1 /* $NetBSD: uvm_map.c,v 1.107 2001/09/21 07:57:35 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 #include "opt_ddb.h" 70 #include "opt_uvmhist.h" 71 #include "opt_sysv.h" 72 73 /* 74 * uvm_map.c: uvm map operations 75 */ 76 77 #include <sys/param.h> 78 #include <sys/systm.h> 79 #include <sys/mman.h> 80 #include <sys/proc.h> 81 #include <sys/malloc.h> 82 #include <sys/pool.h> 83 #include <sys/kernel.h> 84 85 #ifdef SYSVSHM 86 #include <sys/shm.h> 87 #endif 88 89 #define UVM_MAP 90 #include <uvm/uvm.h> 91 92 #ifdef DDB 93 #include <uvm/uvm_ddb.h> 94 #endif 95 96 extern struct vm_map *pager_map; 97 98 struct uvm_cnt uvm_map_call, map_backmerge, map_forwmerge; 99 struct uvm_cnt uvm_mlk_call, uvm_mlk_hint; 100 const char vmmapbsy[] = "vmmapbsy"; 101 102 /* 103 * pool for vmspace structures. 104 */ 105 106 struct pool uvm_vmspace_pool; 107 108 /* 109 * pool for dynamically-allocated map entries. 110 */ 111 112 struct pool uvm_map_entry_pool; 113 struct pool uvm_map_entry_kmem_pool; 114 115 #ifdef PMAP_GROWKERNEL 116 /* 117 * This global represents the end of the kernel virtual address 118 * space. If we want to exceed this, we must grow the kernel 119 * virtual address space dynamically. 120 * 121 * Note, this variable is locked by kernel_map's lock. 122 */ 123 vaddr_t uvm_maxkaddr; 124 #endif 125 126 /* 127 * macros 128 */ 129 130 /* 131 * uvm_map_entry_link: insert entry into a map 132 * 133 * => map must be locked 134 */ 135 #define uvm_map_entry_link(map, after_where, entry) do { \ 136 (map)->nentries++; \ 137 (entry)->prev = (after_where); \ 138 (entry)->next = (after_where)->next; \ 139 (entry)->prev->next = (entry); \ 140 (entry)->next->prev = (entry); \ 141 } while (0) 142 143 /* 144 * uvm_map_entry_unlink: remove entry from a map 145 * 146 * => map must be locked 147 */ 148 #define uvm_map_entry_unlink(map, entry) do { \ 149 (map)->nentries--; \ 150 (entry)->next->prev = (entry)->prev; \ 151 (entry)->prev->next = (entry)->next; \ 152 } while (0) 153 154 /* 155 * SAVE_HINT: saves the specified entry as the hint for future lookups. 156 * 157 * => map need not be locked (protected by hint_lock). 158 */ 159 #define SAVE_HINT(map,check,value) do { \ 160 simple_lock(&(map)->hint_lock); \ 161 if ((map)->hint == (check)) \ 162 (map)->hint = (value); \ 163 simple_unlock(&(map)->hint_lock); \ 164 } while (0) 165 166 /* 167 * VM_MAP_RANGE_CHECK: check and correct range 168 * 169 * => map must at least be read locked 170 */ 171 172 #define VM_MAP_RANGE_CHECK(map, start, end) do { \ 173 if (start < vm_map_min(map)) \ 174 start = vm_map_min(map); \ 175 if (end > vm_map_max(map)) \ 176 end = vm_map_max(map); \ 177 if (start > end) \ 178 start = end; \ 179 } while (0) 180 181 /* 182 * local prototypes 183 */ 184 185 static struct vm_map_entry *uvm_mapent_alloc __P((struct vm_map *)); 186 static void uvm_mapent_copy __P((struct vm_map_entry *, struct vm_map_entry *)); 187 static void uvm_mapent_free __P((struct vm_map_entry *)); 188 static void uvm_map_entry_unwire __P((struct vm_map *, struct vm_map_entry *)); 189 static void uvm_map_reference_amap __P((struct vm_map_entry *, int)); 190 static void uvm_map_unreference_amap __P((struct vm_map_entry *, int)); 191 192 /* 193 * local inlines 194 */ 195 196 /* 197 * uvm_mapent_alloc: allocate a map entry 198 */ 199 200 static __inline struct vm_map_entry * 201 uvm_mapent_alloc(map) 202 struct vm_map *map; 203 { 204 struct vm_map_entry *me; 205 int s; 206 UVMHIST_FUNC("uvm_mapent_alloc"); UVMHIST_CALLED(maphist); 207 208 if (map->flags & VM_MAP_INTRSAFE || cold) { 209 s = splvm(); 210 simple_lock(&uvm.kentry_lock); 211 me = uvm.kentry_free; 212 if (me) uvm.kentry_free = me->next; 213 simple_unlock(&uvm.kentry_lock); 214 splx(s); 215 if (me == NULL) { 216 panic("uvm_mapent_alloc: out of static map entries, " 217 "check MAX_KMAPENT (currently %d)", 218 MAX_KMAPENT); 219 } 220 me->flags = UVM_MAP_STATIC; 221 } else if (map == kernel_map) { 222 me = pool_get(&uvm_map_entry_kmem_pool, PR_WAITOK); 223 me->flags = UVM_MAP_KMEM; 224 } else { 225 me = pool_get(&uvm_map_entry_pool, PR_WAITOK); 226 me->flags = 0; 227 } 228 229 UVMHIST_LOG(maphist, "<- new entry=0x%x [kentry=%d]", me, 230 ((map->flags & VM_MAP_INTRSAFE) != 0 || map == kernel_map), 0, 0); 231 return(me); 232 } 233 234 /* 235 * uvm_mapent_free: free map entry 236 */ 237 238 static __inline void 239 uvm_mapent_free(me) 240 struct vm_map_entry *me; 241 { 242 int s; 243 UVMHIST_FUNC("uvm_mapent_free"); UVMHIST_CALLED(maphist); 244 245 UVMHIST_LOG(maphist,"<- freeing map entry=0x%x [flags=%d]", 246 me, me->flags, 0, 0); 247 if (me->flags & UVM_MAP_STATIC) { 248 s = splvm(); 249 simple_lock(&uvm.kentry_lock); 250 me->next = uvm.kentry_free; 251 uvm.kentry_free = me; 252 simple_unlock(&uvm.kentry_lock); 253 splx(s); 254 } else if (me->flags & UVM_MAP_KMEM) { 255 pool_put(&uvm_map_entry_kmem_pool, me); 256 } else { 257 pool_put(&uvm_map_entry_pool, me); 258 } 259 } 260 261 /* 262 * uvm_mapent_copy: copy a map entry, preserving flags 263 */ 264 265 static __inline void 266 uvm_mapent_copy(src, dst) 267 struct vm_map_entry *src; 268 struct vm_map_entry *dst; 269 { 270 memcpy(dst, src, ((char *)&src->uvm_map_entry_stop_copy) - 271 ((char *)src)); 272 } 273 274 /* 275 * uvm_map_entry_unwire: unwire a map entry 276 * 277 * => map should be locked by caller 278 */ 279 280 static __inline void 281 uvm_map_entry_unwire(map, entry) 282 struct vm_map *map; 283 struct vm_map_entry *entry; 284 { 285 entry->wired_count = 0; 286 uvm_fault_unwire_locked(map, entry->start, entry->end); 287 } 288 289 290 /* 291 * wrapper for calling amap_ref() 292 */ 293 static __inline void 294 uvm_map_reference_amap(entry, flags) 295 struct vm_map_entry *entry; 296 int flags; 297 { 298 amap_ref(entry->aref.ar_amap, entry->aref.ar_pageoff, 299 (entry->end - entry->start) >> PAGE_SHIFT, flags); 300 } 301 302 303 /* 304 * wrapper for calling amap_unref() 305 */ 306 static __inline void 307 uvm_map_unreference_amap(entry, flags) 308 struct vm_map_entry *entry; 309 int flags; 310 { 311 amap_unref(entry->aref.ar_amap, entry->aref.ar_pageoff, 312 (entry->end - entry->start) >> PAGE_SHIFT, flags); 313 } 314 315 316 /* 317 * uvm_map_init: init mapping system at boot time. note that we allocate 318 * and init the static pool of struct vm_map_entry *'s for the kernel here. 319 */ 320 321 void 322 uvm_map_init() 323 { 324 static struct vm_map_entry kernel_map_entry[MAX_KMAPENT]; 325 #if defined(UVMHIST) 326 static struct uvm_history_ent maphistbuf[100]; 327 static struct uvm_history_ent pdhistbuf[100]; 328 #endif 329 int lcv; 330 331 /* 332 * first, init logging system. 333 */ 334 335 UVMHIST_FUNC("uvm_map_init"); 336 UVMHIST_INIT_STATIC(maphist, maphistbuf); 337 UVMHIST_INIT_STATIC(pdhist, pdhistbuf); 338 UVMHIST_CALLED(maphist); 339 UVMHIST_LOG(maphist,"<starting uvm map system>", 0, 0, 0, 0); 340 UVMCNT_INIT(uvm_map_call, UVMCNT_CNT, 0, 341 "# uvm_map() successful calls", 0); 342 UVMCNT_INIT(map_backmerge, UVMCNT_CNT, 0, "# uvm_map() back merges", 0); 343 UVMCNT_INIT(map_forwmerge, UVMCNT_CNT, 0, "# uvm_map() missed forward", 344 0); 345 UVMCNT_INIT(uvm_mlk_call, UVMCNT_CNT, 0, "# map lookup calls", 0); 346 UVMCNT_INIT(uvm_mlk_hint, UVMCNT_CNT, 0, "# map lookup hint hits", 0); 347 348 /* 349 * now set up static pool of kernel map entrys ... 350 */ 351 352 simple_lock_init(&uvm.kentry_lock); 353 uvm.kentry_free = NULL; 354 for (lcv = 0 ; lcv < MAX_KMAPENT ; lcv++) { 355 kernel_map_entry[lcv].next = uvm.kentry_free; 356 uvm.kentry_free = &kernel_map_entry[lcv]; 357 } 358 359 /* 360 * initialize the map-related pools. 361 */ 362 pool_init(&uvm_vmspace_pool, sizeof(struct vmspace), 363 0, 0, 0, "vmsppl", 0, 364 pool_page_alloc_nointr, pool_page_free_nointr, M_VMMAP); 365 pool_init(&uvm_map_entry_pool, sizeof(struct vm_map_entry), 366 0, 0, 0, "vmmpepl", 0, 367 pool_page_alloc_nointr, pool_page_free_nointr, M_VMMAP); 368 pool_init(&uvm_map_entry_kmem_pool, sizeof(struct vm_map_entry), 369 0, 0, 0, "vmmpekpl", 0, NULL, NULL, M_VMMAP); 370 } 371 372 /* 373 * clippers 374 */ 375 376 /* 377 * uvm_map_clip_start: ensure that the entry begins at or after 378 * the starting address, if it doesn't we split the entry. 379 * 380 * => caller should use UVM_MAP_CLIP_START macro rather than calling 381 * this directly 382 * => map must be locked by caller 383 */ 384 385 void 386 uvm_map_clip_start(map, entry, start) 387 struct vm_map *map; 388 struct vm_map_entry *entry; 389 vaddr_t start; 390 { 391 struct vm_map_entry *new_entry; 392 vaddr_t new_adj; 393 394 /* uvm_map_simplify_entry(map, entry); */ /* XXX */ 395 396 /* 397 * Split off the front portion. note that we must insert the new 398 * entry BEFORE this one, so that this entry has the specified 399 * starting address. 400 */ 401 402 new_entry = uvm_mapent_alloc(map); 403 uvm_mapent_copy(entry, new_entry); /* entry -> new_entry */ 404 405 new_entry->end = start; 406 new_adj = start - new_entry->start; 407 if (entry->object.uvm_obj) 408 entry->offset += new_adj; /* shift start over */ 409 entry->start = start; 410 411 if (new_entry->aref.ar_amap) { 412 amap_splitref(&new_entry->aref, &entry->aref, new_adj); 413 } 414 415 uvm_map_entry_link(map, entry->prev, new_entry); 416 417 if (UVM_ET_ISSUBMAP(entry)) { 418 /* ... unlikely to happen, but play it safe */ 419 uvm_map_reference(new_entry->object.sub_map); 420 } else { 421 if (UVM_ET_ISOBJ(entry) && 422 entry->object.uvm_obj->pgops && 423 entry->object.uvm_obj->pgops->pgo_reference) 424 entry->object.uvm_obj->pgops->pgo_reference( 425 entry->object.uvm_obj); 426 } 427 } 428 429 /* 430 * uvm_map_clip_end: ensure that the entry ends at or before 431 * the ending address, if it does't we split the reference 432 * 433 * => caller should use UVM_MAP_CLIP_END macro rather than calling 434 * this directly 435 * => map must be locked by caller 436 */ 437 438 void 439 uvm_map_clip_end(map, entry, end) 440 struct vm_map *map; 441 struct vm_map_entry *entry; 442 vaddr_t end; 443 { 444 struct vm_map_entry * new_entry; 445 vaddr_t new_adj; /* #bytes we move start forward */ 446 447 /* 448 * Create a new entry and insert it 449 * AFTER the specified entry 450 */ 451 452 new_entry = uvm_mapent_alloc(map); 453 uvm_mapent_copy(entry, new_entry); /* entry -> new_entry */ 454 455 new_entry->start = entry->end = end; 456 new_adj = end - entry->start; 457 if (new_entry->object.uvm_obj) 458 new_entry->offset += new_adj; 459 460 if (entry->aref.ar_amap) 461 amap_splitref(&entry->aref, &new_entry->aref, new_adj); 462 463 uvm_map_entry_link(map, entry, new_entry); 464 465 if (UVM_ET_ISSUBMAP(entry)) { 466 /* ... unlikely to happen, but play it safe */ 467 uvm_map_reference(new_entry->object.sub_map); 468 } else { 469 if (UVM_ET_ISOBJ(entry) && 470 entry->object.uvm_obj->pgops && 471 entry->object.uvm_obj->pgops->pgo_reference) 472 entry->object.uvm_obj->pgops->pgo_reference( 473 entry->object.uvm_obj); 474 } 475 } 476 477 478 /* 479 * M A P - m a i n e n t r y p o i n t 480 */ 481 /* 482 * uvm_map: establish a valid mapping in a map 483 * 484 * => assume startp is page aligned. 485 * => assume size is a multiple of PAGE_SIZE. 486 * => assume sys_mmap provides enough of a "hint" to have us skip 487 * over text/data/bss area. 488 * => map must be unlocked (we will lock it) 489 * => <uobj,uoffset> value meanings (4 cases): 490 * [1] <NULL,uoffset> == uoffset is a hint for PMAP_PREFER 491 * [2] <NULL,UVM_UNKNOWN_OFFSET> == don't PMAP_PREFER 492 * [3] <uobj,uoffset> == normal mapping 493 * [4] <uobj,UVM_UNKNOWN_OFFSET> == uvm_map finds offset based on VA 494 * 495 * case [4] is for kernel mappings where we don't know the offset until 496 * we've found a virtual address. note that kernel object offsets are 497 * always relative to vm_map_min(kernel_map). 