1 /* $NetBSD: uvm_map.c,v 1.117 2001/12/31 22:34:40 chs Exp $ */ 2 3 /* 4 * Copyright (c) 1997 Charles D. Cranor and Washington University. 5 * Copyright (c) 1991, 1993, The Regents of the University of California. 6 * 7 * All rights reserved. 8 * 9 * This code is derived from software contributed to Berkeley by 10 * The Mach Operating System project at Carnegie-Mellon University. 11 * 12 * Redistribution and use in source and binary forms, with or without 13 * modification, are permitted provided that the following conditions 14 * are met: 15 * 1. Redistributions of source code must retain the above copyright 16 * notice, this list of conditions and the following disclaimer. 17 * 2. Redistributions in binary form must reproduce the above copyright 18 * notice, this list of conditions and the following disclaimer in the 19 * documentation and/or other materials provided with the distribution. 20 * 3. All advertising materials mentioning features or use of this software 21 * must display the following acknowledgement: 22 * This product includes software developed by Charles D. Cranor, 23 * Washington University, the University of California, Berkeley and 24 * its contributors. 25 * 4. Neither the name of the University nor the names of its contributors 26 * may be used to endorse or promote products derived from this software 27 * without specific prior written permission. 28 * 29 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 30 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 31 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 32 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 33 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 34 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 35 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 36 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 37 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 38 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 39 * SUCH DAMAGE. 40 * 41 * @(#)vm_map.c 8.3 (Berkeley) 1/12/94 42 * from: Id: uvm_map.c,v 1.1.2.27 1998/02/07 01:16:54 chs Exp 43 * 44 * 45 * Copyright (c) 1987, 1990 Carnegie-Mellon University. 46 * All rights reserved. 47 * 48 * Permission to use, copy, modify and distribute this software and 49 * its documentation is hereby granted, provided that both the copyright 50 * notice and this permission notice appear in all copies of the 51 * software, derivative works or modified versions, and any portions 52 * thereof, and that both notices appear in supporting documentation. 53 * 54 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS" 55 * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND 56 * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE. 57 * 58 * Carnegie Mellon requests users of this software to return to 59 * 60 * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU 61 * School of Computer Science 62 * Carnegie Mellon University 63 * Pittsburgh PA 15213-3890 64 * 65 * any improvements or extensions that they make and grant Carnegie the 66 * rights to redistribute these changes. 67 */ 68 69 /* 70 * uvm_map.c: uvm map operations 71 */ 72 73 #include <sys/cdefs.h> 74 __KERNEL_RCSID(0, "$NetBSD: uvm_map.c,v 1.117 2001/12/31 22:34:40 chs Exp $"); 75 76 #include "opt_ddb.h" 77 #include "opt_uvmhist.h" 78 #include "opt_sysv.h" 79 80 #include <sys/param.h> 81 #include <sys/systm.h> 82 #include <sys/mman.h> 83 #include <sys/proc.h> 84 #include <sys/malloc.h> 85 #include <sys/pool.h> 86 #include <sys/kernel.h> 87 #include <sys/mount.h> 88 #include <sys/vnode.h> 89 90 #ifdef SYSVSHM 91 #include <sys/shm.h> 92 #endif 93 94 #define UVM_MAP 95 #include <uvm/uvm.h> 96 97 #ifdef DDB 98 #include <uvm/uvm_ddb.h> 99 #endif 100 101 extern struct vm_map *pager_map; 102 103 struct uvm_cnt uvm_map_call, map_backmerge, map_forwmerge; 104 struct uvm_cnt uvm_mlk_call, uvm_mlk_hint; 105 const char vmmapbsy[] = "vmmapbsy"; 106 107 /* 108 * pool for vmspace structures. 109 */ 110 111 struct pool uvm_vmspace_pool; 112 113 /* 114 * pool for dynamically-allocated map entries. 115 */ 116 117 struct pool uvm_map_entry_pool; 118 struct pool uvm_map_entry_kmem_pool; 119 120 #ifdef PMAP_GROWKERNEL 121 /* 122 * This global represents the end of the kernel virtual address 123 * space. If we want to exceed this, we must grow the kernel 124 * virtual address space dynamically. 125 * 126 * Note, this variable is locked by kernel_map's lock. 127 */ 128 vaddr_t uvm_maxkaddr; 129 #endif 130 131 /* 132 * macros 133 */ 134 135 /* 136 * uvm_map_entry_link: insert entry into a map 137 * 138 * => map must be locked 139 */ 140 #define uvm_map_entry_link(map, after_where, entry) do { \ 141 (map)->nentries++; \ 142 (entry)->prev = (after_where); \ 143 (entry)->next = (after_where)->next; \ 144 (entry)->prev->next = (entry); \ 145 (entry)->next->prev = (entry); \ 146 } while (0) 147 148 /* 149 * uvm_map_entry_unlink: remove entry from a map 150 * 151 * => map must be locked 152 */ 153 #define uvm_map_entry_unlink(map, entry) do { \ 154 (map)->nentries--; \ 155 (entry)->next->prev = (entry)->prev; \ 156 (entry)->prev->next = (entry)->next; \ 157 } while (0) 158 159 /* 160 * SAVE_HINT: saves the specified entry as the hint for future lookups. 161 * 162 * => map need not be locked (protected by hint_lock). 163 */ 164 #define SAVE_HINT(map,check,value) do { \ 165 simple_lock(&(map)->hint_lock); \ 166 if ((map)->hint == (check)) \ 167 (map)->hint = (value); \ 168 simple_unlock(&(map)->hint_lock); \ 169 } while (0) 170 171 /* 172 * VM_MAP_RANGE_CHECK: check and correct range 173 * 174 * => map must at least be read locked 175 */ 176 177 #define VM_MAP_RANGE_CHECK(map, start, end) do { \ 178 if (start < vm_map_min(map)) \ 179 start = vm_map_min(map); \ 180 if (end > vm_map_max(map)) \ 181 end = vm_map_max(map); \ 182 if (start > end) \ 183 start = end; \ 184 } while (0) 185 186 /* 187 * local prototypes 188 */ 189 190 static struct vm_map_entry *uvm_mapent_alloc __P((struct vm_map *)); 191 static void uvm_mapent_copy __P((struct vm_map_entry *, struct vm_map_entry *)); 192 static void uvm_mapent_free __P((struct vm_map_entry *)); 193 static void uvm_map_entry_unwire __P((struct vm_map *, struct vm_map_entry *)); 194 static void uvm_map_reference_amap __P((struct vm_map_entry *, int)); 195 static void uvm_map_unreference_amap __P((struct vm_map_entry *, int)); 196 197 /* 198 * local inlines 199 */ 200 201 /* 202 * uvm_mapent_alloc: allocate a map entry 203 */ 204 205 static __inline struct vm_map_entry * 206 uvm_mapent_alloc(map) 207 struct vm_map *map; 208 { 209 struct vm_map_entry *me; 210 int s; 211 UVMHIST_FUNC("uvm_mapent_alloc"); UVMHIST_CALLED(maphist); 212 213 if (map->flags & VM_MAP_INTRSAFE || cold) { 214 s = 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 == NULL) { 221 panic("uvm_mapent_alloc: out of static map entries, " 222 "check MAX_KMAPENT (currently %d)", 223 MAX_KMAPENT); 224 } 225 me->flags = UVM_MAP_STATIC; 226 } else if (map == kernel_map) { 227 me = pool_get(&uvm_map_entry_kmem_pool, PR_WAITOK); 228 me->flags = UVM_MAP_KMEM; 229 } else { 230 me = pool_get(&uvm_map_entry_pool, PR_WAITOK); 231 me->flags = 0; 232 } 233 234 UVMHIST_LOG(maphist, "<- new entry=0x%x [kentry=%d]", me, 235 ((map->flags & VM_MAP_INTRSAFE) != 0 || map == kernel_map), 0, 0); 236 return(me); 237 } 238 239 /* 240 * uvm_mapent_free: free map entry 241 */ 242 243 static __inline void 244 uvm_mapent_free(me) 245 struct vm_map_entry *me; 246 { 247 int s; 248 UVMHIST_FUNC("uvm_mapent_free"); UVMHIST_CALLED(maphist); 249 250 UVMHIST_LOG(maphist,"<- freeing map entry=0x%x [flags=%d]", 251 me, me->flags, 0, 0); 252 if (me->flags & UVM_MAP_STATIC) { 253 s = splvm(); 254 simple_lock(&uvm.kentry_lock); 255 me->next = uvm.kentry_free; 256 uvm.kentry_free = me; 257 simple_unlock(&uvm.kentry_lock); 258 splx(s); 259 } else if (me->flags & UVM_MAP_KMEM) { 260 pool_put(&uvm_map_entry_kmem_pool, me); 261 } else { 262 pool_put(&uvm_map_entry_pool, me); 263 } 264 } 265 266 /* 267 * uvm_mapent_copy: copy a map entry, preserving flags 268 */ 269 270 static __inline void 271 uvm_mapent_copy(src, dst) 272 struct vm_map_entry *src; 273 struct vm_map_entry *dst; 274 { 275 memcpy(dst, src, ((char *)&src->uvm_map_entry_stop_copy) - 276 ((char *)src)); 277 } 278 279 /* 280 * uvm_map_entry_unwire: unwire a map entry 281 * 282 * => map should be locked by caller 283 */ 284 285 static __inline void 286 uvm_map_entry_unwire(map, entry) 287 struct vm_map *map; 288 struct vm_map_entry *entry; 289 { 290 entry->wired_count = 0; 291 uvm_fault_unwire_locked(map, entry->start, entry->end); 292 } 293 294 295 /* 296 * wrapper for calling amap_ref() 297 */ 298 static __inline void 299 uvm_map_reference_amap(entry, flags) 300 struct vm_map_entry *entry; 301 int flags; 302 { 303 amap_ref(entry->aref.ar_amap, entry->aref.ar_pageoff, 304 (entry->end - entry->start) >> PAGE_SHIFT, flags); 305 } 306 307 308 /* 309 * wrapper for calling amap_unref() 310 */ 311 static __inline void 312 uvm_map_unreference_amap(entry, flags) 313 struct vm_map_entry *entry; 314 int flags; 315 { 316 amap_unref(entry->aref.ar_amap, entry->aref.ar_pageoff, 317 (entry->end - entry->start) >> PAGE_SHIFT, flags); 318 } 319 320 321 /* 322 * uvm_map_init: init mapping system at boot time. note that we allocate 323 * and init the static pool of struct vm_map_entry *'s for the kernel here. 324 */ 325 326 void 327 uvm_map_init() 328 { 329 static struct vm_map_entry kernel_map_entry[MAX_KMAPENT]; 330 #if defined(UVMHIST) 331 static struct uvm_history_ent maphistbuf[100]; 332 static struct uvm_history_ent pdhistbuf[100]; 333 #endif 334 int lcv; 335 336 /* 337 * first, init logging system. 338 */ 339 340 UVMHIST_FUNC("uvm_map_init"); 341 UVMHIST_INIT_STATIC(maphist, maphistbuf); 342 UVMHIST_INIT_STATIC(pdhist, pdhistbuf); 343 UVMHIST_CALLED(maphist); 344 UVMHIST_LOG(maphist,"<starting uvm map system>", 0, 0, 0, 0); 345 UVMCNT_INIT(uvm_map_call, UVMCNT_CNT, 0, 346 "# uvm_map() successful calls", 0); 347 UVMCNT_INIT(map_backmerge, UVMCNT_CNT, 0, "# uvm_map() back merges", 0); 348 UVMCNT_INIT(map_forwmerge, UVMCNT_CNT, 0, "# uvm_map() missed forward", 349 0); 350 UVMCNT_INIT(uvm_mlk_call, UVMCNT_CNT, 0, "# map lookup calls", 0); 351 UVMCNT_INIT(uvm_mlk_hint, UVMCNT_CNT, 0, "# map lookup hint hits", 0); 352 353 /* 354 * now set up static pool of kernel map entrys ... 355 */ 356 357 simple_lock_init(&uvm.kentry_lock); 358 uvm.kentry_free = NULL; 359 for (lcv = 0 ; lcv < MAX_KMAPENT ; lcv++) { 360 kernel_map_entry[lcv].next = uvm.kentry_free; 361 uvm.kentry_free = &kernel_map_entry[lcv]; 362 } 363 364 /* 365 * initialize the map-related pools. 366 */ 367 pool_init(&uvm_vmspace_pool, sizeof(struct vmspace), 368 0, 0, 0, "vmsppl", 0, 369 pool_page_alloc_nointr, pool_page_free_nointr, M_VMMAP); 370 pool_init(&uvm_map_entry_pool, sizeof(struct vm_map_entry), 371 0, 0, 0, "vmmpepl", 0, 372 pool_page_alloc_nointr, pool_page_free_nointr, M_VMMAP); 373 pool_init(&uvm_map_entry_kmem_pool, sizeof(struct vm_map_entry), 374 0, 0, 0, "vmmpekpl", 0, NULL, NULL, M_VMMAP); 375 } 376 377 /* 378 * clippers 379 */ 380 381 /* 382 * uvm_map_clip_start: ensure that the entry begins at or after 383 * the starting address, if it doesn't we split the entry. 384 * 385 * => caller should use UVM_MAP_CLIP_START macro rather than calling 386 * this directly 387 * => map must be locked by caller 388 */ 389 390 void 391 uvm_map_clip_start(map, entry, start) 392 struct vm_map *map; 393 struct vm_map_entry *entry; 394 vaddr_t start; 395 { 396 struct vm_map_entry *new_entry; 397 vaddr_t new_adj; 398 399 /* uvm_map_simplify_entry(map, entry); */ /* XXX */ 400 401 /* 402 * Split off the front portion. note that we must insert the new 403 * entry BEFORE this one, so that this entry has the specified 404 * starting address. 