1 /* $NetBSD: uvm_map.c,v 1.126 2002/11/30 18:28:06 bouyer 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 1.126 2002/11/30 18:28:06 bouyer 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 map_ubackmerge, map_uforwmerge; 104 struct uvm_cnt map_ubimerge, map_unomerge; 105 struct uvm_cnt map_kbackmerge, map_kforwmerge; 106 struct uvm_cnt map_kbimerge, map_knomerge; 107 struct uvm_cnt uvm_map_call, uvm_mlk_call, uvm_mlk_hint; 108 const char vmmapbsy[] = "vmmapbsy"; 109 110 /* 111 * pool for vmspace structures. 112 */ 113 114 struct pool uvm_vmspace_pool; 115 116 /* 117 * pool for dynamically-allocated map entries. 118 */ 119 120 struct pool uvm_map_entry_pool; 121 struct pool uvm_map_entry_kmem_pool; 122 123 #ifdef PMAP_GROWKERNEL 124 /* 125 * This global represents the end of the kernel virtual address 126 * space. If we want to exceed this, we must grow the kernel 127 * virtual address space dynamically. 128 * 129 * Note, this variable is locked by kernel_map's lock. 130 */ 131 vaddr_t uvm_maxkaddr; 132 #endif 133 134 /* 135 * macros 136 */ 137 138 /* 139 * uvm_map_entry_link: insert entry into a map 140 * 141 * => map must be locked 142 */ 143 #define uvm_map_entry_link(map, after_where, entry) do { \ 144 (map)->nentries++; \ 145 (entry)->prev = (after_where); \ 146 (entry)->next = (after_where)->next; \ 147 (entry)->prev->next = (entry); \ 148 (entry)->next->prev = (entry); \ 149 } while (/*CONSTCOND*/ 0) 150 151 /* 152 * uvm_map_entry_unlink: remove entry from a map 153 * 154 * => map must be locked 155 */ 156 #define uvm_map_entry_unlink(map, entry) do { \ 157 (map)->nentries--; \ 158 (entry)->next->prev = (entry)->prev; \ 159 (entry)->prev->next = (entry)->next; \ 160 } while (/*CONSTCOND*/ 0) 161 162 /* 163 * SAVE_HINT: saves the specified entry as the hint for future lookups. 164 * 165 * => map need not be locked (protected by hint_lock). 166 */ 167 #define SAVE_HINT(map,check,value) do { \ 168 simple_lock(&(map)->hint_lock); \ 169 if ((map)->hint == (check)) \ 170 (map)->hint = (value); \ 171 simple_unlock(&(map)->hint_lock); \ 172 } while (/*CONSTCOND*/ 0) 173 174 /* 175 * VM_MAP_RANGE_CHECK: check and correct range 176 * 177 * => map must at least be read locked 178 */ 179 180 #define VM_MAP_RANGE_CHECK(map, start, end) do { \ 181 if (start < vm_map_min(map)) \ 182 start = vm_map_min(map); \ 183 if (end > vm_map_max(map)) \ 184 end = vm_map_max(map); \ 185 if (start > end) \ 186 start = end; \ 187 } while (/*CONSTCOND*/ 0) 188 189 /* 190 * local prototypes 191 */ 192 193 static struct vm_map_entry *uvm_mapent_alloc __P((struct vm_map *, int)); 194 static void uvm_mapent_copy __P((struct vm_map_entry *, struct vm_map_entry *)); 195 static void uvm_mapent_free __P((struct vm_map_entry *)); 196 static void uvm_map_entry_unwire __P((struct vm_map *, struct vm_map_entry *)); 197 static void uvm_map_reference_amap __P((struct vm_map_entry *, int)); 198 static void uvm_map_unreference_amap __P((struct vm_map_entry *, int)); 199 200 /* 201 * local inlines 202 */ 203 204 /* 205 * uvm_mapent_alloc: allocate a map entry 206 */ 207 208 static __inline struct vm_map_entry * 209 uvm_mapent_alloc(map, flags) 210 struct vm_map *map; 211 int flags; 212 { 213 struct vm_map_entry *me; 214 int s; 215 int pflags = (flags & UVM_KMF_NOWAIT) ? PR_NOWAIT : PR_WAITOK; 216 UVMHIST_FUNC("uvm_mapent_alloc"); UVMHIST_CALLED(maphist); 217 218 if (map->flags & VM_MAP_INTRSAFE || cold) { 219 s = splvm(); 220 simple_lock(&uvm.kentry_lock); 221 me = uvm.kentry_free; 222 if (me) uvm.kentry_free = me->next; 223 simple_unlock(&uvm.kentry_lock); 224 splx(s); 225 if (__predict_false(me == NULL)) { 226 panic("uvm_mapent_alloc: out of static map entries, " 227 "check MAX_KMAPENT (currently %d)", 228 MAX_KMAPENT); 229 } 230 me->flags = UVM_MAP_STATIC; 231 } else if (map == kernel_map) { 232 me = pool_get(&uvm_map_entry_kmem_pool, pflags); 233 if (__predict_false(me == NULL)) 234 return NULL; 235 me->flags = UVM_MAP_KMEM; 236 } else { 237 me = pool_get(&uvm_map_entry_pool, pflags); 238 if (__predict_false(me == NULL)) 239 return NULL; 240 me->flags = 0; 241 } 242 243 UVMHIST_LOG(maphist, "<- new entry=0x%x [kentry=%d]", me, 244 ((map->flags & VM_MAP_INTRSAFE) != 0 || map == kernel_map), 0, 0); 245 return(me); 246 } 247 248 /* 249 * uvm_mapent_free: free map entry 250 */ 251 252 static __inline void 253 uvm_mapent_free(me) 254 struct vm_map_entry *me; 255 { 256 int s; 257 UVMHIST_FUNC("uvm_mapent_free"); UVMHIST_CALLED(maphist); 258 259 UVMHIST_LOG(maphist,"<- freeing map entry=0x%x [flags=%d]", 260 me, me->flags, 0, 0); 261 if (me->flags & UVM_MAP_STATIC) { 262 s = splvm(); 263 simple_lock(&uvm.kentry_lock); 264 me->next = uvm.kentry_free; 265 uvm.kentry_free = me; 266 simple_unlock(&uvm.kentry_lock); 267 splx(s); 268 } else if (me->flags & UVM_MAP_KMEM) { 269 pool_put(&uvm_map_entry_kmem_pool, me); 270 } else { 271 pool_put(&uvm_map_entry_pool, me); 272 } 273 } 274 275 /* 276 * uvm_mapent_copy: copy a map entry, preserving flags 277 */ 278 279 static __inline void 280 uvm_mapent_copy(src, dst) 281 struct vm_map_entry *src; 282 struct vm_map_entry *dst; 283 { 284 memcpy(dst, src, ((char *)&src->uvm_map_entry_stop_copy) - 285 ((char *)src)); 286 } 287 288 /* 289 * uvm_map_entry_unwire: unwire a map entry 290 * 291 * => map should be locked by caller 292 */ 293 294 static __inline void 295 uvm_map_entry_unwire(map, entry) 296 struct vm_map *map; 297 struct vm_map_entry *entry; 298 { 299 entry->wired_count = 0; 300 uvm_fault_unwire_locked(map, entry->start, entry->end); 301 } 302 303 304 /* 305 * wrapper for calling amap_ref() 306 */ 307 static __inline void 308 uvm_map_reference_amap(entry, flags) 309 struct vm_map_entry *entry; 310 int flags; 311 { 312 amap_ref(entry->aref.ar_amap, entry->aref.ar_pageoff, 313 (entry->end - entry->start) >> PAGE_SHIFT, flags); 314 } 315 316 317 /* 318 * wrapper for calling amap_unref() 319 */ 320 static __inline void 321 uvm_map_unreference_amap(entry, flags) 322 struct vm_map_entry *entry; 323 int flags; 324 { 325 amap_unref(entry->aref.ar_amap, entry->aref.ar_pageoff, 326 (entry->end - entry->start) >> PAGE_SHIFT, flags); 327 } 328 329 330 /* 331 * uvm_map_init: init mapping system at boot time. note that we allocate 332 * and init the static pool of struct vm_map_entry *'s for the kernel here. 333 */ 334 335 void 336 uvm_map_init() 337 { 338 static struct vm_map_entry kernel_map_entry[MAX_KMAPENT]; 339 #if defined(UVMHIST) 340 static struct uvm_history_ent maphistbuf[100]; 341 static struct uvm_history_ent pdhistbuf[100]; 342 #endif 343 int lcv; 344 345 /* 346 * first, init logging system. 347 */ 348 349 UVMHIST_FUNC("uvm_map_init"); 350 UVMHIST_INIT_STATIC(maphist, maphistbuf); 351 UVMHIST_INIT_STATIC(pdhist, pdhistbuf); 352 UVMHIST_CALLED(maphist); 353 UVMHIST_LOG(maphist,"<starting uvm map system>", 0, 0, 0, 0); 354 UVMCNT_INIT(uvm_map_call, UVMCNT_CNT, 0, 355 "# uvm_map() successful calls", 0); 356 357 UVMCNT_INIT(map_ubackmerge, UVMCNT_CNT, 0, 358 "# uvm_map() back umerges", 0); 359 UVMCNT_INIT(map_uforwmerge, UVMCNT_CNT, 0, 360 "# uvm_map() forward umerges", 0); 361 UVMCNT_INIT(map_ubimerge, UVMCNT_CNT, 0, 362 "# uvm_map() dual umerge", 0); 363 UVMCNT_INIT(map_unomerge, UVMCNT_CNT, 0, 364 "# uvm_map() no umerge", 0); 365 366 UVMCNT_INIT(map_kbackmerge, UVMCNT_CNT, 0, 367 "# uvm_map() back kmerges", 0); 368 UVMCNT_INIT(map_kforwmerge, UVMCNT_CNT, 0, 369 "# uvm_map() forward kmerges", 0); 370 UVMCNT_INIT(map_kbimerge, UVMCNT_CNT, 0, 371 "# uvm_map() dual kmerge", 0); 372 UVMCNT_INIT(map_knomerge, UVMCNT_CNT, 0, 373 "# uvm_map() no kmerge", 0); 374 375 UVMCNT_INIT(uvm_mlk_call, UVMCNT_CNT, 0, "# map lookup calls", 0); 376 UVMCNT_INIT(uvm_mlk_hint, UVMCNT_CNT, 0, "# map lookup hint hits", 0); 377 378 /* 379 * now set up static pool of kernel map entrys ... 380 */ 381 382 simple_lock_init(&uvm.kentry_lock); 383 uvm.kentry_free = NULL; 384 for (lcv = 0 ; lcv < MAX_KMAPENT ; lcv++) { 385 kernel_map_entry[lcv].next = uvm.kentry_free; 386 uvm.kentry_free = &kernel_map_entry[lcv]; 387 } 388 389 /* 390 * initialize the map-related pools. 391 */ 392 pool_init(&uvm_vmspace_pool, sizeof(struct vmspace), 393 0, 0, 0, "vmsppl", &pool_allocator_nointr); 394 pool_init(&uvm_map_entry_pool, sizeof(struct vm_map_entry), 395 0, 0, 0, "vmmpepl", &pool_allocator_nointr); 396 pool_init(&uvm_map_entry_kmem_pool, sizeof(struct vm_map_entry), 397 0, 0, 0, "vmmpekpl", NULL); 398 } 399 400 /* 401 * clippers 402 */ 403 404 /* 405 * uvm_map_clip_start: ensure that the entry begins at or after 406 * the starting address, if it doesn't we split the entry. 407 * 408 * => caller should use UVM_MAP_CLIP_START macro rather than calling 409 * this directly 410 * => map must be locked by caller 411 */ 412 413 void 414 uvm_map_clip_start(map, entry, start) 415 struct vm_map *map; 416 struct vm_map_entry *entry; 417 vaddr_t start; 418 { 419 struct vm_map_entry *new_entry; 420 vaddr_t new_adj; 421 422 /* uvm_map_simplify_entry(map, entry); */ /* XXX */ 423 424 /* 425 * Split off the front portion. note that we must insert the new 426 * entry BEFORE this one, so that this entry has the specified 427 * starting address. 