498 * 499 * => if `align' is non-zero, we try to align the virtual address to 500 * the specified alignment. this is only a hint; if we can't 501 * do it, the address will be unaligned. this is provided as 502 * a mechanism for large pages. 503 * 504 * => XXXCDC: need way to map in external amap? 505 */ 506 507 int 508 uvm_map(map, startp, size, uobj, uoffset, align, flags) 509 struct vm_map *map; 510 vaddr_t *startp; /* IN/OUT */ 511 vsize_t size; 512 struct uvm_object *uobj; 513 voff_t uoffset; 514 vsize_t align; 515 uvm_flag_t flags; 516 { 517 struct vm_map_entry *prev_entry, *new_entry; 518 vm_prot_t prot = UVM_PROTECTION(flags), maxprot = 519 UVM_MAXPROTECTION(flags); 520 vm_inherit_t inherit = UVM_INHERIT(flags); 521 int advice = UVM_ADVICE(flags); 522 UVMHIST_FUNC("uvm_map"); 523 UVMHIST_CALLED(maphist); 524 525 UVMHIST_LOG(maphist, "(map=0x%x, *startp=0x%x, size=%d, flags=0x%x)", 526 map, *startp, size, flags); 527 UVMHIST_LOG(maphist, " uobj/offset 0x%x/%d", uobj, uoffset,0,0); 528 529 /* 530 * detect a popular device driver bug. 531 */ 532 533 KASSERT(curproc != NULL || map->flags & VM_MAP_INTRSAFE); 534 535 /* 536 * check sanity of protection code 537 */ 538 539 if ((prot & maxprot) != prot) { 540 UVMHIST_LOG(maphist, "<- prot. failure: prot=0x%x, max=0x%x", 541 prot, maxprot,0,0); 542 return EACCES; 543 } 544 545 /* 546 * for pager_map, allocate the new entry first to avoid sleeping 547 * for memory while we have the map locked. 548 */ 549 550 new_entry = NULL; 551 if (map == pager_map) { 552 new_entry = uvm_mapent_alloc(map); 553 } 554 555 /* 556 * figure out where to put new VM range 557 */ 558 559 if (vm_map_lock_try(map) == FALSE) { 560 if (flags & UVM_FLAG_TRYLOCK) { 561 if (new_entry) { 562 uvm_mapent_free(new_entry); 563 } 564 return EAGAIN; 565 } 566 vm_map_lock(map); /* could sleep here */ 567 } 568 if ((prev_entry = uvm_map_findspace(map, *startp, size, startp, 569 uobj, uoffset, align, flags)) == NULL) { 570 UVMHIST_LOG(maphist,"<- uvm_map_findspace failed!",0,0,0,0); 571 vm_map_unlock(map); 572 if (new_entry) { 573 uvm_mapent_free(new_entry); 574 } 575 return ENOMEM; 576 } 577 578 #ifdef PMAP_GROWKERNEL 579 { 580 /* 581 * If the kernel pmap can't map the requested space, 582 * then allocate more resources for it. 583 */ 584 if (map == kernel_map && uvm_maxkaddr < (*startp + size)) 585 uvm_maxkaddr = pmap_growkernel(*startp + size); 586 } 587 #endif 588 589 UVMCNT_INCR(uvm_map_call); 590 591 /* 592 * if uobj is null, then uoffset is either a VAC hint for PMAP_PREFER 593 * [typically from uvm_map_reserve] or it is UVM_UNKNOWN_OFFSET. in 594 * either case we want to zero it before storing it in the map entry 595 * (because it looks strange and confusing when debugging...) 596 * 597 * if uobj is not null 598 * if uoffset is not UVM_UNKNOWN_OFFSET then we have a normal mapping 599 * and we do not need to change uoffset. 600 * if uoffset is UVM_UNKNOWN_OFFSET then we need to find the offset 601 * now (based on the starting address of the map). this case is 602 * for kernel object mappings where we don't know the offset until 603 * the virtual address is found (with uvm_map_findspace). the 604 * offset is the distance we are from the start of the map. 605 */ 606 607 if (uobj == NULL) { 608 uoffset = 0; 609 } else { 610 if (uoffset == UVM_UNKNOWN_OFFSET) { 611 KASSERT(UVM_OBJ_IS_KERN_OBJECT(uobj)); 612 uoffset = *startp - vm_map_min(kernel_map); 613 } 614 } 615 616 /* 617 * try and insert in map by extending previous entry, if possible. 618 * XXX: we don't try and pull back the next entry. might be useful 619 * for a stack, but we are currently allocating our stack in advance. 620 */ 621 622 if ((flags & UVM_FLAG_NOMERGE) == 0 && 623 prev_entry->end == *startp && prev_entry != &map->header && 624 prev_entry->object.uvm_obj == uobj) { 625 626 if (uobj && prev_entry->offset + 627 (prev_entry->end - prev_entry->start) != uoffset) 628 goto nomerge; 629 630 if (UVM_ET_ISSUBMAP(prev_entry)) 631 goto nomerge; 632 633 if (prev_entry->protection != prot || 634 prev_entry->max_protection != maxprot) 635 goto nomerge; 636 637 if (prev_entry->inheritance != inherit || 638 prev_entry->advice != advice) 639 goto nomerge; 640 641 /* wiring status must match (new area is unwired) */ 642 if (VM_MAPENT_ISWIRED(prev_entry)) 643 goto nomerge; 644 645 /* 646 * can't extend a shared amap. note: no need to lock amap to 647 * look at refs since we don't care about its exact value. 648 * if it is one (i.e. we have only reference) it will stay there 649 */ 650 651 if (prev_entry->aref.ar_amap && 652 amap_refs(prev_entry->aref.ar_amap) != 1) { 653 goto nomerge; 654 } 655 656 /* got it! */ 657 658 UVMCNT_INCR(map_backmerge); 659 UVMHIST_LOG(maphist," starting back merge", 0, 0, 0, 0); 660 661 /* 662 * drop our reference to uobj since we are extending a reference 663 * that we already have (the ref count can not drop to zero). 664 */ 665 if (uobj && uobj->pgops->pgo_detach) 666 uobj->pgops->pgo_detach(uobj); 667 668 if (prev_entry->aref.ar_amap) { 669 amap_extend(prev_entry, size); 670 } 671 672 prev_entry->end += size; 673 map->size += size; 674 675 UVMHIST_LOG(maphist,"<- done (via backmerge)!", 0, 0, 0, 0); 676 vm_map_unlock(map); 677 if (new_entry) { 678 uvm_mapent_free(new_entry); 679 } 680 return 0; 681 } 682 683 nomerge: 684 UVMHIST_LOG(maphist," allocating new map entry", 0, 0, 0, 0); 685 686 /* 687 * check for possible forward merge (which we don't do) and count 688 * the number of times we missed a *possible* chance to merge more 689 */ 690 691 if ((flags & UVM_FLAG_NOMERGE) == 0 && 692 prev_entry->next != &map->header && 693 prev_entry->next->start == (*startp + size)) 694 UVMCNT_INCR(map_forwmerge); 695 696 /* 697 * allocate new entry and link it in. 698 */ 699 700 if (new_entry == NULL) { 701 new_entry = uvm_mapent_alloc(map); 702 } 703 new_entry->start = *startp; 704 new_entry->end = new_entry->start + size; 705 new_entry->object.uvm_obj = uobj; 706 new_entry->offset = uoffset; 707 708 if (uobj) 709 new_entry->etype = UVM_ET_OBJ; 710 else 711 new_entry->etype = 0; 712 713 if (flags & UVM_FLAG_COPYONW) { 714 new_entry->etype |= UVM_ET_COPYONWRITE; 715 if ((flags & UVM_FLAG_OVERLAY) == 0) 716 new_entry->etype |= UVM_ET_NEEDSCOPY; 717 } 718 719 new_entry->protection = prot; 720 new_entry->max_protection = maxprot; 721 new_entry->inheritance = inherit; 722 new_entry->wired_count = 0; 723 new_entry->advice = advice; 724 if (flags & UVM_FLAG_OVERLAY) { 725 726 /* 727 * to_add: for BSS we overallocate a little since we 728 * are likely to extend 729 */ 730 731 vaddr_t to_add = (flags & UVM_FLAG_AMAPPAD) ? 732 UVM_AMAP_CHUNK << PAGE_SHIFT : 0; 733 struct vm_amap *amap = amap_alloc(size, to_add, M_WAITOK); 734 new_entry->aref.ar_pageoff = 0; 735 new_entry->aref.ar_amap = amap; 736 } else { 737 new_entry->aref.ar_pageoff = 0; 738 new_entry->aref.ar_amap = NULL; 739 } 740 uvm_map_entry_link(map, prev_entry, new_entry); 741 map->size += size; 742 743 /* 744 * Update the free space hint 745 */ 746 747 if ((map->first_free == prev_entry) && 748 (prev_entry->end >= new_entry->start)) 749 map->first_free = new_entry; 750 751 UVMHIST_LOG(maphist,"<- done!", 0, 0, 0, 0); 752 vm_map_unlock(map); 753 return 0; 754 } 755 756 /* 757 * uvm_map_lookup_entry: find map entry at or before an address 758 * 759 * => map must at least be read-locked by caller 760 * => entry is returned in "entry" 761 * => return value is true if address is in the returned entry 762 */ 763 764 boolean_t 765 uvm_map_lookup_entry(map, address, entry) 766 struct vm_map *map; 767 vaddr_t address; 768 struct vm_map_entry **entry; /* OUT */ 769 { 770 struct vm_map_entry *cur; 771 struct vm_map_entry *last; 772 UVMHIST_FUNC("uvm_map_lookup_entry"); 773 UVMHIST_CALLED(maphist); 774 775 UVMHIST_LOG(maphist,"(map=0x%x,addr=0x%x,ent=0x%x)", 776 map, address, entry, 0); 777 778 /* 779 * start looking either from the head of the 780 * list, or from the hint. 781 */ 782 783 simple_lock(&map->hint_lock); 784 cur = map->hint; 785 simple_unlock(&map->hint_lock); 786 787 if (cur == &map->header) 788 cur = cur->next; 789 790 UVMCNT_INCR(uvm_mlk_call); 791 if (address >= cur->start) { 792 793 /* 794 * go from hint to end of list. 795 * 796 * but first, make a quick check to see if 797 * we are already looking at the entry we 798 * want (which is usually the case). 799 * note also that we don't need to save the hint 800 * here... it is the same hint (unless we are 801 * at the header, in which case the hint didn't 802 * buy us anything anyway). 803 */ 804 805 last = &map->header; 806 if ((cur != last) && (cur->end > address)) { 807 UVMCNT_INCR(uvm_mlk_hint); 808 *entry = cur; 809 UVMHIST_LOG(maphist,"<- got it via hint (0x%x)", 810 cur, 0, 0, 0); 811 return (TRUE); 812 } 813 } else { 814 815 /* 816 * go from start to hint, *inclusively* 817 */ 818 819 last = cur->next; 820 cur = map->header.next; 821 } 822 823 /* 824 * search linearly 825 */ 826 827 while (cur != last) { 828 if (cur->end > address) { 829 if (address >= cur->start) { 830 /* 831 * save this lookup for future 832 * hints, and return 833 */ 834 835 *entry = cur; 836 SAVE_HINT(map, map->hint, cur); 837 UVMHIST_LOG(maphist,"<- search got it (0x%x)", 838 cur, 0, 0, 0); 839 return (TRUE); 840 } 841 break; 842 } 843 cur = cur->next; 844 } 845 *entry = cur->prev; 846 SAVE_HINT(map, map->hint, *entry); 847 UVMHIST_LOG(maphist,"<- failed!",0,0,0,0); 848 return (FALSE); 849 } 850 851 /* 852 * uvm_map_findspace: find "length" sized space in "map". 853 * 854 * => "hint" is a hint about where we want it, unless FINDSPACE_FIXED is 855 * set (in which case we insist on using "hint"). 856 * => "result" is VA returned 857 * => uobj/uoffset are to be used to handle VAC alignment, if required 858 * => if `align' is non-zero, we attempt to align to that value. 859 * => caller must at least have read-locked map 860 * => returns NULL on failure, or pointer to prev. map entry if success 861 * => note this is a cross between the old vm_map_findspace and vm_map_find 862 */ 863 864 struct vm_map_entry * 865 uvm_map_findspace(map, hint, length, result, uobj, uoffset, align, flags) 866 struct vm_map *map; 867 vaddr_t hint; 868 vsize_t length; 869 vaddr_t *result; /* OUT */ 870 struct uvm_object *uobj; 871 voff_t uoffset; 872 vsize_t align; 873 int flags; 874 { 875 struct vm_map_entry *entry, *next, *tmp; 876 vaddr_t end, orig_hint; 877 UVMHIST_FUNC("uvm_map_findspace"); 878 UVMHIST_CALLED(maphist); 879 880 UVMHIST_LOG(maphist, "(map=0x%x, hint=0x%x, len=%d, flags=0x%x)", 881 map, hint, length, flags); 882 KASSERT((align & (align - 1)) == 0); 883 KASSERT((flags & UVM_FLAG_FIXED) == 0 || align == 0); 884 885 /* 886 * remember the original hint. if we are aligning, then we 887 * may have to try again with no alignment constraint if 888 * we fail the first time. 889 */ 890 891 orig_hint = hint; 892 if (hint < map->min_offset) { /* check ranges ... */ 893 if (flags & UVM_FLAG_FIXED) { 894 UVMHIST_LOG(maphist,"<- VA below map range",0,0,0,0); 895 return(NULL); 896 } 897 hint = map->min_offset; 898 } 899 if (hint > map->max_offset) { 900 UVMHIST_LOG(maphist,"<- VA 0x%x > range [0x%x->0x%x]", 901 hint, map->min_offset, map->max_offset, 0); 902 return(NULL); 903 } 904 905 /* 906 * Look for the first possible address; if there's already 907 * something at this address, we have to start after it. 908 */ 909 910 if ((flags & UVM_FLAG_FIXED) == 0 && hint == map->min_offset) { 911 if ((entry = map->first_free) != &map->header) 912 hint = entry->end; 913 } else { 914 if (uvm_map_lookup_entry(map, hint, &tmp)) { 915 /* "hint" address already in use ... */ 916 if (flags & UVM_FLAG_FIXED) { 917 UVMHIST_LOG(maphist,"<- fixed & VA in use", 918 0, 0, 0, 0); 919 return(NULL); 920 } 921 hint = tmp->end; 922 } 923 entry = tmp; 924 } 925 926 /* 927 * Look through the rest of the map, trying to fit a new region in 928 * the gap between existing regions, or after the very last region. 