1 /* $NetBSD: uvm_map.c,v 1.79 2000/06/27 17:29:26 mrg 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 84 #ifdef SYSVSHM 85 #include <sys/shm.h> 86 #endif 87 88 #define UVM_MAP 89 #include <uvm/uvm.h> 90 91 #ifdef DDB 92 #include <uvm/uvm_ddb.h> 93 #endif 94 95 96 struct uvm_cnt uvm_map_call, map_backmerge, map_forwmerge; 97 struct uvm_cnt uvm_mlk_call, uvm_mlk_hint; 98 99 /* 100 * pool for vmspace structures. 101 */ 102 103 struct pool uvm_vmspace_pool; 104 105 /* 106 * pool for dynamically-allocated map entries. 107 */ 108 109 struct pool uvm_map_entry_pool; 110 111 #ifdef PMAP_GROWKERNEL 112 /* 113 * This global represents the end of the kernel virtual address 114 * space. If we want to exceed this, we must grow the kernel 115 * virtual address space dynamically. 116 * 117 * Note, this variable is locked by kernel_map's lock. 118 */ 119 vaddr_t uvm_maxkaddr; 120 #endif 121 122 /* 123 * macros 124 */ 125 126 /* 127 * uvm_map_entry_link: insert entry into a map 128 * 129 * => map must be locked 130 */ 131 #define uvm_map_entry_link(map, after_where, entry) do { \ 132 (map)->nentries++; \ 133 (entry)->prev = (after_where); \ 134 (entry)->next = (after_where)->next; \ 135 (entry)->prev->next = (entry); \ 136 (entry)->next->prev = (entry); \ 137 } while (0) 138 139 /* 140 * uvm_map_entry_unlink: remove entry from a map 141 * 142 * => map must be locked 143 */ 144 #define uvm_map_entry_unlink(map, entry) do { \ 145 (map)->nentries--; \ 146 (entry)->next->prev = (entry)->prev; \ 147 (entry)->prev->next = (entry)->next; \ 148 } while (0) 149 150 /* 151 * SAVE_HINT: saves the specified entry as the hint for future lookups. 152 * 153 * => map need not be locked (protected by hint_lock). 154 */ 155 #define SAVE_HINT(map,value) do { \ 156 simple_lock(&(map)->hint_lock); \ 157 (map)->hint = (value); \ 158 simple_unlock(&(map)->hint_lock); \ 159 } while (0) 160 161 /* 162 * VM_MAP_RANGE_CHECK: check and correct range 163 * 164 * => map must at least be read locked 165 */ 166 167 #define VM_MAP_RANGE_CHECK(map, start, end) do { \ 168 if (start < vm_map_min(map)) \ 169 start = vm_map_min(map); \ 170 if (end > vm_map_max(map)) \ 171 end = vm_map_max(map); \ 172 if (start > end) \ 173 start = end; \ 174 } while (0) 175 176 /* 177 * local prototypes 178 */ 179 180 static vm_map_entry_t uvm_mapent_alloc __P((vm_map_t)); 181 static void uvm_mapent_copy __P((vm_map_entry_t,vm_map_entry_t)); 182 static void uvm_mapent_free __P((vm_map_entry_t)); 183 static void uvm_map_entry_unwire __P((vm_map_t, vm_map_entry_t)); 184 185 /* 186 * local inlines 187 */ 188 189 /* 190 * uvm_mapent_alloc: allocate a map entry 191 * 192 * => XXX: static pool for kernel map? 193 */ 194 195 static __inline vm_map_entry_t 196 uvm_mapent_alloc(map) 197 vm_map_t map; 198 { 199 vm_map_entry_t me; 200 int s; 201 UVMHIST_FUNC("uvm_mapent_alloc"); 202 UVMHIST_CALLED(maphist); 203 204 if ((map->flags & VM_MAP_INTRSAFE) == 0 && 205 map != kernel_map && kernel_map != NULL /* XXX */) { 206 me = pool_get(&uvm_map_entry_pool, PR_WAITOK); 207 me->flags = 0; 208 /* me can't be null, wait ok */ 209 } else { 210 s = splimp(); /* protect kentry_free list with splimp */ 211 simple_lock(&uvm.kentry_lock); 212 me = uvm.kentry_free; 213 if (me) uvm.kentry_free = me->next; 214 simple_unlock(&uvm.kentry_lock); 215 splx(s); 216 if (!me) 217 panic("mapent_alloc: out of static map entries, check MAX_KMAPENT"); 218 me->flags = UVM_MAP_STATIC; 219 } 220 221 UVMHIST_LOG(maphist, "<- new entry=0x%x [kentry=%d]", 222 me, ((map->flags & VM_MAP_INTRSAFE) != 0 || map == kernel_map) 223 ? TRUE : FALSE, 0, 0); 224 return(me); 225 } 226 227 /* 228 * uvm_mapent_free: free map entry 229 * 230 * => XXX: static pool for kernel map? 231 */ 232 233 static __inline void 234 uvm_mapent_free(me) 235 vm_map_entry_t me; 236 { 237 int s; 238 UVMHIST_FUNC("uvm_mapent_free"); 239 UVMHIST_CALLED(maphist); 240 UVMHIST_LOG(maphist,"<- freeing map entry=0x%x [flags=%d]", 241 me, me->flags, 0, 0); 242 if ((me->flags & UVM_MAP_STATIC) == 0) { 243 pool_put(&uvm_map_entry_pool, me); 244 } else { 245 s = splimp(); /* protect kentry_free list with splimp */ 246 simple_lock(&uvm.kentry_lock); 247 me->next = uvm.kentry_free; 248 uvm.kentry_free = me; 249 simple_unlock(&uvm.kentry_lock); 250 splx(s); 251 } 252 } 253 254 /* 255 * uvm_mapent_copy: copy a map entry, preserving flags 256 */ 257 258 static __inline void 259 uvm_mapent_copy(src, dst) 260 vm_map_entry_t src; 261 vm_map_entry_t dst; 262 { 263 264 memcpy(dst, src, ((char *)&src->uvm_map_entry_stop_copy) - ((char*)src)); 265 } 266 267 /* 268 * uvm_map_entry_unwire: unwire a map entry 269 * 270 * => map should be locked by caller 271 */ 272 273 static __inline void 274 uvm_map_entry_unwire(map, entry) 275 vm_map_t map; 276 vm_map_entry_t entry; 277 { 278 279 entry->wired_count = 0; 280 uvm_fault_unwire_locked(map, entry->start, entry->end); 281 } 282 283 /* 284 * uvm_map_init: init mapping system at boot time. note that we allocate 285 * and init the static pool of vm_map_entry_t's for the kernel here. 286 */ 287 288 void 289 uvm_map_init() 290 { 291 static struct vm_map_entry kernel_map_entry[MAX_KMAPENT]; 292 #if defined(UVMHIST) 293 static struct uvm_history_ent maphistbuf[100]; 294 static struct uvm_history_ent pdhistbuf[100]; 295 #endif 296 int lcv; 297 298 /* 299 * first, init logging system. 300 */ 301 302 UVMHIST_FUNC("uvm_map_init"); 303 UVMHIST_INIT_STATIC(maphist, maphistbuf); 304 UVMHIST_INIT_STATIC(pdhist, pdhistbuf); 305 UVMHIST_CALLED(maphist); 306 UVMHIST_LOG(maphist,"<starting uvm map system>", 0, 0, 0, 0); 307 UVMCNT_INIT(uvm_map_call, UVMCNT_CNT, 0, 308 "# uvm_map() successful calls", 0); 309 UVMCNT_INIT(map_backmerge, UVMCNT_CNT, 0, "# uvm_map() back merges", 0); 310 UVMCNT_INIT(map_forwmerge, UVMCNT_CNT, 0, "# uvm_map() missed forward", 311 0); 312 UVMCNT_INIT(uvm_mlk_call, UVMCNT_CNT, 0, "# map lookup calls", 0); 313 UVMCNT_INIT(uvm_mlk_hint, UVMCNT_CNT, 0, "# map lookup hint hits", 0); 314 315 /* 316 * now set up static pool of kernel map entrys ... 317 */ 318 319 simple_lock_init(&uvm.kentry_lock); 320 uvm.kentry_free = NULL; 321 for (lcv = 0 ; lcv < MAX_KMAPENT ; lcv++) { 322 kernel_map_entry[lcv].next = uvm.kentry_free; 323 uvm.kentry_free = &kernel_map_entry[lcv]; 324 } 325 326 /* 327 * initialize the map-related pools. 328 */ 329 pool_init(&uvm_vmspace_pool, sizeof(struct vmspace), 330 0, 0, 0, "vmsppl", 0, 331 pool_page_alloc_nointr, pool_page_free_nointr, M_VMMAP); 332 pool_init(&uvm_map_entry_pool, sizeof(struct vm_map_entry), 333 0, 0, 0, "vmmpepl", 0, 334 pool_page_alloc_nointr, pool_page_free_nointr, M_VMMAP); 335 } 336 337 /* 338 * clippers 339 */ 340 341 /* 342 * uvm_map_clip_start: ensure that the entry begins at or after 343 * the starting address, if it doesn't we split the entry. 344 * 345 * => caller should use UVM_MAP_CLIP_START macro rather than calling 346 * this directly 347 * => map must be locked by caller 348 */ 349 350 void uvm_map_clip_start(map, entry, start) 351 vm_map_t map; 352 vm_map_entry_t entry; 353 vaddr_t start; 354 { 355 vm_map_entry_t new_entry; 356 vaddr_t new_adj; 357 358 /* uvm_map_simplify_entry(map, entry); */ /* XXX */ 359 360 /* 361 * Split off the front portion. note that we must insert the new 362 * entry BEFORE this one, so that this entry has the specified 363 * starting address. 364 */ 365 366 new_entry = uvm_mapent_alloc(map); 367 uvm_mapent_copy(entry, new_entry); /* entry -> new_entry */ 368 369 new_entry->end = start; 370 new_adj = start - new_entry->start; 371 if (entry->object.uvm_obj) 372 entry->offset += new_adj; /* shift start over */ 373 entry->start = start; 374 375 if (new_entry->aref.ar_amap) { 376 amap_splitref(&new_entry->aref, &entry->aref, new_adj); 377 } 378 379 uvm_map_entry_link(map, entry->prev, new_entry); 380 381 if (UVM_ET_ISSUBMAP(entry)) { 382 /* ... unlikely to happen, but play it safe */ 383 uvm_map_reference(new_entry->object.sub_map); 384 } else { 385 if (UVM_ET_ISOBJ(entry) && 386 entry->object.uvm_obj->pgops && 387 entry->object.uvm_obj->pgops->pgo_reference) 388 entry->object.uvm_obj->pgops->pgo_reference( 389 entry->object.uvm_obj); 390 } 391 } 392 393 /* 394 * uvm_map_clip_end: ensure that the entry ends at or before 395 * the ending address, if it does't we split the reference 396 * 397 * => caller should use UVM_MAP_CLIP_END macro rather than calling 398 * this directly 399 * => map must be locked by caller 400 */ 401 402 void 403 uvm_map_clip_end(map, entry, end) 404 vm_map_t map; 405 vm_map_entry_t entry; 406 vaddr_t end; 407 { 408 vm_map_entry_t new_entry; 409 vaddr_t new_adj; /* #bytes we move start forward */ 410 411 /* 412 * Create a new entry and insert it 413 * AFTER the specified entry 414 */ 415 416 new_entry = uvm_mapent_alloc(map); 417 uvm_mapent_copy(entry, new_entry); /* entry -> new_entry */ 418 419 new_entry->start = entry->end = end; 420 new_adj = end - entry->start; 421 if (new_entry->object.uvm_obj) 422 new_entry->offset += new_adj; 423 424 if (entry->aref.ar_amap) 425 amap_splitref(&entry->aref, &new_entry->aref, new_adj); 426 427 uvm_map_entry_link(map, entry, new_entry); 428 429 if (UVM_ET_ISSUBMAP(entry)) { 430 /* ... unlikely to happen, but play it safe */ 431 uvm_map_reference(new_entry->object.sub_map); 432 } else { 433 if (UVM_ET_ISOBJ(entry) && 434 entry->object.uvm_obj->pgops && 435 entry->object.uvm_obj->pgops->pgo_reference) 436 entry->object.uvm_obj->pgops->pgo_reference( 437 entry->object.uvm_obj); 438 } 439 } 440 441 442 /* 443 * M A P - m a i n e n t r y p o i n t 444 */ 445 /* 446 * uvm_map: establish a valid mapping in a map 447 * 448 * => assume startp is page aligned. 449 * => assume size is a multiple of PAGE_SIZE. 450 * => assume sys_mmap provides enough of a "hint" to have us skip 451 * over text/data/bss area. 452 * => map must be unlocked (we will lock it) 453 * => <uobj,uoffset> value meanings (4 cases): 454 * [1] <NULL,uoffset> == uoffset is a hint for PMAP_PREFER 455 * [2] <NULL,UVM_UNKNOWN_OFFSET> == don't PMAP_PREFER 456 * [3] <uobj,uoffset> == normal mapping 457 * [4] <uobj,UVM_UNKNOWN_OFFSET> == uvm_map finds offset based on VA 458 * 459 * case [4] is for kernel mappings where we don't know the offset until 460 * we've found a virtual address. note that kernel object offsets are 461 * always relative to vm_map_min(kernel_map). 462 * => XXXCDC: need way to map in external amap? 463 */ 464 465 int 466 uvm_map(map, startp, size, uobj, uoffset, flags) 467 vm_map_t map; 468 vaddr_t *startp; /* IN/OUT */ 469 vsize_t size; 470 struct uvm_object *uobj; 471 voff_t uoffset; 472 uvm_flag_t flags; 473 { 474 vm_map_entry_t prev_entry, new_entry; 475 vm_prot_t prot = UVM_PROTECTION(flags), maxprot = 476 UVM_MAXPROTECTION(flags); 477 vm_inherit_t inherit = UVM_INHERIT(flags); 478 int advice = UVM_ADVICE(flags); 479 UVMHIST_FUNC("uvm_map"); 480 UVMHIST_CALLED(maphist); 481 482 UVMHIST_LOG(maphist, "(map=0x%x, *startp=0x%x, size=%d, flags=0x%x)", 483 map, *startp, size, flags); 484 UVMHIST_LOG(maphist, " uobj/offset 0x%x/%d", uobj, uoffset,0,0); 485 486 /* 487 * step 0: sanity check of protection code 488 */ 489 490 if ((prot & maxprot) != prot) { 491 UVMHIST_LOG(maphist, "<- prot. failure: prot=0x%x, max=0x%x", 492 prot, maxprot,0,0); 493 return(KERN_PROTECTION_FAILURE); 494 } 495 496 /* 497 * step 1: figure out where to put new VM range 498 */ 499 500 if (vm_map_lock_try(map) == FALSE) { 501 if (flags & UVM_FLAG_TRYLOCK) 502 return(KERN_FAILURE); 503 vm_map_lock(map); /* could sleep here */ 504 } 505 if ((prev_entry = uvm_map_findspace(map, *startp, size, startp, 506 uobj, uoffset, flags & UVM_FLAG_FIXED)) == NULL) { 507 UVMHIST_LOG(maphist,"<- uvm_map_findspace failed!",0,0,0,0); 508 vm_map_unlock(map); 509 return (KERN_NO_SPACE); 510 } 511 512 #ifdef PMAP_GROWKERNEL 513 { 514 /* 515 * If the kernel pmap can't map the requested space, 516 * then allocate more resources for it. 517 */ 518 if (map == kernel_map && uvm_maxkaddr < (*startp + size)) 519 uvm_maxkaddr = pmap_growkernel(*startp + size); 520 } 521 #endif 522 523 UVMCNT_INCR(uvm_map_call); 524 525 /* 526 * if uobj is null, then uoffset is either a VAC hint for PMAP_PREFER 527 * [typically from uvm_map_reserve] or it is UVM_UNKNOWN_OFFSET. in 528 * either case we want to zero it before storing it in the map entry 529 * (because it looks strange and confusing when debugging...) 