405 */ 406 407 new_entry = uvm_mapent_alloc(map); 408 uvm_mapent_copy(entry, new_entry); /* entry -> new_entry */ 409 410 new_entry->end = start; 411 new_adj = start - new_entry->start; 412 if (entry->object.uvm_obj) 413 entry->offset += new_adj; /* shift start over */ 414 entry->start = start; 415 416 if (new_entry->aref.ar_amap) { 417 amap_splitref(&new_entry->aref, &entry->aref, new_adj); 418 } 419 420 uvm_map_entry_link(map, entry->prev, new_entry); 421 422 if (UVM_ET_ISSUBMAP(entry)) { 423 /* ... unlikely to happen, but play it safe */ 424 uvm_map_reference(new_entry->object.sub_map); 425 } else { 426 if (UVM_ET_ISOBJ(entry) && 427 entry->object.uvm_obj->pgops && 428 entry->object.uvm_obj->pgops->pgo_reference) 429 entry->object.uvm_obj->pgops->pgo_reference( 430 entry->object.uvm_obj); 431 } 432 } 433 434 /* 435 * uvm_map_clip_end: ensure that the entry ends at or before 436 * the ending address, if it does't we split the reference 437 * 438 * => caller should use UVM_MAP_CLIP_END macro rather than calling 439 * this directly 440 * => map must be locked by caller 441 */ 442 443 void 444 uvm_map_clip_end(map, entry, end) 445 struct vm_map *map; 446 struct vm_map_entry *entry; 447 vaddr_t end; 448 { 449 struct vm_map_entry * new_entry; 450 vaddr_t new_adj; /* #bytes we move start forward */ 451 452 /* 453 * Create a new entry and insert it 454 * AFTER the specified entry 455 */ 456 457 new_entry = uvm_mapent_alloc(map); 458 uvm_mapent_copy(entry, new_entry); /* entry -> new_entry */ 459 460 new_entry->start = entry->end = end; 461 new_adj = end - entry->start; 462 if (new_entry->object.uvm_obj) 463 new_entry->offset += new_adj; 464 465 if (entry->aref.ar_amap) 466 amap_splitref(&entry->aref, &new_entry->aref, new_adj); 467 468 uvm_map_entry_link(map, entry, new_entry); 469 470 if (UVM_ET_ISSUBMAP(entry)) { 471 /* ... unlikely to happen, but play it safe */ 472 uvm_map_reference(new_entry->object.sub_map); 473 } else { 474 if (UVM_ET_ISOBJ(entry) && 475 entry->object.uvm_obj->pgops && 476 entry->object.uvm_obj->pgops->pgo_reference) 477 entry->object.uvm_obj->pgops->pgo_reference( 478 entry->object.uvm_obj); 479 } 480 } 481 482 483 /* 484 * M A P - m a i n e n t r y p o i n t 485 */ 486 /* 487 * uvm_map: establish a valid mapping in a map 488 * 489 * => assume startp is page aligned. 490 * => assume size is a multiple of PAGE_SIZE. 491 * => assume sys_mmap provides enough of a "hint" to have us skip 492 * over text/data/bss area. 493 * => map must be unlocked (we will lock it) 494 * => <uobj,uoffset> value meanings (4 cases): 495 * [1] <NULL,uoffset> == uoffset is a hint for PMAP_PREFER 496 * [2] <NULL,UVM_UNKNOWN_OFFSET> == don't PMAP_PREFER 497 * [3] <uobj,uoffset> == normal mapping 498 * [4] <uobj,UVM_UNKNOWN_OFFSET> == uvm_map finds offset based on VA 499 * 500 * case [4] is for kernel mappings where we don't know the offset until 501 * we've found a virtual address. note that kernel object offsets are 502 * always relative to vm_map_min(kernel_map). 503 * 504 * => if `align' is non-zero, we try to align the virtual address to 505 * the specified alignment. this is only a hint; if we can't 506 * do it, the address will be unaligned. this is provided as 507 * a mechanism for large pages. 508 * 509 * => XXXCDC: need way to map in external amap? 510 */ 511 512 int 513 uvm_map(map, startp, size, uobj, uoffset, align, flags) 514 struct vm_map *map; 515 vaddr_t *startp; /* IN/OUT */ 516 vsize_t size; 517 struct uvm_object *uobj; 518 voff_t uoffset; 519 vsize_t align; 520 uvm_flag_t flags; 521 { 522 struct vm_map_entry *prev_entry, *new_entry; 523 vm_prot_t prot = UVM_PROTECTION(flags), maxprot = 524 UVM_MAXPROTECTION(flags); 525 vm_inherit_t inherit = UVM_INHERIT(flags); 526 int advice = UVM_ADVICE(flags); 527 UVMHIST_FUNC("uvm_map"); 528 UVMHIST_CALLED(maphist); 529 530 UVMHIST_LOG(maphist, "(map=0x%x, *startp=0x%x, size=%d, flags=0x%x)", 531 map, *startp, size, flags); 532 UVMHIST_LOG(maphist, " uobj/offset 0x%x/%d", uobj, uoffset,0,0); 533 534 /* 535 * detect a popular device driver bug. 536 */ 537 538 KASSERT(curproc != NULL || map->flags & VM_MAP_INTRSAFE); 539 540 /* 541 * check sanity of protection code 542 */ 543 544 if ((prot & maxprot) != prot) { 545 UVMHIST_LOG(maphist, "<- prot. failure: prot=0x%x, max=0x%x", 546 prot, maxprot,0,0); 547 return EACCES; 548 } 549 550 /* 551 * for pager_map, allocate the new entry first to avoid sleeping 552 * for memory while we have the map locked. 553 */ 554 555 new_entry = NULL; 556 if (map == pager_map) { 557 new_entry = uvm_mapent_alloc(map); 558 } 559 560 /* 561 * figure out where to put new VM range 562 */ 563 564 if (vm_map_lock_try(map) == FALSE) { 565 if (flags & UVM_FLAG_TRYLOCK) { 566 if (new_entry) { 567 uvm_mapent_free(new_entry); 568 } 569 return EAGAIN; 570 } 571 vm_map_lock(map); /* could sleep here */ 572 } 573 if ((prev_entry = uvm_map_findspace(map, *startp, size, startp, 574 uobj, uoffset, align, flags)) == NULL) { 575 UVMHIST_LOG(maphist,"<- uvm_map_findspace failed!",0,0,0,0); 576 vm_map_unlock(map); 577 if (new_entry) { 578 uvm_mapent_free(new_entry); 579 } 580 return ENOMEM; 581 } 582 583 #ifdef PMAP_GROWKERNEL 584 { 585 /* 586 * If the kernel pmap can't map the requested space, 587 * then allocate more resources for it. 588 */ 589 if (map == kernel_map && uvm_maxkaddr < (*startp + size)) 590 uvm_maxkaddr = pmap_growkernel(*startp + size); 591 } 592 #endif 593 594 UVMCNT_INCR(uvm_map_call); 595 596 /* 597 * if uobj is null, then uoffset is either a VAC hint for PMAP_PREFER 598 * [typically from uvm_map_reserve] or it is UVM_UNKNOWN_OFFSET. in 599 * either case we want to zero it before storing it in the map entry 600 * (because it looks strange and confusing when debugging...) 601 * 602 * if uobj is not null 603 * if uoffset is not UVM_UNKNOWN_OFFSET then we have a normal mapping 604 * and we do not need to change uoffset. 605 * if uoffset is UVM_UNKNOWN_OFFSET then we need to find the offset 606 * now (based on the starting address of the map). this case is 607 * for kernel object mappings where we don't know the offset until 608 * the virtual address is found (with uvm_map_findspace). the 609 * offset is the distance we are from the start of the map. 610 */ 611 612 if (uobj == NULL) { 613 uoffset = 0; 614 } else { 615 if (uoffset == UVM_UNKNOWN_OFFSET) { 616 KASSERT(UVM_OBJ_IS_KERN_OBJECT(uobj)); 617 uoffset = *startp - vm_map_min(kernel_map); 618 } 619 } 620 621 /* 622 * try and insert in map by extending previous entry, if possible. 623 * XXX: we don't try and pull back the next entry. might be useful 624 * for a stack, but we are currently allocating our stack in advance. 625 */ 626 627 if ((flags & UVM_FLAG_NOMERGE) == 0 && 628 prev_entry->end == *startp && prev_entry != &map->header && 629 prev_entry->object.uvm_obj == uobj) { 630 631 if (uobj && prev_entry->offset + 632 (prev_entry->end - prev_entry->start) != uoffset) 633 goto nomerge; 634 635 if (UVM_ET_ISSUBMAP(prev_entry)) 636 goto nomerge; 637 638 if (prev_entry->protection != prot || 639 prev_entry->max_protection != maxprot) 640 goto nomerge; 641 642 if (prev_entry->inheritance != inherit || 643 prev_entry->advice != advice) 644 goto nomerge; 645 646 /* wiring status must match (new area is unwired) */ 647 if (VM_MAPENT_ISWIRED(prev_entry)) 648 goto nomerge; 649 650 /* 651 * can't extend a shared amap. note: no need to lock amap to 652 * look at refs since we don't care about its exact value. 653 * if it is one (i.e. we have only reference) it will stay there 654 */ 655 656 if (prev_entry->aref.ar_amap && 657 amap_refs(prev_entry->aref.ar_amap) != 1) { 658 goto nomerge; 659 } 660 661 /* got it! */ 662 663 UVMCNT_INCR(map_backmerge); 664 UVMHIST_LOG(maphist," starting back merge", 0, 0, 0, 0); 665 666 /* 667 * drop our reference to uobj since we are extending a reference 668 * that we already have (the ref count can not drop to zero). 669 */ 670 if (uobj && uobj->pgops->pgo_detach) 671 uobj->pgops->pgo_detach(uobj); 672 673 if (prev_entry->aref.ar_amap) { 674 amap_extend(prev_entry, size); 675 } 676 677 prev_entry->end += size; 678 map->size += size; 679 680 UVMHIST_LOG(maphist,"<- done (via backmerge)!", 0, 0, 0, 0); 681 vm_map_unlock(map); 682 if (new_entry) { 683 uvm_mapent_free(new_entry); 684 } 685 return 0; 686 } 687 688 nomerge: 689 UVMHIST_LOG(maphist," allocating new map entry", 0, 0, 0, 0); 690 691 /* 692 * check for possible forward merge (which we don't do) and count 693 * the number of times we missed a *possible* chance to merge more 694 */ 695 696 if ((flags & UVM_FLAG_NOMERGE) == 0 && 697 prev_entry->next != &map->header && 698 prev_entry->next->start == (*startp + size)) 699 UVMCNT_INCR(map_forwmerge); 700 701 /* 702 * allocate new entry and link it in. 703 */ 704 705 if (new_entry == NULL) { 706 new_entry = uvm_mapent_alloc(map); 707 } 708 new_entry->start = *startp; 709 new_entry->end = new_entry->start + size; 710 new_entry->object.uvm_obj = uobj; 711 new_entry->offset = uoffset; 712 713 if (uobj) 714 new_entry->etype = UVM_ET_OBJ; 715 else 716 new_entry->etype = 0; 717 718 if (flags & UVM_FLAG_COPYONW) { 719 new_entry->etype |= UVM_ET_COPYONWRITE; 720 if ((flags & UVM_FLAG_OVERLAY) == 0) 721 new_entry->etype |= UVM_ET_NEEDSCOPY; 722 } 723 724 new_entry->protection = prot; 725 new_entry->max_protection = maxprot; 726 new_entry->inheritance = inherit; 727 new_entry->wired_count = 0; 728 new_entry->advice = advice; 729 if (flags & UVM_FLAG_OVERLAY) { 730 731 /* 732 * to_add: for BSS we overallocate a little since we 733 * are likely to extend 734 */ 735 736 vaddr_t to_add = (flags & UVM_FLAG_AMAPPAD) ? 737 UVM_AMAP_CHUNK << PAGE_SHIFT : 0; 738 struct vm_amap *amap = amap_alloc(size, to_add, M_WAITOK); 739 new_entry->aref.ar_pageoff = 0; 740 new_entry->aref.ar_amap = amap; 741 } else { 742 new_entry->aref.ar_pageoff = 0; 743 new_entry->aref.ar_amap = NULL; 744 } 745 uvm_map_entry_link(map, prev_entry, new_entry); 746 map->size += size; 747 748 /* 749 * Update the free space hint 750 */ 751 752 if ((map->first_free == prev_entry) && 753 (prev_entry->end >= new_entry->start)) 754 map->first_free = new_entry; 755 756 UVMHIST_LOG(maphist,"<- done!", 0, 0, 0, 0); 757 vm_map_unlock(map); 758 return 0; 759 } 760 761 /* 762 * uvm_map_lookup_entry: find map entry at or before an address 763 * 764 * => map must at least be read-locked by caller 765 * => entry is returned in "entry" 766 * => return value is true if address is in the returned entry 767 */ 768 769 boolean_t 770 uvm_map_lookup_entry(map, address, entry) 771 struct vm_map *map; 772 vaddr_t address; 773 struct vm_map_entry **entry; /* OUT */ 774 { 775 struct vm_map_entry *cur; 776 struct vm_map_entry *last; 777 UVMHIST_FUNC("uvm_map_lookup_entry"); 778 UVMHIST_CALLED(maphist); 779 780 UVMHIST_LOG(maphist,"(map=0x%x,addr=0x%x,ent=0x%x)", 781 map, address, entry, 0); 782 783 /* 784 * start looking either from the head of the 785 * list, or from the hint. 786 */ 787 788 simple_lock(&map->hint_lock); 789 cur = map->hint; 790 simple_unlock(&map->hint_lock); 791 792 if (cur == &map->header) 793 cur = cur->next; 794 795 UVMCNT_INCR(uvm_mlk_call); 796 if (address >= cur->start) { 797 798 /* 799 * go from hint to end of list. 800 * 801 * but first, make a quick check to see if 802 * we are already looking at the entry we 803 * want (which is usually the case). 804 * note also that we don't need to save the hint 805 * here... it is the same hint (unless we are 806 * at the header, in which case the hint didn't 807 * buy us anything anyway). 