929 * note: entry->end = base VA of current gap, 930 * next->start = VA of end of current gap 931 */ 932 933 for (;; hint = (entry = next)->end) { 934 935 /* 936 * Find the end of the proposed new region. Be sure we didn't 937 * go beyond the end of the map, or wrap around the address; 938 * if so, we lose. Otherwise, if this is the last entry, or 939 * if the proposed new region fits before the next entry, we 940 * win. 941 */ 942 943 #ifdef PMAP_PREFER 944 /* 945 * push hint forward as needed to avoid VAC alias problems. 946 * we only do this if a valid offset is specified. 947 */ 948 949 if ((flags & UVM_FLAG_FIXED) == 0 && 950 uoffset != UVM_UNKNOWN_OFFSET) 951 PMAP_PREFER(uoffset, &hint); 952 #endif 953 if (align != 0) { 954 if ((hint & (align - 1)) != 0) 955 hint = roundup(hint, align); 956 /* 957 * XXX Should we PMAP_PREFER() here again? 958 */ 959 } 960 end = hint + length; 961 if (end > map->max_offset || end < hint) { 962 UVMHIST_LOG(maphist,"<- failed (off end)", 0,0,0,0); 963 if (align != 0) { 964 UVMHIST_LOG(maphist, 965 "calling recursively, no align", 966 0,0,0,0); 967 return (uvm_map_findspace(map, orig_hint, 968 length, result, uobj, uoffset, 0, flags)); 969 } 970 return (NULL); 971 } 972 next = entry->next; 973 if (next == &map->header || next->start >= end) 974 break; 975 if (flags & UVM_FLAG_FIXED) { 976 UVMHIST_LOG(maphist,"<- fixed mapping failed", 0,0,0,0); 977 return(NULL); /* only one shot at it ... */ 978 } 979 } 980 SAVE_HINT(map, map->hint, entry); 981 *result = hint; 982 UVMHIST_LOG(maphist,"<- got it! (result=0x%x)", hint, 0,0,0); 983 return (entry); 984 } 985 986 /* 987 * U N M A P - m a i n h e l p e r f u n c t i o n s 988 */ 989 990 /* 991 * uvm_unmap_remove: remove mappings from a vm_map (from "start" up to "stop") 992 * 993 * => caller must check alignment and size 994 * => map must be locked by caller 995 * => we return a list of map entries that we've remove from the map 996 * in "entry_list" 997 */ 998 999 void 1000 uvm_unmap_remove(map, start, end, entry_list) 1001 struct vm_map *map; 1002 vaddr_t start, end; 1003 struct vm_map_entry **entry_list; /* OUT */ 1004 { 1005 struct vm_map_entry *entry, *first_entry, *next; 1006 vaddr_t len; 1007 UVMHIST_FUNC("uvm_unmap_remove"); UVMHIST_CALLED(maphist); 1008 1009 UVMHIST_LOG(maphist,"(map=0x%x, start=0x%x, end=0x%x)", 1010 map, start, end, 0); 1011 VM_MAP_RANGE_CHECK(map, start, end); 1012 1013 /* 1014 * find first entry 1015 */ 1016 1017 if (uvm_map_lookup_entry(map, start, &first_entry) == TRUE) { 1018 /* clip and go... */ 1019 entry = first_entry; 1020 UVM_MAP_CLIP_START(map, entry, start); 1021 /* critical! prevents stale hint */ 1022 SAVE_HINT(map, entry, entry->prev); 1023 } else { 1024 entry = first_entry->next; 1025 } 1026 1027 /* 1028 * Save the free space hint 1029 */ 1030 1031 if (map->first_free->start >= start) 1032 map->first_free = entry->prev; 1033 1034 /* 1035 * note: we now re-use first_entry for a different task. we remove 1036 * a number of map entries from the map and save them in a linked 1037 * list headed by "first_entry". once we remove them from the map 1038 * the caller should unlock the map and drop the references to the 1039 * backing objects [c.f. uvm_unmap_detach]. the object is to 1040 * separate unmapping from reference dropping. why? 1041 * [1] the map has to be locked for unmapping 1042 * [2] the map need not be locked for reference dropping 1043 * [3] dropping references may trigger pager I/O, and if we hit 1044 * a pager that does synchronous I/O we may have to wait for it. 1045 * [4] we would like all waiting for I/O to occur with maps unlocked 1046 * so that we don't block other threads. 1047 */ 1048 1049 first_entry = NULL; 1050 *entry_list = NULL; 1051 1052 /* 1053 * break up the area into map entry sized regions and unmap. note 1054 * that all mappings have to be removed before we can even consider 1055 * dropping references to amaps or VM objects (otherwise we could end 1056 * up with a mapping to a page on the free list which would be very bad) 1057 */ 1058 1059 while ((entry != &map->header) && (entry->start < end)) { 1060 UVM_MAP_CLIP_END(map, entry, end); 1061 next = entry->next; 1062 len = entry->end - entry->start; 1063 1064 /* 1065 * unwire before removing addresses from the pmap; otherwise 1066 * unwiring will put the entries back into the pmap (XXX). 1067 */ 1068 1069 if (VM_MAPENT_ISWIRED(entry)) { 1070 uvm_map_entry_unwire(map, entry); 1071 } 1072 if ((map->flags & VM_MAP_PAGEABLE) == 0) { 1073 1074 /* 1075 * if the map is non-pageable, any pages mapped there 1076 * must be wired and entered with pmap_kenter_pa(), 1077 * and we should free any such pages immediately. 1078 * this is mostly used for kmem_map and mb_map. 1079 */ 1080 1081 uvm_km_pgremove_intrsafe(entry->start, entry->end); 1082 pmap_kremove(entry->start, len); 1083 } else if (UVM_ET_ISOBJ(entry) && 1084 UVM_OBJ_IS_KERN_OBJECT(entry->object.uvm_obj)) { 1085 KASSERT(vm_map_pmap(map) == pmap_kernel()); 1086 1087 /* 1088 * note: kernel object mappings are currently used in 1089 * two ways: 1090 * [1] "normal" mappings of pages in the kernel object 1091 * [2] uvm_km_valloc'd allocations in which we 1092 * pmap_enter in some non-kernel-object page 1093 * (e.g. vmapbuf). 1094 * 1095 * for case [1], we need to remove the mapping from 1096 * the pmap and then remove the page from the kernel 1097 * object (because, once pages in a kernel object are 1098 * unmapped they are no longer needed, unlike, say, 1099 * a vnode where you might want the data to persist 1100 * until flushed out of a queue). 1101 * 1102 * for case [2], we need to remove the mapping from 1103 * the pmap. there shouldn't be any pages at the 1104 * specified offset in the kernel object [but it 1105 * doesn't hurt to call uvm_km_pgremove just to be 1106 * safe?] 1107 * 1108 * uvm_km_pgremove currently does the following: 1109 * for pages in the kernel object in range: 1110 * - drops the swap slot 1111 * - uvm_pagefree the page 1112 */ 1113 1114 /* 1115 * remove mappings from pmap and drop the pages 1116 * from the object. offsets are always relative 1117 * to vm_map_min(kernel_map). 1118 */ 1119 1120 pmap_remove(pmap_kernel(), entry->start, 1121 entry->start + len); 1122 uvm_km_pgremove(entry->object.uvm_obj, 1123 entry->start - vm_map_min(kernel_map), 1124 entry->end - vm_map_min(kernel_map)); 1125 1126 /* 1127 * null out kernel_object reference, we've just 1128 * dropped it 1129 */ 1130 1131 entry->etype &= ~UVM_ET_OBJ; 1132 entry->object.uvm_obj = NULL; 1133 } else if (UVM_ET_ISOBJ(entry) || entry->aref.ar_amap) { 1134 1135 /* 1136 * remove mappings the standard way. 1137 */ 1138 1139 pmap_remove(map->pmap, entry->start, entry->end); 1140 } 1141 1142 /* 1143 * remove entry from map and put it on our list of entries 1144 * that we've nuked. then go to next entry. 1145 */ 1146 1147 UVMHIST_LOG(maphist, " removed map entry 0x%x", entry, 0, 0,0); 1148 1149 /* critical! prevents stale hint */ 1150 SAVE_HINT(map, entry, entry->prev); 1151 1152 uvm_map_entry_unlink(map, entry); 1153 map->size -= len; 1154 entry->next = first_entry; 1155 first_entry = entry; 1156 entry = next; 1157 } 1158 pmap_update(vm_map_pmap(map)); 1159 1160 /* 1161 * now we've cleaned up the map and are ready for the caller to drop 1162 * references to the mapped objects. 1163 */ 1164 1165 *entry_list = first_entry; 1166 UVMHIST_LOG(maphist,"<- done!", 0, 0, 0, 0); 1167 } 1168 1169 /* 1170 * uvm_unmap_detach: drop references in a chain of map entries 1171 * 1172 * => we will free the map entries as we traverse the list. 1173 */ 1174 1175 void 1176 uvm_unmap_detach(first_entry, flags) 1177 struct vm_map_entry *first_entry; 1178 int flags; 1179 { 1180 struct vm_map_entry *next_entry; 1181 UVMHIST_FUNC("uvm_unmap_detach"); UVMHIST_CALLED(maphist); 1182 1183 while (first_entry) { 1184 KASSERT(!VM_MAPENT_ISWIRED(first_entry)); 1185 UVMHIST_LOG(maphist, 1186 " detach 0x%x: amap=0x%x, obj=0x%x, submap?=%d", 1187 first_entry, first_entry->aref.ar_amap, 1188 first_entry->object.uvm_obj, 1189 UVM_ET_ISSUBMAP(first_entry)); 1190 1191 /* 1192 * drop reference to amap, if we've got one 1193 */ 1194 1195 if (first_entry->aref.ar_amap) 1196 uvm_map_unreference_amap(first_entry, flags); 1197 1198 /* 1199 * drop reference to our backing object, if we've got one 1200 */ 1201 1202 if (UVM_ET_ISSUBMAP(first_entry)) { 1203 /* ... unlikely to happen, but play it safe */ 1204 uvm_map_deallocate(first_entry->object.sub_map); 1205 } else { 1206 if (UVM_ET_ISOBJ(first_entry) && 1207 first_entry->object.uvm_obj->pgops->pgo_detach) 1208 first_entry->object.uvm_obj->pgops-> 1209 pgo_detach(first_entry->object.uvm_obj); 1210 } 1211 next_entry = first_entry->next; 1212 uvm_mapent_free(first_entry); 1213 first_entry = next_entry; 1214 } 1215 UVMHIST_LOG(maphist, "<- done", 0,0,0,0); 1216 } 1217 1218 /* 1219 * E X T R A C T I O N F U N C T I O N S 1220 */ 1221 1222 /* 1223 * uvm_map_reserve: reserve space in a vm_map for future use. 1224 * 1225 * => we reserve space in a map by putting a dummy map entry in the 1226 * map (dummy means obj=NULL, amap=NULL, prot=VM_PROT_NONE) 1227 * => map should be unlocked (we will write lock it) 1228 * => we return true if we were able to reserve space 1229 * => XXXCDC: should be inline? 1230 */ 1231 1232 int 1233 uvm_map_reserve(map, size, offset, align, raddr) 1234 struct vm_map *map; 1235 vsize_t size; 1236 vaddr_t offset; /* hint for pmap_prefer */ 1237 vsize_t align; /* alignment hint */ 1238 vaddr_t *raddr; /* IN:hint, OUT: reserved VA */ 1239 { 1240 UVMHIST_FUNC("uvm_map_reserve"); UVMHIST_CALLED(maphist); 1241 1242 UVMHIST_LOG(maphist, "(map=0x%x, size=0x%x, offset=0x%x,addr=0x%x)", 1243 map,size,offset,raddr); 1244 1245 size = round_page(size); 1246 if (*raddr < vm_map_min(map)) 1247 *raddr = vm_map_min(map); /* hint */ 1248 1249 /* 1250 * reserve some virtual space. 1251 */ 1252 1253 if (uvm_map(map, raddr, size, NULL, offset, 0, 1254 UVM_MAPFLAG(UVM_PROT_NONE, UVM_PROT_NONE, UVM_INH_NONE, 1255 UVM_ADV_RANDOM, UVM_FLAG_NOMERGE)) != 0) { 1256 UVMHIST_LOG(maphist, "<- done (no VM)", 0,0,0,0); 1257 return (FALSE); 1258 } 1259 1260 UVMHIST_LOG(maphist, "<- done (*raddr=0x%x)", *raddr,0,0,0); 1261 return (TRUE); 1262 } 1263 1264 /* 1265 * uvm_map_replace: replace a reserved (blank) area of memory with 1266 * real mappings. 