530 * 531 * if uobj is not null 532 * if uoffset is not UVM_UNKNOWN_OFFSET then we have a normal mapping 533 * and we do not need to change uoffset. 534 * if uoffset is UVM_UNKNOWN_OFFSET then we need to find the offset 535 * now (based on the starting address of the map). this case is 536 * for kernel object mappings where we don't know the offset until 537 * the virtual address is found (with uvm_map_findspace). the 538 * offset is the distance we are from the start of the map. 539 */ 540 541 if (uobj == NULL) { 542 uoffset = 0; 543 } else { 544 if (uoffset == UVM_UNKNOWN_OFFSET) { 545 #ifdef DIAGNOSTIC 546 if (UVM_OBJ_IS_KERN_OBJECT(uobj) == 0) 547 panic("uvm_map: unknown offset with " 548 "non-kernel object"); 549 #endif 550 uoffset = *startp - vm_map_min(kernel_map); 551 } 552 } 553 554 /* 555 * step 2: try and insert in map by extending previous entry, if 556 * possible 557 * XXX: we don't try and pull back the next entry. might be useful 558 * for a stack, but we are currently allocating our stack in advance. 559 */ 560 561 if ((flags & UVM_FLAG_NOMERGE) == 0 && 562 prev_entry->end == *startp && prev_entry != &map->header && 563 prev_entry->object.uvm_obj == uobj) { 564 565 if (uobj && prev_entry->offset + 566 (prev_entry->end - prev_entry->start) != uoffset) 567 goto step3; 568 569 if (UVM_ET_ISSUBMAP(prev_entry)) 570 goto step3; 571 572 if (prev_entry->protection != prot || 573 prev_entry->max_protection != maxprot) 574 goto step3; 575 576 if (prev_entry->inheritance != inherit || 577 prev_entry->advice != advice) 578 goto step3; 579 580 /* wiring status must match (new area is unwired) */ 581 if (VM_MAPENT_ISWIRED(prev_entry)) 582 goto step3; 583 584 /* 585 * can't extend a shared amap. note: no need to lock amap to 586 * look at refs since we don't care about its exact value. 587 * if it is one (i.e. we have only reference) it will stay there 588 */ 589 590 if (prev_entry->aref.ar_amap && 591 amap_refs(prev_entry->aref.ar_amap) != 1) { 592 goto step3; 593 } 594 595 /* got it! */ 596 597 UVMCNT_INCR(map_backmerge); 598 UVMHIST_LOG(maphist," starting back merge", 0, 0, 0, 0); 599 600 /* 601 * drop our reference to uobj since we are extending a reference 602 * that we already have (the ref count can not drop to zero). 603 */ 604 if (uobj && uobj->pgops->pgo_detach) 605 uobj->pgops->pgo_detach(uobj); 606 607 if (prev_entry->aref.ar_amap) { 608 amap_extend(prev_entry, size); 609 } 610 611 prev_entry->end += size; 612 map->size += size; 613 614 UVMHIST_LOG(maphist,"<- done (via backmerge)!", 0, 0, 0, 0); 615 vm_map_unlock(map); 616 return (KERN_SUCCESS); 617 618 } 619 step3: 620 UVMHIST_LOG(maphist," allocating new map entry", 0, 0, 0, 0); 621 622 /* 623 * check for possible forward merge (which we don't do) and count 624 * the number of times we missed a *possible* chance to merge more 625 */ 626 627 if ((flags & UVM_FLAG_NOMERGE) == 0 && 628 prev_entry->next != &map->header && 629 prev_entry->next->start == (*startp + size)) 630 UVMCNT_INCR(map_forwmerge); 631 632 /* 633 * step 3: allocate new entry and link it in 634 */ 635 636 new_entry = uvm_mapent_alloc(map); 637 new_entry->start = *startp; 638 new_entry->end = new_entry->start + size; 639 new_entry->object.uvm_obj = uobj; 640 new_entry->offset = uoffset; 641 642 if (uobj) 643 new_entry->etype = UVM_ET_OBJ; 644 else 645 new_entry->etype = 0; 646 647 if (flags & UVM_FLAG_COPYONW) { 648 new_entry->etype |= UVM_ET_COPYONWRITE; 649 if ((flags & UVM_FLAG_OVERLAY) == 0) 650 new_entry->etype |= UVM_ET_NEEDSCOPY; 651 } 652 653 new_entry->protection = prot; 654 new_entry->max_protection = maxprot; 655 new_entry->inheritance = inherit; 656 new_entry->wired_count = 0; 657 new_entry->advice = advice; 658 if (flags & UVM_FLAG_OVERLAY) { 659 /* 660 * to_add: for BSS we overallocate a little since we 661 * are likely to extend 662 */ 663 vaddr_t to_add = (flags & UVM_FLAG_AMAPPAD) ? 664 UVM_AMAP_CHUNK << PAGE_SHIFT : 0; 665 struct vm_amap *amap = amap_alloc(size, to_add, M_WAITOK); 666 new_entry->aref.ar_pageoff = 0; 667 new_entry->aref.ar_amap = amap; 668 } else { 669 new_entry->aref.ar_pageoff = 0; 670 new_entry->aref.ar_amap = NULL; 671 } 672 673 uvm_map_entry_link(map, prev_entry, new_entry); 674 675 map->size += size; 676 677 /* 678 * Update the free space hint 679 */ 680 681 if ((map->first_free == prev_entry) && 682 (prev_entry->end >= new_entry->start)) 683 map->first_free = new_entry; 684 685 UVMHIST_LOG(maphist,"<- done!", 0, 0, 0, 0); 686 vm_map_unlock(map); 687 return(KERN_SUCCESS); 688 } 689 690 /* 691 * uvm_map_lookup_entry: find map entry at or before an address 692 * 693 * => map must at least be read-locked by caller 694 * => entry is returned in "entry" 695 * => return value is true if address is in the returned entry 696 */ 697 698 boolean_t 699 uvm_map_lookup_entry(map, address, entry) 700 vm_map_t map; 701 vaddr_t address; 702 vm_map_entry_t *entry; /* OUT */ 703 { 704 vm_map_entry_t cur; 705 vm_map_entry_t last; 706 UVMHIST_FUNC("uvm_map_lookup_entry"); 707 UVMHIST_CALLED(maphist); 708 709 UVMHIST_LOG(maphist,"(map=0x%x,addr=0x%x,ent=0x%x)", 710 map, address, entry, 0); 711 712 /* 713 * start looking either from the head of the 714 * list, or from the hint. 715 */ 716 717 simple_lock(&map->hint_lock); 718 cur = map->hint; 719 simple_unlock(&map->hint_lock); 720 721 if (cur == &map->header) 722 cur = cur->next; 723 724 UVMCNT_INCR(uvm_mlk_call); 725 if (address >= cur->start) { 726 /* 727 * go from hint to end of list. 728 * 729 * but first, make a quick check to see if 730 * we are already looking at the entry we 731 * want (which is usually the case). 732 * note also that we don't need to save the hint 733 * here... it is the same hint (unless we are 734 * at the header, in which case the hint didn't 735 * buy us anything anyway). 736 */ 737 last = &map->header; 738 if ((cur != last) && (cur->end > address)) { 739 UVMCNT_INCR(uvm_mlk_hint); 740 *entry = cur; 741 UVMHIST_LOG(maphist,"<- got it via hint (0x%x)", 742 cur, 0, 0, 0); 743 return (TRUE); 744 } 745 } else { 746 /* 747 * go from start to hint, *inclusively* 748 */ 749 last = cur->next; 750 cur = map->header.next; 751 } 752 753 /* 754 * search linearly 755 */ 756 757 while (cur != last) { 758 if (cur->end > address) { 759 if (address >= cur->start) { 760 /* 761 * save this lookup for future 762 * hints, and return 763 */ 764 765 *entry = cur; 766 SAVE_HINT(map, cur); 767 UVMHIST_LOG(maphist,"<- search got it (0x%x)", 768 cur, 0, 0, 0); 769 return (TRUE); 770 } 771 break; 772 } 773 cur = cur->next; 774 } 775 *entry = cur->prev; 776 SAVE_HINT(map, *entry); 777 UVMHIST_LOG(maphist,"<- failed!",0,0,0,0); 778 return (FALSE); 779 } 780 781 782 /* 783 * uvm_map_findspace: find "length" sized space in "map". 784 * 785 * => "hint" is a hint about where we want it, unless fixed is true 786 * (in which case we insist on using "hint"). 787 * => "result" is VA returned 788 * => uobj/uoffset are to be used to handle VAC alignment, if required 789 * => caller must at least have read-locked map 790 * => returns NULL on failure, or pointer to prev. map entry if success 791 * => note this is a cross between the old vm_map_findspace and vm_map_find 792 */ 793 794 vm_map_entry_t 795 uvm_map_findspace(map, hint, length, result, uobj, uoffset, fixed) 796 vm_map_t map; 797 vaddr_t hint; 798 vsize_t length; 799 vaddr_t *result; /* OUT */ 800 struct uvm_object *uobj; 801 voff_t uoffset; 802 boolean_t fixed; 803 { 804 vm_map_entry_t entry, next, tmp; 805 vaddr_t end; 806 UVMHIST_FUNC("uvm_map_findspace"); 807 UVMHIST_CALLED(maphist); 808 809 UVMHIST_LOG(maphist, "(map=0x%x, hint=0x%x, len=%d, fixed=%d)", 810 map, hint, length, fixed); 811 812 if (hint < map->min_offset) { /* check ranges ... */ 813 if (fixed) { 814 UVMHIST_LOG(maphist,"<- VA below map range",0,0,0,0); 815 return(NULL); 816 } 817 hint = map->min_offset; 818 } 819 if (hint > map->max_offset) { 820 UVMHIST_LOG(maphist,"<- VA 0x%x > range [0x%x->0x%x]", 821 hint, map->min_offset, map->max_offset, 0); 822 return(NULL); 823 } 824 825 /* 826 * Look for the first possible address; if there's already 827 * something at this address, we have to start after it. 828 */ 829 830 if (!fixed && hint == map->min_offset) { 831 if ((entry = map->first_free) != &map->header) 832 hint = entry->end; 833 } else { 834 if (uvm_map_lookup_entry(map, hint, &tmp)) { 835 /* "hint" address already in use ... */ 836 if (fixed) { 837 UVMHIST_LOG(maphist,"<- fixed & VA in use", 838 0, 0, 0, 0); 839 return(NULL); 840 } 841 hint = tmp->end; 842 } 843 entry = tmp; 844 } 845 846 /* 847 * Look through the rest of the map, trying to fit a new region in 848 * the gap between existing regions, or after the very last region. 849 * note: entry->end = base VA of current gap, 850 * next->start = VA of end of current gap 851 */ 852 for (;; hint = (entry = next)->end) { 853 /* 854 * Find the end of the proposed new region. Be sure we didn't 855 * go beyond the end of the map, or wrap around the address; 856 * if so, we lose. Otherwise, if this is the last entry, or 857 * if the proposed new region fits before the next entry, we 858 * win. 859 */ 860 861 #ifdef PMAP_PREFER 862 /* 863 * push hint forward as needed to avoid VAC alias problems. 864 * we only do this if a valid offset is specified. 865 */ 866 if (!fixed && uoffset != UVM_UNKNOWN_OFFSET) 867 PMAP_PREFER(uoffset, &hint); 868 #endif 869 end = hint + length; 870 if (end > map->max_offset || end < hint) { 871 UVMHIST_LOG(maphist,"<- failed (off end)", 0,0,0,0); 872 return (NULL); 873 } 874 next = entry->next; 875 if (next == &map->header || next->start >= end) 876 break; 877 if (fixed) { 878 UVMHIST_LOG(maphist,"<- fixed mapping failed", 0,0,0,0); 879 return(NULL); /* only one shot at it ... */ 880 } 881 } 882 SAVE_HINT(map, entry); 883 *result = hint; 884 UVMHIST_LOG(maphist,"<- got it! (result=0x%x)", hint, 0,0,0); 885 return (entry); 886 } 887 888 /* 889 * U N M A P - m a i n h e l p e r f u n c t i o n s 890 */ 891 892 /* 893 * uvm_unmap_remove: remove mappings from a vm_map (from "start" up to "stop") 894 * 895 * => caller must check alignment and size 896 * => map must be locked by caller 897 * => we return a list of map entries that we've remove from the map 898 * in "entry_list" 899 */ 900 901 int 902 uvm_unmap_remove(map, start, end, entry_list) 903 vm_map_t map; 904 vaddr_t start,end; 905 vm_map_entry_t *entry_list; /* OUT */ 906 { 907 vm_map_entry_t entry, first_entry, next; 908 vaddr_t len; 909 UVMHIST_FUNC("uvm_unmap_remove"); 910 UVMHIST_CALLED(maphist); 911 912 UVMHIST_LOG(maphist,"(map=0x%x, start=0x%x, end=0x%x)", 913 map, start, end, 0); 914 915 VM_MAP_RANGE_CHECK(map, start, end); 916 917 /* 918 * find first entry 919 */ 920 if (uvm_map_lookup_entry(map, start, &first_entry) == TRUE) { 921 /* clip and go... */ 922 entry = first_entry; 923 UVM_MAP_CLIP_START(map, entry, start); 924 /* critical! prevents stale hint */ 925 SAVE_HINT(map, entry->prev); 926 927 } else { 928 entry = first_entry->next; 929 } 930 931 /* 932 * Save the free space hint 933 */ 934 935 if (map->first_free->start >= start) 936 map->first_free = entry->prev; 937 938 /* 939 * note: we now re-use first_entry for a different task. we remove 940 * a number of map entries from the map and save them in a linked 941 * list headed by "first_entry". once we remove them from the map 942 * the caller should unlock the map and drop the references to the 943 * backing objects [c.f. uvm_unmap_detach]. the object is to 944 * seperate unmapping from reference dropping. why? 945 * [1] the map has to be locked for unmapping 946 * [2] the map need not be locked for reference dropping 947 * [3] dropping references may trigger pager I/O, and if we hit 948 * a pager that does synchronous I/O we may have to wait for it. 949 * [4] we would like all waiting for I/O to occur with maps unlocked 950 * so that we don't block other threads. 