static struct vm_map_entry * 257 uvm_kmapent_alloc(struct vm_map *, int); 258 static void uvm_kmapent_free(struct vm_map_entry *); 259 static void uvm_map_entry_unwire(struct vm_map *, struct vm_map_entry *); 260 static void uvm_map_reference_amap(struct vm_map_entry *, int); 261 static int uvm_map_space_avail(vaddr_t *, vsize_t, voff_t, vsize_t, int, 262 struct vm_map_entry *); 263 static void uvm_map_unreference_amap(struct vm_map_entry *, int); 264 265 int _uvm_tree_sanity(struct vm_map *, const char *); 266 static vsize_t uvm_rb_subtree_space(const struct vm_map_entry *); 267 268 static __inline int 269 uvm_compare(const struct vm_map_entry *a, const struct vm_map_entry *b) 270 { 271 272 if (a->start < b->start) 273 return (-1); 274 else if (a->start > b->start) 275 return (1); 276 277 return (0); 278 } 279 280 static __inline void 281 uvm_rb_augment(struct vm_map_entry *entry) 282 { 283 284 entry->space = uvm_rb_subtree_space(entry); 285 } 286 287 RB_PROTOTYPE(uvm_tree, vm_map_entry, rb_entry, uvm_compare); 288 289 RB_GENERATE(uvm_tree, vm_map_entry, rb_entry, uvm_compare); 290 291 static __inline vsize_t 292 uvm_rb_space(const struct vm_map *map, const struct vm_map_entry *entry) 293 { 294 /* XXX map is not used */ 295 296 KASSERT(entry->next != NULL); 297 return entry->next->start - entry->end; 298 } 299 300 static vsize_t 301 uvm_rb_subtree_space(const struct vm_map_entry *entry) 302 { 303 vaddr_t space, tmp; 304 305 space = entry->ownspace; 306 if (RB_LEFT(entry, rb_entry)) { 307 tmp = RB_LEFT(entry, rb_entry)->space; 308 if (tmp > space) 309 space = tmp; 310 } 311 312 if (RB_RIGHT(entry, rb_entry)) { 313 tmp = RB_RIGHT(entry, rb_entry)->space; 314 if (tmp > space) 315 space = tmp; 316 } 317 318 return (space); 319 } 320 321 static __inline void 322 uvm_rb_fixup(struct vm_map *map, struct vm_map_entry *entry) 323 { 324 /* We need to traverse to the very top */ 325 do { 326 entry->ownspace = uvm_rb_space(map, entry); 327 entry->space = uvm_rb_subtree_space(entry); 328 } while ((entry = RB_PARENT(entry, rb_entry)) != NULL); 329 } 330 331 static __inline void 332 uvm_rb_insert(struct vm_map *map, struct vm_map_entry *entry) 333 { 334 vaddr_t space = uvm_rb_space(map, entry); 335 struct vm_map_entry *tmp; 336 337 entry->ownspace = entry->space = space; 338 tmp = RB_INSERT(uvm_tree, &(map)->rbhead, entry); 339 #ifdef DIAGNOSTIC 340 if (tmp != NULL) 341 panic("uvm_rb_insert: duplicate entry?"); 342 #endif 343 uvm_rb_fixup(map, entry); 344 if (entry->prev != &map->header) 345 uvm_rb_fixup(map, entry->prev); 346 } 347 348 static __inline void 349 uvm_rb_remove(struct vm_map *map, struct vm_map_entry *entry) 350 { 351 struct vm_map_entry *parent; 352 353 parent = RB_PARENT(entry, rb_entry); 354 RB_REMOVE(uvm_tree, &(map)->rbhead, entry); 355 if (entry->prev != &map->header) 356 uvm_rb_fixup(map, entry->prev); 357 if (parent) 358 uvm_rb_fixup(map, parent); 359 } 360 361 #ifdef DEBUG 362 int uvm_debug_check_rbtree = 0; 363 #define uvm_tree_sanity(x,y) \ 364 if (uvm_debug_check_rbtree) \ 365 _uvm_tree_sanity(x,y) 366 #else 367 #define uvm_tree_sanity(x,y) 368 #endif 369 370 int 371 _uvm_tree_sanity(struct vm_map *map, const char *name) 372 { 373 struct vm_map_entry *tmp, *trtmp; 374 int n = 0, i = 1; 375 376 RB_FOREACH(tmp, uvm_tree, &map->rbhead) { 377 if (tmp->ownspace != uvm_rb_space(map, tmp)) { 378 printf("%s: %d/%d ownspace %lx != %lx %s\n", 379 name, n + 1, map->nentries, 380 (ulong)tmp->ownspace, (ulong)uvm_rb_space(map, tmp), 381 tmp->next == &map->header ? "(last)" : ""); 382 goto error; 383 } 384 } 385 trtmp = NULL; 386 RB_FOREACH(tmp, uvm_tree, &map->rbhead) { 387 if (tmp->space != uvm_rb_subtree_space(tmp)) { 388 printf("%s: space %lx != %lx\n", 389 name, (ulong)tmp->space, 390 (ulong)uvm_rb_subtree_space(tmp)); 391 goto error; 392 } 393 if (trtmp != NULL && trtmp->start >= tmp->start) { 394 printf("%s: corrupt: 0x%lx >= 0x%lx\n", 395 name, trtmp->start, tmp->start); 396 goto error; 397 } 398 n++; 399 400 trtmp = tmp; 401 } 402 403 if (n != map->nentries) { 404 printf("%s: nentries: %d vs %d\n", 405 name, n, map->nentries); 406 goto error; 407 } 408 409 for (tmp = map->header.next; tmp && tmp != &map->header; 410 tmp = tmp->next, i++) { 411 trtmp = RB_FIND(uvm_tree, &map->rbhead, tmp); 412 if (trtmp != tmp) { 413 printf("%s: lookup: %d: %p - %p: %p\n", 414 name, i, tmp, trtmp, 415 RB_PARENT(tmp, rb_entry)); 416 goto error; 417 } 418 } 419 420 return (0); 421 error: 422 #ifdef DDB 423 /* handy breakpoint location for error case */ 424 __asm(".globl treesanity_label\ntreesanity_label:"); 425 #endif 426 return (-1); 427 } 428 429 /* 430 * local inlines 431 */ 432 433 static __inline struct vm_map *uvm_kmapent_map(struct vm_map_entry *); 434 435 /* 436 * uvm_mapent_alloc: allocate a map entry 437 */ 438 439 static __inline struct vm_map_entry * 440 uvm_mapent_alloc(struct vm_map *map, int flags) 441 { 442 struct vm_map_entry *me; 443 int pflags = (flags & UVM_FLAG_NOWAIT) ? PR_NOWAIT : PR_WAITOK; 444 UVMHIST_FUNC("uvm_mapent_alloc"); UVMHIST_CALLED(maphist); 445 446 if (VM_MAP_USE_KMAPENT(map)) { 447 me = uvm_kmapent_alloc(map, flags); 448 } else { 449 me = pool_get(&uvm_map_entry_pool, pflags); 450 if (__predict_false(me == NULL)) 451 return NULL; 452 me->flags = 0; 453 } 454 455 UVMHIST_LOG(maphist, "<- new entry=0x%x [kentry=%d]", me, 456 ((map->flags & VM_MAP_INTRSAFE) != 0 || map == kernel_map), 0, 0); 457 return (me); 458 } 459 460 /* 461 * uvm_mapent_alloc_split: allocate a map entry for clipping. 462 */ 463 464 static __inline struct vm_map_entry * 465 uvm_mapent_alloc_split(struct vm_map *map, 466 const struct vm_map_entry *old_entry, int flags, 467 struct uvm_mapent_reservation *umr) 468 { 469 struct vm_map_entry *me; 470 471 KASSERT(!VM_MAP_USE_KMAPENT(map) || 472 (old_entry->flags & UVM_MAP_QUANTUM) || !UMR_EMPTY(umr)); 473 474 if (old_entry->flags & UVM_MAP_QUANTUM) { 475 int s; 476 struct vm_map_kernel *vmk = vm_map_to_kernel(map); 477 478 s = splvm(); 479 simple_lock(&uvm.kentry_lock); 480 me = vmk->vmk_merged_entries; 481 KASSERT(me); 482 vmk->vmk_merged_entries = me->next; 483 simple_unlock(&uvm.kentry_lock); 484 splx(s); 485 KASSERT(me->flags & UVM_MAP_QUANTUM); 486 } else { 487 me = uvm_mapent_alloc(map, flags); 488 } 489 490 return me; 491 } 492 493 /* 494 * uvm_mapent_free: free map entry 495 */ 496 497 static __inline void 498 uvm_mapent_free(struct vm_map_entry *me) 499 { 500 UVMHIST_FUNC("uvm_mapent_free"); UVMHIST_CALLED(maphist); 501 502 UVMHIST_LOG(maphist,"<- freeing map entry=0x%x [flags=%d]", 503 me, me->flags, 0, 0); 504 if (me->flags & UVM_MAP_KERNEL) { 505 uvm_kmapent_free(me); 506 } else { 507 pool_put(&uvm_map_entry_pool, me); 508 } 509 } 510 511 /* 512 * uvm_mapent_free_merge: free merged map entry 513 * 514 * => keep the entry if needed. 515 * => caller shouldn't hold map locked. 516 */ 517 518 static __inline void 519 uvm_mapent_free_merged(struct vm_map *map, struct vm_map_entry *me) 520 { 521 522 KASSERT(!(me->flags & UVM_MAP_KERNEL) || uvm_kmapent_map(me) == map); 523 524 if (me->flags & UVM_MAP_QUANTUM) { 525 /* 526 * keep this entry for later splitting. 527 */ 528 struct vm_map_kernel *vmk; 529 int s; 530 531 KASSERT(VM_MAP_IS_KERNEL(map)); 532 KASSERT(!VM_MAP_USE_KMAPENT(map) || 533 (me->flags & UVM_MAP_KERNEL)); 534 535 vmk = vm_map_to_kernel(map); 536 s = splvm(); 537 simple_lock(&uvm.kentry_lock); 538 me->next = vmk->vmk_merged_entries; 539 vmk->vmk_merged_entries = me; 540 simple_unlock(&uvm.kentry_lock); 541 splx(s); 542 } else { 543 uvm_mapent_free(me); 544 } 545 } 546 547 /* 548 * uvm_mapent_copy: copy a map entry, preserving flags 549 */ 550 551 static __inline void 552 uvm_mapent_copy(struct vm_map_entry *src, struct vm_map_entry *dst) 553 { 554 555 memcpy(dst, src, ((char *)&src->uvm_map_entry_stop_copy) - 556 ((char *)src)); 557 } 558 559 /* 560 * uvm_map_entry_unwire: unwire a map entry 561 * 562 * => map should be locked by caller 563 */ 564 565 static __inline void 566 uvm_map_entry_unwire(struct vm_map *map, struct vm_map_entry *entry) 567 { 568 569 entry->wired_count = 0; 570 uvm_fault_unwire_locked(map, entry->start, entry->end); 571 } 572 573 574 /* 575 * wrapper for calling amap_ref() 576 */ 577 static __inline void 578 uvm_map_reference_amap(struct vm_map_entry *entry, int flags) 579 { 580 581 amap_ref(entry->aref.ar_amap, entry->aref.ar_pageoff, 582 (entry->end - entry->start) >> PAGE_SHIFT, flags); 583 } 584 585 586 /* 587 * wrapper for calling amap_unref() 588 */ 589 static __inline void 590 uvm_map_unreference_amap(struct vm_map_entry *entry, int flags) 591 { 592 593 amap_unref(entry->aref.ar_amap, entry->aref.ar_pageoff, 594 (entry->end - entry->start) >> PAGE_SHIFT, flags); 595 } 596 597 598 /* 599 * uvm_map_init: init mapping system at boot time. note that we allocate 600 * and init the static pool of struct vm_map_entry *'s for the kernel here. 601 */ 602 603 void 604 uvm_map_init(void) 605 { 606 #if defined(UVMHIST) 607 static struct uvm_history_ent maphistbuf[100]; 608 static struct uvm_history_ent pdhistbuf[100]; 609 #endif 610 611 /* 612 * first, init logging system. 613 */ 614 615 UVMHIST_FUNC("uvm_map_init"); 616 UVMHIST_INIT_STATIC(maphist, maphistbuf); 617 UVMHIST_INIT_STATIC(pdhist, pdhistbuf); 618 UVMHIST_CALLED(maphist); 619 UVMHIST_LOG(maphist,"<starting uvm map system>", 0, 0, 0, 0); 620 621 /* 622 * initialize the global lock for kernel map entry. 623 * 624 * XXX is it worth to have per-map lock instead? 625 */ 626 627 simple_lock_init(&uvm.kentry_lock); 628 } 629 630 /* 631 * clippers 632 */ 633 634 /* 635 * uvm_map_clip_start: ensure that the entry begins at or after 636 * the starting address, if it doesn't we split the entry. 637 * 638 * => caller should use UVM_MAP_CLIP_START macro rather than calling 639 * this directly 640 * => map must be locked by caller 641 */ 642 643 void 644 uvm_map_clip_start(struct vm_map *map, struct vm_map_entry *entry, 645 vaddr_t start, struct uvm_mapent_reservation *umr) 646 { 647 struct vm_map_entry *new_entry; 648 vaddr_t new_adj; 649 650 /* uvm_map_simplify_entry(map, entry); */ /* XXX */ 651 652 uvm_tree_sanity(map, "clip_start entry"); 653 654 /* 655 * Split off the front portion. note that we must insert the new 656 * entry BEFORE this one, so that this entry has the specified 657 * starting address. 658 */ 659 new_entry = uvm_mapent_alloc_split(map, entry, 0, umr); 660 uvm_mapent_copy(entry, new_entry); /* entry -> new_entry */ 661 662 new_entry->end = start; 663 new_adj = start - new_entry->start; 664 if (entry->object.uvm_obj) 665 entry->offset += new_adj; /* shift start over */ 666 667 /* Does not change order for the RB tree */ 668 entry->start = start; 669 670 if (new_entry->aref.ar_amap) { 671 amap_splitref(&new_entry->aref, &entry->aref, new_adj); 672 } 673 674 uvm_map_entry_link(map, entry->prev, new_entry); 675 676 if (UVM_ET_ISSUBMAP(entry)) { 677 /* ... unlikely to happen, but play it safe */ 678 uvm_map_reference(new_entry->object.sub_map); 679 } else { 680 if (UVM_ET_ISOBJ(entry) && 681 entry->object.uvm_obj->pgops && 682 entry->object.uvm_obj->pgops->pgo_reference) 683 entry->object.uvm_obj->pgops->pgo_reference( 684 entry->object.uvm_obj); 685 } 686 687 uvm_tree_sanity(map, "clip_start leave"); 688 } 689 690 /* 691 * uvm_map_clip_end: ensure that the entry ends at or before 692 * the ending address, if it does't we split the reference 693 * 694 * => caller should use UVM_MAP_CLIP_END macro rather than calling 695 * this directly 696 * => map must be locked by caller 697 */ 698 699 void 700 uvm_map_clip_end(struct vm_map *map, struct vm_map_entry *entry, vaddr_t end, 701 struct uvm_mapent_reservation *umr) 702 { 703 struct vm_map_entry * new_entry; 704 vaddr_t new_adj; /* #bytes we move start forward */ 705 706 uvm_tree_sanity(map, "clip_end entry"); 707 708 /* 709 * Create a new entry and insert it 710 * AFTER the specified entry 711 */ 712 new_entry = uvm_mapent_alloc_split(map, entry, 0, umr); 713 uvm_mapent_copy(entry, new_entry); /* entry -> new_entry */ 714 715 new_entry->start = entry->end = end; 716 new_adj = end - entry->start; 717 if (new_entry->object.uvm_obj) 718 new_entry->offset += new_adj; 719 720 if (entry->aref.ar_amap) 721 amap_splitref(&entry->aref, &new_entry->aref, new_adj); 722 723 uvm_rb_fixup(map, entry); 724 725 uvm_map_entry_link(map, entry, new_entry); 726 727 if (UVM_ET_ISSUBMAP(entry)) { 728 /* ... unlikely to happen, but play it safe */ 729 uvm_map_reference(new_entry->object.sub_map); 730 } else { 731 if (UVM_ET_ISOBJ(entry) && 732 entry->object.uvm_obj->pgops && 733 entry->object.uvm_obj->pgops->pgo_reference) 734 entry->object.uvm_obj->pgops->pgo_reference( 735 entry->object.uvm_obj); 736 } 737 738 uvm_tree_sanity(map, "clip_end leave"); 739 } 740 741 742 /* 743 * M A P - m a i n e n t r y p o i n t 744 */ 745 /* 746 * uvm_map: establish a valid mapping in a map 747 * 748 * => assume startp is page aligned. 749 * => assume size is a multiple of PAGE_SIZE. 750 * => assume sys_mmap provides enough of a "hint" to have us skip 751 * over text/data/bss area. 752 * => map must be unlocked (we will lock it) 753 * => <uobj,uoffset> value meanings (4 cases): 754 * [1] <NULL,uoffset> == uoffset is a hint for PMAP_PREFER 755 * [2] <NULL,UVM_UNKNOWN_OFFSET> == don't PMAP_PREFER 756 * [3] <uobj,uoffset> == normal mapping 757 * [4] <uobj,UVM_UNKNOWN_OFFSET> == uvm_map finds offset based on VA 758 * 759 * case [4] is for kernel mappings where we don't know the offset until 760 * we've found a virtual address. note that kernel object offsets are 761 * always relative to vm_map_min(kernel_map). 762 * 763 * => if `align' is non-zero, we align the virtual address to the specified 764 * alignment. 765 * this is provided as a mechanism for large pages. 766 * 767 * => XXXCDC: need way to map in external amap? 768 */ 769 770 int 771 uvm_map(struct vm_map *map, vaddr_t *startp /* IN/OUT */, vsize_t size, 772 struct uvm_object *uobj, voff_t uoffset, vsize_t align, uvm_flag_t flags) 773 { 774 struct uvm_map_args args; 775 struct vm_map_entry *new_entry; 776 int error; 777 778 KASSERT((flags & UVM_FLAG_QUANTUM) == 0 || VM_MAP_IS_KERNEL(map)); 779 780 /* 781 * for pager_map, allocate the new entry first to avoid sleeping 782 * for memory while we have the map locked. 