428 */ 429 430 new_entry = uvm_mapent_alloc(map, 0); 431 uvm_mapent_copy(entry, new_entry); /* entry -> new_entry */ 432 433 new_entry->end = start; 434 new_adj = start - new_entry->start; 435 if (entry->object.uvm_obj) 436 entry->offset += new_adj; /* shift start over */ 437 entry->start = start; 438 439 if (new_entry->aref.ar_amap) { 440 amap_splitref(&new_entry->aref, &entry->aref, new_adj); 441 } 442 443 uvm_map_entry_link(map, entry->prev, new_entry); 444 445 if (UVM_ET_ISSUBMAP(entry)) { 446 /* ... unlikely to happen, but play it safe */ 447 uvm_map_reference(new_entry->object.sub_map); 448 } else { 449 if (UVM_ET_ISOBJ(entry) && 450 entry->object.uvm_obj->pgops && 451 entry->object.uvm_obj->pgops->pgo_reference) 452 entry->object.uvm_obj->pgops->pgo_reference( 453 entry->object.uvm_obj); 454 } 455 } 456 457 /* 458 * uvm_map_clip_end: ensure that the entry ends at or before 459 * the ending address, if it does't we split the reference 460 * 461 * => caller should use UVM_MAP_CLIP_END macro rather than calling 462 * this directly 463 * => map must be locked by caller 464 */ 465 466 void 467 uvm_map_clip_end(map, entry, end) 468 struct vm_map *map; 469 struct vm_map_entry *entry; 470 vaddr_t end; 471 { 472 struct vm_map_entry * new_entry; 473 vaddr_t new_adj; /* #bytes we move start forward */ 474 475 /* 476 * Create a new entry and insert it 477 * AFTER the specified entry 478 */ 479 480 new_entry = uvm_mapent_alloc(map, 0); 481 uvm_mapent_copy(entry, new_entry); /* entry -> new_entry */ 482 483 new_entry->start = entry->end = end; 484 new_adj = end - entry->start; 485 if (new_entry->object.uvm_obj) 486 new_entry->offset += new_adj; 487 488 if (entry->aref.ar_amap) 489 amap_splitref(&entry->aref, &new_entry->aref, new_adj); 490 491 uvm_map_entry_link(map, entry, new_entry); 492 493 if (UVM_ET_ISSUBMAP(entry)) { 494 /* ... unlikely to happen, but play it safe */ 495 uvm_map_reference(new_entry->object.sub_map); 496 } else { 497 if (UVM_ET_ISOBJ(entry) && 498 entry->object.uvm_obj->pgops && 499 entry->object.uvm_obj->pgops->pgo_reference) 500 entry->object.uvm_obj->pgops->pgo_reference( 501 entry->object.uvm_obj); 502 } 503 } 504 505 506 /* 507 * M A P - m a i n e n t r y p o i n t 508 */ 509 /* 510 * uvm_map: establish a valid mapping in a map 511 * 512 * => assume startp is page aligned. 513 * => assume size is a multiple of PAGE_SIZE. 514 * => assume sys_mmap provides enough of a "hint" to have us skip 515 * over text/data/bss area. 516 * => map must be unlocked (we will lock it) 517 * => <uobj,uoffset> value meanings (4 cases): 518 * [1] <NULL,uoffset> == uoffset is a hint for PMAP_PREFER 519 * [2] <NULL,UVM_UNKNOWN_OFFSET> == don't PMAP_PREFER 520 * [3] <uobj,uoffset> == normal mapping 521 * [4] <uobj,UVM_UNKNOWN_OFFSET> == uvm_map finds offset based on VA 522 * 523 * case [4] is for kernel mappings where we don't know the offset until 524 * we've found a virtual address. note that kernel object offsets are 525 * always relative to vm_map_min(kernel_map). 526 * 527 * => if `align' is non-zero, we try to align the virtual address to 528 * the specified alignment. this is only a hint; if we can't 529 * do it, the address will be unaligned. this is provided as 530 * a mechanism for large pages. 531 * 532 * => XXXCDC: need way to map in external amap? 533 */ 534 535 int 536 uvm_map(map, startp, size, uobj, uoffset, align, flags) 537 struct vm_map *map; 538 vaddr_t *startp; /* IN/OUT */ 539 vsize_t size; 540 struct uvm_object *uobj; 541 voff_t uoffset; 542 vsize_t align; 543 uvm_flag_t flags; 544 { 545 struct vm_map_entry *prev_entry, *new_entry; 546 const int amapwaitflag = (flags & UVM_KMF_NOWAIT) ? 547 AMAP_EXTEND_NOWAIT : 0; 548 vm_prot_t prot = UVM_PROTECTION(flags), maxprot = 549 UVM_MAXPROTECTION(flags); 550 vm_inherit_t inherit = UVM_INHERIT(flags); 551 int advice = UVM_ADVICE(flags); 552 int error, merged = 0, kmap = (vm_map_pmap(map) == pmap_kernel()); 553 UVMHIST_FUNC("uvm_map"); 554 UVMHIST_CALLED(maphist); 555 556 UVMHIST_LOG(maphist, "(map=0x%x, *startp=0x%x, size=%d, flags=0x%x)", 557 map, *startp, size, flags); 558 UVMHIST_LOG(maphist, " uobj/offset 0x%x/%d", uobj, uoffset,0,0); 559 560 /* 561 * detect a popular device driver bug. 562 */ 563 564 KASSERT(curproc != NULL || map->flags & VM_MAP_INTRSAFE); 565 566 /* 567 * check sanity of protection code 568 */ 569 570 if ((prot & maxprot) != prot) { 571 UVMHIST_LOG(maphist, "<- prot. failure: prot=0x%x, max=0x%x", 572 prot, maxprot,0,0); 573 return EACCES; 574 } 575 576 /* 577 * for pager_map, allocate the new entry first to avoid sleeping 578 * for memory while we have the map locked. 579 */ 580 581 new_entry = NULL; 582 if (map == pager_map) { 583 new_entry = uvm_mapent_alloc(map, (flags & UVM_KMF_NOWAIT)); 584 if (__predict_false(new_entry == NULL)) 585 return ENOMEM; 586 } 587 588 /* 589 * figure out where to put new VM range 590 */ 591 592 if (vm_map_lock_try(map) == FALSE) { 593 if (flags & UVM_FLAG_TRYLOCK) { 594 if (new_entry) { 595 uvm_mapent_free(new_entry); 596 } 597 return EAGAIN; 598 } 599 vm_map_lock(map); /* could sleep here */ 600 } 601 if ((prev_entry = uvm_map_findspace(map, *startp, size, startp, 602 uobj, uoffset, align, flags)) == NULL) { 603 UVMHIST_LOG(maphist,"<- uvm_map_findspace failed!",0,0,0,0); 604 vm_map_unlock(map); 605 if (new_entry) { 606 uvm_mapent_free(new_entry); 607 } 608 return ENOMEM; 609 } 610 611 #ifdef PMAP_GROWKERNEL 612 { 613 /* 614 * If the kernel pmap can't map the requested space, 615 * then allocate more resources for it. 616 */ 617 if (map == kernel_map && uvm_maxkaddr < (*startp + size)) 618 uvm_maxkaddr = pmap_growkernel(*startp + size); 619 } 620 #endif 621 622 UVMCNT_INCR(uvm_map_call); 623 624 /* 625 * if uobj is null, then uoffset is either a VAC hint for PMAP_PREFER 626 * [typically from uvm_map_reserve] or it is UVM_UNKNOWN_OFFSET. in 627 * either case we want to zero it before storing it in the map entry 628 * (because it looks strange and confusing when debugging...) 