951 */ 952 first_entry = NULL; 953 *entry_list = NULL; /* to be safe */ 954 955 /* 956 * break up the area into map entry sized regions and unmap. note 957 * that all mappings have to be removed before we can even consider 958 * dropping references to amaps or VM objects (otherwise we could end 959 * up with a mapping to a page on the free list which would be very bad) 960 */ 961 962 while ((entry != &map->header) && (entry->start < end)) { 963 964 UVM_MAP_CLIP_END(map, entry, end); 965 next = entry->next; 966 len = entry->end - entry->start; 967 968 /* 969 * unwire before removing addresses from the pmap; otherwise 970 * unwiring will put the entries back into the pmap (XXX). 971 */ 972 973 if (VM_MAPENT_ISWIRED(entry)) 974 uvm_map_entry_unwire(map, entry); 975 976 /* 977 * special case: handle mappings to anonymous kernel objects. 978 * we want to free these pages right away... 979 */ 980 if (UVM_ET_ISOBJ(entry) && 981 UVM_OBJ_IS_KERN_OBJECT(entry->object.uvm_obj)) { 982 #ifdef DIAGNOSTIC 983 if (vm_map_pmap(map) != pmap_kernel()) 984 panic("uvm_unmap_remove: kernel object " 985 "mapped by non-kernel map"); 986 #endif 987 988 /* 989 * note: kernel object mappings are currently used in 990 * two ways: 991 * [1] "normal" mappings of pages in the kernel object 992 * [2] uvm_km_valloc'd allocations in which we 993 * pmap_enter in some non-kernel-object page 994 * (e.g. vmapbuf). 995 * 996 * for case [1], we need to remove the mapping from 997 * the pmap and then remove the page from the kernel 998 * object (because, once pages in a kernel object are 999 * unmapped they are no longer needed, unlike, say, 1000 * a vnode where you might want the data to persist 1001 * until flushed out of a queue). 1002 * 1003 * for case [2], we need to remove the mapping from 1004 * the pmap. there shouldn't be any pages at the 1005 * specified offset in the kernel object [but it 1006 * doesn't hurt to call uvm_km_pgremove just to be 1007 * safe?] 1008 * 1009 * uvm_km_pgremove currently does the following: 1010 * for pages in the kernel object in range: 1011 * - drops the swap slot 1012 * - uvm_pagefree the page 1013 * 1014 * note there is version of uvm_km_pgremove() that 1015 * is used for "intrsafe" objects. 1016 */ 1017 1018 /* 1019 * remove mappings from pmap and drop the pages 1020 * from the object. offsets are always relative 1021 * to vm_map_min(kernel_map). 1022 */ 1023 if (UVM_OBJ_IS_INTRSAFE_OBJECT(entry->object.uvm_obj)) { 1024 pmap_kremove(entry->start, len); 1025 uvm_km_pgremove_intrsafe(entry->object.uvm_obj, 1026 entry->start - vm_map_min(kernel_map), 1027 entry->end - vm_map_min(kernel_map)); 1028 } else { 1029 pmap_remove(pmap_kernel(), entry->start, 1030 entry->start + len); 1031 uvm_km_pgremove(entry->object.uvm_obj, 1032 entry->start - vm_map_min(kernel_map), 1033 entry->end - vm_map_min(kernel_map)); 1034 } 1035 1036 /* 1037 * null out kernel_object reference, we've just 1038 * dropped it 1039 */ 1040 entry->etype &= ~UVM_ET_OBJ; 1041 entry->object.uvm_obj = NULL; /* to be safe */ 1042 1043 } else { 1044 /* 1045 * remove mappings the standard way. 1046 */ 1047 pmap_remove(map->pmap, entry->start, entry->end); 1048 } 1049 1050 /* 1051 * remove entry from map and put it on our list of entries 1052 * that we've nuked. then go do next entry. 1053 */ 1054 UVMHIST_LOG(maphist, " removed map entry 0x%x", entry, 0, 0,0); 1055 uvm_map_entry_unlink(map, entry); 1056 map->size -= len; 1057 entry->next = first_entry; 1058 first_entry = entry; 1059 entry = next; /* next entry, please */ 1060 } 1061 1062 /* 1063 * now we've cleaned up the map and are ready for the caller to drop 1064 * references to the mapped objects. 1065 */ 1066 1067 *entry_list = first_entry; 1068 UVMHIST_LOG(maphist,"<- done!", 0, 0, 0, 0); 1069 return(KERN_SUCCESS); 1070 } 1071 1072 /* 1073 * uvm_unmap_detach: drop references in a chain of map entries 1074 * 1075 * => we will free the map entries as we traverse the list. 1076 */ 1077 1078 void 1079 uvm_unmap_detach(first_entry, amap_unref_flags) 1080 vm_map_entry_t first_entry; 1081 int amap_unref_flags; 1082 { 1083 vm_map_entry_t next_entry; 1084 UVMHIST_FUNC("uvm_unmap_detach"); UVMHIST_CALLED(maphist); 1085 1086 while (first_entry) { 1087 1088 #ifdef DIAGNOSTIC 1089 /* 1090 * sanity check 1091 */ 1092 /* was part of vm_map_entry_delete() */ 1093 if (VM_MAPENT_ISWIRED(first_entry)) 1094 panic("unmap: still wired!"); 1095 #endif 1096 1097 UVMHIST_LOG(maphist, 1098 " detach 0x%x: amap=0x%x, obj=0x%x, submap?=%d", 1099 first_entry, first_entry->aref.ar_amap, 1100 first_entry->object.uvm_obj, 1101 UVM_ET_ISSUBMAP(first_entry)); 1102 1103 /* 1104 * drop reference to amap, if we've got one 1105 */ 1106 1107 if (first_entry->aref.ar_amap) 1108 amap_unref(first_entry, amap_unref_flags); 1109 1110 /* 1111 * drop reference to our backing object, if we've got one 1112 */ 1113 1114 if (UVM_ET_ISSUBMAP(first_entry)) { 1115 /* ... unlikely to happen, but play it safe */ 1116 uvm_map_deallocate(first_entry->object.sub_map); 1117 } else { 1118 if (UVM_ET_ISOBJ(first_entry) && 1119 first_entry->object.uvm_obj->pgops->pgo_detach) 1120 first_entry->object.uvm_obj->pgops-> 1121 pgo_detach(first_entry->object.uvm_obj); 1122 } 1123 1124 /* 1125 * next entry 1126 */ 1127 next_entry = first_entry->next; 1128 uvm_mapent_free(first_entry); 1129 first_entry = next_entry; 1130 } 1131 1132 /* 1133 * done! 1134 */ 1135 UVMHIST_LOG(maphist, "<- done", 0,0,0,0); 1136 return; 1137 } 1138 1139 /* 1140 * E X T R A C T I O N F U N C T I O N S 1141 */ 1142 1143 /* 1144 * uvm_map_reserve: reserve space in a vm_map for future use. 1145 * 1146 * => we reserve space in a map by putting a dummy map entry in the 1147 * map (dummy means obj=NULL, amap=NULL, prot=VM_PROT_NONE) 1148 * => map should be unlocked (we will write lock it) 1149 * => we return true if we were able to reserve space 1150 * => XXXCDC: should be inline? 1151 */ 1152 1153 int 1154 uvm_map_reserve(map, size, offset, raddr) 1155 vm_map_t map; 1156 vsize_t size; 1157 vaddr_t offset; /* hint for pmap_prefer */ 1158 vaddr_t *raddr; /* IN:hint, OUT: reserved VA */ 1159 { 1160 UVMHIST_FUNC("uvm_map_reserve"); UVMHIST_CALLED(maphist); 1161 1162 UVMHIST_LOG(maphist, "(map=0x%x, size=0x%x, offset=0x%x,addr=0x%x)", 1163 map,size,offset,raddr); 1164 1165 size = round_page(size); 1166 if (*raddr < vm_map_min(map)) 1167 *raddr = vm_map_min(map); /* hint */ 1168 1169 /* 1170 * reserve some virtual space. 1171 */ 1172 1173 if (uvm_map(map, raddr, size, NULL, offset, 1174 UVM_MAPFLAG(UVM_PROT_NONE, UVM_PROT_NONE, UVM_INH_NONE, 1175 UVM_ADV_RANDOM, UVM_FLAG_NOMERGE)) != KERN_SUCCESS) { 1176 UVMHIST_LOG(maphist, "<- done (no VM)", 0,0,0,0); 1177 return (FALSE); 1178 } 1179 1180 UVMHIST_LOG(maphist, "<- done (*raddr=0x%x)", *raddr,0,0,0); 1181 return (TRUE); 1182 } 1183 1184 /* 1185 * uvm_map_replace: replace a reserved (blank) area of memory with 1186 * real mappings. 1187 * 1188 * => caller must WRITE-LOCK the map 1189 * => we return TRUE if replacement was a success 1190 * => we expect the newents chain to have nnewents entrys on it and 1191 * we expect newents->prev to point to the last entry on the list 1192 * => note newents is allowed to be NULL 1193 */ 1194 1195 int 1196 uvm_map_replace(map, start, end, newents, nnewents) 1197 struct vm_map *map; 1198 vaddr_t start, end; 1199 vm_map_entry_t newents; 1200 int nnewents; 1201 { 1202 vm_map_entry_t oldent, last; 1203 UVMHIST_FUNC("uvm_map_replace"); 1204 UVMHIST_CALLED(maphist); 1205 1206 /* 1207 * first find the blank map entry at the specified address 1208 */ 1209 1210 if (!uvm_map_lookup_entry(map, start, &oldent)) { 1211 return(FALSE); 1212 } 1213 1214 /* 1215 * check to make sure we have a proper blank entry 1216 */ 1217 1218 if (oldent->start != start || oldent->end != end || 1219 oldent->object.uvm_obj != NULL || oldent->aref.ar_amap != NULL) { 1220 return (FALSE); 1221 } 1222 1223 #ifdef DIAGNOSTIC 1224 /* 1225 * sanity check the newents chain 1226 */ 1227 { 1228 vm_map_entry_t tmpent = newents; 1229 int nent = 0; 1230 vaddr_t cur = start; 1231 1232 while (tmpent) { 1233 nent++; 1234 if (tmpent->start < cur) 1235 panic("uvm_map_replace1"); 1236 if (tmpent->start > tmpent->end || tmpent->end > end) { 1237 printf("tmpent->start=0x%lx, tmpent->end=0x%lx, end=0x%lx\n", 1238 tmpent->start, tmpent->end, end); 1239 panic("uvm_map_replace2"); 1240 } 1241 cur = tmpent->end; 1242 if (tmpent->next) { 1243 if (tmpent->next->prev != tmpent) 1244 panic("uvm_map_replace3"); 1245 } else { 1246 if (newents->prev != tmpent) 1247 panic("uvm_map_replace4"); 1248 } 1249 tmpent = tmpent->next; 1250 } 1251 if (nent != nnewents) 1252 panic("uvm_map_replace5"); 1253 } 1254 #endif 1255 1256 /* 1257 * map entry is a valid blank! replace it. (this does all the 1258 * work of map entry link/unlink...). 1259 */ 1260 1261 if (newents) { 1262 1263 last = newents->prev; /* we expect this */ 1264 1265 /* critical: flush stale hints out of map */ 1266 SAVE_HINT(map, newents); 1267 if (map->first_free == oldent) 1268 map->first_free = last; 1269 1270 last->next = oldent->next; 1271 last->next->prev = last; 1272 newents->prev = oldent->prev; 1273 newents->prev->next = newents; 1274 map->nentries = map->nentries + (nnewents - 1); 1275 1276 } else { 1277 1278 /* critical: flush stale hints out of map */ 1279 SAVE_HINT(map, oldent->prev); 1280 if (map->first_free == oldent) 1281 map->first_free = oldent->prev; 1282 1283 /* NULL list of new entries: just remove the old one */ 1284 uvm_map_entry_unlink(map, oldent); 1285 } 1286 1287 1288 /* 1289 * now we can free the old blank entry, unlock the map and return. 1290 */ 1291 1292 uvm_mapent_free(oldent); 1293 return(TRUE); 1294 } 1295 1296 /* 1297 * uvm_map_extract: extract a mapping from a map and put it somewhere 1298 * (maybe removing the old mapping) 1299 * 1300 * => maps should be unlocked (we will write lock them) 1301 * => returns 0 on success, error code otherwise 1302 * => start must be page aligned 1303 * => len must be page sized 1304 * => flags: 1305 * UVM_EXTRACT_REMOVE: remove mappings from srcmap 1306 * UVM_EXTRACT_CONTIG: abort if unmapped area (advisory only) 1307 * UVM_EXTRACT_QREF: for a temporary extraction do quick obj refs 1308 * UVM_EXTRACT_FIXPROT: set prot to maxprot as we go 1309 * >>>NOTE: if you set REMOVE, you are not allowed to use CONTIG or QREF!