1267 * 1268 * => caller must WRITE-LOCK the map 1269 * => we return TRUE if replacement was a success 1270 * => we expect the newents chain to have nnewents entrys on it and 1271 * we expect newents->prev to point to the last entry on the list 1272 * => note newents is allowed to be NULL 1273 */ 1274 1275 int 1276 uvm_map_replace(map, start, end, newents, nnewents) 1277 struct vm_map *map; 1278 vaddr_t start, end; 1279 struct vm_map_entry *newents; 1280 int nnewents; 1281 { 1282 struct vm_map_entry *oldent, *last; 1283 1284 /* 1285 * first find the blank map entry at the specified address 1286 */ 1287 1288 if (!uvm_map_lookup_entry(map, start, &oldent)) { 1289 return(FALSE); 1290 } 1291 1292 /* 1293 * check to make sure we have a proper blank entry 1294 */ 1295 1296 if (oldent->start != start || oldent->end != end || 1297 oldent->object.uvm_obj != NULL || oldent->aref.ar_amap != NULL) { 1298 return (FALSE); 1299 } 1300 1301 #ifdef DIAGNOSTIC 1302 1303 /* 1304 * sanity check the newents chain 1305 */ 1306 1307 { 1308 struct vm_map_entry *tmpent = newents; 1309 int nent = 0; 1310 vaddr_t cur = start; 1311 1312 while (tmpent) { 1313 nent++; 1314 if (tmpent->start < cur) 1315 panic("uvm_map_replace1"); 1316 if (tmpent->start > tmpent->end || tmpent->end > end) { 1317 printf("tmpent->start=0x%lx, tmpent->end=0x%lx, end=0x%lx\n", 1318 tmpent->start, tmpent->end, end); 1319 panic("uvm_map_replace2"); 1320 } 1321 cur = tmpent->end; 1322 if (tmpent->next) { 1323 if (tmpent->next->prev != tmpent) 1324 panic("uvm_map_replace3"); 1325 } else { 1326 if (newents->prev != tmpent) 1327 panic("uvm_map_replace4"); 1328 } 1329 tmpent = tmpent->next; 1330 } 1331 if (nent != nnewents) 1332 panic("uvm_map_replace5"); 1333 } 1334 #endif 1335 1336 /* 1337 * map entry is a valid blank! replace it. (this does all the 1338 * work of map entry link/unlink...). 1339 */ 1340 1341 if (newents) { 1342 last = newents->prev; 1343 1344 /* critical: flush stale hints out of map */ 1345 SAVE_HINT(map, map->hint, newents); 1346 if (map->first_free == oldent) 1347 map->first_free = last; 1348 1349 last->next = oldent->next; 1350 last->next->prev = last; 1351 newents->prev = oldent->prev; 1352 newents->prev->next = newents; 1353 map->nentries = map->nentries + (nnewents - 1); 1354 1355 } else { 1356 1357 /* critical: flush stale hints out of map */ 1358 SAVE_HINT(map, map->hint, oldent->prev); 1359 if (map->first_free == oldent) 1360 map->first_free = oldent->prev; 1361 1362 /* NULL list of new entries: just remove the old one */ 1363 uvm_map_entry_unlink(map, oldent); 1364 } 1365 1366 1367 /* 1368 * now we can free the old blank entry, unlock the map and return. 1369 */ 1370 1371 uvm_mapent_free(oldent); 1372 return(TRUE); 1373 } 1374 1375 /* 1376 * uvm_map_extract: extract a mapping from a map and put it somewhere 1377 * (maybe removing the old mapping) 1378 * 1379 * => maps should be unlocked (we will write lock them) 1380 * => returns 0 on success, error code otherwise 1381 * => start must be page aligned 1382 * => len must be page sized 1383 * => flags: 1384 * UVM_EXTRACT_REMOVE: remove mappings from srcmap 1385 * UVM_EXTRACT_CONTIG: abort if unmapped area (advisory only) 1386 * UVM_EXTRACT_QREF: for a temporary extraction do quick obj refs 1387 * UVM_EXTRACT_FIXPROT: set prot to maxprot as we go 1388 * >>>NOTE: if you set REMOVE, you are not allowed to use CONTIG or QREF!<<< 1389 * >>>NOTE: QREF's must be unmapped via the QREF path, thus should only 1390 * be used from within the kernel in a kernel level map <<< 1391 */ 1392 1393 int 1394 uvm_map_extract(srcmap, start, len, dstmap, dstaddrp, flags) 1395 struct vm_map *srcmap, *dstmap; 1396 vaddr_t start, *dstaddrp; 1397 vsize_t len; 1398 int flags; 1399 { 1400 vaddr_t dstaddr, end, newend, oldoffset, fudge, orig_fudge, 1401 oldstart; 1402 struct vm_map_entry *chain, *endchain, *entry, *orig_entry, *newentry, 1403 *deadentry, *oldentry; 1404 vsize_t elen; 1405 int nchain, error, copy_ok; 1406 UVMHIST_FUNC("uvm_map_extract"); UVMHIST_CALLED(maphist); 1407 1408 UVMHIST_LOG(maphist,"(srcmap=0x%x,start=0x%x, len=0x%x", srcmap, start, 1409 len,0); 1410 UVMHIST_LOG(maphist," ...,dstmap=0x%x, flags=0x%x)", dstmap,flags,0,0); 1411 1412 /* 1413 * step 0: sanity check: start must be on a page boundary, length 1414 * must be page sized. can't ask for CONTIG/QREF if you asked for 1415 * REMOVE. 1416 */ 1417 1418 KASSERT((start & PAGE_MASK) == 0 && (len & PAGE_MASK) == 0); 1419 KASSERT((flags & UVM_EXTRACT_REMOVE) == 0 || 1420 (flags & (UVM_EXTRACT_CONTIG|UVM_EXTRACT_QREF)) == 0); 1421 1422 /* 1423 * step 1: reserve space in the target map for the extracted area 1424 */ 1425 1426 dstaddr = vm_map_min(dstmap); 1427 if (uvm_map_reserve(dstmap, len, start, 0, &dstaddr) == FALSE) 1428 return(ENOMEM); 1429 *dstaddrp = dstaddr; /* pass address back to caller */ 1430 UVMHIST_LOG(maphist, " dstaddr=0x%x", dstaddr,0,0,0); 1431 1432 /* 1433 * step 2: setup for the extraction process loop by init'ing the 1434 * map entry chain, locking src map, and looking up the first useful 1435 * entry in the map. 1436 */ 1437 1438 end = start + len; 1439 newend = dstaddr + len; 1440 chain = endchain = NULL; 1441 nchain = 0; 1442 vm_map_lock(srcmap); 1443 1444 if (uvm_map_lookup_entry(srcmap, start, &entry)) { 1445 1446 /* "start" is within an entry */ 1447 if (flags & UVM_EXTRACT_QREF) { 1448 1449 /* 1450 * for quick references we don't clip the entry, so 1451 * the entry may map space "before" the starting 1452 * virtual address... this is the "fudge" factor 1453 * (which can be non-zero only the first time 1454 * through the "while" loop in step 3). 1455 */ 1456 1457 fudge = start - entry->start; 1458 } else { 1459 1460 /* 1461 * normal reference: we clip the map to fit (thus 1462 * fudge is zero) 1463 */ 1464 1465 UVM_MAP_CLIP_START(srcmap, entry, start); 1466 SAVE_HINT(srcmap, srcmap->hint, entry->prev); 1467 fudge = 0; 1468 } 1469 } else { 1470 1471 /* "start" is not within an entry ... skip to next entry */ 1472 if (flags & UVM_EXTRACT_CONTIG) { 1473 error = EINVAL; 1474 goto bad; /* definite hole here ... */ 1475 } 1476 1477 entry = entry->next; 1478 fudge = 0; 1479 } 1480 1481 /* save values from srcmap for step 6 */ 1482 orig_entry = entry; 1483 orig_fudge = fudge; 1484 1485 /* 1486 * step 3: now start looping through the map entries, extracting 1487 * as we go. 1488 */ 1489 1490 while (entry->start < end && entry != &srcmap->header) { 1491 1492 /* if we are not doing a quick reference, clip it */ 1493 if ((flags & UVM_EXTRACT_QREF) == 0) 1494 UVM_MAP_CLIP_END(srcmap, entry, end); 1495 1496 /* clear needs_copy (allow chunking) */ 1497 if (UVM_ET_ISNEEDSCOPY(entry)) { 1498 if (fudge) 1499 oldstart = entry->start; 1500 else 1501 oldstart = 0; /* XXX: gcc */ 1502 amap_copy(srcmap, entry, M_NOWAIT, TRUE, start, end); 1503 if (UVM_ET_ISNEEDSCOPY(entry)) { /* failed? */ 1504 error = ENOMEM; 1505 goto bad; 1506 } 1507 1508 /* amap_copy could clip (during chunk)! update fudge */ 1509 if (fudge) { 1510 fudge = fudge - (entry->start - oldstart); 1511 orig_fudge = fudge; 1512 } 1513 } 1514 1515 /* calculate the offset of this from "start" */ 1516 oldoffset = (entry->start + fudge) - start; 1517 1518 /* allocate a new map entry */ 1519 newentry = uvm_mapent_alloc(dstmap); 1520 if (newentry == NULL) { 1521 error = ENOMEM; 1522 goto bad; 1523 } 1524 1525 /* set up new map entry */ 1526 newentry->next = NULL; 1527 newentry->prev = endchain; 1528 newentry->start = dstaddr + oldoffset; 1529 newentry->end = 1530 newentry->start + (entry->end - (entry->start + fudge)); 1531 if (newentry->end > newend || newentry->end < newentry->start) 1532 newentry->end = newend; 1533 newentry->object.uvm_obj = entry->object.uvm_obj; 1534 if (newentry->object.uvm_obj) { 1535 if (newentry->object.uvm_obj->pgops->pgo_reference) 1536 newentry->object.uvm_obj->pgops-> 1537 pgo_reference(newentry->object.uvm_obj); 1538 newentry->offset = entry->offset + fudge; 1539 } else { 1540 newentry->offset = 0; 1541 } 1542 newentry->etype = entry->etype; 1543 newentry->protection = (flags & UVM_EXTRACT_FIXPROT) ? 1544 entry->max_protection : entry->protection; 1545 newentry->max_protection = entry->max_protection; 1546 newentry->inheritance = entry->inheritance; 1547 newentry->wired_count = 0; 1548 newentry->aref.ar_amap = entry->aref.ar_amap; 1549 if (newentry->aref.ar_amap) { 1550 newentry->aref.ar_pageoff = 1551 entry->aref.ar_pageoff + (fudge >> PAGE_SHIFT); 1552 uvm_map_reference_amap(newentry, AMAP_SHARED | 1553 ((flags & UVM_EXTRACT_QREF) ? AMAP_REFALL : 0)); 1554 } else { 1555 newentry->aref.ar_pageoff = 0; 1556 } 1557 newentry->advice = entry->advice; 1558 1559 /* now link it on the chain */ 1560 nchain++; 1561 if (endchain == NULL) { 1562 chain = endchain = newentry; 1563 } else { 1564 endchain->next = newentry; 1565 endchain = newentry; 1566 } 1567 1568 /* end of 'while' loop! */ 1569 if ((flags & UVM_EXTRACT_CONTIG) && entry->end < end && 1570 (entry->next == &srcmap->header || 1571 entry->next->start != entry->end)) { 1572 error = EINVAL; 1573 goto bad; 1574 } 1575 entry = entry->next; 1576 fudge = 0; 1577 } 1578 1579 /* 1580 * step 4: close off chain (in format expected by uvm_map_replace) 1581 */ 1582 1583 if (chain) 1584 chain->prev = endchain; 1585 1586 /* 1587 * step 5: attempt to lock the dest map so we can pmap_copy. 1588 * note usage of copy_ok: 1589 * 1 => dstmap locked, pmap_copy ok, and we "replace" here (step 5) 1590 * 0 => dstmap unlocked, NO pmap_copy, and we will "replace" in step 7 1591 */ 1592 1593 if (srcmap == dstmap || vm_map_lock_try(dstmap) == TRUE) { 1594 copy_ok = 1; 1595 if (!uvm_map_replace(dstmap, dstaddr, dstaddr+len, chain, 1596 nchain)) { 1597 if (srcmap != dstmap) 1598 vm_map_unlock(dstmap); 1599 error = EIO; 1600 goto bad; 1601 } 1602 } else { 1603 copy_ok = 0; 1604 /* replace defered until step 7 */ 1605 } 1606 1607 /* 1608 * step 6: traverse the srcmap a second time to do the following: 1609 * - if we got a lock on the dstmap do pmap_copy 1610 * - if UVM_EXTRACT_REMOVE remove the entries 1611 * we make use of orig_entry and orig_fudge (saved in step 2) 1612 */ 1613 1614 if (copy_ok || (flags & UVM_EXTRACT_REMOVE)) { 1615 1616 /* purge possible stale hints from srcmap */ 1617 if (flags & UVM_EXTRACT_REMOVE) { 1618 SAVE_HINT(srcmap, srcmap->hint, orig_entry->prev); 1619 if (srcmap->first_free->start >= start) 1620 srcmap->first_free = orig_entry->prev; 1621 } 1622 1623 entry = orig_entry; 1624 fudge = orig_fudge; 1625 deadentry = NULL; /* for UVM_EXTRACT_REMOVE */ 1626 1627 while (entry->start < end && entry != &srcmap->header) { 1628 if (copy_ok) { 1629 oldoffset = (entry->start + fudge) - start; 1630 elen = MIN(end, entry->end) - 1631 (entry->start + fudge); 1632 pmap_copy(dstmap->pmap, srcmap->pmap, 1633 dstaddr + oldoffset, elen, 1634 entry->start + fudge); 1635 } 1636 1637 /* we advance "entry" in the following if statement */ 1638 if (flags & UVM_EXTRACT_REMOVE) { 1639 pmap_remove(srcmap->pmap, entry->start, 1640 entry->end); 1641 oldentry = entry; /* save entry */ 1642 entry = entry->next; /* advance */ 1643 uvm_map_entry_unlink(srcmap, oldentry); 1644 /* add to dead list */ 1645 oldentry->next = deadentry; 1646 deadentry = oldentry; 1647 } else { 1648 entry = entry->next; /* advance */ 1649 } 1650 1651 /* end of 'while' loop */ 1652 fudge = 0; 1653 } 1654 pmap_update(srcmap->pmap); 1655 1656 /* 1657 * unlock dstmap. we will dispose of deadentry in 1658 * step 7 if needed 1659 */ 1660 1661 if (copy_ok && srcmap != dstmap) 1662 vm_map_unlock(dstmap); 1663 1664 } else { 1665 deadentry = NULL; 1666 } 1667 1668 /* 1669 * step 7: we are done with the source map, unlock. if copy_ok 1670 * is 0 then we have not replaced the dummy mapping in dstmap yet 1671 * and we need to do so now. 1672 */ 1673 1674 vm_map_unlock(srcmap); 1675 if ((flags & UVM_EXTRACT_REMOVE) && deadentry) 1676 uvm_unmap_detach(deadentry, 0); /* dispose of old entries */ 1677 1678 /* now do the replacement if we didn't do it in step 5 */ 1679 if (copy_ok == 0) { 1680 vm_map_lock(dstmap); 1681 error = uvm_map_replace(dstmap, dstaddr, dstaddr+len, chain, 1682 nchain); 1683 vm_map_unlock(dstmap); 1684 1685 if (error == FALSE) { 1686 error = EIO; 1687 goto bad2; 1688 } 1689 } 1690 return(0); 1691 1692 /* 1693 * bad: failure recovery 1694 */ 1695 bad: 1696 vm_map_unlock(srcmap); 1697 bad2: /* src already unlocked */ 1698 if (chain) 1699 uvm_unmap_detach(chain, 1700 (flags & UVM_EXTRACT_QREF) ? AMAP_REFALL : 0); 1701 uvm_unmap(dstmap, dstaddr, dstaddr+len); /* ??? */ 1702 return(error); 1703 } 1704 1705 /* end of extraction functions */ 1706 1707 /* 1708 * uvm_map_submap: punch down part of a map into a submap 1709 * 1710 * => only the kernel_map is allowed to be submapped 1711 * => the purpose of submapping is to break up the locking granularity 1712 * of a larger map 1713 * => the range specified must have been mapped previously with a uvm_map() 1714 * call [with uobj==NULL] to create a blank map entry in the main map. 1715 * [And it had better still be blank!] 