1 /* $NetBSD: uvm_map.c,v 1.100 2001/07/22 13:34:12 wiz 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 const char vmmapbsy[] = "vmmapbsy"; 99 100 /* 101 * pool for vmspace structures. 102 */ 103 104 struct pool uvm_vmspace_pool; 105 106 /* 107 * pool for dynamically-allocated map entries. 108 */ 109 110 struct pool uvm_map_entry_pool; 111 112 #ifdef PMAP_GROWKERNEL 113 /* 114 * This global represents the end of the kernel virtual address 115 * space. If we want to exceed this, we must grow the kernel 116 * virtual address space dynamically. 117 * 118 * Note, this variable is locked by kernel_map's lock. 119 */ 120 vaddr_t uvm_maxkaddr; 121 #endif 122 123 /* 124 * macros 125 */ 126 127 /* 128 * uvm_map_entry_link: insert entry into a map 129 * 130 * => map must be locked 131 */ 132 #define uvm_map_entry_link(map, after_where, entry) do { \ 133 (map)->nentries++; \ 134 (entry)->prev = (after_where); \ 135 (entry)->next = (after_where)->next; \ 136 (entry)->prev->next = (entry); \ 137 (entry)->next->prev = (entry); \ 138 } while (0) 139 140 /* 141 * uvm_map_entry_unlink: remove entry from a map 142 * 143 * => map must be locked 144 */ 145 #define uvm_map_entry_unlink(map, entry) do { \ 146 (map)->nentries--; \ 147 (entry)->next->prev = (entry)->prev; \ 148 (entry)->prev->next = (entry)->next; \ 149 } while (0) 150 151 /* 152 * SAVE_HINT: saves the specified entry as the hint for future lookups. 153 * 154 * => map need not be locked (protected by hint_lock). 155 */ 156 #define SAVE_HINT(map,check,value) do { \ 157 simple_lock(&(map)->hint_lock); \ 158 if ((map)->hint == (check)) \ 159 (map)->hint = (value); \ 160 simple_unlock(&(map)->hint_lock); \ 161 } while (0) 162 163 /* 164 * VM_MAP_RANGE_CHECK: check and correct range 165 * 166 * => map must at least be read locked 167 */ 168 169 #define VM_MAP_RANGE_CHECK(map, start, end) do { \ 170 if (start < vm_map_min(map)) \ 171 start = vm_map_min(map); \ 172 if (end > vm_map_max(map)) \ 173 end = vm_map_max(map); \ 174 if (start > end) \ 175 start = end; \ 176 } while (0) 177 178 /* 179 * local prototypes 180 */ 181 182 static struct vm_map_entry *uvm_mapent_alloc __P((struct vm_map *)); 183 static void uvm_mapent_copy __P((struct vm_map_entry *, struct vm_map_entry *)); 184 static void uvm_mapent_free __P((struct vm_map_entry *)); 185 static void uvm_map_entry_unwire __P((struct vm_map *, struct vm_map_entry *)); 186 static void uvm_map_reference_amap __P((struct vm_map_entry *, int)); 187 static void uvm_map_unreference_amap __P((struct vm_map_entry *, int)); 188 189 /* 190 * local inlines 191 */ 192 193 /* 194 * uvm_mapent_alloc: allocate a map entry 195 * 196 * => XXX: static pool for kernel map? 197 */ 198 199 static __inline struct vm_map_entry * 200 uvm_mapent_alloc(map) 201 struct vm_map *map; 202 { 203 struct vm_map_entry *me; 204 int s; 205 UVMHIST_FUNC("uvm_mapent_alloc"); 206 UVMHIST_CALLED(maphist); 207 208 if ((map->flags & VM_MAP_INTRSAFE) == 0 && 209 map != kernel_map && kernel_map != NULL /* XXX */) { 210 me = pool_get(&uvm_map_entry_pool, PR_WAITOK); 211 me->flags = 0; 212 /* me can't be null, wait ok */ 213 } else { 214 s = splvm(); /* protect kentry_free list with splvm */ 215 simple_lock(&uvm.kentry_lock); 216 me = uvm.kentry_free; 217 if (me) uvm.kentry_free = me->next; 218 simple_unlock(&uvm.kentry_lock); 219 splx(s); 220 if (!me) 221 panic("mapent_alloc: out of static map entries, check MAX_KMAPENT"); 222 me->flags = UVM_MAP_STATIC; 223 } 224 225 UVMHIST_LOG(maphist, "<- new entry=0x%x [kentry=%d]", 226 me, ((map->flags & VM_MAP_INTRSAFE) != 0 || map == kernel_map) 227 ? TRUE : FALSE, 0, 0); 228 return(me); 229 } 230 231 /* 232 * uvm_mapent_free: free map entry 233 * 234 * => XXX: static pool for kernel map? 235 */ 236 237 static __inline void 238 uvm_mapent_free(me) 239 struct vm_map_entry *me; 240 { 241 int s; 242 UVMHIST_FUNC("uvm_mapent_free"); 243 UVMHIST_CALLED(maphist); 244 UVMHIST_LOG(maphist,"<- freeing map entry=0x%x [flags=%d]", 245 me, me->flags, 0, 0); 246 if ((me->flags & UVM_MAP_STATIC) == 0) { 247 pool_put(&uvm_map_entry_pool, me); 248 } else { 249 s = splvm(); /* protect kentry_free list with splvm */ 250 simple_lock(&uvm.kentry_lock); 251 me->next = uvm.kentry_free; 252 uvm.kentry_free = me; 253 simple_unlock(&uvm.kentry_lock); 254 splx(s); 255 } 256 } 257 258 /* 259 * uvm_mapent_copy: copy a map entry, preserving flags 260 */ 261 262 static __inline void 263 uvm_mapent_copy(src, dst) 264 struct vm_map_entry *src; 265 struct vm_map_entry *dst; 266 { 267 memcpy(dst, src, 268 ((char *)&src->uvm_map_entry_stop_copy) - ((char *)src)); 269 } 270 271 /* 272 * uvm_map_entry_unwire: unwire a map entry 273 * 274 * => map should be locked by caller 275 */ 276 277 static __inline void 278 uvm_map_entry_unwire(map, entry) 279 struct vm_map *map; 280 struct vm_map_entry *entry; 281 { 282 entry->wired_count = 0; 283 uvm_fault_unwire_locked(map, entry->start, entry->end); 284 } 285 286 287 /* 288 * wrapper for calling amap_ref() 289 */ 290 static __inline void 291 uvm_map_reference_amap(entry, flags) 292 struct vm_map_entry *entry; 293 int flags; 294 { 295 amap_ref(entry->aref.ar_amap, entry->aref.ar_pageoff, 296 (entry->end - entry->start) >> PAGE_SHIFT, flags); 297 } 298 299 300 /* 301 * wrapper for calling amap_unref() 302 */ 303 static __inline void 304 uvm_map_unreference_amap(entry, flags) 305 struct vm_map_entry *entry; 306 int flags; 307 { 308 amap_unref(entry->aref.ar_amap, entry->aref.ar_pageoff, 309 (entry->end - entry->start) >> PAGE_SHIFT, flags); 310 } 311 312 313 /* 314 * uvm_map_init: init mapping system at boot time. note that we allocate 315 * and init the static pool of struct vm_map_entry *'s for the kernel here. 316 */ 317 318 void 319 uvm_map_init() 320 { 321 static struct vm_map_entry kernel_map_entry[MAX_KMAPENT]; 322 #if defined(UVMHIST) 323 static struct uvm_history_ent maphistbuf[100]; 324 static struct uvm_history_ent pdhistbuf[100]; 325 #endif 326 int lcv; 327 328 /* 329 * first, init logging system. 330 */ 331 332 UVMHIST_FUNC("uvm_map_init"); 333 UVMHIST_INIT_STATIC(maphist, maphistbuf); 334 UVMHIST_INIT_STATIC(pdhist, pdhistbuf); 335 UVMHIST_CALLED(maphist); 336 UVMHIST_LOG(maphist,"<starting uvm map system>", 0, 0, 0, 0); 337 UVMCNT_INIT(uvm_map_call, UVMCNT_CNT, 0, 338 "# uvm_map() successful calls", 0); 339 UVMCNT_INIT(map_backmerge, UVMCNT_CNT, 0, "# uvm_map() back merges", 0); 340 UVMCNT_INIT(map_forwmerge, UVMCNT_CNT, 0, "# uvm_map() missed forward", 341 0); 342 UVMCNT_INIT(uvm_mlk_call, UVMCNT_CNT, 0, "# map lookup calls", 0); 343 UVMCNT_INIT(uvm_mlk_hint, UVMCNT_CNT, 0, "# map lookup hint hits", 0); 344 345 /* 346 * now set up static pool of kernel map entrys ... 347 */ 348 349 simple_lock_init(&uvm.kentry_lock); 350 uvm.kentry_free = NULL; 351 for (lcv = 0 ; lcv < MAX_KMAPENT ; lcv++) { 352 kernel_map_entry[lcv].next = uvm.kentry_free; 353 uvm.kentry_free = &kernel_map_entry[lcv]; 354 } 355 356 /* 357 * initialize the map-related pools. 358 */ 359 pool_init(&uvm_vmspace_pool, sizeof(struct vmspace), 360 0, 0, 0, "vmsppl", 0, 361 pool_page_alloc_nointr, pool_page_free_nointr, M_VMMAP); 362 pool_init(&uvm_map_entry_pool, sizeof(struct vm_map_entry), 363 0, 0, 0, "vmmpepl", 0, 364 pool_page_alloc_nointr, pool_page_free_nointr, M_VMMAP); 365 } 366 367 /* 368 * clippers 369 */ 370 371 /* 372 * uvm_map_clip_start: ensure that the entry begins at or after 373 * the starting address, if it doesn't we split the entry. 374 * 375 * => caller should use UVM_MAP_CLIP_START macro rather than calling 376 * this directly 377 * => map must be locked by caller 378 */ 379 380 void 381 uvm_map_clip_start(map, entry, start) 382 struct vm_map *map; 383 struct vm_map_entry *entry; 384 vaddr_t start; 385 { 386 struct vm_map_entry *new_entry; 387 vaddr_t new_adj; 388 389 /* uvm_map_simplify_entry(map, entry); */ /* XXX */ 390 391 /* 392 * Split off the front portion. note that we must insert the new 393 * entry BEFORE this one, so that this entry has the specified 394 * starting address. 395 */ 396 397 new_entry = uvm_mapent_alloc(map); 398 uvm_mapent_copy(entry, new_entry); /* entry -> new_entry */ 399 400 new_entry->end = start; 401 new_adj = start - new_entry->start; 402 if (entry->object.uvm_obj) 403 entry->offset += new_adj; /* shift start over */ 404 entry->start = start; 405 406 if (new_entry->aref.ar_amap) { 407 amap_splitref(&new_entry->aref, &entry->aref, new_adj); 408 } 409 410 uvm_map_entry_link(map, entry->prev, new_entry); 411 412 if (UVM_ET_ISSUBMAP(entry)) { 413 /* ... unlikely to happen, but play it safe */ 414 uvm_map_reference(new_entry->object.sub_map); 415 } else { 416 if (UVM_ET_ISOBJ(entry) && 417 entry->object.uvm_obj->pgops && 418 entry->object.uvm_obj->pgops->pgo_reference) 419 entry->object.uvm_obj->pgops->pgo_reference( 420 entry->object.uvm_obj); 421 } 422 } 423 424 /* 425 * uvm_map_clip_end: ensure that the entry ends at or before 426 * the ending address, if it does't we split the reference 427 * 428 * => caller should use UVM_MAP_CLIP_END macro rather than calling 429 * this directly 430 * => map must be locked by caller 431 */ 432 433 void 434 uvm_map_clip_end(map, entry, end) 435 struct vm_map *map; 436 struct vm_map_entry *entry; 437 vaddr_t end; 438 { 439 struct vm_map_entry * new_entry; 440 vaddr_t new_adj; /* #bytes we move start forward */ 441 442 /* 443 * Create a new entry and insert it 444 * AFTER the specified entry 445 */ 446 447 new_entry = uvm_mapent_alloc(map); 448 uvm_mapent_copy(entry, new_entry); /* entry -> new_entry */ 449 450 new_entry->start = entry->end = end; 451 new_adj = end - entry->start; 452 if (new_entry->object.uvm_obj) 453 new_entry->offset += new_adj; 454 455 if (entry->aref.ar_amap) 456 amap_splitref(&entry->aref, &new_entry->aref, new_adj); 457 458 uvm_map_entry_link(map, entry, new_entry); 459 460 if (UVM_ET_ISSUBMAP(entry)) { 461 /* ... unlikely to happen, but play it safe */ 462 uvm_map_reference(new_entry->object.sub_map); 463 } else { 464 if (UVM_ET_ISOBJ(entry) && 465 entry->object.uvm_obj->pgops && 466 entry->object.uvm_obj->pgops->pgo_reference) 467 entry->object.uvm_obj->pgops->pgo_reference( 468 entry->object.uvm_obj); 469 } 470 } 471 472 473 /* 474 * M A P - m a i n e n t r y p o i n t 475 */ 476 /* 477 * uvm_map: establish a valid mapping in a map 478 * 479 * => assume startp is page aligned. 480 * => assume size is a multiple of PAGE_SIZE. 481 * => assume sys_mmap provides enough of a "hint" to have us skip 482 * over text/data/bss area. 483 * => map must be unlocked (we will lock it) 484 * => <uobj,uoffset> value meanings (4 cases): 485 * [1] <NULL,uoffset> == uoffset is a hint for PMAP_PREFER 486 * [2] <NULL,UVM_UNKNOWN_OFFSET> == don't PMAP_PREFER 487 * [3] <uobj,uoffset> == normal mapping 488 * [4] <uobj,UVM_UNKNOWN_OFFSET> == uvm_map finds offset based on VA 489 * 490 * case [4] is for kernel mappings where we don't know the offset until 491 * we've found a virtual address. note that kernel object offsets are 492 * always relative to vm_map_min(kernel_map). 493 * 494 * => if `align' is non-zero, we try to align the virtual address to 495 * the specified alignment. this is only a hint; if we can't 496 * do it, the address will be unaligned. this is provided as 497 * a mechanism for large pages. 