<<< 1310 * >>>NOTE: QREF's must be unmapped via the QREF path, thus should only 1311 * be used from within the kernel in a kernel level map <<< 1312 */ 1313 1314 int 1315 uvm_map_extract(srcmap, start, len, dstmap, dstaddrp, flags) 1316 vm_map_t srcmap, dstmap; 1317 vaddr_t start, *dstaddrp; 1318 vsize_t len; 1319 int flags; 1320 { 1321 vaddr_t dstaddr, end, newend, oldoffset, fudge, orig_fudge, 1322 oldstart; 1323 vm_map_entry_t chain, endchain, entry, orig_entry, newentry, deadentry; 1324 vm_map_entry_t oldentry; 1325 vsize_t elen; 1326 int nchain, error, copy_ok; 1327 UVMHIST_FUNC("uvm_map_extract"); UVMHIST_CALLED(maphist); 1328 UVMHIST_LOG(maphist,"(srcmap=0x%x,start=0x%x, len=0x%x", srcmap, start, 1329 len,0); 1330 UVMHIST_LOG(maphist," ...,dstmap=0x%x, flags=0x%x)", dstmap,flags,0,0); 1331 1332 #ifdef DIAGNOSTIC 1333 /* 1334 * step 0: sanity check: start must be on a page boundary, length 1335 * must be page sized. can't ask for CONTIG/QREF if you asked for 1336 * REMOVE. 1337 */ 1338 if ((start & PAGE_MASK) || (len & PAGE_MASK)) 1339 panic("uvm_map_extract1"); 1340 if (flags & UVM_EXTRACT_REMOVE) 1341 if (flags & (UVM_EXTRACT_CONTIG|UVM_EXTRACT_QREF)) 1342 panic("uvm_map_extract2"); 1343 #endif 1344 1345 1346 /* 1347 * step 1: reserve space in the target map for the extracted area 1348 */ 1349 1350 dstaddr = vm_map_min(dstmap); 1351 if (uvm_map_reserve(dstmap, len, start, &dstaddr) == FALSE) 1352 return(ENOMEM); 1353 *dstaddrp = dstaddr; /* pass address back to caller */ 1354 UVMHIST_LOG(maphist, " dstaddr=0x%x", dstaddr,0,0,0); 1355 1356 1357 /* 1358 * step 2: setup for the extraction process loop by init'ing the 1359 * map entry chain, locking src map, and looking up the first useful 1360 * entry in the map. 1361 */ 1362 1363 end = start + len; 1364 newend = dstaddr + len; 1365 chain = endchain = NULL; 1366 nchain = 0; 1367 vm_map_lock(srcmap); 1368 1369 if (uvm_map_lookup_entry(srcmap, start, &entry)) { 1370 1371 /* "start" is within an entry */ 1372 if (flags & UVM_EXTRACT_QREF) { 1373 /* 1374 * for quick references we don't clip the entry, so 1375 * the entry may map space "before" the starting 1376 * virtual address... this is the "fudge" factor 1377 * (which can be non-zero only the first time 1378 * through the "while" loop in step 3). 1379 */ 1380 fudge = start - entry->start; 1381 } else { 1382 /* 1383 * normal reference: we clip the map to fit (thus 1384 * fudge is zero) 1385 */ 1386 UVM_MAP_CLIP_START(srcmap, entry, start); 1387 SAVE_HINT(srcmap, entry->prev); 1388 fudge = 0; 1389 } 1390 1391 } else { 1392 1393 /* "start" is not within an entry ... skip to next entry */ 1394 if (flags & UVM_EXTRACT_CONTIG) { 1395 error = EINVAL; 1396 goto bad; /* definite hole here ... */ 1397 } 1398 1399 entry = entry->next; 1400 fudge = 0; 1401 } 1402 /* save values from srcmap for step 6 */ 1403 orig_entry = entry; 1404 orig_fudge = fudge; 1405 1406 1407 /* 1408 * step 3: now start looping through the map entries, extracting 1409 * as we go. 1410 */ 1411 1412 while (entry->start < end && entry != &srcmap->header) { 1413 1414 /* if we are not doing a quick reference, clip it */ 1415 if ((flags & UVM_EXTRACT_QREF) == 0) 1416 UVM_MAP_CLIP_END(srcmap, entry, end); 1417 1418 /* clear needs_copy (allow chunking) */ 1419 if (UVM_ET_ISNEEDSCOPY(entry)) { 1420 if (fudge) 1421 oldstart = entry->start; 1422 else 1423 oldstart = 0; /* XXX: gcc */ 1424 amap_copy(srcmap, entry, M_NOWAIT, TRUE, start, end); 1425 if (UVM_ET_ISNEEDSCOPY(entry)) { /* failed? */ 1426 error = ENOMEM; 1427 goto bad; 1428 } 1429 /* amap_copy could clip (during chunk)! update fudge */ 1430 if (fudge) { 1431 fudge = fudge - (entry->start - oldstart); 1432 orig_fudge = fudge; 1433 } 1434 } 1435 1436 /* calculate the offset of this from "start" */ 1437 oldoffset = (entry->start + fudge) - start; 1438 1439 /* allocate a new map entry */ 1440 newentry = uvm_mapent_alloc(dstmap); 1441 if (newentry == NULL) { 1442 error = ENOMEM; 1443 goto bad; 1444 } 1445 1446 /* set up new map entry */ 1447 newentry->next = NULL; 1448 newentry->prev = endchain; 1449 newentry->start = dstaddr + oldoffset; 1450 newentry->end = 1451 newentry->start + (entry->end - (entry->start + fudge)); 1452 if (newentry->end > newend || newentry->end < newentry->start) 1453 newentry->end = newend; 1454 newentry->object.uvm_obj = entry->object.uvm_obj; 1455 if (newentry->object.uvm_obj) { 1456 if (newentry->object.uvm_obj->pgops->pgo_reference) 1457 newentry->object.uvm_obj->pgops-> 1458 pgo_reference(newentry->object.uvm_obj); 1459 newentry->offset = entry->offset + fudge; 1460 } else { 1461 newentry->offset = 0; 1462 } 1463 newentry->etype = entry->etype; 1464 newentry->protection = (flags & UVM_EXTRACT_FIXPROT) ? 1465 entry->max_protection : entry->protection; 1466 newentry->max_protection = entry->max_protection; 1467 newentry->inheritance = entry->inheritance; 1468 newentry->wired_count = 0; 1469 newentry->aref.ar_amap = entry->aref.ar_amap; 1470 if (newentry->aref.ar_amap) { 1471 newentry->aref.ar_pageoff = 1472 entry->aref.ar_pageoff + (fudge >> PAGE_SHIFT); 1473 amap_ref(newentry, AMAP_SHARED | 1474 ((flags & UVM_EXTRACT_QREF) ? AMAP_REFALL : 0)); 1475 } else { 1476 newentry->aref.ar_pageoff = 0; 1477 } 1478 newentry->advice = entry->advice; 1479 1480 /* now link it on the chain */ 1481 nchain++; 1482 if (endchain == NULL) { 1483 chain = endchain = newentry; 1484 } else { 1485 endchain->next = newentry; 1486 endchain = newentry; 1487 } 1488 1489 /* end of 'while' loop! */ 1490 if ((flags & UVM_EXTRACT_CONTIG) && entry->end < end && 1491 (entry->next == &srcmap->header || 1492 entry->next->start != entry->end)) { 1493 error = EINVAL; 1494 goto bad; 1495 } 1496 entry = entry->next; 1497 fudge = 0; 1498 } 1499 1500 1501 /* 1502 * step 4: close off chain (in format expected by uvm_map_replace) 1503 */ 1504 1505 if (chain) 1506 chain->prev = endchain; 1507 1508 1509 /* 1510 * step 5: attempt to lock the dest map so we can pmap_copy. 1511 * note usage of copy_ok: 1512 * 1 => dstmap locked, pmap_copy ok, and we "replace" here (step 5) 1513 * 0 => dstmap unlocked, NO pmap_copy, and we will "replace" in step 7 1514 */ 1515 1516 if (srcmap == dstmap || vm_map_lock_try(dstmap) == TRUE) { 1517 1518 copy_ok = 1; 1519 if (!uvm_map_replace(dstmap, dstaddr, dstaddr+len, chain, 1520 nchain)) { 1521 if (srcmap != dstmap) 1522 vm_map_unlock(dstmap); 1523 error = EIO; 1524 goto bad; 1525 } 1526 1527 } else { 1528 1529 copy_ok = 0; 1530 /* replace defered until step 7 */ 1531 1532 } 1533 1534 1535 /* 1536 * step 6: traverse the srcmap a second time to do the following: 1537 * - if we got a lock on the dstmap do pmap_copy 1538 * - if UVM_EXTRACT_REMOVE remove the entries 1539 * we make use of orig_entry and orig_fudge (saved in step 2) 1540 */ 1541 1542 if (copy_ok || (flags & UVM_EXTRACT_REMOVE)) { 1543 1544 /* purge possible stale hints from srcmap */ 1545 if (flags & UVM_EXTRACT_REMOVE) { 1546 SAVE_HINT(srcmap, orig_entry->prev); 1547 if (srcmap->first_free->start >= start) 1548 srcmap->first_free = orig_entry->prev; 1549 } 1550 1551 entry = orig_entry; 1552 fudge = orig_fudge; 1553 deadentry = NULL; /* for UVM_EXTRACT_REMOVE */ 1554 1555 while (entry->start < end && entry != &srcmap->header) { 1556 1557 if (copy_ok) { 1558 oldoffset = (entry->start + fudge) - start; 1559 elen = min(end, entry->end) - 1560 (entry->start + fudge); 1561 pmap_copy(dstmap->pmap, srcmap->pmap, 1562 dstaddr + oldoffset, elen, 1563 entry->start + fudge); 1564 } 1565 1566 /* we advance "entry" in the following if statement */ 1567 if (flags & UVM_EXTRACT_REMOVE) { 1568 pmap_remove(srcmap->pmap, entry->start, 1569 entry->end); 1570 oldentry = entry; /* save entry */ 1571 entry = entry->next; /* advance */ 1572 uvm_map_entry_unlink(srcmap, oldentry); 1573 /* add to dead list */ 1574 oldentry->next = deadentry; 1575 deadentry = oldentry; 1576 } else { 1577 entry = entry->next; /* advance */ 1578 } 1579 1580 /* end of 'while' loop */ 1581 fudge = 0; 1582 } 1583 1584 /* 1585 * unlock dstmap. we will dispose of deadentry in 1586 * step 7 if needed 1587 */ 1588 if (copy_ok && srcmap != dstmap) 1589 vm_map_unlock(dstmap); 1590 1591 } 1592 else 1593 deadentry = NULL; /* XXX: gcc */ 1594 1595 /* 1596 * step 7: we are done with the source map, unlock. if copy_ok 1597 * is 0 then we have not replaced the dummy mapping in dstmap yet 1598 * and we need to do so now. 1599 */ 1600 1601 vm_map_unlock(srcmap); 1602 if ((flags & UVM_EXTRACT_REMOVE) && deadentry) 1603 uvm_unmap_detach(deadentry, 0); /* dispose of old entries */ 1604 1605 /* now do the replacement if we didn't do it in step 5 */ 1606 if (copy_ok == 0) { 1607 vm_map_lock(dstmap); 1608 error = uvm_map_replace(dstmap, dstaddr, dstaddr+len, chain, 1609 nchain); 1610 vm_map_unlock(dstmap); 1611 1612 if (error == FALSE) { 1613 error = EIO; 1614 goto bad2; 1615 } 1616 } 1617 1618 /* 1619 * done! 1620 */ 1621 return(0); 1622 1623 /* 1624 * bad: failure recovery 1625 */ 1626 bad: 1627 vm_map_unlock(srcmap); 1628 bad2: /* src already unlocked */ 1629 if (chain) 1630 uvm_unmap_detach(chain, 1631 (flags & UVM_EXTRACT_QREF) ? AMAP_REFALL : 0); 1632 uvm_unmap(dstmap, dstaddr, dstaddr+len); /* ??? */ 1633 return(error); 1634 } 1635 1636 /* end of extraction functions */ 1637 1638 /* 1639 * uvm_map_submap: punch down part of a map into a submap 1640 * 1641 * => only the kernel_map is allowed to be submapped 1642 * => the purpose of submapping is to break up the locking granularity 1643 * of a larger map 1644 * => the range specified must have been mapped previously with a uvm_map() 1645 * call [with uobj==NULL] to create a blank map entry in the main map. 1646 * [And it had better still be blank!] 1647 * => maps which contain submaps should never be copied or forked. 1648 * => to remove a submap, use uvm_unmap() on the main map 1649 * and then uvm_map_deallocate() the submap. 1650 * => main map must be unlocked. 1651 * => submap must have been init'd and have a zero reference count. 1652 * [need not be locked as we don't actually reference it] 1653 */ 1654 1655 int 1656 uvm_map_submap(map, start, end, submap) 1657 vm_map_t map, submap; 1658 vaddr_t start, end; 1659 { 1660 vm_map_entry_t entry; 1661 int result; 1662 UVMHIST_FUNC("uvm_map_submap"); UVMHIST_CALLED(maphist); 1663 1664 vm_map_lock(map); 1665 1666 VM_MAP_RANGE_CHECK(map, start, end); 1667 1668 if (uvm_map_lookup_entry(map, start, &entry)) { 1669 UVM_MAP_CLIP_START(map, entry, start); 1670 UVM_MAP_CLIP_END(map, entry, end); /* to be safe */ 1671 } 1672 else { 1673 entry = NULL; 1674 } 1675 1676 if (entry != NULL && 1677 entry->start == start && entry->end == end && 1678 entry->object.uvm_obj == NULL && entry->aref.ar_amap == NULL && 1679 !UVM_ET_ISCOPYONWRITE(entry) && !UVM_ET_ISNEEDSCOPY(entry)) { 1680 1681 /* 1682 * doit! 1683 */ 1684 entry->etype |= UVM_ET_SUBMAP; 1685 entry->object.sub_map = submap; 1686 entry->offset = 0; 1687 uvm_map_reference(submap); 1688 result = KERN_SUCCESS; 1689 } else { 1690 result = KERN_INVALID_ARGUMENT; 1691 } 1692 vm_map_unlock(map); 1693 1694 return(result); 1695 } 1696 1697 1698 /* 1699 * uvm_map_protect: change map protection 1700 * 1701 * => set_max means set max_protection. 1702 * => map must be unlocked. 