783 * 784 * besides, because we allocates entries for in-kernel maps 785 * a bit differently (cf. uvm_kmapent_alloc/free), we need to 786 * allocate them before locking the map. 787 */ 788 789 new_entry = NULL; 790 if (VM_MAP_USE_KMAPENT(map) || (flags & UVM_FLAG_QUANTUM) || 791 map == pager_map) { 792 new_entry = uvm_mapent_alloc(map, (flags & UVM_FLAG_NOWAIT)); 793 if (__predict_false(new_entry == NULL)) 794 return ENOMEM; 795 if (flags & UVM_FLAG_QUANTUM) 796 new_entry->flags |= UVM_MAP_QUANTUM; 797 } 798 if (map == pager_map) 799 flags |= UVM_FLAG_NOMERGE; 800 801 error = uvm_map_prepare(map, *startp, size, uobj, uoffset, align, 802 flags, &args); 803 if (!error) { 804 error = uvm_map_enter(map, &args, new_entry); 805 *startp = args.uma_start; 806 } 807 808 return error; 809 } 810 811 int 812 uvm_map_prepare(struct vm_map *map, vaddr_t start, vsize_t size, 813 struct uvm_object *uobj, voff_t uoffset, vsize_t align, uvm_flag_t flags, 814 struct uvm_map_args *args) 815 { 816 struct vm_map_entry *prev_entry; 817 vm_prot_t prot = UVM_PROTECTION(flags); 818 vm_prot_t maxprot = UVM_MAXPROTECTION(flags); 819 820 UVMHIST_FUNC("uvm_map_prepare"); 821 UVMHIST_CALLED(maphist); 822 823 UVMHIST_LOG(maphist, "(map=0x%x, start=0x%x, size=%d, flags=0x%x)", 824 map, start, size, flags); 825 UVMHIST_LOG(maphist, " uobj/offset 0x%x/%d", uobj, uoffset,0,0); 826 827 /* 828 * detect a popular device driver bug. 829 */ 830 831 KASSERT(doing_shutdown || curlwp != NULL || 832 (map->flags & VM_MAP_INTRSAFE)); 833 834 /* 835 * zero-sized mapping doesn't make any sense. 836 */ 837 KASSERT(size > 0); 838 839 KASSERT((~flags & (UVM_FLAG_NOWAIT | UVM_FLAG_WAITVA)) != 0); 840 841 uvm_tree_sanity(map, "map entry"); 842 843 /* 844 * check sanity of protection code 845 */ 846 847 if ((prot & maxprot) != prot) { 848 UVMHIST_LOG(maphist, "<- prot. failure: prot=0x%x, max=0x%x", 849 prot, maxprot,0,0); 850 return EACCES; 851 } 852 853 /* 854 * figure out where to put new VM range 855 */ 856 857 retry: 858 if (vm_map_lock_try(map) == FALSE) { 859 if (flags & UVM_FLAG_TRYLOCK) { 860 return EAGAIN; 861 } 862 vm_map_lock(map); /* could sleep here */ 863 } 864 if ((prev_entry = uvm_map_findspace(map, start, size, &start, 865 uobj, uoffset, align, flags)) == NULL) { 866 unsigned int timestamp; 867 868 if ((flags & UVM_FLAG_WAITVA) == 0) { 869 UVMHIST_LOG(maphist,"<- uvm_map_findspace failed!", 870 0,0,0,0); 871 vm_map_unlock(map); 872 return ENOMEM; 873 } 874 timestamp = map->timestamp; 875 UVMHIST_LOG(maphist,"waiting va timestamp=0x%x", 876 timestamp,0,0,0); 877 simple_lock(&map->flags_lock); 878 map->flags |= VM_MAP_WANTVA; 879 simple_unlock(&map->flags_lock); 880 vm_map_unlock(map); 881 882 /* 883 * wait until someone does unmap. 884 * XXX fragile locking 885 */ 886 887 simple_lock(&map->flags_lock); 888 while ((map->flags & VM_MAP_WANTVA) != 0 && 889 map->timestamp == timestamp) { 890 ltsleep(&map->header, PVM, "vmmapva", 0, 891 &map->flags_lock); 892 } 893 simple_unlock(&map->flags_lock); 894 goto retry; 895 } 896 897 #ifdef PMAP_GROWKERNEL 898 /* 899 * If the kernel pmap can't map the requested space, 900 * then allocate more resources for it. 901 */ 902 if (map == kernel_map && uvm_maxkaddr < (start + size)) 903 uvm_maxkaddr = pmap_growkernel(start + size); 904 #endif 905 906 UVMCNT_INCR(uvm_map_call); 907 908 /* 909 * if uobj is null, then uoffset is either a VAC hint for PMAP_PREFER 910 * [typically from uvm_map_reserve] or it is UVM_UNKNOWN_OFFSET. in 911 * either case we want to zero it before storing it in the map entry 912 * (because it looks strange and confusing when debugging...) 913 * 914 * if uobj is not null 915 * if uoffset is not UVM_UNKNOWN_OFFSET then we have a normal mapping 916 * and we do not need to change uoffset. 917 * if uoffset is UVM_UNKNOWN_OFFSET then we need to find the offset 918 * now (based on the starting address of the map). this case is 919 * for kernel object mappings where we don't know the offset until 920 * the virtual address is found (with uvm_map_findspace). the 921 * offset is the distance we are from the start of the map. 922 */ 923 924 if (uobj == NULL) { 925 uoffset = 0; 926 } else { 927 if (uoffset == UVM_UNKNOWN_OFFSET) { 928 KASSERT(UVM_OBJ_IS_KERN_OBJECT(uobj)); 929 uoffset = start - vm_map_min(kernel_map); 930 } 931 } 932 933 args->uma_flags = flags; 934 args->uma_prev = prev_entry; 935 args->uma_start = start; 936 args->uma_size = size; 937 args->uma_uobj = uobj; 938 args->uma_uoffset = uoffset; 939 940 return 0; 941 } 942 943 int 944 uvm_map_enter(struct vm_map *map, const struct uvm_map_args *args, 945 struct vm_map_entry *new_entry) 946 { 947 struct vm_map_entry *prev_entry = args->uma_prev; 948 struct vm_map_entry *dead = NULL; 949 950 const uvm_flag_t flags = args->uma_flags; 951 const vm_prot_t prot = UVM_PROTECTION(flags); 952 const vm_prot_t maxprot = UVM_MAXPROTECTION(flags); 953 const vm_inherit_t inherit = UVM_INHERIT(flags); 954 const int amapwaitflag = (flags & UVM_FLAG_NOWAIT) ? 955 AMAP_EXTEND_NOWAIT : 0; 956 const int advice = UVM_ADVICE(flags); 957 const int meflagmask = UVM_MAP_NOMERGE | UVM_MAP_QUANTUM; 958 const int meflagval = (flags & UVM_FLAG_QUANTUM) ? 959 UVM_MAP_QUANTUM : 0; 960 961 vaddr_t start = args->uma_start; 962 vsize_t size = args->uma_size; 963 struct uvm_object *uobj = args->uma_uobj; 964 voff_t uoffset = args->uma_uoffset; 965 966 const int kmap = (vm_map_pmap(map) == pmap_kernel()); 967 int merged = 0; 968 int error; 969 int newetype; 970 971 UVMHIST_FUNC("uvm_map_enter"); 972 UVMHIST_CALLED(maphist); 973 974 UVMHIST_LOG(maphist, "(map=0x%x, start=0x%x, size=%d, flags=0x%x)", 975 map, start, size, flags); 976 UVMHIST_LOG(maphist, " uobj/offset 0x%x/%d", uobj, uoffset,0,0); 977 978 if (flags & UVM_FLAG_QUANTUM) { 979 KASSERT(new_entry); 980 KASSERT(new_entry->flags & UVM_MAP_QUANTUM); 981 } 982 983 if (uobj) 984 newetype = UVM_ET_OBJ; 985 else 986 newetype = 0; 987 988 if (flags & UVM_FLAG_COPYONW) { 989 newetype |= UVM_ET_COPYONWRITE; 990 if ((flags & UVM_FLAG_OVERLAY) == 0) 991 newetype |= UVM_ET_NEEDSCOPY; 992 } 993 994 /* 995 * try and insert in map by extending previous entry, if possible. 996 * XXX: we don't try and pull back the next entry. might be useful 997 * for a stack, but we are currently allocating our stack in advance. 998 */ 999 1000 if (flags & UVM_FLAG_NOMERGE) 1001 goto nomerge; 1002 1003 if (prev_entry->etype == newetype && 1004 prev_entry->end == start && 1005 prev_entry != &map->header && 1006 prev_entry->object.uvm_obj == uobj) { 1007 1008 if ((prev_entry->flags & meflagmask) != meflagval) 1009 goto forwardmerge; 1010 1011 if (uobj && prev_entry->offset + 1012 (prev_entry->end - prev_entry->start) != uoffset) 1013 goto forwardmerge; 1014 1015 if (prev_entry->protection != prot || 1016 prev_entry->max_protection != maxprot) 1017 goto forwardmerge; 1018 1019 if (prev_entry->inheritance != inherit || 1020 prev_entry->advice != advice) 1021 goto forwardmerge; 1022 1023 /* wiring status must match (new area is unwired) */ 1024 if (VM_MAPENT_ISWIRED(prev_entry)) 1025 goto forwardmerge; 1026 1027 /* 1028 * can't extend a shared amap. note: no need to lock amap to 1029 * look at refs since we don't care about its exact value. 1030 * if it is one (i.e. we have only reference) it will stay there 1031 */ 1032 1033 if (prev_entry->aref.ar_amap && 1034 amap_refs(prev_entry->aref.ar_amap) != 1) { 1035 goto forwardmerge; 1036 } 1037 1038 if (prev_entry->aref.ar_amap) { 1039 error = amap_extend(prev_entry, size, 1040 amapwaitflag | AMAP_EXTEND_FORWARDS); 1041 if (error) 1042 goto done; 1043 } 1044 1045 if (kmap) 1046 UVMCNT_INCR(map_kbackmerge); 1047 else 1048 UVMCNT_INCR(map_ubackmerge); 1049 UVMHIST_LOG(maphist," starting back merge", 0, 0, 0, 0); 1050 1051 /* 1052 * drop our reference to uobj since we are extending a reference 1053 * that we already have (the ref count can not drop to zero). 1054 */ 1055 1056 if (uobj && uobj->pgops->pgo_detach) 1057 uobj->pgops->pgo_detach(uobj); 1058 1059 prev_entry->end += size; 1060 uvm_rb_fixup(map, prev_entry); 1061 1062 uvm_tree_sanity(map, "map backmerged"); 1063 1064 UVMHIST_LOG(maphist,"<- done (via backmerge)!", 0, 0, 0, 0); 1065 merged++; 1066 } 1067 1068 forwardmerge: 1069 if (prev_entry->next->etype == newetype && 1070 prev_entry->next->start == (start + size) && 1071 prev_entry->next != &map->header && 1072 prev_entry->next->object.uvm_obj == uobj) { 1073 1074 if ((prev_entry->next->flags & meflagmask) != meflagval) 1075 goto nomerge; 1076 1077 if (uobj && prev_entry->next->offset != uoffset + size) 1078 goto nomerge; 1079 1080 if (prev_entry->next->protection != prot || 1081 prev_entry->next->max_protection != maxprot) 1082 goto nomerge; 1083 1084 if (prev_entry->next->inheritance != inherit || 1085 prev_entry->next->advice != advice) 1086 goto nomerge; 1087 1088 /* wiring status must match (new area is unwired) */ 1089 if (VM_MAPENT_ISWIRED(prev_entry->next)) 1090 goto nomerge; 1091 1092 /* 1093 * can't extend a shared amap. note: no need to lock amap to 1094 * look at refs since we don't care about its exact value. 1095 * if it is one (i.e. we have only reference) it will stay there. 1096 * 1097 * note that we also can't merge two amaps, so if we 1098 * merged with the previous entry which has an amap, 1099 * and the next entry also has an amap, we give up. 1100 * 1101 * Interesting cases: 1102 * amap, new, amap -> give up second merge (single fwd extend) 1103 * amap, new, none -> double forward extend (extend again here) 1104 * none, new, amap -> double backward extend (done here) 1105 * uobj, new, amap -> single backward extend (done here) 1106 * 1107 * XXX should we attempt to deal with someone refilling 1108 * the deallocated region between two entries that are 1109 * backed by the same amap (ie, arefs is 2, "prev" and 1110 * "next" refer to it, and adding this allocation will 1111 * close the hole, thus restoring arefs to 1 and 1112 * deallocating the "next" vm_map_entry)? -- @@@ 1113 */ 1114 1115 if (prev_entry->next->aref.ar_amap && 1116 (amap_refs(prev_entry->next->aref.ar_amap) != 1 || 1117 (merged && prev_entry->aref.ar_amap))) { 1118 goto nomerge; 1119 } 1120 1121 if (merged) { 1122 /* 1123 * Try to extend the amap of the previous entry to 1124 * cover the next entry as well. If it doesn't work 1125 * just skip on, don't actually give up, since we've 1126 * already completed the back merge. 1127 */ 1128 if (prev_entry->aref.ar_amap) { 1129 if (amap_extend(prev_entry, 1130 prev_entry->next->end - 1131 prev_entry->next->start, 1132 amapwaitflag | AMAP_EXTEND_FORWARDS)) 1133 goto nomerge; 1134 } 1135 1136 /* 1137 * Try to extend the amap of the *next* entry 1138 * back to cover the new allocation *and* the 1139 * previous entry as well (the previous merge 1140 * didn't have an amap already otherwise we 1141 * wouldn't be checking here for an amap). If 1142 * it doesn't work just skip on, again, don't 1143 * actually give up, since we've already 1144 * completed the back merge. 1145 */ 1146 else if (prev_entry->next->aref.ar_amap) { 1147 if (amap_extend(prev_entry->next, 1148 prev_entry->end - 1149 prev_entry->start, 1150 amapwaitflag | AMAP_EXTEND_BACKWARDS)) 1151 goto nomerge; 1152 } 1153 } else { 1154 /* 1155 * Pull the next entry's amap backwards to cover this 1156 * new allocation. 1157 */ 1158 if (prev_entry->next->aref.ar_amap) { 1159 error = amap_extend(prev_entry->next, size, 1160 amapwaitflag | AMAP_EXTEND_BACKWARDS); 1161 if (error) 1162 goto done; 1163 } 1164 } 1165 1166 if (merged) { 1167 if (kmap) { 1168 UVMCNT_DECR(map_kbackmerge); 1169 UVMCNT_INCR(map_kbimerge); 1170 } else { 1171 UVMCNT_DECR(map_ubackmerge); 1172 UVMCNT_INCR(map_ubimerge); 1173 } 1174 } else { 1175 if (kmap) 1176 UVMCNT_INCR(map_kforwmerge); 1177 else 1178 UVMCNT_INCR(map_uforwmerge); 1179 } 1180 UVMHIST_LOG(maphist," starting forward merge", 0, 0, 0, 0); 1181 1182 /* 1183 * drop our reference to uobj since we are extending a reference 1184 * that we already have (the ref count can not drop to zero). 1185 * (if merged, we've already detached) 1186 */ 1187 if (uobj && uobj->pgops->pgo_detach && !merged) 1188 uobj->pgops->pgo_detach(uobj); 1189 1190 if (merged) { 1191 dead = prev_entry->next; 1192 prev_entry->end = dead->end; 1193 uvm_map_entry_unlink(map, dead); 1194 if (dead->aref.ar_amap != NULL) { 1195 prev_entry->aref = dead->aref; 1196 dead->aref.ar_amap = NULL; 1197 } 1198 } else { 1199 prev_entry->next->start -= size; 1200 if (prev_entry != &map->header) 1201 uvm_rb_fixup(map, prev_entry); 1202 if (uobj) 1203 prev_entry->next->offset = uoffset; 1204 } 1205 1206 uvm_tree_sanity(map, "map forwardmerged"); 1207 1208 UVMHIST_LOG(maphist,"<- done forwardmerge", 0, 0, 0, 0); 1209 merged++; 1210 } 1211 1212 nomerge: 1213 if (!merged) { 1214 UVMHIST_LOG(maphist," allocating new map entry", 0, 0, 0, 0); 1215 if (kmap) 1216 UVMCNT_INCR(map_knomerge); 1217 else 1218 UVMCNT_INCR(map_unomerge); 1219 1220 /* 1221 * allocate new entry and link it in. 1222 */ 1223 1224 if (new_entry == NULL) { 1225 new_entry = uvm_mapent_alloc(map, 1226 (flags & UVM_FLAG_NOWAIT)); 1227 if (__predict_false(new_entry == NULL)) { 1228 error = ENOMEM; 1229 goto done; 1230 } 1231 } 1232 new_entry->start = start; 1233 new_entry->end = new_entry->start + size; 1234 new_entry->object.uvm_obj = uobj; 1235 new_entry->offset = uoffset; 1236 1237 new_entry->etype = newetype; 1238 1239 if (flags & UVM_FLAG_NOMERGE) { 1240 new_entry->flags |= UVM_MAP_NOMERGE; 1241 } 1242 1243 new_entry->protection = prot; 1244 new_entry->max_protection = maxprot; 1245 new_entry->inheritance = inherit; 1246 new_entry->wired_count = 0; 1247 new_entry->advice = advice; 1248 if (flags & UVM_FLAG_OVERLAY) { 1249 1250 /* 1251 * to_add: for BSS we overallocate a little since we 1252 * are likely to extend 1253 */ 1254 1255 vaddr_t to_add = (flags & UVM_FLAG_AMAPPAD) ? 1256 UVM_AMAP_CHUNK << PAGE_SHIFT : 0; 1257 struct vm_amap *amap = amap_alloc(size, to_add, 1258 (flags & UVM_FLAG_NOWAIT) ? M_NOWAIT : M_WAITOK); 1259 if (__predict_false(amap == NULL)) { 1260 error = ENOMEM; 1261 goto done; 1262 } 1263 new_entry->aref.ar_pageoff = 0; 1264 new_entry->aref.ar_amap = amap; 1265 } else { 1266 new_entry->aref.ar_pageoff = 0; 1267 new_entry->aref.ar_amap = NULL; 1268 } 1269 uvm_map_entry_link(map, prev_entry, new_entry); 1270 1271 /* 1272 * Update the free space hint 1273 */ 1274 1275 if ((map->first_free == prev_entry) && 1276 (prev_entry->end >= new_entry->start)) 1277 map->first_free = new_entry; 1278 1279 new_entry = NULL; 1280 } 1281 1282 map->size += size; 1283 1284 UVMHIST_LOG(maphist,"<- done!", 0, 0, 0, 0); 1285 1286 error = 0; 1287 done: 1288 vm_map_unlock(map); 1289 if (new_entry) { 1290 if (error == 0) { 1291 KDASSERT(merged); 1292 uvm_mapent_free_merged(map, new_entry); 1293 } else { 1294 uvm_mapent_free(new_entry); 1295 } 1296 } 1297 if (dead) { 1298 KDASSERT(merged); 1299 uvm_mapent_free_merged(map, dead); 1300 } 1301 return error; 1302 } 1303 1304 /* 1305 * uvm_map_lookup_entry: find map entry at or before an address 1306 * 1307 * => map must at least be read-locked by caller 1308 * => entry is returned in "entry" 1309 * => return value is true if address is in the returned entry 1310 */ 1311 1312 boolean_t 1313 uvm_map_lookup_entry(struct vm_map *map, vaddr_t address, 1314 struct vm_map_entry **entry /* OUT */) 1315 { 1316 struct vm_map_entry *cur; 1317 boolean_t use_tree = FALSE; 1318 UVMHIST_FUNC("uvm_map_lookup_entry"); 1319 UVMHIST_CALLED(maphist); 1320 1321 UVMHIST_LOG(maphist,"(map=0x%x,addr=0x%x,ent=0x%x)", 1322 map, address, entry, 0); 1323 1324 /* 1325 * start looking either from the head of the 1326 * list, or from the hint. 1327 */ 1328 1329 simple_lock(&map->hint_lock); 1330 cur = map->hint; 1331 simple_unlock(&map->hint_lock); 1332 1333 if (cur == &map->header) 1334 cur = cur->next; 1335 1336 UVMCNT_INCR(uvm_mlk_call); 1337 if (address >= cur->start) { 1338 1339 /* 1340 * go from hint to end of list. 1341 * 1342 * but first, make a quick check to see if 1343 * we are already looking at the entry we 1344 * want (which is usually the case). 