629 * 630 * if uobj is not null 631 * if uoffset is not UVM_UNKNOWN_OFFSET then we have a normal mapping 632 * and we do not need to change uoffset. 633 * if uoffset is UVM_UNKNOWN_OFFSET then we need to find the offset 634 * now (based on the starting address of the map). this case is 635 * for kernel object mappings where we don't know the offset until 636 * the virtual address is found (with uvm_map_findspace). the 637 * offset is the distance we are from the start of the map. 638 */ 639 640 if (uobj == NULL) { 641 uoffset = 0; 642 } else { 643 if (uoffset == UVM_UNKNOWN_OFFSET) { 644 KASSERT(UVM_OBJ_IS_KERN_OBJECT(uobj)); 645 uoffset = *startp - vm_map_min(kernel_map); 646 } 647 } 648 649 /* 650 * try and insert in map by extending previous entry, if possible. 651 * XXX: we don't try and pull back the next entry. might be useful 652 * for a stack, but we are currently allocating our stack in advance. 653 */ 654 655 if (flags & UVM_FLAG_NOMERGE) 656 goto nomerge; 657 658 if (prev_entry->end == *startp && 659 prev_entry != &map->header && 660 prev_entry->object.uvm_obj == uobj) { 661 662 if (uobj && prev_entry->offset + 663 (prev_entry->end - prev_entry->start) != uoffset) 664 goto forwardmerge; 665 666 if (UVM_ET_ISSUBMAP(prev_entry)) 667 goto forwardmerge; 668 669 if (prev_entry->protection != prot || 670 prev_entry->max_protection != maxprot) 671 goto forwardmerge; 672 673 if (prev_entry->inheritance != inherit || 674 prev_entry->advice != advice) 675 goto forwardmerge; 676 677 /* wiring status must match (new area is unwired) */ 678 if (VM_MAPENT_ISWIRED(prev_entry)) 679 goto forwardmerge; 680 681 /* 682 * can't extend a shared amap. note: no need to lock amap to 683 * look at refs since we don't care about its exact value. 684 * if it is one (i.e. we have only reference) it will stay there 685 */ 686 687 if (prev_entry->aref.ar_amap && 688 amap_refs(prev_entry->aref.ar_amap) != 1) { 689 goto forwardmerge; 690 } 691 692 if (prev_entry->aref.ar_amap) { 693 error = amap_extend(prev_entry, size, 694 amapwaitflag | AMAP_EXTEND_FORWARDS); 695 if (error) { 696 vm_map_unlock(map); 697 if (new_entry) { 698 uvm_mapent_free(new_entry); 699 } 700 return error; 701 } 702 } 703 704 if (kmap) 705 UVMCNT_INCR(map_kbackmerge); 706 else 707 UVMCNT_INCR(map_ubackmerge); 708 UVMHIST_LOG(maphist," starting back merge", 0, 0, 0, 0); 709 710 /* 711 * drop our reference to uobj since we are extending a reference 712 * that we already have (the ref count can not drop to zero). 713 */ 714 715 if (uobj && uobj->pgops->pgo_detach) 716 uobj->pgops->pgo_detach(uobj); 717 718 prev_entry->end += size; 719 map->size += size; 720 721 UVMHIST_LOG(maphist,"<- done (via backmerge)!", 0, 0, 0, 0); 722 if (new_entry) { 723 uvm_mapent_free(new_entry); 724 new_entry = NULL; 725 } 726 merged++; 727 } 728 729 forwardmerge: 730 if (prev_entry->next->start == (*startp + size) && 731 prev_entry->next != &map->header && 732 prev_entry->next->object.uvm_obj == uobj) { 733 734 if (uobj && prev_entry->next->offset != uoffset + size) 735 goto nomerge; 736 737 if (UVM_ET_ISSUBMAP(prev_entry->next)) 738 goto nomerge; 739 740 if (prev_entry->next->protection != prot || 741 prev_entry->next->max_protection != maxprot) 742 goto nomerge; 743 744 if (prev_entry->next->inheritance != inherit || 745 prev_entry->next->advice != advice) 746 goto nomerge; 747 748 /* wiring status must match (new area is unwired) */ 749 if (VM_MAPENT_ISWIRED(prev_entry->next)) 750 goto nomerge; 751 752 /* 753 * can't extend a shared amap. note: no need to lock amap to 754 * look at refs since we don't care about its exact value. 755 * if it is one (i.e. we have only reference) it will stay there. 756 * 757 * note that we also can't merge two amaps, so if we 758 * merged with the previous entry which has an amap, 759 * and the next entry also has an amap, we give up. 760 * 761 * Interesting cases: 762 * amap, new, amap -> give up second merge (single fwd extend) 763 * amap, new, none -> double forward extend (extend again here) 764 * none, new, amap -> double backward extend (done here) 765 * uobj, new, amap -> single backward extend (done here) 766 * 767 * XXX should we attempt to deal with someone refilling 768 * the deallocated region between two entries that are 769 * backed by the same amap (ie, arefs is 2, "prev" and 770 * "next" refer to it, and adding this allocation will 771 * close the hole, thus restoring arefs to 1 and 772 * deallocating the "next" vm_map_entry)? -- @@@ 773 */ 774 775 if (prev_entry->next->aref.ar_amap && 776 (amap_refs(prev_entry->next->aref.ar_amap) != 1 || 777 (merged && prev_entry->aref.ar_amap))) { 778 goto nomerge; 779 } 780 781 if (merged) { 782 /* 783 * Try to extend the amap of the previous entry to 784 * cover the next entry as well. If it doesn't work 785 * just skip on, don't actually give up, since we've 786 * already completed the back merge. 