808 */ 809 810 last = &map->header; 811 if ((cur != last) && (cur->end > address)) { 812 UVMCNT_INCR(uvm_mlk_hint); 813 *entry = cur; 814 UVMHIST_LOG(maphist,"<- got it via hint (0x%x)", 815 cur, 0, 0, 0); 816 return (TRUE); 817 } 818 } else { 819 820 /* 821 * go from start to hint, *inclusively* 822 */ 823 824 last = cur->next; 825 cur = map->header.next; 826 } 827 828 /* 829 * search linearly 830 */ 831 832 while (cur != last) { 833 if (cur->end > address) { 834 if (address >= cur->start) { 835 /* 836 * save this lookup for future 837 * hints, and return 838 */ 839 840 *entry = cur; 841 SAVE_HINT(map, map->hint, cur); 842 UVMHIST_LOG(maphist,"<- search got it (0x%x)", 843 cur, 0, 0, 0); 844 return (TRUE); 845 } 846 break; 847 } 848 cur = cur->next; 849 } 850 *entry = cur->prev; 851 SAVE_HINT(map, map->hint, *entry); 852 UVMHIST_LOG(maphist,"<- failed!",0,0,0,0); 853 return (FALSE); 854 } 855 856 /* 857 * uvm_map_findspace: find "length" sized space in "map". 858 * 859 * => "hint" is a hint about where we want it, unless FINDSPACE_FIXED is 860 * set (in which case we insist on using "hint"). 861 * => "result" is VA returned 862 * => uobj/uoffset are to be used to handle VAC alignment, if required 863 * => if `align' is non-zero, we attempt to align to that value. 864 * => caller must at least have read-locked map 865 * => returns NULL on failure, or pointer to prev. map entry if success 866 * => note this is a cross between the old vm_map_findspace and vm_map_find 867 */ 868 869 struct vm_map_entry * 870 uvm_map_findspace(map, hint, length, result, uobj, uoffset, align, flags) 871 struct vm_map *map; 872 vaddr_t hint; 873 vsize_t length; 874 vaddr_t *result; /* OUT */ 875 struct uvm_object *uobj; 876 voff_t uoffset; 877 vsize_t align; 878 int flags; 879 { 880 struct vm_map_entry *entry, *next, *tmp; 881 vaddr_t end, orig_hint; 882 UVMHIST_FUNC("uvm_map_findspace"); 883 UVMHIST_CALLED(maphist); 884 885 UVMHIST_LOG(maphist, "(map=0x%x, hint=0x%x, len=%d, flags=0x%x)", 886 map, hint, length, flags); 887 KASSERT((align & (align - 1)) == 0); 888 KASSERT((flags & UVM_FLAG_FIXED) == 0 || align == 0); 889 890 /* 891 * remember the original hint. if we are aligning, then we 892 * may have to try again with no alignment constraint if 893 * we fail the first time. 894 */ 895 896 orig_hint = hint; 897 if (hint < map->min_offset) { /* check ranges ... */ 898 if (flags & UVM_FLAG_FIXED) { 899 UVMHIST_LOG(maphist,"<- VA below map range",0,0,0,0); 900 return(NULL); 901 } 902 hint = map->min_offset; 903 } 904 if (hint > map->max_offset) { 905 UVMHIST_LOG(maphist,"<- VA 0x%x > range [0x%x->0x%x]", 906 hint, map->min_offset, map->max_offset, 0); 907 return(NULL); 908 } 909 910 /* 911 * Look for the first possible address; if there's already 912 * something at this address, we have to start after it. 913 */ 914 915 if ((flags & UVM_FLAG_FIXED) == 0 && hint == map->min_offset) { 916 if ((entry = map->first_free) != &map->header) 917 hint = entry->end; 918 } else { 919 if (uvm_map_lookup_entry(map, hint, &tmp)) { 920 /* "hint" address already in use ... */ 921 if (flags & UVM_FLAG_FIXED) { 922 UVMHIST_LOG(maphist,"<- fixed & VA in use", 923 0, 0, 0, 0); 924 return(NULL); 925 } 926 hint = tmp->end; 927 } 928 entry = tmp; 929 } 930 931 /* 932 * Look through the rest of the map, trying to fit a new region in 933 * the gap between existing regions, or after the very last region. 934 * note: entry->end = base VA of current gap, 935 * next->start = VA of end of current gap 936 */ 937 938 for (;; hint = (entry = next)->end) { 939 940 /* 941 * Find the end of the proposed new region. Be sure we didn't 942 * go beyond the end of the map, or wrap around the address; 943 * if so, we lose. Otherwise, if this is the last entry, or 944 * if the proposed new region fits before the next entry, we 945 * win. 946 */ 947 948 #ifdef PMAP_PREFER 949 /* 950 * push hint forward as needed to avoid VAC alias problems. 951 * we only do this if a valid offset is specified. 952 */ 953 954 if ((flags & UVM_FLAG_FIXED) == 0 && 955 uoffset != UVM_UNKNOWN_OFFSET) 956 PMAP_PREFER(uoffset, &hint); 957 #endif 958 if (align != 0) { 959 if ((hint & (align - 1)) != 0) 960 hint = roundup(hint, align); 961 /* 962 * XXX Should we PMAP_PREFER() here again? 963 */ 964 } 965 end = hint + length; 966 if (end > map->max_offset || end < hint) { 967 UVMHIST_LOG(maphist,"<- failed (off end)", 0,0,0,0); 968 if (align != 0) { 969 UVMHIST_LOG(maphist, 970 "calling recursively, no align", 971 0,0,0,0); 972 return (uvm_map_findspace(map, orig_hint, 973 length, result, uobj, uoffset, 0, flags)); 974 } 975 return (NULL); 976 } 977 next = entry->next; 978 if (next == &map->header || next->start >= end) 979 break; 980 if (flags & UVM_FLAG_FIXED) { 981 UVMHIST_LOG(maphist,"<- fixed mapping failed", 0,0,0,0); 982 return(NULL); /* only one shot at it ... */ 983 } 984 } 985 SAVE_HINT(map, map->hint, entry); 986 *result = hint; 987 UVMHIST_LOG(maphist,"<- got it! (result=0x%x)", hint, 0,0,0); 988 return (entry); 989 } 990 991 /* 992 * U N M A P - m a i n h e l p e r f u n c t i o n s 993 */ 994 995 /* 996 * uvm_unmap_remove: remove mappings from a vm_map (from "start" up to "stop") 997 * 998 * => caller must check alignment and size 999 * => map must be locked by caller 1000 * => we return a list of map entries that we've remove from the map 1001 * in "entry_list" 1002 */ 1003 1004 void 1005 uvm_unmap_remove(map, start, end, entry_list) 1006 struct vm_map *map; 1007 vaddr_t start, end; 1008 struct vm_map_entry **entry_list; /* OUT */ 1009 { 1010 struct vm_map_entry *entry, *first_entry, *next; 1011 vaddr_t len; 1012 UVMHIST_FUNC("uvm_unmap_remove"); UVMHIST_CALLED(maphist); 1013 1014 UVMHIST_LOG(maphist,"(map=0x%x, start=0x%x, end=0x%x)", 1015 map, start, end, 0); 1016 VM_MAP_RANGE_CHECK(map, start, end); 1017 1018 /* 1019 * find first entry 1020 */ 1021 1022 if (uvm_map_lookup_entry(map, start, &first_entry) == TRUE) { 1023 /* clip and go... */ 1024 entry = first_entry; 1025 UVM_MAP_CLIP_START(map, entry, start); 1026 /* critical! prevents stale hint */ 1027 SAVE_HINT(map, entry, entry->prev); 1028 } else { 1029 entry = first_entry->next; 1030 } 1031 1032 /* 1033 * Save the free space hint 1034 */ 1035 1036 if (map->first_free->start >= start) 1037 map->first_free = entry->prev; 1038 1039 /* 1040 * note: we now re-use first_entry for a different task. we remove 1041 * a number of map entries from the map and save them in a linked 1042 * list headed by "first_entry". once we remove them from the map 1043 * the caller should unlock the map and drop the references to the 1044 * backing objects [c.f. uvm_unmap_detach]. the object is to 1045 * separate unmapping from reference dropping. why? 1046 * [1] the map has to be locked for unmapping 1047 * [2] the map need not be locked for reference dropping 1048 * [3] dropping references may trigger pager I/O, and if we hit 1049 * a pager that does synchronous I/O we may have to wait for it. 1050 * [4] we would like all waiting for I/O to occur with maps unlocked 1051 * so that we don't block other threads. 1052 */ 1053 1054 first_entry = NULL; 1055 *entry_list = NULL; 1056 1057 /* 1058 * break up the area into map entry sized regions and unmap. note 1059 * that all mappings have to be removed before we can even consider 1060 * dropping references to amaps or VM objects (otherwise we could end 1061 * up with a mapping to a page on the free list which would be very bad) 1062 */ 1063 1064 while ((entry != &map->header) && (entry->start < end)) { 1065 UVM_MAP_CLIP_END(map, entry, end); 1066 next = entry->next; 1067 len = entry->end - entry->start; 1068 1069 /* 1070 * unwire before removing addresses from the pmap; otherwise 1071 * unwiring will put the entries back into the pmap (XXX). 1072 */ 1073 1074 if (VM_MAPENT_ISWIRED(entry)) { 1075 uvm_map_entry_unwire(map, entry); 1076 } 1077 if ((map->flags & VM_MAP_PAGEABLE) == 0) { 1078 1079 /* 1080 * if the map is non-pageable, any pages mapped there 1081 * must be wired and entered with pmap_kenter_pa(), 1082 * and we should free any such pages immediately. 1083 * this is mostly used for kmem_map and mb_map. 1084 */ 1085 1086 uvm_km_pgremove_intrsafe(entry->start, entry->end); 1087 pmap_kremove(entry->start, len); 1088 } else if (UVM_ET_ISOBJ(entry) && 1089 UVM_OBJ_IS_KERN_OBJECT(entry->object.uvm_obj)) { 1090 KASSERT(vm_map_pmap(map) == pmap_kernel()); 1091 1092 /* 1093 * note: kernel object mappings are currently used in 1094 * two ways: 1095 * [1] "normal" mappings of pages in the kernel object 1096 * [2] uvm_km_valloc'd allocations in which we 1097 * pmap_enter in some non-kernel-object page 1098 * (e.g. vmapbuf). 1099 * 1100 * for case [1], we need to remove the mapping from 1101 * the pmap and then remove the page from the kernel 1102 * object (because, once pages in a kernel object are 1103 * unmapped they are no longer needed, unlike, say, 1104 * a vnode where you might want the data to persist 1105 * until flushed out of a queue). 1106 * 1107 * for case [2], we need to remove the mapping from 1108 * the pmap. there shouldn't be any pages at the 1109 * specified offset in the kernel object [but it 1110 * doesn't hurt to call uvm_km_pgremove just to be 1111 * safe?] 1112 * 1113 * uvm_km_pgremove currently does the following: 1114 * for pages in the kernel object in range: 1115 * - drops the swap slot 1116 * - uvm_pagefree the page 1117 */ 1118 1119 /* 1120 * remove mappings from pmap and drop the pages 1121 * from the object. offsets are always relative 1122 * to vm_map_min(kernel_map). 1123 */ 1124 1125 pmap_remove(pmap_kernel(), entry->start, 1126 entry->start + len); 1127 uvm_km_pgremove(entry->object.uvm_obj, 1128 entry->start - vm_map_min(kernel_map), 1129 entry->end - vm_map_min(kernel_map)); 1130 1131 /* 1132 * null out kernel_object reference, we've just 1133 * dropped it 1134 */ 1135 1136 entry->etype &= ~UVM_ET_OBJ; 1137 entry->object.uvm_obj = NULL; 1138 } else if (UVM_ET_ISOBJ(entry) || entry->aref.ar_amap) { 1139 1140 /* 1141 * remove mappings the standard way. 1142 */ 1143 1144 pmap_remove(map->pmap, entry->start, entry->end); 1145 } 1146 1147 /* 1148 * remove entry from map and put it on our list of entries 1149 * that we've nuked. then go to next entry. 1150 */ 1151 1152 UVMHIST_LOG(maphist, " removed map entry 0x%x", entry, 0, 0,0); 1153 1154 /* critical! prevents stale hint */ 1155 SAVE_HINT(map, entry, entry->prev); 1156 1157 uvm_map_entry_unlink(map, entry); 1158 map->size -= len; 1159 entry->next = first_entry; 1160 first_entry = entry; 1161 entry = next; 1162 } 1163 pmap_update(vm_map_pmap(map)); 1164 1165 /* 1166 * now we've cleaned up the map and are ready for the caller to drop 1167 * references to the mapped objects. 1168 */ 1169 1170 *entry_list = first_entry; 1171 UVMHIST_LOG(maphist,"<- done!", 0, 0, 0, 0); 1172 } 1173 1174 /* 1175 * uvm_unmap_detach: drop references in a chain of map entries 1176 * 1177 * => we will free the map entries as we traverse the list. 1178 */ 1179 1180 void 1181 uvm_unmap_detach(first_entry, flags) 1182 struct vm_map_entry *first_entry; 1183 int flags; 1184 { 1185 struct vm_map_entry *next_entry; 1186 UVMHIST_FUNC("uvm_unmap_detach"); UVMHIST_CALLED(maphist); 1187 1188 while (first_entry) { 1189 KASSERT(!VM_MAPENT_ISWIRED(first_entry)); 1190 UVMHIST_LOG(maphist, 1191 " detach 0x%x: amap=0x%x, obj=0x%x, submap?=%d", 1192 first_entry, first_entry->aref.ar_amap, 1193 first_entry->object.uvm_obj, 1194 UVM_ET_ISSUBMAP(first_entry)); 1195 1196 /* 1197 * drop reference to amap, if we've got one 1198 */ 1199 1200 if (first_entry->aref.ar_amap) 1201 uvm_map_unreference_amap(first_entry, flags); 1202 1203 /* 1204 * drop reference to our backing object, if we've got one 1205 */ 1206 1207 if (UVM_ET_ISSUBMAP(first_entry)) { 1208 /* ... unlikely to happen, but play it safe */ 1209 uvm_map_deallocate(first_entry->object.sub_map); 1210 } else { 1211 if (UVM_ET_ISOBJ(first_entry) && 1212 first_entry->object.uvm_obj->pgops->pgo_detach) 1213 first_entry->object.uvm_obj->pgops-> 1214 pgo_detach(first_entry->object.uvm_obj); 1215 } 1216 next_entry = first_entry->next; 1217 uvm_mapent_free(first_entry); 1218 first_entry = next_entry; 1219 } 1220 UVMHIST_LOG(maphist, "<- done", 0,0,0,0); 1221 } 1222 1223 /* 1224 * E X T R A C T I O N F U N C T I O N S 1225 */ 1226 1227 /* 1228 * uvm_map_reserve: reserve space in a vm_map for future use. 1229 * 1230 * => we reserve space in a map by putting a dummy map entry in the 1231 * map (dummy means obj=NULL, amap=NULL, prot=VM_PROT_NONE) 1232 * => map should be unlocked (we will write lock it) 1233 * => we return true if we were able to reserve space 1234 * => XXXCDC: should be inline? 