1716 * => maps which contain submaps should never be copied or forked. 1717 * => to remove a submap, use uvm_unmap() on the main map 1718 * and then uvm_map_deallocate() the submap. 1719 * => main map must be unlocked. 1720 * => submap must have been init'd and have a zero reference count. 1721 * [need not be locked as we don't actually reference it] 1722 */ 1723 1724 int 1725 uvm_map_submap(map, start, end, submap) 1726 struct vm_map *map, *submap; 1727 vaddr_t start, end; 1728 { 1729 struct vm_map_entry *entry; 1730 int error; 1731 1732 vm_map_lock(map); 1733 VM_MAP_RANGE_CHECK(map, start, end); 1734 1735 if (uvm_map_lookup_entry(map, start, &entry)) { 1736 UVM_MAP_CLIP_START(map, entry, start); 1737 UVM_MAP_CLIP_END(map, entry, end); /* to be safe */ 1738 } else { 1739 entry = NULL; 1740 } 1741 1742 if (entry != NULL && 1743 entry->start == start && entry->end == end && 1744 entry->object.uvm_obj == NULL && entry->aref.ar_amap == NULL && 1745 !UVM_ET_ISCOPYONWRITE(entry) && !UVM_ET_ISNEEDSCOPY(entry)) { 1746 entry->etype |= UVM_ET_SUBMAP; 1747 entry->object.sub_map = submap; 1748 entry->offset = 0; 1749 uvm_map_reference(submap); 1750 error = 0; 1751 } else { 1752 error = EINVAL; 1753 } 1754 vm_map_unlock(map); 1755 return error; 1756 } 1757 1758 1759 /* 1760 * uvm_map_protect: change map protection 1761 * 1762 * => set_max means set max_protection. 1763 * => map must be unlocked. 1764 */ 1765 1766 #define MASK(entry) (UVM_ET_ISCOPYONWRITE(entry) ? \ 1767 ~VM_PROT_WRITE : VM_PROT_ALL) 1768 1769 int 1770 uvm_map_protect(map, start, end, new_prot, set_max) 1771 struct vm_map *map; 1772 vaddr_t start, end; 1773 vm_prot_t new_prot; 1774 boolean_t set_max; 1775 { 1776 struct vm_map_entry *current, *entry; 1777 int error = 0; 1778 UVMHIST_FUNC("uvm_map_protect"); UVMHIST_CALLED(maphist); 1779 UVMHIST_LOG(maphist,"(map=0x%x,start=0x%x,end=0x%x,new_prot=0x%x)", 1780 map, start, end, new_prot); 1781 1782 vm_map_lock(map); 1783 VM_MAP_RANGE_CHECK(map, start, end); 1784 if (uvm_map_lookup_entry(map, start, &entry)) { 1785 UVM_MAP_CLIP_START(map, entry, start); 1786 } else { 1787 entry = entry->next; 1788 } 1789 1790 /* 1791 * make a first pass to check for protection violations. 1792 */ 1793 1794 current = entry; 1795 while ((current != &map->header) && (current->start < end)) { 1796 if (UVM_ET_ISSUBMAP(current)) { 1797 error = EINVAL; 1798 goto out; 1799 } 1800 if ((new_prot & current->max_protection) != new_prot) { 1801 error = EACCES; 1802 goto out; 1803 } 1804 current = current->next; 1805 } 1806 1807 /* go back and fix up protections (no need to clip this time). */ 1808 1809 current = entry; 1810 while ((current != &map->header) && (current->start < end)) { 1811 vm_prot_t old_prot; 1812 1813 UVM_MAP_CLIP_END(map, current, end); 1814 old_prot = current->protection; 1815 if (set_max) 1816 current->protection = 1817 (current->max_protection = new_prot) & old_prot; 1818 else 1819 current->protection = new_prot; 1820 1821 /* 1822 * update physical map if necessary. worry about copy-on-write 1823 * here -- CHECK THIS XXX 1824 */ 1825 1826 if (current->protection != old_prot) { 1827 /* update pmap! */ 1828 pmap_protect(map->pmap, current->start, current->end, 1829 current->protection & MASK(entry)); 1830 } 1831 1832 /* 1833 * If the map is configured to lock any future mappings, 1834 * wire this entry now if the old protection was VM_PROT_NONE 1835 * and the new protection is not VM_PROT_NONE. 1836 */ 1837 1838 if ((map->flags & VM_MAP_WIREFUTURE) != 0 && 1839 VM_MAPENT_ISWIRED(entry) == 0 && 1840 old_prot == VM_PROT_NONE && 1841 new_prot != VM_PROT_NONE) { 1842 if (uvm_map_pageable(map, entry->start, 1843 entry->end, FALSE, 1844 UVM_LK_ENTER|UVM_LK_EXIT) != 0) { 1845 1846 /* 1847 * If locking the entry fails, remember the 1848 * error if it's the first one. Note we 1849 * still continue setting the protection in 1850 * the map, but will return the error 1851 * condition regardless. 1852 * 1853 * XXX Ignore what the actual error is, 1854 * XXX just call it a resource shortage 1855 * XXX so that it doesn't get confused 1856 * XXX what uvm_map_protect() itself would 1857 * XXX normally return. 1858 */ 1859 1860 error = ENOMEM; 1861 } 1862 } 1863 current = current->next; 1864 } 1865 pmap_update(map->pmap); 1866 1867 out: 1868 vm_map_unlock(map); 1869 UVMHIST_LOG(maphist, "<- done, error=%d",error,0,0,0); 1870 return error; 1871 } 1872 1873 #undef MASK 1874 1875 /* 1876 * uvm_map_inherit: set inheritance code for range of addrs in map. 1877 * 1878 * => map must be unlocked 1879 * => note that the inherit code is used during a "fork". see fork 1880 * code for details. 1881 */ 1882 1883 int 1884 uvm_map_inherit(map, start, end, new_inheritance) 1885 struct vm_map *map; 1886 vaddr_t start; 1887 vaddr_t end; 1888 vm_inherit_t new_inheritance; 1889 { 1890 struct vm_map_entry *entry, *temp_entry; 1891 UVMHIST_FUNC("uvm_map_inherit"); UVMHIST_CALLED(maphist); 1892 UVMHIST_LOG(maphist,"(map=0x%x,start=0x%x,end=0x%x,new_inh=0x%x)", 1893 map, start, end, new_inheritance); 1894 1895 switch (new_inheritance) { 1896 case MAP_INHERIT_NONE: 1897 case MAP_INHERIT_COPY: 1898 case MAP_INHERIT_SHARE: 1899 break; 1900 default: 1901 UVMHIST_LOG(maphist,"<- done (INVALID ARG)",0,0,0,0); 1902 return EINVAL; 1903 } 1904 1905 vm_map_lock(map); 1906 VM_MAP_RANGE_CHECK(map, start, end); 1907 if (uvm_map_lookup_entry(map, start, &temp_entry)) { 1908 entry = temp_entry; 1909 UVM_MAP_CLIP_START(map, entry, start); 1910 } else { 1911 entry = temp_entry->next; 1912 } 1913 while ((entry != &map->header) && (entry->start < end)) { 1914 UVM_MAP_CLIP_END(map, entry, end); 1915 entry->inheritance = new_inheritance; 1916 entry = entry->next; 1917 } 1918 vm_map_unlock(map); 1919 UVMHIST_LOG(maphist,"<- done (OK)",0,0,0,0); 1920 return 0; 1921 } 1922 1923 /* 1924 * uvm_map_advice: set advice code for range of addrs in map. 1925 * 1926 * => map must be unlocked 1927 */ 1928 1929 int 1930 uvm_map_advice(map, start, end, new_advice) 1931 struct vm_map *map; 1932 vaddr_t start; 1933 vaddr_t end; 1934 int new_advice; 1935 { 1936 struct vm_map_entry *entry, *temp_entry; 1937 UVMHIST_FUNC("uvm_map_advice"); UVMHIST_CALLED(maphist); 1938 UVMHIST_LOG(maphist,"(map=0x%x,start=0x%x,end=0x%x,new_adv=0x%x)", 1939 map, start, end, new_advice); 1940 1941 vm_map_lock(map); 1942 VM_MAP_RANGE_CHECK(map, start, end); 1943 if (uvm_map_lookup_entry(map, start, &temp_entry)) { 1944 entry = temp_entry; 1945 UVM_MAP_CLIP_START(map, entry, start); 1946 } else { 1947 entry = temp_entry->next; 1948 } 1949 1950 /* 1951 * XXXJRT: disallow holes? 1952 */ 1953 1954 while ((entry != &map->header) && (entry->start < end)) { 1955 UVM_MAP_CLIP_END(map, entry, end); 1956 1957 switch (new_advice) { 1958 case MADV_NORMAL: 1959 case MADV_RANDOM: 1960 case MADV_SEQUENTIAL: 1961 /* nothing special here */ 1962 break; 1963 1964 default: 1965 vm_map_unlock(map); 1966 UVMHIST_LOG(maphist,"<- done (INVALID ARG)",0,0,0,0); 1967 return EINVAL; 1968 } 1969 entry->advice = new_advice; 1970 entry = entry->next; 1971 } 1972 1973 vm_map_unlock(map); 1974 UVMHIST_LOG(maphist,"<- done (OK)",0,0,0,0); 1975 return 0; 1976 } 1977 1978 /* 1979 * uvm_map_pageable: sets the pageability of a range in a map. 1980 * 1981 * => wires map entries. should not be used for transient page locking. 1982 * for that, use uvm_fault_wire()/uvm_fault_unwire() (see uvm_vslock()). 1983 * => regions sepcified as not pageable require lock-down (wired) memory 1984 * and page tables. 1985 * => map must never be read-locked 1986 * => if islocked is TRUE, map is already write-locked 1987 * => we always unlock the map, since we must downgrade to a read-lock 1988 * to call uvm_fault_wire() 1989 * => XXXCDC: check this and try and clean it up. 1990 */ 1991 1992 int 1993 uvm_map_pageable(map, start, end, new_pageable, lockflags) 1994 struct vm_map *map; 1995 vaddr_t start, end; 1996 boolean_t new_pageable; 1997 int lockflags; 1998 { 1999 struct vm_map_entry *entry, *start_entry, *failed_entry; 2000 int rv; 2001 #ifdef DIAGNOSTIC 2002 u_int timestamp_save; 2003 #endif 2004 UVMHIST_FUNC("uvm_map_pageable"); UVMHIST_CALLED(maphist); 2005 UVMHIST_LOG(maphist,"(map=0x%x,start=0x%x,end=0x%x,new_pageable=0x%x)", 2006 map, start, end, new_pageable); 2007 KASSERT(map->flags & VM_MAP_PAGEABLE); 2008 2009 if ((lockflags & UVM_LK_ENTER) == 0) 2010 vm_map_lock(map); 2011 VM_MAP_RANGE_CHECK(map, start, end); 2012 2013 /* 2014 * only one pageability change may take place at one time, since 2015 * uvm_fault_wire assumes it will be called only once for each 2016 * wiring/unwiring. therefore, we have to make sure we're actually 2017 * changing the pageability for the entire region. we do so before 2018 * making any changes. 2019 */ 2020 2021 if (uvm_map_lookup_entry(map, start, &start_entry) == FALSE) { 2022 if ((lockflags & UVM_LK_EXIT) == 0) 2023 vm_map_unlock(map); 2024 2025 UVMHIST_LOG(maphist,"<- done (fault)",0,0,0,0); 2026 return EFAULT; 2027 } 2028 entry = start_entry; 2029 2030 /* 2031 * handle wiring and unwiring separately. 2032 */ 2033 2034 if (new_pageable) { /* unwire */ 2035 UVM_MAP_CLIP_START(map, entry, start); 2036 2037 /* 2038 * unwiring. first ensure that the range to be unwired is 2039 * really wired down and that there are no holes. 2040 */ 2041 2042 while ((entry != &map->header) && (entry->start < end)) { 2043 if (entry->wired_count == 0 || 2044 (entry->end < end && 2045 (entry->next == &map->header || 2046 entry->next->start > entry->end))) { 2047 if ((lockflags & UVM_LK_EXIT) == 0) 2048 vm_map_unlock(map); 2049 UVMHIST_LOG(maphist, "<- done (INVAL)",0,0,0,0); 2050 return EINVAL; 2051 } 2052 entry = entry->next; 2053 } 2054 2055 /* 2056 * POSIX 1003.1b - a single munlock call unlocks a region, 2057 * regardless of the number of mlock calls made on that 2058 * region. 2059 */ 2060 2061 entry = start_entry; 2062 while ((entry != &map->header) && (entry->start < end)) { 2063 UVM_MAP_CLIP_END(map, entry, end); 2064 if (VM_MAPENT_ISWIRED(entry)) 2065 uvm_map_entry_unwire(map, entry); 2066 entry = entry->next; 2067 } 2068 if ((lockflags & UVM_LK_EXIT) == 0) 2069 vm_map_unlock(map); 2070 UVMHIST_LOG(maphist,"<- done (OK UNWIRE)",0,0,0,0); 2071 return 0; 2072 } 2073 2074 /* 2075 * wire case: in two passes [XXXCDC: ugly block of code here] 2076 * 2077 * 1: holding the write lock, we create any anonymous maps that need 2078 * to be created. then we clip each map entry to the region to 2079 * be wired and increment its wiring count. 2080 * 2081 * 2: we downgrade to a read lock, and call uvm_fault_wire to fault 2082 * in the pages for any newly wired area (wired_count == 1). 