787 */ 788 if (prev_entry->aref.ar_amap) { 789 if (amap_extend(prev_entry, 790 prev_entry->next->end - 791 prev_entry->next->start, 792 amapwaitflag | AMAP_EXTEND_FORWARDS)) 793 goto nomerge; 794 } 795 796 /* 797 * Try to extend the amap of the *next* entry 798 * back to cover the new allocation *and* the 799 * previous entry as well (the previous merge 800 * didn't have an amap already otherwise we 801 * wouldn't be checking here for an amap). If 802 * it doesn't work just skip on, again, don't 803 * actually give up, since we've already 804 * completed the back merge. 805 */ 806 else if (prev_entry->next->aref.ar_amap) { 807 if (amap_extend(prev_entry->next, 808 prev_entry->end - 809 prev_entry->start + size, 810 amapwaitflag | AMAP_EXTEND_BACKWARDS)) 811 goto nomerge; 812 } 813 } else { 814 /* 815 * Pull the next entry's amap backwards to cover this 816 * new allocation. 817 */ 818 if (prev_entry->next->aref.ar_amap) { 819 error = amap_extend(prev_entry->next, size, 820 amapwaitflag | AMAP_EXTEND_BACKWARDS); 821 if (error) { 822 vm_map_unlock(map); 823 if (new_entry) { 824 uvm_mapent_free(new_entry); 825 } 826 return error; 827 } 828 } 829 } 830 831 if (merged) { 832 if (kmap) { 833 UVMCNT_DECR(map_kbackmerge); 834 UVMCNT_INCR(map_kbimerge); 835 } else { 836 UVMCNT_DECR(map_ubackmerge); 837 UVMCNT_INCR(map_ubimerge); 838 } 839 } else { 840 if (kmap) 841 UVMCNT_INCR(map_kforwmerge); 842 else 843 UVMCNT_INCR(map_uforwmerge); 844 } 845 UVMHIST_LOG(maphist," starting forward merge", 0, 0, 0, 0); 846 847 /* 848 * drop our reference to uobj since we are extending a reference 849 * that we already have (the ref count can not drop to zero). 850 * (if merged, we've already detached) 851 */ 852 if (uobj && uobj->pgops->pgo_detach && !merged) 853 uobj->pgops->pgo_detach(uobj); 854 855 if (merged) { 856 struct vm_map_entry *dead = prev_entry->next; 857 prev_entry->end = dead->end; 858 uvm_map_entry_unlink(map, dead); 859 if (dead->aref.ar_amap != NULL) { 860 prev_entry->aref = dead->aref; 861 dead->aref.ar_amap = NULL; 862 } 863 uvm_mapent_free(dead); 864 } else { 865 prev_entry->next->start -= size; 866 map->size += size; 867 if (uobj) 868 prev_entry->next->offset = uoffset; 869 } 870 871 UVMHIST_LOG(maphist,"<- done forwardmerge", 0, 0, 0, 0); 872 if (new_entry) { 873 uvm_mapent_free(new_entry); 874 new_entry = NULL; 875 } 876 merged++; 877 } 878 879 nomerge: 880 if (!merged) { 881 UVMHIST_LOG(maphist," allocating new map entry", 0, 0, 0, 0); 882 if (kmap) 883 UVMCNT_INCR(map_knomerge); 884 else 885 UVMCNT_INCR(map_unomerge); 886 887 /* 888 * allocate new entry and link it in. 889 */ 890 891 if (new_entry == NULL) { 892 new_entry = uvm_mapent_alloc(map, 893 (flags & UVM_KMF_NOWAIT)); 894 if (__predict_false(new_entry == NULL)) { 895 vm_map_unlock(map); 896 return ENOMEM; 897 } 898 } 899 new_entry->start = *startp; 900 new_entry->end = new_entry->start + size; 901 new_entry->object.uvm_obj = uobj; 902 new_entry->offset = uoffset; 903 904 if (uobj) 905 new_entry->etype = UVM_ET_OBJ; 906 else 907 new_entry->etype = 0; 908 909 if (flags & UVM_FLAG_COPYONW) { 910 new_entry->etype |= UVM_ET_COPYONWRITE; 911 if ((flags & UVM_FLAG_OVERLAY) == 0) 912 new_entry->etype |= UVM_ET_NEEDSCOPY; 913 } 914 915 new_entry->protection = prot; 916 new_entry->max_protection = maxprot; 917 new_entry->inheritance = inherit; 918 new_entry->wired_count = 0; 919 new_entry->advice = advice; 920 if (flags & UVM_FLAG_OVERLAY) { 921 922 /* 923 * to_add: for BSS we overallocate a little since we 924 * are likely to extend 925 */ 926 927 vaddr_t to_add = (flags & UVM_FLAG_AMAPPAD) ? 