1345 * note also that we don't need to save the hint 1346 * here... it is the same hint (unless we are 1347 * at the header, in which case the hint didn't 1348 * buy us anything anyway). 1349 */ 1350 1351 if (cur != &map->header && cur->end > address) { 1352 UVMCNT_INCR(uvm_mlk_hint); 1353 *entry = cur; 1354 UVMHIST_LOG(maphist,"<- got it via hint (0x%x)", 1355 cur, 0, 0, 0); 1356 return (TRUE); 1357 } 1358 1359 if (map->nentries > 30) 1360 use_tree = TRUE; 1361 } else { 1362 1363 /* 1364 * invalid hint. use tree. 1365 */ 1366 use_tree = TRUE; 1367 } 1368 1369 uvm_tree_sanity(map, __func__); 1370 1371 if (use_tree) { 1372 struct vm_map_entry *prev = &map->header; 1373 cur = RB_ROOT(&map->rbhead); 1374 1375 /* 1376 * Simple lookup in the tree. Happens when the hint is 1377 * invalid, or nentries reach a threshold. 1378 */ 1379 while (cur) { 1380 if (address >= cur->start) { 1381 if (address < cur->end) { 1382 *entry = cur; 1383 goto got; 1384 } 1385 prev = cur; 1386 cur = RB_RIGHT(cur, rb_entry); 1387 } else 1388 cur = RB_LEFT(cur, rb_entry); 1389 } 1390 *entry = prev; 1391 goto failed; 1392 } 1393 1394 /* 1395 * search linearly 1396 */ 1397 1398 while (cur != &map->header) { 1399 if (cur->end > address) { 1400 if (address >= cur->start) { 1401 /* 1402 * save this lookup for future 1403 * hints, and return 1404 */ 1405 1406 *entry = cur; 1407 got: 1408 SAVE_HINT(map, map->hint, *entry); 1409 UVMHIST_LOG(maphist,"<- search got it (0x%x)", 1410 cur, 0, 0, 0); 1411 KDASSERT((*entry)->start <= address); 1412 KDASSERT(address < (*entry)->end); 1413 return (TRUE); 1414 } 1415 break; 1416 } 1417 cur = cur->next; 1418 } 1419 *entry = cur->prev; 1420 failed: 1421 SAVE_HINT(map, map->hint, *entry); 1422 UVMHIST_LOG(maphist,"<- failed!",0,0,0,0); 1423 KDASSERT((*entry) == &map->header || (*entry)->end <= address); 1424 KDASSERT((*entry)->next == &map->header || 1425 address < (*entry)->next->start); 1426 return (FALSE); 1427 } 1428 1429 /* 1430 * See if the range between start and start + length fits in the gap 1431 * entry->next->start and entry->end. Returns 1 if fits, 0 if doesn't 1432 * fit, and -1 address wraps around. 1433 */ 1434 static __inline int 1435 uvm_map_space_avail(vaddr_t *start, vsize_t length, voff_t uoffset, 1436 vsize_t align, int topdown, struct vm_map_entry *entry) 1437 { 1438 vaddr_t end; 1439 1440 #ifdef PMAP_PREFER 1441 /* 1442 * push start address forward as needed to avoid VAC alias problems. 1443 * we only do this if a valid offset is specified. 1444 */ 1445 1446 if (uoffset != UVM_UNKNOWN_OFFSET) 1447 PMAP_PREFER(uoffset, start, length, topdown); 1448 #endif 1449 if (align != 0) { 1450 if ((*start & (align - 1)) != 0) { 1451 if (topdown) 1452 *start &= ~(align - 1); 1453 else 1454 *start = roundup(*start, align); 1455 } 1456 /* 1457 * XXX Should we PMAP_PREFER() here again? 1458 * eh...i think we're okay 1459 */ 1460 } 1461 1462 /* 1463 * Find the end of the proposed new region. Be sure we didn't 1464 * wrap around the address; if so, we lose. Otherwise, if the 1465 * proposed new region fits before the next entry, we win. 1466 */ 1467 1468 end = *start + length; 1469 if (end < *start) 1470 return (-1); 1471 1472 if (entry->next->start >= end && *start >= entry->end) 1473 return (1); 1474 1475 return (0); 1476 } 1477 1478 /* 1479 * uvm_map_findspace: find "length" sized space in "map". 1480 * 1481 * => "hint" is a hint about where we want it, unless UVM_FLAG_FIXED is 1482 * set in "flags" (in which case we insist on using "hint"). 1483 * => "result" is VA returned 1484 * => uobj/uoffset are to be used to handle VAC alignment, if required 1485 * => if "align" is non-zero, we attempt to align to that value. 1486 * => caller must at least have read-locked map 1487 * => returns NULL on failure, or pointer to prev. map entry if success 1488 * => note this is a cross between the old vm_map_findspace and vm_map_find 1489 */ 1490 1491 struct vm_map_entry * 1492 uvm_map_findspace(struct vm_map *map, vaddr_t hint, vsize_t length, 1493 vaddr_t *result /* OUT */, struct uvm_object *uobj, voff_t uoffset, 1494 vsize_t align, int flags) 1495 { 1496 struct vm_map_entry *entry; 1497 struct vm_map_entry *child, *prev, *tmp; 1498 vaddr_t orig_hint; 1499 const int topdown = map->flags & VM_MAP_TOPDOWN; 1500 UVMHIST_FUNC("uvm_map_findspace"); 1501 UVMHIST_CALLED(maphist); 1502 1503 UVMHIST_LOG(maphist, "(map=0x%x, hint=0x%x, len=%d, flags=0x%x)", 1504 map, hint, length, flags); 1505 KASSERT((align & (align - 1)) == 0); 1506 KASSERT((flags & UVM_FLAG_FIXED) == 0 || align == 0); 1507 1508 uvm_tree_sanity(map, "map_findspace entry"); 1509 1510 /* 1511 * remember the original hint. if we are aligning, then we 1512 * may have to try again with no alignment constraint if 1513 * we fail the first time. 1514 */ 1515 1516 orig_hint = hint; 1517 if (hint < map->min_offset) { /* check ranges ... */ 1518 if (flags & UVM_FLAG_FIXED) { 1519 UVMHIST_LOG(maphist,"<- VA below map range",0,0,0,0); 1520 return (NULL); 1521 } 1522 hint = map->min_offset; 1523 } 1524 if (hint > map->max_offset) { 1525 UVMHIST_LOG(maphist,"<- VA 0x%x > range [0x%x->0x%x]", 1526 hint, map->min_offset, map->max_offset, 0); 1527 return (NULL); 1528 } 1529 1530 /* 1531 * Look for the first possible address; if there's already 1532 * something at this address, we have to start after it. 1533 */ 1534 1535 /* 1536 * @@@: there are four, no, eight cases to consider. 1537 * 1538 * 0: found, fixed, bottom up -> fail 1539 * 1: found, fixed, top down -> fail 1540 * 2: found, not fixed, bottom up -> start after entry->end, 1541 * loop up 1542 * 3: found, not fixed, top down -> start before entry->start, 1543 * loop down 1544 * 4: not found, fixed, bottom up -> check entry->next->start, fail 1545 * 5: not found, fixed, top down -> check entry->next->start, fail 1546 * 6: not found, not fixed, bottom up -> check entry->next->start, 1547 * loop up 1548 * 7: not found, not fixed, top down -> check entry->next->start, 1549 * loop down 1550 * 1551 * as you can see, it reduces to roughly five cases, and that 1552 * adding top down mapping only adds one unique case (without 1553 * it, there would be four cases). 1554 */ 1555 1556 if ((flags & UVM_FLAG_FIXED) == 0 && hint == map->min_offset) { 1557 entry = map->first_free; 1558 } else { 1559 if (uvm_map_lookup_entry(map, hint, &entry)) { 1560 /* "hint" address already in use ... */ 1561 if (flags & UVM_FLAG_FIXED) { 1562 UVMHIST_LOG(maphist, "<- fixed & VA in use", 1563 0, 0, 0, 0); 1564 return (NULL); 1565 } 1566 if (topdown) 1567 /* Start from lower gap. */ 1568 entry = entry->prev; 1569 } else if (flags & UVM_FLAG_FIXED) { 1570 if (entry->next->start >= hint + length && 1571 hint + length > hint) 1572 goto found; 1573 1574 /* "hint" address is gap but too small */ 1575 UVMHIST_LOG(maphist, "<- fixed mapping failed", 1576 0, 0, 0, 0); 1577 return (NULL); /* only one shot at it ... */ 1578 } else { 1579 /* 1580 * See if given hint fits in this gap. 1581 */ 1582 switch (uvm_map_space_avail(&hint, length, 1583 uoffset, align, topdown, entry)) { 1584 case 1: 1585 goto found; 1586 case -1: 1587 goto wraparound; 1588 } 1589 1590 if (topdown) { 1591 /* 1592 * Still there is a chance to fit 1593 * if hint > entry->end. 1594 */ 1595 } else { 1596 /* Start from higher gap. */ 1597 entry = entry->next; 1598 if (entry == &map->header) 1599 goto notfound; 1600 goto nextgap; 1601 } 1602 } 1603 } 1604 1605 /* 1606 * Note that all UVM_FLAGS_FIXED case is already handled. 1607 */ 1608 KDASSERT((flags & UVM_FLAG_FIXED) == 0); 1609 1610 /* Try to find the space in the red-black tree */ 1611 1612 /* Check slot before any entry */ 1613 hint = topdown ? entry->next->start - length : entry->end; 1614 switch (uvm_map_space_avail(&hint, length, uoffset, align, 1615 topdown, entry)) { 1616 case 1: 1617 goto found; 1618 case -1: 1619 goto wraparound; 1620 } 1621 1622 nextgap: 1623 KDASSERT((flags & UVM_FLAG_FIXED) == 0); 1624 /* If there is not enough space in the whole tree, we fail */ 1625 tmp = RB_ROOT(&map->rbhead); 1626 if (tmp == NULL || tmp->space < length) 1627 goto notfound; 1628 1629 prev = NULL; /* previous candidate */ 1630 1631 /* Find an entry close to hint that has enough space */ 1632 for (; tmp;) { 1633 KASSERT(tmp->next->start == tmp->end + tmp->ownspace); 1634 if (topdown) { 1635 if (tmp->next->start < hint + length && 1636 (prev == NULL || tmp->end > prev->end)) { 1637 if (tmp->ownspace >= length) 1638 prev = tmp; 1639 else if ((child = RB_LEFT(tmp, rb_entry)) 1640 != NULL && child->space >= length) 1641 prev = tmp; 1642 } 1643 } else { 1644 if (tmp->end >= hint && 1645 (prev == NULL || tmp->end < prev->end)) { 1646 if (tmp->ownspace >= length) 1647 prev = tmp; 1648 else if ((child = RB_RIGHT(tmp, rb_entry)) 1649 != NULL && child->space >= length) 1650 prev = tmp; 1651 } 1652 } 1653 if (tmp->next->start < hint + length) 1654 child = RB_RIGHT(tmp, rb_entry); 1655 else if (tmp->end > hint) 1656 child = RB_LEFT(tmp, rb_entry); 1657 else { 1658 if (tmp->ownspace >= length) 1659 break; 1660 if (topdown) 1661 child = RB_LEFT(tmp, rb_entry); 1662 else 1663 child = RB_RIGHT(tmp, rb_entry); 1664 } 1665 if (child == NULL || child->space < length) 1666 break; 1667 tmp = child; 1668 } 1669 1670 if (tmp != NULL && tmp->start < hint && hint < tmp->next->start) { 1671 /* 1672 * Check if the entry that we found satifies the 1673 * space requirement 1674 */ 1675 if (topdown) { 1676 if (hint > tmp->next->start - length) 1677 hint = tmp->next->start - length; 1678 } else { 1679 if (hint < tmp->end) 1680 hint = tmp->end; 1681 } 1682 switch (uvm_map_space_avail(&hint, length, uoffset, align, 1683 topdown, tmp)) { 1684 case 1: 1685 entry = tmp; 1686 goto found; 1687 case -1: 1688 goto wraparound; 1689 } 1690 if (tmp->ownspace >= length) 1691 goto listsearch; 1692 } 1693 if (prev == NULL) 1694 goto notfound; 1695 1696 if (topdown) { 1697 KASSERT(orig_hint >= prev->next->start - length || 1698 prev->next->start - length > prev->next->start); 1699 hint = prev->next->start - length; 1700 } else { 1701 KASSERT(orig_hint <= prev->end); 1702 hint = prev->end; 1703 } 1704 switch (uvm_map_space_avail(&hint, length, uoffset, align, 1705 topdown, prev)) { 1706 case 1: 1707 entry = prev; 1708 goto found; 1709 case -1: 1710 goto wraparound; 1711 } 1712 if (prev->ownspace >= length) 1713 goto listsearch; 1714 1715 if (topdown) 1716 tmp = RB_LEFT(prev, rb_entry); 1717 else 1718 tmp = RB_RIGHT(prev, rb_entry); 1719 for (;;) { 1720 KASSERT(tmp && tmp->space >= length); 1721 if (topdown) 1722 child = RB_RIGHT(tmp, rb_entry); 1723 else 1724 child = RB_LEFT(tmp, rb_entry); 1725 if (child && child->space >= length) { 1726 tmp = child; 1727 continue; 1728 } 1729 if (tmp->ownspace >= length) 1730 break; 1731 if (topdown) 1732 tmp = RB_LEFT(tmp, rb_entry); 1733 else 1734 tmp = RB_RIGHT(tmp, rb_entry); 1735 } 1736 1737 if (topdown) { 1738 KASSERT(orig_hint >= tmp->next->start - length || 1739 tmp->next->start - length > tmp->next->start); 1740 hint = tmp->next->start - length; 1741 } else { 1742 KASSERT(orig_hint <= tmp->end); 1743 hint = tmp->end; 1744 } 1745 switch (uvm_map_space_avail(&hint, length, uoffset, align, 1746 topdown, tmp)) { 1747 case 1: 1748 entry = tmp; 1749 goto found; 1750 case -1: 1751 goto wraparound; 1752 } 1753 1754 /* 1755 * The tree fails to find an entry because of offset or alignment 1756 * restrictions. Search the list instead. 1757 */ 1758 listsearch: 1759 /* 1760 * Look through the rest of the map, trying to fit a new region in 1761 * the gap between existing regions, or after the very last region. 1762 * note: entry->end = base VA of current gap, 1763 * entry->next->start = VA of end of current gap 1764 */ 1765 1766 for (;;) { 1767 /* Update hint for current gap. */ 1768 hint = topdown ? entry->next->start - length : entry->end; 1769 1770 /* See if it fits. */ 1771 switch (uvm_map_space_avail(&hint, length, uoffset, align, 1772 topdown, entry)) { 1773 case 1: 1774 goto found; 1775 case -1: 1776 goto wraparound; 1777 } 1778 1779 /* Advance to next/previous gap */ 1780 if (topdown) { 1781 if (entry == &map->header) { 1782 UVMHIST_LOG(maphist, "<- failed (off start)", 1783 0,0,0,0); 1784 goto notfound; 1785 } 1786 entry = entry->prev; 1787 } else { 1788 entry = entry->next; 1789 if (entry == &map->header) { 1790 UVMHIST_LOG(maphist, "<- failed (off end)", 1791 0,0,0,0); 1792 goto notfound; 1793 } 1794 } 1795 } 1796 1797 found: 1798 SAVE_HINT(map, map->hint, entry); 1799 *result = hint; 1800 UVMHIST_LOG(maphist,"<- got it! (result=0x%x)", hint, 0,0,0); 1801 KASSERT( topdown || hint >= orig_hint); 1802 KASSERT(!topdown || hint <= orig_hint); 1803 KASSERT(entry->end <= hint); 1804 KASSERT(hint + length <= entry->next->start); 1805 return (entry); 1806 1807 wraparound: 1808 UVMHIST_LOG(maphist, "<- failed (wrap around)", 0,0,0,0); 1809 1810 return (NULL); 1811 1812 notfound: 1813 UVMHIST_LOG(maphist, "<- failed (notfound)", 0,0,0,0); 1814 1815 return (NULL); 1816 } 1817 1818 /* 1819 * U N M A P - m a i n h e l p e r f u n c t i o n s 1820 */ 1821 1822 /* 1823 * uvm_unmap_remove: remove mappings from a vm_map (from "start" up to "stop") 1824 * 1825 * => caller must check alignment and size 1826 * => map must be locked by caller 1827 * => we return a list of map entries that we've remove from the map 1828 * in "entry_list" 1829 */ 1830 1831 void 1832 uvm_unmap_remove(struct vm_map *map, vaddr_t start, vaddr_t end, 1833 struct vm_map_entry **entry_list /* OUT */, 1834 struct uvm_mapent_reservation *umr) 1835 { 1836 struct vm_map_entry *entry, *first_entry, *next; 1837 vaddr_t len; 1838 UVMHIST_FUNC("uvm_unmap_remove"); UVMHIST_CALLED(maphist); 1839 1840 UVMHIST_LOG(maphist,"(map=0x%x, start=0x%x, end=0x%x)", 1841 map, start, end, 0); 1842 VM_MAP_RANGE_CHECK(map, start, end); 1843 1844 uvm_tree_sanity(map, "unmap_remove entry"); 1845 1846 /* 1847 * find first entry 1848 */ 1849 1850 if (uvm_map_lookup_entry(map, start, &first_entry) == TRUE) { 1851 /* clip and go... */ 1852 entry = first_entry; 1853 UVM_MAP_CLIP_START(map, entry, start, umr); 1854 /* critical! prevents stale hint */ 1855 SAVE_HINT(map, entry, entry->prev); 1856 } else { 1857 entry = first_entry->next; 1858 } 1859 1860 /* 1861 * Save the free space hint 1862 */ 1863 1864 if (map->first_free->start >= start) 1865 map->first_free = entry->prev; 1866 1867 /* 1868 * note: we now re-use first_entry for a different task. we remove 1869 * a number of map entries from the map and save them in a linked 1870 * list headed by "first_entry". once we remove them from the map 1871 * the caller should unlock the map and drop the references to the 1872 * backing objects [c.f. uvm_unmap_detach]. the object is to 1873 * separate unmapping from reference dropping. why? 1874 * [1] the map has to be locked for unmapping 1875 * [2] the map need not be locked for reference dropping 1876 * [3] dropping references may trigger pager I/O, and if we hit 1877 * a pager that does synchronous I/O we may have to wait for it. 1878 * [4] we would like all waiting for I/O to occur with maps unlocked 1879 * so that we don't block other threads. 