498 * 499 * => XXXCDC: need way to map in external amap? 500 */ 501 502 int 503 uvm_map(map, startp, size, uobj, uoffset, align, flags) 504 struct vm_map *map; 505 vaddr_t *startp; /* IN/OUT */ 506 vsize_t size; 507 struct uvm_object *uobj; 508 voff_t uoffset; 509 vsize_t align; 510 uvm_flag_t flags; 511 { 512 struct vm_map_entry *prev_entry, *new_entry; 513 vm_prot_t prot = UVM_PROTECTION(flags), maxprot = 514 UVM_MAXPROTECTION(flags); 515 vm_inherit_t inherit = UVM_INHERIT(flags); 516 int advice = UVM_ADVICE(flags); 517 UVMHIST_FUNC("uvm_map"); 518 UVMHIST_CALLED(maphist); 519 520 UVMHIST_LOG(maphist, "(map=0x%x, *startp=0x%x, size=%d, flags=0x%x)", 521 map, *startp, size, flags); 522 UVMHIST_LOG(maphist, " uobj/offset 0x%x/%d", uobj, uoffset,0,0); 523 524 /* 525 * step 0: sanity check of protection code 526 */ 527 528 if ((prot & maxprot) != prot) { 529 UVMHIST_LOG(maphist, "<- prot. failure: prot=0x%x, max=0x%x", 530 prot, maxprot,0,0); 531 return EACCES; 532 } 533 534 /* 535 * step 1: figure out where to put new VM range 536 */ 537 538 if (vm_map_lock_try(map) == FALSE) { 539 if (flags & UVM_FLAG_TRYLOCK) 540 return EAGAIN; 541 vm_map_lock(map); /* could sleep here */ 542 } 543 if ((prev_entry = uvm_map_findspace(map, *startp, size, startp, 544 uobj, uoffset, align, flags)) == NULL) { 545 UVMHIST_LOG(maphist,"<- uvm_map_findspace failed!",0,0,0,0); 546 vm_map_unlock(map); 547 return ENOMEM; 548 } 549 550 #ifdef PMAP_GROWKERNEL 551 { 552 /* 553 * If the kernel pmap can't map the requested space, 554 * then allocate more resources for it. 555 */ 556 if (map == kernel_map && uvm_maxkaddr < (*startp + size)) 557 uvm_maxkaddr = pmap_growkernel(*startp + size); 558 } 559 #endif 560 561 UVMCNT_INCR(uvm_map_call); 562 563 /* 564 * if uobj is null, then uoffset is either a VAC hint for PMAP_PREFER 565 * [typically from uvm_map_reserve] or it is UVM_UNKNOWN_OFFSET. in 566 * either case we want to zero it before storing it in the map entry 567 * (because it looks strange and confusing when debugging...) 568 * 569 * if uobj is not null 570 * if uoffset is not UVM_UNKNOWN_OFFSET then we have a normal mapping 571 * and we do not need to change uoffset. 572 * if uoffset is UVM_UNKNOWN_OFFSET then we need to find the offset 573 * now (based on the starting address of the map). this case is 574 * for kernel object mappings where we don't know the offset until 575 * the virtual address is found (with uvm_map_findspace). the 576 * offset is the distance we are from the start of the map. 577 */ 578 579 if (uobj == NULL) { 580 uoffset = 0; 581 } else { 582 if (uoffset == UVM_UNKNOWN_OFFSET) { 583 KASSERT(UVM_OBJ_IS_KERN_OBJECT(uobj)); 584 uoffset = *startp - vm_map_min(kernel_map); 585 } 586 } 587 588 /* 589 * step 2: try and insert in map by extending previous entry, if 590 * possible 591 * XXX: we don't try and pull back the next entry. might be useful 592 * for a stack, but we are currently allocating our stack in advance. 593 */ 594 595 if ((flags & UVM_FLAG_NOMERGE) == 0 && 596 prev_entry->end == *startp && prev_entry != &map->header && 597 prev_entry->object.uvm_obj == uobj) { 598 599 if (uobj && prev_entry->offset + 600 (prev_entry->end - prev_entry->start) != uoffset) 601 goto step3; 602 603 if (UVM_ET_ISSUBMAP(prev_entry)) 604 goto step3; 605 606 if (prev_entry->protection != prot || 607 prev_entry->max_protection != maxprot) 608 goto step3; 609 610 if (prev_entry->inheritance != inherit || 611 prev_entry->advice != advice) 612 goto step3; 613 614 /* wiring status must match (new area is unwired) */ 615 if (VM_MAPENT_ISWIRED(prev_entry)) 616 goto step3; 617 618 /* 619 * can't extend a shared amap. note: no need to lock amap to 620 * look at refs since we don't care about its exact value. 621 * if it is one (i.e. we have only reference) it will stay there 622 */ 623 624 if (prev_entry->aref.ar_amap && 625 amap_refs(prev_entry->aref.ar_amap) != 1) { 626 goto step3; 627 } 628 629 /* got it! */ 630 631 UVMCNT_INCR(map_backmerge); 632 UVMHIST_LOG(maphist," starting back merge", 0, 0, 0, 0); 633 634 /* 635 * drop our reference to uobj since we are extending a reference 636 * that we already have (the ref count can not drop to zero). 637 */ 638 if (uobj && uobj->pgops->pgo_detach) 639 uobj->pgops->pgo_detach(uobj); 640 641 if (prev_entry->aref.ar_amap) { 642 amap_extend(prev_entry, size); 643 } 644 645 prev_entry->end += size; 646 map->size += size; 647 648 UVMHIST_LOG(maphist,"<- done (via backmerge)!", 0, 0, 0, 0); 649 vm_map_unlock(map); 650 return 0; 651 652 } 653 step3: 654 UVMHIST_LOG(maphist," allocating new map entry", 0, 0, 0, 0); 655 656 /* 657 * check for possible forward merge (which we don't do) and count 658 * the number of times we missed a *possible* chance to merge more 659 */ 660 661 if ((flags & UVM_FLAG_NOMERGE) == 0 && 662 prev_entry->next != &map->header && 663 prev_entry->next->start == (*startp + size)) 664 UVMCNT_INCR(map_forwmerge); 665 666 /* 667 * step 3: allocate new entry and link it in 668 */ 669 670 new_entry = uvm_mapent_alloc(map); 671 new_entry->start = *startp; 672 new_entry->end = new_entry->start + size; 673 new_entry->object.uvm_obj = uobj; 674 new_entry->offset = uoffset; 675 676 if (uobj) 677 new_entry->etype = UVM_ET_OBJ; 678 else 679 new_entry->etype = 0; 680 681 if (flags & UVM_FLAG_COPYONW) { 682 new_entry->etype |= UVM_ET_COPYONWRITE; 683 if ((flags & UVM_FLAG_OVERLAY) == 0) 684 new_entry->etype |= UVM_ET_NEEDSCOPY; 685 } 686 687 new_entry->protection = prot; 688 new_entry->max_protection = maxprot; 689 new_entry->inheritance = inherit; 690 new_entry->wired_count = 0; 691 new_entry->advice = advice; 692 if (flags & UVM_FLAG_OVERLAY) { 693 /* 694 * to_add: for BSS we overallocate a little since we 695 * are likely to extend 696 */ 697 vaddr_t to_add = (flags & UVM_FLAG_AMAPPAD) ? 698 UVM_AMAP_CHUNK << PAGE_SHIFT : 0; 699 struct vm_amap *amap = amap_alloc(size, to_add, M_WAITOK); 700 new_entry->aref.ar_pageoff = 0; 701 new_entry->aref.ar_amap = amap; 702 } else { 703 new_entry->aref.ar_pageoff = 0; 704 new_entry->aref.ar_amap = NULL; 705 } 706 uvm_map_entry_link(map, prev_entry, new_entry); 707 map->size += size; 708 709 /* 710 * Update the free space hint 711 */ 712 713 if ((map->first_free == prev_entry) && 714 (prev_entry->end >= new_entry->start)) 715 map->first_free = new_entry; 716 717 UVMHIST_LOG(maphist,"<- done!", 0, 0, 0, 0); 718 vm_map_unlock(map); 719 return 0; 720 } 721 722 /* 723 * uvm_map_lookup_entry: find map entry at or before an address 724 * 725 * => map must at least be read-locked by caller 726 * => entry is returned in "entry" 727 * => return value is true if address is in the returned entry 728 */ 729 730 boolean_t 731 uvm_map_lookup_entry(map, address, entry) 732 struct vm_map *map; 733 vaddr_t address; 734 struct vm_map_entry **entry; /* OUT */ 735 { 736 struct vm_map_entry *cur; 737 struct vm_map_entry *last; 738 UVMHIST_FUNC("uvm_map_lookup_entry"); 739 UVMHIST_CALLED(maphist); 740 741 UVMHIST_LOG(maphist,"(map=0x%x,addr=0x%x,ent=0x%x)", 742 map, address, entry, 0); 743 744 /* 745 * start looking either from the head of the 746 * list, or from the hint. 747 */ 748 749 simple_lock(&map->hint_lock); 750 cur = map->hint; 751 simple_unlock(&map->hint_lock); 752 753 if (cur == &map->header) 754 cur = cur->next; 755 756 UVMCNT_INCR(uvm_mlk_call); 757 if (address >= cur->start) { 758 759 /* 760 * go from hint to end of list. 761 * 762 * but first, make a quick check to see if 763 * we are already looking at the entry we 764 * want (which is usually the case). 765 * note also that we don't need to save the hint 766 * here... it is the same hint (unless we are 767 * at the header, in which case the hint didn't 768 * buy us anything anyway). 769 */ 770 771 last = &map->header; 772 if ((cur != last) && (cur->end > address)) { 773 UVMCNT_INCR(uvm_mlk_hint); 774 *entry = cur; 775 UVMHIST_LOG(maphist,"<- got it via hint (0x%x)", 776 cur, 0, 0, 0); 777 return (TRUE); 778 } 779 } else { 780 781 /* 782 * go from start to hint, *inclusively* 783 */ 784 785 last = cur->next; 786 cur = map->header.next; 787 } 788 789 /* 790 * search linearly 791 */ 792 793 while (cur != last) { 794 if (cur->end > address) { 795 if (address >= cur->start) { 796 /* 797 * save this lookup for future 798 * hints, and return 799 */ 800 801 *entry = cur; 802 SAVE_HINT(map, map->hint, cur); 803 UVMHIST_LOG(maphist,"<- search got it (0x%x)", 804 cur, 0, 0, 0); 805 return (TRUE); 806 } 807 break; 808 } 809 cur = cur->next; 810 } 811 *entry = cur->prev; 812 SAVE_HINT(map, map->hint, *entry); 813 UVMHIST_LOG(maphist,"<- failed!",0,0,0,0); 814 return (FALSE); 815 } 816 817 /* 818 * uvm_map_findspace: find "length" sized space in "map". 819 * 820 * => "hint" is a hint about where we want it, unless FINDSPACE_FIXED is 821 * set (in which case we insist on using "hint"). 822 * => "result" is VA returned 823 * => uobj/uoffset are to be used to handle VAC alignment, if required 824 * => if `align' is non-zero, we attempt to align to that value. 825 * => caller must at least have read-locked map 826 * => returns NULL on failure, or pointer to prev. map entry if success 827 * => note this is a cross between the old vm_map_findspace and vm_map_find 828 */ 829 830 struct vm_map_entry * 831 uvm_map_findspace(map, hint, length, result, uobj, uoffset, align, flags) 832 struct vm_map *map; 833 vaddr_t hint; 834 vsize_t length; 835 vaddr_t *result; /* OUT */ 836 struct uvm_object *uobj; 837 voff_t uoffset; 838 vsize_t align; 839 int flags; 840 { 841 struct vm_map_entry *entry, *next, *tmp; 842 vaddr_t end, orig_hint; 843 UVMHIST_FUNC("uvm_map_findspace"); 844 UVMHIST_CALLED(maphist); 845 846 UVMHIST_LOG(maphist, "(map=0x%x, hint=0x%x, len=%d, flags=0x%x)", 847 map, hint, length, flags); 848 KASSERT((align & (align - 1)) == 0); 849 KASSERT((flags & UVM_FLAG_FIXED) == 0 || align == 0); 850 851 /* 852 * remember the original hint. if we are aligning, then we 853 * may have to try again with no alignment constraint if 854 * we fail the first time. 855 */ 856 857 orig_hint = hint; 858 if (hint < map->min_offset) { /* check ranges ... */ 859 if (flags & UVM_FLAG_FIXED) { 860 UVMHIST_LOG(maphist,"<- VA below map range",0,0,0,0); 861 return(NULL); 862 } 863 hint = map->min_offset; 864 } 865 if (hint > map->max_offset) { 866 UVMHIST_LOG(maphist,"<- VA 0x%x > range [0x%x->0x%x]", 867 hint, map->min_offset, map->max_offset, 0); 868 return(NULL); 869 } 870 871 /* 872 * Look for the first possible address; if there's already 873 * something at this address, we have to start after it. 874 */ 875 876 if ((flags & UVM_FLAG_FIXED) == 0 && hint == map->min_offset) { 877 if ((entry = map->first_free) != &map->header) 878 hint = entry->end; 879 } else { 880 if (uvm_map_lookup_entry(map, hint, &tmp)) { 881 /* "hint" address already in use ... */ 882 if (flags & UVM_FLAG_FIXED) { 883 UVMHIST_LOG(maphist,"<- fixed & VA in use", 884 0, 0, 0, 0); 885 return(NULL); 886 } 887 hint = tmp->end; 888 } 889 entry = tmp; 890 } 891 892 /* 893 * Look through the rest of the map, trying to fit a new region in 894 * the gap between existing regions, or after the very last region. 