928 UVM_AMAP_CHUNK << PAGE_SHIFT : 0; 929 struct vm_amap *amap = amap_alloc(size, to_add, 930 (flags & UVM_KMF_NOWAIT) ? M_NOWAIT : M_WAITOK); 931 if (__predict_false(amap == NULL)) { 932 vm_map_unlock(map); 933 uvm_mapent_free(new_entry); 934 return ENOMEM; 935 } 936 new_entry->aref.ar_pageoff = 0; 937 new_entry->aref.ar_amap = amap; 938 } else { 939 new_entry->aref.ar_pageoff = 0; 940 new_entry->aref.ar_amap = NULL; 941 } 942 uvm_map_entry_link(map, prev_entry, new_entry); 943 map->size += size; 944 945 /* 946 * Update the free space hint 947 */ 948 949 if ((map->first_free == prev_entry) && 950 (prev_entry->end >= new_entry->start)) 951 map->first_free = new_entry; 952 } 953 954 UVMHIST_LOG(maphist,"<- done!", 0, 0, 0, 0); 955 vm_map_unlock(map); 956 return 0; 957 } 958 959 /* 960 * uvm_map_lookup_entry: find map entry at or before an address 961 * 962 * => map must at least be read-locked by caller 963 * => entry is returned in "entry" 964 * => return value is true if address is in the returned entry 965 */ 966 967 boolean_t 968 uvm_map_lookup_entry(map, address, entry) 969 struct vm_map *map; 970 vaddr_t address; 971 struct vm_map_entry **entry; /* OUT */ 972 { 973 struct vm_map_entry *cur; 974 struct vm_map_entry *last; 975 UVMHIST_FUNC("uvm_map_lookup_entry"); 976 UVMHIST_CALLED(maphist); 977 978 UVMHIST_LOG(maphist,"(map=0x%x,addr=0x%x,ent=0x%x)", 979 map, address, entry, 0); 980 981 /* 982 * start looking either from the head of the 983 * list, or from the hint. 984 */ 985 986 simple_lock(&map->hint_lock); 987 cur = map->hint; 988 simple_unlock(&map->hint_lock); 989 990 if (cur == &map->header) 991 cur = cur->next; 992 993 UVMCNT_INCR(uvm_mlk_call); 994 if (address >= cur->start) { 995 996 /* 997 * go from hint to end of list. 998 * 999 * but first, make a quick check to see if 1000 * we are already looking at the entry we 1001 * want (which is usually the case). 1002 * note also that we don't need to save the hint 1003 * here... it is the same hint (unless we are 1004 * at the header, in which case the hint didn't 1005 * buy us anything anyway). 1006 */ 1007 1008 last = &map->header; 1009 if ((cur != last) && (cur->end > address)) { 1010 UVMCNT_INCR(uvm_mlk_hint); 1011 *entry = cur; 1012 UVMHIST_LOG(maphist,"<- got it via hint (0x%x)", 1013 cur, 0, 0, 0); 1014 return (TRUE); 1015 } 1016 } else { 1017 1018 /* 1019 * go from start to hint, *inclusively* 1020 */ 1021 1022 last = cur->next; 1023 cur = map->header.next; 1024 } 1025 1026 /* 1027 * search linearly 1028 */ 1029 1030 while (cur != last) { 1031 if (cur->end > address) { 1032 if (address >= cur->start) { 1033 /* 1034 * save this lookup for future 1035 * hints, and return 1036 */ 1037 1038 *entry = cur; 1039 SAVE_HINT(map, map->hint, cur); 1040 UVMHIST_LOG(maphist,"<- search got it (0x%x)", 1041 cur, 0, 0, 0); 1042 return (TRUE); 1043 } 1044 break; 1045 } 1046 cur = cur->next; 1047 } 1048 *entry = cur->prev; 1049 SAVE_HINT(map, map->hint, *entry); 1050 UVMHIST_LOG(maphist,"<- failed!",0,0,0,0); 1051 return (FALSE); 1052 } 1053 1054 /* 1055 * uvm_map_findspace: find "length" sized space in "map". 1056 * 1057 * => "hint" is a hint about where we want it, unless FINDSPACE_FIXED is 1058 * set (in which case we insist on using "hint"). 1059 * => "result" is VA returned 1060 * => uobj/uoffset are to be used to handle VAC alignment, if required 1061 * => if `align' is non-zero, we attempt to align to that value. 1062 * => caller must at least have read-locked map 1063 * => returns NULL on failure, or pointer to prev. map entry if success 1064 * => note this is a cross between the old vm_map_findspace and vm_map_find 1065 */ 1066 1067 struct vm_map_entry * 1068 uvm_map_findspace(map, hint, length, result, uobj, uoffset, align, flags) 1069 struct vm_map *map; 1070 vaddr_t hint; 1071 vsize_t length; 1072 vaddr_t *result; /* OUT */ 1073 struct uvm_object *uobj; 1074 voff_t uoffset; 1075 vsize_t align; 1076 int flags; 1077 { 1078 struct vm_map_entry *entry, *next, *tmp; 1079 vaddr_t end, orig_hint; 1080 UVMHIST_FUNC("uvm_map_findspace"); 1081 UVMHIST_CALLED(maphist); 1082 1083 UVMHIST_LOG(maphist, "(map=0x%x, hint=0x%x, len=%d, flags=0x%x)", 1084 map, hint, length, flags); 1085 KASSERT((align & (align - 1)) == 0); 1086 KASSERT((flags & UVM_FLAG_FIXED) == 0 || align == 0); 1087 1088 /* 1089 * remember the original hint. if we are aligning, then we 1090 * may have to try again with no alignment constraint if 1091 * we fail the first time. 1092 */ 1093 1094 orig_hint = hint; 1095 if (hint < map->min_offset) { /* check ranges ... */ 1096 if (flags & UVM_FLAG_FIXED) { 1097 UVMHIST_LOG(maphist,"<- VA below map range",0,0,0,0); 1098 return(NULL); 1099 } 1100 hint = map->min_offset; 1101 } 1102 if (hint > map->max_offset) { 1103 UVMHIST_LOG(maphist,"<- VA 0x%x > range [0x%x->0x%x]", 1104 hint, map->min_offset, map->max_offset, 0); 1105 return(NULL); 1106 } 1107 1108 /* 1109 * Look for the first possible address; if there's already 1110 * something at this address, we have to start after it. 1111 */ 1112 1113 if ((flags & UVM_FLAG_FIXED) == 0 && hint == map->min_offset) { 1114 if ((entry = map->first_free) != &map->header) 1115 hint = entry->end; 1116 } else { 1117 if (uvm_map_lookup_entry(map, hint, &tmp)) { 1118 /* "hint" address already in use ... */ 1119 if (flags & UVM_FLAG_FIXED) { 1120 UVMHIST_LOG(maphist,"<- fixed & VA in use", 1121 0, 0, 0, 0); 1122 return(NULL); 1123 } 1124 hint = tmp->end; 1125 } 1126 entry = tmp; 1127 } 1128 1129 /* 1130 * Look through the rest of the map, trying to fit a new region in 1131 * the gap between existing regions, or after the very last region. 