1880 */ 1881 1882 first_entry = NULL; 1883 *entry_list = NULL; 1884 1885 /* 1886 * break up the area into map entry sized regions and unmap. note 1887 * that all mappings have to be removed before we can even consider 1888 * dropping references to amaps or VM objects (otherwise we could end 1889 * up with a mapping to a page on the free list which would be very bad) 1890 */ 1891 1892 while ((entry != &map->header) && (entry->start < end)) { 1893 KASSERT((entry->flags & UVM_MAP_FIRST) == 0); 1894 1895 UVM_MAP_CLIP_END(map, entry, end, umr); 1896 next = entry->next; 1897 len = entry->end - entry->start; 1898 1899 /* 1900 * unwire before removing addresses from the pmap; otherwise 1901 * unwiring will put the entries back into the pmap (XXX). 1902 */ 1903 1904 if (VM_MAPENT_ISWIRED(entry)) { 1905 uvm_map_entry_unwire(map, entry); 1906 } 1907 if ((map->flags & VM_MAP_PAGEABLE) == 0) { 1908 1909 /* 1910 * if the map is non-pageable, any pages mapped there 1911 * must be wired and entered with pmap_kenter_pa(), 1912 * and we should free any such pages immediately. 1913 * this is mostly used for kmem_map and mb_map. 1914 */ 1915 1916 if ((entry->flags & UVM_MAP_KMAPENT) == 0) { 1917 uvm_km_pgremove_intrsafe(entry->start, 1918 entry->end); 1919 pmap_kremove(entry->start, len); 1920 } 1921 } else if (UVM_ET_ISOBJ(entry) && 1922 UVM_OBJ_IS_KERN_OBJECT(entry->object.uvm_obj)) { 1923 KASSERT(vm_map_pmap(map) == pmap_kernel()); 1924 1925 /* 1926 * note: kernel object mappings are currently used in 1927 * two ways: 1928 * [1] "normal" mappings of pages in the kernel object 1929 * [2] uvm_km_valloc'd allocations in which we 1930 * pmap_enter in some non-kernel-object page 1931 * (e.g. vmapbuf). 1932 * 1933 * for case [1], we need to remove the mapping from 1934 * the pmap and then remove the page from the kernel 1935 * object (because, once pages in a kernel object are 1936 * unmapped they are no longer needed, unlike, say, 1937 * a vnode where you might want the data to persist 1938 * until flushed out of a queue). 1939 * 1940 * for case [2], we need to remove the mapping from 1941 * the pmap. there shouldn't be any pages at the 1942 * specified offset in the kernel object [but it 1943 * doesn't hurt to call uvm_km_pgremove just to be 1944 * safe?] 1945 * 1946 * uvm_km_pgremove currently does the following: 1947 * for pages in the kernel object in range: 1948 * - drops the swap slot 1949 * - uvm_pagefree the page 1950 */ 1951 1952 /* 1953 * remove mappings from pmap and drop the pages 1954 * from the object. offsets are always relative 1955 * to vm_map_min(kernel_map). 1956 */ 1957 1958 pmap_remove(pmap_kernel(), entry->start, 1959 entry->start + len); 1960 uvm_km_pgremove(entry->object.uvm_obj, 1961 entry->start - vm_map_min(kernel_map), 1962 entry->end - vm_map_min(kernel_map)); 1963 1964 /* 1965 * null out kernel_object reference, we've just 1966 * dropped it 1967 */ 1968 1969 entry->etype &= ~UVM_ET_OBJ; 1970 entry->object.uvm_obj = NULL; 1971 } else if (UVM_ET_ISOBJ(entry) || entry->aref.ar_amap) { 1972 1973 /* 1974 * remove mappings the standard way. 1975 */ 1976 1977 pmap_remove(map->pmap, entry->start, entry->end); 1978 } 1979 1980 #if defined(DEBUG) 1981 if ((entry->flags & UVM_MAP_KMAPENT) == 0) { 1982 1983 /* 1984 * check if there's remaining mapping, 1985 * which is a bug in caller. 1986 */ 1987 1988 vaddr_t va; 1989 for (va = entry->start; va < entry->end; 1990 va += PAGE_SIZE) { 1991 if (pmap_extract(vm_map_pmap(map), va, NULL)) { 1992 panic("uvm_unmap_remove: has mapping"); 1993 } 1994 } 1995 } 1996 #endif /* defined(DEBUG) */ 1997 1998 /* 1999 * remove entry from map and put it on our list of entries 2000 * that we've nuked. then go to next entry. 2001 */ 2002 2003 UVMHIST_LOG(maphist, " removed map entry 0x%x", entry, 0, 0,0); 2004 2005 /* critical! prevents stale hint */ 2006 SAVE_HINT(map, entry, entry->prev); 2007 2008 uvm_map_entry_unlink(map, entry); 2009 KASSERT(map->size >= len); 2010 map->size -= len; 2011 entry->prev = NULL; 2012 entry->next = first_entry; 2013 first_entry = entry; 2014 entry = next; 2015 } 2016 if ((map->flags & VM_MAP_DYING) == 0) { 2017 pmap_update(vm_map_pmap(map)); 2018 } 2019 2020 uvm_tree_sanity(map, "unmap_remove leave"); 2021 2022 /* 2023 * now we've cleaned up the map and are ready for the caller to drop 2024 * references to the mapped objects. 2025 */ 2026 2027 *entry_list = first_entry; 2028 UVMHIST_LOG(maphist,"<- done!", 0, 0, 0, 0); 2029 2030 simple_lock(&map->flags_lock); 2031 if (map->flags & VM_MAP_WANTVA) { 2032 map->flags &= ~VM_MAP_WANTVA; 2033 wakeup(&map->header); 2034 } 2035 simple_unlock(&map->flags_lock); 2036 } 2037 2038 /* 2039 * uvm_unmap_detach: drop references in a chain of map entries 2040 * 2041 * => we will free the map entries as we traverse the list. 2042 */ 2043 2044 void 2045 uvm_unmap_detach(struct vm_map_entry *first_entry, int flags) 2046 { 2047 struct vm_map_entry *next_entry; 2048 UVMHIST_FUNC("uvm_unmap_detach"); UVMHIST_CALLED(maphist); 2049 2050 while (first_entry) { 2051 KASSERT(!VM_MAPENT_ISWIRED(first_entry)); 2052 UVMHIST_LOG(maphist, 2053 " detach 0x%x: amap=0x%x, obj=0x%x, submap?=%d", 2054 first_entry, first_entry->aref.ar_amap, 2055 first_entry->object.uvm_obj, 2056 UVM_ET_ISSUBMAP(first_entry)); 2057 2058 /* 2059 * drop reference to amap, if we've got one 2060 */ 2061 2062 if (first_entry->aref.ar_amap) 2063 uvm_map_unreference_amap(first_entry, flags); 2064 2065 /* 2066 * drop reference to our backing object, if we've got one 2067 */ 2068 2069 KASSERT(!UVM_ET_ISSUBMAP(first_entry)); 2070 if (UVM_ET_ISOBJ(first_entry) && 2071 first_entry->object.uvm_obj->pgops->pgo_detach) { 2072 (*first_entry->object.uvm_obj->pgops->pgo_detach) 2073 (first_entry->object.uvm_obj); 2074 } 2075 next_entry = first_entry->next; 2076 uvm_mapent_free(first_entry); 2077 first_entry = next_entry; 2078 } 2079 UVMHIST_LOG(maphist, "<- done", 0,0,0,0); 2080 } 2081 2082 /* 2083 * E X T R A C T I O N F U N C T I O N S 2084 */ 2085 2086 /* 2087 * uvm_map_reserve: reserve space in a vm_map for future use. 2088 * 2089 * => we reserve space in a map by putting a dummy map entry in the 2090 * map (dummy means obj=NULL, amap=NULL, prot=VM_PROT_NONE) 2091 * => map should be unlocked (we will write lock it) 2092 * => we return true if we were able to reserve space 2093 * => XXXCDC: should be inline? 2094 */ 2095 2096 int 2097 uvm_map_reserve(struct vm_map *map, vsize_t size, 2098 vaddr_t offset /* hint for pmap_prefer */, 2099 vsize_t align /* alignment hint */, 2100 vaddr_t *raddr /* IN:hint, OUT: reserved VA */) 2101 { 2102 UVMHIST_FUNC("uvm_map_reserve"); UVMHIST_CALLED(maphist); 2103 2104 UVMHIST_LOG(maphist, "(map=0x%x, size=0x%x, offset=0x%x,addr=0x%x)", 2105 map,size,offset,raddr); 2106 2107 size = round_page(size); 2108 if (*raddr < vm_map_min(map)) 2109 *raddr = vm_map_min(map); /* hint */ 2110 2111 /* 2112 * reserve some virtual space. 2113 */ 2114 2115 if (uvm_map(map, raddr, size, NULL, offset, 0, 2116 UVM_MAPFLAG(UVM_PROT_NONE, UVM_PROT_NONE, UVM_INH_NONE, 2117 UVM_ADV_RANDOM, UVM_FLAG_NOMERGE)) != 0) { 2118 UVMHIST_LOG(maphist, "<- done (no VM)", 0,0,0,0); 2119 return (FALSE); 2120 } 2121 2122 UVMHIST_LOG(maphist, "<- done (*raddr=0x%x)", *raddr,0,0,0); 2123 return (TRUE); 2124 } 2125 2126 /* 2127 * uvm_map_replace: replace a reserved (blank) area of memory with 2128 * real mappings. 2129 * 2130 * => caller must WRITE-LOCK the map 2131 * => we return TRUE if replacement was a success 2132 * => we expect the newents chain to have nnewents entrys on it and 2133 * we expect newents->prev to point to the last entry on the list 2134 * => note newents is allowed to be NULL 2135 */ 2136 2137 int 2138 uvm_map_replace(struct vm_map *map, vaddr_t start, vaddr_t end, 2139 struct vm_map_entry *newents, int nnewents) 2140 { 2141 struct vm_map_entry *oldent, *last; 2142 2143 uvm_tree_sanity(map, "map_replace entry"); 2144 2145 /* 2146 * first find the blank map entry at the specified address 2147 */ 2148 2149 if (!uvm_map_lookup_entry(map, start, &oldent)) { 2150 return (FALSE); 2151 } 2152 2153 /* 2154 * check to make sure we have a proper blank entry 2155 */ 2156 2157 if (oldent->start != start || oldent->end != end || 2158 oldent->object.uvm_obj != NULL || oldent->aref.ar_amap != NULL) { 2159 return (FALSE); 2160 } 2161 2162 #ifdef DIAGNOSTIC 2163 2164 /* 2165 * sanity check the newents chain 2166 */ 2167 2168 { 2169 struct vm_map_entry *tmpent = newents; 2170 int nent = 0; 2171 vaddr_t cur = start; 2172 2173 while (tmpent) { 2174 nent++; 2175 if (tmpent->start < cur) 2176 panic("uvm_map_replace1"); 2177 if (tmpent->start > tmpent->end || tmpent->end > end) { 2178 printf("tmpent->start=0x%lx, tmpent->end=0x%lx, end=0x%lx\n", 2179 tmpent->start, tmpent->end, end); 2180 panic("uvm_map_replace2"); 2181 } 2182 cur = tmpent->end; 2183 if (tmpent->next) { 2184 if (tmpent->next->prev != tmpent) 2185 panic("uvm_map_replace3"); 2186 } else { 2187 if (newents->prev != tmpent) 2188 panic("uvm_map_replace4"); 2189 } 2190 tmpent = tmpent->next; 2191 } 2192 if (nent != nnewents) 2193 panic("uvm_map_replace5"); 2194 } 2195 #endif 2196 2197 /* 2198 * map entry is a valid blank! replace it. (this does all the 2199 * work of map entry link/unlink...). 2200 */ 2201 2202 if (newents) { 2203 last = newents->prev; 2204 2205 /* critical: flush stale hints out of map */ 2206 SAVE_HINT(map, map->hint, newents); 2207 if (map->first_free == oldent) 2208 map->first_free = last; 2209 2210 last->next = oldent->next; 2211 last->next->prev = last; 2212 2213 /* Fix RB tree */ 2214 uvm_rb_remove(map, oldent); 2215 2216 newents->prev = oldent->prev; 2217 newents->prev->next = newents; 2218 map->nentries = map->nentries + (nnewents - 1); 2219 2220 /* Fixup the RB tree */ 2221 { 2222 int i; 2223 struct vm_map_entry *tmp; 2224 2225 tmp = newents; 2226 for (i = 0; i < nnewents && tmp; i++) { 2227 uvm_rb_insert(map, tmp); 2228 tmp = tmp->next; 2229 } 2230 } 2231 } else { 2232 2233 /* critical: flush stale hints out of map */ 2234 SAVE_HINT(map, map->hint, oldent->prev); 2235 if (map->first_free == oldent) 2236 map->first_free = oldent->prev; 2237 2238 /* NULL list of new entries: just remove the old one */ 2239 uvm_map_entry_unlink(map, oldent); 2240 } 2241 2242 uvm_tree_sanity(map, "map_replace leave"); 2243 2244 /* 2245 * now we can free the old blank entry, unlock the map and return. 2246 */ 2247 2248 uvm_mapent_free(oldent); 2249 return (TRUE); 2250 } 2251 2252 /* 2253 * uvm_map_extract: extract a mapping from a map and put it somewhere 2254 * (maybe removing the old mapping) 2255 * 2256 * => maps should be unlocked (we will write lock them) 2257 * => returns 0 on success, error code otherwise 2258 * => start must be page aligned 2259 * => len must be page sized 2260 * => flags: 2261 * UVM_EXTRACT_REMOVE: remove mappings from srcmap 2262 * UVM_EXTRACT_CONTIG: abort if unmapped area (advisory only) 2263 * UVM_EXTRACT_QREF: for a temporary extraction do quick obj refs 2264 * UVM_EXTRACT_FIXPROT: set prot to maxprot as we go 2265 * >>>NOTE: if you set REMOVE, you are not allowed to use CONTIG or QREF!<<< 2266 * >>>NOTE: QREF's must be unmapped via the QREF path, thus should only 2267 * be used from within the kernel in a kernel level map <<< 2268 */ 2269 2270 int 2271 uvm_map_extract(struct vm_map *srcmap, vaddr_t start, vsize_t len, 2272 struct vm_map *dstmap, vaddr_t *dstaddrp, int flags) 2273 { 2274 vaddr_t dstaddr, end, newend, oldoffset, fudge, orig_fudge; 2275 struct vm_map_entry *chain, *endchain, *entry, *orig_entry, *newentry, 2276 *deadentry, *oldentry; 2277 vsize_t elen; 2278 int nchain, error, copy_ok; 2279 UVMHIST_FUNC("uvm_map_extract"); UVMHIST_CALLED(maphist); 2280 2281 UVMHIST_LOG(maphist,"(srcmap=0x%x,start=0x%x, len=0x%x", srcmap, start, 2282 len,0); 2283 UVMHIST_LOG(maphist," ...,dstmap=0x%x, flags=0x%x)", dstmap,flags,0,0); 2284 2285 uvm_tree_sanity(srcmap, "map_extract src enter"); 2286 uvm_tree_sanity(dstmap, "map_extract dst enter"); 2287 2288 /* 2289 * step 0: sanity check: start must be on a page boundary, length 2290 * must be page sized. can't ask for CONTIG/QREF if you asked for 2291 * REMOVE. 2292 */ 2293 2294 KASSERT((start & PAGE_MASK) == 0 && (len & PAGE_MASK) == 0); 2295 KASSERT((flags & UVM_EXTRACT_REMOVE) == 0 || 2296 (flags & (UVM_EXTRACT_CONTIG|UVM_EXTRACT_QREF)) == 0); 2297 2298 /* 2299 * step 1: reserve space in the target map for the extracted area 2300 */ 2301 2302 dstaddr = vm_map_min(dstmap); 2303 if (uvm_map_reserve(dstmap, len, start, 0, &dstaddr) == FALSE) 2304 return (ENOMEM); 2305 *dstaddrp = dstaddr; /* pass address back to caller */ 2306 UVMHIST_LOG(maphist, " dstaddr=0x%x", dstaddr,0,0,0); 2307 2308 /* 2309 * step 2: setup for the extraction process loop by init'ing the 2310 * map entry chain, locking src map, and looking up the first useful 2311 * entry in the map. 2312 */ 2313 2314 end = start + len; 2315 newend = dstaddr + len; 2316 chain = endchain = NULL; 2317 nchain = 0; 2318 vm_map_lock(srcmap); 2319 2320 if (uvm_map_lookup_entry(srcmap, start, &entry)) { 2321 2322 /* "start" is within an entry */ 2323 if (flags & UVM_EXTRACT_QREF) { 2324 2325 /* 2326 * for quick references we don't clip the entry, so 2327 * the entry may map space "before" the starting 2328 * virtual address... this is the "fudge" factor 2329 * (which can be non-zero only the first time 2330 * through the "while" loop in step 3). 