1235 */ 1236 1237 int 1238 uvm_map_reserve(map, size, offset, align, raddr) 1239 struct vm_map *map; 1240 vsize_t size; 1241 vaddr_t offset; /* hint for pmap_prefer */ 1242 vsize_t align; /* alignment hint */ 1243 vaddr_t *raddr; /* IN:hint, OUT: reserved VA */ 1244 { 1245 UVMHIST_FUNC("uvm_map_reserve"); UVMHIST_CALLED(maphist); 1246 1247 UVMHIST_LOG(maphist, "(map=0x%x, size=0x%x, offset=0x%x,addr=0x%x)", 1248 map,size,offset,raddr); 1249 1250 size = round_page(size); 1251 if (*raddr < vm_map_min(map)) 1252 *raddr = vm_map_min(map); /* hint */ 1253 1254 /* 1255 * reserve some virtual space. 1256 */ 1257 1258 if (uvm_map(map, raddr, size, NULL, offset, 0, 1259 UVM_MAPFLAG(UVM_PROT_NONE, UVM_PROT_NONE, UVM_INH_NONE, 1260 UVM_ADV_RANDOM, UVM_FLAG_NOMERGE)) != 0) { 1261 UVMHIST_LOG(maphist, "<- done (no VM)", 0,0,0,0); 1262 return (FALSE); 1263 } 1264 1265 UVMHIST_LOG(maphist, "<- done (*raddr=0x%x)", *raddr,0,0,0); 1266 return (TRUE); 1267 } 1268 1269 /* 1270 * uvm_map_replace: replace a reserved (blank) area of memory with 1271 * real mappings. 1272 * 1273 * => caller must WRITE-LOCK the map 1274 * => we return TRUE if replacement was a success 1275 * => we expect the newents chain to have nnewents entrys on it and 1276 * we expect newents->prev to point to the last entry on the list 1277 * => note newents is allowed to be NULL 1278 */ 1279 1280 int 1281 uvm_map_replace(map, start, end, newents, nnewents) 1282 struct vm_map *map; 1283 vaddr_t start, end; 1284 struct vm_map_entry *newents; 1285 int nnewents; 1286 { 1287 struct vm_map_entry *oldent, *last; 1288 1289 /* 1290 * first find the blank map entry at the specified address 1291 */ 1292 1293 if (!uvm_map_lookup_entry(map, start, &oldent)) { 1294 return(FALSE); 1295 } 1296 1297 /* 1298 * check to make sure we have a proper blank entry 1299 */ 1300 1301 if (oldent->start != start || oldent->end != end || 1302 oldent->object.uvm_obj != NULL || oldent->aref.ar_amap != NULL) { 1303 return (FALSE); 1304 } 1305 1306 #ifdef DIAGNOSTIC 1307 1308 /* 1309 * sanity check the newents chain 1310 */ 1311 1312 { 1313 struct vm_map_entry *tmpent = newents; 1314 int nent = 0; 1315 vaddr_t cur = start; 1316 1317 while (tmpent) { 1318 nent++; 1319 if (tmpent->start < cur) 1320 panic("uvm_map_replace1"); 1321 if (tmpent->start > tmpent->end || tmpent->end > end) { 1322 printf("tmpent->start=0x%lx, tmpent->end=0x%lx, end=0x%lx\n", 1323 tmpent->start, tmpent->end, end); 1324 panic("uvm_map_replace2"); 1325 } 1326 cur = tmpent->end; 1327 if (tmpent->next) { 1328 if (tmpent->next->prev != tmpent) 1329 panic("uvm_map_replace3"); 1330 } else { 1331 if (newents->prev != tmpent) 1332 panic("uvm_map_replace4"); 1333 } 1334 tmpent = tmpent->next; 1335 } 1336 if (nent != nnewents) 1337 panic("uvm_map_replace5"); 1338 } 1339 #endif 1340 1341 /* 1342 * map entry is a valid blank! replace it. (this does all the 1343 * work of map entry link/unlink...). 1344 */ 1345 1346 if (newents) { 1347 last = newents->prev; 1348 1349 /* critical: flush stale hints out of map */ 1350 SAVE_HINT(map, map->hint, newents); 1351 if (map->first_free == oldent) 1352 map->first_free = last; 1353 1354 last->next = oldent->next; 1355 last->next->prev = last; 1356 newents->prev = oldent->prev; 1357 newents->prev->next = newents; 1358 map->nentries = map->nentries + (nnewents - 1); 1359 1360 } else { 1361 1362 /* critical: flush stale hints out of map */ 1363 SAVE_HINT(map, map->hint, oldent->prev); 1364 if (map->first_free == oldent) 1365 map->first_free = oldent->prev; 1366 1367 /* NULL list of new entries: just remove the old one */ 1368 uvm_map_entry_unlink(map, oldent); 1369 } 1370 1371 1372 /* 1373 * now we can free the old blank entry, unlock the map and return. 1374 */ 1375 1376 uvm_mapent_free(oldent); 1377 return(TRUE); 1378 } 1379 1380 /* 1381 * uvm_map_extract: extract a mapping from a map and put it somewhere 1382 * (maybe removing the old mapping) 1383 * 1384 * => maps should be unlocked (we will write lock them) 1385 * => returns 0 on success, error code otherwise 1386 * => start must be page aligned 1387 * => len must be page sized 1388 * => flags: 1389 * UVM_EXTRACT_REMOVE: remove mappings from srcmap 1390 * UVM_EXTRACT_CONTIG: abort if unmapped area (advisory only) 1391 * UVM_EXTRACT_QREF: for a temporary extraction do quick obj refs 1392 * UVM_EXTRACT_FIXPROT: set prot to maxprot as we go 1393 * >>>NOTE: if you set REMOVE, you are not allowed to use CONTIG or QREF!<<< 1394 * >>>NOTE: QREF's must be unmapped via the QREF path, thus should only 1395 * be used from within the kernel in a kernel level map <<< 1396 */ 1397 1398 int 1399 uvm_map_extract(srcmap, start, len, dstmap, dstaddrp, flags) 1400 struct vm_map *srcmap, *dstmap; 1401 vaddr_t start, *dstaddrp; 1402 vsize_t len; 1403 int flags; 1404 { 1405 vaddr_t dstaddr, end, newend, oldoffset, fudge, orig_fudge, 1406 oldstart; 1407 struct vm_map_entry *chain, *endchain, *entry, *orig_entry, *newentry, 1408 *deadentry, *oldentry; 1409 vsize_t elen; 1410 int nchain, error, copy_ok; 1411 UVMHIST_FUNC("uvm_map_extract"); UVMHIST_CALLED(maphist); 1412 1413 UVMHIST_LOG(maphist,"(srcmap=0x%x,start=0x%x, len=0x%x", srcmap, start, 1414 len,0); 1415 UVMHIST_LOG(maphist," ...,dstmap=0x%x, flags=0x%x)", dstmap,flags,0,0); 1416 1417 /* 1418 * step 0: sanity check: start must be on a page boundary, length 1419 * must be page sized. can't ask for CONTIG/QREF if you asked for 1420 * REMOVE. 1421 */ 1422 1423 KASSERT((start & PAGE_MASK) == 0 && (len & PAGE_MASK) == 0); 1424 KASSERT((flags & UVM_EXTRACT_REMOVE) == 0 || 1425 (flags & (UVM_EXTRACT_CONTIG|UVM_EXTRACT_QREF)) == 0); 1426 1427 /* 1428 * step 1: reserve space in the target map for the extracted area 1429 */ 1430 1431 dstaddr = vm_map_min(dstmap); 1432 if (uvm_map_reserve(dstmap, len, start, 0, &dstaddr) == FALSE) 1433 return(ENOMEM); 1434 *dstaddrp = dstaddr; /* pass address back to caller */ 1435 UVMHIST_LOG(maphist, " dstaddr=0x%x", dstaddr,0,0,0); 1436 1437 /* 1438 * step 2: setup for the extraction process loop by init'ing the 1439 * map entry chain, locking src map, and looking up the first useful 1440 * entry in the map. 1441 */ 1442 1443 end = start + len; 1444 newend = dstaddr + len; 1445 chain = endchain = NULL; 1446 nchain = 0; 1447 vm_map_lock(srcmap); 1448 1449 if (uvm_map_lookup_entry(srcmap, start, &entry)) { 1450 1451 /* "start" is within an entry */ 1452 if (flags & UVM_EXTRACT_QREF) { 1453 1454 /* 1455 * for quick references we don't clip the entry, so 1456 * the entry may map space "before" the starting 1457 * virtual address... this is the "fudge" factor 1458 * (which can be non-zero only the first time 1459 * through the "while" loop in step 3). 1460 */ 1461 1462 fudge = start - entry->start; 1463 } else { 1464 1465 /* 1466 * normal reference: we clip the map to fit (thus 1467 * fudge is zero) 1468 */ 1469 1470 UVM_MAP_CLIP_START(srcmap, entry, start); 1471 SAVE_HINT(srcmap, srcmap->hint, entry->prev); 1472 fudge = 0; 1473 } 1474 } else { 1475 1476 /* "start" is not within an entry ... skip to next entry */ 1477 if (flags & UVM_EXTRACT_CONTIG) { 1478 error = EINVAL; 1479 goto bad; /* definite hole here ... */ 1480 } 1481 1482 entry = entry->next; 1483 fudge = 0; 1484 } 1485 1486 /* save values from srcmap for step 6 */ 1487 orig_entry = entry; 1488 orig_fudge = fudge; 1489 1490 /* 1491 * step 3: now start looping through the map entries, extracting 1492 * as we go. 1493 */ 1494 1495 while (entry->start < end && entry != &srcmap->header) { 1496 1497 /* if we are not doing a quick reference, clip it */ 1498 if ((flags & UVM_EXTRACT_QREF) == 0) 1499 UVM_MAP_CLIP_END(srcmap, entry, end); 1500 1501 /* clear needs_copy (allow chunking) */ 1502 if (UVM_ET_ISNEEDSCOPY(entry)) { 1503 if (fudge) 1504 oldstart = entry->start; 1505 else 1506 oldstart = 0; /* XXX: gcc */ 1507 amap_copy(srcmap, entry, M_NOWAIT, TRUE, start, end); 1508 if (UVM_ET_ISNEEDSCOPY(entry)) { /* failed? */ 1509 error = ENOMEM; 1510 goto bad; 1511 } 1512 1513 /* amap_copy could clip (during chunk)! update fudge */ 1514 if (fudge) { 1515 fudge = fudge - (entry->start - oldstart); 1516 orig_fudge = fudge; 1517 } 1518 } 1519 1520 /* calculate the offset of this from "start" */ 1521 oldoffset = (entry->start + fudge) - start; 1522 1523 /* allocate a new map entry */ 1524 newentry = uvm_mapent_alloc(dstmap); 1525 if (newentry == NULL) { 1526 error = ENOMEM; 1527 goto bad; 1528 } 1529 1530 /* set up new map entry */ 1531 newentry->next = NULL; 1532 newentry->prev = endchain; 1533 newentry->start = dstaddr + oldoffset; 1534 newentry->end = 1535 newentry->start + (entry->end - (entry->start + fudge)); 1536 if (newentry->end > newend || newentry->end < newentry->start) 1537 newentry->end = newend; 1538 newentry->object.uvm_obj = entry->object.uvm_obj; 1539 if (newentry->object.uvm_obj) { 1540 if (newentry->object.uvm_obj->pgops->pgo_reference) 1541 newentry->object.uvm_obj->pgops-> 1542 pgo_reference(newentry->object.uvm_obj); 1543 newentry->offset = entry->offset + fudge; 1544 } else { 1545 newentry->offset = 0; 1546 } 1547 newentry->etype = entry->etype; 1548 newentry->protection = (flags & UVM_EXTRACT_FIXPROT) ? 1549 entry->max_protection : entry->protection; 1550 newentry->max_protection = entry->max_protection; 1551 newentry->inheritance = entry->inheritance; 1552 newentry->wired_count = 0; 1553 newentry->aref.ar_amap = entry->aref.ar_amap; 1554 if (newentry->aref.ar_amap) { 1555 newentry->aref.ar_pageoff = 1556 entry->aref.ar_pageoff + (fudge >> PAGE_SHIFT); 1557 uvm_map_reference_amap(newentry, AMAP_SHARED | 1558 ((flags & UVM_EXTRACT_QREF) ? AMAP_REFALL : 0)); 1559 } else { 1560 newentry->aref.ar_pageoff = 0; 1561 } 1562 newentry->advice = entry->advice; 1563 1564 /* now link it on the chain */ 1565 nchain++; 1566 if (endchain == NULL) { 1567 chain = endchain = newentry; 1568 } else { 1569 endchain->next = newentry; 1570 endchain = newentry; 1571 } 1572 1573 /* end of 'while' loop! */ 1574 if ((flags & UVM_EXTRACT_CONTIG) && entry->end < end && 1575 (entry->next == &srcmap->header || 1576 entry->next->start != entry->end)) { 1577 error = EINVAL; 1578 goto bad; 1579 } 1580 entry = entry->next; 1581 fudge = 0; 1582 } 1583 1584 /* 1585 * step 4: close off chain (in format expected by uvm_map_replace) 1586 */ 1587 1588 if (chain) 1589 chain->prev = endchain; 1590 1591 /* 1592 * step 5: attempt to lock the dest map so we can pmap_copy. 1593 * note usage of copy_ok: 1594 * 1 => dstmap locked, pmap_copy ok, and we "replace" here (step 5) 1595 * 0 => dstmap unlocked, NO pmap_copy, and we will "replace" in step 7 1596 */ 1597 1598 if (srcmap == dstmap || vm_map_lock_try(dstmap) == TRUE) { 1599 copy_ok = 1; 1600 if (!uvm_map_replace(dstmap, dstaddr, dstaddr+len, chain, 1601 nchain)) { 1602 if (srcmap != dstmap) 1603 vm_map_unlock(dstmap); 1604 error = EIO; 1605 goto bad; 1606 } 1607 } else { 1608 copy_ok = 0; 1609 /* replace defered until step 7 */ 1610 } 1611 1612 /* 1613 * step 6: traverse the srcmap a second time to do the following: 1614 * - if we got a lock on the dstmap do pmap_copy 1615 * - if UVM_EXTRACT_REMOVE remove the entries 1616 * we make use of orig_entry and orig_fudge (saved in step 2) 1617 */ 1618 1619 if (copy_ok || (flags & UVM_EXTRACT_REMOVE)) { 1620 1621 /* purge possible stale