895 * note: entry->end = base VA of current gap, 896 * next->start = VA of end of current gap 897 */ 898 899 for (;; hint = (entry = next)->end) { 900 901 /* 902 * Find the end of the proposed new region. Be sure we didn't 903 * go beyond the end of the map, or wrap around the address; 904 * if so, we lose. Otherwise, if this is the last entry, or 905 * if the proposed new region fits before the next entry, we 906 * win. 907 */ 908 909 #ifdef PMAP_PREFER 910 /* 911 * push hint forward as needed to avoid VAC alias problems. 912 * we only do this if a valid offset is specified. 913 */ 914 915 if ((flags & UVM_FLAG_FIXED) == 0 && 916 uoffset != UVM_UNKNOWN_OFFSET) 917 PMAP_PREFER(uoffset, &hint); 918 #endif 919 if (align != 0) { 920 if ((hint & (align - 1)) != 0) 921 hint = roundup(hint, align); 922 /* 923 * XXX Should we PMAP_PREFER() here again? 924 */ 925 } 926 end = hint + length; 927 if (end > map->max_offset || end < hint) { 928 UVMHIST_LOG(maphist,"<- failed (off end)", 0,0,0,0); 929 if (align != 0) { 930 UVMHIST_LOG(maphist, 931 "calling recursively, no align", 932 0,0,0,0); 933 return (uvm_map_findspace(map, orig_hint, 934 length, result, uobj, uoffset, 0, flags)); 935 } 936 return (NULL); 937 } 938 next = entry->next; 939 if (next == &map->header || next->start >= end) 940 break; 941 if (flags & UVM_FLAG_FIXED) { 942 UVMHIST_LOG(maphist,"<- fixed mapping failed", 0,0,0,0); 943 return(NULL); /* only one shot at it ... */ 944 } 945 } 946 SAVE_HINT(map, map->hint, entry); 947 *result = hint; 948 UVMHIST_LOG(maphist,"<- got it! (result=0x%x)", hint, 0,0,0); 949 return (entry); 950 } 951 952 /* 953 * U N M A P - m a i n h e l p e r f u n c t i o n s 954 */ 955 956 /* 957 * uvm_unmap_remove: remove mappings from a vm_map (from "start" up to "stop") 958 * 959 * => caller must check alignment and size 960 * => map must be locked by caller 961 * => we return a list of map entries that we've remove from the map 962 * in "entry_list" 963 */ 964 965 void 966 uvm_unmap_remove(map, start, end, entry_list) 967 struct vm_map *map; 968 vaddr_t start, end; 969 struct vm_map_entry **entry_list; /* OUT */ 970 { 971 struct vm_map_entry *entry, *first_entry, *next; 972 vaddr_t len; 973 UVMHIST_FUNC("uvm_unmap_remove"); UVMHIST_CALLED(maphist); 974 975 UVMHIST_LOG(maphist,"(map=0x%x, start=0x%x, end=0x%x)", 976 map, start, end, 0); 977 978 VM_MAP_RANGE_CHECK(map, start, end); 979 980 /* 981 * find first entry 982 */ 983 984 if (uvm_map_lookup_entry(map, start, &first_entry) == TRUE) { 985 /* clip and go... */ 986 entry = first_entry; 987 UVM_MAP_CLIP_START(map, entry, start); 988 /* critical! prevents stale hint */ 989 SAVE_HINT(map, entry, entry->prev); 990 } else { 991 entry = first_entry->next; 992 } 993 994 /* 995 * Save the free space hint 996 */ 997 998 if (map->first_free->start >= start) 999 map->first_free = entry->prev; 1000 1001 /* 1002 * note: we now re-use first_entry for a different task. we remove 1003 * a number of map entries from the map and save them in a linked 1004 * list headed by "first_entry". once we remove them from the map 1005 * the caller should unlock the map and drop the references to the 1006 * backing objects [c.f. uvm_unmap_detach]. the object is to 1007 * separate unmapping from reference dropping. why? 1008 * [1] the map has to be locked for unmapping 1009 * [2] the map need not be locked for reference dropping 1010 * [3] dropping references may trigger pager I/O, and if we hit 1011 * a pager that does synchronous I/O we may have to wait for it. 1012 * [4] we would like all waiting for I/O to occur with maps unlocked 1013 * so that we don't block other threads. 1014 */ 1015 1016 first_entry = NULL; 1017 *entry_list = NULL; /* to be safe */ 1018 1019 /* 1020 * break up the area into map entry sized regions and unmap. note 1021 * that all mappings have to be removed before we can even consider 1022 * dropping references to amaps or VM objects (otherwise we could end 1023 * up with a mapping to a page on the free list which would be very bad) 1024 */ 1025 1026 while ((entry != &map->header) && (entry->start < end)) { 1027 1028 UVM_MAP_CLIP_END(map, entry, end); 1029 next = entry->next; 1030 len = entry->end - entry->start; 1031 1032 /* 1033 * unwire before removing addresses from the pmap; otherwise 1034 * unwiring will put the entries back into the pmap (XXX). 1035 */ 1036 1037 if (VM_MAPENT_ISWIRED(entry)) 1038 uvm_map_entry_unwire(map, entry); 1039 1040 /* 1041 * special case: handle mappings to anonymous kernel objects. 1042 * we want to free these pages right away... 1043 */ 1044 1045 if (UVM_ET_ISOBJ(entry) && 1046 UVM_OBJ_IS_KERN_OBJECT(entry->object.uvm_obj)) { 1047 KASSERT(vm_map_pmap(map) == pmap_kernel()); 1048 1049 /* 1050 * note: kernel object mappings are currently used in 1051 * two ways: 1052 * [1] "normal" mappings of pages in the kernel object 1053 * [2] uvm_km_valloc'd allocations in which we 1054 * pmap_enter in some non-kernel-object page 1055 * (e.g. vmapbuf). 1056 * 1057 * for case [1], we need to remove the mapping from 1058 * the pmap and then remove the page from the kernel 1059 * object (because, once pages in a kernel object are 1060 * unmapped they are no longer needed, unlike, say, 1061 * a vnode where you might want the data to persist 1062 * until flushed out of a queue). 1063 * 1064 * for case [2], we need to remove the mapping from 1065 * the pmap. there shouldn't be any pages at the 1066 * specified offset in the kernel object [but it 1067 * doesn't hurt to call uvm_km_pgremove just to be 1068 * safe?] 1069 * 1070 * uvm_km_pgremove currently does the following: 1071 * for pages in the kernel object in range: 1072 * - drops the swap slot 1073 * - uvm_pagefree the page 1074 * 1075 * note there is version of uvm_km_pgremove() that 1076 * is used for "intrsafe" objects. 1077 */ 1078 1079 /* 1080 * remove mappings from pmap and drop the pages 1081 * from the object. offsets are always relative 1082 * to vm_map_min(kernel_map). 1083 */ 1084 1085 if (UVM_OBJ_IS_INTRSAFE_OBJECT(entry->object.uvm_obj)) { 1086 pmap_kremove(entry->start, len); 1087 uvm_km_pgremove_intrsafe(entry->object.uvm_obj, 1088 entry->start - vm_map_min(kernel_map), 1089 entry->end - vm_map_min(kernel_map)); 1090 } else { 1091 pmap_remove(pmap_kernel(), entry->start, 1092 entry->start + len); 1093 uvm_km_pgremove(entry->object.uvm_obj, 1094 entry->start - vm_map_min(kernel_map), 1095 entry->end - vm_map_min(kernel_map)); 1096 } 1097 1098 /* 1099 * null out kernel_object reference, we've just 1100 * dropped it 1101 */ 1102 1103 entry->etype &= ~UVM_ET_OBJ; 1104 entry->object.uvm_obj = NULL; /* to be safe */ 1105 1106 } else { 1107 1108 /* 1109 * remove mappings the standard way. 1110 */ 1111 1112 pmap_remove(map->pmap, entry->start, entry->end); 1113 } 1114 1115 /* 1116 * remove entry from map and put it on our list of entries 1117 * that we've nuked. then go do next entry. 1118 */ 1119 1120 UVMHIST_LOG(maphist, " removed map entry 0x%x", entry, 0, 0,0); 1121 1122 /* critical! prevents stale hint */ 1123 SAVE_HINT(map, entry, entry->prev); 1124 1125 uvm_map_entry_unlink(map, entry); 1126 map->size -= len; 1127 entry->next = first_entry; 1128 first_entry = entry; 1129 entry = next; /* next entry, please */ 1130 } 1131 pmap_update(); 1132 1133 /* 1134 * now we've cleaned up the map and are ready for the caller to drop 1135 * references to the mapped objects. 1136 */ 1137 1138 *entry_list = first_entry; 1139 UVMHIST_LOG(maphist,"<- done!", 0, 0, 0, 0); 1140 } 1141 1142 /* 1143 * uvm_unmap_detach: drop references in a chain of map entries 1144 * 1145 * => we will free the map entries as we traverse the list. 1146 */ 1147 1148 void 1149 uvm_unmap_detach(first_entry, flags) 1150 struct vm_map_entry *first_entry; 1151 int flags; 1152 { 1153 struct vm_map_entry *next_entry; 1154 UVMHIST_FUNC("uvm_unmap_detach"); UVMHIST_CALLED(maphist); 1155 1156 while (first_entry) { 1157 KASSERT(!VM_MAPENT_ISWIRED(first_entry)); 1158 UVMHIST_LOG(maphist, 1159 " detach 0x%x: amap=0x%x, obj=0x%x, submap?=%d", 1160 first_entry, first_entry->aref.ar_amap, 1161 first_entry->object.uvm_obj, 1162 UVM_ET_ISSUBMAP(first_entry)); 1163 1164 /* 1165 * drop reference to amap, if we've got one 1166 */ 1167 1168 if (first_entry->aref.ar_amap) 1169 uvm_map_unreference_amap(first_entry, flags); 1170 1171 /* 1172 * drop reference to our backing object, if we've got one 1173 */ 1174 1175 if (UVM_ET_ISSUBMAP(first_entry)) { 1176 /* ... unlikely to happen, but play it safe */ 1177 uvm_map_deallocate(first_entry->object.sub_map); 1178 } else { 1179 if (UVM_ET_ISOBJ(first_entry) && 1180 first_entry->object.uvm_obj->pgops->pgo_detach) 1181 first_entry->object.uvm_obj->pgops-> 1182 pgo_detach(first_entry->object.uvm_obj); 1183 } 1184 next_entry = first_entry->next; 1185 uvm_mapent_free(first_entry); 1186 first_entry = next_entry; 1187 } 1188 UVMHIST_LOG(maphist, "<- done", 0,0,0,0); 1189 } 1190 1191 /* 1192 * E X T R A C T I O N F U N C T I O N S 1193 */ 1194 1195 /* 1196 * uvm_map_reserve: reserve space in a vm_map for future use. 1197 * 1198 * => we reserve space in a map by putting a dummy map entry in the 1199 * map (dummy means obj=NULL, amap=NULL, prot=VM_PROT_NONE) 1200 * => map should be unlocked (we will write lock it) 1201 * => we return true if we were able to reserve space 1202 * => XXXCDC: should be inline? 1203 */ 1204 1205 int 1206 uvm_map_reserve(map, size, offset, align, raddr) 1207 struct vm_map *map; 1208 vsize_t size; 1209 vaddr_t offset; /* hint for pmap_prefer */ 1210 vsize_t align; /* alignment hint */ 1211 vaddr_t *raddr; /* IN:hint, OUT: reserved VA */ 1212 { 1213 UVMHIST_FUNC("uvm_map_reserve"); UVMHIST_CALLED(maphist); 1214 1215 UVMHIST_LOG(maphist, "(map=0x%x, size=0x%x, offset=0x%x,addr=0x%x)", 1216 map,size,offset,raddr); 1217 1218 size = round_page(size); 1219 if (*raddr < vm_map_min(map)) 1220 *raddr = vm_map_min(map); /* hint */ 1221 1222 /* 1223 * reserve some virtual space. 1224 */ 1225 1226 if (uvm_map(map, raddr, size, NULL, offset, 0, 1227 UVM_MAPFLAG(UVM_PROT_NONE, UVM_PROT_NONE, UVM_INH_NONE, 1228 UVM_ADV_RANDOM, UVM_FLAG_NOMERGE)) != 0) { 1229 UVMHIST_LOG(maphist, "<- done (no VM)", 0,0,0,0); 1230 return (FALSE); 1231 } 1232 1233 UVMHIST_LOG(maphist, "<- done (*raddr=0x%x)", *raddr,0,0,0); 1234 return (TRUE); 1235 } 1236 1237 /* 1238 * uvm_map_replace: replace a reserved (blank) area of memory with 1239 * real mappings. 1240 * 1241 * => caller must WRITE-LOCK the map 1242 * => we return TRUE if replacement was a success 1243 * => we expect the newents chain to have nnewents entrys on it and 1244 * we expect newents->prev to point to the last entry on the list 1245 * => note newents is allowed to be NULL 1246 */ 1247 1248 int 1249 uvm_map_replace(map, start, end, newents, nnewents) 1250 struct vm_map *map; 1251 vaddr_t start, end; 1252 struct vm_map_entry *newents; 1253 int nnewents; 1254 { 1255 struct vm_map_entry *oldent, *last; 1256 1257 /* 1258 * first find the blank map entry at the specified address 1259 */ 1260 1261 if (!uvm_map_lookup_entry(map, start, &oldent)) { 1262 return(FALSE); 1263 } 1264 1265 /* 1266 * check to make sure we have a proper blank entry 1267 */ 1268 1269 if (oldent->start != start || oldent->end != end || 1270 oldent->object.uvm_obj != NULL || oldent->aref.ar_amap != NULL) { 1271 return (FALSE); 1272 } 1273 1274 #ifdef DIAGNOSTIC 1275 1276 /* 1277 * sanity check the newents chain 1278 */ 1279 1280 { 1281 struct vm_map_entry *tmpent = newents; 1282 int nent = 0; 1283 vaddr_t cur = start; 1284 1285 while (tmpent) { 1286 nent++; 1287 if (tmpent->start < cur) 1288 panic("uvm_map_replace1"); 1289 if (tmpent->start > tmpent->end || tmpent->end > end) { 1290 printf("tmpent->start=0x%lx, tmpent->end=0x%lx, end=0x%lx\n", 1291 tmpent->start, tmpent->end, end); 1292 panic("uvm_map_replace2"); 1293 } 1294 cur = tmpent->end; 1295 if (tmpent->next) { 1296 if (tmpent->next->prev != tmpent) 1297 panic("uvm_map_replace3"); 1298 } else { 1299 if (newents->prev != tmpent) 1300 panic("uvm_map_replace4"); 1301 } 1302 tmpent = tmpent->next; 1303 } 1304 if (nent != nnewents) 1305 panic("uvm_map_replace5"); 1306 } 1307 #endif 1308 1309 /* 1310 * map entry is a valid blank! replace it. (this does all the 1311 * work of map entry link/unlink...). 