2331 */ 2332 2333 fudge = start - entry->start; 2334 } else { 2335 2336 /* 2337 * normal reference: we clip the map to fit (thus 2338 * fudge is zero) 2339 */ 2340 2341 UVM_MAP_CLIP_START(srcmap, entry, start, NULL); 2342 SAVE_HINT(srcmap, srcmap->hint, entry->prev); 2343 fudge = 0; 2344 } 2345 } else { 2346 2347 /* "start" is not within an entry ... skip to next entry */ 2348 if (flags & UVM_EXTRACT_CONTIG) { 2349 error = EINVAL; 2350 goto bad; /* definite hole here ... */ 2351 } 2352 2353 entry = entry->next; 2354 fudge = 0; 2355 } 2356 2357 /* save values from srcmap for step 6 */ 2358 orig_entry = entry; 2359 orig_fudge = fudge; 2360 2361 /* 2362 * step 3: now start looping through the map entries, extracting 2363 * as we go. 2364 */ 2365 2366 while (entry->start < end && entry != &srcmap->header) { 2367 2368 /* if we are not doing a quick reference, clip it */ 2369 if ((flags & UVM_EXTRACT_QREF) == 0) 2370 UVM_MAP_CLIP_END(srcmap, entry, end, NULL); 2371 2372 /* clear needs_copy (allow chunking) */ 2373 if (UVM_ET_ISNEEDSCOPY(entry)) { 2374 amap_copy(srcmap, entry, M_NOWAIT, TRUE, start, end); 2375 if (UVM_ET_ISNEEDSCOPY(entry)) { /* failed? */ 2376 error = ENOMEM; 2377 goto bad; 2378 } 2379 2380 /* amap_copy could clip (during chunk)! update fudge */ 2381 if (fudge) { 2382 fudge = start - entry->start; 2383 orig_fudge = fudge; 2384 } 2385 } 2386 2387 /* calculate the offset of this from "start" */ 2388 oldoffset = (entry->start + fudge) - start; 2389 2390 /* allocate a new map entry */ 2391 newentry = uvm_mapent_alloc(dstmap, 0); 2392 if (newentry == NULL) { 2393 error = ENOMEM; 2394 goto bad; 2395 } 2396 2397 /* set up new map entry */ 2398 newentry->next = NULL; 2399 newentry->prev = endchain; 2400 newentry->start = dstaddr + oldoffset; 2401 newentry->end = 2402 newentry->start + (entry->end - (entry->start + fudge)); 2403 if (newentry->end > newend || newentry->end < newentry->start) 2404 newentry->end = newend; 2405 newentry->object.uvm_obj = entry->object.uvm_obj; 2406 if (newentry->object.uvm_obj) { 2407 if (newentry->object.uvm_obj->pgops->pgo_reference) 2408 newentry->object.uvm_obj->pgops-> 2409 pgo_reference(newentry->object.uvm_obj); 2410 newentry->offset = entry->offset + fudge; 2411 } else { 2412 newentry->offset = 0; 2413 } 2414 newentry->etype = entry->etype; 2415 newentry->protection = (flags & UVM_EXTRACT_FIXPROT) ? 2416 entry->max_protection : entry->protection; 2417 newentry->max_protection = entry->max_protection; 2418 newentry->inheritance = entry->inheritance; 2419 newentry->wired_count = 0; 2420 newentry->aref.ar_amap = entry->aref.ar_amap; 2421 if (newentry->aref.ar_amap) { 2422 newentry->aref.ar_pageoff = 2423 entry->aref.ar_pageoff + (fudge >> PAGE_SHIFT); 2424 uvm_map_reference_amap(newentry, AMAP_SHARED | 2425 ((flags & UVM_EXTRACT_QREF) ? AMAP_REFALL : 0)); 2426 } else { 2427 newentry->aref.ar_pageoff = 0; 2428 } 2429 newentry->advice = entry->advice; 2430 2431 /* now link it on the chain */ 2432 nchain++; 2433 if (endchain == NULL) { 2434 chain = endchain = newentry; 2435 } else { 2436 endchain->next = newentry; 2437 endchain = newentry; 2438 } 2439 2440 /* end of 'while' loop! */ 2441 if ((flags & UVM_EXTRACT_CONTIG) && entry->end < end && 2442 (entry->next == &srcmap->header || 2443 entry->next->start != entry->end)) { 2444 error = EINVAL; 2445 goto bad; 2446 } 2447 entry = entry->next; 2448 fudge = 0; 2449 } 2450 2451 /* 2452 * step 4: close off chain (in format expected by uvm_map_replace) 2453 */ 2454 2455 if (chain) 2456 chain->prev = endchain; 2457 2458 /* 2459 * step 5: attempt to lock the dest map so we can pmap_copy. 2460 * note usage of copy_ok: 2461 * 1 => dstmap locked, pmap_copy ok, and we "replace" here (step 5) 2462 * 0 => dstmap unlocked, NO pmap_copy, and we will "replace" in step 7 2463 */ 2464 2465 if (srcmap == dstmap || vm_map_lock_try(dstmap) == TRUE) { 2466 copy_ok = 1; 2467 if (!uvm_map_replace(dstmap, dstaddr, dstaddr+len, chain, 2468 nchain)) { 2469 if (srcmap != dstmap) 2470 vm_map_unlock(dstmap); 2471 error = EIO; 2472 goto bad; 2473 } 2474 } else { 2475 copy_ok = 0; 2476 /* replace defered until step 7 */ 2477 } 2478 2479 /* 2480 * step 6: traverse the srcmap a second time to do the following: 2481 * - if we got a lock on the dstmap do pmap_copy 2482 * - if UVM_EXTRACT_REMOVE remove the entries 2483 * we make use of orig_entry and orig_fudge (saved in step 2) 2484 */ 2485 2486 if (copy_ok || (flags & UVM_EXTRACT_REMOVE)) { 2487 2488 /* purge possible stale hints from srcmap */ 2489 if (flags & UVM_EXTRACT_REMOVE) { 2490 SAVE_HINT(srcmap, srcmap->hint, orig_entry->prev); 2491 if (srcmap->first_free->start >= start) 2492 srcmap->first_free = orig_entry->prev; 2493 } 2494 2495 entry = orig_entry; 2496 fudge = orig_fudge; 2497 deadentry = NULL; /* for UVM_EXTRACT_REMOVE */ 2498 2499 while (entry->start < end && entry != &srcmap->header) { 2500 if (copy_ok) { 2501 oldoffset = (entry->start + fudge) - start; 2502 elen = MIN(end, entry->end) - 2503 (entry->start + fudge); 2504 pmap_copy(dstmap->pmap, srcmap->pmap, 2505 dstaddr + oldoffset, elen, 2506 entry->start + fudge); 2507 } 2508 2509 /* we advance "entry" in the following if statement */ 2510 if (flags & UVM_EXTRACT_REMOVE) { 2511 pmap_remove(srcmap->pmap, entry->start, 2512 entry->end); 2513 oldentry = entry; /* save entry */ 2514 entry = entry->next; /* advance */ 2515 uvm_map_entry_unlink(srcmap, oldentry); 2516 /* add to dead list */ 2517 oldentry->next = deadentry; 2518 deadentry = oldentry; 2519 } else { 2520 entry = entry->next; /* advance */ 2521 } 2522 2523 /* end of 'while' loop */ 2524 fudge = 0; 2525 } 2526 pmap_update(srcmap->pmap); 2527 2528 /* 2529 * unlock dstmap. we will dispose of deadentry in 2530 * step 7 if needed 2531 */ 2532 2533 if (copy_ok && srcmap != dstmap) 2534 vm_map_unlock(dstmap); 2535 2536 } else { 2537 deadentry = NULL; 2538 } 2539 2540 /* 2541 * step 7: we are done with the source map, unlock. if copy_ok 2542 * is 0 then we have not replaced the dummy mapping in dstmap yet 2543 * and we need to do so now. 2544 */ 2545 2546 vm_map_unlock(srcmap); 2547 if ((flags & UVM_EXTRACT_REMOVE) && deadentry) 2548 uvm_unmap_detach(deadentry, 0); /* dispose of old entries */ 2549 2550 /* now do the replacement if we didn't do it in step 5 */ 2551 if (copy_ok == 0) { 2552 vm_map_lock(dstmap); 2553 error = uvm_map_replace(dstmap, dstaddr, dstaddr+len, chain, 2554 nchain); 2555 vm_map_unlock(dstmap); 2556 2557 if (error == FALSE) { 2558 error = EIO; 2559 goto bad2; 2560 } 2561 } 2562 2563 uvm_tree_sanity(srcmap, "map_extract src leave"); 2564 uvm_tree_sanity(dstmap, "map_extract dst leave"); 2565 2566 return (0); 2567 2568 /* 2569 * bad: failure recovery 2570 */ 2571 bad: 2572 vm_map_unlock(srcmap); 2573 bad2: /* src already unlocked */ 2574 if (chain) 2575 uvm_unmap_detach(chain, 2576 (flags & UVM_EXTRACT_QREF) ? AMAP_REFALL : 0); 2577 2578 uvm_tree_sanity(srcmap, "map_extract src err leave"); 2579 uvm_tree_sanity(dstmap, "map_extract dst err leave"); 2580 2581 uvm_unmap(dstmap, dstaddr, dstaddr+len); /* ??? */ 2582 return (error); 2583 } 2584 2585 /* end of extraction functions */ 2586 2587 /* 2588 * uvm_map_submap: punch down part of a map into a submap 2589 * 2590 * => only the kernel_map is allowed to be submapped 2591 * => the purpose of submapping is to break up the locking granularity 2592 * of a larger map 2593 * => the range specified must have been mapped previously with a uvm_map() 2594 * call [with uobj==NULL] to create a blank map entry in the main map. 2595 * [And it had better still be blank!] 2596 * => maps which contain submaps should never be copied or forked. 2597 * => to remove a submap, use uvm_unmap() on the main map 2598 * and then uvm_map_deallocate() the submap. 2599 * => main map must be unlocked. 2600 * => submap must have been init'd and have a zero reference count. 2601 * [need not be locked as we don't actually reference it] 2602 */ 2603 2604 int 2605 uvm_map_submap(struct vm_map *map, vaddr_t start, vaddr_t end, 2606 struct vm_map *submap) 2607 { 2608 struct vm_map_entry *entry; 2609 struct uvm_mapent_reservation umr; 2610 int error; 2611 2612 uvm_mapent_reserve(map, &umr, 2, 0); 2613 2614 vm_map_lock(map); 2615 VM_MAP_RANGE_CHECK(map, start, end); 2616 2617 if (uvm_map_lookup_entry(map, start, &entry)) { 2618 UVM_MAP_CLIP_START(map, entry, start, &umr); 2619 UVM_MAP_CLIP_END(map, entry, end, &umr); /* to be safe */ 2620 } else { 2621 entry = NULL; 2622 } 2623 2624 if (entry != NULL && 2625 entry->start == start && entry->end == end && 2626 entry->object.uvm_obj == NULL && entry->aref.ar_amap == NULL && 2627 !UVM_ET_ISCOPYONWRITE(entry) && !UVM_ET_ISNEEDSCOPY(entry)) { 2628 entry->etype |= UVM_ET_SUBMAP; 2629 entry->object.sub_map = submap; 2630 entry->offset = 0; 2631 uvm_map_reference(submap); 2632 error = 0; 2633 } else { 2634 error = EINVAL; 2635 } 2636 vm_map_unlock(map); 2637 2638 uvm_mapent_unreserve(map, &umr); 2639 2640 return error; 2641 } 2642 2643 /* 2644 * uvm_map_setup_kernel: init in-kernel map 2645 * 2646 * => map must not be in service yet. 2647 */ 2648 2649 void 2650 uvm_map_setup_kernel(struct vm_map_kernel *map, 2651 vaddr_t min, vaddr_t max, int flags) 2652 { 2653 2654 uvm_map_setup(&map->vmk_map, min, max, flags); 2655 2656 LIST_INIT(&map->vmk_kentry_free); 2657 map->vmk_merged_entries = NULL; 2658 } 2659 2660 2661 /* 2662 * uvm_map_protect: change map protection 2663 * 2664 * => set_max means set max_protection. 2665 * => map must be unlocked. 2666 */ 2667 2668 #define MASK(entry) (UVM_ET_ISCOPYONWRITE(entry) ? \ 2669 ~VM_PROT_WRITE : VM_PROT_ALL) 2670 2671 int 2672 uvm_map_protect(struct vm_map *map, vaddr_t start, vaddr_t end, 2673 vm_prot_t new_prot, boolean_t set_max) 2674 { 2675 struct vm_map_entry *current, *entry; 2676 int error = 0; 2677 UVMHIST_FUNC("uvm_map_protect"); UVMHIST_CALLED(maphist); 2678 UVMHIST_LOG(maphist,"(map=0x%x,start=0x%x,end=0x%x,new_prot=0x%x)", 2679 map, start, end, new_prot); 2680 2681 vm_map_lock(map); 2682 VM_MAP_RANGE_CHECK(map, start, end); 2683 if (uvm_map_lookup_entry(map, start, &entry)) { 2684 UVM_MAP_CLIP_START(map, entry, start, NULL); 2685 } else { 2686 entry = entry->next; 2687 } 2688 2689 /* 2690 * make a first pass to check for protection violations. 2691 */ 2692 2693 current = entry; 2694 while ((current != &map->header) && (current->start < end)) { 2695 if (UVM_ET_ISSUBMAP(current)) { 2696 error = EINVAL; 2697 goto out; 2698 } 2699 if ((new_prot & current->max_protection) != new_prot) { 2700 error = EACCES; 2701 goto out; 2702 } 2703 /* 2704 * Don't allow VM_PROT_EXECUTE to be set on entries that 2705 * point to vnodes that are associated with a NOEXEC file 2706 * system. 2707 */ 2708 if (UVM_ET_ISOBJ(current) && 2709 UVM_OBJ_IS_VNODE(current->object.uvm_obj)) { 2710 struct vnode *vp = 2711 (struct vnode *) current->object.uvm_obj; 2712 2713 if ((new_prot & VM_PROT_EXECUTE) != 0 && 2714 (vp->v_mount->mnt_flag & MNT_NOEXEC) != 0) { 2715 error = EACCES; 2716 goto out; 2717 } 2718 } 2719 current = current->next; 2720 } 2721 2722 /* go back and fix up protections (no need to clip this time). */ 2723 2724 current = entry; 2725 while ((current != &map->header) && (current->start < end)) { 2726 vm_prot_t old_prot; 2727 2728 UVM_MAP_CLIP_END(map, current, end, NULL); 2729 old_prot = current->protection; 2730 if (set_max) 2731 current->protection = 2732 (current->max_protection = new_prot) & old_prot; 2733 else 2734 current->protection = new_prot; 2735 2736 /* 2737 * update physical map if necessary. worry about copy-on-write 2738 * here -- CHECK THIS XXX 2739 */ 2740 2741 if (current->protection != old_prot) { 2742 /* update pmap! */ 2743 pmap_protect(map->pmap, current->start, current->end, 2744 current->protection & MASK(entry)); 2745 2746 /* 2747 * If this entry points at a vnode, and the 2748 * protection includes VM_PROT_EXECUTE, mark 2749 * the vnode as VEXECMAP. 2750 */ 2751 if (UVM_ET_ISOBJ(current)) { 2752 struct uvm_object *uobj = 2753 current->object.uvm_obj; 2754 2755 if (UVM_OBJ_IS_VNODE(uobj) && 2756 (current->protection & VM_PROT_EXECUTE)) 2757 vn_markexec((struct vnode *) uobj); 2758 } 2759 } 2760 2761 /* 2762 * If the map is configured to lock any future mappings, 2763 * wire this entry now if the old protection was VM_PROT_NONE 2764 * and the new protection is not VM_PROT_NONE. 2765 */ 2766 2767 if ((map->flags & VM_MAP_WIREFUTURE) != 0 && 2768 VM_MAPENT_ISWIRED(entry) == 0 && 2769 old_prot == VM_PROT_NONE && 2770 new_prot != VM_PROT_NONE) { 2771 if (uvm_map_pageable(map, entry->start, 2772 entry->end, FALSE, 2773 UVM_LK_ENTER|UVM_LK_EXIT) != 0) { 2774 2775 /* 2776 * If locking the entry fails, remember the 2777 * error if it's the first one. Note we 2778 * still continue setting the protection in 2779 * the map, but will return the error 2780 * condition regardless. 2781 * 2782 * XXX Ignore what the actual error is, 2783 * XXX just call it a resource shortage 2784 * XXX so that it doesn't get confused 2785 * XXX what uvm_map_protect() itself would 2786 * XXX normally return. 2787 */ 2788 2789 error = ENOMEM; 2790 } 2791 } 2792 current = current->next; 2793 } 2794 pmap_update(map->pmap); 2795 2796 out: 2797 vm_map_unlock(map); 2798 2799 UVMHIST_LOG(maphist, "<- done, error=%d",error,0,0,0); 2800 return error; 2801 } 2802 2803 #undef MASK 2804 2805 /* 2806 * uvm_map_inherit: set inheritance code for range of addrs in map. 2807 * 2808 * => map must be unlocked 2809 * => note that the inherit code is used during a "fork". see fork 2810 * code for details. 2811 */ 2812 2813 int 2814 uvm_map_inherit(struct vm_map *map, vaddr_t start, vaddr_t end, 2815 vm_inherit_t new_inheritance) 2816 { 2817 struct vm_map_entry *entry, *temp_entry; 2818 UVMHIST_FUNC("uvm_map_inherit"); UVMHIST_CALLED(maphist); 2819 UVMHIST_LOG(maphist,"(map=0x%x,start=0x%x,end=0x%x,new_inh=0x%x)", 2820 map, start, end, new_inheritance); 2821 2822 switch (new_inheritance) { 2823 case MAP_INHERIT_NONE: 2824 case MAP_INHERIT_COPY: 2825 case MAP_INHERIT_SHARE: 2826 break; 2827 default: 2828 UVMHIST_LOG(maphist,"<- done (INVALID ARG)",0,0,0,0); 2829 return EINVAL; 2830 } 2831 2832 vm_map_lock(map); 2833 VM_MAP_RANGE_CHECK(map, start, end); 2834 if (uvm_map_lookup_entry(map, start, &temp_entry)) { 2835 entry = temp_entry; 2836 UVM_MAP_CLIP_START(map, entry, start, NULL); 2837 } else { 2838 entry = temp_entry->next; 2839 } 2840 while ((entry != &map->header) && (entry->start < end)) { 2841 UVM_MAP_CLIP_END(map, entry, end, NULL); 2842 entry->inheritance = new_inheritance; 2843 entry = entry->next; 2844 } 2845 vm_map_unlock(map); 2846 UVMHIST_LOG(maphist,"<- done (OK)",0,0,0,0); 2847 return 0; 2848 } 2849 2850 /* 2851 * uvm_map_advice: set advice code for range of addrs in map. 2852 * 2853 * => map must be unlocked 2854 */ 2855 2856 int 2857 uvm_map_advice(struct vm_map *map, vaddr_t start, vaddr_t end, int new_advice) 2858 { 2859 struct vm_map_entry *entry, *temp_entry; 2860 UVMHIST_FUNC("uvm_map_advice"); UVMHIST_CALLED(maphist); 2861 UVMHIST_LOG(maphist,"(map=0x%x,start=0x%x,end=0x%x,new_adv=0x%x)", 2862 map, start, end, new_advice); 2863 2864 vm_map_lock(map); 2865 VM_MAP_RANGE_CHECK(map, start, end); 2866 if (uvm_map_lookup_entry(map, start, &temp_entry)) { 2867 entry = temp_entry; 2868 UVM_MAP_CLIP_START(map, entry, start, NULL); 2869 } else { 2870 entry = temp_entry->next; 2871 } 2872 2873 /* 2874 * XXXJRT: disallow holes? 