2083 * 2084 * downgrading to a read lock for uvm_fault_wire avoids a possible 2085 * deadlock with another thread that may have faulted on one of 2086 * the pages to be wired (it would mark the page busy, blocking 2087 * us, then in turn block on the map lock that we hold). because 2088 * of problems in the recursive lock package, we cannot upgrade 2089 * to a write lock in vm_map_lookup. thus, any actions that 2090 * require the write lock must be done beforehand. because we 2091 * keep the read lock on the map, the copy-on-write status of the 2092 * entries we modify here cannot change. 2093 */ 2094 2095 while ((entry != &map->header) && (entry->start < end)) { 2096 if (VM_MAPENT_ISWIRED(entry) == 0) { /* not already wired? */ 2097 2098 /* 2099 * perform actions of vm_map_lookup that need the 2100 * write lock on the map: create an anonymous map 2101 * for a copy-on-write region, or an anonymous map 2102 * for a zero-fill region. (XXXCDC: submap case 2103 * ok?) 2104 */ 2105 2106 if (!UVM_ET_ISSUBMAP(entry)) { /* not submap */ 2107 if (UVM_ET_ISNEEDSCOPY(entry) && 2108 ((entry->protection & VM_PROT_WRITE) || 2109 (entry->object.uvm_obj == NULL))) { 2110 amap_copy(map, entry, M_WAITOK, TRUE, 2111 start, end); 2112 /* XXXCDC: wait OK? */ 2113 } 2114 } 2115 } 2116 UVM_MAP_CLIP_START(map, entry, start); 2117 UVM_MAP_CLIP_END(map, entry, end); 2118 entry->wired_count++; 2119 2120 /* 2121 * Check for holes 2122 */ 2123 2124 if (entry->protection == VM_PROT_NONE || 2125 (entry->end < end && 2126 (entry->next == &map->header || 2127 entry->next->start > entry->end))) { 2128 2129 /* 2130 * found one. amap creation actions do not need to 2131 * be undone, but the wired counts need to be restored. 2132 */ 2133 2134 while (entry != &map->header && entry->end > start) { 2135 entry->wired_count--; 2136 entry = entry->prev; 2137 } 2138 if ((lockflags & UVM_LK_EXIT) == 0) 2139 vm_map_unlock(map); 2140 UVMHIST_LOG(maphist,"<- done (INVALID WIRE)",0,0,0,0); 2141 return EINVAL; 2142 } 2143 entry = entry->next; 2144 } 2145 2146 /* 2147 * Pass 2. 2148 */ 2149 2150 #ifdef DIAGNOSTIC 2151 timestamp_save = map->timestamp; 2152 #endif 2153 vm_map_busy(map); 2154 vm_map_downgrade(map); 2155 2156 rv = 0; 2157 entry = start_entry; 2158 while (entry != &map->header && entry->start < end) { 2159 if (entry->wired_count == 1) { 2160 rv = uvm_fault_wire(map, entry->start, entry->end, 2161 entry->protection); 2162 if (rv) { 2163 2164 /* 2165 * wiring failed. break out of the loop. 2166 * we'll clean up the map below, once we 2167 * have a write lock again. 2168 */ 2169 2170 break; 2171 } 2172 } 2173 entry = entry->next; 2174 } 2175 2176 if (rv) { /* failed? */ 2177 2178 /* 2179 * Get back to an exclusive (write) lock. 2180 */ 2181 2182 vm_map_upgrade(map); 2183 vm_map_unbusy(map); 2184 2185 #ifdef DIAGNOSTIC 2186 if (timestamp_save != map->timestamp) 2187 panic("uvm_map_pageable: stale map"); 2188 #endif 2189 2190 /* 2191 * first drop the wiring count on all the entries 2192 * which haven't actually been wired yet. 2193 */ 2194 2195 failed_entry = entry; 2196 while (entry != &map->header && entry->start < end) { 2197 entry->wired_count--; 2198 entry = entry->next; 2199 } 2200 2201 /* 2202 * now, unwire all the entries that were successfully 2203 * wired above. 2204 */ 2205 2206 entry = start_entry; 2207 while (entry != failed_entry) { 2208 entry->wired_count--; 2209 if (VM_MAPENT_ISWIRED(entry) == 0) 2210 uvm_map_entry_unwire(map, entry); 2211 entry = entry->next; 2212 } 2213 if ((lockflags & UVM_LK_EXIT) == 0) 2214 vm_map_unlock(map); 2215 UVMHIST_LOG(maphist, "<- done (RV=%d)", rv,0,0,0); 2216 return(rv); 2217 } 2218 2219 /* We are holding a read lock here. */ 2220 if ((lockflags & UVM_LK_EXIT) == 0) { 2221 vm_map_unbusy(map); 2222 vm_map_unlock_read(map); 2223 } else { 2224 2225 /* 2226 * Get back to an exclusive (write) lock. 2227 */ 2228 2229 vm_map_upgrade(map); 2230 vm_map_unbusy(map); 2231 } 2232 2233 UVMHIST_LOG(maphist,"<- done (OK WIRE)",0,0,0,0); 2234 return 0; 2235 } 2236 2237 /* 2238 * uvm_map_pageable_all: special case of uvm_map_pageable - affects 2239 * all mapped regions. 2240 * 2241 * => map must not be locked. 2242 * => if no flags are specified, all regions are unwired. 2243 * => XXXJRT: has some of the same problems as uvm_map_pageable() above. 2244 */ 2245 2246 int 2247 uvm_map_pageable_all(map, flags, limit) 2248 struct vm_map *map; 2249 int flags; 2250 vsize_t limit; 2251 { 2252 struct vm_map_entry *entry, *failed_entry; 2253 vsize_t size; 2254 int rv; 2255 #ifdef DIAGNOSTIC 2256 u_int timestamp_save; 2257 #endif 2258 UVMHIST_FUNC("uvm_map_pageable_all"); UVMHIST_CALLED(maphist); 2259 UVMHIST_LOG(maphist,"(map=0x%x,flags=0x%x)", map, flags, 0, 0); 2260 2261 KASSERT(map->flags & VM_MAP_PAGEABLE); 2262 2263 vm_map_lock(map); 2264 2265 /* 2266 * handle wiring and unwiring separately. 2267 */ 2268 2269 if (flags == 0) { /* unwire */ 2270 2271 /* 2272 * POSIX 1003.1b -- munlockall unlocks all regions, 2273 * regardless of how many times mlockall has been called. 2274 */ 2275 2276 for (entry = map->header.next; entry != &map->header; 2277 entry = entry->next) { 2278 if (VM_MAPENT_ISWIRED(entry)) 2279 uvm_map_entry_unwire(map, entry); 2280 } 2281 vm_map_modflags(map, 0, VM_MAP_WIREFUTURE); 2282 vm_map_unlock(map); 2283 UVMHIST_LOG(maphist,"<- done (OK UNWIRE)",0,0,0,0); 2284 return 0; 2285 } 2286 2287 if (flags & MCL_FUTURE) { 2288 2289 /* 2290 * must wire all future mappings; remember this. 2291 */ 2292 2293 vm_map_modflags(map, VM_MAP_WIREFUTURE, 0); 2294 } 2295 2296 if ((flags & MCL_CURRENT) == 0) { 2297 2298 /* 2299 * no more work to do! 2300 */ 2301 2302 UVMHIST_LOG(maphist,"<- done (OK no wire)",0,0,0,0); 2303 vm_map_unlock(map); 2304 return 0; 2305 } 2306 2307 /* 2308 * wire case: in three passes [XXXCDC: ugly block of code here] 2309 * 2310 * 1: holding the write lock, count all pages mapped by non-wired 2311 * entries. if this would cause us to go over our limit, we fail. 2312 * 2313 * 2: still holding the write lock, we create any anonymous maps that 2314 * need to be created. then we increment its wiring count. 2315 * 2316 * 3: we downgrade to a read lock, and call uvm_fault_wire to fault 2317 * in the pages for any newly wired area (wired_count == 1). 2318 * 2319 * downgrading to a read lock for uvm_fault_wire avoids a possible 2320 * deadlock with another thread that may have faulted on one of 2321 * the pages to be wired (it would mark the page busy, blocking 2322 * us, then in turn block on the map lock that we hold). because 2323 * of problems in the recursive lock package, we cannot upgrade 2324 * to a write lock in vm_map_lookup. thus, any actions that 2325 * require the write lock must be done beforehand. because we 2326 * keep the read lock on the map, the copy-on-write status of the 2327 * entries we modify here cannot change. 2328 */ 2329 2330 for (size = 0, entry = map->header.next; entry != &map->header; 2331 entry = entry->next) { 2332 if (entry->protection != VM_PROT_NONE && 2333 VM_MAPENT_ISWIRED(entry) == 0) { /* not already wired? */ 2334 size += entry->end - entry->start; 2335 } 2336 } 2337 2338 if (atop(size) + uvmexp.wired > uvmexp.wiredmax) { 2339 vm_map_unlock(map); 2340 return ENOMEM; 2341 } 2342 2343 /* XXX non-pmap_wired_count case must be handled by caller */ 2344 #ifdef pmap_wired_count 2345 if (limit != 0 && 2346 (size + ptoa(pmap_wired_count(vm_map_pmap(map))) > limit)) { 2347 vm_map_unlock(map); 2348 return ENOMEM; 2349 } 2350 #endif 2351 2352 /* 2353 * Pass 2. 2354 */ 2355 2356 for (entry = map->header.next; entry != &map->header; 2357 entry = entry->next) { 2358 if (entry->protection == VM_PROT_NONE) 2359 continue; 2360 if (VM_MAPENT_ISWIRED(entry) == 0) { /* not already wired? */ 2361 2362 /* 2363 * perform actions of vm_map_lookup that need the 2364 * write lock on the map: create an anonymous map 2365 * for a copy-on-write region, or an anonymous map 2366 * for a zero-fill region. (XXXCDC: submap case 2367 * ok?) 2368 */ 2369 2370 if (!UVM_ET_ISSUBMAP(entry)) { /* not submap */ 2371 if (UVM_ET_ISNEEDSCOPY(entry) && 2372 ((entry->protection & VM_PROT_WRITE) || 2373 (entry->object.uvm_obj == NULL))) { 2374 amap_copy(map, entry, M_WAITOK, TRUE, 2375 entry->start, entry->end); 2376 /* XXXCDC: wait OK? */ 2377 } 2378 } 2379 } 2380 entry->wired_count++; 2381 } 2382 2383 /* 2384 * Pass 3. 2385 */ 2386 2387 #ifdef DIAGNOSTIC 2388 timestamp_save = map->timestamp; 2389 #endif 2390 vm_map_busy(map); 2391 vm_map_downgrade(map); 2392 2393 rv = 0; 2394 for (entry = map->header.next; entry != &map->header; 2395 entry = entry->next) { 2396 if (entry->wired_count == 1) { 2397 rv = uvm_fault_wire(map, entry->start, entry->end, 2398 entry->protection); 2399 if (rv) { 2400 2401 /* 2402 * wiring failed. break out of the loop. 2403 * we'll clean up the map below, once we 2404 * have a write lock again. 2405 */ 2406 2407 break; 2408 } 2409 } 2410 } 2411 2412 if (rv) { 2413 2414 /* 2415 * Get back an exclusive (write) lock. 2416 */ 2417 2418 vm_map_upgrade(map); 2419 vm_map_unbusy(map); 2420 2421 #ifdef DIAGNOSTIC 2422 if (timestamp_save != map->timestamp) 2423 panic("uvm_map_pageable_all: stale map"); 2424 #endif 2425 2426 /* 2427 * first drop the wiring count on all the entries 2428 * which haven't actually been wired yet. 2429 * 2430 * Skip VM_PROT_NONE entries like we did above. 2431 */ 2432 2433 failed_entry = entry; 2434 for (/* nothing */; entry != &map->header; 2435 entry = entry->next) { 2436 if (entry->protection == VM_PROT_NONE) 2437 continue; 2438 entry->wired_count--; 2439 } 2440 2441 /* 2442 * now, unwire all the entries that were successfully 2443 * wired above. 2444 * 2445 * Skip VM_PROT_NONE entries like we did above. 2446 */ 2447 2448 for (entry = map->header.next; entry != failed_entry; 2449 entry = entry->next) { 2450 if (entry->protection == VM_PROT_NONE) 2451 continue; 2452 entry->wired_count--; 2453 if (VM_MAPENT_ISWIRED(entry)) 2454 uvm_map_entry_unwire(map, entry); 2455 } 2456 vm_map_unlock(map); 2457 UVMHIST_LOG(maphist,"<- done (RV=%d)", rv,0,0,0); 2458 return (rv); 2459 } 2460 2461 /* We are holding a read lock here. */ 2462 vm_map_unbusy(map); 2463 vm_map_unlock_read(map); 2464 2465 UVMHIST_LOG(maphist,"<- done (OK WIRE)",0,0,0,0); 2466 return 0; 2467 } 2468 2469 /* 2470 * uvm_map_clean: clean out a map range 2471 * 2472 * => valid flags: 2473 * if (flags & PGO_CLEANIT): dirty pages are cleaned first 2474 * if (flags & PGO_SYNCIO): dirty pages are written synchronously 2475 * if (flags & PGO_DEACTIVATE): any cached pages are deactivated after clean 2476 * if (flags & PGO_FREE): any cached pages are freed after clean 2477 * => returns an error if any part of the specified range isn't mapped 2478 * => never a need to flush amap layer since the anonymous memory has 2479 * no permanent home, but may deactivate pages there 2480 * => called from sys_msync() and sys_madvise() 2481 * => caller must not write-lock map (read OK). 2482 * => we may sleep while cleaning if SYNCIO [with map read-locked] 2483 */ 2484 2485 int 2486 uvm_map_clean(map, start, end, flags) 2487 struct vm_map *map; 2488 vaddr_t start, end; 2489 int flags; 2490 { 2491 struct vm_map_entry *current, *entry; 2492 struct uvm_object *uobj; 2493 struct vm_amap *amap; 2494 struct vm_anon *anon; 2495 struct vm_page *pg; 2496 vaddr_t offset; 2497 vsize_t size; 2498 int error, refs; 2499 UVMHIST_FUNC("uvm_map_clean"); UVMHIST_CALLED(maphist); 2500 2501 UVMHIST_LOG(maphist,"(map=0x%x,start=0x%x,end=0x%x,flags=0x%x)", 2502 map, start, end, flags); 2503 KASSERT((flags & (PGO_FREE|PGO_DEACTIVATE)) != 2504 (PGO_FREE|PGO_DEACTIVATE)); 2505 2506 vm_map_lock_read(map); 2507 VM_MAP_RANGE_CHECK(map, start, end); 2508 if (uvm_map_lookup_entry(map, start, &entry) == FALSE) { 2509 vm_map_unlock_read(map); 2510 return EFAULT; 2511 } 2512 2513 /* 2514 * Make a first pass to check for holes. 