1312 */ 1313 1314 if (newents) { 1315 last = newents->prev; 1316 1317 /* critical: flush stale hints out of map */ 1318 SAVE_HINT(map, map->hint, newents); 1319 if (map->first_free == oldent) 1320 map->first_free = last; 1321 1322 last->next = oldent->next; 1323 last->next->prev = last; 1324 newents->prev = oldent->prev; 1325 newents->prev->next = newents; 1326 map->nentries = map->nentries + (nnewents - 1); 1327 1328 } else { 1329 1330 /* critical: flush stale hints out of map */ 1331 SAVE_HINT(map, map->hint, oldent->prev); 1332 if (map->first_free == oldent) 1333 map->first_free = oldent->prev; 1334 1335 /* NULL list of new entries: just remove the old one */ 1336 uvm_map_entry_unlink(map, oldent); 1337 } 1338 1339 1340 /* 1341 * now we can free the old blank entry, unlock the map and return. 1342 */ 1343 1344 uvm_mapent_free(oldent); 1345 return(TRUE); 1346 } 1347 1348 /* 1349 * uvm_map_extract: extract a mapping from a map and put it somewhere 1350 * (maybe removing the old mapping) 1351 * 1352 * => maps should be unlocked (we will write lock them) 1353 * => returns 0 on success, error code otherwise 1354 * => start must be page aligned 1355 * => len must be page sized 1356 * => flags: 1357 * UVM_EXTRACT_REMOVE: remove mappings from srcmap 1358 * UVM_EXTRACT_CONTIG: abort if unmapped area (advisory only) 1359 * UVM_EXTRACT_QREF: for a temporary extraction do quick obj refs 1360 * UVM_EXTRACT_FIXPROT: set prot to maxprot as we go 1361 * >>>NOTE: if you set REMOVE, you are not allowed to use CONTIG or QREF!<<< 1362 * >>>NOTE: QREF's must be unmapped via the QREF path, thus should only 1363 * be used from within the kernel in a kernel level map <<< 1364 */ 1365 1366 int 1367 uvm_map_extract(srcmap, start, len, dstmap, dstaddrp, flags) 1368 struct vm_map *srcmap, *dstmap; 1369 vaddr_t start, *dstaddrp; 1370 vsize_t len; 1371 int flags; 1372 { 1373 vaddr_t dstaddr, end, newend, oldoffset, fudge, orig_fudge, 1374 oldstart; 1375 struct vm_map_entry *chain, *endchain, *entry, *orig_entry, *newentry, 1376 *deadentry, *oldentry; 1377 vsize_t elen; 1378 int nchain, error, copy_ok; 1379 UVMHIST_FUNC("uvm_map_extract"); UVMHIST_CALLED(maphist); 1380 1381 UVMHIST_LOG(maphist,"(srcmap=0x%x,start=0x%x, len=0x%x", srcmap, start, 1382 len,0); 1383 UVMHIST_LOG(maphist," ...,dstmap=0x%x, flags=0x%x)", dstmap,flags,0,0); 1384 1385 /* 1386 * step 0: sanity check: start must be on a page boundary, length 1387 * must be page sized. can't ask for CONTIG/QREF if you asked for 1388 * REMOVE. 1389 */ 1390 1391 KASSERT((start & PAGE_MASK) == 0 && (len & PAGE_MASK) == 0); 1392 KASSERT((flags & UVM_EXTRACT_REMOVE) == 0 || 1393 (flags & (UVM_EXTRACT_CONTIG|UVM_EXTRACT_QREF)) == 0); 1394 1395 /* 1396 * step 1: reserve space in the target map for the extracted area 1397 */ 1398 1399 dstaddr = vm_map_min(dstmap); 1400 if (uvm_map_reserve(dstmap, len, start, 0, &dstaddr) == FALSE) 1401 return(ENOMEM); 1402 *dstaddrp = dstaddr; /* pass address back to caller */ 1403 UVMHIST_LOG(maphist, " dstaddr=0x%x", dstaddr,0,0,0); 1404 1405 /* 1406 * step 2: setup for the extraction process loop by init'ing the 1407 * map entry chain, locking src map, and looking up the first useful 1408 * entry in the map. 1409 */ 1410 1411 end = start + len; 1412 newend = dstaddr + len; 1413 chain = endchain = NULL; 1414 nchain = 0; 1415 vm_map_lock(srcmap); 1416 1417 if (uvm_map_lookup_entry(srcmap, start, &entry)) { 1418 1419 /* "start" is within an entry */ 1420 if (flags & UVM_EXTRACT_QREF) { 1421 1422 /* 1423 * for quick references we don't clip the entry, so 1424 * the entry may map space "before" the starting 1425 * virtual address... this is the "fudge" factor 1426 * (which can be non-zero only the first time 1427 * through the "while" loop in step 3). 1428 */ 1429 1430 fudge = start - entry->start; 1431 } else { 1432 1433 /* 1434 * normal reference: we clip the map to fit (thus 1435 * fudge is zero) 1436 */ 1437 1438 UVM_MAP_CLIP_START(srcmap, entry, start); 1439 SAVE_HINT(srcmap, srcmap->hint, entry->prev); 1440 fudge = 0; 1441 } 1442 } else { 1443 1444 /* "start" is not within an entry ... skip to next entry */ 1445 if (flags & UVM_EXTRACT_CONTIG) { 1446 error = EINVAL; 1447 goto bad; /* definite hole here ... */ 1448 } 1449 1450 entry = entry->next; 1451 fudge = 0; 1452 } 1453 1454 /* save values from srcmap for step 6 */ 1455 orig_entry = entry; 1456 orig_fudge = fudge; 1457 1458 /* 1459 * step 3: now start looping through the map entries, extracting 1460 * as we go. 1461 */ 1462 1463 while (entry->start < end && entry != &srcmap->header) { 1464 1465 /* if we are not doing a quick reference, clip it */ 1466 if ((flags & UVM_EXTRACT_QREF) == 0) 1467 UVM_MAP_CLIP_END(srcmap, entry, end); 1468 1469 /* clear needs_copy (allow chunking) */ 1470 if (UVM_ET_ISNEEDSCOPY(entry)) { 1471 if (fudge) 1472 oldstart = entry->start; 1473 else 1474 oldstart = 0; /* XXX: gcc */ 1475 amap_copy(srcmap, entry, M_NOWAIT, TRUE, start, end); 1476 if (UVM_ET_ISNEEDSCOPY(entry)) { /* failed? */ 1477 error = ENOMEM; 1478 goto bad; 1479 } 1480 1481 /* amap_copy could clip (during chunk)! update fudge */ 1482 if (fudge) { 1483 fudge = fudge - (entry->start - oldstart); 1484 orig_fudge = fudge; 1485 } 1486 } 1487 1488 /* calculate the offset of this from "start" */ 1489 oldoffset = (entry->start + fudge) - start; 1490 1491 /* allocate a new map entry */ 1492 newentry = uvm_mapent_alloc(dstmap); 1493 if (newentry == NULL) { 1494 error = ENOMEM; 1495 goto bad; 1496 } 1497 1498 /* set up new map entry */ 1499 newentry->next = NULL; 1500 newentry->prev = endchain; 1501 newentry->start = dstaddr + oldoffset; 1502 newentry->end = 1503 newentry->start + (entry->end - (entry->start + fudge)); 1504 if (newentry->end > newend || newentry->end < newentry->start) 1505 newentry->end = newend; 1506 newentry->object.uvm_obj = entry->object.uvm_obj; 1507 if (newentry->object.uvm_obj) { 1508 if (newentry->object.uvm_obj->pgops->pgo_reference) 1509 newentry->object.uvm_obj->pgops-> 1510 pgo_reference(newentry->object.uvm_obj); 1511 newentry->offset = entry->offset + fudge; 1512 } else { 1513 newentry->offset = 0; 1514 } 1515 newentry->etype = entry->etype; 1516 newentry->protection = (flags & UVM_EXTRACT_FIXPROT) ? 1517 entry->max_protection : entry->protection; 1518 newentry->max_protection = entry->max_protection; 1519 newentry->inheritance = entry->inheritance; 1520 newentry->wired_count = 0; 1521 newentry->aref.ar_amap = entry->aref.ar_amap; 1522 if (newentry->aref.ar_amap) { 1523 newentry->aref.ar_pageoff = 1524 entry->aref.ar_pageoff + (fudge >> PAGE_SHIFT); 1525 uvm_map_reference_amap(newentry, AMAP_SHARED | 1526 ((flags & UVM_EXTRACT_QREF) ? AMAP_REFALL : 0)); 1527 } else { 1528 newentry->aref.ar_pageoff = 0; 1529 } 1530 newentry->advice = entry->advice; 1531 1532 /* now link it on the chain */ 1533 nchain++; 1534 if (endchain == NULL) { 1535 chain = endchain = newentry; 1536 } else { 1537 endchain->next = newentry; 1538 endchain = newentry; 1539 } 1540 1541 /* end of 'while' loop! */ 1542 if ((flags & UVM_EXTRACT_CONTIG) && entry->end < end && 1543 (entry->next == &srcmap->header || 1544 entry->next->start != entry->end)) { 1545 error = EINVAL; 1546 goto bad; 1547 } 1548 entry = entry->next; 1549 fudge = 0; 1550 } 1551 1552 /* 1553 * step 4: close off chain (in format expected by uvm_map_replace) 1554 */ 1555 1556 if (chain) 1557 chain->prev = endchain; 1558 1559 /* 1560 * step 5: attempt to lock the dest map so we can pmap_copy. 1561 * note usage of copy_ok: 1562 * 1 => dstmap locked, pmap_copy ok, and we "replace" here (step 5) 1563 * 0 => dstmap unlocked, NO pmap_copy, and we will "replace" in step 7 1564 */ 1565 1566 if (srcmap == dstmap || vm_map_lock_try(dstmap) == TRUE) { 1567 copy_ok = 1; 1568 if (!uvm_map_replace(dstmap, dstaddr, dstaddr+len, chain, 1569 nchain)) { 1570 if (srcmap != dstmap) 1571 vm_map_unlock(dstmap); 1572 error = EIO; 1573 goto bad; 1574 } 1575 } else { 1576 copy_ok = 0; 1577 /* replace defered until step 7 */ 1578 } 1579 1580 /* 1581 * step 6: traverse the srcmap a second time to do the following: 1582 * - if we got a lock on the dstmap do pmap_copy 1583 * - if UVM_EXTRACT_REMOVE remove the entries 1584 * we make use of orig_entry and orig_fudge (saved in step 2) 1585 */ 1586 1587 if (copy_ok || (flags & UVM_EXTRACT_REMOVE)) { 1588 1589 /* purge possible stale hints from srcmap */ 1590 if (flags & UVM_EXTRACT_REMOVE) { 1591 SAVE_HINT(srcmap, srcmap->hint, orig_entry->prev); 1592 if (srcmap->first_free->start >= start) 1593 srcmap->first_free = orig_entry->prev; 1594 } 1595 1596 entry = orig_entry; 1597 fudge = orig_fudge; 1598 deadentry = NULL; /* for UVM_EXTRACT_REMOVE */ 1599 1600 while (entry->start < end && entry != &srcmap->header) { 1601 if (copy_ok) { 1602 oldoffset = (entry->start + fudge) - start; 1603 elen = MIN(end, entry->end) - 1604 (entry->start + fudge); 1605 pmap_copy(dstmap->pmap, srcmap->pmap, 1606 dstaddr + oldoffset, elen, 1607 entry->start + fudge); 1608 } 1609 1610 /* we advance "entry" in the following if statement */ 1611 if (flags & UVM_EXTRACT_REMOVE) { 1612 pmap_remove(srcmap->pmap, entry->start, 1613 entry->end); 1614 oldentry = entry; /* save entry */ 1615 entry = entry->next; /* advance */ 1616 uvm_map_entry_unlink(srcmap, oldentry); 1617 /* add to dead list */ 1618 oldentry->next = deadentry; 1619 deadentry = oldentry; 1620 } else { 1621 entry = entry->next; /* advance */ 1622 } 1623 1624 /* end of 'while' loop */ 1625 fudge = 0; 1626 } 1627 pmap_update(); 1628 1629 /* 1630 * unlock dstmap. we will dispose of deadentry in 1631 * step 7 if needed 1632 */ 1633 1634 if (copy_ok && srcmap != dstmap) 1635 vm_map_unlock(dstmap); 1636 1637 } else { 1638 deadentry = NULL; 1639 } 1640 1641 /* 1642 * step 7: we are done with the source map, unlock. if copy_ok 1643 * is 0 then we have not replaced the dummy mapping in dstmap yet 1644 * and we need to do so now. 1645 */ 1646 1647 vm_map_unlock(srcmap); 1648 if ((flags & UVM_EXTRACT_REMOVE) && deadentry) 1649 uvm_unmap_detach(deadentry, 0); /* dispose of old entries */ 1650 1651 /* now do the replacement if we didn't do it in step 5 */ 1652 if (copy_ok == 0) { 1653 vm_map_lock(dstmap); 1654 error = uvm_map_replace(dstmap, dstaddr, dstaddr+len, chain, 1655 nchain); 1656 vm_map_unlock(dstmap); 1657 1658 if (error == FALSE) { 1659 error = EIO; 1660 goto bad2; 1661 } 1662 } 1663 return(0); 1664 1665 /* 1666 * bad: failure recovery 1667 */ 1668 bad: 1669 vm_map_unlock(srcmap); 1670 bad2: /* src already unlocked */ 1671 if (chain) 1672 uvm_unmap_detach(chain, 1673 (flags & UVM_EXTRACT_QREF) ? AMAP_REFALL : 0); 1674 uvm_unmap(dstmap, dstaddr, dstaddr+len); /* ??? */ 1675 return(error); 1676 } 1677 1678 /* end of extraction functions */ 1679 1680 /* 1681 * uvm_map_submap: punch down part of a map into a submap 1682 * 1683 * => only the kernel_map is allowed to be submapped 1684 * => the purpose of submapping is to break up the locking granularity 1685 * of a larger map 1686 * => the range specified must have been mapped previously with a uvm_map() 1687 * call [with uobj==NULL] to create a blank map entry in the main map. 1688 * [And it had better still be blank!] 