2875 */ 2876 2877 while ((entry != &map->header) && (entry->start < end)) { 2878 UVM_MAP_CLIP_END(map, entry, end, NULL); 2879 2880 switch (new_advice) { 2881 case MADV_NORMAL: 2882 case MADV_RANDOM: 2883 case MADV_SEQUENTIAL: 2884 /* nothing special here */ 2885 break; 2886 2887 default: 2888 vm_map_unlock(map); 2889 UVMHIST_LOG(maphist,"<- done (INVALID ARG)",0,0,0,0); 2890 return EINVAL; 2891 } 2892 entry->advice = new_advice; 2893 entry = entry->next; 2894 } 2895 2896 vm_map_unlock(map); 2897 UVMHIST_LOG(maphist,"<- done (OK)",0,0,0,0); 2898 return 0; 2899 } 2900 2901 /* 2902 * uvm_map_pageable: sets the pageability of a range in a map. 2903 * 2904 * => wires map entries. should not be used for transient page locking. 2905 * for that, use uvm_fault_wire()/uvm_fault_unwire() (see uvm_vslock()). 2906 * => regions sepcified as not pageable require lock-down (wired) memory 2907 * and page tables. 2908 * => map must never be read-locked 2909 * => if islocked is TRUE, map is already write-locked 2910 * => we always unlock the map, since we must downgrade to a read-lock 2911 * to call uvm_fault_wire() 2912 * => XXXCDC: check this and try and clean it up. 2913 */ 2914 2915 int 2916 uvm_map_pageable(struct vm_map *map, vaddr_t start, vaddr_t end, 2917 boolean_t new_pageable, int lockflags) 2918 { 2919 struct vm_map_entry *entry, *start_entry, *failed_entry; 2920 int rv; 2921 #ifdef DIAGNOSTIC 2922 u_int timestamp_save; 2923 #endif 2924 UVMHIST_FUNC("uvm_map_pageable"); UVMHIST_CALLED(maphist); 2925 UVMHIST_LOG(maphist,"(map=0x%x,start=0x%x,end=0x%x,new_pageable=0x%x)", 2926 map, start, end, new_pageable); 2927 KASSERT(map->flags & VM_MAP_PAGEABLE); 2928 2929 if ((lockflags & UVM_LK_ENTER) == 0) 2930 vm_map_lock(map); 2931 VM_MAP_RANGE_CHECK(map, start, end); 2932 2933 /* 2934 * only one pageability change may take place at one time, since 2935 * uvm_fault_wire assumes it will be called only once for each 2936 * wiring/unwiring. therefore, we have to make sure we're actually 2937 * changing the pageability for the entire region. we do so before 2938 * making any changes. 2939 */ 2940 2941 if (uvm_map_lookup_entry(map, start, &start_entry) == FALSE) { 2942 if ((lockflags & UVM_LK_EXIT) == 0) 2943 vm_map_unlock(map); 2944 2945 UVMHIST_LOG(maphist,"<- done (fault)",0,0,0,0); 2946 return EFAULT; 2947 } 2948 entry = start_entry; 2949 2950 /* 2951 * handle wiring and unwiring separately. 2952 */ 2953 2954 if (new_pageable) { /* unwire */ 2955 UVM_MAP_CLIP_START(map, entry, start, NULL); 2956 2957 /* 2958 * unwiring. first ensure that the range to be unwired is 2959 * really wired down and that there are no holes. 2960 */ 2961 2962 while ((entry != &map->header) && (entry->start < end)) { 2963 if (entry->wired_count == 0 || 2964 (entry->end < end && 2965 (entry->next == &map->header || 2966 entry->next->start > entry->end))) { 2967 if ((lockflags & UVM_LK_EXIT) == 0) 2968 vm_map_unlock(map); 2969 UVMHIST_LOG(maphist, "<- done (INVAL)",0,0,0,0); 2970 return EINVAL; 2971 } 2972 entry = entry->next; 2973 } 2974 2975 /* 2976 * POSIX 1003.1b - a single munlock call unlocks a region, 2977 * regardless of the number of mlock calls made on that 2978 * region. 2979 */ 2980 2981 entry = start_entry; 2982 while ((entry != &map->header) && (entry->start < end)) { 2983 UVM_MAP_CLIP_END(map, entry, end, NULL); 2984 if (VM_MAPENT_ISWIRED(entry)) 2985 uvm_map_entry_unwire(map, entry); 2986 entry = entry->next; 2987 } 2988 if ((lockflags & UVM_LK_EXIT) == 0) 2989 vm_map_unlock(map); 2990 UVMHIST_LOG(maphist,"<- done (OK UNWIRE)",0,0,0,0); 2991 return 0; 2992 } 2993 2994 /* 2995 * wire case: in two passes [XXXCDC: ugly block of code here] 2996 * 2997 * 1: holding the write lock, we create any anonymous maps that need 2998 * to be created. then we clip each map entry to the region to 2999 * be wired and increment its wiring count. 3000 * 3001 * 2: we downgrade to a read lock, and call uvm_fault_wire to fault 3002 * in the pages for any newly wired area (wired_count == 1). 3003 * 3004 * downgrading to a read lock for uvm_fault_wire avoids a possible 3005 * deadlock with another thread that may have faulted on one of 3006 * the pages to be wired (it would mark the page busy, blocking 3007 * us, then in turn block on the map lock that we hold). because 3008 * of problems in the recursive lock package, we cannot upgrade 3009 * to a write lock in vm_map_lookup. thus, any actions that 3010 * require the write lock must be done beforehand. because we 3011 * keep the read lock on the map, the copy-on-write status of the 3012 * entries we modify here cannot change. 3013 */ 3014 3015 while ((entry != &map->header) && (entry->start < end)) { 3016 if (VM_MAPENT_ISWIRED(entry) == 0) { /* not already wired? */ 3017 3018 /* 3019 * perform actions of vm_map_lookup that need the 3020 * write lock on the map: create an anonymous map 3021 * for a copy-on-write region, or an anonymous map 3022 * for a zero-fill region. (XXXCDC: submap case 3023 * ok?) 3024 */ 3025 3026 if (!UVM_ET_ISSUBMAP(entry)) { /* not submap */ 3027 if (UVM_ET_ISNEEDSCOPY(entry) && 3028 ((entry->max_protection & VM_PROT_WRITE) || 3029 (entry->object.uvm_obj == NULL))) { 3030 amap_copy(map, entry, M_WAITOK, TRUE, 3031 start, end); 3032 /* XXXCDC: wait OK? */ 3033 } 3034 } 3035 } 3036 UVM_MAP_CLIP_START(map, entry, start, NULL); 3037 UVM_MAP_CLIP_END(map, entry, end, NULL); 3038 entry->wired_count++; 3039 3040 /* 3041 * Check for holes 3042 */ 3043 3044 if (entry->protection == VM_PROT_NONE || 3045 (entry->end < end && 3046 (entry->next == &map->header || 3047 entry->next->start > entry->end))) { 3048 3049 /* 3050 * found one. amap creation actions do not need to 3051 * be undone, but the wired counts need to be restored. 3052 */ 3053 3054 while (entry != &map->header && entry->end > start) { 3055 entry->wired_count--; 3056 entry = entry->prev; 3057 } 3058 if ((lockflags & UVM_LK_EXIT) == 0) 3059 vm_map_unlock(map); 3060 UVMHIST_LOG(maphist,"<- done (INVALID WIRE)",0,0,0,0); 3061 return EINVAL; 3062 } 3063 entry = entry->next; 3064 } 3065 3066 /* 3067 * Pass 2. 3068 */ 3069 3070 #ifdef DIAGNOSTIC 3071 timestamp_save = map->timestamp; 3072 #endif 3073 vm_map_busy(map); 3074 vm_map_downgrade(map); 3075 3076 rv = 0; 3077 entry = start_entry; 3078 while (entry != &map->header && entry->start < end) { 3079 if (entry->wired_count == 1) { 3080 rv = uvm_fault_wire(map, entry->start, entry->end, 3081 VM_FAULT_WIREMAX, entry->max_protection); 3082 if (rv) { 3083 3084 /* 3085 * wiring failed. break out of the loop. 3086 * we'll clean up the map below, once we 3087 * have a write lock again. 3088 */ 3089 3090 break; 3091 } 3092 } 3093 entry = entry->next; 3094 } 3095 3096 if (rv) { /* failed? */ 3097 3098 /* 3099 * Get back to an exclusive (write) lock. 3100 */ 3101 3102 vm_map_upgrade(map); 3103 vm_map_unbusy(map); 3104 3105 #ifdef DIAGNOSTIC 3106 if (timestamp_save != map->timestamp) 3107 panic("uvm_map_pageable: stale map"); 3108 #endif 3109 3110 /* 3111 * first drop the wiring count on all the entries 3112 * which haven't actually been wired yet. 3113 */ 3114 3115 failed_entry = entry; 3116 while (entry != &map->header && entry->start < end) { 3117 entry->wired_count--; 3118 entry = entry->next; 3119 } 3120 3121 /* 3122 * now, unwire all the entries that were successfully 3123 * wired above. 3124 */ 3125 3126 entry = start_entry; 3127 while (entry != failed_entry) { 3128 entry->wired_count--; 3129 if (VM_MAPENT_ISWIRED(entry) == 0) 3130 uvm_map_entry_unwire(map, entry); 3131 entry = entry->next; 3132 } 3133 if ((lockflags & UVM_LK_EXIT) == 0) 3134 vm_map_unlock(map); 3135 UVMHIST_LOG(maphist, "<- done (RV=%d)", rv,0,0,0); 3136 return (rv); 3137 } 3138 3139 /* We are holding a read lock here. */ 3140 if ((lockflags & UVM_LK_EXIT) == 0) { 3141 vm_map_unbusy(map); 3142 vm_map_unlock_read(map); 3143 } else { 3144 3145 /* 3146 * Get back to an exclusive (write) lock. 3147 */ 3148 3149 vm_map_upgrade(map); 3150 vm_map_unbusy(map); 3151 } 3152 3153 UVMHIST_LOG(maphist,"<- done (OK WIRE)",0,0,0,0); 3154 return 0; 3155 } 3156 3157 /* 3158 * uvm_map_pageable_all: special case of uvm_map_pageable - affects 3159 * all mapped regions. 3160 * 3161 * => map must not be locked. 3162 * => if no flags are specified, all regions are unwired. 3163 * => XXXJRT: has some of the same problems as uvm_map_pageable() above. 3164 */ 3165 3166 int 3167 uvm_map_pageable_all(struct vm_map *map, int flags, vsize_t limit) 3168 { 3169 struct vm_map_entry *entry, *failed_entry; 3170 vsize_t size; 3171 int rv; 3172 #ifdef DIAGNOSTIC 3173 u_int timestamp_save; 3174 #endif 3175 UVMHIST_FUNC("uvm_map_pageable_all"); UVMHIST_CALLED(maphist); 3176 UVMHIST_LOG(maphist,"(map=0x%x,flags=0x%x)", map, flags, 0, 0); 3177 3178 KASSERT(map->flags & VM_MAP_PAGEABLE); 3179 3180 vm_map_lock(map); 3181 3182 /* 3183 * handle wiring and unwiring separately. 3184 */ 3185 3186 if (flags == 0) { /* unwire */ 3187 3188 /* 3189 * POSIX 1003.1b -- munlockall unlocks all regions, 3190 * regardless of how many times mlockall has been called. 3191 */ 3192 3193 for (entry = map->header.next; entry != &map->header; 3194 entry = entry->next) { 3195 if (VM_MAPENT_ISWIRED(entry)) 3196 uvm_map_entry_unwire(map, entry); 3197 } 3198 vm_map_modflags(map, 0, VM_MAP_WIREFUTURE); 3199 vm_map_unlock(map); 3200 UVMHIST_LOG(maphist,"<- done (OK UNWIRE)",0,0,0,0); 3201 return 0; 3202 } 3203 3204 if (flags & MCL_FUTURE) { 3205 3206 /* 3207 * must wire all future mappings; remember this. 3208 */ 3209 3210 vm_map_modflags(map, VM_MAP_WIREFUTURE, 0); 3211 } 3212 3213 if ((flags & MCL_CURRENT) == 0) { 3214 3215 /* 3216 * no more work to do! 3217 */ 3218 3219 UVMHIST_LOG(maphist,"<- done (OK no wire)",0,0,0,0); 3220 vm_map_unlock(map); 3221 return 0; 3222 } 3223 3224 /* 3225 * wire case: in three passes [XXXCDC: ugly block of code here] 3226 * 3227 * 1: holding the write lock, count all pages mapped by non-wired 3228 * entries. if this would cause us to go over our limit, we fail. 3229 * 3230 * 2: still holding the write lock, we create any anonymous maps that 3231 * need to be created. then we increment its wiring count. 3232 * 3233 * 3: we downgrade to a read lock, and call uvm_fault_wire to fault 3234 * in the pages for any newly wired area (wired_count == 1). 3235 * 3236 * downgrading to a read lock for uvm_fault_wire avoids a possible 3237 * deadlock with another thread that may have faulted on one of 3238 * the pages to be wired (it would mark the page busy, blocking 3239 * us, then in turn block on the map lock that we hold). because 3240 * of problems in the recursive lock package, we cannot upgrade 3241 * to a write lock in vm_map_lookup. thus, any actions that 3242 * require the write lock must be done beforehand. because we 3243 * keep the read lock on the map, the copy-on-write status of the 3244 * entries we modify here cannot change. 3245 */ 3246 3247 for (size = 0, entry = map->header.next; entry != &map->header; 3248 entry = entry->next) { 3249 if (entry->protection != VM_PROT_NONE && 3250 VM_MAPENT_ISWIRED(entry) == 0) { /* not already wired? */ 3251 size += entry->end - entry->start; 3252 } 3253 } 3254 3255 if (atop(size) + uvmexp.wired > uvmexp.wiredmax) { 3256 vm_map_unlock(map); 3257 return ENOMEM; 3258 } 3259 3260 if (limit != 0 && 3261 (size + ptoa(pmap_wired_count(vm_map_pmap(map))) > limit)) { 3262 vm_map_unlock(map); 3263 return ENOMEM; 3264 } 3265 3266 /* 3267 * Pass 2. 3268 */ 3269 3270 for (entry = map->header.next; entry != &map->header; 3271 entry = entry->next) { 3272 if (entry->protection == VM_PROT_NONE) 3273 continue; 3274 if (VM_MAPENT_ISWIRED(entry) == 0) { /* not already wired? */ 3275 3276 /* 3277 * perform actions of vm_map_lookup that need the 3278 * write lock on the map: create an anonymous map 3279 * for a copy-on-write region, or an anonymous map 3280 * for a zero-fill region. (XXXCDC: submap case 3281 * ok?) 3282 */ 3283 3284 if (!UVM_ET_ISSUBMAP(entry)) { /* not submap */ 3285 if (UVM_ET_ISNEEDSCOPY(entry) && 3286 ((entry->max_protection & VM_PROT_WRITE) || 3287 (entry->object.uvm_obj == NULL))) { 3288 amap_copy(map, entry, M_WAITOK, TRUE, 3289 entry->start, entry->end); 3290 /* XXXCDC: wait OK? */ 3291 } 3292 } 3293 } 3294 entry->wired_count++; 3295 } 3296 3297 /* 3298 * Pass 3. 3299 */ 3300 3301 #ifdef DIAGNOSTIC 3302 timestamp_save = map->timestamp; 3303 #endif 3304 vm_map_busy(map); 3305 vm_map_downgrade(map); 3306 3307 rv = 0; 3308 for (entry = map->header.next; entry != &map->header; 3309 entry = entry->next) { 3310 if (entry->wired_count == 1) { 3311 rv = uvm_fault_wire(map, entry->start, entry->end, 3312 VM_FAULT_WIREMAX, entry->max_protection); 3313 if (rv) { 3314 3315 /* 3316 * wiring failed. break out of the loop. 3317 * we'll clean up the map below, once we 3318 * have a write lock again. 3319 */ 3320 3321 break; 3322 } 3323 } 3324 } 3325 3326 if (rv) { 3327 3328 /* 3329 * Get back an exclusive (write) lock. 3330 */ 3331 3332 vm_map_upgrade(map); 3333 vm_map_unbusy(map); 3334 3335 #ifdef DIAGNOSTIC 3336 if (timestamp_save != map->timestamp) 3337 panic("uvm_map_pageable_all: stale map"); 3338 #endif 3339 3340 /* 3341 * first drop the wiring count on all the entries 3342 * which haven't actually been wired yet. 3343 * 3344 * Skip VM_PROT_NONE entries like we did above. 3345 */ 3346 3347 failed_entry = entry; 3348 for (/* nothing */; entry != &map->header; 3349 entry = entry->next) { 3350 if (entry->protection == VM_PROT_NONE) 3351 continue; 3352 entry->wired_count--; 3353 } 3354 3355 /* 3356 * now, unwire all the entries that were successfully 3357 * wired above. 3358 * 3359 * Skip VM_PROT_NONE entries like we did above. 3360 */ 3361 3362 for (entry = map->header.next; entry != failed_entry; 3363 entry = entry->next) { 3364 if (entry->protection == VM_PROT_NONE) 3365 continue; 3366 entry->wired_count--; 3367 if (VM_MAPENT_ISWIRED(entry)) 3368 uvm_map_entry_unwire(map, entry); 3369 } 3370 vm_map_unlock(map); 3371 UVMHIST_LOG(maphist,"<- done (RV=%d)", rv,0,0,0); 3372 return (rv); 3373 } 3374 3375 /* We are holding a read lock here. */ 3376 vm_map_unbusy(map); 3377 vm_map_unlock_read(map); 3378 3379 UVMHIST_LOG(maphist,"<- done (OK WIRE)",0,0,0,0); 3380 return 0; 3381 } 3382 3383 /* 3384 * uvm_map_clean: clean out a map range 3385 * 3386 * => valid flags: 3387 * if (flags & PGO_CLEANIT): dirty pages are cleaned first 3388 * if (flags & PGO_SYNCIO): dirty pages are written synchronously 3389 * if (flags & PGO_DEACTIVATE): any cached pages are deactivated after clean 3390 * if (flags & PGO_FREE): any cached pages are freed after clean 3391 * => returns an error if any part of the specified range isn't mapped 3392 * => never a need to flush amap layer since the anonymous memory has 3393 * no permanent home, but may deactivate pages there 3394 * => called from sys_msync() and sys_madvise() 3395 * => caller must not write-lock map (read OK). 