1132 * note: entry->end = base VA of current gap, 1133 * next->start = VA of end of current gap 1134 */ 1135 1136 for (;; hint = (entry = next)->end) { 1137 1138 /* 1139 * Find the end of the proposed new region. Be sure we didn't 1140 * go beyond the end of the map, or wrap around the address; 1141 * if so, we lose. Otherwise, if this is the last entry, or 1142 * if the proposed new region fits before the next entry, we 1143 * win. 1144 */ 1145 1146 #ifdef PMAP_PREFER 1147 /* 1148 * push hint forward as needed to avoid VAC alias problems. 1149 * we only do this if a valid offset is specified. 1150 */ 1151 1152 if ((flags & UVM_FLAG_FIXED) == 0 && 1153 uoffset != UVM_UNKNOWN_OFFSET) 1154 PMAP_PREFER(uoffset, &hint); 1155 #endif 1156 if (align != 0) { 1157 if ((hint & (align - 1)) != 0) 1158 hint = roundup(hint, align); 1159 /* 1160 * XXX Should we PMAP_PREFER() here again? 1161 */ 1162 } 1163 end = hint + length; 1164 if (end > map->max_offset || end < hint) { 1165 UVMHIST_LOG(maphist,"<- failed (off end)", 0,0,0,0); 1166 if (align != 0) { 1167 UVMHIST_LOG(maphist, 1168 "calling recursively, no align", 1169 0,0,0,0); 1170 return (uvm_map_findspace(map, orig_hint, 1171 length, result, uobj, uoffset, 0, flags)); 1172 } 1173 return (NULL); 1174 } 1175 next = entry->next; 1176 if (next == &map->header || next->start >= end) 1177 break; 1178 if (flags & UVM_FLAG_FIXED) { 1179 UVMHIST_LOG(maphist,"<- fixed mapping failed", 0,0,0,0); 1180 return(NULL); /* only one shot at it ... */ 1181 } 1182 } 1183 SAVE_HINT(map, map->hint, entry); 1184 *result = hint; 1185 UVMHIST_LOG(maphist,"<- got it! (result=0x%x)", hint, 0,0,0); 1186 return (entry); 1187 } 1188 1189 /* 1190 * U N M A P - m a i n h e l p e r f u n c t i o n s 1191 */ 1192 1193 /* 1194 * uvm_unmap_remove: remove mappings from a vm_map (from "start" up to "stop") 1195 * 1196 * => caller must check alignment and size 1197 * => map must be locked by caller 1198 * => we return a list of map entries that we've remove from the map 1199 * in "entry_list" 1200 */ 1201 1202 void 1203 uvm_unmap_remove(map, start, end, entry_list) 1204 struct vm_map *map; 1205 vaddr_t start, end; 1206 struct vm_map_entry **entry_list; /* OUT */ 1207 { 1208 struct vm_map_entry *entry, *first_entry, *next; 1209 vaddr_t len; 1210 UVMHIST_FUNC("uvm_unmap_remove"); UVMHIST_CALLED(maphist); 1211 1212 UVMHIST_LOG(maphist,"(map=0x%x, start=0x%x, end=0x%x)", 1213 map, start, end, 0); 1214 VM_MAP_RANGE_CHECK(map, start, end); 1215 1216 /* 1217 * find first entry 1218 */ 1219 1220 if (uvm_map_lookup_entry(map, start, &first_entry) == TRUE) { 1221 /* clip and go... */ 1222 entry = first_entry; 1223 UVM_MAP_CLIP_START(map, entry, start); 1224 /* critical! prevents stale hint */ 1225 SAVE_HINT(map, entry, entry->prev); 1226 } else { 1227 entry = first_entry->next; 1228 } 1229 1230 /* 1231 * Save the free space hint 1232 */ 1233 1234 if (map->first_free->start >= start) 1235 map->first_free = entry->prev; 1236 1237 /* 1238 * note: we now re-use first_entry for a different task. we remove 1239 * a number of map entries from the map and save them in a linked 1240 * list headed by "first_entry". once we remove them from the map 1241 * the caller should unlock the map and drop the references to the 1242 * backing objects [c.f. uvm_unmap_detach]. the object is to 1243 * separate unmapping from reference dropping. why? 1244 * [1] the map has to be locked for unmapping 1245 * [2] the map need not be locked for reference dropping 1246 * [3] dropping references may trigger pager I/O, and if we hit 1247 * a pager that does synchronous I/O we may have to wait for it. 1248 * [4] we would like all waiting for I/O to occur with maps unlocked 1249 * so that we don't block other threads. 1250 */ 1251 1252 first_entry = NULL; 1253 *entry_list = NULL; 1254 1255 /* 1256 * break up the area into map entry sized regions and unmap. note 1257 * that all mappings have to be removed before we can even consider 1258 * dropping references to amaps or VM objects (otherwise we could end 1259 * up with a mapping to a page on the free list which would be very bad) 1260 */ 1261 1262 while ((entry != &map->header) && (entry->start < end)) { 1263 UVM_MAP_CLIP_END(map, entry, end); 1264 next = entry->next; 1265 len = entry->end - entry->start; 1266 1267 /* 1268 * unwire before removing addresses from the pmap; otherwise 1269 * unwiring will put the entries back into the pmap (XXX). 1270 */ 1271 1272 if (VM_MAPENT_ISWIRED(entry)) { 1273 uvm_map_entry_unwire(map, entry); 1274 } 1275 if ((map->flags & VM_MAP_PAGEABLE) == 0) { 1276 1277 /* 1278 * if the map is non-pageable, any pages mapped there 1279 * must be wired and entered with pmap_kenter_pa(), 1280 * and we should free any such pages immediately. 