1689 * => maps which contain submaps should never be copied or forked. 1690 * => to remove a submap, use uvm_unmap() on the main map 1691 * and then uvm_map_deallocate() the submap. 1692 * => main map must be unlocked. 1693 * => submap must have been init'd and have a zero reference count. 1694 * [need not be locked as we don't actually reference it] 1695 */ 1696 1697 int 1698 uvm_map_submap(map, start, end, submap) 1699 struct vm_map *map, *submap; 1700 vaddr_t start, end; 1701 { 1702 struct vm_map_entry *entry; 1703 int error; 1704 1705 vm_map_lock(map); 1706 VM_MAP_RANGE_CHECK(map, start, end); 1707 1708 if (uvm_map_lookup_entry(map, start, &entry)) { 1709 UVM_MAP_CLIP_START(map, entry, start); 1710 UVM_MAP_CLIP_END(map, entry, end); /* to be safe */ 1711 } else { 1712 entry = NULL; 1713 } 1714 1715 if (entry != NULL && 1716 entry->start == start && entry->end == end && 1717 entry->object.uvm_obj == NULL && entry->aref.ar_amap == NULL && 1718 !UVM_ET_ISCOPYONWRITE(entry) && !UVM_ET_ISNEEDSCOPY(entry)) { 1719 entry->etype |= UVM_ET_SUBMAP; 1720 entry->object.sub_map = submap; 1721 entry->offset = 0; 1722 uvm_map_reference(submap); 1723 error = 0; 1724 } else { 1725 error = EINVAL; 1726 } 1727 vm_map_unlock(map); 1728 return error; 1729 } 1730 1731 1732 /* 1733 * uvm_map_protect: change map protection 1734 * 1735 * => set_max means set max_protection. 1736 * => map must be unlocked. 1737 */ 1738 1739 #define MASK(entry) (UVM_ET_ISCOPYONWRITE(entry) ? \ 1740 ~VM_PROT_WRITE : VM_PROT_ALL) 1741 1742 int 1743 uvm_map_protect(map, start, end, new_prot, set_max) 1744 struct vm_map *map; 1745 vaddr_t start, end; 1746 vm_prot_t new_prot; 1747 boolean_t set_max; 1748 { 1749 struct vm_map_entry *current, *entry; 1750 int error = 0; 1751 UVMHIST_FUNC("uvm_map_protect"); UVMHIST_CALLED(maphist); 1752 UVMHIST_LOG(maphist,"(map=0x%x,start=0x%x,end=0x%x,new_prot=0x%x)", 1753 map, start, end, new_prot); 1754 1755 vm_map_lock(map); 1756 VM_MAP_RANGE_CHECK(map, start, end); 1757 if (uvm_map_lookup_entry(map, start, &entry)) { 1758 UVM_MAP_CLIP_START(map, entry, start); 1759 } else { 1760 entry = entry->next; 1761 } 1762 1763 /* 1764 * make a first pass to check for protection violations. 1765 */ 1766 1767 current = entry; 1768 while ((current != &map->header) && (current->start < end)) { 1769 if (UVM_ET_ISSUBMAP(current)) { 1770 error = EINVAL; 1771 goto out; 1772 } 1773 if ((new_prot & current->max_protection) != new_prot) { 1774 error = EACCES; 1775 goto out; 1776 } 1777 current = current->next; 1778 } 1779 1780 /* go back and fix up protections (no need to clip this time). */ 1781 1782 current = entry; 1783 while ((current != &map->header) && (current->start < end)) { 1784 vm_prot_t old_prot; 1785 1786 UVM_MAP_CLIP_END(map, current, end); 1787 old_prot = current->protection; 1788 if (set_max) 1789 current->protection = 1790 (current->max_protection = new_prot) & old_prot; 1791 else 1792 current->protection = new_prot; 1793 1794 /* 1795 * update physical map if necessary. worry about copy-on-write 1796 * here -- CHECK THIS XXX 1797 */ 1798 1799 if (current->protection != old_prot) { 1800 /* update pmap! */ 1801 pmap_protect(map->pmap, current->start, current->end, 1802 current->protection & MASK(entry)); 1803 } 1804 1805 /* 1806 * If the map is configured to lock any future mappings, 1807 * wire this entry now if the old protection was VM_PROT_NONE 1808 * and the new protection is not VM_PROT_NONE. 1809 */ 1810 1811 if ((map->flags & VM_MAP_WIREFUTURE) != 0 && 1812 VM_MAPENT_ISWIRED(entry) == 0 && 1813 old_prot == VM_PROT_NONE && 1814 new_prot != VM_PROT_NONE) { 1815 if (uvm_map_pageable(map, entry->start, 1816 entry->end, FALSE, 1817 UVM_LK_ENTER|UVM_LK_EXIT) != 0) { 1818 1819 /* 1820 * If locking the entry fails, remember the 1821 * error if it's the first one. Note we 1822 * still continue setting the protection in 1823 * the map, but will return the error 1824 * condition regardless. 1825 * 1826 * XXX Ignore what the actual error is, 1827 * XXX just call it a resource shortage 1828 * XXX so that it doesn't get confused 1829 * XXX what uvm_map_protect() itself would 1830 * XXX normally return. 1831 */ 1832 1833 error = ENOMEM; 1834 } 1835 } 1836 current = current->next; 1837 } 1838 pmap_update(); 1839 1840 out: 1841 vm_map_unlock(map); 1842 UVMHIST_LOG(maphist, "<- done, error=%d",error,0,0,0); 1843 return error; 1844 } 1845 1846 #undef MASK 1847 1848 /* 1849 * uvm_map_inherit: set inheritance code for range of addrs in map. 1850 * 1851 * => map must be unlocked 1852 * => note that the inherit code is used during a "fork". see fork 1853 * code for details. 1854 */ 1855 1856 int 1857 uvm_map_inherit(map, start, end, new_inheritance) 1858 struct vm_map *map; 1859 vaddr_t start; 1860 vaddr_t end; 1861 vm_inherit_t new_inheritance; 1862 { 1863 struct vm_map_entry *entry, *temp_entry; 1864 UVMHIST_FUNC("uvm_map_inherit"); UVMHIST_CALLED(maphist); 1865 UVMHIST_LOG(maphist,"(map=0x%x,start=0x%x,end=0x%x,new_inh=0x%x)", 1866 map, start, end, new_inheritance); 1867 1868 switch (new_inheritance) { 1869 case MAP_INHERIT_NONE: 1870 case MAP_INHERIT_COPY: 1871 case MAP_INHERIT_SHARE: 1872 break; 1873 default: 1874 UVMHIST_LOG(maphist,"<- done (INVALID ARG)",0,0,0,0); 1875 return EINVAL; 1876 } 1877 1878 vm_map_lock(map); 1879 VM_MAP_RANGE_CHECK(map, start, end); 1880 if (uvm_map_lookup_entry(map, start, &temp_entry)) { 1881 entry = temp_entry; 1882 UVM_MAP_CLIP_START(map, entry, start); 1883 } else { 1884 entry = temp_entry->next; 1885 } 1886 while ((entry != &map->header) && (entry->start < end)) { 1887 UVM_MAP_CLIP_END(map, entry, end); 1888 entry->inheritance = new_inheritance; 1889 entry = entry->next; 1890 } 1891 vm_map_unlock(map); 1892 UVMHIST_LOG(maphist,"<- done (OK)",0,0,0,0); 1893 return 0; 1894 } 1895 1896 /* 1897 * uvm_map_advice: set advice code for range of addrs in map. 1898 * 1899 * => map must be unlocked 1900 */ 1901 1902 int 1903 uvm_map_advice(map, start, end, new_advice) 1904 struct vm_map *map; 1905 vaddr_t start; 1906 vaddr_t end; 1907 int new_advice; 1908 { 1909 struct vm_map_entry *entry, *temp_entry; 1910 UVMHIST_FUNC("uvm_map_advice"); UVMHIST_CALLED(maphist); 1911 UVMHIST_LOG(maphist,"(map=0x%x,start=0x%x,end=0x%x,new_adv=0x%x)", 1912 map, start, end, new_advice); 1913 1914 vm_map_lock(map); 1915 VM_MAP_RANGE_CHECK(map, start, end); 1916 if (uvm_map_lookup_entry(map, start, &temp_entry)) { 1917 entry = temp_entry; 1918 UVM_MAP_CLIP_START(map, entry, start); 1919 } else { 1920 entry = temp_entry->next; 1921 } 1922 1923 /* 1924 * XXXJRT: disallow holes? 1925 */ 1926 1927 while ((entry != &map->header) && (entry->start < end)) { 1928 UVM_MAP_CLIP_END(map, entry, end); 1929 1930 switch (new_advice) { 1931 case MADV_NORMAL: 1932 case MADV_RANDOM: 1933 case MADV_SEQUENTIAL: 1934 /* nothing special here */ 1935 break; 1936 1937 default: 1938 vm_map_unlock(map); 1939 UVMHIST_LOG(maphist,"<- done (INVALID ARG)",0,0,0,0); 1940 return EINVAL; 1941 } 1942 entry->advice = new_advice; 1943 entry = entry->next; 1944 } 1945 1946 vm_map_unlock(map); 1947 UVMHIST_LOG(maphist,"<- done (OK)",0,0,0,0); 1948 return 0; 1949 } 1950 1951 /* 1952 * uvm_map_pageable: sets the pageability of a range in a map. 1953 * 1954 * => wires map entries. should not be used for transient page locking. 1955 * for that, use uvm_fault_wire()/uvm_fault_unwire() (see uvm_vslock()). 1956 * => regions sepcified as not pageable require lock-down (wired) memory 1957 * and page tables. 1958 * => map must never be read-locked 1959 * => if islocked is TRUE, map is already write-locked 1960 * => we always unlock the map, since we must downgrade to a read-lock 1961 * to call uvm_fault_wire() 1962 * => XXXCDC: check this and try and clean it up. 1963 */ 1964 1965 int 1966 uvm_map_pageable(map, start, end, new_pageable, lockflags) 1967 struct vm_map *map; 1968 vaddr_t start, end; 1969 boolean_t new_pageable; 1970 int lockflags; 1971 { 1972 struct vm_map_entry *entry, *start_entry, *failed_entry; 1973 int rv; 1974 #ifdef DIAGNOSTIC 1975 u_int timestamp_save; 1976 #endif 1977 UVMHIST_FUNC("uvm_map_pageable"); UVMHIST_CALLED(maphist); 1978 UVMHIST_LOG(maphist,"(map=0x%x,start=0x%x,end=0x%x,new_pageable=0x%x)", 1979 map, start, end, new_pageable); 1980 KASSERT(map->flags & VM_MAP_PAGEABLE); 1981 1982 if ((lockflags & UVM_LK_ENTER) == 0) 1983 vm_map_lock(map); 1984 VM_MAP_RANGE_CHECK(map, start, end); 1985 1986 /* 1987 * only one pageability change may take place at one time, since 1988 * uvm_fault_wire assumes it will be called only once for each 1989 * wiring/unwiring. therefore, we have to make sure we're actually 1990 * changing the pageability for the entire region. we do so before 1991 * making any changes. 1992 */ 1993 1994 if (uvm_map_lookup_entry(map, start, &start_entry) == FALSE) { 1995 if ((lockflags & UVM_LK_EXIT) == 0) 1996 vm_map_unlock(map); 1997 1998 UVMHIST_LOG(maphist,"<- done (fault)",0,0,0,0); 1999 return EFAULT; 2000 } 2001 entry = start_entry; 2002 2003 /* 2004 * handle wiring and unwiring separately. 2005 */ 2006 2007 if (new_pageable) { /* unwire */ 2008 UVM_MAP_CLIP_START(map, entry, start); 2009 2010 /* 2011 * unwiring. first ensure that the range to be unwired is 2012 * really wired down and that there are no holes. 2013 */ 2014 2015 while ((entry != &map->header) && (entry->start < end)) { 2016 if (entry->wired_count == 0 || 2017 (entry->end < end && 2018 (entry->next == &map->header || 2019 entry->next->start > entry->end))) { 2020 if ((lockflags & UVM_LK_EXIT) == 0) 2021 vm_map_unlock(map); 2022 UVMHIST_LOG(maphist, "<- done (INVAL)",0,0,0,0); 2023 return EINVAL; 2024 } 2025 entry = entry->next; 2026 } 2027 2028 /* 2029 * POSIX 1003.1b - a single munlock call unlocks a region, 2030 * regardless of the number of mlock calls made on that 2031 * region. 2032 */ 2033 2034 entry = start_entry; 2035 while ((entry != &map->header) && (entry->start < end)) { 2036 UVM_MAP_CLIP_END(map, entry, end); 2037 if (VM_MAPENT_ISWIRED(entry)) 2038 uvm_map_entry_unwire(map, entry); 2039 entry = entry->next; 2040 } 2041 if ((lockflags & UVM_LK_EXIT) == 0) 2042 vm_map_unlock(map); 2043 UVMHIST_LOG(maphist,"<- done (OK UNWIRE)",0,0,0,0); 2044 return 0; 2045 } 2046 2047 /* 2048 * wire case: in two passes [XXXCDC: ugly block of code here] 2049 * 2050 * 1: holding the write lock, we create any anonymous maps that need 2051 * to be created. then we clip each map entry to the region to 2052 * be wired and increment its wiring count. 2053 * 2054 * 2: we downgrade to a read lock, and call uvm_fault_wire to fault 2055 * in the pages for any newly wired area (wired_count == 1). 