hints from srcmap */ 1622 if (flags & UVM_EXTRACT_REMOVE) { 1623 SAVE_HINT(srcmap, srcmap->hint, orig_entry->prev); 1624 if (srcmap->first_free->start >= start) 1625 srcmap->first_free = orig_entry->prev; 1626 } 1627 1628 entry = orig_entry; 1629 fudge = orig_fudge; 1630 deadentry = NULL; /* for UVM_EXTRACT_REMOVE */ 1631 1632 while (entry->start < end && entry != &srcmap->header) { 1633 if (copy_ok) { 1634 oldoffset = (entry->start + fudge) - start; 1635 elen = MIN(end, entry->end) - 1636 (entry->start + fudge); 1637 pmap_copy(dstmap->pmap, srcmap->pmap, 1638 dstaddr + oldoffset, elen, 1639 entry->start + fudge); 1640 } 1641 1642 /* we advance "entry" in the following if statement */ 1643 if (flags & UVM_EXTRACT_REMOVE) { 1644 pmap_remove(srcmap->pmap, entry->start, 1645 entry->end); 1646 oldentry = entry; /* save entry */ 1647 entry = entry->next; /* advance */ 1648 uvm_map_entry_unlink(srcmap, oldentry); 1649 /* add to dead list */ 1650 oldentry->next = deadentry; 1651 deadentry = oldentry; 1652 } else { 1653 entry = entry->next; /* advance */ 1654 } 1655 1656 /* end of 'while' loop */ 1657 fudge = 0; 1658 } 1659 pmap_update(srcmap->pmap); 1660 1661 /* 1662 * unlock dstmap. we will dispose of deadentry in 1663 * step 7 if needed 1664 */ 1665 1666 if (copy_ok && srcmap != dstmap) 1667 vm_map_unlock(dstmap); 1668 1669 } else { 1670 deadentry = NULL; 1671 } 1672 1673 /* 1674 * step 7: we are done with the source map, unlock. if copy_ok 1675 * is 0 then we have not replaced the dummy mapping in dstmap yet 1676 * and we need to do so now. 1677 */ 1678 1679 vm_map_unlock(srcmap); 1680 if ((flags & UVM_EXTRACT_REMOVE) && deadentry) 1681 uvm_unmap_detach(deadentry, 0); /* dispose of old entries */ 1682 1683 /* now do the replacement if we didn't do it in step 5 */ 1684 if (copy_ok == 0) { 1685 vm_map_lock(dstmap); 1686 error = uvm_map_replace(dstmap, dstaddr, dstaddr+len, chain, 1687 nchain); 1688 vm_map_unlock(dstmap); 1689 1690 if (error == FALSE) { 1691 error = EIO; 1692 goto bad2; 1693 } 1694 } 1695 return(0); 1696 1697 /* 1698 * bad: failure recovery 1699 */ 1700 bad: 1701 vm_map_unlock(srcmap); 1702 bad2: /* src already unlocked */ 1703 if (chain) 1704 uvm_unmap_detach(chain, 1705 (flags & UVM_EXTRACT_QREF) ? AMAP_REFALL : 0); 1706 uvm_unmap(dstmap, dstaddr, dstaddr+len); /* ??? */ 1707 return(error); 1708 } 1709 1710 /* end of extraction functions */ 1711 1712 /* 1713 * uvm_map_submap: punch down part of a map into a submap 1714 * 1715 * => only the kernel_map is allowed to be submapped 1716 * => the purpose of submapping is to break up the locking granularity 1717 * of a larger map 1718 * => the range specified must have been mapped previously with a uvm_map() 1719 * call [with uobj==NULL] to create a blank map entry in the main map. 1720 * [And it had better still be blank!] 1721 * => maps which contain submaps should never be copied or forked. 1722 * => to remove a submap, use uvm_unmap() on the main map 1723 * and then uvm_map_deallocate() the submap. 1724 * => main map must be unlocked. 1725 * => submap must have been init'd and have a zero reference count. 1726 * [need not be locked as we don't actually reference it] 1727 */ 1728 1729 int 1730 uvm_map_submap(map, start, end, submap) 1731 struct vm_map *map, *submap; 1732 vaddr_t start, end; 1733 { 1734 struct vm_map_entry *entry; 1735 int error; 1736 1737 vm_map_lock(map); 1738 VM_MAP_RANGE_CHECK(map, start, end); 1739 1740 if (uvm_map_lookup_entry(map, start, &entry)) { 1741 UVM_MAP_CLIP_START(map, entry, start); 1742 UVM_MAP_CLIP_END(map, entry, end); /* to be safe */ 1743 } else { 1744 entry = NULL; 1745 } 1746 1747 if (entry != NULL && 1748 entry->start == start && entry->end == end && 1749 entry->object.uvm_obj == NULL && entry->aref.ar_amap == NULL && 1750 !UVM_ET_ISCOPYONWRITE(entry) && !UVM_ET_ISNEEDSCOPY(entry)) { 1751 entry->etype |= UVM_ET_SUBMAP; 1752 entry->object.sub_map = submap; 1753 entry->offset = 0; 1754 uvm_map_reference(submap); 1755 error = 0; 1756 } else { 1757 error = EINVAL; 1758 } 1759 vm_map_unlock(map); 1760 return error; 1761 } 1762 1763 1764 /* 1765 * uvm_map_protect: change map protection 1766 * 1767 * => set_max means set max_protection. 1768 * => map must be unlocked. 1769 */ 1770 1771 #define MASK(entry) (UVM_ET_ISCOPYONWRITE(entry) ? \ 1772 ~VM_PROT_WRITE : VM_PROT_ALL) 1773 1774 int 1775 uvm_map_protect(map, start, end, new_prot, set_max) 1776 struct vm_map *map; 1777 vaddr_t start, end; 1778 vm_prot_t new_prot; 1779 boolean_t set_max; 1780 { 1781 struct vm_map_entry *current, *entry; 1782 int error = 0; 1783 UVMHIST_FUNC("uvm_map_protect"); UVMHIST_CALLED(maphist); 1784 UVMHIST_LOG(maphist,"(map=0x%x,start=0x%x,end=0x%x,new_prot=0x%x)", 1785 map, start, end, new_prot); 1786 1787 vm_map_lock(map); 1788 VM_MAP_RANGE_CHECK(map, start, end); 1789 if (uvm_map_lookup_entry(map, start, &entry)) { 1790 UVM_MAP_CLIP_START(map, entry, start); 1791 } else { 1792 entry = entry->next; 1793 } 1794 1795 /* 1796 * make a first pass to check for protection violations. 1797 */ 1798 1799 current = entry; 1800 while ((current != &map->header) && (current->start < end)) { 1801 if (UVM_ET_ISSUBMAP(current)) { 1802 error = EINVAL; 1803 goto out; 1804 } 1805 if ((new_prot & current->max_protection) != new_prot) { 1806 error = EACCES; 1807 goto out; 1808 } 1809 /* 1810 * Don't allow VM_PROT_EXECUTE to be set on entries that 1811 * point to vnodes that are associated with a NOEXEC file 1812 * system. 1813 */ 1814 if (UVM_ET_ISOBJ(current) && 1815 UVM_OBJ_IS_VNODE(current->object.uvm_obj)) { 1816 struct vnode *vp = 1817 (struct vnode *) current->object.uvm_obj; 1818 1819 if ((new_prot & VM_PROT_EXECUTE) != 0 && 1820 (vp->v_mount->mnt_flag & MNT_NOEXEC) != 0) { 1821 error = EACCES; 1822 goto out; 1823 } 1824 } 1825 current = current->next; 1826 } 1827 1828 /* go back and fix up protections (no need to clip this time). */ 1829 1830 current = entry; 1831 while ((current != &map->header) && (current->start < end)) { 1832 vm_prot_t old_prot; 1833 1834 UVM_MAP_CLIP_END(map, current, end); 1835 old_prot = current->protection; 1836 if (set_max) 1837 current->protection = 1838 (current->max_protection = new_prot) & old_prot; 1839 else 1840 current->protection = new_prot; 1841 1842 /* 1843 * update physical map if necessary. worry about copy-on-write 1844 * here -- CHECK THIS XXX 1845 */ 1846 1847 if (current->protection != old_prot) { 1848 /* update pmap! */ 1849 pmap_protect(map->pmap, current->start, current->end, 1850 current->protection & MASK(entry)); 1851 1852 /* 1853 * If this entry points at a vnode, and the 1854 * protection includes VM_PROT_EXECUTE, mark 1855 * the vnode as VEXECMAP. 1856 */ 1857 if (UVM_ET_ISOBJ(current)) { 1858 struct uvm_object *uobj = 1859 current->object.uvm_obj; 1860 1861 if (UVM_OBJ_IS_VNODE(uobj) && 1862 (current->protection & VM_PROT_EXECUTE)) 1863 vn_markexec((struct vnode *) uobj); 1864 } 1865 } 1866 1867 /* 1868 * If the map is configured to lock any future mappings, 1869 * wire this entry now if the old protection was VM_PROT_NONE 1870 * and the new protection is not VM_PROT_NONE. 1871 */ 1872 1873 if ((map->flags & VM_MAP_WIREFUTURE) != 0 && 1874 VM_MAPENT_ISWIRED(entry) == 0 && 1875 old_prot == VM_PROT_NONE && 1876 new_prot != VM_PROT_NONE) { 1877 if (uvm_map_pageable(map, entry->start, 1878 entry->end, FALSE, 1879 UVM_LK_ENTER|UVM_LK_EXIT) != 0) { 1880 1881 /* 1882 * If locking the entry fails, remember the 1883 * error if it's the first one. Note we 1884 * still continue setting the protection in 1885 * the map, but will return the error 1886 * condition regardless. 1887 * 1888 * XXX Ignore what the actual error is, 1889 * XXX just call it a resource shortage 1890 * XXX so that it doesn't get confused 1891 * XXX what uvm_map_protect() itself would 1892 * XXX normally return. 1893 */ 1894 1895 error = ENOMEM; 1896 } 1897 } 1898 current = current->next; 1899 } 1900 pmap_update(map->pmap); 1901 1902 out: 1903 vm_map_unlock(map); 1904 UVMHIST_LOG(maphist, "<- done, error=%d",error,0,0,0); 1905 return error; 1906 } 1907 1908 #undef MASK 1909 1910 /* 1911 * uvm_map_inherit: set inheritance code for range of addrs in map. 1912 * 1913 * => map must be unlocked 1914 * => note that the inherit code is used during a "fork". see fork 1915 * code for details. 1916 */ 1917 1918 int 1919 uvm_map_inherit(map, start, end, new_inheritance) 1920 struct vm_map *map; 1921 vaddr_t start; 1922 vaddr_t end; 1923 vm_inherit_t new_inheritance; 1924 { 1925 struct vm_map_entry *entry, *temp_entry; 1926 UVMHIST_FUNC("uvm_map_inherit"); UVMHIST_CALLED(maphist); 1927 UVMHIST_LOG(maphist,"(map=0x%x,start=0x%x,end=0x%x,new_inh=0x%x)", 1928 map, start, end, new_inheritance); 1929 1930 switch (new_inheritance) { 1931 case MAP_INHERIT_NONE: 1932 case MAP_INHERIT_COPY: 1933 case MAP_INHERIT_SHARE: 1934 break; 1935 default: 1936 UVMHIST_LOG(maphist,"<- done (INVALID ARG)",0,0,0,0); 1937 return EINVAL; 1938 } 1939 1940 vm_map_lock(map); 1941 VM_MAP_RANGE_CHECK(map, start, end); 1942 if (uvm_map_lookup_entry(map, start, &temp_entry)) { 1943 entry = temp_entry; 1944 UVM_MAP_CLIP_START(map, entry, start); 1945 } else { 1946 entry = temp_entry->next; 1947 } 1948 while ((entry != &map->header) && (entry->start < end)) { 1949 UVM_MAP_CLIP_END(map, entry, end); 1950 entry->inheritance = new_inheritance; 1951 entry = entry->next; 1952 } 1953 vm_map_unlock(map); 1954 UVMHIST_LOG(maphist,"<- done (OK)",0,0,0,0); 1955 return 0; 1956 } 1957 1958 /* 1959 * uvm_map_advice: set advice code for range of addrs in map. 1960 * 1961 * => map must be unlocked 1962 */ 1963 1964 int 1965 uvm_map_advice(map, start, end, new_advice) 1966 struct vm_map *map; 1967 vaddr_t start; 1968 vaddr_t end; 1969 int new_advice; 1970 { 1971 struct vm_map_entry *entry, *temp_entry; 1972 UVMHIST_FUNC("uvm_map_advice"); UVMHIST_CALLED(maphist); 1973 UVMHIST_LOG(maphist,"(map=0x%x,start=0x%x,end=0x%x,new_adv=0x%x)", 1974 map, start, end, new_advice); 1975 1976 vm_map_lock(map); 1977 VM_MAP_RANGE_CHECK(map, start, end); 1978 if (uvm_map_lookup_entry(map, start, &temp_entry)) { 1979 entry = temp_entry; 1980 UVM_MAP_CLIP_START(map, entry, start); 1981 } else { 1982 entry = temp_entry->next; 1983 } 1984 1985 /* 1986 * XXXJRT: disallow holes? 1987 */ 1988 1989 while ((entry != &map->header) && (entry->start < end)) { 1990 UVM_MAP_CLIP_END(map, entry, end); 1991 1992 switch (new_advice) { 1993 case MADV_NORMAL: 1994 case MADV_RANDOM: 1995 case MADV_SEQUENTIAL: 1996 /* nothing special here */ 1997 break; 1998 1999 default: 2000 vm_map_unlock(map); 2001 UVMHIST_LOG(maphist,"<- done (INVALID ARG)",0,0,0,0); 2002 return EINVAL; 2003 } 2004 entry->advice = new_advice; 2005 entry = entry->next; 2006 } 2007 2008 vm_map_unlock(map); 2009 UVMHIST_LOG(maphist,"<- done (OK)",0,0,0,0); 2010 return 0; 2011 } 2012 2013 /* 2014 * uvm_map_pageable: sets the pageability of a range in a map. 2015 * 2016 * => wires map entries. should not be used for transient page locking. 2017 * for that, use uvm_fault_wire()/uvm_fault_unwire() (see uvm_vslock()). 2018 * => regions sepcified as not pageable require lock-down (wired) memory 2019 * and page tables. 2020 * => map must never be read-locked 2021 * => if islocked is TRUE, map is already write-locked 2022 * => we always unlock the map, since we must downgrade to a read-lock 2023 * to call uvm_fault_wire() 2024 * => XXXCDC: check this and try and clean it up. 2025 */ 2026 2027 int 2028 uvm_map_pageable(map, start, end, new_pageable, lockflags) 2029 struct vm_map *map; 2030 vaddr_t start, end; 2031 boolean_t new_pageable; 2032 int lockflags; 2033 { 2034 struct vm_map_entry *entry, *start_entry, *failed_entry; 2035 int rv; 2036 #ifdef DIAGNOSTIC 2037 u_int timestamp_save; 2038 #endif 2039 UVMHIST_FUNC("uvm_map_pageable"); UVMHIST_CALLED(maphist); 2040 UVMHIST_LOG(maphist,"(map=0x%x,start=0x%x,end=0x%x,new_pageable=0x%x)", 2041 map, start, end, new_pageable); 2042 KASSERT(map->flags & VM_MAP_PAGEABLE); 2043 2044 if ((lockflags & UVM_LK_ENTER) == 0) 2045 vm_map_lock(map); 2046 VM_MAP_RANGE_CHECK(map, start, end); 2047 2048 /* 2049 * only one pageability change may take place at one time, since 2050 * uvm_fault_wire assumes it will be called only once for each 2051 * wiring/unwiring. therefore, we have to make sure we're actually 2052 * changing the pageability for the entire region. we do so before 2053 * making any changes. 