2515 */ 2516 2517 for (current = entry; current->start < end; current = current->next) { 2518 if (UVM_ET_ISSUBMAP(current)) { 2519 vm_map_unlock_read(map); 2520 return EINVAL; 2521 } 2522 if (end <= current->end) { 2523 break; 2524 } 2525 if (current->end != current->next->start) { 2526 vm_map_unlock_read(map); 2527 return EFAULT; 2528 } 2529 } 2530 2531 error = 0; 2532 for (current = entry; start < end; current = current->next) { 2533 amap = current->aref.ar_amap; /* top layer */ 2534 uobj = current->object.uvm_obj; /* bottom layer */ 2535 KASSERT(start >= current->start); 2536 2537 /* 2538 * No amap cleaning necessary if: 2539 * 2540 * (1) There's no amap. 2541 * 2542 * (2) We're not deactivating or freeing pages. 2543 */ 2544 2545 if (amap == NULL || (flags & (PGO_DEACTIVATE|PGO_FREE)) == 0) 2546 goto flush_object; 2547 2548 amap_lock(amap); 2549 offset = start - current->start; 2550 size = MIN(end, current->end) - start; 2551 for ( ; size != 0; size -= PAGE_SIZE, offset += PAGE_SIZE) { 2552 anon = amap_lookup(¤t->aref, offset); 2553 if (anon == NULL) 2554 continue; 2555 2556 simple_lock(&anon->an_lock); 2557 2558 pg = anon->u.an_page; 2559 if (pg == NULL) { 2560 simple_unlock(&anon->an_lock); 2561 continue; 2562 } 2563 2564 switch (flags & (PGO_CLEANIT|PGO_FREE|PGO_DEACTIVATE)) { 2565 2566 /* 2567 * XXX In these first 3 cases, we always just 2568 * XXX deactivate the page. We may want to 2569 * XXX handle the different cases more 2570 * XXX specifically, in the future. 2571 */ 2572 2573 case PGO_CLEANIT|PGO_FREE: 2574 case PGO_CLEANIT|PGO_DEACTIVATE: 2575 case PGO_DEACTIVATE: 2576 deactivate_it: 2577 /* skip the page if it's loaned or wired */ 2578 if (pg->loan_count != 0 || 2579 pg->wire_count != 0) { 2580 simple_unlock(&anon->an_lock); 2581 continue; 2582 } 2583 2584 uvm_lock_pageq(); 2585 2586 /* 2587 * skip the page if it's not actually owned 2588 * by the anon (may simply be loaned to the 2589 * anon). 2590 */ 2591 2592 if ((pg->pqflags & PQ_ANON) == 0) { 2593 KASSERT(pg->uobject == NULL); 2594 uvm_unlock_pageq(); 2595 simple_unlock(&anon->an_lock); 2596 continue; 2597 } 2598 KASSERT(pg->uanon == anon); 2599 2600 /* ...and deactivate the page. */ 2601 pmap_clear_reference(pg); 2602 uvm_pagedeactivate(pg); 2603 2604 uvm_unlock_pageq(); 2605 simple_unlock(&anon->an_lock); 2606 continue; 2607 2608 case PGO_FREE: 2609 2610 /* 2611 * If there are multiple references to 2612 * the amap, just deactivate the page. 2613 */ 2614 2615 if (amap_refs(amap) > 1) 2616 goto deactivate_it; 2617 2618 /* XXX skip the page if it's wired */ 2619 if (pg->wire_count != 0) { 2620 simple_unlock(&anon->an_lock); 2621 continue; 2622 } 2623 amap_unadd(¤t->aref, offset); 2624 refs = --anon->an_ref; 2625 simple_unlock(&anon->an_lock); 2626 if (refs == 0) 2627 uvm_anfree(anon); 2628 continue; 2629 } 2630 } 2631 amap_unlock(amap); 2632 2633 flush_object: 2634 /* 2635 * flush pages if we've got a valid backing object. 2636 */ 2637 2638 offset = current->offset + (start - current->start); 2639 size = MIN(end, current->end) - start; 2640 if (uobj != NULL) { 2641 simple_lock(&uobj->vmobjlock); 2642 error = (uobj->pgops->pgo_put)(uobj, offset, 2643 offset + size, flags); 2644 } 2645 start += size; 2646 } 2647 vm_map_unlock_read(map); 2648 return (error); 2649 } 2650 2651 2652 /* 2653 * uvm_map_checkprot: check protection in map 2654 * 2655 * => must allow specified protection in a fully allocated region. 2656 * => map must be read or write locked by caller. 2657 */ 2658 2659 boolean_t 2660 uvm_map_checkprot(map, start, end, protection) 2661 struct vm_map * map; 2662 vaddr_t start, end; 2663 vm_prot_t protection; 2664 { 2665 struct vm_map_entry *entry; 2666 struct vm_map_entry *tmp_entry; 2667 2668 if (!uvm_map_lookup_entry(map, start, &tmp_entry)) { 2669 return(FALSE); 2670 } 2671 entry = tmp_entry; 2672 while (start < end) { 2673 if (entry == &map->header) { 2674 return(FALSE); 2675 } 2676 2677 /* 2678 * no holes allowed 2679 */ 2680 2681 if (start < entry->start) { 2682 return(FALSE); 2683 } 2684 2685 /* 2686 * check protection associated with entry 2687 */ 2688 2689 if ((entry->protection & protection) != protection) { 2690 return(FALSE); 2691 } 2692 start = entry->end; 2693 entry = entry->next; 2694 } 2695 return(TRUE); 2696 } 2697 2698 /* 2699 * uvmspace_alloc: allocate a vmspace structure. 2700 * 2701 * - structure includes vm_map and pmap 2702 * - XXX: no locking on this structure 2703 * - refcnt set to 1, rest must be init'd by caller 2704 */ 2705 struct vmspace * 2706 uvmspace_alloc(min, max) 2707 vaddr_t min, max; 2708 { 2709 struct vmspace *vm; 2710 UVMHIST_FUNC("uvmspace_alloc"); UVMHIST_CALLED(maphist); 2711 2712 vm = pool_get(&uvm_vmspace_pool, PR_WAITOK); 2713 uvmspace_init(vm, NULL, min, max); 2714 UVMHIST_LOG(maphist,"<- done (vm=0x%x)", vm,0,0,0); 2715 return (vm); 2716 } 2717 2718 /* 2719 * uvmspace_init: initialize a vmspace structure. 2720 * 2721 * - XXX: no locking on this structure 2722 * - refcnt set to 1, rest must me init'd by caller 2723 */ 2724 void 2725 uvmspace_init(vm, pmap, min, max) 2726 struct vmspace *vm; 2727 struct pmap *pmap; 2728 vaddr_t min, max; 2729 { 2730 UVMHIST_FUNC("uvmspace_init"); UVMHIST_CALLED(maphist); 2731 2732 memset(vm, 0, sizeof(*vm)); 2733 uvm_map_setup(&vm->vm_map, min, max, VM_MAP_PAGEABLE); 2734 if (pmap) 2735 pmap_reference(pmap); 2736 else 2737 pmap = pmap_create(); 2738 vm->vm_map.pmap = pmap; 2739 vm->vm_refcnt = 1; 2740 UVMHIST_LOG(maphist,"<- done",0,0,0,0); 2741 } 2742 2743 /* 2744 * uvmspace_share: share a vmspace between two proceses 2745 * 2746 * - XXX: no locking on vmspace 2747 * - used for vfork, threads(?) 2748 */ 2749 2750 void 2751 uvmspace_share(p1, p2) 2752 struct proc *p1, *p2; 2753 { 2754 p2->p_vmspace = p1->p_vmspace; 2755 p1->p_vmspace->vm_refcnt++; 2756 } 2757 2758 /* 2759 * uvmspace_unshare: ensure that process "p" has its own, unshared, vmspace 2760 * 2761 * - XXX: no locking on vmspace 2762 */ 2763 2764 void 2765 uvmspace_unshare(p) 2766 struct proc *p; 2767 { 2768 struct vmspace *nvm, *ovm = p->p_vmspace; 2769 2770 if (ovm->vm_refcnt == 1) 2771 /* nothing to do: vmspace isn't shared in the first place */ 2772 return; 2773 2774 /* make a new vmspace, still holding old one */ 2775 nvm = uvmspace_fork(ovm); 2776 2777 pmap_deactivate(p); /* unbind old vmspace */ 2778 p->p_vmspace = nvm; 2779 pmap_activate(p); /* switch to new vmspace */ 2780 2781 uvmspace_free(ovm); /* drop reference to old vmspace */ 2782 } 2783 2784 /* 2785 * uvmspace_exec: the process wants to exec a new program 2786 * 2787 * - XXX: no locking on vmspace 2788 */ 2789 2790 void 2791 uvmspace_exec(p, start, end) 2792 struct proc *p; 2793 vaddr_t start, end; 2794 { 2795 struct vmspace *nvm, *ovm = p->p_vmspace; 2796 struct vm_map *map = &ovm->vm_map; 2797 2798 #ifdef __sparc__ 2799 /* XXX cgd 960926: the sparc #ifdef should be a MD hook */ 2800 kill_user_windows(p); /* before stack addresses go away */ 2801 #endif 2802 2803 /* 2804 * see if more than one process is using this vmspace... 2805 */ 2806 2807 if (ovm->vm_refcnt == 1) { 2808 2809 /* 2810 * if p is the only process using its vmspace then we can safely 2811 * recycle that vmspace for the program that is being exec'd. 2812 */ 2813 2814 #ifdef SYSVSHM 2815 /* 2816 * SYSV SHM semantics require us to kill all segments on an exec 2817 */ 2818 2819 if (ovm->vm_shm) 2820 shmexit(ovm); 2821 #endif 2822 2823 /* 2824 * POSIX 1003.1b -- "lock future mappings" is revoked 2825 * when a process execs another program image. 2826 */ 2827 2828 vm_map_lock(map); 2829 vm_map_modflags(map, 0, VM_MAP_WIREFUTURE); 2830 vm_map_unlock(map); 2831 2832 /* 2833 * now unmap the old program 2834 */ 2835 2836 uvm_unmap(map, map->min_offset, map->max_offset); 2837 2838 /* 2839 * resize the map 2840 */ 2841 2842 vm_map_lock(map); 2843 map->min_offset = start; 2844 map->max_offset = end; 2845 vm_map_unlock(map); 2846 } else { 2847 2848 /* 2849 * p's vmspace is being shared, so we can't reuse it for p since 2850 * it is still being used for others. allocate a new vmspace 2851 * for p 2852 */ 2853 2854 nvm = uvmspace_alloc(start, end); 2855 2856 /* 2857 * install new vmspace and drop our ref to the old one. 2858 */ 2859 2860 pmap_deactivate(p); 2861 p->p_vmspace = nvm; 2862 pmap_activate(p); 2863 2864 uvmspace_free(ovm); 2865 } 2866 } 2867 2868 /* 2869 * uvmspace_free: free a vmspace data structure 2870 * 2871 * - XXX: no locking on vmspace 2872 */ 2873 2874 void 2875 uvmspace_free(vm) 2876 struct vmspace *vm; 2877 { 2878 struct vm_map_entry *dead_entries; 2879 UVMHIST_FUNC("uvmspace_free"); UVMHIST_CALLED(maphist); 2880 2881 UVMHIST_LOG(maphist,"(vm=0x%x) ref=%d", vm, vm->vm_refcnt,0,0); 2882 if (--vm->vm_refcnt == 0) { 2883 2884 /* 2885 * lock the map, to wait out all other references to it. delete 2886 * all of the mappings and pages they hold, then call the pmap 2887 * module to reclaim anything left. 2888 */ 2889 2890 #ifdef SYSVSHM 2891 /* Get rid of any SYSV shared memory segments. */ 2892 if (vm->vm_shm != NULL) 2893 shmexit(vm); 2894 #endif 2895 vm_map_lock(&vm->vm_map); 2896 if (vm->vm_map.nentries) { 2897 uvm_unmap_remove(&vm->vm_map, 2898 vm->vm_map.min_offset, vm->vm_map.max_offset, 2899 &dead_entries); 2900 if (dead_entries != NULL) 2901 uvm_unmap_detach(dead_entries, 0); 2902 } 2903 pmap_destroy(vm->vm_map.pmap); 2904 vm->vm_map.pmap = NULL; 2905 pool_put(&uvm_vmspace_pool, vm); 2906 } 2907 UVMHIST_LOG(maphist,"<- done", 0,0,0,0); 2908 } 2909 2910 /* 2911 * F O R K - m a i n e n t r y p o i n t 2912 */ 2913 /* 2914 * uvmspace_fork: fork a process' main map 2915 * 2916 * => create a new vmspace for child process from parent. 2917 * => parent's map must not be locked. 2918 */ 2919 2920 struct vmspace * 2921 uvmspace_fork(vm1) 2922 struct vmspace *vm1; 2923 { 2924 struct vmspace *vm2; 2925 struct vm_map *old_map = &vm1->vm_map; 2926 struct vm_map *new_map; 2927 struct vm_map_entry *old_entry; 2928 struct vm_map_entry *new_entry; 2929 pmap_t new_pmap; 2930 boolean_t protect_child; 2931 UVMHIST_FUNC("uvmspace_fork"); UVMHIST_CALLED(maphist); 2932 2933 vm_map_lock(old_map); 2934 2935 vm2 = uvmspace_alloc(old_map->min_offset, old_map->max_offset); 2936 memcpy(&vm2->vm_startcopy, &vm1->vm_startcopy, 2937 (caddr_t) (vm1 + 1) - (caddr_t) &vm1->vm_startcopy); 2938 new_map = &vm2->vm_map; /* XXX */ 2939 new_pmap = new_map->pmap; 2940 2941 old_entry = old_map->header.next; 2942 2943 /* 2944 * go entry-by-entry 2945 */ 2946 2947 while (old_entry != &old_map->header) { 2948 2949 /* 2950 * first, some sanity checks on the old entry 2951 */ 2952 2953 KASSERT(!UVM_ET_ISSUBMAP(old_entry)); 2954 KASSERT(UVM_ET_ISCOPYONWRITE(old_entry) || 2955 !UVM_ET_ISNEEDSCOPY(old_entry)); 2956 2957 switch (old_entry->inheritance) { 2958 case MAP_INHERIT_NONE: 2959 2960 /* 2961 * drop the mapping 2962 */ 2963 2964 break; 2965 2966 case MAP_INHERIT_SHARE: 2967 2968 /* 2969 * share the mapping: this means we want the old and 2970 * new entries to share amaps and backing objects. 2971 */ 2972 /* 2973 * if the old_entry needs a new amap (due to prev fork) 2974 * then we need to allocate it now so that we have 2975 * something we own to share with the new_entry. [in 2976 * other words, we need to clear needs_copy] 2977 */ 2978 2979 if (UVM_ET_ISNEEDSCOPY(old_entry)) { 2980 /* get our own amap, clears needs_copy */ 2981 amap_copy(old_map, old_entry, M_WAITOK, FALSE, 2982 0, 0); 2983 /* XXXCDC: WAITOK??? */ 2984 } 2985 2986 new_entry = uvm_mapent_alloc(new_map); 2987 /* old_entry -> new_entry */ 2988 uvm_mapent_copy(old_entry, new_entry); 2989 2990 /* new pmap has nothing wired in it */ 2991 new_entry->wired_count = 0; 2992 2993 /* 2994 * gain reference to object backing the map (can't 2995 * be a submap, already checked this case). 