1703 */ 1704 1705 #define MASK(entry) (UVM_ET_ISCOPYONWRITE(entry) ? \ 1706 ~VM_PROT_WRITE : VM_PROT_ALL) 1707 #define max(a,b) ((a) > (b) ? (a) : (b)) 1708 1709 int 1710 uvm_map_protect(map, start, end, new_prot, set_max) 1711 vm_map_t map; 1712 vaddr_t start, end; 1713 vm_prot_t new_prot; 1714 boolean_t set_max; 1715 { 1716 vm_map_entry_t current, entry; 1717 int rv = KERN_SUCCESS; 1718 UVMHIST_FUNC("uvm_map_protect"); UVMHIST_CALLED(maphist); 1719 UVMHIST_LOG(maphist,"(map=0x%x,start=0x%x,end=0x%x,new_prot=0x%x)", 1720 map, start, end, new_prot); 1721 1722 vm_map_lock(map); 1723 1724 VM_MAP_RANGE_CHECK(map, start, end); 1725 1726 if (uvm_map_lookup_entry(map, start, &entry)) { 1727 UVM_MAP_CLIP_START(map, entry, start); 1728 } else { 1729 entry = entry->next; 1730 } 1731 1732 /* 1733 * make a first pass to check for protection violations. 1734 */ 1735 1736 current = entry; 1737 while ((current != &map->header) && (current->start < end)) { 1738 if (UVM_ET_ISSUBMAP(current)) { 1739 rv = KERN_INVALID_ARGUMENT; 1740 goto out; 1741 } 1742 if ((new_prot & current->max_protection) != new_prot) { 1743 rv = KERN_PROTECTION_FAILURE; 1744 goto out; 1745 } 1746 current = current->next; 1747 } 1748 1749 /* go back and fix up protections (no need to clip this time). */ 1750 1751 current = entry; 1752 1753 while ((current != &map->header) && (current->start < end)) { 1754 vm_prot_t old_prot; 1755 1756 UVM_MAP_CLIP_END(map, current, end); 1757 1758 old_prot = current->protection; 1759 if (set_max) 1760 current->protection = 1761 (current->max_protection = new_prot) & old_prot; 1762 else 1763 current->protection = new_prot; 1764 1765 /* 1766 * update physical map if necessary. worry about copy-on-write 1767 * here -- CHECK THIS XXX 1768 */ 1769 1770 if (current->protection != old_prot) { 1771 /* update pmap! */ 1772 pmap_protect(map->pmap, current->start, current->end, 1773 current->protection & MASK(entry)); 1774 } 1775 1776 /* 1777 * If the map is configured to lock any future mappings, 1778 * wire this entry now if the old protection was VM_PROT_NONE 1779 * and the new protection is not VM_PROT_NONE. 1780 */ 1781 1782 if ((map->flags & VM_MAP_WIREFUTURE) != 0 && 1783 VM_MAPENT_ISWIRED(entry) == 0 && 1784 old_prot == VM_PROT_NONE && 1785 new_prot != VM_PROT_NONE) { 1786 if (uvm_map_pageable(map, entry->start, 1787 entry->end, FALSE, 1788 UVM_LK_ENTER|UVM_LK_EXIT) != KERN_SUCCESS) { 1789 /* 1790 * If locking the entry fails, remember the 1791 * error if it's the first one. Note we 1792 * still continue setting the protection in 1793 * the map, but will return the resource 1794 * shortage condition regardless. 1795 * 1796 * XXX Ignore what the actual error is, 1797 * XXX just call it a resource shortage 1798 * XXX so that it doesn't get confused 1799 * XXX what uvm_map_protect() itself would 1800 * XXX normally return. 1801 */ 1802 rv = KERN_RESOURCE_SHORTAGE; 1803 } 1804 } 1805 1806 current = current->next; 1807 } 1808 1809 out: 1810 vm_map_unlock(map); 1811 UVMHIST_LOG(maphist, "<- done, rv=%d",rv,0,0,0); 1812 return (rv); 1813 } 1814 1815 #undef max 1816 #undef MASK 1817 1818 /* 1819 * uvm_map_inherit: set inheritance code for range of addrs in map. 1820 * 1821 * => map must be unlocked 1822 * => note that the inherit code is used during a "fork". see fork 1823 * code for details. 1824 */ 1825 1826 int 1827 uvm_map_inherit(map, start, end, new_inheritance) 1828 vm_map_t map; 1829 vaddr_t start; 1830 vaddr_t end; 1831 vm_inherit_t new_inheritance; 1832 { 1833 vm_map_entry_t entry, temp_entry; 1834 UVMHIST_FUNC("uvm_map_inherit"); UVMHIST_CALLED(maphist); 1835 UVMHIST_LOG(maphist,"(map=0x%x,start=0x%x,end=0x%x,new_inh=0x%x)", 1836 map, start, end, new_inheritance); 1837 1838 switch (new_inheritance) { 1839 case VM_INHERIT_NONE: 1840 case VM_INHERIT_COPY: 1841 case VM_INHERIT_SHARE: 1842 break; 1843 default: 1844 UVMHIST_LOG(maphist,"<- done (INVALID ARG)",0,0,0,0); 1845 return (KERN_INVALID_ARGUMENT); 1846 } 1847 1848 vm_map_lock(map); 1849 1850 VM_MAP_RANGE_CHECK(map, start, end); 1851 1852 if (uvm_map_lookup_entry(map, start, &temp_entry)) { 1853 entry = temp_entry; 1854 UVM_MAP_CLIP_START(map, entry, start); 1855 } else { 1856 entry = temp_entry->next; 1857 } 1858 1859 while ((entry != &map->header) && (entry->start < end)) { 1860 UVM_MAP_CLIP_END(map, entry, end); 1861 1862 entry->inheritance = new_inheritance; 1863 1864 entry = entry->next; 1865 } 1866 1867 vm_map_unlock(map); 1868 UVMHIST_LOG(maphist,"<- done (OK)",0,0,0,0); 1869 return(KERN_SUCCESS); 1870 } 1871 1872 /* 1873 * uvm_map_advice: set advice code for range of addrs in map. 1874 * 1875 * => map must be unlocked 1876 */ 1877 1878 int 1879 uvm_map_advice(map, start, end, new_advice) 1880 vm_map_t map; 1881 vaddr_t start; 1882 vaddr_t end; 1883 int new_advice; 1884 { 1885 vm_map_entry_t entry, temp_entry; 1886 UVMHIST_FUNC("uvm_map_advice"); UVMHIST_CALLED(maphist); 1887 UVMHIST_LOG(maphist,"(map=0x%x,start=0x%x,end=0x%x,new_adv=0x%x)", 1888 map, start, end, new_advice); 1889 1890 vm_map_lock(map); 1891 1892 VM_MAP_RANGE_CHECK(map, start, end); 1893 1894 if (uvm_map_lookup_entry(map, start, &temp_entry)) { 1895 entry = temp_entry; 1896 UVM_MAP_CLIP_START(map, entry, start); 1897 } else { 1898 entry = temp_entry->next; 1899 } 1900 1901 /* 1902 * XXXJRT: disallow holes? 1903 */ 1904 1905 while ((entry != &map->header) && (entry->start < end)) { 1906 UVM_MAP_CLIP_END(map, entry, end); 1907 1908 switch (new_advice) { 1909 case MADV_NORMAL: 1910 case MADV_RANDOM: 1911 case MADV_SEQUENTIAL: 1912 /* nothing special here */ 1913 break; 1914 1915 default: 1916 vm_map_unlock(map); 1917 UVMHIST_LOG(maphist,"<- done (INVALID ARG)",0,0,0,0); 1918 return (KERN_INVALID_ARGUMENT); 1919 } 1920 1921 1922 entry->advice = new_advice; 1923 1924 entry = entry->next; 1925 } 1926 1927 vm_map_unlock(map); 1928 UVMHIST_LOG(maphist,"<- done (OK)",0,0,0,0); 1929 return (KERN_SUCCESS); 1930 } 1931 1932 /* 1933 * uvm_map_pageable: sets the pageability of a range in a map. 1934 * 1935 * => wires map entries. should not be used for transient page locking. 1936 * for that, use uvm_fault_wire()/uvm_fault_unwire() (see uvm_vslock()). 1937 * => regions sepcified as not pageable require lock-down (wired) memory 1938 * and page tables. 1939 * => map must never be read-locked 1940 * => if islocked is TRUE, map is already write-locked 1941 * => we always unlock the map, since we must downgrade to a read-lock 1942 * to call uvm_fault_wire() 1943 * => XXXCDC: check this and try and clean it up. 1944 */ 1945 1946 int 1947 uvm_map_pageable(map, start, end, new_pageable, lockflags) 1948 vm_map_t map; 1949 vaddr_t start, end; 1950 boolean_t new_pageable; 1951 int lockflags; 1952 { 1953 vm_map_entry_t entry, start_entry, failed_entry; 1954 int rv; 1955 #ifdef DIAGNOSTIC 1956 u_int timestamp_save; 1957 #endif 1958 UVMHIST_FUNC("uvm_map_pageable"); UVMHIST_CALLED(maphist); 1959 UVMHIST_LOG(maphist,"(map=0x%x,start=0x%x,end=0x%x,new_pageable=0x%x)", 1960 map, start, end, new_pageable); 1961 1962 #ifdef DIAGNOSTIC 1963 if ((map->flags & VM_MAP_PAGEABLE) == 0) 1964 panic("uvm_map_pageable: map %p not pageable", map); 1965 #endif 1966 1967 if ((lockflags & UVM_LK_ENTER) == 0) 1968 vm_map_lock(map); 1969 1970 VM_MAP_RANGE_CHECK(map, start, end); 1971 1972 /* 1973 * only one pageability change may take place at one time, since 1974 * uvm_fault_wire assumes it will be called only once for each 1975 * wiring/unwiring. therefore, we have to make sure we're actually 1976 * changing the pageability for the entire region. we do so before 1977 * making any changes. 1978 */ 1979 1980 if (uvm_map_lookup_entry(map, start, &start_entry) == FALSE) { 1981 if ((lockflags & UVM_LK_EXIT) == 0) 1982 vm_map_unlock(map); 1983 1984 UVMHIST_LOG(maphist,"<- done (INVALID ARG)",0,0,0,0); 1985 return (KERN_INVALID_ADDRESS); 1986 } 1987 entry = start_entry; 1988 1989 /* 1990 * handle wiring and unwiring seperately. 1991 */ 1992 1993 if (new_pageable) { /* unwire */ 1994 UVM_MAP_CLIP_START(map, entry, start); 1995 /* 1996 * unwiring. first ensure that the range to be unwired is 1997 * really wired down and that there are no holes. 1998 */ 1999 while ((entry != &map->header) && (entry->start < end)) { 2000 if (entry->wired_count == 0 || 2001 (entry->end < end && 2002 (entry->next == &map->header || 2003 entry->next->start > entry->end))) { 2004 if ((lockflags & UVM_LK_EXIT) == 0) 2005 vm_map_unlock(map); 2006 UVMHIST_LOG(maphist, 2007 "<- done (INVALID UNWIRE ARG)",0,0,0,0); 2008 return (KERN_INVALID_ARGUMENT); 2009 } 2010 entry = entry->next; 2011 } 2012 2013 /* 2014 * POSIX 1003.1b - a single munlock call unlocks a region, 2015 * regardless of the number of mlock calls made on that 2016 * region. 2017 */ 2018 entry = start_entry; 2019 while ((entry != &map->header) && (entry->start < end)) { 2020 UVM_MAP_CLIP_END(map, entry, end); 2021 if (VM_MAPENT_ISWIRED(entry)) 2022 uvm_map_entry_unwire(map, entry); 2023 entry = entry->next; 2024 } 2025 if ((lockflags & UVM_LK_EXIT) == 0) 2026 vm_map_unlock(map); 2027 UVMHIST_LOG(maphist,"<- done (OK UNWIRE)",0,0,0,0); 2028 return(KERN_SUCCESS); 2029 2030 /* 2031 * end of unwire case! 2032 */ 2033 } 2034 2035 /* 2036 * wire case: in two passes [XXXCDC: ugly block of code here] 2037 * 2038 * 1: holding the write lock, we create any anonymous maps that need 2039 * to be created. then we clip each map entry to the region to 2040 * be wired and increment its wiring count. 2041 * 2042 * 2: we downgrade to a read lock, and call uvm_fault_wire to fault 2043 * in the pages for any newly wired area (wired_count == 1). 2044 * 2045 * downgrading to a read lock for uvm_fault_wire avoids a possible 2046 * deadlock with another thread that may have faulted on one of 2047 * the pages to be wired (it would mark the page busy, blocking 2048 * us, then in turn block on the map lock that we hold). because 2049 * of problems in the recursive lock package, we cannot upgrade 2050 * to a write lock in vm_map_lookup. thus, any actions that 2051 * require the write lock must be done beforehand. because we 2052 * keep the read lock on the map, the copy-on-write status of the 2053 * entries we modify here cannot change. 2054 */ 2055 2056 while ((entry != &map->header) && (entry->start < end)) { 2057 if (VM_MAPENT_ISWIRED(entry) == 0) { /* not already wired? */ 2058 /* 2059 * perform actions of vm_map_lookup that need the 2060 * write lock on the map: create an anonymous map 2061 * for a copy-on-write region, or an anonymous map 2062 * for a zero-fill region. (XXXCDC: submap case 2063 * ok?) 2064 */ 2065 if (!UVM_ET_ISSUBMAP(entry)) { /* not submap */ 2066 if (UVM_ET_ISNEEDSCOPY(entry) && 2067 ((entry->protection & VM_PROT_WRITE) || 2068 (entry->object.uvm_obj == NULL))) { 2069 amap_copy(map, entry, M_WAITOK, TRUE, 2070 start, end); 2071 /* XXXCDC: wait OK? */ 2072 } 2073 } 2074 } 2075 UVM_MAP_CLIP_START(map, entry, start); 2076 UVM_MAP_CLIP_END(map, entry, end); 2077 entry->wired_count++; 2078 2079 /* 2080 * Check for holes 2081 */ 2082 if (entry->protection == VM_PROT_NONE || 2083 (entry->end < end && 2084 (entry->next == &map->header || 2085 entry->next->start > entry->end))) { 2086 /* 2087 * found one. amap creation actions do not need to 2088 * be undone, but the wired counts need to be restored. 2089 */ 2090 while (entry != &map->header && entry->end > start) { 2091 entry->wired_count--; 2092 entry = entry->prev; 2093 } 2094 if ((lockflags & UVM_LK_EXIT) == 0) 2095 vm_map_unlock(map); 2096 UVMHIST_LOG(maphist,"<- done (INVALID WIRE)",0,0,0,0); 2097 return (KERN_INVALID_ARGUMENT); 2098 } 2099 entry = entry->next; 2100 } 2101 2102 /* 2103 * Pass 2. 