2054 */ 2055 2056 if (uvm_map_lookup_entry(map, start, &start_entry) == FALSE) { 2057 if ((lockflags & UVM_LK_EXIT) == 0) 2058 vm_map_unlock(map); 2059 2060 UVMHIST_LOG(maphist,"<- done (fault)",0,0,0,0); 2061 return EFAULT; 2062 } 2063 entry = start_entry; 2064 2065 /* 2066 * handle wiring and unwiring separately. 2067 */ 2068 2069 if (new_pageable) { /* unwire */ 2070 UVM_MAP_CLIP_START(map, entry, start); 2071 2072 /* 2073 * unwiring. first ensure that the range to be unwired is 2074 * really wired down and that there are no holes. 2075 */ 2076 2077 while ((entry != &map->header) && (entry->start < end)) { 2078 if (entry->wired_count == 0 || 2079 (entry->end < end && 2080 (entry->next == &map->header || 2081 entry->next->start > entry->end))) { 2082 if ((lockflags & UVM_LK_EXIT) == 0) 2083 vm_map_unlock(map); 2084 UVMHIST_LOG(maphist, "<- done (INVAL)",0,0,0,0); 2085 return EINVAL; 2086 } 2087 entry = entry->next; 2088 } 2089 2090 /* 2091 * POSIX 1003.1b - a single munlock call unlocks a region, 2092 * regardless of the number of mlock calls made on that 2093 * region. 2094 */ 2095 2096 entry = start_entry; 2097 while ((entry != &map->header) && (entry->start < end)) { 2098 UVM_MAP_CLIP_END(map, entry, end); 2099 if (VM_MAPENT_ISWIRED(entry)) 2100 uvm_map_entry_unwire(map, entry); 2101 entry = entry->next; 2102 } 2103 if ((lockflags & UVM_LK_EXIT) == 0) 2104 vm_map_unlock(map); 2105 UVMHIST_LOG(maphist,"<- done (OK UNWIRE)",0,0,0,0); 2106 return 0; 2107 } 2108 2109 /* 2110 * wire case: in two passes [XXXCDC: ugly block of code here] 2111 * 2112 * 1: holding the write lock, we create any anonymous maps that need 2113 * to be created. then we clip each map entry to the region to 2114 * be wired and increment its wiring count. 2115 * 2116 * 2: we downgrade to a read lock, and call uvm_fault_wire to fault 2117 * in the pages for any newly wired area (wired_count == 1). 2118 * 2119 * downgrading to a read lock for uvm_fault_wire avoids a possible 2120 * deadlock with another thread that may have faulted on one of 2121 * the pages to be wired (it would mark the page busy, blocking 2122 * us, then in turn block on the map lock that we hold). because 2123 * of problems in the recursive lock package, we cannot upgrade 2124 * to a write lock in vm_map_lookup. thus, any actions that 2125 * require the write lock must be done beforehand. because we 2126 * keep the read lock on the map, the copy-on-write status of the 2127 * entries we modify here cannot change. 2128 */ 2129 2130 while ((entry != &map->header) && (entry->start < end)) { 2131 if (VM_MAPENT_ISWIRED(entry) == 0) { /* not already wired? */ 2132 2133 /* 2134 * perform actions of vm_map_lookup that need the 2135 * write lock on the map: create an anonymous map 2136 * for a copy-on-write region, or an anonymous map 2137 * for a zero-fill region. (XXXCDC: submap case 2138 * ok?) 2139 */ 2140 2141 if (!UVM_ET_ISSUBMAP(entry)) { /* not submap */ 2142 if (UVM_ET_ISNEEDSCOPY(entry) && 2143 ((entry->max_protection & VM_PROT_WRITE) || 2144 (entry->object.uvm_obj == NULL))) { 2145 amap_copy(map, entry, M_WAITOK, TRUE, 2146 start, end); 2147 /* XXXCDC: wait OK? */ 2148 } 2149 } 2150 } 2151 UVM_MAP_CLIP_START(map, entry, start); 2152 UVM_MAP_CLIP_END(map, entry, end); 2153 entry->wired_count++; 2154 2155 /* 2156 * Check for holes 2157 */ 2158 2159 if (entry->protection == VM_PROT_NONE || 2160 (entry->end < end && 2161 (entry->next == &map->header || 2162 entry->next->start > entry->end))) { 2163 2164 /* 2165 * found one. amap creation actions do not need to 2166 * be undone, but the wired counts need to be restored. 2167 */ 2168 2169 while (entry != &map->header && entry->end > start) { 2170 entry->wired_count--; 2171 entry = entry->prev; 2172 } 2173 if ((lockflags & UVM_LK_EXIT) == 0) 2174 vm_map_unlock(map); 2175 UVMHIST_LOG(maphist,"<- done (INVALID WIRE)",0,0,0,0); 2176 return EINVAL; 2177 } 2178 entry = entry->next; 2179 } 2180 2181 /* 2182 * Pass 2. 2183 */ 2184 2185 #ifdef DIAGNOSTIC 2186 timestamp_save = map->timestamp; 2187 #endif 2188 vm_map_busy(map); 2189 vm_map_downgrade(map); 2190 2191 rv = 0; 2192 entry = start_entry; 2193 while (entry != &map->header && entry->start < end) { 2194 if (entry->wired_count == 1) { 2195 rv = uvm_fault_wire(map, entry->start, entry->end, 2196 VM_FAULT_WIREMAX, entry->max_protection); 2197 if (rv) { 2198 2199 /* 2200 * wiring failed. break out of the loop. 2201 * we'll clean up the map below, once we 2202 * have a write lock again. 2203 */ 2204 2205 break; 2206 } 2207 } 2208 entry = entry->next; 2209 } 2210 2211 if (rv) { /* failed? */ 2212 2213 /* 2214 * Get back to an exclusive (write) lock. 2215 */ 2216 2217 vm_map_upgrade(map); 2218 vm_map_unbusy(map); 2219 2220 #ifdef DIAGNOSTIC 2221 if (timestamp_save != map->timestamp) 2222 panic("uvm_map_pageable: stale map"); 2223 #endif 2224 2225 /* 2226 * first drop the wiring count on all the entries 2227 * which haven't actually been wired yet. 2228 */ 2229 2230 failed_entry = entry; 2231 while (entry != &map->header && entry->start < end) { 2232 entry->wired_count--; 2233 entry = entry->next; 2234 } 2235 2236 /* 2237 * now, unwire all the entries that were successfully 2238 * wired above. 2239 */ 2240 2241 entry = start_entry; 2242 while (entry != failed_entry) { 2243 entry->wired_count--; 2244 if (VM_MAPENT_ISWIRED(entry) == 0) 2245 uvm_map_entry_unwire(map, entry); 2246 entry = entry->next; 2247 } 2248 if ((lockflags & UVM_LK_EXIT) == 0) 2249 vm_map_unlock(map); 2250 UVMHIST_LOG(maphist, "<- done (RV=%d)", rv,0,0,0); 2251 return(rv); 2252 } 2253 2254 /* We are holding a read lock here. */ 2255 if ((lockflags & UVM_LK_EXIT) == 0) { 2256 vm_map_unbusy(map); 2257 vm_map_unlock_read(map); 2258 } else { 2259 2260 /* 2261 * Get back to an exclusive (write) lock. 2262 */ 2263 2264 vm_map_upgrade(map); 2265 vm_map_unbusy(map); 2266 } 2267 2268 UVMHIST_LOG(maphist,"<- done (OK WIRE)",0,0,0,0); 2269 return 0; 2270 } 2271 2272 /* 2273 * uvm_map_pageable_all: special case of uvm_map_pageable - affects 2274 * all mapped regions. 2275 * 2276 * => map must not be locked. 2277 * => if no flags are specified, all regions are unwired. 2278 * => XXXJRT: has some of the same problems as uvm_map_pageable() above. 2279 */ 2280 2281 int 2282 uvm_map_pageable_all(map, flags, limit) 2283 struct vm_map *map; 2284 int flags; 2285 vsize_t limit; 2286 { 2287 struct vm_map_entry *entry, *failed_entry; 2288 vsize_t size; 2289 int rv; 2290 #ifdef DIAGNOSTIC 2291 u_int timestamp_save; 2292 #endif 2293 UVMHIST_FUNC("uvm_map_pageable_all"); UVMHIST_CALLED(maphist); 2294 UVMHIST_LOG(maphist,"(map=0x%x,flags=0x%x)", map, flags, 0, 0); 2295 2296 KASSERT(map->flags & VM_MAP_PAGEABLE); 2297 2298 vm_map_lock(map); 2299 2300 /* 2301 * handle wiring and unwiring separately. 2302 */ 2303 2304 if (flags == 0) { /* unwire */ 2305 2306 /* 2307 * POSIX 1003.1b -- munlockall unlocks all regions, 2308 * regardless of how many times mlockall has been called. 2309 */ 2310 2311 for (entry = map->header.next; entry != &map->header; 2312 entry = entry->next) { 2313 if (VM_MAPENT_ISWIRED(entry)) 2314 uvm_map_entry_unwire(map, entry); 2315 } 2316 vm_map_modflags(map, 0, VM_MAP_WIREFUTURE); 2317 vm_map_unlock(map); 2318 UVMHIST_LOG(maphist,"<- done (OK UNWIRE)",0,0,0,0); 2319 return 0; 2320 } 2321 2322 if (flags & MCL_FUTURE) { 2323 2324 /* 2325 * must wire all future mappings; remember this. 2326 */ 2327 2328 vm_map_modflags(map, VM_MAP_WIREFUTURE, 0); 2329 } 2330 2331 if ((flags & MCL_CURRENT) == 0) { 2332 2333 /* 2334 * no more work to do! 2335 */ 2336 2337 UVMHIST_LOG(maphist,"<- done (OK no wire)",0,0,0,0); 2338 vm_map_unlock(map); 2339 return 0; 2340 } 2341 2342 /* 2343 * wire case: in three passes [XXXCDC: ugly block of code here] 2344 * 2345 * 1: holding the write lock, count all pages mapped by non-wired 2346 * entries. if this would cause us to go over our limit, we fail. 2347 * 2348 * 2: still holding the write lock, we create any anonymous maps that 2349 * need to be created. then we increment its wiring count. 2350 * 2351 * 3: we downgrade to a read lock, and call uvm_fault_wire to fault 2352 * in the pages for any newly wired area (wired_count == 1). 2353 * 2354 * downgrading to a read lock for uvm_fault_wire avoids a possible 2355 * deadlock with another thread that may have faulted on one of 2356 * the pages to be wired (it would mark the page busy, blocking 2357 * us, then in turn block on the map lock that we hold). because 2358 * of problems in the recursive lock package, we cannot upgrade 2359 * to a write lock in vm_map_lookup. thus, any actions that 2360 * require the write lock must be done beforehand. because we 2361 * keep the read lock on the map, the copy-on-write status of the 2362 * entries we modify here cannot change. 2363 */ 2364 2365 for (size = 0, entry = map->header.next; entry != &map->header; 2366 entry = entry->next) { 2367 if (entry->protection != VM_PROT_NONE && 2368 VM_MAPENT_ISWIRED(entry) == 0) { /* not already wired? */ 2369 size += entry->end - entry->start; 2370 } 2371 } 2372 2373 if (atop(size) + uvmexp.wired > uvmexp.wiredmax) { 2374 vm_map_unlock(map); 2375 return ENOMEM; 2376 } 2377 2378 /* XXX non-pmap_wired_count case must be handled by caller */ 2379 #ifdef pmap_wired_count 2380 if (limit != 0 && 2381 (size + ptoa(pmap_wired_count(vm_map_pmap(map))) > limit)) { 2382 vm_map_unlock(map); 2383 return ENOMEM; 2384 } 2385 #endif 2386 2387 /* 2388 * Pass 2. 2389 */ 2390 2391 for (entry = map->header.next; entry != &map->header; 2392 entry = entry->next) { 2393 if (entry->protection == VM_PROT_NONE) 2394 continue; 2395 if (VM_MAPENT_ISWIRED(entry) == 0) { /* not already wired? */ 2396 2397 /* 2398 * perform actions of vm_map_lookup that need the 2399 * write lock on the map: create an anonymous map 2400 * for a copy-on-write region, or an anonymous map 2401 * for a zero-fill region. (XXXCDC: submap case 2402 * ok?) 2403 */ 2404 2405 if (!UVM_ET_ISSUBMAP(entry)) { /* not submap */ 2406 if (UVM_ET_ISNEEDSCOPY(entry) && 2407 ((entry->max_protection & VM_PROT_WRITE) || 2408 (entry->object.uvm_obj == NULL))) { 2409 amap_copy(map, entry, M_WAITOK, TRUE, 2410 entry->start, entry->end); 2411 /* XXXCDC: wait OK? */ 2412 } 2413 } 2414 } 2415 entry->wired_count++; 2416 } 2417 2418 /* 2419 * Pass 3. 2420 */ 2421 2422 #ifdef DIAGNOSTIC 2423 timestamp_save = map->timestamp; 2424 #endif 2425 vm_map_busy(map); 2426 vm_map_downgrade(map); 2427 2428 rv = 0; 2429 for (entry = map->header.next; entry != &map->header; 2430 entry = entry->next) { 2431 if (entry->wired_count == 1) { 2432 rv = uvm_fault_wire(map, entry->start, entry->end, 2433 VM_FAULT_WIREMAX, entry->max_protection); 2434 if (rv) { 2435 2436 /* 2437 * wiring failed. break out of the loop. 2438 * we'll clean up the map below, once we 2439 * have a write lock again. 2440 */ 2441 2442 break; 2443 } 2444 } 2445 } 2446 2447 if (rv) { 2448 2449 /* 2450 * Get back an exclusive (write) lock. 2451 */ 2452 2453 vm_map_upgrade(map); 2454 vm_map_unbusy(map); 2455 2456 #ifdef DIAGNOSTIC 2457 if (timestamp_save != map->timestamp) 2458 panic("uvm_map_pageable_all: stale map"); 2459 #endif 2460 2461 /* 2462 * first drop the wiring count on all the entries 2463 * which haven't actually been wired yet. 2464 * 2465 * Skip VM_PROT_NONE entries like we did above. 