2996 */ 2997 2998 if (new_entry->aref.ar_amap) 2999 uvm_map_reference_amap(new_entry, AMAP_SHARED); 3000 3001 if (new_entry->object.uvm_obj && 3002 new_entry->object.uvm_obj->pgops->pgo_reference) 3003 new_entry->object.uvm_obj-> 3004 pgops->pgo_reference( 3005 new_entry->object.uvm_obj); 3006 3007 /* insert entry at end of new_map's entry list */ 3008 uvm_map_entry_link(new_map, new_map->header.prev, 3009 new_entry); 3010 3011 /* 3012 * pmap_copy the mappings: this routine is optional 3013 * but if it is there it will reduce the number of 3014 * page faults in the new proc. 3015 */ 3016 3017 pmap_copy(new_pmap, old_map->pmap, new_entry->start, 3018 (old_entry->end - old_entry->start), 3019 old_entry->start); 3020 3021 break; 3022 3023 case MAP_INHERIT_COPY: 3024 3025 /* 3026 * copy-on-write the mapping (using mmap's 3027 * MAP_PRIVATE semantics) 3028 * 3029 * allocate new_entry, adjust reference counts. 3030 * (note that new references are read-only). 3031 */ 3032 3033 new_entry = uvm_mapent_alloc(new_map); 3034 /* old_entry -> new_entry */ 3035 uvm_mapent_copy(old_entry, new_entry); 3036 3037 if (new_entry->aref.ar_amap) 3038 uvm_map_reference_amap(new_entry, 0); 3039 3040 if (new_entry->object.uvm_obj && 3041 new_entry->object.uvm_obj->pgops->pgo_reference) 3042 new_entry->object.uvm_obj->pgops->pgo_reference 3043 (new_entry->object.uvm_obj); 3044 3045 /* new pmap has nothing wired in it */ 3046 new_entry->wired_count = 0; 3047 3048 new_entry->etype |= 3049 (UVM_ET_COPYONWRITE|UVM_ET_NEEDSCOPY); 3050 uvm_map_entry_link(new_map, new_map->header.prev, 3051 new_entry); 3052 3053 /* 3054 * the new entry will need an amap. it will either 3055 * need to be copied from the old entry or created 3056 * from scratch (if the old entry does not have an 3057 * amap). can we defer this process until later 3058 * (by setting "needs_copy") or do we need to copy 3059 * the amap now? 3060 * 3061 * we must copy the amap now if any of the following 3062 * conditions hold: 3063 * 1. the old entry has an amap and that amap is 3064 * being shared. this means that the old (parent) 3065 * process is sharing the amap with another 3066 * process. if we do not clear needs_copy here 3067 * we will end up in a situation where both the 3068 * parent and child process are refering to the 3069 * same amap with "needs_copy" set. if the 3070 * parent write-faults, the fault routine will 3071 * clear "needs_copy" in the parent by allocating 3072 * a new amap. this is wrong because the 3073 * parent is supposed to be sharing the old amap 3074 * and the new amap will break that. 3075 * 3076 * 2. if the old entry has an amap and a non-zero 3077 * wire count then we are going to have to call 3078 * amap_cow_now to avoid page faults in the 3079 * parent process. since amap_cow_now requires 3080 * "needs_copy" to be clear we might as well 3081 * clear it here as well. 3082 * 3083 */ 3084 3085 if (old_entry->aref.ar_amap != NULL) { 3086 if ((amap_flags(old_entry->aref.ar_amap) & 3087 AMAP_SHARED) != 0 || 3088 VM_MAPENT_ISWIRED(old_entry)) { 3089 3090 amap_copy(new_map, new_entry, M_WAITOK, 3091 FALSE, 0, 0); 3092 /* XXXCDC: M_WAITOK ... ok? */ 3093 } 3094 } 3095 3096 /* 3097 * if the parent's entry is wired down, then the 3098 * parent process does not want page faults on 3099 * access to that memory. this means that we 3100 * cannot do copy-on-write because we can't write 3101 * protect the old entry. in this case we 3102 * resolve all copy-on-write faults now, using 3103 * amap_cow_now. note that we have already 3104 * allocated any needed amap (above). 3105 */ 3106 3107 if (VM_MAPENT_ISWIRED(old_entry)) { 3108 3109 /* 3110 * resolve all copy-on-write faults now 3111 * (note that there is nothing to do if 3112 * the old mapping does not have an amap). 3113 * XXX: is it worthwhile to bother with pmap_copy 3114 * in this case? 3115 */ 3116 if (old_entry->aref.ar_amap) 3117 amap_cow_now(new_map, new_entry); 3118 3119 } else { 3120 3121 /* 3122 * setup mappings to trigger copy-on-write faults 3123 * we must write-protect the parent if it has 3124 * an amap and it is not already "needs_copy"... 3125 * if it is already "needs_copy" then the parent 3126 * has already been write-protected by a previous 3127 * fork operation. 3128 * 3129 * if we do not write-protect the parent, then 3130 * we must be sure to write-protect the child 3131 * after the pmap_copy() operation. 3132 * 3133 * XXX: pmap_copy should have some way of telling 3134 * us that it didn't do anything so we can avoid 3135 * calling pmap_protect needlessly. 3136 */ 3137 3138 if (old_entry->aref.ar_amap) { 3139 3140 if (!UVM_ET_ISNEEDSCOPY(old_entry)) { 3141 if (old_entry->max_protection & VM_PROT_WRITE) { 3142 pmap_protect(old_map->pmap, 3143 old_entry->start, 3144 old_entry->end, 3145 old_entry->protection & 3146 ~VM_PROT_WRITE); 3147 pmap_update(old_map->pmap); 3148 } 3149 old_entry->etype |= UVM_ET_NEEDSCOPY; 3150 } 3151 3152 /* 3153 * parent must now be write-protected 3154 */ 3155 protect_child = FALSE; 3156 } else { 3157 3158 /* 3159 * we only need to protect the child if the 3160 * parent has write access. 3161 */ 3162 if (old_entry->max_protection & VM_PROT_WRITE) 3163 protect_child = TRUE; 3164 else 3165 protect_child = FALSE; 3166 3167 } 3168 3169 /* 3170 * copy the mappings 3171 * XXX: need a way to tell if this does anything 3172 */ 3173 3174 pmap_copy(new_pmap, old_map->pmap, 3175 new_entry->start, 3176 (old_entry->end - old_entry->start), 3177 old_entry->start); 3178 3179 /* 3180 * protect the child's mappings if necessary 3181 */ 3182 if (protect_child) { 3183 pmap_protect(new_pmap, new_entry->start, 3184 new_entry->end, 3185 new_entry->protection & 3186 ~VM_PROT_WRITE); 3187 pmap_update(new_pmap); 3188 } 3189 3190 } 3191 break; 3192 } /* end of switch statement */ 3193 old_entry = old_entry->next; 3194 } 3195 3196 new_map->size = old_map->size; 3197 vm_map_unlock(old_map); 3198 3199 #ifdef SYSVSHM 3200 if (vm1->vm_shm) 3201 shmfork(vm1, vm2); 3202 #endif 3203 3204 #ifdef PMAP_FORK 3205 pmap_fork(vm1->vm_map.pmap, vm2->vm_map.pmap); 3206 #endif 3207 3208 UVMHIST_LOG(maphist,"<- done",0,0,0,0); 3209 return(vm2); 3210 } 3211 3212 3213 #if defined(DDB) 3214 3215 /* 3216 * DDB hooks 3217 */ 3218 3219 /* 3220 * uvm_map_printit: actually prints the map 3221 */ 3222 3223 void 3224 uvm_map_printit(map, full, pr) 3225 struct vm_map *map; 3226 boolean_t full; 3227 void (*pr) __P((const char *, ...)); 3228 { 3229 struct vm_map_entry *entry; 3230 3231 (*pr)("MAP %p: [0x%lx->0x%lx]\n", map, map->min_offset,map->max_offset); 3232 (*pr)("\t#ent=%d, sz=%d, ref=%d, version=%d, flags=0x%x\n", 3233 map->nentries, map->size, map->ref_count, map->timestamp, 3234 map->flags); 3235 #ifdef pmap_resident_count 3236 (*pr)("\tpmap=%p(resident=%d)\n", map->pmap, 3237 pmap_resident_count(map->pmap)); 3238 #else 3239 /* XXXCDC: this should be required ... */ 3240 (*pr)("\tpmap=%p(resident=<<NOT SUPPORTED!!!>>)\n", map->pmap); 3241 #endif 3242 if (!full) 3243 return; 3244 for (entry = map->header.next; entry != &map->header; 3245 entry = entry->next) { 3246 (*pr)(" - %p: 0x%lx->0x%lx: obj=%p/0x%llx, amap=%p/%d\n", 3247 entry, entry->start, entry->end, entry->object.uvm_obj, 3248 (long long)entry->offset, entry->aref.ar_amap, 3249 entry->aref.ar_pageoff); 3250 (*pr)( 3251 "\tsubmap=%c, cow=%c, nc=%c, prot(max)=%d/%d, inh=%d, " 3252 "wc=%d, adv=%d\n", 3253 (entry->etype & UVM_ET_SUBMAP) ? 'T' : 'F', 3254 (entry->etype & UVM_ET_COPYONWRITE) ? 'T' : 'F', 3255 (entry->etype & UVM_ET_NEEDSCOPY) ? 'T' : 'F', 3256 entry->protection, entry->max_protection, 3257 entry->inheritance, entry->wired_count, entry->advice); 3258 } 3259 } 3260 3261 /* 3262 * uvm_object_printit: actually prints the object 3263 */ 3264 3265 void 3266 uvm_object_printit(uobj, full, pr) 3267 struct uvm_object *uobj; 3268 boolean_t full; 3269 void (*pr) __P((const char *, ...)); 3270 { 3271 struct vm_page *pg; 3272 int cnt = 0; 3273 3274 (*pr)("OBJECT %p: locked=%d, pgops=%p, npages=%d, ", 3275 uobj, uobj->vmobjlock.lock_data, uobj->pgops, uobj->uo_npages); 3276 if (UVM_OBJ_IS_KERN_OBJECT(uobj)) 3277 (*pr)("refs=<SYSTEM>\n"); 3278 else 3279 (*pr)("refs=%d\n", uobj->uo_refs); 3280 3281 if (!full) { 3282 return; 3283 } 3284 (*pr)(" PAGES <pg,offset>:\n "); 3285 TAILQ_FOREACH(pg, &uobj->memq, listq) { 3286 cnt++; 3287 (*pr)("<%p,0x%llx> ", pg, (long long)pg->offset); 3288 if ((cnt % 3) == 0) { 3289 (*pr)("\n "); 3290 } 3291 } 3292 if ((cnt % 3) != 0) { 3293 (*pr)("\n"); 3294 } 3295 } 3296 3297 /* 3298 * uvm_page_printit: actually print the page 3299 */ 3300 3301 static const char page_flagbits[] = 3302 "\20\1BUSY\2WANTED\3TABLED\4CLEAN\5PAGEOUT\6RELEASED\7FAKE\10RDONLY" 3303 "\11ZERO\15PAGER1"; 3304 static const char page_pqflagbits[] = 3305 "\20\1FREE\2INACTIVE\3ACTIVE\5ANON\6AOBJ"; 3306 3307 void 3308 uvm_page_printit(pg, full, pr) 3309 struct vm_page *pg; 3310 boolean_t full; 3311 void (*pr) __P((const char *, ...)); 3312 { 3313 struct vm_page *tpg; 3314 struct uvm_object *uobj; 3315 struct pglist *pgl; 3316 char pgbuf[128]; 3317 char pqbuf[128]; 3318 3319 (*pr)("PAGE %p:\n", pg); 3320 bitmask_snprintf(pg->flags, page_flagbits, pgbuf, sizeof(pgbuf)); 3321 bitmask_snprintf(pg->pqflags, page_pqflagbits, pqbuf, sizeof(pqbuf)); 3322 (*pr)(" flags=%s, pqflags=%s, wire_count=%d, pa=0x%lx\n", 3323 pgbuf, pqbuf, pg->wire_count, (long)pg->phys_addr); 3324 (*pr)(" uobject=%p, uanon=%p, offset=0x%llx loan_count=%d\n", 3325 pg->uobject, pg->uanon, (long long)pg->offset, pg->loan_count); 3326 #if defined(UVM_PAGE_TRKOWN) 3327 if (pg->flags & PG_BUSY) 3328 (*pr)(" owning process = %d, tag=%s\n", 3329 pg->owner, pg->owner_tag); 3330 else 3331 (*pr)(" page not busy, no owner\n"); 3332 #else 3333 (*pr)(" [page ownership tracking disabled]\n"); 3334 #endif 3335 3336 if (!full) 3337 return; 3338 3339 /* cross-verify object/anon */ 3340 if ((pg->pqflags & PQ_FREE) == 0) { 3341 if (pg->pqflags & PQ_ANON) { 3342 if (pg->uanon == NULL || pg->uanon->u.an_page != pg) 3343 (*pr)(" >>> ANON DOES NOT POINT HERE <<< (%p)\n", 3344 (pg->uanon) ? pg->uanon->u.an_page : NULL); 3345 else 3346 (*pr)(" anon backpointer is OK\n"); 3347 } else { 3348 uobj = pg->uobject; 3349 if (uobj) { 3350 (*pr)(" checking object list\n"); 3351 TAILQ_FOREACH(tpg, &uobj->memq, listq) { 3352 if (tpg == pg) { 3353 break; 3354 } 3355 } 3356 if (tpg) 3357 (*pr)(" page found on object list\n"); 3358 else 3359 (*pr)(" >>> PAGE NOT FOUND ON OBJECT LIST! <<<\n"); 3360 } 3361 } 3362 } 3363 3364 /* cross-verify page queue */ 3365 if (pg->pqflags & PQ_FREE) { 3366 int fl = uvm_page_lookup_freelist(pg); 3367 int color = VM_PGCOLOR_BUCKET(pg); 3368 pgl = &uvm.page_free[fl].pgfl_buckets[color].pgfl_queues[ 3369 ((pg)->flags & PG_ZERO) ? PGFL_ZEROS : PGFL_UNKNOWN]; 3370 } else if (pg->pqflags & PQ_INACTIVE) { 3371 pgl = &uvm.page_inactive; 3372 } else if (pg->pqflags & PQ_ACTIVE) { 3373 pgl = &uvm.page_active; 3374 } else { 3375 pgl = NULL; 3376 } 3377 3378 if (pgl) { 3379 (*pr)(" checking pageq list\n"); 3380 TAILQ_FOREACH(tpg, pgl, pageq) { 3381 if (tpg == pg) { 3382 break; 3383 } 3384 } 3385 if (tpg) 3386 (*pr)(" page found on pageq list\n"); 3387 else 3388 (*pr)(" >>> PAGE NOT FOUND ON PAGEQ LIST! <<<\n"); 3389 } 3390 } 3391 #endif 3392