2056 * 2057 * downgrading to a read lock for uvm_fault_wire avoids a possible 2058 * deadlock with another thread that may have faulted on one of 2059 * the pages to be wired (it would mark the page busy, blocking 2060 * us, then in turn block on the map lock that we hold). because 2061 * of problems in the recursive lock package, we cannot upgrade 2062 * to a write lock in vm_map_lookup. thus, any actions that 2063 * require the write lock must be done beforehand. because we 2064 * keep the read lock on the map, the copy-on-write status of the 2065 * entries we modify here cannot change. 2066 */ 2067 2068 while ((entry != &map->header) && (entry->start < end)) { 2069 if (VM_MAPENT_ISWIRED(entry) == 0) { /* not already wired? */ 2070 2071 /* 2072 * perform actions of vm_map_lookup that need the 2073 * write lock on the map: create an anonymous map 2074 * for a copy-on-write region, or an anonymous map 2075 * for a zero-fill region. (XXXCDC: submap case 2076 * ok?) 2077 */ 2078 2079 if (!UVM_ET_ISSUBMAP(entry)) { /* not submap */ 2080 if (UVM_ET_ISNEEDSCOPY(entry) && 2081 ((entry->protection & VM_PROT_WRITE) || 2082 (entry->object.uvm_obj == NULL))) { 2083 amap_copy(map, entry, M_WAITOK, TRUE, 2084 start, end); 2085 /* XXXCDC: wait OK? */ 2086 } 2087 } 2088 } 2089 UVM_MAP_CLIP_START(map, entry, start); 2090 UVM_MAP_CLIP_END(map, entry, end); 2091 entry->wired_count++; 2092 2093 /* 2094 * Check for holes 2095 */ 2096 2097 if (entry->protection == VM_PROT_NONE || 2098 (entry->end < end && 2099 (entry->next == &map->header || 2100 entry->next->start > entry->end))) { 2101 2102 /* 2103 * found one. amap creation actions do not need to 2104 * be undone, but the wired counts need to be restored. 2105 */ 2106 2107 while (entry != &map->header && entry->end > start) { 2108 entry->wired_count--; 2109 entry = entry->prev; 2110 } 2111 if ((lockflags & UVM_LK_EXIT) == 0) 2112 vm_map_unlock(map); 2113 UVMHIST_LOG(maphist,"<- done (INVALID WIRE)",0,0,0,0); 2114 return EINVAL; 2115 } 2116 entry = entry->next; 2117 } 2118 2119 /* 2120 * Pass 2. 2121 */ 2122 2123 #ifdef DIAGNOSTIC 2124 timestamp_save = map->timestamp; 2125 #endif 2126 vm_map_busy(map); 2127 vm_map_downgrade(map); 2128 2129 rv = 0; 2130 entry = start_entry; 2131 while (entry != &map->header && entry->start < end) { 2132 if (entry->wired_count == 1) { 2133 rv = uvm_fault_wire(map, entry->start, entry->end, 2134 entry->protection); 2135 if (rv) { 2136 2137 /* 2138 * wiring failed. break out of the loop. 2139 * we'll clean up the map below, once we 2140 * have a write lock again. 2141 */ 2142 2143 break; 2144 } 2145 } 2146 entry = entry->next; 2147 } 2148 2149 if (rv) { /* failed? */ 2150 2151 /* 2152 * Get back to an exclusive (write) lock. 2153 */ 2154 2155 vm_map_upgrade(map); 2156 vm_map_unbusy(map); 2157 2158 #ifdef DIAGNOSTIC 2159 if (timestamp_save != map->timestamp) 2160 panic("uvm_map_pageable: stale map"); 2161 #endif 2162 2163 /* 2164 * first drop the wiring count on all the entries 2165 * which haven't actually been wired yet. 2166 */ 2167 2168 failed_entry = entry; 2169 while (entry != &map->header && entry->start < end) { 2170 entry->wired_count--; 2171 entry = entry->next; 2172 } 2173 2174 /* 2175 * now, unwire all the entries that were successfully 2176 * wired above. 2177 */ 2178 2179 entry = start_entry; 2180 while (entry != failed_entry) { 2181 entry->wired_count--; 2182 if (VM_MAPENT_ISWIRED(entry) == 0) 2183 uvm_map_entry_unwire(map, entry); 2184 entry = entry->next; 2185 } 2186 if ((lockflags & UVM_LK_EXIT) == 0) 2187 vm_map_unlock(map); 2188 UVMHIST_LOG(maphist, "<- done (RV=%d)", rv,0,0,0); 2189 return(rv); 2190 } 2191 2192 /* We are holding a read lock here. */ 2193 if ((lockflags & UVM_LK_EXIT) == 0) { 2194 vm_map_unbusy(map); 2195 vm_map_unlock_read(map); 2196 } else { 2197 2198 /* 2199 * Get back to an exclusive (write) lock. 2200 */ 2201 2202 vm_map_upgrade(map); 2203 vm_map_unbusy(map); 2204 } 2205 2206 UVMHIST_LOG(maphist,"<- done (OK WIRE)",0,0,0,0); 2207 return 0; 2208 } 2209 2210 /* 2211 * uvm_map_pageable_all: special case of uvm_map_pageable - affects 2212 * all mapped regions. 2213 * 2214 * => map must not be locked. 2215 * => if no flags are specified, all regions are unwired. 2216 * => XXXJRT: has some of the same problems as uvm_map_pageable() above. 2217 */ 2218 2219 int 2220 uvm_map_pageable_all(map, flags, limit) 2221 struct vm_map *map; 2222 int flags; 2223 vsize_t limit; 2224 { 2225 struct vm_map_entry *entry, *failed_entry; 2226 vsize_t size; 2227 int rv; 2228 #ifdef DIAGNOSTIC 2229 u_int timestamp_save; 2230 #endif 2231 UVMHIST_FUNC("uvm_map_pageable_all"); UVMHIST_CALLED(maphist); 2232 UVMHIST_LOG(maphist,"(map=0x%x,flags=0x%x)", map, flags, 0, 0); 2233 2234 KASSERT(map->flags & VM_MAP_PAGEABLE); 2235 2236 vm_map_lock(map); 2237 2238 /* 2239 * handle wiring and unwiring separately. 2240 */ 2241 2242 if (flags == 0) { /* unwire */ 2243 2244 /* 2245 * POSIX 1003.1b -- munlockall unlocks all regions, 2246 * regardless of how many times mlockall has been called. 2247 */ 2248 2249 for (entry = map->header.next; entry != &map->header; 2250 entry = entry->next) { 2251 if (VM_MAPENT_ISWIRED(entry)) 2252 uvm_map_entry_unwire(map, entry); 2253 } 2254 vm_map_modflags(map, 0, VM_MAP_WIREFUTURE); 2255 vm_map_unlock(map); 2256 UVMHIST_LOG(maphist,"<- done (OK UNWIRE)",0,0,0,0); 2257 return 0; 2258 } 2259 2260 if (flags & MCL_FUTURE) { 2261 2262 /* 2263 * must wire all future mappings; remember this. 2264 */ 2265 2266 vm_map_modflags(map, VM_MAP_WIREFUTURE, 0); 2267 } 2268 2269 if ((flags & MCL_CURRENT) == 0) { 2270 2271 /* 2272 * no more work to do! 2273 */ 2274 2275 UVMHIST_LOG(maphist,"<- done (OK no wire)",0,0,0,0); 2276 vm_map_unlock(map); 2277 return 0; 2278 } 2279 2280 /* 2281 * wire case: in three passes [XXXCDC: ugly block of code here] 2282 * 2283 * 1: holding the write lock, count all pages mapped by non-wired 2284 * entries. if this would cause us to go over our limit, we fail. 2285 * 2286 * 2: still holding the write lock, we create any anonymous maps that 2287 * need to be created. then we increment its wiring count. 2288 * 2289 * 3: we downgrade to a read lock, and call uvm_fault_wire to fault 2290 * in the pages for any newly wired area (wired_count == 1). 2291 * 2292 * downgrading to a read lock for uvm_fault_wire avoids a possible 2293 * deadlock with another thread that may have faulted on one of 2294 * the pages to be wired (it would mark the page busy, blocking 2295 * us, then in turn block on the map lock that we hold). because 2296 * of problems in the recursive lock package, we cannot upgrade 2297 * to a write lock in vm_map_lookup. thus, any actions that 2298 * require the write lock must be done beforehand. because we 2299 * keep the read lock on the map, the copy-on-write status of the 2300 * entries we modify here cannot change. 2301 */ 2302 2303 for (size = 0, entry = map->header.next; entry != &map->header; 2304 entry = entry->next) { 2305 if (entry->protection != VM_PROT_NONE && 2306 VM_MAPENT_ISWIRED(entry) == 0) { /* not already wired? */ 2307 size += entry->end - entry->start; 2308 } 2309 } 2310 2311 if (atop(size) + uvmexp.wired > uvmexp.wiredmax) { 2312 vm_map_unlock(map); 2313 return ENOMEM; 2314 } 2315 2316 /* XXX non-pmap_wired_count case must be handled by caller */ 2317 #ifdef pmap_wired_count 2318 if (limit != 0 && 2319 (size + ptoa(pmap_wired_count(vm_map_pmap(map))) > limit)) { 2320 vm_map_unlock(map); 2321 return ENOMEM; 2322 } 2323 #endif 2324 2325 /* 2326 * Pass 2. 2327 */ 2328 2329 for (entry = map->header.next; entry != &map->header; 2330 entry = entry->next) { 2331 if (entry->protection == VM_PROT_NONE) 2332 continue; 2333 if (VM_MAPENT_ISWIRED(entry) == 0) { /* not already wired? */ 2334 2335 /* 2336 * perform actions of vm_map_lookup that need the 2337 * write lock on the map: create an anonymous map 2338 * for a copy-on-write region, or an anonymous map 2339 * for a zero-fill region. (XXXCDC: submap case 2340 * ok?) 2341 */ 2342 2343 if (!UVM_ET_ISSUBMAP(entry)) { /* not submap */ 2344 if (UVM_ET_ISNEEDSCOPY(entry) && 2345 ((entry->protection & VM_PROT_WRITE) || 2346 (entry->object.uvm_obj == NULL))) { 2347 amap_copy(map, entry, M_WAITOK, TRUE, 2348 entry->start, entry->end); 2349 /* XXXCDC: wait OK? */ 2350 } 2351 } 2352 } 2353 entry->wired_count++; 2354 } 2355 2356 /* 2357 * Pass 3. 2358 */ 2359 2360 #ifdef DIAGNOSTIC 2361 timestamp_save = map->timestamp; 2362 #endif 2363 vm_map_busy(map); 2364 vm_map_downgrade(map); 2365 2366 rv = 0; 2367 for (entry = map->header.next; entry != &map->header; 2368 entry = entry->next) { 2369 if (entry->wired_count == 1) { 2370 rv = uvm_fault_wire(map, entry->start, entry->end, 2371 entry->protection); 2372 if (rv) { 2373 2374 /* 2375 * wiring failed. break out of the loop. 2376 * we'll clean up the map below, once we 2377 * have a write lock again. 2378 */ 2379 2380 break; 2381 } 2382 } 2383 } 2384 2385 if (rv) { 2386 2387 /* 2388 * Get back an exclusive (write) lock. 2389 */ 2390 2391 vm_map_upgrade(map); 2392 vm_map_unbusy(map); 2393 2394 #ifdef DIAGNOSTIC 2395 if (timestamp_save != map->timestamp) 2396 panic("uvm_map_pageable_all: stale map"); 2397 #endif 2398 2399 /* 2400 * first drop the wiring count on all the entries 2401 * which haven't actually been wired yet. 2402 * 2403 * Skip VM_PROT_NONE entries like we did above. 2404 */ 2405 2406 failed_entry = entry; 2407 for (/* nothing */; entry != &map->header; 2408 entry = entry->next) { 2409 if (entry->protection == VM_PROT_NONE) 2410 continue; 2411 entry->wired_count--; 2412 } 2413 2414 /* 2415 * now, unwire all the entries that were successfully 2416 * wired above. 2417 * 2418 * Skip VM_PROT_NONE entries like we did above. 2419 */ 2420 2421 for (entry = map->header.next; entry != failed_entry; 2422 entry = entry->next) { 2423 if (entry->protection == VM_PROT_NONE) 2424 continue; 2425 entry->wired_count--; 2426 if (VM_MAPENT_ISWIRED(entry)) 2427 uvm_map_entry_unwire(map, entry); 2428 } 2429 vm_map_unlock(map); 2430 UVMHIST_LOG(maphist,"<- done (RV=%d)", rv,0,0,0); 2431 return (rv); 2432 } 2433 2434 /* We are holding a read lock here. */ 2435 vm_map_unbusy(map); 2436 vm_map_unlock_read(map); 2437 2438 UVMHIST_LOG(maphist,"<- done (OK WIRE)",0,0,0,0); 2439 return 0; 2440 } 2441 2442 /* 2443 * uvm_map_clean: clean out a map range 2444 * 2445 * => valid flags: 2446 * if (flags & PGO_CLEANIT): dirty pages are cleaned first 2447 * if (flags & PGO_SYNCIO): dirty pages are written synchronously 2448 * if (flags & PGO_DEACTIVATE): any cached pages are deactivated after clean 2449 * if (flags & PGO_FREE): any cached pages are freed after clean 2450 * => returns an error if any part of the specified range isn't mapped 2451 * => never a need to flush amap layer since the anonymous memory has 2452 * no permanent home, but may deactivate pages there 2453 * => called from sys_msync() and sys_madvise() 2454 * => caller must not write-lock map (read OK). 2455 * => we may sleep while cleaning if SYNCIO [with map read-locked] 2456 */ 2457 2458 int 2459 uvm_map_clean(map, start, end, flags) 2460 struct vm_map *map; 2461 vaddr_t start, end; 2462 int flags; 2463 { 2464 struct vm_map_entry *current, *entry; 2465 struct uvm_object *uobj; 2466 struct vm_amap *amap; 2467 struct vm_anon *anon; 2468 struct vm_page *pg; 2469 vaddr_t offset; 2470 vsize_t size; 2471 int rv, error, refs; 2472 UVMHIST_FUNC("uvm_map_clean"); UVMHIST_CALLED(maphist); 2473 2474 UVMHIST_LOG(maphist,"(map=0x%x,start=0x%x,end=0x%x,flags=0x%x)", 2475 map, start, end, flags); 2476 KASSERT((flags & (PGO_FREE|PGO_DEACTIVATE)) != 2477 (PGO_FREE|PGO_DEACTIVATE)); 2478 2479 vm_map_lock_read(map); 2480 VM_MAP_RANGE_CHECK(map, start, end); 2481 if (uvm_map_lookup_entry(map, start, &entry) == FALSE) { 2482 vm_map_unlock_read(map); 2483 return EFAULT; 2484 } 2485 2486 /* 2487 * Make a first pass to check for holes. 