2466 */ 2467 2468 failed_entry = entry; 2469 for (/* nothing */; entry != &map->header; 2470 entry = entry->next) { 2471 if (entry->protection == VM_PROT_NONE) 2472 continue; 2473 entry->wired_count--; 2474 } 2475 2476 /* 2477 * now, unwire all the entries that were successfully 2478 * wired above. 2479 * 2480 * Skip VM_PROT_NONE entries like we did above. 2481 */ 2482 2483 for (entry = map->header.next; entry != failed_entry; 2484 entry = entry->next) { 2485 if (entry->protection == VM_PROT_NONE) 2486 continue; 2487 entry->wired_count--; 2488 if (VM_MAPENT_ISWIRED(entry)) 2489 uvm_map_entry_unwire(map, entry); 2490 } 2491 vm_map_unlock(map); 2492 UVMHIST_LOG(maphist,"<- done (RV=%d)", rv,0,0,0); 2493 return (rv); 2494 } 2495 2496 /* We are holding a read lock here. */ 2497 vm_map_unbusy(map); 2498 vm_map_unlock_read(map); 2499 2500 UVMHIST_LOG(maphist,"<- done (OK WIRE)",0,0,0,0); 2501 return 0; 2502 } 2503 2504 /* 2505 * uvm_map_clean: clean out a map range 2506 * 2507 * => valid flags: 2508 * if (flags & PGO_CLEANIT): dirty pages are cleaned first 2509 * if (flags & PGO_SYNCIO): dirty pages are written synchronously 2510 * if (flags & PGO_DEACTIVATE): any cached pages are deactivated after clean 2511 * if (flags & PGO_FREE): any cached pages are freed after clean 2512 * => returns an error if any part of the specified range isn't mapped 2513 * => never a need to flush amap layer since the anonymous memory has 2514 * no permanent home, but may deactivate pages there 2515 * => called from sys_msync() and sys_madvise() 2516 * => caller must not write-lock map (read OK). 2517 * => we may sleep while cleaning if SYNCIO [with map read-locked] 2518 */ 2519 2520 int 2521 uvm_map_clean(map, start, end, flags) 2522 struct vm_map *map; 2523 vaddr_t start, end; 2524 int flags; 2525 { 2526 struct vm_map_entry *current, *entry; 2527 struct uvm_object *uobj; 2528 struct vm_amap *amap; 2529 struct vm_anon *anon; 2530 struct vm_page *pg; 2531 vaddr_t offset; 2532 vsize_t size; 2533 int error, refs; 2534 UVMHIST_FUNC("uvm_map_clean"); UVMHIST_CALLED(maphist); 2535 2536 UVMHIST_LOG(maphist,"(map=0x%x,start=0x%x,end=0x%x,flags=0x%x)", 2537 map, start, end, flags); 2538 KASSERT((flags & (PGO_FREE|PGO_DEACTIVATE)) != 2539 (PGO_FREE|PGO_DEACTIVATE)); 2540 2541 vm_map_lock_read(map); 2542 VM_MAP_RANGE_CHECK(map, start, end); 2543 if (uvm_map_lookup_entry(map, start, &entry) == FALSE) { 2544 vm_map_unlock_read(map); 2545 return EFAULT; 2546 } 2547 2548 /* 2549 * Make a first pass to check for holes. 2550 */ 2551 2552 for (current = entry; current->start < end; current = current->next) { 2553 if (UVM_ET_ISSUBMAP(current)) { 2554 vm_map_unlock_read(map); 2555 return EINVAL; 2556 } 2557 if (end <= current->end) { 2558 break; 2559 } 2560 if (current->end != current->next->start) { 2561 vm_map_unlock_read(map); 2562 return EFAULT; 2563 } 2564 } 2565 2566 error = 0; 2567 for (current = entry; start < end; current = current->next) { 2568 amap = current->aref.ar_amap; /* top layer */ 2569 uobj = current->object.uvm_obj; /* bottom layer */ 2570 KASSERT(start >= current->start); 2571 2572 /* 2573 * No amap cleaning necessary if: 2574 * 2575 * (1) There's no amap. 2576 * 2577 * (2) We're not deactivating or freeing pages. 2578 */ 2579 2580 if (amap == NULL || (flags & (PGO_DEACTIVATE|PGO_FREE)) == 0) 2581 goto flush_object; 2582 2583 amap_lock(amap); 2584 offset = start - current->start; 2585 size = MIN(end, current->end) - start; 2586 for ( ; size != 0; size -= PAGE_SIZE, offset += PAGE_SIZE) { 2587 anon = amap_lookup(¤t->aref, offset); 2588 if (anon == NULL) 2589 continue; 2590 2591 simple_lock(&anon->an_lock); 2592 pg = anon->u.an_page; 2593 if (pg == NULL) { 2594 simple_unlock(&anon->an_lock); 2595 continue; 2596 } 2597 2598 switch (flags & (PGO_CLEANIT|PGO_FREE|PGO_DEACTIVATE)) { 2599 2600 /* 2601 * In these first 3 cases, we just deactivate the page. 2602 */ 2603 2604 case PGO_CLEANIT|PGO_FREE: 2605 case PGO_CLEANIT|PGO_DEACTIVATE: 2606 case PGO_DEACTIVATE: 2607 deactivate_it: 2608 /* 2609 * skip the page if it's loaned or wired, 2610 * since it shouldn't be on a paging queue 2611 * at all in these cases. 2612 */ 2613 2614 uvm_lock_pageq(); 2615 if (pg->loan_count != 0 || 2616 pg->wire_count != 0) { 2617 uvm_unlock_pageq(); 2618 simple_unlock(&anon->an_lock); 2619 continue; 2620 } 2621 KASSERT(pg->uanon == anon); 2622 pmap_clear_reference(pg); 2623 uvm_pagedeactivate(pg); 2624 uvm_unlock_pageq(); 2625 simple_unlock(&anon->an_lock); 2626 continue; 2627 2628 case PGO_FREE: 2629 2630 /* 2631 * If there are multiple references to 2632 * the amap, just deactivate the page. 2633 */ 2634 2635 if (amap_refs(amap) > 1) 2636 goto deactivate_it; 2637 2638 /* skip the page if it's wired */ 2639 if (pg->wire_count != 0) { 2640 simple_unlock(&anon->an_lock); 2641 continue; 2642 } 2643 amap_unadd(¤t->aref, offset); 2644 refs = --anon->an_ref; 2645 simple_unlock(&anon->an_lock); 2646 if (refs == 0) 2647 uvm_anfree(anon); 2648 continue; 2649 } 2650 } 2651 amap_unlock(amap); 2652 2653 flush_object: 2654 /* 2655 * flush pages if we've got a valid backing object. 2656 * note that we must always clean object pages before 2657 * freeing them since otherwise we could reveal stale 2658 * data from files. 2659 */ 2660 2661 offset = current->offset + (start - current->start); 2662 size = MIN(end, current->end) - start; 2663 if (uobj != NULL) { 2664 simple_lock(&uobj->vmobjlock); 2665 error = (uobj->pgops->pgo_put)(uobj, offset, 2666 offset + size, flags | PGO_CLEANIT); 2667 } 2668 start += size; 2669 } 2670 vm_map_unlock_read(map); 2671 return (error); 2672 } 2673 2674 2675 /* 2676 * uvm_map_checkprot: check protection in map 2677 * 2678 * => must allow specified protection in a fully allocated region. 2679 * => map must be read or write locked by caller. 2680 */ 2681 2682 boolean_t 2683 uvm_map_checkprot(map, start, end, protection) 2684 struct vm_map * map; 2685 vaddr_t start, end; 2686 vm_prot_t protection; 2687 { 2688 struct vm_map_entry *entry; 2689 struct vm_map_entry *tmp_entry; 2690 2691 if (!uvm_map_lookup_entry(map, start, &tmp_entry)) { 2692 return(FALSE); 2693 } 2694 entry = tmp_entry; 2695 while (start < end) { 2696 if (entry == &map->header) { 2697 return(FALSE); 2698 } 2699 2700 /* 2701 * no holes allowed 2702 */ 2703 2704 if (start < entry->start) { 2705 return(FALSE); 2706 } 2707 2708 /* 2709 * check protection associated with entry 2710 */ 2711 2712 if ((entry->protection & protection) != protection) { 2713 return(FALSE); 2714 } 2715 start = entry->end; 2716 entry = entry->next; 2717 } 2718 return(TRUE); 2719 } 2720 2721 /* 2722 * uvmspace_alloc: allocate a vmspace structure. 2723 * 2724 * - structure includes vm_map and pmap 2725 * - XXX: no locking on this structure 2726 * - refcnt set to 1, rest must be init'd by caller 2727 */ 2728 struct vmspace * 2729 uvmspace_alloc(min, max) 2730 vaddr_t min, max; 2731 { 2732 struct vmspace *vm; 2733 UVMHIST_FUNC("uvmspace_alloc"); UVMHIST_CALLED(maphist); 2734 2735 vm = pool_get(&uvm_vmspace_pool, PR_WAITOK); 2736 uvmspace_init(vm, NULL, min, max); 2737 UVMHIST_LOG(maphist,"<- done (vm=0x%x)", vm,0,0,0); 2738 return (vm); 2739 } 2740 2741 /* 2742 * uvmspace_init: initialize a vmspace structure. 2743 * 2744 * - XXX: no locking on this structure 2745 * - refcnt set to 1, rest must me init'd by caller 2746 */ 2747 void 2748 uvmspace_init(vm, pmap, min, max) 2749 struct vmspace *vm; 2750 struct pmap *pmap; 2751 vaddr_t min, max; 2752 { 2753 UVMHIST_FUNC("uvmspace_init"); UVMHIST_CALLED(maphist); 2754 2755 memset(vm, 0, sizeof(*vm)); 2756 uvm_map_setup(&vm->vm_map, min, max, VM_MAP_PAGEABLE); 2757 if (pmap) 2758 pmap_reference(pmap); 2759 else 2760 pmap = pmap_create(); 2761 vm->vm_map.pmap = pmap; 2762 vm->vm_refcnt = 1; 2763 UVMHIST_LOG(maphist,"<- done",0,0,0,0); 2764 } 2765 2766 /* 2767 * uvmspace_share: share a vmspace between two proceses 2768 * 2769 * - XXX: no locking on vmspace 2770 * - used for vfork, threads(?) 2771 */ 2772 2773 void 2774 uvmspace_share(p1, p2) 2775 struct proc *p1, *p2; 2776 { 2777 p2->p_vmspace = p1->p_vmspace; 2778 p1->p_vmspace->vm_refcnt++; 2779 } 2780 2781 /* 2782 * uvmspace_unshare: ensure that process "p" has its own, unshared, vmspace 2783 * 2784 * - XXX: no locking on vmspace 2785 */ 2786 2787 void 2788 uvmspace_unshare(p) 2789 struct proc *p; 2790 { 2791 struct vmspace *nvm, *ovm = p->p_vmspace; 2792 2793 if (ovm->vm_refcnt == 1) 2794 /* nothing to do: vmspace isn't shared in the first place */ 2795 return; 2796 2797 /* make a new vmspace, still holding old one */ 2798 nvm = uvmspace_fork(ovm); 2799 2800 pmap_deactivate(p); /* unbind old vmspace */ 2801 p->p_vmspace = nvm; 2802 pmap_activate(p); /* switch to new vmspace */ 2803 2804 uvmspace_free(ovm); /* drop reference to old vmspace */ 2805 } 2806 2807 /* 2808 * uvmspace_exec: the process wants to exec a new program 2809 * 2810 * - XXX: no locking on vmspace 2811 */ 2812 2813 void 2814 uvmspace_exec(p, start, end) 2815 struct proc *p; 2816 vaddr_t start, end; 2817 { 2818 struct vmspace *nvm, *ovm = p->p_vmspace; 2819 struct vm_map *map = &ovm->vm_map; 2820 2821 #ifdef __sparc__ 2822 /* XXX cgd 960926: the sparc #ifdef should be a MD hook */ 2823 kill_user_windows(p); /* before stack addresses go away */ 2824 #endif 2825 2826 /* 2827 * see if more than one process is using this vmspace... 2828 */ 2829 2830 if (ovm->vm_refcnt == 1) { 2831 2832 /* 2833 * if p is the only process using its vmspace then we can safely 2834 * recycle that vmspace for the program that is being exec'd. 2835 */ 2836 2837 #ifdef SYSVSHM 2838 /* 2839 * SYSV SHM semantics require us to kill all segments on an exec 2840 */ 2841 2842 if (ovm->vm_shm) 2843 shmexit(ovm); 2844 #endif 2845 2846 /* 2847 * POSIX 1003.1b -- "lock future mappings" is revoked 2848 * when a process execs another program image. 2849 */ 2850 2851 vm_map_lock(map); 2852 vm_map_modflags(map, 0, VM_MAP_WIREFUTURE); 2853 vm_map_unlock(map); 2854 2855 /* 2856 * now unmap the old program 2857 */ 2858 2859 uvm_unmap(map, map->min_offset, map->max_offset); 2860 2861 /* 2862 * resize the map 2863 */ 2864 2865 vm_map_lock(map); 2866 map->min_offset = start; 2867 map->max_offset = end; 2868 vm_map_unlock(map); 2869 } else { 2870 2871 /* 2872 * p's vmspace is being shared, so we can't reuse it for p since 2873 * it is still being used for others. allocate a new vmspace 2874 * for p 2875 */ 2876 2877 nvm = uvmspace_alloc(start, end); 2878 2879 /* 2880 * install new vmspace and drop our ref to the old one. 2881 */ 2882 2883 pmap_deactivate(p); 2884 p->p_vmspace = nvm; 2885 pmap_activate(p); 2886 2887 uvmspace_free(ovm); 2888 } 2889 } 2890 2891 /* 2892 * uvmspace_free: free a vmspace data structure 2893 * 2894 * - XXX: no locking on vmspace 2895 */ 2896 2897 void 2898 uvmspace_free(vm) 2899 struct vmspace *vm; 2900 { 2901 struct vm_map_entry *dead_entries; 2902 UVMHIST_FUNC("uvmspace_free"); UVMHIST_CALLED(maphist); 2903 2904 UVMHIST_LOG(maphist,"(vm=0x%x) ref=%d", vm, vm->vm_refcnt,0,0); 2905 if (--vm->vm_refcnt == 0) { 2906 2907 /* 2908 * lock the map, to wait out all other references to it. delete 2909 * all of the mappings and pages they hold, then call the pmap 2910 * module to reclaim anything left. 2911 */ 2912 2913 #ifdef SYSVSHM 2914 /* Get rid of any SYSV shared memory segments. */ 2915 if (vm->vm_shm != NULL) 2916 shmexit(vm); 2917 #endif 2918 vm_map_lock(&vm->vm_map); 2919 if (vm->vm_map.nentries) { 2920 uvm_unmap_remove(&vm->vm_map, 2921 vm->vm_map.min_offset, vm->vm_map.max_offset, 2922 &dead_entries); 2923 if (dead_entries != NULL) 2924 uvm_unmap_detach(dead_entries, 0); 2925 } 2926 pmap_destroy(vm->vm_map.pmap); 2927 vm->vm_map.pmap = NULL; 2928 pool_put(&uvm_vmspace_pool, vm); 2929 } 2930 UVMHIST_LOG(maphist,"<- done", 0,0,0,0); 2931 } 2932 2933 /* 2934 * F O R K - m a i n e n t r y p o i n t 2935 */ 2936 /* 2937 * uvmspace_fork: fork a process' main map 2938 * 2939 * => create a new vmspace for child process from parent. 