2104 */ 2105 2106 #ifdef DIAGNOSTIC 2107 timestamp_save = map->timestamp; 2108 #endif 2109 vm_map_busy(map); 2110 vm_map_downgrade(map); 2111 2112 rv = 0; 2113 entry = start_entry; 2114 while (entry != &map->header && entry->start < end) { 2115 if (entry->wired_count == 1) { 2116 rv = uvm_fault_wire(map, entry->start, entry->end, 2117 entry->protection); 2118 if (rv) { 2119 /* 2120 * wiring failed. break out of the loop. 2121 * we'll clean up the map below, once we 2122 * have a write lock again. 2123 */ 2124 break; 2125 } 2126 } 2127 entry = entry->next; 2128 } 2129 2130 if (rv) { /* failed? */ 2131 /* 2132 * Get back to an exclusive (write) lock. 2133 */ 2134 vm_map_upgrade(map); 2135 vm_map_unbusy(map); 2136 2137 #ifdef DIAGNOSTIC 2138 if (timestamp_save != map->timestamp) 2139 panic("uvm_map_pageable: stale map"); 2140 #endif 2141 2142 /* 2143 * first drop the wiring count on all the entries 2144 * which haven't actually been wired yet. 2145 */ 2146 failed_entry = entry; 2147 while (entry != &map->header && entry->start < end) { 2148 entry->wired_count--; 2149 entry = entry->next; 2150 } 2151 2152 /* 2153 * now, unwire all the entries that were successfully 2154 * wired above. 2155 */ 2156 entry = start_entry; 2157 while (entry != failed_entry) { 2158 entry->wired_count--; 2159 if (VM_MAPENT_ISWIRED(entry) == 0) 2160 uvm_map_entry_unwire(map, entry); 2161 entry = entry->next; 2162 } 2163 if ((lockflags & UVM_LK_EXIT) == 0) 2164 vm_map_unlock(map); 2165 UVMHIST_LOG(maphist, "<- done (RV=%d)", rv,0,0,0); 2166 return(rv); 2167 } 2168 2169 /* We are holding a read lock here. */ 2170 if ((lockflags & UVM_LK_EXIT) == 0) { 2171 vm_map_unbusy(map); 2172 vm_map_unlock_read(map); 2173 } else { 2174 /* 2175 * Get back to an exclusive (write) lock. 2176 */ 2177 vm_map_upgrade(map); 2178 vm_map_unbusy(map); 2179 } 2180 2181 UVMHIST_LOG(maphist,"<- done (OK WIRE)",0,0,0,0); 2182 return(KERN_SUCCESS); 2183 } 2184 2185 /* 2186 * uvm_map_pageable_all: special case of uvm_map_pageable - affects 2187 * all mapped regions. 2188 * 2189 * => map must not be locked. 2190 * => if no flags are specified, all regions are unwired. 2191 * => XXXJRT: has some of the same problems as uvm_map_pageable() above. 2192 */ 2193 2194 int 2195 uvm_map_pageable_all(map, flags, limit) 2196 vm_map_t map; 2197 int flags; 2198 vsize_t limit; 2199 { 2200 vm_map_entry_t entry, failed_entry; 2201 vsize_t size; 2202 int rv; 2203 #ifdef DIAGNOSTIC 2204 u_int timestamp_save; 2205 #endif 2206 UVMHIST_FUNC("uvm_map_pageable_all"); UVMHIST_CALLED(maphist); 2207 UVMHIST_LOG(maphist,"(map=0x%x,flags=0x%x)", map, flags, 0, 0); 2208 2209 #ifdef DIAGNOSTIC 2210 if ((map->flags & VM_MAP_PAGEABLE) == 0) 2211 panic("uvm_map_pageable_all: map %p not pageable", map); 2212 #endif 2213 2214 vm_map_lock(map); 2215 2216 /* 2217 * handle wiring and unwiring separately. 2218 */ 2219 2220 if (flags == 0) { /* unwire */ 2221 /* 2222 * POSIX 1003.1b -- munlockall unlocks all regions, 2223 * regardless of how many times mlockall has been called. 2224 */ 2225 for (entry = map->header.next; entry != &map->header; 2226 entry = entry->next) { 2227 if (VM_MAPENT_ISWIRED(entry)) 2228 uvm_map_entry_unwire(map, entry); 2229 } 2230 vm_map_modflags(map, 0, VM_MAP_WIREFUTURE); 2231 vm_map_unlock(map); 2232 UVMHIST_LOG(maphist,"<- done (OK UNWIRE)",0,0,0,0); 2233 return (KERN_SUCCESS); 2234 2235 /* 2236 * end of unwire case! 2237 */ 2238 } 2239 2240 if (flags & MCL_FUTURE) { 2241 /* 2242 * must wire all future mappings; remember this. 2243 */ 2244 vm_map_modflags(map, VM_MAP_WIREFUTURE, 0); 2245 } 2246 2247 if ((flags & MCL_CURRENT) == 0) { 2248 /* 2249 * no more work to do! 2250 */ 2251 UVMHIST_LOG(maphist,"<- done (OK no wire)",0,0,0,0); 2252 vm_map_unlock(map); 2253 return (KERN_SUCCESS); 2254 } 2255 2256 /* 2257 * wire case: in three passes [XXXCDC: ugly block of code here] 2258 * 2259 * 1: holding the write lock, count all pages mapped by non-wired 2260 * entries. if this would cause us to go over our limit, we fail. 2261 * 2262 * 2: still holding the write lock, we create any anonymous maps that 2263 * need to be created. then we increment its wiring count. 2264 * 2265 * 3: we downgrade to a read lock, and call uvm_fault_wire to fault 2266 * in the pages for any newly wired area (wired_count == 1). 2267 * 2268 * downgrading to a read lock for uvm_fault_wire avoids a possible 2269 * deadlock with another thread that may have faulted on one of 2270 * the pages to be wired (it would mark the page busy, blocking 2271 * us, then in turn block on the map lock that we hold). because 2272 * of problems in the recursive lock package, we cannot upgrade 2273 * to a write lock in vm_map_lookup. thus, any actions that 2274 * require the write lock must be done beforehand. because we 2275 * keep the read lock on the map, the copy-on-write status of the 2276 * entries we modify here cannot change. 2277 */ 2278 2279 for (size = 0, entry = map->header.next; entry != &map->header; 2280 entry = entry->next) { 2281 if (entry->protection != VM_PROT_NONE && 2282 VM_MAPENT_ISWIRED(entry) == 0) { /* not already wired? */ 2283 size += entry->end - entry->start; 2284 } 2285 } 2286 2287 if (atop(size) + uvmexp.wired > uvmexp.wiredmax) { 2288 vm_map_unlock(map); 2289 return (KERN_NO_SPACE); /* XXX overloaded */ 2290 } 2291 2292 /* XXX non-pmap_wired_count case must be handled by caller */ 2293 #ifdef pmap_wired_count 2294 if (limit != 0 && 2295 (size + ptoa(pmap_wired_count(vm_map_pmap(map))) > limit)) { 2296 vm_map_unlock(map); 2297 return (KERN_NO_SPACE); /* XXX overloaded */ 2298 } 2299 #endif 2300 2301 /* 2302 * Pass 2. 2303 */ 2304 2305 for (entry = map->header.next; entry != &map->header; 2306 entry = entry->next) { 2307 if (entry->protection == VM_PROT_NONE) 2308 continue; 2309 if (VM_MAPENT_ISWIRED(entry) == 0) { /* not already wired? */ 2310 /* 2311 * perform actions of vm_map_lookup that need the 2312 * write lock on the map: create an anonymous map 2313 * for a copy-on-write region, or an anonymous map 2314 * for a zero-fill region. (XXXCDC: submap case 2315 * ok?) 2316 */ 2317 if (!UVM_ET_ISSUBMAP(entry)) { /* not submap */ 2318 if (UVM_ET_ISNEEDSCOPY(entry) && 2319 ((entry->protection & VM_PROT_WRITE) || 2320 (entry->object.uvm_obj == NULL))) { 2321 amap_copy(map, entry, M_WAITOK, TRUE, 2322 entry->start, entry->end); 2323 /* XXXCDC: wait OK? */ 2324 } 2325 } 2326 } 2327 entry->wired_count++; 2328 } 2329 2330 /* 2331 * Pass 3. 2332 */ 2333 2334 #ifdef DIAGNOSTIC 2335 timestamp_save = map->timestamp; 2336 #endif 2337 vm_map_busy(map); 2338 vm_map_downgrade(map); 2339 2340 rv = KERN_SUCCESS; 2341 for (entry = map->header.next; entry != &map->header; 2342 entry = entry->next) { 2343 if (entry->wired_count == 1) { 2344 rv = uvm_fault_wire(map, entry->start, entry->end, 2345 entry->protection); 2346 if (rv) { 2347 /* 2348 * wiring failed. break out of the loop. 2349 * we'll clean up the map below, once we 2350 * have a write lock again. 2351 */ 2352 break; 2353 } 2354 } 2355 } 2356 2357 if (rv) { /* failed? */ 2358 /* 2359 * Get back an exclusive (write) lock. 2360 */ 2361 vm_map_upgrade(map); 2362 vm_map_unbusy(map); 2363 2364 #ifdef DIAGNOSTIC 2365 if (timestamp_save != map->timestamp) 2366 panic("uvm_map_pageable_all: stale map"); 2367 #endif 2368 2369 /* 2370 * first drop the wiring count on all the entries 2371 * which haven't actually been wired yet. 2372 * 2373 * Skip VM_PROT_NONE entries like we did above. 2374 */ 2375 failed_entry = entry; 2376 for (/* nothing */; entry != &map->header; 2377 entry = entry->next) { 2378 if (entry->protection == VM_PROT_NONE) 2379 continue; 2380 entry->wired_count--; 2381 } 2382 2383 /* 2384 * now, unwire all the entries that were successfully 2385 * wired above. 2386 * 2387 * Skip VM_PROT_NONE entries like we did above. 2388 */ 2389 for (entry = map->header.next; entry != failed_entry; 2390 entry = entry->next) { 2391 if (entry->protection == VM_PROT_NONE) 2392 continue; 2393 entry->wired_count--; 2394 if (VM_MAPENT_ISWIRED(entry)) 2395 uvm_map_entry_unwire(map, entry); 2396 } 2397 vm_map_unlock(map); 2398 UVMHIST_LOG(maphist,"<- done (RV=%d)", rv,0,0,0); 2399 return (rv); 2400 } 2401 2402 /* We are holding a read lock here. */ 2403 vm_map_unbusy(map); 2404 vm_map_unlock_read(map); 2405 2406 UVMHIST_LOG(maphist,"<- done (OK WIRE)",0,0,0,0); 2407 return (KERN_SUCCESS); 2408 } 2409 2410 /* 2411 * uvm_map_clean: clean out a map range 2412 * 2413 * => valid flags: 2414 * if (flags & PGO_CLEANIT): dirty pages are cleaned first 2415 * if (flags & PGO_SYNCIO): dirty pages are written synchronously 2416 * if (flags & PGO_DEACTIVATE): any cached pages are deactivated after clean 2417 * if (flags & PGO_FREE): any cached pages are freed after clean 2418 * => returns an error if any part of the specified range isn't mapped 2419 * => never a need to flush amap layer since the anonymous memory has 2420 * no permanent home, but may deactivate pages there 2421 * => called from sys_msync() and sys_madvise() 2422 * => caller must not write-lock map (read OK). 2423 * => we may sleep while cleaning if SYNCIO [with map read-locked] 2424 */ 2425 2426 int amap_clean_works = 1; /* XXX for now, just in case... */ 2427 2428 int 2429 uvm_map_clean(map, start, end, flags) 2430 vm_map_t map; 2431 vaddr_t start, end; 2432 int flags; 2433 { 2434 vm_map_entry_t current, entry; 2435 struct uvm_object *uobj; 2436 struct vm_amap *amap; 2437 struct vm_anon *anon; 2438 struct vm_page *pg; 2439 vaddr_t offset; 2440 vsize_t size; 2441 int rv, error, refs; 2442 UVMHIST_FUNC("uvm_map_clean"); UVMHIST_CALLED(maphist); 2443 UVMHIST_LOG(maphist,"(map=0x%x,start=0x%x,end=0x%x,flags=0x%x)", 2444 map, start, end, flags); 2445 2446 #ifdef DIAGNOSTIC 2447 if ((flags & (PGO_FREE|PGO_DEACTIVATE)) == (PGO_FREE|PGO_DEACTIVATE)) 2448 panic("uvm_map_clean: FREE and DEACTIVATE"); 2449 #endif 2450 2451 vm_map_lock_read(map); 2452 VM_MAP_RANGE_CHECK(map, start, end); 2453 if (uvm_map_lookup_entry(map, start, &entry) == FALSE) { 2454 vm_map_unlock_read(map); 2455 return(KERN_INVALID_ADDRESS); 2456 } 2457 2458 /* 2459 * Make a first pass to check for holes. 2460 */ 2461 for (current = entry; current->start < end; current = current->next) { 2462 if (UVM_ET_ISSUBMAP(current)) { 2463 vm_map_unlock_read(map); 2464 return (KERN_INVALID_ARGUMENT); 2465 } 2466 if (end > current->end && (current->next == &map->header || 2467 current->end != current->next->start)) { 2468 vm_map_unlock_read(map); 2469 return (KERN_INVALID_ADDRESS); 2470 } 2471 } 2472 2473 error = KERN_SUCCESS; 2474 2475 for (current = entry; current->start < end; current = current->next) { 2476 amap = current->aref.ar_amap; /* top layer */ 2477 uobj = current->object.uvm_obj; /* bottom layer */ 2478 2479 #ifdef DIAGNOSTIC 2480 if (start < current->start) 2481 panic("uvm_map_clean: hole"); 2482 #endif 2483 2484 /* 2485 * No amap cleaning necessary if: 2486 * 2487 * (1) There's no amap. 2488 * 2489 * (2) We're not deactivating or freeing pages. 2490 */ 2491 if (amap == NULL || 2492 (flags & (PGO_DEACTIVATE|PGO_FREE)) == 0) 2493 goto flush_object; 2494 2495 /* XXX for now, just in case... */ 2496 if (amap_clean_works == 0) 2497 goto flush_object; 2498 2499 amap_lock(amap); 2500 2501 offset = start - current->start; 2502 size = (end <= current->end ? end : current->end) - 2503 start; 2504 2505 for (/* nothing */; size != 0; size -= PAGE_SIZE, 2506 offset += PAGE_SIZE) { 2507 anon = amap_lookup(¤t->aref, offset); 2508 if (anon == NULL) 2509 continue; 2510 2511 simple_lock(&anon->an_lock); 2512 2513 pg = anon->u.an_page; 2514 if (pg == NULL) { 2515 simple_unlock(&anon->an_lock); 2516 continue; 2517 } 2518 2519 switch (flags & (PGO_CLEANIT|PGO_FREE|PGO_DEACTIVATE)) { 2520 /* 2521 * XXX In these first 3 cases, we always just 2522 * XXX deactivate the page. We may want to 2523 * XXX handle the different cases more 2524 * XXX specifically, in the future. 