2488 */ 2489 2490 for (current = entry; current->start < end; current = current->next) { 2491 if (UVM_ET_ISSUBMAP(current)) { 2492 vm_map_unlock_read(map); 2493 return EINVAL; 2494 } 2495 if (end <= current->end) { 2496 break; 2497 } 2498 if (current->end != current->next->start) { 2499 vm_map_unlock_read(map); 2500 return EFAULT; 2501 } 2502 } 2503 2504 error = 0; 2505 for (current = entry; start < end; current = current->next) { 2506 amap = current->aref.ar_amap; /* top layer */ 2507 uobj = current->object.uvm_obj; /* bottom layer */ 2508 KASSERT(start >= current->start); 2509 2510 /* 2511 * No amap cleaning necessary if: 2512 * 2513 * (1) There's no amap. 2514 * 2515 * (2) We're not deactivating or freeing pages. 2516 */ 2517 2518 if (amap == NULL || (flags & (PGO_DEACTIVATE|PGO_FREE)) == 0) 2519 goto flush_object; 2520 2521 amap_lock(amap); 2522 offset = start - current->start; 2523 size = MIN(end, current->end) - start; 2524 for ( ; size != 0; size -= PAGE_SIZE, offset += PAGE_SIZE) { 2525 anon = amap_lookup(¤t->aref, offset); 2526 if (anon == NULL) 2527 continue; 2528 2529 simple_lock(&anon->an_lock); 2530 2531 pg = anon->u.an_page; 2532 if (pg == NULL) { 2533 simple_unlock(&anon->an_lock); 2534 continue; 2535 } 2536 2537 switch (flags & (PGO_CLEANIT|PGO_FREE|PGO_DEACTIVATE)) { 2538 2539 /* 2540 * XXX In these first 3 cases, we always just 2541 * XXX deactivate the page. We may want to 2542 * XXX handle the different cases more 2543 * XXX specifically, in the future. 2544 */ 2545 2546 case PGO_CLEANIT|PGO_FREE: 2547 case PGO_CLEANIT|PGO_DEACTIVATE: 2548 case PGO_DEACTIVATE: 2549 deactivate_it: 2550 /* skip the page if it's loaned or wired */ 2551 if (pg->loan_count != 0 || 2552 pg->wire_count != 0) { 2553 simple_unlock(&anon->an_lock); 2554 continue; 2555 } 2556 2557 uvm_lock_pageq(); 2558 2559 /* 2560 * skip the page if it's not actually owned 2561 * by the anon (may simply be loaned to the 2562 * anon). 2563 */ 2564 2565 if ((pg->pqflags & PQ_ANON) == 0) { 2566 KASSERT(pg->uobject == NULL); 2567 uvm_unlock_pageq(); 2568 simple_unlock(&anon->an_lock); 2569 continue; 2570 } 2571 KASSERT(pg->uanon == anon); 2572 2573 /* ...and deactivate the page. */ 2574 pmap_clear_reference(pg); 2575 uvm_pagedeactivate(pg); 2576 2577 uvm_unlock_pageq(); 2578 simple_unlock(&anon->an_lock); 2579 continue; 2580 2581 case PGO_FREE: 2582 2583 /* 2584 * If there are multiple references to 2585 * the amap, just deactivate the page. 2586 */ 2587 2588 if (amap_refs(amap) > 1) 2589 goto deactivate_it; 2590 2591 /* XXX skip the page if it's wired */ 2592 if (pg->wire_count != 0) { 2593 simple_unlock(&anon->an_lock); 2594 continue; 2595 } 2596 amap_unadd(¤t->aref, offset); 2597 refs = --anon->an_ref; 2598 simple_unlock(&anon->an_lock); 2599 if (refs == 0) 2600 uvm_anfree(anon); 2601 continue; 2602 2603 default: 2604 panic("uvm_map_clean: wierd flags"); 2605 } 2606 } 2607 amap_unlock(amap); 2608 2609 flush_object: 2610 /* 2611 * flush pages if we've got a valid backing object. 2612 */ 2613 2614 offset = current->offset + (start - current->start); 2615 size = MIN(end, current->end) - start; 2616 if (uobj != NULL) { 2617 simple_lock(&uobj->vmobjlock); 2618 rv = uobj->pgops->pgo_flush(uobj, offset, 2619 offset + size, flags); 2620 simple_unlock(&uobj->vmobjlock); 2621 2622 if (rv == FALSE) 2623 error = EIO; 2624 } 2625 start += size; 2626 } 2627 vm_map_unlock_read(map); 2628 return (error); 2629 } 2630 2631 2632 /* 2633 * uvm_map_checkprot: check protection in map 2634 * 2635 * => must allow specified protection in a fully allocated region. 2636 * => map must be read or write locked by caller. 2637 */ 2638 2639 boolean_t 2640 uvm_map_checkprot(map, start, end, protection) 2641 struct vm_map * map; 2642 vaddr_t start, end; 2643 vm_prot_t protection; 2644 { 2645 struct vm_map_entry *entry; 2646 struct vm_map_entry *tmp_entry; 2647 2648 if (!uvm_map_lookup_entry(map, start, &tmp_entry)) { 2649 return(FALSE); 2650 } 2651 entry = tmp_entry; 2652 while (start < end) { 2653 if (entry == &map->header) { 2654 return(FALSE); 2655 } 2656 2657 /* 2658 * no holes allowed 2659 */ 2660 2661 if (start < entry->start) { 2662 return(FALSE); 2663 } 2664 2665 /* 2666 * check protection associated with entry 2667 */ 2668 2669 if ((entry->protection & protection) != protection) { 2670 return(FALSE); 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 uvm_map_setup(&vm->vm_map, min, max, pageable ? VM_MAP_PAGEABLE : 0); 2716 if (pmap) 2717 pmap_reference(pmap); 2718 else 2719 pmap = pmap_create(); 2720 vm->vm_map.pmap = pmap; 2721 vm->vm_refcnt = 1; 2722 UVMHIST_LOG(maphist,"<- done",0,0,0,0); 2723 } 2724 2725 /* 2726 * uvmspace_share: share a vmspace between two proceses 2727 * 2728 * - XXX: no locking on vmspace 2729 * - used for vfork, threads(?) 2730 */ 2731 2732 void 2733 uvmspace_share(p1, p2) 2734 struct proc *p1, *p2; 2735 { 2736 p2->p_vmspace = p1->p_vmspace; 2737 p1->p_vmspace->vm_refcnt++; 2738 } 2739 2740 /* 2741 * uvmspace_unshare: ensure that process "p" has its own, unshared, vmspace 2742 * 2743 * - XXX: no locking on vmspace 2744 */ 2745 2746 void 2747 uvmspace_unshare(p) 2748 struct proc *p; 2749 { 2750 struct vmspace *nvm, *ovm = p->p_vmspace; 2751 2752 if (ovm->vm_refcnt == 1) 2753 /* nothing to do: vmspace isn't shared in the first place */ 2754 return; 2755 2756 /* make a new vmspace, still holding old one */ 2757 nvm = uvmspace_fork(ovm); 2758 2759 pmap_deactivate(p); /* unbind old vmspace */ 2760 p->p_vmspace = nvm; 2761 pmap_activate(p); /* switch to new vmspace */ 2762 2763 uvmspace_free(ovm); /* drop reference to old vmspace */ 2764 } 2765 2766 /* 2767 * uvmspace_exec: the process wants to exec a new program 2768 * 2769 * - XXX: no locking on vmspace 2770 */ 2771 2772 void 2773 uvmspace_exec(p, start, end) 2774 struct proc *p; 2775 vaddr_t start, end; 2776 { 2777 struct vmspace *nvm, *ovm = p->p_vmspace; 2778 struct vm_map *map = &ovm->vm_map; 2779 2780 #ifdef __sparc__ 2781 /* XXX cgd 960926: the sparc #ifdef should be a MD hook */ 2782 kill_user_windows(p); /* before stack addresses go away */ 2783 #endif 2784 2785 /* 2786 * see if more than one process is using this vmspace... 2787 */ 2788 2789 if (ovm->vm_refcnt == 1) { 2790 2791 /* 2792 * if p is the only process using its vmspace then we can safely 2793 * recycle that vmspace for the program that is being exec'd. 2794 */ 2795 2796 #ifdef SYSVSHM 2797 /* 2798 * SYSV SHM semantics require us to kill all segments on an exec 2799 */ 2800 2801 if (ovm->vm_shm) 2802 shmexit(ovm); 2803 #endif 2804 2805 /* 2806 * POSIX 1003.1b -- "lock future mappings" is revoked 2807 * when a process execs another program image. 2808 */ 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 2818 uvm_unmap(map, map->min_offset, map->max_offset); 2819 2820 /* 2821 * resize the map 2822 */ 2823 2824 vm_map_lock(map); 2825 map->min_offset = start; 2826 map->max_offset = end; 2827 vm_map_unlock(map); 2828 } else { 2829 2830 /* 2831 * p's vmspace is being shared, so we can't reuse it for p since 2832 * it is still being used for others. allocate a new vmspace 2833 * for p 2834 */ 2835 2836 nvm = uvmspace_alloc(start, end, 2837 (map->flags & VM_MAP_PAGEABLE) ? TRUE : FALSE); 2838 2839 /* 2840 * install new vmspace and drop our ref to the old one. 2841 */ 2842 2843 pmap_deactivate(p); 2844 p->p_vmspace = nvm; 2845 pmap_activate(p); 2846 2847 uvmspace_free(ovm); 2848 } 2849 } 2850 2851 /* 2852 * uvmspace_free: free a vmspace data structure 2853 * 2854 * - XXX: no locking on vmspace 2855 */ 2856 2857 void 2858 uvmspace_free(vm) 2859 struct vmspace *vm; 2860 { 2861 struct vm_map_entry *dead_entries; 2862 UVMHIST_FUNC("uvmspace_free"); UVMHIST_CALLED(maphist); 2863 2864 UVMHIST_LOG(maphist,"(vm=0x%x) ref=%d", vm, vm->vm_refcnt,0,0); 2865 if (--vm->vm_refcnt == 0) { 2866 2867 /* 2868 * lock the map, to wait out all other references to it. delete 2869 * all of the mappings and pages they hold, then call the pmap 2870 * module to reclaim anything left. 2871 */ 2872 2873 #ifdef SYSVSHM 2874 /* Get rid of any SYSV shared memory segments. */ 2875 if (vm->vm_shm != NULL) 2876 shmexit(vm); 2877 #endif 2878 vm_map_lock(&vm->vm_map); 2879 if (vm->vm_map.nentries) { 2880 uvm_unmap_remove(&vm->vm_map, 2881 vm->vm_map.min_offset, vm->vm_map.max_offset, 2882 &dead_entries); 2883 if (dead_entries != NULL) 2884 uvm_unmap_detach(dead_entries, 0); 2885 } 2886 pmap_destroy(vm->vm_map.pmap); 2887 vm->vm_map.pmap = NULL; 2888 pool_put(&uvm_vmspace_pool, vm); 2889 } 2890 UVMHIST_LOG(maphist,"<- done", 0,0,0,0); 2891 } 2892 2893 /* 2894 * F O R K - m a i n e n t r y p o i n t 2895 */ 2896 /* 2897 * uvmspace_fork: fork a process' main map 2898 * 2899 * => create a new vmspace for child process from parent. 2900 * => parent's map must not be locked. 2901 */ 2902 2903 struct vmspace * 2904 uvmspace_fork(vm1) 2905 struct vmspace *vm1; 2906 { 2907 struct vmspace *vm2; 2908 struct vm_map *old_map = &vm1->vm_map; 2909 struct vm_map *new_map; 2910 struct vm_map_entry *old_entry; 2911 struct vm_map_entry *new_entry; 2912 pmap_t new_pmap; 2913 boolean_t protect_child; 2914 UVMHIST_FUNC("uvmspace_fork"); UVMHIST_CALLED(maphist); 2915 2916 vm_map_lock(old_map); 2917 2918 vm2 = uvmspace_alloc(old_map->min_offset, old_map->max_offset, 2919 (old_map->flags & VM_MAP_PAGEABLE) ? TRUE : FALSE); 2920 memcpy(&vm2->vm_startcopy, &vm1->vm_startcopy, 2921 (caddr_t) (vm1 + 1) - (caddr_t) &vm1->vm_startcopy); 2922 new_map = &vm2->vm_map; /* XXX */ 2923 new_pmap = new_map->pmap; 2924 2925 old_entry = old_map->header.next; 2926 2927 /* 2928 * go entry-by-entry 2929 */ 2930 2931 while (old_entry != &old_map->header) { 2932 2933 /* 2934 * first, some sanity checks on the old entry 2935 */ 2936 2937 KASSERT(!UVM_ET_ISSUBMAP(old_entry)); 2938 KASSERT(UVM_ET_ISCOPYONWRITE(old_entry) || 2939 !UVM_ET_ISNEEDSCOPY(old_entry)); 2940 2941 switch (old_entry->inheritance) { 2942 case MAP_INHERIT_NONE: 2943 2944 /* 2945 * drop the mapping 2946 */ 2947 2948 break; 2949 2950 case MAP_INHERIT_SHARE: 2951 2952 /* 2953 * share the mapping: this means we want the old and 2954 * new entries to share amaps and backing objects. 2955 */ 2956 /* 2957 * if the old_entry needs a new amap (due to prev fork) 2958 * then we need to allocate it now so that we have 2959 * something we own to share with the new_entry. [in 2960 * other words, we need to clear needs_copy] 2961 */ 2962 2963 if (UVM_ET_ISNEEDSCOPY(old_entry)) { 2964 /* get our own amap, clears needs_copy */ 2965 amap_copy(old_map, old_entry, M_WAITOK, FALSE, 2966 0, 0); 2967 /* XXXCDC: WAITOK??? */ 2968 } 2969 2970 new_entry = uvm_mapent_alloc(new_map); 2971 /* old_entry -> new_entry */ 2972 uvm_mapent_copy(old_entry, new_entry); 2973 2974 /* new pmap has nothing wired in it */ 2975 new_entry->wired_count = 0; 2976 2977 /* 2978 * gain reference to object backing the map (can't 2979 * be a submap, already checked this case). 