2940 * => parent's map must not be locked. 2941 */ 2942 2943 struct vmspace * 2944 uvmspace_fork(vm1) 2945 struct vmspace *vm1; 2946 { 2947 struct vmspace *vm2; 2948 struct vm_map *old_map = &vm1->vm_map; 2949 struct vm_map *new_map; 2950 struct vm_map_entry *old_entry; 2951 struct vm_map_entry *new_entry; 2952 pmap_t new_pmap; 2953 UVMHIST_FUNC("uvmspace_fork"); UVMHIST_CALLED(maphist); 2954 2955 vm_map_lock(old_map); 2956 2957 vm2 = uvmspace_alloc(old_map->min_offset, old_map->max_offset); 2958 memcpy(&vm2->vm_startcopy, &vm1->vm_startcopy, 2959 (caddr_t) (vm1 + 1) - (caddr_t) &vm1->vm_startcopy); 2960 new_map = &vm2->vm_map; /* XXX */ 2961 new_pmap = new_map->pmap; 2962 2963 old_entry = old_map->header.next; 2964 2965 /* 2966 * go entry-by-entry 2967 */ 2968 2969 while (old_entry != &old_map->header) { 2970 2971 /* 2972 * first, some sanity checks on the old entry 2973 */ 2974 2975 KASSERT(!UVM_ET_ISSUBMAP(old_entry)); 2976 KASSERT(UVM_ET_ISCOPYONWRITE(old_entry) || 2977 !UVM_ET_ISNEEDSCOPY(old_entry)); 2978 2979 switch (old_entry->inheritance) { 2980 case MAP_INHERIT_NONE: 2981 2982 /* 2983 * drop the mapping 2984 */ 2985 2986 break; 2987 2988 case MAP_INHERIT_SHARE: 2989 2990 /* 2991 * share the mapping: this means we want the old and 2992 * new entries to share amaps and backing objects. 2993 */ 2994 /* 2995 * if the old_entry needs a new amap (due to prev fork) 2996 * then we need to allocate it now so that we have 2997 * something we own to share with the new_entry. [in 2998 * other words, we need to clear needs_copy] 2999 */ 3000 3001 if (UVM_ET_ISNEEDSCOPY(old_entry)) { 3002 /* get our own amap, clears needs_copy */ 3003 amap_copy(old_map, old_entry, M_WAITOK, FALSE, 3004 0, 0); 3005 /* XXXCDC: WAITOK??? */ 3006 } 3007 3008 new_entry = uvm_mapent_alloc(new_map); 3009 /* old_entry -> new_entry */ 3010 uvm_mapent_copy(old_entry, new_entry); 3011 3012 /* new pmap has nothing wired in it */ 3013 new_entry->wired_count = 0; 3014 3015 /* 3016 * gain reference to object backing the map (can't 3017 * be a submap, already checked this case). 3018 */ 3019 3020 if (new_entry->aref.ar_amap) 3021 uvm_map_reference_amap(new_entry, AMAP_SHARED); 3022 3023 if (new_entry->object.uvm_obj && 3024 new_entry->object.uvm_obj->pgops->pgo_reference) 3025 new_entry->object.uvm_obj-> 3026 pgops->pgo_reference( 3027 new_entry->object.uvm_obj); 3028 3029 /* insert entry at end of new_map's entry list */ 3030 uvm_map_entry_link(new_map, new_map->header.prev, 3031 new_entry); 3032 3033 break; 3034 3035 case MAP_INHERIT_COPY: 3036 3037 /* 3038 * copy-on-write the mapping (using mmap's 3039 * MAP_PRIVATE semantics) 3040 * 3041 * allocate new_entry, adjust reference counts. 3042 * (note that new references are read-only). 3043 */ 3044 3045 new_entry = uvm_mapent_alloc(new_map); 3046 /* old_entry -> new_entry */ 3047 uvm_mapent_copy(old_entry, new_entry); 3048 3049 if (new_entry->aref.ar_amap) 3050 uvm_map_reference_amap(new_entry, 0); 3051 3052 if (new_entry->object.uvm_obj && 3053 new_entry->object.uvm_obj->pgops->pgo_reference) 3054 new_entry->object.uvm_obj->pgops->pgo_reference 3055 (new_entry->object.uvm_obj); 3056 3057 /* new pmap has nothing wired in it */ 3058 new_entry->wired_count = 0; 3059 3060 new_entry->etype |= 3061 (UVM_ET_COPYONWRITE|UVM_ET_NEEDSCOPY); 3062 uvm_map_entry_link(new_map, new_map->header.prev, 3063 new_entry); 3064 3065 /* 3066 * the new entry will need an amap. it will either 3067 * need to be copied from the old entry or created 3068 * from scratch (if the old entry does not have an 3069 * amap). can we defer this process until later 3070 * (by setting "needs_copy") or do we need to copy 3071 * the amap now? 3072 * 3073 * we must copy the amap now if any of the following 3074 * conditions hold: 3075 * 1. the old entry has an amap and that amap is 3076 * being shared. this means that the old (parent) 3077 * process is sharing the amap with another 3078 * process. if we do not clear needs_copy here 3079 * we will end up in a situation where both the 3080 * parent and child process are refering to the 3081 * same amap with "needs_copy" set. if the 3082 * parent write-faults, the fault routine will 3083 * clear "needs_copy" in the parent by allocating 3084 * a new amap. this is wrong because the 3085 * parent is supposed to be sharing the old amap 3086 * and the new amap will break that. 3087 * 3088 * 2. if the old entry has an amap and a non-zero 3089 * wire count then we are going to have to call 3090 * amap_cow_now to avoid page faults in the 3091 * parent process. since amap_cow_now requires 3092 * "needs_copy" to be clear we might as well 3093 * clear it here as well. 3094 * 3095 */ 3096 3097 if (old_entry->aref.ar_amap != NULL) { 3098 if ((amap_flags(old_entry->aref.ar_amap) & 3099 AMAP_SHARED) != 0 || 3100 VM_MAPENT_ISWIRED(old_entry)) { 3101 3102 amap_copy(new_map, new_entry, M_WAITOK, 3103 FALSE, 0, 0); 3104 /* XXXCDC: M_WAITOK ... ok? */ 3105 } 3106 } 3107 3108 /* 3109 * if the parent's entry is wired down, then the 3110 * parent process does not want page faults on 3111 * access to that memory. this means that we 3112 * cannot do copy-on-write because we can't write 3113 * protect the old entry. in this case we 3114 * resolve all copy-on-write faults now, using 3115 * amap_cow_now. note that we have already 3116 * allocated any needed amap (above). 3117 */ 3118 3119 if (VM_MAPENT_ISWIRED(old_entry)) { 3120 3121 /* 3122 * resolve all copy-on-write faults now 3123 * (note that there is nothing to do if 3124 * the old mapping does not have an amap). 3125 */ 3126 if (old_entry->aref.ar_amap) 3127 amap_cow_now(new_map, new_entry); 3128 3129 } else { 3130 3131 /* 3132 * setup mappings to trigger copy-on-write faults 3133 * we must write-protect the parent if it has 3134 * an amap and it is not already "needs_copy"... 3135 * if it is already "needs_copy" then the parent 3136 * has already been write-protected by a previous 3137 * fork operation. 3138 */ 3139 3140 if (old_entry->aref.ar_amap && 3141 !UVM_ET_ISNEEDSCOPY(old_entry)) { 3142 if (old_entry->max_protection & VM_PROT_WRITE) { 3143 pmap_protect(old_map->pmap, 3144 old_entry->start, 3145 old_entry->end, 3146 old_entry->protection & 3147 ~VM_PROT_WRITE); 3148 pmap_update(old_map->pmap); 3149 } 3150 old_entry->etype |= UVM_ET_NEEDSCOPY; 3151 } 3152 } 3153 break; 3154 } /* end of switch statement */ 3155 old_entry = old_entry->next; 3156 } 3157 3158 new_map->size = old_map->size; 3159 vm_map_unlock(old_map); 3160 3161 #ifdef SYSVSHM 3162 if (vm1->vm_shm) 3163 shmfork(vm1, vm2); 3164 #endif 3165 3166 #ifdef PMAP_FORK 3167 pmap_fork(vm1->vm_map.pmap, vm2->vm_map.pmap); 3168 #endif 3169 3170 UVMHIST_LOG(maphist,"<- done",0,0,0,0); 3171 return(vm2); 3172 } 3173 3174 3175 #if defined(DDB) 3176 3177 /* 3178 * DDB hooks 3179 */ 3180 3181 /* 3182 * uvm_map_printit: actually prints the map 3183 */ 3184 3185 void 3186 uvm_map_printit(map, full, pr) 3187 struct vm_map *map; 3188 boolean_t full; 3189 void (*pr) __P((const char *, ...)); 3190 { 3191 struct vm_map_entry *entry; 3192 3193 (*pr)("MAP %p: [0x%lx->0x%lx]\n", map, map->min_offset,map->max_offset); 3194 (*pr)("\t#ent=%d, sz=%d, ref=%d, version=%d, flags=0x%x\n", 3195 map->nentries, map->size, map->ref_count, map->timestamp, 3196 map->flags); 3197 (*pr)("\tpmap=%p(resident=%d)\n", map->pmap, 3198 pmap_resident_count(map->pmap)); 3199 if (!full) 3200 return; 3201 for (entry = map->header.next; entry != &map->header; 3202 entry = entry->next) { 3203 (*pr)(" - %p: 0x%lx->0x%lx: obj=%p/0x%llx, amap=%p/%d\n", 3204 entry, entry->start, entry->end, entry->object.uvm_obj, 3205 (long long)entry->offset, entry->aref.ar_amap, 3206 entry->aref.ar_pageoff); 3207 (*pr)( 3208 "\tsubmap=%c, cow=%c, nc=%c, prot(max)=%d/%d, inh=%d, " 3209 "wc=%d, adv=%d\n", 3210 (entry->etype & UVM_ET_SUBMAP) ? 'T' : 'F', 3211 (entry->etype & UVM_ET_COPYONWRITE) ? 'T' : 'F', 3212 (entry->etype & UVM_ET_NEEDSCOPY) ? 'T' : 'F', 3213 entry->protection, entry->max_protection, 3214 entry->inheritance, entry->wired_count, entry->advice); 3215 } 3216 } 3217 3218 /* 3219 * uvm_object_printit: actually prints the object 3220 */ 3221 3222 void 3223 uvm_object_printit(uobj, full, pr) 3224 struct uvm_object *uobj; 3225 boolean_t full; 3226 void (*pr) __P((const char *, ...)); 3227 { 3228 struct vm_page *pg; 3229 int cnt = 0; 3230 3231 (*pr)("OBJECT %p: locked=%d, pgops=%p, npages=%d, ", 3232 uobj, uobj->vmobjlock.lock_data, uobj->pgops, uobj->uo_npages); 3233 if (UVM_OBJ_IS_KERN_OBJECT(uobj)) 3234 (*pr)("refs=<SYSTEM>\n"); 3235 else 3236 (*pr)("refs=%d\n", uobj->uo_refs); 3237 3238 if (!full) { 3239 return; 3240 } 3241 (*pr)(" PAGES <pg,offset>:\n "); 3242 TAILQ_FOREACH(pg, &uobj->memq, listq) { 3243 cnt++; 3244 (*pr)("<%p,0x%llx> ", pg, (long long)pg->offset); 3245 if ((cnt % 3) == 0) { 3246 (*pr)("\n "); 3247 } 3248 } 3249 if ((cnt % 3) != 0) { 3250 (*pr)("\n"); 3251 } 3252 } 3253 3254 /* 3255 * uvm_page_printit: actually print the page 3256 */ 3257 3258 static const char page_flagbits[] = 3259 "\20\1BUSY\2WANTED\3TABLED\4CLEAN\5PAGEOUT\6RELEASED\7FAKE\10RDONLY" 3260 "\11ZERO\15PAGER1"; 3261 static const char page_pqflagbits[] = 3262 "\20\1FREE\2INACTIVE\3ACTIVE\5ANON\6AOBJ"; 3263 3264 void 3265 uvm_page_printit(pg, full, pr) 3266 struct vm_page *pg; 3267 boolean_t full; 3268 void (*pr) __P((const char *, ...)); 3269 { 3270 struct vm_page *tpg; 3271 struct uvm_object *uobj; 3272 struct pglist *pgl; 3273 char pgbuf[128]; 3274 char pqbuf[128]; 3275 3276 (*pr)("PAGE %p:\n", pg); 3277 bitmask_snprintf(pg->flags, page_flagbits, pgbuf, sizeof(pgbuf)); 3278 bitmask_snprintf(pg->pqflags, page_pqflagbits, pqbuf, sizeof(pqbuf)); 3279 (*pr)(" flags=%s, pqflags=%s, wire_count=%d, pa=0x%lx\n", 3280 pgbuf, pqbuf, pg->wire_count, (long)pg->phys_addr); 3281 (*pr)(" uobject=%p, uanon=%p, offset=0x%llx loan_count=%d\n", 3282 pg->uobject, pg->uanon, (long long)pg->offset, pg->loan_count); 3283 #if defined(UVM_PAGE_TRKOWN) 3284 if (pg->flags & PG_BUSY) 3285 (*pr)(" owning process = %d, tag=%s\n", 3286 pg->owner, pg->owner_tag); 3287 else 3288 (*pr)(" page not busy, no owner\n"); 3289 #else 3290 (*pr)(" [page ownership tracking disabled]\n"); 3291 #endif 3292 3293 if (!full) 3294 return; 3295 3296 /* cross-verify object/anon */ 3297 if ((pg->pqflags & PQ_FREE) == 0) { 3298 if (pg->pqflags & PQ_ANON) { 3299 if (pg->uanon == NULL || pg->uanon->u.an_page != pg) 3300 (*pr)(" >>> ANON DOES NOT POINT HERE <<< (%p)\n", 3301 (pg->uanon) ? pg->uanon->u.an_page : NULL); 3302 else 3303 (*pr)(" anon backpointer is OK\n"); 3304 } else { 3305 uobj = pg->uobject; 3306 if (uobj) { 3307 (*pr)(" checking object list\n"); 3308 TAILQ_FOREACH(tpg, &uobj->memq, listq) { 3309 if (tpg == pg) { 3310 break; 3311 } 3312 } 3313 if (tpg) 3314 (*pr)(" page found on object list\n"); 3315 else 3316 (*pr)(" >>> PAGE NOT FOUND ON OBJECT LIST! <<<\n"); 3317 } 3318 } 3319 } 3320 3321 /* cross-verify page queue */ 3322 if (pg->pqflags & PQ_FREE) { 3323 int fl = uvm_page_lookup_freelist(pg); 3324 int color = VM_PGCOLOR_BUCKET(pg); 3325 pgl = &uvm.page_free[fl].pgfl_buckets[color].pgfl_queues[ 3326 ((pg)->flags & PG_ZERO) ? PGFL_ZEROS : PGFL_UNKNOWN]; 3327 } else if (pg->pqflags & PQ_INACTIVE) { 3328 pgl = &uvm.page_inactive; 3329 } else if (pg->pqflags & PQ_ACTIVE) { 3330 pgl = &uvm.page_active; 3331 } else { 3332 pgl = NULL; 3333 } 3334 3335 if (pgl) { 3336 (*pr)(" checking pageq list\n"); 3337 TAILQ_FOREACH(tpg, pgl, pageq) { 3338 if (tpg == pg) { 3339 break; 3340 } 3341 } 3342 if (tpg) 3343 (*pr)(" page found on pageq list\n"); 3344 else 3345 (*pr)(" >>> PAGE NOT FOUND ON PAGEQ LIST! <<<\n"); 3346 } 3347 } 3348 #endif 3349