2525 */ 2526 case PGO_CLEANIT|PGO_FREE: 2527 case PGO_CLEANIT|PGO_DEACTIVATE: 2528 case PGO_DEACTIVATE: 2529 deactivate_it: 2530 /* skip the page if it's loaned or wired */ 2531 if (pg->loan_count != 0 || 2532 pg->wire_count != 0) { 2533 simple_unlock(&anon->an_lock); 2534 continue; 2535 } 2536 2537 uvm_lock_pageq(); 2538 2539 /* 2540 * skip the page if it's not actually owned 2541 * by the anon (may simply be loaned to the 2542 * anon). 2543 */ 2544 if ((pg->pqflags & PQ_ANON) == 0) { 2545 #ifdef DIAGNOSTIC 2546 if (pg->uobject != NULL) 2547 panic("uvm_map_clean: " 2548 "page anon vs. object " 2549 "inconsistency"); 2550 #endif 2551 uvm_unlock_pageq(); 2552 simple_unlock(&anon->an_lock); 2553 continue; 2554 } 2555 2556 #ifdef DIAGNOSTIC 2557 if (pg->uanon != anon) 2558 panic("uvm_map_clean: anon " 2559 "inconsistency"); 2560 #endif 2561 2562 /* zap all mappings for the page. */ 2563 pmap_page_protect(pg, VM_PROT_NONE); 2564 2565 /* ...and deactivate the page. */ 2566 uvm_pagedeactivate(pg); 2567 2568 uvm_unlock_pageq(); 2569 simple_unlock(&anon->an_lock); 2570 continue; 2571 2572 case PGO_FREE: 2573 /* 2574 * If there are multiple references to 2575 * the amap, just deactivate the page. 2576 */ 2577 if (amap_refs(amap) > 1) 2578 goto deactivate_it; 2579 2580 /* XXX skip the page if it's wired */ 2581 if (pg->wire_count != 0) { 2582 simple_unlock(&anon->an_lock); 2583 continue; 2584 } 2585 amap_unadd(¤t->aref, offset); 2586 refs = --anon->an_ref; 2587 simple_unlock(&anon->an_lock); 2588 if (refs == 0) 2589 uvm_anfree(anon); 2590 continue; 2591 2592 default: 2593 panic("uvm_map_clean: wierd flags"); 2594 } 2595 #ifdef DIAGNOSTIC 2596 panic("uvm_map_clean: unreachable code"); 2597 #endif 2598 } 2599 2600 amap_unlock(amap); 2601 2602 flush_object: 2603 /* 2604 * flush pages if we've got a valid backing object. 2605 */ 2606 2607 offset = current->offset + (start - current->start); 2608 size = (end <= current->end ? end : current->end) - start; 2609 2610 if (uobj != NULL) { 2611 simple_lock(&uobj->vmobjlock); 2612 rv = uobj->pgops->pgo_flush(uobj, offset, 2613 offset + size, flags); 2614 simple_unlock(&uobj->vmobjlock); 2615 2616 if (rv == FALSE) 2617 error = KERN_FAILURE; 2618 } 2619 start += size; 2620 } 2621 2622 vm_map_unlock_read(map); 2623 return (error); 2624 } 2625 2626 2627 /* 2628 * uvm_map_checkprot: check protection in map 2629 * 2630 * => must allow specified protection in a fully allocated region. 2631 * => map must be read or write locked by caller. 2632 */ 2633 2634 boolean_t 2635 uvm_map_checkprot(map, start, end, protection) 2636 vm_map_t map; 2637 vaddr_t start, end; 2638 vm_prot_t protection; 2639 { 2640 vm_map_entry_t entry; 2641 vm_map_entry_t tmp_entry; 2642 2643 if (!uvm_map_lookup_entry(map, start, &tmp_entry)) { 2644 return(FALSE); 2645 } 2646 2647 entry = tmp_entry; 2648 2649 while (start < end) { 2650 if (entry == &map->header) { 2651 return(FALSE); 2652 } 2653 2654 /* 2655 * no holes allowed 2656 */ 2657 2658 if (start < entry->start) { 2659 return(FALSE); 2660 } 2661 2662 /* 2663 * check protection associated with entry 2664 */ 2665 2666 if ((entry->protection & protection) != protection) { 2667 return(FALSE); 2668 } 2669 2670 /* go to next entry */ 2671 2672 start = entry->end; 2673 entry = entry->next; 2674 } 2675 return(TRUE); 2676 } 2677 2678 /* 2679 * uvmspace_alloc: allocate a vmspace structure. 2680 * 2681 * - structure includes vm_map and pmap 2682 * - XXX: no locking on this structure 2683 * - refcnt set to 1, rest must be init'd by caller 2684 */ 2685 struct vmspace * 2686 uvmspace_alloc(min, max, pageable) 2687 vaddr_t min, max; 2688 int pageable; 2689 { 2690 struct vmspace *vm; 2691 UVMHIST_FUNC("uvmspace_alloc"); UVMHIST_CALLED(maphist); 2692 2693 vm = pool_get(&uvm_vmspace_pool, PR_WAITOK); 2694 uvmspace_init(vm, NULL, min, max, pageable); 2695 UVMHIST_LOG(maphist,"<- done (vm=0x%x)", vm,0,0,0); 2696 return (vm); 2697 } 2698 2699 /* 2700 * uvmspace_init: initialize a vmspace structure. 2701 * 2702 * - XXX: no locking on this structure 2703 * - refcnt set to 1, rest must me init'd by caller 2704 */ 2705 void 2706 uvmspace_init(vm, pmap, min, max, pageable) 2707 struct vmspace *vm; 2708 struct pmap *pmap; 2709 vaddr_t min, max; 2710 boolean_t pageable; 2711 { 2712 UVMHIST_FUNC("uvmspace_init"); UVMHIST_CALLED(maphist); 2713 2714 memset(vm, 0, sizeof(*vm)); 2715 2716 uvm_map_setup(&vm->vm_map, min, max, pageable ? VM_MAP_PAGEABLE : 0); 2717 2718 if (pmap) 2719 pmap_reference(pmap); 2720 else 2721 pmap = pmap_create(); 2722 vm->vm_map.pmap = pmap; 2723 2724 vm->vm_refcnt = 1; 2725 UVMHIST_LOG(maphist,"<- done",0,0,0,0); 2726 } 2727 2728 /* 2729 * uvmspace_share: share a vmspace between two proceses 2730 * 2731 * - XXX: no locking on vmspace 2732 * - used for vfork, threads(?) 2733 */ 2734 2735 void 2736 uvmspace_share(p1, p2) 2737 struct proc *p1, *p2; 2738 { 2739 p2->p_vmspace = p1->p_vmspace; 2740 p1->p_vmspace->vm_refcnt++; 2741 } 2742 2743 /* 2744 * uvmspace_unshare: ensure that process "p" has its own, unshared, vmspace 2745 * 2746 * - XXX: no locking on vmspace 2747 */ 2748 2749 void 2750 uvmspace_unshare(p) 2751 struct proc *p; 2752 { 2753 struct vmspace *nvm, *ovm = p->p_vmspace; 2754 2755 if (ovm->vm_refcnt == 1) 2756 /* nothing to do: vmspace isn't shared in the first place */ 2757 return; 2758 2759 /* make a new vmspace, still holding old one */ 2760 nvm = uvmspace_fork(ovm); 2761 2762 pmap_deactivate(p); /* unbind old vmspace */ 2763 p->p_vmspace = nvm; 2764 pmap_activate(p); /* switch to new vmspace */ 2765 2766 uvmspace_free(ovm); /* drop reference to old vmspace */ 2767 } 2768 2769 /* 2770 * uvmspace_exec: the process wants to exec a new program 2771 * 2772 * - XXX: no locking on vmspace 2773 */ 2774 2775 void 2776 uvmspace_exec(p) 2777 struct proc *p; 2778 { 2779 struct vmspace *nvm, *ovm = p->p_vmspace; 2780 vm_map_t map = &ovm->vm_map; 2781 2782 #ifdef __sparc__ 2783 /* XXX cgd 960926: the sparc #ifdef should be a MD hook */ 2784 kill_user_windows(p); /* before stack addresses go away */ 2785 #endif 2786 2787 /* 2788 * see if more than one process is using this vmspace... 2789 */ 2790 2791 if (ovm->vm_refcnt == 1) { 2792 2793 /* 2794 * if p is the only process using its vmspace then we can safely 2795 * recycle that vmspace for the program that is being exec'd. 2796 */ 2797 2798 #ifdef SYSVSHM 2799 /* 2800 * SYSV SHM semantics require us to kill all segments on an exec 2801 */ 2802 if (ovm->vm_shm) 2803 shmexit(ovm); 2804 #endif 2805 2806 /* 2807 * POSIX 1003.1b -- "lock future mappings" is revoked 2808 * when a process execs another program image. 2809 */ 2810 vm_map_lock(map); 2811 vm_map_modflags(map, 0, VM_MAP_WIREFUTURE); 2812 vm_map_unlock(map); 2813 2814 /* 2815 * now unmap the old program 2816 */ 2817 uvm_unmap(map, VM_MIN_ADDRESS, VM_MAXUSER_ADDRESS); 2818 2819 } else { 2820 2821 /* 2822 * p's vmspace is being shared, so we can't reuse it for p since 2823 * it is still being used for others. allocate a new vmspace 2824 * for p 2825 */ 2826 nvm = uvmspace_alloc(map->min_offset, map->max_offset, 2827 (map->flags & VM_MAP_PAGEABLE) ? TRUE : FALSE); 2828 2829 /* 2830 * install new vmspace and drop our ref to the old one. 2831 */ 2832 2833 pmap_deactivate(p); 2834 p->p_vmspace = nvm; 2835 pmap_activate(p); 2836 2837 uvmspace_free(ovm); 2838 } 2839 } 2840 2841 /* 2842 * uvmspace_free: free a vmspace data structure 2843 * 2844 * - XXX: no locking on vmspace 2845 */ 2846 2847 void 2848 uvmspace_free(vm) 2849 struct vmspace *vm; 2850 { 2851 vm_map_entry_t dead_entries; 2852 UVMHIST_FUNC("uvmspace_free"); UVMHIST_CALLED(maphist); 2853 2854 UVMHIST_LOG(maphist,"(vm=0x%x) ref=%d", vm, vm->vm_refcnt,0,0); 2855 if (--vm->vm_refcnt == 0) { 2856 /* 2857 * lock the map, to wait out all other references to it. delete 2858 * all of the mappings and pages they hold, then call the pmap 2859 * module to reclaim anything left. 2860 */ 2861 vm_map_lock(&vm->vm_map); 2862 if (vm->vm_map.nentries) { 2863 (void)uvm_unmap_remove(&vm->vm_map, 2864 vm->vm_map.min_offset, vm->vm_map.max_offset, 2865 &dead_entries); 2866 if (dead_entries != NULL) 2867 uvm_unmap_detach(dead_entries, 0); 2868 } 2869 pmap_destroy(vm->vm_map.pmap); 2870 vm->vm_map.pmap = NULL; 2871 pool_put(&uvm_vmspace_pool, vm); 2872 } 2873 UVMHIST_LOG(maphist,"<- done", 0,0,0,0); 2874 } 2875 2876 /* 2877 * F O R K - m a i n e n t r y p o i n t 2878 */ 2879 /* 2880 * uvmspace_fork: fork a process' main map 2881 * 2882 * => create a new vmspace for child process from parent. 2883 * => parent's map must not be locked. 2884 */ 2885 2886 struct vmspace * 2887 uvmspace_fork(vm1) 2888 struct vmspace *vm1; 2889 { 2890 struct vmspace *vm2; 2891 vm_map_t old_map = &vm1->vm_map; 2892 vm_map_t new_map; 2893 vm_map_entry_t old_entry; 2894 vm_map_entry_t new_entry; 2895 pmap_t new_pmap; 2896 boolean_t protect_child; 2897 UVMHIST_FUNC("uvmspace_fork"); UVMHIST_CALLED(maphist); 2898 2899 vm_map_lock(old_map); 2900 2901 vm2 = uvmspace_alloc(old_map->min_offset, old_map->max_offset, 2902 (old_map->flags & VM_MAP_PAGEABLE) ? TRUE : FALSE); 2903 memcpy(&vm2->vm_startcopy, &vm1->vm_startcopy, 2904 (caddr_t) (vm1 + 1) - (caddr_t) &vm1->vm_startcopy); 2905 new_map = &vm2->vm_map; /* XXX */ 2906 new_pmap = new_map->pmap; 2907 2908 old_entry = old_map->header.next; 2909 2910 /* 2911 * go entry-by-entry 2912 */ 2913 2914 while (old_entry != &old_map->header) { 2915 2916 /* 2917 * first, some sanity checks on the old entry 2918 */ 2919 if (UVM_ET_ISSUBMAP(old_entry)) 2920 panic("fork: encountered a submap during fork (illegal)"); 2921 2922 if (!UVM_ET_ISCOPYONWRITE(old_entry) && 2923 UVM_ET_ISNEEDSCOPY(old_entry)) 2924 panic("fork: non-copy_on_write map entry marked needs_copy (illegal)"); 2925 2926 2927 switch (old_entry->inheritance) { 2928 case VM_INHERIT_NONE: 2929 /* 2930 * drop the mapping 2931 */ 2932 break; 2933 2934 case VM_INHERIT_SHARE: 2935 /* 2936 * share the mapping: this means we want the old and 2937 * new entries to share amaps and backing objects. 2938 */ 2939 2940 /* 2941 * if the old_entry needs a new amap (due to prev fork) 2942 * then we need to allocate it now so that we have 2943 * something we own to share with the new_entry. [in 2944 * other words, we need to clear needs_copy] 2945 */ 2946 2947 if (UVM_ET_ISNEEDSCOPY(old_entry)) { 2948 /* get our own amap, clears needs_copy */ 2949 amap_copy(old_map, old_entry, M_WAITOK, FALSE, 2950 0, 0); 2951 /* XXXCDC: WAITOK??? */ 2952 } 2953 2954 new_entry = uvm_mapent_alloc(new_map); 2955 /* old_entry -> new_entry */ 2956 uvm_mapent_copy(old_entry, new_entry); 2957 2958 /* new pmap has nothing wired in it */ 2959 new_entry->wired_count = 0; 2960 2961 /* 2962 * gain reference to object backing the map (can't 2963 * be a submap, already checked this case). 