3396 * => we may sleep while cleaning if SYNCIO [with map read-locked] 3397 */ 3398 3399 int 3400 uvm_map_clean(struct vm_map *map, vaddr_t start, vaddr_t end, int flags) 3401 { 3402 struct vm_map_entry *current, *entry; 3403 struct uvm_object *uobj; 3404 struct vm_amap *amap; 3405 struct vm_anon *anon; 3406 struct vm_page *pg; 3407 vaddr_t offset; 3408 vsize_t size; 3409 int error, refs; 3410 UVMHIST_FUNC("uvm_map_clean"); UVMHIST_CALLED(maphist); 3411 3412 UVMHIST_LOG(maphist,"(map=0x%x,start=0x%x,end=0x%x,flags=0x%x)", 3413 map, start, end, flags); 3414 KASSERT((flags & (PGO_FREE|PGO_DEACTIVATE)) != 3415 (PGO_FREE|PGO_DEACTIVATE)); 3416 3417 vm_map_lock_read(map); 3418 VM_MAP_RANGE_CHECK(map, start, end); 3419 if (uvm_map_lookup_entry(map, start, &entry) == FALSE) { 3420 vm_map_unlock_read(map); 3421 return EFAULT; 3422 } 3423 3424 /* 3425 * Make a first pass to check for holes. 3426 */ 3427 3428 for (current = entry; current->start < end; current = current->next) { 3429 if (UVM_ET_ISSUBMAP(current)) { 3430 vm_map_unlock_read(map); 3431 return EINVAL; 3432 } 3433 if (end <= current->end) { 3434 break; 3435 } 3436 if (current->end != current->next->start) { 3437 vm_map_unlock_read(map); 3438 return EFAULT; 3439 } 3440 } 3441 3442 error = 0; 3443 for (current = entry; start < end; current = current->next) { 3444 amap = current->aref.ar_amap; /* top layer */ 3445 uobj = current->object.uvm_obj; /* bottom layer */ 3446 KASSERT(start >= current->start); 3447 3448 /* 3449 * No amap cleaning necessary if: 3450 * 3451 * (1) There's no amap. 3452 * 3453 * (2) We're not deactivating or freeing pages. 3454 */ 3455 3456 if (amap == NULL || (flags & (PGO_DEACTIVATE|PGO_FREE)) == 0) 3457 goto flush_object; 3458 3459 amap_lock(amap); 3460 offset = start - current->start; 3461 size = MIN(end, current->end) - start; 3462 for ( ; size != 0; size -= PAGE_SIZE, offset += PAGE_SIZE) { 3463 anon = amap_lookup(¤t->aref, offset); 3464 if (anon == NULL) 3465 continue; 3466 3467 simple_lock(&anon->an_lock); 3468 pg = anon->u.an_page; 3469 if (pg == NULL) { 3470 simple_unlock(&anon->an_lock); 3471 continue; 3472 } 3473 3474 switch (flags & (PGO_CLEANIT|PGO_FREE|PGO_DEACTIVATE)) { 3475 3476 /* 3477 * In these first 3 cases, we just deactivate the page. 3478 */ 3479 3480 case PGO_CLEANIT|PGO_FREE: 3481 case PGO_CLEANIT|PGO_DEACTIVATE: 3482 case PGO_DEACTIVATE: 3483 deactivate_it: 3484 /* 3485 * skip the page if it's loaned or wired, 3486 * since it shouldn't be on a paging queue 3487 * at all in these cases. 3488 */ 3489 3490 uvm_lock_pageq(); 3491 if (pg->loan_count != 0 || 3492 pg->wire_count != 0) { 3493 uvm_unlock_pageq(); 3494 simple_unlock(&anon->an_lock); 3495 continue; 3496 } 3497 KASSERT(pg->uanon == anon); 3498 pmap_clear_reference(pg); 3499 uvm_pagedeactivate(pg); 3500 uvm_unlock_pageq(); 3501 simple_unlock(&anon->an_lock); 3502 continue; 3503 3504 case PGO_FREE: 3505 3506 /* 3507 * If there are multiple references to 3508 * the amap, just deactivate the page. 3509 */ 3510 3511 if (amap_refs(amap) > 1) 3512 goto deactivate_it; 3513 3514 /* skip the page if it's wired */ 3515 if (pg->wire_count != 0) { 3516 simple_unlock(&anon->an_lock); 3517 continue; 3518 } 3519 amap_unadd(¤t->aref, offset); 3520 refs = --anon->an_ref; 3521 simple_unlock(&anon->an_lock); 3522 if (refs == 0) 3523 uvm_anfree(anon); 3524 continue; 3525 } 3526 } 3527 amap_unlock(amap); 3528 3529 flush_object: 3530 /* 3531 * flush pages if we've got a valid backing object. 3532 * note that we must always clean object pages before 3533 * freeing them since otherwise we could reveal stale 3534 * data from files. 3535 */ 3536 3537 offset = current->offset + (start - current->start); 3538 size = MIN(end, current->end) - start; 3539 if (uobj != NULL) { 3540 simple_lock(&uobj->vmobjlock); 3541 if (uobj->pgops->pgo_put != NULL) 3542 error = (uobj->pgops->pgo_put)(uobj, offset, 3543 offset + size, flags | PGO_CLEANIT); 3544 else 3545 error = 0; 3546 } 3547 start += size; 3548 } 3549 vm_map_unlock_read(map); 3550 return (error); 3551 } 3552 3553 3554 /* 3555 * uvm_map_checkprot: check protection in map 3556 * 3557 * => must allow specified protection in a fully allocated region. 3558 * => map must be read or write locked by caller. 3559 */ 3560 3561 boolean_t 3562 uvm_map_checkprot(struct vm_map *map, vaddr_t start, vaddr_t end, 3563 vm_prot_t protection) 3564 { 3565 struct vm_map_entry *entry; 3566 struct vm_map_entry *tmp_entry; 3567 3568 if (!uvm_map_lookup_entry(map, start, &tmp_entry)) { 3569 return (FALSE); 3570 } 3571 entry = tmp_entry; 3572 while (start < end) { 3573 if (entry == &map->header) { 3574 return (FALSE); 3575 } 3576 3577 /* 3578 * no holes allowed 3579 */ 3580 3581 if (start < entry->start) { 3582 return (FALSE); 3583 } 3584 3585 /* 3586 * check protection associated with entry 3587 */ 3588 3589 if ((entry->protection & protection) != protection) { 3590 return (FALSE); 3591 } 3592 start = entry->end; 3593 entry = entry->next; 3594 } 3595 return (TRUE); 3596 } 3597 3598 /* 3599 * uvmspace_alloc: allocate a vmspace structure. 3600 * 3601 * - structure includes vm_map and pmap 3602 * - XXX: no locking on this structure 3603 * - refcnt set to 1, rest must be init'd by caller 3604 */ 3605 struct vmspace * 3606 uvmspace_alloc(vaddr_t min, vaddr_t max) 3607 { 3608 struct vmspace *vm; 3609 UVMHIST_FUNC("uvmspace_alloc"); UVMHIST_CALLED(maphist); 3610 3611 vm = pool_get(&uvm_vmspace_pool, PR_WAITOK); 3612 uvmspace_init(vm, NULL, min, max); 3613 UVMHIST_LOG(maphist,"<- done (vm=0x%x)", vm,0,0,0); 3614 return (vm); 3615 } 3616 3617 /* 3618 * uvmspace_init: initialize a vmspace structure. 3619 * 3620 * - XXX: no locking on this structure 3621 * - refcnt set to 1, rest must be init'd by caller 3622 */ 3623 void 3624 uvmspace_init(struct vmspace *vm, struct pmap *pmap, vaddr_t min, vaddr_t max) 3625 { 3626 UVMHIST_FUNC("uvmspace_init"); UVMHIST_CALLED(maphist); 3627 3628 memset(vm, 0, sizeof(*vm)); 3629 uvm_map_setup(&vm->vm_map, min, max, VM_MAP_PAGEABLE 3630 #ifdef __USING_TOPDOWN_VM 3631 | VM_MAP_TOPDOWN 3632 #endif 3633 ); 3634 if (pmap) 3635 pmap_reference(pmap); 3636 else 3637 pmap = pmap_create(); 3638 vm->vm_map.pmap = pmap; 3639 vm->vm_refcnt = 1; 3640 UVMHIST_LOG(maphist,"<- done",0,0,0,0); 3641 } 3642 3643 /* 3644 * uvmspace_share: share a vmspace between two processes 3645 * 3646 * - used for vfork, threads(?) 3647 */ 3648 3649 void 3650 uvmspace_share(struct proc *p1, struct proc *p2) 3651 { 3652 struct simplelock *slock = &p1->p_vmspace->vm_map.ref_lock; 3653 3654 p2->p_vmspace = p1->p_vmspace; 3655 simple_lock(slock); 3656 p1->p_vmspace->vm_refcnt++; 3657 simple_unlock(slock); 3658 } 3659 3660 /* 3661 * uvmspace_unshare: ensure that process "p" has its own, unshared, vmspace 3662 * 3663 * - XXX: no locking on vmspace 3664 */ 3665 3666 void 3667 uvmspace_unshare(struct lwp *l) 3668 { 3669 struct proc *p = l->l_proc; 3670 struct vmspace *nvm, *ovm = p->p_vmspace; 3671 3672 if (ovm->vm_refcnt == 1) 3673 /* nothing to do: vmspace isn't shared in the first place */ 3674 return; 3675 3676 /* make a new vmspace, still holding old one */ 3677 nvm = uvmspace_fork(ovm); 3678 3679 pmap_deactivate(l); /* unbind old vmspace */ 3680 p->p_vmspace = nvm; 3681 pmap_activate(l); /* switch to new vmspace */ 3682 3683 uvmspace_free(ovm); /* drop reference to old vmspace */ 3684 } 3685 3686 /* 3687 * uvmspace_exec: the process wants to exec a new program 3688 */ 3689 3690 void 3691 uvmspace_exec(struct lwp *l, vaddr_t start, vaddr_t end) 3692 { 3693 struct proc *p = l->l_proc; 3694 struct vmspace *nvm, *ovm = p->p_vmspace; 3695 struct vm_map *map = &ovm->vm_map; 3696 3697 #ifdef __sparc__ 3698 /* XXX cgd 960926: the sparc #ifdef should be a MD hook */ 3699 kill_user_windows(l); /* before stack addresses go away */ 3700 #endif 3701 3702 /* 3703 * see if more than one process is using this vmspace... 3704 */ 3705 3706 if (ovm->vm_refcnt == 1) { 3707 3708 /* 3709 * if p is the only process using its vmspace then we can safely 3710 * recycle that vmspace for the program that is being exec'd. 3711 */ 3712 3713 #ifdef SYSVSHM 3714 /* 3715 * SYSV SHM semantics require us to kill all segments on an exec 3716 */ 3717 3718 if (ovm->vm_shm) 3719 shmexit(ovm); 3720 #endif 3721 3722 /* 3723 * POSIX 1003.1b -- "lock future mappings" is revoked 3724 * when a process execs another program image. 3725 */ 3726 3727 vm_map_modflags(map, 0, VM_MAP_WIREFUTURE); 3728 3729 /* 3730 * now unmap the old program 3731 */ 3732 3733 pmap_remove_all(map->pmap); 3734 uvm_unmap(map, map->min_offset, map->max_offset); 3735 KASSERT(map->header.prev == &map->header); 3736 KASSERT(map->nentries == 0); 3737 3738 /* 3739 * resize the map 3740 */ 3741 3742 map->min_offset = start; 3743 map->max_offset = end; 3744 } else { 3745 3746 /* 3747 * p's vmspace is being shared, so we can't reuse it for p since 3748 * it is still being used for others. allocate a new vmspace 3749 * for p 3750 */ 3751 3752 nvm = uvmspace_alloc(start, end); 3753 3754 /* 3755 * install new vmspace and drop our ref to the old one. 3756 */ 3757 3758 pmap_deactivate(l); 3759 p->p_vmspace = nvm; 3760 pmap_activate(l); 3761 3762 uvmspace_free(ovm); 3763 } 3764 } 3765 3766 /* 3767 * uvmspace_free: free a vmspace data structure 3768 */ 3769 3770 void 3771 uvmspace_free(struct vmspace *vm) 3772 { 3773 struct vm_map_entry *dead_entries; 3774 struct vm_map *map = &vm->vm_map; 3775 int n; 3776 3777 UVMHIST_FUNC("uvmspace_free"); UVMHIST_CALLED(maphist); 3778 3779 UVMHIST_LOG(maphist,"(vm=0x%x) ref=%d", vm, vm->vm_refcnt,0,0); 3780 simple_lock(&map->ref_lock); 3781 n = --vm->vm_refcnt; 3782 simple_unlock(&map->ref_lock); 3783 if (n > 0) 3784 return; 3785 3786 /* 3787 * at this point, there should be no other references to the map. 3788 * delete all of the mappings, then destroy the pmap. 3789 */ 3790 3791 map->flags |= VM_MAP_DYING; 3792 pmap_remove_all(map->pmap); 3793 #ifdef SYSVSHM 3794 /* Get rid of any SYSV shared memory segments. */ 3795 if (vm->vm_shm != NULL) 3796 shmexit(vm); 3797 #endif 3798 if (map->nentries) { 3799 uvm_unmap_remove(map, map->min_offset, map->max_offset, 3800 &dead_entries, NULL); 3801 if (dead_entries != NULL) 3802 uvm_unmap_detach(dead_entries, 0); 3803 } 3804 KASSERT(map->nentries == 0); 3805 KASSERT(map->size == 0); 3806 pmap_destroy(map->pmap); 3807 pool_put(&uvm_vmspace_pool, vm); 3808 } 3809 3810 /* 3811 * F O R K - m a i n e n t r y p o i n t 3812 */ 3813 /* 3814 * uvmspace_fork: fork a process' main map 3815 * 3816 * => create a new vmspace for child process from parent. 3817 * => parent's map must not be locked. 3818 */ 3819 3820 struct vmspace * 3821 uvmspace_fork(struct vmspace *vm1) 3822 { 3823 struct vmspace *vm2; 3824 struct vm_map *old_map = &vm1->vm_map; 3825 struct vm_map *new_map; 3826 struct vm_map_entry *old_entry; 3827 struct vm_map_entry *new_entry; 3828 UVMHIST_FUNC("uvmspace_fork"); UVMHIST_CALLED(maphist); 3829 3830 vm_map_lock(old_map); 3831 3832 vm2 = uvmspace_alloc(old_map->min_offset, old_map->max_offset); 3833 memcpy(&vm2->vm_startcopy, &vm1->vm_startcopy, 3834 (caddr_t) (vm1 + 1) - (caddr_t) &vm1->vm_startcopy); 3835 new_map = &vm2->vm_map; /* XXX */ 3836 3837 old_entry = old_map->header.next; 3838 new_map->size = old_map->size; 3839 3840 /* 3841 * go entry-by-entry 3842 */ 3843 3844 while (old_entry != &old_map->header) { 3845 3846 /* 3847 * first, some sanity checks on the old entry 3848 */ 3849 3850 KASSERT(!UVM_ET_ISSUBMAP(old_entry)); 3851 KASSERT(UVM_ET_ISCOPYONWRITE(old_entry) || 3852 !UVM_ET_ISNEEDSCOPY(old_entry)); 3853 3854 switch (old_entry->inheritance) { 3855 case MAP_INHERIT_NONE: 3856 3857 /* 3858 * drop the mapping, modify size 3859 */ 3860 new_map->size -= old_entry->end - old_entry->start; 3861 break; 3862 3863 case MAP_INHERIT_SHARE: 3864 3865 /* 3866 * share the mapping: this means we want the old and 3867 * new entries to share amaps and backing objects. 3868 */ 3869 /* 3870 * if the old_entry needs a new amap (due to prev fork) 3871 * then we need to allocate it now so that we have 3872 * something we own to share with the new_entry. [in 3873 * other words, we need to clear needs_copy] 3874 */ 3875 3876 if (UVM_ET_ISNEEDSCOPY(old_entry)) { 3877 /* get our own amap, clears needs_copy */ 3878 amap_copy(old_map, old_entry, M_WAITOK, FALSE, 3879 0, 0); 3880 /* XXXCDC: WAITOK??? */ 3881 } 3882 3883 new_entry = uvm_mapent_alloc(new_map, 0); 3884 /* old_entry -> new_entry */ 3885 uvm_mapent_copy(old_entry, new_entry); 3886 3887 /* new pmap has nothing wired in it */ 3888 new_entry->wired_count = 0; 3889 3890 /* 3891 * gain reference to object backing the map (can't 3892 * be a submap, already checked this case). 3893 */ 3894 3895 if (new_entry->aref.ar_amap) 3896 uvm_map_reference_amap(new_entry, AMAP_SHARED); 3897 3898 if (new_entry->object.uvm_obj && 3899 new_entry->object.uvm_obj->pgops->pgo_reference) 3900 new_entry->object.uvm_obj-> 3901 pgops->pgo_reference( 3902 new_entry->object.uvm_obj); 3903 3904 /* insert entry at end of new_map's entry list */ 3905 uvm_map_entry_link(new_map, new_map->header.prev, 3906 new_entry); 3907 3908 break; 3909 3910 case MAP_INHERIT_COPY: 3911 3912 /* 3913 * copy-on-write the mapping (using mmap's 3914 * MAP_PRIVATE semantics) 3915 * 3916 * allocate new_entry, adjust reference counts. 3917 * (note that new references are read-only). 3918 */ 3919 3920 new_entry = uvm_mapent_alloc(new_map, 0); 3921 /* old_entry -> new_entry */ 3922 uvm_mapent_copy(old_entry, new_entry); 3923 3924 if (new_entry->aref.ar_amap) 3925 uvm_map_reference_amap(new_entry, 0); 3926 3927 if (new_entry->object.uvm_obj && 3928 new_entry->object.uvm_obj->pgops->pgo_reference) 3929 new_entry->object.uvm_obj->pgops->pgo_reference 3930 (new_entry->object.uvm_obj); 3931 3932 /* new pmap has nothing wired in it */ 3933 new_entry->wired_count = 0; 3934 3935 new_entry->etype |= 3936 (UVM_ET_COPYONWRITE|UVM_ET_NEEDSCOPY); 3937 uvm_map_entry_link(new_map, new_map->header.prev, 3938 new_entry); 3939 3940 /* 3941 * the new entry will need an amap. it will either 3942 * need to be copied from the old entry or created 3943 * from scratch (if the old entry does not have an 3944 * amap). can we defer this process until later 3945 * (by setting "needs_copy") or do we need to copy 3946 * the amap now? 3947 * 3948 * we must copy the amap now if any of the following 3949 * conditions hold: 3950 * 1. the old entry has an amap and that amap is 3951 * being shared. this means that the old (parent) 3952 * process is sharing the amap with another 3953 * process. if we do not clear needs_copy here 3954 * we will end up in a situation where both the 3955 * parent and child process are refering to the 3956 * same amap with "needs_copy" set. if the 3957 * parent write-faults, the fault routine will 3958 * clear "needs_copy" in the parent by allocating 3959 * a new amap. this is wrong because the 3960 * parent is supposed to be sharing the old amap 3961 * and the new amap will break that. 3962 * 3963 * 2. if the old entry has an amap and a non-zero 3964 * wire count then we are going to have to call 3965 * amap_cow_now to avoid page faults in the 3966 * parent process. since amap_cow_now requires 3967 * "needs_copy" to be clear we might as well 3968 * clear it here as well. 3969 * 3970 */ 3971 3972 if (old_entry->aref.ar_amap != NULL) { 3973 if ((amap_flags(old_entry->aref.ar_amap) & 3974 AMAP_SHARED) != 0 || 3975 VM_MAPENT_ISWIRED(old_entry)) { 3976 3977 amap_copy(new_map, new_entry, M_WAITOK, 3978 FALSE, 0, 0); 3979 /* XXXCDC: M_WAITOK ... ok? */ 3980 } 3981 } 3982 3983 /* 3984 * if the parent's entry is wired down, then the 3985 * parent process does not want page faults on 3986 * access to that memory. this means that we 3987 * cannot do copy-on-write because we can't write 3988 * protect the old entry. in this case we 3989 * resolve all copy-on-write faults now, using 3990 * amap_cow_now. note that we have already 3991 * allocated any needed amap (above). 