2980 */ 2981 2982 if (new_entry->aref.ar_amap) 2983 uvm_map_reference_amap(new_entry, AMAP_SHARED); 2984 2985 if (new_entry->object.uvm_obj && 2986 new_entry->object.uvm_obj->pgops->pgo_reference) 2987 new_entry->object.uvm_obj-> 2988 pgops->pgo_reference( 2989 new_entry->object.uvm_obj); 2990 2991 /* insert entry at end of new_map's entry list */ 2992 uvm_map_entry_link(new_map, new_map->header.prev, 2993 new_entry); 2994 2995 /* 2996 * pmap_copy the mappings: this routine is optional 2997 * but if it is there it will reduce the number of 2998 * page faults in the new proc. 2999 */ 3000 3001 pmap_copy(new_pmap, old_map->pmap, new_entry->start, 3002 (old_entry->end - old_entry->start), 3003 old_entry->start); 3004 3005 break; 3006 3007 case MAP_INHERIT_COPY: 3008 3009 /* 3010 * copy-on-write the mapping (using mmap's 3011 * MAP_PRIVATE semantics) 3012 * 3013 * allocate new_entry, adjust reference counts. 3014 * (note that new references are read-only). 3015 */ 3016 3017 new_entry = uvm_mapent_alloc(new_map); 3018 /* old_entry -> new_entry */ 3019 uvm_mapent_copy(old_entry, new_entry); 3020 3021 if (new_entry->aref.ar_amap) 3022 uvm_map_reference_amap(new_entry, 0); 3023 3024 if (new_entry->object.uvm_obj && 3025 new_entry->object.uvm_obj->pgops->pgo_reference) 3026 new_entry->object.uvm_obj->pgops->pgo_reference 3027 (new_entry->object.uvm_obj); 3028 3029 /* new pmap has nothing wired in it */ 3030 new_entry->wired_count = 0; 3031 3032 new_entry->etype |= 3033 (UVM_ET_COPYONWRITE|UVM_ET_NEEDSCOPY); 3034 uvm_map_entry_link(new_map, new_map->header.prev, 3035 new_entry); 3036 3037 /* 3038 * the new entry will need an amap. it will either 3039 * need to be copied from the old entry or created 3040 * from scratch (if the old entry does not have an 3041 * amap). can we defer this process until later 3042 * (by setting "needs_copy") or do we need to copy 3043 * the amap now? 3044 * 3045 * we must copy the amap now if any of the following 3046 * conditions hold: 3047 * 1. the old entry has an amap and that amap is 3048 * being shared. this means that the old (parent) 3049 * process is sharing the amap with another 3050 * process. if we do not clear needs_copy here 3051 * we will end up in a situation where both the 3052 * parent and child process are refering to the 3053 * same amap with "needs_copy" set. if the 3054 * parent write-faults, the fault routine will 3055 * clear "needs_copy" in the parent by allocating 3056 * a new amap. this is wrong because the 3057 * parent is supposed to be sharing the old amap 3058 * and the new amap will break that. 3059 * 3060 * 2. if the old entry has an amap and a non-zero 3061 * wire count then we are going to have to call 3062 * amap_cow_now to avoid page faults in the 3063 * parent process. since amap_cow_now requires 3064 * "needs_copy" to be clear we might as well 3065 * clear it here as well. 3066 * 3067 */ 3068 3069 if (old_entry->aref.ar_amap != NULL) { 3070 if ((amap_flags(old_entry->aref.ar_amap) & 3071 AMAP_SHARED) != 0 || 3072 VM_MAPENT_ISWIRED(old_entry)) { 3073 3074 amap_copy(new_map, new_entry, M_WAITOK, 3075 FALSE, 0, 0); 3076 /* XXXCDC: M_WAITOK ... ok? */ 3077 } 3078 } 3079 3080 /* 3081 * if the parent's entry is wired down, then the 3082 * parent process does not want page faults on 3083 * access to that memory. this means that we 3084 * cannot do copy-on-write because we can't write 3085 * protect the old entry. in this case we 3086 * resolve all copy-on-write faults now, using 3087 * amap_cow_now. note that we have already 3088 * allocated any needed amap (above). 3089 */ 3090 3091 if (VM_MAPENT_ISWIRED(old_entry)) { 3092 3093 /* 3094 * resolve all copy-on-write faults now 3095 * (note that there is nothing to do if 3096 * the old mapping does not have an amap). 3097 * XXX: is it worthwhile to bother with pmap_copy 3098 * in this case? 3099 */ 3100 if (old_entry->aref.ar_amap) 3101 amap_cow_now(new_map, new_entry); 3102 3103 } else { 3104 3105 /* 3106 * setup mappings to trigger copy-on-write faults 3107 * we must write-protect the parent if it has 3108 * an amap and it is not already "needs_copy"... 3109 * if it is already "needs_copy" then the parent 3110 * has already been write-protected by a previous 3111 * fork operation. 3112 * 3113 * if we do not write-protect the parent, then 3114 * we must be sure to write-protect the child 3115 * after the pmap_copy() operation. 3116 * 3117 * XXX: pmap_copy should have some way of telling 3118 * us that it didn't do anything so we can avoid 3119 * calling pmap_protect needlessly. 3120 */ 3121 3122 if (old_entry->aref.ar_amap) { 3123 3124 if (!UVM_ET_ISNEEDSCOPY(old_entry)) { 3125 if (old_entry->max_protection & VM_PROT_WRITE) { 3126 pmap_protect(old_map->pmap, 3127 old_entry->start, 3128 old_entry->end, 3129 old_entry->protection & 3130 ~VM_PROT_WRITE); 3131 pmap_update(); 3132 } 3133 old_entry->etype |= UVM_ET_NEEDSCOPY; 3134 } 3135 3136 /* 3137 * parent must now be write-protected 3138 */ 3139 protect_child = FALSE; 3140 } else { 3141 3142 /* 3143 * we only need to protect the child if the 3144 * parent has write access. 3145 */ 3146 if (old_entry->max_protection & VM_PROT_WRITE) 3147 protect_child = TRUE; 3148 else 3149 protect_child = FALSE; 3150 3151 } 3152 3153 /* 3154 * copy the mappings 3155 * XXX: need a way to tell if this does anything 3156 */ 3157 3158 pmap_copy(new_pmap, old_map->pmap, 3159 new_entry->start, 3160 (old_entry->end - old_entry->start), 3161 old_entry->start); 3162 3163 /* 3164 * protect the child's mappings if necessary 3165 */ 3166 if (protect_child) { 3167 pmap_protect(new_pmap, new_entry->start, 3168 new_entry->end, 3169 new_entry->protection & 3170 ~VM_PROT_WRITE); 3171 pmap_update(); 3172 } 3173 3174 } 3175 break; 3176 } /* end of switch statement */ 3177 old_entry = old_entry->next; 3178 } 3179 3180 new_map->size = old_map->size; 3181 vm_map_unlock(old_map); 3182 3183 #ifdef SYSVSHM 3184 if (vm1->vm_shm) 3185 shmfork(vm1, vm2); 3186 #endif 3187 3188 #ifdef PMAP_FORK 3189 pmap_fork(vm1->vm_map.pmap, vm2->vm_map.pmap); 3190 #endif 3191 3192 UVMHIST_LOG(maphist,"<- done",0,0,0,0); 3193 return(vm2); 3194 } 3195 3196 3197 #if defined(DDB) 3198 3199 /* 3200 * DDB hooks 3201 */ 3202 3203 /* 3204 * uvm_map_printit: actually prints the map 3205 */ 3206 3207 void 3208 uvm_map_printit(map, full, pr) 3209 struct vm_map *map; 3210 boolean_t full; 3211 void (*pr) __P((const char *, ...)); 3212 { 3213 struct vm_map_entry *entry; 3214 3215 (*pr)("MAP %p: [0x%lx->0x%lx]\n", map, map->min_offset,map->max_offset); 3216 (*pr)("\t#ent=%d, sz=%d, ref=%d, version=%d, flags=0x%x\n", 3217 map->nentries, map->size, map->ref_count, map->timestamp, 3218 map->flags); 3219 #ifdef pmap_resident_count 3220 (*pr)("\tpmap=%p(resident=%d)\n", map->pmap, 3221 pmap_resident_count(map->pmap)); 3222 #else 3223 /* XXXCDC: this should be required ... */ 3224 (*pr)("\tpmap=%p(resident=<<NOT SUPPORTED!!!>>)\n", map->pmap); 3225 #endif 3226 if (!full) 3227 return; 3228 for (entry = map->header.next; entry != &map->header; 3229 entry = entry->next) { 3230 (*pr)(" - %p: 0x%lx->0x%lx: obj=%p/0x%llx, amap=%p/%d\n", 3231 entry, entry->start, entry->end, entry->object.uvm_obj, 3232 (long long)entry->offset, entry->aref.ar_amap, 3233 entry->aref.ar_pageoff); 3234 (*pr)( 3235 "\tsubmap=%c, cow=%c, nc=%c, prot(max)=%d/%d, inh=%d, " 3236 "wc=%d, adv=%d\n", 3237 (entry->etype & UVM_ET_SUBMAP) ? 'T' : 'F', 3238 (entry->etype & UVM_ET_COPYONWRITE) ? 'T' : 'F', 3239 (entry->etype & UVM_ET_NEEDSCOPY) ? 'T' : 'F', 3240 entry->protection, entry->max_protection, 3241 entry->inheritance, entry->wired_count, entry->advice); 3242 } 3243 } 3244 3245 /* 3246 * uvm_object_printit: actually prints the object 3247 */ 3248 3249 void 3250 uvm_object_printit(uobj, full, pr) 3251 struct uvm_object *uobj; 3252 boolean_t full; 3253 void (*pr) __P((const char *, ...)); 3254 { 3255 struct vm_page *pg; 3256 int cnt = 0; 3257 3258 (*pr)("OBJECT %p: locked=%d, pgops=%p, npages=%d, ", 3259 uobj, uobj->vmobjlock.lock_data, uobj->pgops, uobj->uo_npages); 3260 if (UVM_OBJ_IS_KERN_OBJECT(uobj)) 3261 (*pr)("refs=<SYSTEM>\n"); 3262 else 3263 (*pr)("refs=%d\n", uobj->uo_refs); 3264 3265 if (!full) { 3266 return; 3267 } 3268 (*pr)(" PAGES <pg,offset>:\n "); 3269 for (pg = TAILQ_FIRST(&uobj->memq); 3270 pg != NULL; 3271 pg = TAILQ_NEXT(pg, listq), cnt++) { 3272 (*pr)("<%p,0x%llx> ", pg, (long long)pg->offset); 3273 if ((cnt % 3) == 2) { 3274 (*pr)("\n "); 3275 } 3276 } 3277 if ((cnt % 3) != 2) { 3278 (*pr)("\n"); 3279 } 3280 } 3281 3282 /* 3283 * uvm_page_printit: actually print the page 3284 */ 3285 3286 static const char page_flagbits[] = 3287 "\20\1BUSY\2WANTED\3TABLED\4CLEAN\5CLEANCHK\6RELEASED\7FAKE\10RDONLY" 3288 "\11ZERO\15PAGER1"; 3289 static const char page_pqflagbits[] = 3290 "\20\1FREE\2INACTIVE\3ACTIVE\4LAUNDRY\5ANON\6AOBJ"; 3291 3292 void 3293 uvm_page_printit(pg, full, pr) 3294 struct vm_page *pg; 3295 boolean_t full; 3296 void (*pr) __P((const char *, ...)); 3297 { 3298 struct vm_page *tpg; 3299 struct uvm_object *uobj; 3300 struct pglist *pgl; 3301 char pgbuf[128]; 3302 char pqbuf[128]; 3303 3304 (*pr)("PAGE %p:\n", pg); 3305 bitmask_snprintf(pg->flags, page_flagbits, pgbuf, sizeof(pgbuf)); 3306 bitmask_snprintf(pg->pqflags, page_pqflagbits, pqbuf, sizeof(pqbuf)); 3307 (*pr)(" flags=%s, pqflags=%s, vers=%d, wire_count=%d, pa=0x%lx\n", 3308 pgbuf, pqbuf, pg->version, pg->wire_count, (long)pg->phys_addr); 3309 (*pr)(" uobject=%p, uanon=%p, offset=0x%llx loan_count=%d\n", 3310 pg->uobject, pg->uanon, (long long)pg->offset, pg->loan_count); 3311 #if defined(UVM_PAGE_TRKOWN) 3312 if (pg->flags & PG_BUSY) 3313 (*pr)(" owning process = %d, tag=%s\n", 3314 pg->owner, pg->owner_tag); 3315 else 3316 (*pr)(" page not busy, no owner\n"); 3317 #else 3318 (*pr)(" [page ownership tracking disabled]\n"); 3319 #endif 3320 3321 if (!full) 3322 return; 3323 3324 /* cross-verify object/anon */ 3325 if ((pg->pqflags & PQ_FREE) == 0) { 3326 if (pg->pqflags & PQ_ANON) { 3327 if (pg->uanon == NULL || pg->uanon->u.an_page != pg) 3328 (*pr)(" >>> ANON DOES NOT POINT HERE <<< (%p)\n", 3329 (pg->uanon) ? pg->uanon->u.an_page : NULL); 3330 else 3331 (*pr)(" anon backpointer is OK\n"); 3332 } else { 3333 uobj = pg->uobject; 3334 if (uobj) { 3335 (*pr)(" checking object list\n"); 3336 TAILQ_FOREACH(tpg, &uobj->memq, listq) { 3337 if (tpg == pg) { 3338 break; 3339 } 3340 } 3341 if (tpg) 3342 (*pr)(" page found on object list\n"); 3343 else 3344 (*pr)(" >>> PAGE NOT FOUND ON OBJECT LIST! <<<\n"); 3345 } 3346 } 3347 } 3348 3349 /* cross-verify page queue */ 3350 if (pg->pqflags & PQ_FREE) { 3351 int fl = uvm_page_lookup_freelist(pg); 3352 int color = VM_PGCOLOR_BUCKET(pg); 3353 pgl = &uvm.page_free[fl].pgfl_buckets[color].pgfl_queues[ 3354 ((pg)->flags & PG_ZERO) ? PGFL_ZEROS : PGFL_UNKNOWN]; 3355 } else if (pg->pqflags & PQ_INACTIVE) { 3356 pgl = &uvm.page_inactive; 3357 } else if (pg->pqflags & PQ_ACTIVE) { 3358 pgl = &uvm.page_active; 3359 } else { 3360 pgl = NULL; 3361 } 3362 3363 if (pgl) { 3364 (*pr)(" checking pageq list\n"); 3365 TAILQ_FOREACH(tpg, pgl, pageq) { 3366 if (tpg == pg) { 3367 break; 3368 } 3369 } 3370 if (tpg) 3371 (*pr)(" page found on pageq list\n"); 3372 else 3373 (*pr)(" >>> PAGE NOT FOUND ON PAGEQ LIST! <<<\n"); 3374 } 3375 } 3376 #endif 3377