2964 */ 2965 if (new_entry->aref.ar_amap) 2966 /* share reference */ 2967 amap_ref(new_entry, AMAP_SHARED); 2968 2969 if (new_entry->object.uvm_obj && 2970 new_entry->object.uvm_obj->pgops->pgo_reference) 2971 new_entry->object.uvm_obj-> 2972 pgops->pgo_reference( 2973 new_entry->object.uvm_obj); 2974 2975 /* insert entry at end of new_map's entry list */ 2976 uvm_map_entry_link(new_map, new_map->header.prev, 2977 new_entry); 2978 2979 /* 2980 * pmap_copy the mappings: this routine is optional 2981 * but if it is there it will reduce the number of 2982 * page faults in the new proc. 2983 */ 2984 2985 pmap_copy(new_pmap, old_map->pmap, new_entry->start, 2986 (old_entry->end - old_entry->start), 2987 old_entry->start); 2988 2989 break; 2990 2991 case VM_INHERIT_COPY: 2992 2993 /* 2994 * copy-on-write the mapping (using mmap's 2995 * MAP_PRIVATE semantics) 2996 * 2997 * allocate new_entry, adjust reference counts. 2998 * (note that new references are read-only). 2999 */ 3000 3001 new_entry = uvm_mapent_alloc(new_map); 3002 /* old_entry -> new_entry */ 3003 uvm_mapent_copy(old_entry, new_entry); 3004 3005 if (new_entry->aref.ar_amap) 3006 amap_ref(new_entry, 0); 3007 3008 if (new_entry->object.uvm_obj && 3009 new_entry->object.uvm_obj->pgops->pgo_reference) 3010 new_entry->object.uvm_obj->pgops->pgo_reference 3011 (new_entry->object.uvm_obj); 3012 3013 /* new pmap has nothing wired in it */ 3014 new_entry->wired_count = 0; 3015 3016 new_entry->etype |= 3017 (UVM_ET_COPYONWRITE|UVM_ET_NEEDSCOPY); 3018 uvm_map_entry_link(new_map, new_map->header.prev, 3019 new_entry); 3020 3021 /* 3022 * the new entry will need an amap. it will either 3023 * need to be copied from the old entry or created 3024 * from scratch (if the old entry does not have an 3025 * amap). can we defer this process until later 3026 * (by setting "needs_copy") or do we need to copy 3027 * the amap now? 3028 * 3029 * we must copy the amap now if any of the following 3030 * conditions hold: 3031 * 1. the old entry has an amap and that amap is 3032 * being shared. this means that the old (parent) 3033 * process is sharing the amap with another 3034 * process. if we do not clear needs_copy here 3035 * we will end up in a situation where both the 3036 * parent and child process are refering to the 3037 * same amap with "needs_copy" set. if the 3038 * parent write-faults, the fault routine will 3039 * clear "needs_copy" in the parent by allocating 3040 * a new amap. this is wrong because the 3041 * parent is supposed to be sharing the old amap 3042 * and the new amap will break that. 3043 * 3044 * 2. if the old entry has an amap and a non-zero 3045 * wire count then we are going to have to call 3046 * amap_cow_now to avoid page faults in the 3047 * parent process. since amap_cow_now requires 3048 * "needs_copy" to be clear we might as well 3049 * clear it here as well. 3050 * 3051 */ 3052 3053 if (old_entry->aref.ar_amap != NULL) { 3054 3055 if ((amap_flags(old_entry->aref.ar_amap) & 3056 AMAP_SHARED) != 0 || 3057 VM_MAPENT_ISWIRED(old_entry)) { 3058 3059 amap_copy(new_map, new_entry, M_WAITOK, FALSE, 3060 0, 0); 3061 /* XXXCDC: M_WAITOK ... ok? */ 3062 } 3063 } 3064 3065 /* 3066 * if the parent's entry is wired down, then the 3067 * parent process does not want page faults on 3068 * access to that memory. this means that we 3069 * cannot do copy-on-write because we can't write 3070 * protect the old entry. in this case we 3071 * resolve all copy-on-write faults now, using 3072 * amap_cow_now. note that we have already 3073 * allocated any needed amap (above). 3074 */ 3075 3076 if (VM_MAPENT_ISWIRED(old_entry)) { 3077 3078 /* 3079 * resolve all copy-on-write faults now 3080 * (note that there is nothing to do if 3081 * the old mapping does not have an amap). 3082 * XXX: is it worthwhile to bother with pmap_copy 3083 * in this case? 3084 */ 3085 if (old_entry->aref.ar_amap) 3086 amap_cow_now(new_map, new_entry); 3087 3088 } else { 3089 3090 /* 3091 * setup mappings to trigger copy-on-write faults 3092 * we must write-protect the parent if it has 3093 * an amap and it is not already "needs_copy"... 3094 * if it is already "needs_copy" then the parent 3095 * has already been write-protected by a previous 3096 * fork operation. 3097 * 3098 * if we do not write-protect the parent, then 3099 * we must be sure to write-protect the child 3100 * after the pmap_copy() operation. 3101 * 3102 * XXX: pmap_copy should have some way of telling 3103 * us that it didn't do anything so we can avoid 3104 * calling pmap_protect needlessly. 3105 */ 3106 3107 if (old_entry->aref.ar_amap) { 3108 3109 if (!UVM_ET_ISNEEDSCOPY(old_entry)) { 3110 if (old_entry->max_protection & VM_PROT_WRITE) { 3111 pmap_protect(old_map->pmap, 3112 old_entry->start, 3113 old_entry->end, 3114 old_entry->protection & 3115 ~VM_PROT_WRITE); 3116 } 3117 old_entry->etype |= UVM_ET_NEEDSCOPY; 3118 } 3119 3120 /* 3121 * parent must now be write-protected 3122 */ 3123 protect_child = FALSE; 3124 } else { 3125 3126 /* 3127 * we only need to protect the child if the 3128 * parent has write access. 3129 */ 3130 if (old_entry->max_protection & VM_PROT_WRITE) 3131 protect_child = TRUE; 3132 else 3133 protect_child = FALSE; 3134 3135 } 3136 3137 /* 3138 * copy the mappings 3139 * XXX: need a way to tell if this does anything 3140 */ 3141 3142 pmap_copy(new_pmap, old_map->pmap, 3143 new_entry->start, 3144 (old_entry->end - old_entry->start), 3145 old_entry->start); 3146 3147 /* 3148 * protect the child's mappings if necessary 3149 */ 3150 if (protect_child) { 3151 pmap_protect(new_pmap, new_entry->start, 3152 new_entry->end, 3153 new_entry->protection & 3154 ~VM_PROT_WRITE); 3155 } 3156 3157 } 3158 break; 3159 } /* end of switch statement */ 3160 old_entry = old_entry->next; 3161 } 3162 3163 new_map->size = old_map->size; 3164 vm_map_unlock(old_map); 3165 3166 #ifdef SYSVSHM 3167 if (vm1->vm_shm) 3168 shmfork(vm1, vm2); 3169 #endif 3170 3171 #ifdef PMAP_FORK 3172 pmap_fork(vm1->vm_map.pmap, vm2->vm_map.pmap); 3173 #endif 3174 3175 UVMHIST_LOG(maphist,"<- done",0,0,0,0); 3176 return(vm2); 3177 } 3178 3179 3180 #if defined(DDB) 3181 3182 /* 3183 * DDB hooks 3184 */ 3185 3186 /* 3187 * uvm_map_print: print out a map 3188 */ 3189 3190 void 3191 uvm_map_print(map, full) 3192 vm_map_t map; 3193 boolean_t full; 3194 { 3195 3196 uvm_map_printit(map, full, printf); 3197 } 3198 3199 /* 3200 * uvm_map_printit: actually prints the map 3201 */ 3202 3203 void 3204 uvm_map_printit(map, full, pr) 3205 vm_map_t map; 3206 boolean_t full; 3207 void (*pr) __P((const char *, ...)); 3208 { 3209 vm_map_entry_t entry; 3210 3211 (*pr)("MAP %p: [0x%lx->0x%lx]\n", map, map->min_offset,map->max_offset); 3212 (*pr)("\t#ent=%d, sz=%d, ref=%d, version=%d, flags=0x%x\n", 3213 map->nentries, map->size, map->ref_count, map->timestamp, 3214 map->flags); 3215 #ifdef pmap_resident_count 3216 (*pr)("\tpmap=%p(resident=%d)\n", map->pmap, 3217 pmap_resident_count(map->pmap)); 3218 #else 3219 /* XXXCDC: this should be required ... */ 3220 (*pr)("\tpmap=%p(resident=<<NOT SUPPORTED!!!>>)\n", map->pmap); 3221 #endif 3222 if (!full) 3223 return; 3224 for (entry = map->header.next; entry != &map->header; 3225 entry = entry->next) { 3226 (*pr)(" - %p: 0x%lx->0x%lx: obj=%p/0x%llx, amap=%p/%d\n", 3227 entry, entry->start, entry->end, entry->object.uvm_obj, 3228 (long long)entry->offset, entry->aref.ar_amap, entry->aref.ar_pageoff); 3229 (*pr)( 3230 "\tsubmap=%c, cow=%c, nc=%c, prot(max)=%d/%d, inh=%d, wc=%d, adv=%d\n", 3231 (entry->etype & UVM_ET_SUBMAP) ? 'T' : 'F', 3232 (entry->etype & UVM_ET_COPYONWRITE) ? 'T' : 'F', 3233 (entry->etype & UVM_ET_NEEDSCOPY) ? 'T' : 'F', 3234 entry->protection, entry->max_protection, 3235 entry->inheritance, entry->wired_count, entry->advice); 3236 } 3237 } 3238 3239 /* 3240 * uvm_object_print: print out an object 3241 */ 3242 3243 void 3244 uvm_object_print(uobj, full) 3245 struct uvm_object *uobj; 3246 boolean_t full; 3247 { 3248 3249 uvm_object_printit(uobj, full, printf); 3250 } 3251 3252 /* 3253 * uvm_object_printit: actually prints the object 3254 */ 3255 3256 void 3257 uvm_object_printit(uobj, full, pr) 3258 struct uvm_object *uobj; 3259 boolean_t full; 3260 void (*pr) __P((const char *, ...)); 3261 { 3262 struct vm_page *pg; 3263 int cnt = 0; 3264 3265 (*pr)("OBJECT %p: locked=%d, pgops=%p, npages=%d, ", 3266 uobj, uobj->vmobjlock.lock_data, uobj->pgops, uobj->uo_npages); 3267 if (UVM_OBJ_IS_KERN_OBJECT(uobj)) 3268 (*pr)("refs=<SYSTEM>\n"); 3269 else 3270 (*pr)("refs=%d\n", uobj->uo_refs); 3271 3272 if (!full) { 3273 return; 3274 } 3275 (*pr)(" PAGES <pg,offset>:\n "); 3276 for (pg = TAILQ_FIRST(&uobj->memq); 3277 pg != NULL; 3278 pg = TAILQ_NEXT(pg, listq), cnt++) { 3279 (*pr)("<%p,0x%lx> ", pg, pg->offset); 3280 if ((cnt % 3) == 2) { 3281 (*pr)("\n "); 3282 } 3283 } 3284 if ((cnt % 3) != 2) { 3285 (*pr)("\n"); 3286 } 3287 } 3288 3289 const char page_flagbits[] = 3290 "\20\4CLEAN\5BUSY\6WANTED\7TABLED\12FAKE\13FILLED\14DIRTY\15RELEASED" 3291 "\16FAULTING\17CLEANCHK"; 3292 const char page_pqflagbits[] = 3293 "\20\1FREE\2INACTIVE\3ACTIVE\4LAUNDRY\5ANON\6AOBJ"; 3294 3295 /* 3296 * uvm_page_print: print out a page 3297 */ 3298 3299 void 3300 uvm_page_print(pg, full) 3301 struct vm_page *pg; 3302 boolean_t full; 3303 { 3304 3305 uvm_page_printit(pg, full, printf); 3306 } 3307 3308 /* 3309 * uvm_page_printit: actually print the page 3310 */ 3311 3312 void 3313 uvm_page_printit(pg, full, pr) 3314 struct vm_page *pg; 3315 boolean_t full; 3316 void (*pr) __P((const char *, ...)); 3317 { 3318 struct vm_page *lcv; 3319 struct uvm_object *uobj; 3320 struct pglist *pgl; 3321 char pgbuf[128]; 3322 char pqbuf[128]; 3323 3324 (*pr)("PAGE %p:\n", pg); 3325 bitmask_snprintf(pg->flags, page_flagbits, pgbuf, sizeof(pgbuf)); 3326 bitmask_snprintf(pg->pqflags, page_pqflagbits, pqbuf, sizeof(pqbuf)); 3327 (*pr)(" flags=%s, pqflags=%s, vers=%d, wire_count=%d, pa=0x%lx\n", 3328 pgbuf, pqbuf, pg->version, pg->wire_count, (long)pg->phys_addr); 3329 (*pr)(" uobject=%p, uanon=%p, offset=0x%lx loan_count=%d\n", 3330 pg->uobject, pg->uanon, pg->offset, pg->loan_count); 3331 #if defined(UVM_PAGE_TRKOWN) 3332 if (pg->flags & PG_BUSY) 3333 (*pr)(" owning process = %d, tag=%s\n", 3334 pg->owner, pg->owner_tag); 3335 else 3336 (*pr)(" page not busy, no owner\n"); 3337 #else 3338 (*pr)(" [page ownership tracking disabled]\n"); 3339 #endif 3340 3341 if (!full) 3342 return; 3343 3344 /* cross-verify object/anon */ 3345 if ((pg->pqflags & PQ_FREE) == 0) { 3346 if (pg->pqflags & PQ_ANON) { 3347 if (pg->uanon == NULL || pg->uanon->u.an_page != pg) 3348 (*pr)(" >>> ANON DOES NOT POINT HERE <<< (%p)\n", 3349 (pg->uanon) ? pg->uanon->u.an_page : NULL); 3350 else 3351 (*pr)(" anon backpointer is OK\n"); 3352 } else { 3353 uobj = pg->uobject; 3354 if (uobj) { 3355 (*pr)(" checking object list\n"); 3356 for (lcv = uobj->memq.tqh_first ; lcv ; 3357 lcv = lcv->listq.tqe_next) { 3358 if (lcv == pg) break; 3359 } 3360 if (lcv) 3361 (*pr)(" page found on object list\n"); 3362 else 3363 (*pr)(" >>> PAGE NOT FOUND ON OBJECT LIST! <<<\n"); 3364 } 3365 } 3366 } 3367 3368 /* cross-verify page queue */ 3369 if (pg->pqflags & PQ_FREE) { 3370 int fl = uvm_page_lookup_freelist(pg); 3371 pgl = &uvm.page_free[fl].pgfl_queues[((pg)->flags & PG_ZERO) ? 3372 PGFL_ZEROS : PGFL_UNKNOWN]; 3373 } 3374 else if (pg->pqflags & PQ_INACTIVE) 3375 pgl = (pg->pqflags & PQ_SWAPBACKED) ? 3376 &uvm.page_inactive_swp : &uvm.page_inactive_obj; 3377 else if (pg->pqflags & PQ_ACTIVE) 3378 pgl = &uvm.page_active; 3379 else 3380 pgl = NULL; 3381 3382 if (pgl) { 3383 (*pr)(" checking pageq list\n"); 3384 for (lcv = pgl->tqh_first ; lcv ; lcv = lcv->pageq.tqe_next) { 3385 if (lcv == pg) break; 3386 } 3387 if (lcv) 3388 (*pr)(" page found on pageq list\n"); 3389 else 3390 (*pr)(" >>> PAGE NOT FOUND ON PAGEQ LIST! <<<\n"); 3391 } 3392 } 3393 #endif 3394