3992 */ 3993 3994 if (VM_MAPENT_ISWIRED(old_entry)) { 3995 3996 /* 3997 * resolve all copy-on-write faults now 3998 * (note that there is nothing to do if 3999 * the old mapping does not have an amap). 4000 */ 4001 if (old_entry->aref.ar_amap) 4002 amap_cow_now(new_map, new_entry); 4003 4004 } else { 4005 4006 /* 4007 * setup mappings to trigger copy-on-write faults 4008 * we must write-protect the parent if it has 4009 * an amap and it is not already "needs_copy"... 4010 * if it is already "needs_copy" then the parent 4011 * has already been write-protected by a previous 4012 * fork operation. 4013 */ 4014 4015 if (old_entry->aref.ar_amap && 4016 !UVM_ET_ISNEEDSCOPY(old_entry)) { 4017 if (old_entry->max_protection & VM_PROT_WRITE) { 4018 pmap_protect(old_map->pmap, 4019 old_entry->start, 4020 old_entry->end, 4021 old_entry->protection & 4022 ~VM_PROT_WRITE); 4023 pmap_update(old_map->pmap); 4024 } 4025 old_entry->etype |= UVM_ET_NEEDSCOPY; 4026 } 4027 } 4028 break; 4029 } /* end of switch statement */ 4030 old_entry = old_entry->next; 4031 } 4032 4033 vm_map_unlock(old_map); 4034 4035 #ifdef SYSVSHM 4036 if (vm1->vm_shm) 4037 shmfork(vm1, vm2); 4038 #endif 4039 4040 #ifdef PMAP_FORK 4041 pmap_fork(vm1->vm_map.pmap, vm2->vm_map.pmap); 4042 #endif 4043 4044 UVMHIST_LOG(maphist,"<- done",0,0,0,0); 4045 return (vm2); 4046 } 4047 4048 4049 /* 4050 * in-kernel map entry allocation. 4051 */ 4052 4053 int ukh_alloc, ukh_free; 4054 int uke_alloc, uke_free; 4055 4056 struct uvm_kmapent_hdr { 4057 LIST_ENTRY(uvm_kmapent_hdr) ukh_listq; 4058 int ukh_nused; 4059 struct vm_map_entry *ukh_freelist; 4060 struct vm_map *ukh_map; 4061 struct vm_map_entry ukh_entries[0]; 4062 }; 4063 4064 #define UVM_KMAPENT_CHUNK \ 4065 ((PAGE_SIZE - sizeof(struct uvm_kmapent_hdr)) \ 4066 / sizeof(struct vm_map_entry)) 4067 4068 #define UVM_KHDR_FIND(entry) \ 4069 ((struct uvm_kmapent_hdr *)(((vaddr_t)entry) & ~PAGE_MASK)) 4070 4071 static __inline struct vm_map_entry *uvm_kmapent_get(struct uvm_kmapent_hdr *); 4072 static __inline void uvm_kmapent_put(struct uvm_kmapent_hdr *, 4073 struct vm_map_entry *); 4074 4075 static __inline struct vm_map * 4076 uvm_kmapent_map(struct vm_map_entry *entry) 4077 { 4078 const struct uvm_kmapent_hdr *ukh; 4079 4080 ukh = UVM_KHDR_FIND(entry); 4081 return ukh->ukh_map; 4082 } 4083 4084 static __inline struct vm_map_entry * 4085 uvm_kmapent_get(struct uvm_kmapent_hdr *ukh) 4086 { 4087 struct vm_map_entry *entry; 4088 4089 KASSERT(ukh->ukh_nused <= UVM_KMAPENT_CHUNK); 4090 KASSERT(ukh->ukh_nused >= 0); 4091 4092 entry = ukh->ukh_freelist; 4093 if (entry) { 4094 KASSERT((entry->flags & (UVM_MAP_KERNEL | UVM_MAP_KMAPENT)) 4095 == UVM_MAP_KERNEL); 4096 ukh->ukh_freelist = entry->next; 4097 ukh->ukh_nused++; 4098 KASSERT(ukh->ukh_nused <= UVM_KMAPENT_CHUNK); 4099 } else { 4100 KASSERT(ukh->ukh_nused == UVM_KMAPENT_CHUNK); 4101 } 4102 4103 return entry; 4104 } 4105 4106 static __inline void 4107 uvm_kmapent_put(struct uvm_kmapent_hdr *ukh, struct vm_map_entry *entry) 4108 { 4109 4110 KASSERT((entry->flags & (UVM_MAP_KERNEL | UVM_MAP_KMAPENT)) 4111 == UVM_MAP_KERNEL); 4112 KASSERT(ukh->ukh_nused <= UVM_KMAPENT_CHUNK); 4113 KASSERT(ukh->ukh_nused > 0); 4114 KASSERT(ukh->ukh_freelist != NULL || 4115 ukh->ukh_nused == UVM_KMAPENT_CHUNK); 4116 KASSERT(ukh->ukh_freelist == NULL || 4117 ukh->ukh_nused < UVM_KMAPENT_CHUNK); 4118 4119 ukh->ukh_nused--; 4120 entry->next = ukh->ukh_freelist; 4121 ukh->ukh_freelist = entry; 4122 } 4123 4124 /* 4125 * uvm_kmapent_alloc: allocate a map entry for in-kernel map 4126 */ 4127 4128 static struct vm_map_entry * 4129 uvm_kmapent_alloc(struct vm_map *map, int flags) 4130 { 4131 struct vm_page *pg; 4132 struct uvm_map_args args; 4133 struct uvm_kmapent_hdr *ukh; 4134 struct vm_map_entry *entry; 4135 uvm_flag_t mapflags = UVM_MAPFLAG(UVM_PROT_ALL, UVM_PROT_ALL, 4136 UVM_INH_NONE, UVM_ADV_RANDOM, flags | UVM_FLAG_NOMERGE); 4137 vaddr_t va; 4138 int error; 4139 int i; 4140 int s; 4141 4142 KDASSERT(UVM_KMAPENT_CHUNK > 2); 4143 KDASSERT(kernel_map != NULL); 4144 KASSERT(vm_map_pmap(map) == pmap_kernel()); 4145 4146 uke_alloc++; 4147 entry = NULL; 4148 again: 4149 /* 4150 * try to grab an entry from freelist. 4151 */ 4152 s = splvm(); 4153 simple_lock(&uvm.kentry_lock); 4154 ukh = LIST_FIRST(&vm_map_to_kernel(map)->vmk_kentry_free); 4155 if (ukh) { 4156 entry = uvm_kmapent_get(ukh); 4157 if (ukh->ukh_nused == UVM_KMAPENT_CHUNK) 4158 LIST_REMOVE(ukh, ukh_listq); 4159 } 4160 simple_unlock(&uvm.kentry_lock); 4161 splx(s); 4162 4163 if (entry) 4164 return entry; 4165 4166 /* 4167 * there's no free entry for this vm_map. 4168 * now we need to allocate some vm_map_entry. 4169 * for simplicity, always allocate one page chunk of them at once. 4170 */ 4171 4172 pg = uvm_pagealloc(NULL, 0, NULL, 0); 4173 if (__predict_false(pg == NULL)) { 4174 if (flags & UVM_FLAG_NOWAIT) 4175 return NULL; 4176 uvm_wait("kme_alloc"); 4177 goto again; 4178 } 4179 4180 error = uvm_map_prepare(map, 0, PAGE_SIZE, NULL, 0, 0, mapflags, &args); 4181 if (error) { 4182 uvm_pagefree(pg); 4183 return NULL; 4184 } 4185 4186 va = args.uma_start; 4187 4188 pmap_kenter_pa(va, VM_PAGE_TO_PHYS(pg), VM_PROT_READ|VM_PROT_WRITE); 4189 pmap_update(vm_map_pmap(map)); 4190 4191 ukh = (void *)va; 4192 4193 /* 4194 * use the first entry for ukh itsself. 4195 */ 4196 4197 entry = &ukh->ukh_entries[0]; 4198 entry->flags = UVM_MAP_KERNEL | UVM_MAP_KMAPENT; 4199 error = uvm_map_enter(map, &args, entry); 4200 KASSERT(error == 0); 4201 4202 ukh->ukh_nused = UVM_KMAPENT_CHUNK; 4203 ukh->ukh_map = map; 4204 ukh->ukh_freelist = NULL; 4205 for (i = UVM_KMAPENT_CHUNK - 1; i >= 2; i--) { 4206 struct vm_map_entry *entry = &ukh->ukh_entries[i]; 4207 4208 entry->flags = UVM_MAP_KERNEL; 4209 uvm_kmapent_put(ukh, entry); 4210 } 4211 KASSERT(ukh->ukh_nused == 2); 4212 4213 s = splvm(); 4214 simple_lock(&uvm.kentry_lock); 4215 LIST_INSERT_HEAD(&vm_map_to_kernel(map)->vmk_kentry_free, 4216 ukh, ukh_listq); 4217 simple_unlock(&uvm.kentry_lock); 4218 splx(s); 4219 4220 /* 4221 * return second entry. 4222 */ 4223 4224 entry = &ukh->ukh_entries[1]; 4225 entry->flags = UVM_MAP_KERNEL; 4226 ukh_alloc++; 4227 return entry; 4228 } 4229 4230 /* 4231 * uvm_mapent_free: free map entry for in-kernel map 4232 */ 4233 4234 static void 4235 uvm_kmapent_free(struct vm_map_entry *entry) 4236 { 4237 struct uvm_kmapent_hdr *ukh; 4238 struct vm_page *pg; 4239 struct vm_map *map; 4240 struct pmap *pmap; 4241 vaddr_t va; 4242 paddr_t pa; 4243 struct vm_map_entry *deadentry; 4244 int s; 4245 4246 uke_free++; 4247 ukh = UVM_KHDR_FIND(entry); 4248 map = ukh->ukh_map; 4249 4250 s = splvm(); 4251 simple_lock(&uvm.kentry_lock); 4252 uvm_kmapent_put(ukh, entry); 4253 if (ukh->ukh_nused > 1) { 4254 if (ukh->ukh_nused == UVM_KMAPENT_CHUNK - 1) 4255 LIST_INSERT_HEAD( 4256 &vm_map_to_kernel(map)->vmk_kentry_free, 4257 ukh, ukh_listq); 4258 simple_unlock(&uvm.kentry_lock); 4259 splx(s); 4260 return; 4261 } 4262 4263 /* 4264 * now we can free this ukh. 4265 * 4266 * however, keep an empty ukh to avoid ping-pong. 4267 */ 4268 4269 if (LIST_FIRST(&vm_map_to_kernel(map)->vmk_kentry_free) == ukh && 4270 LIST_NEXT(ukh, ukh_listq) == NULL) { 4271 simple_unlock(&uvm.kentry_lock); 4272 splx(s); 4273 return; 4274 } 4275 LIST_REMOVE(ukh, ukh_listq); 4276 simple_unlock(&uvm.kentry_lock); 4277 splx(s); 4278 4279 KASSERT(ukh->ukh_nused == 1); 4280 4281 /* 4282 * remove map entry for ukh itsself. 4283 */ 4284 4285 va = (vaddr_t)ukh; 4286 KASSERT((va & PAGE_MASK) == 0); 4287 uvm_unmap_remove(map, va, va + PAGE_SIZE, &deadentry, NULL); 4288 KASSERT(deadentry->flags & UVM_MAP_KERNEL); 4289 KASSERT(deadentry->flags & UVM_MAP_KMAPENT); 4290 KASSERT(deadentry->next == NULL); 4291 KASSERT(deadentry == &ukh->ukh_entries[0]); 4292 4293 /* 4294 * unmap the page from pmap and free it. 4295 */ 4296 4297 pmap = vm_map_pmap(map); 4298 KASSERT(pmap == pmap_kernel()); 4299 if (!pmap_extract(pmap, va, &pa)) 4300 panic("%s: no mapping", __func__); 4301 pmap_kremove(va, PAGE_SIZE); 4302 pg = PHYS_TO_VM_PAGE(pa); 4303 uvm_pagefree(pg); 4304 ukh_free++; 4305 } 4306 4307 /* 4308 * map entry reservation 4309 */ 4310 4311 /* 4312 * uvm_mapent_reserve: reserve map entries for clipping before locking map. 4313 * 4314 * => needed when unmapping entries allocated without UVM_FLAG_QUANTUM. 4315 * => caller shouldn't hold map locked. 4316 */ 4317 int 4318 uvm_mapent_reserve(struct vm_map *map, struct uvm_mapent_reservation *umr, 4319 int nentries, int flags) 4320 { 4321 4322 umr->umr_nentries = 0; 4323 4324 if ((flags & UVM_FLAG_QUANTUM) != 0) 4325 return 0; 4326 4327 if (!VM_MAP_USE_KMAPENT(map)) 4328 return 0; 4329 4330 while (nentries--) { 4331 struct vm_map_entry *ent; 4332 ent = uvm_kmapent_alloc(map, flags); 4333 if (!ent) { 4334 uvm_mapent_unreserve(map, umr); 4335 return ENOMEM; 4336 } 4337 UMR_PUTENTRY(umr, ent); 4338 } 4339 4340 return 0; 4341 } 4342 4343 /* 4344 * uvm_mapent_unreserve: 4345 * 4346 * => caller shouldn't hold map locked. 4347 * => never fail or sleep. 4348 */ 4349 void 4350 uvm_mapent_unreserve(struct vm_map *map, struct uvm_mapent_reservation *umr) 4351 { 4352 4353 while (!UMR_EMPTY(umr)) 4354 uvm_kmapent_free(UMR_GETENTRY(umr)); 4355 } 4356 4357 #if defined(DDB) 4358 4359 /* 4360 * DDB hooks 4361 */ 4362 4363 /* 4364 * uvm_map_printit: actually prints the map 4365 */ 4366 4367 void 4368 uvm_map_printit(struct vm_map *map, boolean_t full, 4369 void (*pr)(const char *, ...)) 4370 { 4371 struct vm_map_entry *entry; 4372 4373 (*pr)("MAP %p: [0x%lx->0x%lx]\n", map, map->min_offset,map->max_offset); 4374 (*pr)("\t#ent=%d, sz=%d, ref=%d, version=%d, flags=0x%x\n", 4375 map->nentries, map->size, map->ref_count, map->timestamp, 4376 map->flags); 4377 (*pr)("\tpmap=%p(resident=%ld, wired=%ld)\n", map->pmap, 4378 pmap_resident_count(map->pmap), pmap_wired_count(map->pmap)); 4379 if (!full) 4380 return; 4381 for (entry = map->header.next; entry != &map->header; 4382 entry = entry->next) { 4383 (*pr)(" - %p: 0x%lx->0x%lx: obj=%p/0x%llx, amap=%p/%d\n", 4384 entry, entry->start, entry->end, entry->object.uvm_obj, 4385 (long long)entry->offset, entry->aref.ar_amap, 4386 entry->aref.ar_pageoff); 4387 (*pr)( 4388 "\tsubmap=%c, cow=%c, nc=%c, prot(max)=%d/%d, inh=%d, " 4389 "wc=%d, adv=%d\n", 4390 (entry->etype & UVM_ET_SUBMAP) ? 'T' : 'F', 4391 (entry->etype & UVM_ET_COPYONWRITE) ? 'T' : 'F', 4392 (entry->etype & UVM_ET_NEEDSCOPY) ? 'T' : 'F', 4393 entry->protection, entry->max_protection, 4394 entry->inheritance, entry->wired_count, entry->advice); 4395 } 4396 } 4397 4398 /* 4399 * uvm_object_printit: actually prints the object 4400 */ 4401 4402 void 4403 uvm_object_printit(struct uvm_object *uobj, boolean_t full, 4404 void (*pr)(const char *, ...)) 4405 { 4406 struct vm_page *pg; 4407 int cnt = 0; 4408 4409 (*pr)("OBJECT %p: locked=%d, pgops=%p, npages=%d, ", 4410 uobj, uobj->vmobjlock.lock_data, uobj->pgops, uobj->uo_npages); 4411 if (UVM_OBJ_IS_KERN_OBJECT(uobj)) 4412 (*pr)("refs=<SYSTEM>\n"); 4413 else 4414 (*pr)("refs=%d\n", uobj->uo_refs); 4415 4416 if (!full) { 4417 return; 4418 } 4419 (*pr)(" PAGES <pg,offset>:\n "); 4420 TAILQ_FOREACH(pg, &uobj->memq, listq) { 4421 cnt++; 4422 (*pr)("<%p,0x%llx> ", pg, (long long)pg->offset); 4423 if ((cnt % 3) == 0) { 4424 (*pr)("\n "); 4425 } 4426 } 4427 if ((cnt % 3) != 0) { 4428 (*pr)("\n"); 4429 } 4430 } 4431 4432 /* 4433 * uvm_page_printit: actually print the page 4434 */ 4435 4436 static const char page_flagbits[] = 4437 "\20\1BUSY\2WANTED\3TABLED\4CLEAN\5PAGEOUT\6RELEASED\7FAKE\10RDONLY" 4438 "\11ZERO\15PAGER1"; 4439 static const char page_pqflagbits[] = 4440 "\20\1FREE\2INACTIVE\3ACTIVE\5ANON\6AOBJ"; 4441 4442 void 4443 uvm_page_printit(struct vm_page *pg, boolean_t full, 4444 void (*pr)(const char *, ...)) 4445 { 4446 struct vm_page *tpg; 4447 struct uvm_object *uobj; 4448 struct pglist *pgl; 4449 char pgbuf[128]; 4450 char pqbuf[128]; 4451 4452 (*pr)("PAGE %p:\n", pg); 4453 bitmask_snprintf(pg->flags, page_flagbits, pgbuf, sizeof(pgbuf)); 4454 bitmask_snprintf(pg->pqflags, page_pqflagbits, pqbuf, sizeof(pqbuf)); 4455 (*pr)(" flags=%s, pqflags=%s, wire_count=%d, pa=0x%lx\n", 4456 pgbuf, pqbuf, pg->wire_count, (long)VM_PAGE_TO_PHYS(pg)); 4457 (*pr)(" uobject=%p, uanon=%p, offset=0x%llx loan_count=%d\n", 4458 pg->uobject, pg->uanon, (long long)pg->offset, pg->loan_count); 4459 #if defined(UVM_PAGE_TRKOWN) 4460 if (pg->flags & PG_BUSY) 4461 (*pr)(" owning process = %d, tag=%s\n", 4462 pg->owner, pg->owner_tag); 4463 else 4464 (*pr)(" page not busy, no owner\n"); 4465 #else 4466 (*pr)(" [page ownership tracking disabled]\n"); 4467 #endif 4468 4469 if (!full) 4470 return; 4471 4472 /* cross-verify object/anon */ 4473 if ((pg->pqflags & PQ_FREE) == 0) { 4474 if (pg->pqflags & PQ_ANON) { 4475 if (pg->uanon == NULL || pg->uanon->u.an_page != pg) 4476 (*pr)(" >>> ANON DOES NOT POINT HERE <<< (%p)\n", 4477 (pg->uanon) ? pg->uanon->u.an_page : NULL); 4478 else 4479 (*pr)(" anon backpointer is OK\n"); 4480 } else { 4481 uobj = pg->uobject; 4482 if (uobj) { 4483 (*pr)(" checking object list\n"); 4484 TAILQ_FOREACH(tpg, &uobj->memq, listq) { 4485 if (tpg == pg) { 4486 break; 4487 } 4488 } 4489 if (tpg) 4490 (*pr)(" page found on object list\n"); 4491 else 4492 (*pr)(" >>> PAGE NOT FOUND ON OBJECT LIST! <<<\n"); 4493 } 4494 } 4495 } 4496 4497 /* cross-verify page queue */ 4498 if (pg->pqflags & PQ_FREE) { 4499 int fl = uvm_page_lookup_freelist(pg); 4500 int color = VM_PGCOLOR_BUCKET(pg); 4501 pgl = &uvm.page_free[fl].pgfl_buckets[color].pgfl_queues[ 4502 ((pg)->flags & PG_ZERO) ? PGFL_ZEROS : PGFL_UNKNOWN]; 4503 } else if (pg->pqflags & PQ_INACTIVE) { 4504 pgl = &uvm.page_inactive; 4505 } else if (pg->pqflags & PQ_ACTIVE) { 4506 pgl = &uvm.page_active; 4507 } else { 4508 pgl = NULL; 4509 } 4510 4511 if (pgl) { 4512 (*pr)(" checking pageq list\n"); 4513 TAILQ_FOREACH(tpg, pgl, pageq) { 4514 if (tpg == pg) { 4515 break; 4516 } 4517 } 4518 if (tpg) 4519 (*pr)(" page found on pageq list\n"); 4520 else 4521 (*pr)(" >>> PAGE NOT FOUND ON PAGEQ LIST! <<<\n"); 4522 } 4523 } 4524 #endif 4525