1281 * this is mostly used for kmem_map and mb_map. 1282 */ 1283 1284 uvm_km_pgremove_intrsafe(entry->start, entry->end); 1285 pmap_kremove(entry->start, len); 1286 } else if (UVM_ET_ISOBJ(entry) && 1287 UVM_OBJ_IS_KERN_OBJECT(entry->object.uvm_obj)) { 1288 KASSERT(vm_map_pmap(map) == pmap_kernel()); 1289 1290 /* 1291 * note: kernel object mappings are currently used in 1292 * two ways: 1293 * [1] "normal" mappings of pages in the kernel object 1294 * [2] uvm_km_valloc'd allocations in which we 1295 * pmap_enter in some non-kernel-object page 1296 * (e.g. vmapbuf). 1297 * 1298 * for case [1], we need to remove the mapping from 1299 * the pmap and then remove the page from the kernel 1300 * object (because, once pages in a kernel object are 1301 * unmapped they are no longer needed, unlike, say, 1302 * a vnode where you might want the data to persist 1303 * until flushed out of a queue). 1304 * 1305 * for case [2], we need to remove the mapping from 1306 * the pmap. there shouldn't be any pages at the 1307 * specified offset in the kernel object [but it 1308 * doesn't hurt to call uvm_km_pgremove just to be 1309 * safe?] 1310 * 1311 * uvm_km_pgremove currently does the following: 1312 * for pages in the kernel object in range: 1313 * - drops the swap slot 1314 * - uvm_pagefree the page 1315 */ 1316 1317 /* 1318 * remove mappings from pmap and drop the pages 1319 * from the object. offsets are always relative 1320 * to vm_map_min(kernel_map). 1321 */ 1322 1323 pmap_remove(pmap_kernel(), entry->start, 1324 entry->start + len); 1325 uvm_km_pgremove(entry->object.uvm_obj, 1326 entry->start - vm_map_min(kernel_map), 1327 entry->end - vm_map_min(kernel_map)); 1328 1329 /* 1330 * null out kernel_object reference, we've just 1331 * dropped it 1332 */ 1333 1334 entry->etype &= ~UVM_ET_OBJ; 1335 entry->object.uvm_obj = NULL; 1336 } else if (UVM_ET_ISOBJ(entry) || entry->aref.ar_amap) { 1337 1338 /* 1339 * remove mappings the standard way. 1340 */ 1341 1342 pmap_remove(map->pmap, entry->start, entry->end); 1343 } 1344 1345 /* 1346 * remove entry from map and put it on our list of entries 1347 * that we've nuked. then go to next entry. 1348 */ 1349 1350 UVMHIST_LOG(maphist, " removed map entry 0x%x", entry, 0, 0,0); 1351 1352 /* critical! prevents stale hint */ 1353 SAVE_HINT(map, entry, entry->prev); 1354 1355 uvm_map_entry_unlink(map, entry); 1356 map->size -= len; 1357 entry->next = first_entry; 1358 first_entry = entry; 1359 entry = next; 1360 } 1361 if ((map->flags & VM_MAP_DYING) == 0) { 1362 pmap_update(vm_map_pmap(map)); 1363 } 1364 1365 /* 1366 * now we've cleaned up the map and are ready for the caller to drop 1367 * references to the mapped objects. 1368 */ 1369 1370 *entry_list = first_entry; 1371 UVMHIST_LOG(maphist,"<- done!", 0, 0, 0, 0); 1372 } 1373 1374 /* 1375 * uvm_unmap_detach: drop references in a chain of map entries 1376 * 1377 * => we will free the map entries as we traverse the list. 1378 */ 1379 1380 void 1381 uvm_unmap_detach(first_entry, flags) 1382 struct vm_map_entry *first_entry; 1383 int flags; 1384 { 1385 struct vm_map_entry *next_entry; 1386 UVMHIST_FUNC("uvm_unmap_detach"); UVMHIST_CALLED(maphist); 1387 1388 while (first_entry) { 1389 KASSERT(!VM_MAPENT_ISWIRED(first_entry)); 1390 UVMHIST_LOG(maphist, 1391 " detach 0x%x: amap=0x%x, obj=0x%x, submap?=%d", 1392 first_entry, first_entry->aref.ar_amap, 1393 first_entry->object.uvm_obj, 1394 UVM_ET_ISSUBMAP(first_entry)); 1395 1396 /* 1397 * drop reference to amap, if we've got one 1398 */ 1399 1400 if (first_entry->aref.ar_amap) 1401 uvm_map_unreference_amap(first_entry, flags); 1402 1403 /* 1404 * drop reference to our backing object, if we've got one 1405 */ 1406 1407 KASSERT(!UVM_ET_ISSUBMAP(first_entry)); 1408 if (UVM_ET_ISOBJ(first_entry) && 1409 first_entry->object.uvm_obj->pgops->pgo_detach) { 1410 (*first_entry->object.uvm_obj->pgops->pgo_detach) 1411 (first_entry->object.uvm_obj); 1412 } 1413 next_entry = first_entry->next; 1414 uvm_mapent_free(first_entry); 1415 first_entry = next_entry; 1416 } 1417 UVMHIST_LOG(maphist, "<- done", 0,0,0,0); 1418 } 1419 1420 /* 1421 * E X T R A C T I O N F U N C T I O N S 1422 */ 1423 1424 /* 1425 * uvm_map_reserve: reserve space in a vm_map for future use. 1426 * 1427 * => we reserve space in a map by putting a dummy map entry in the 1428 * map (dummy means obj=NULL, amap=NULL, prot=VM_PROT_NONE) 1429 * => map should be unlocked (we will write lock it) 1430 * => we return true if we were able to reserve space 1431 * => XXXCDC: should be inline? 1432 */ 1433 1434 int 1435 uvm_map_reserve(map, size, offset, align, raddr) 1436 struct vm_map *map; 1437 vsize_t size; 1438 vaddr_t offset; /* hint for pmap_prefer */ 1439 vsize_t align; /* alignment hint */ 1440 vaddr_t *raddr; /* IN:hint, OUT: reserved VA */ 1441 { 1442 UVMHIST_FUNC("uvm_map_reserve"); UVMHIST_CALLED(maphist); 1443 1444 UVMHIST_LOG(maphist, "(map=0x%x, size=0x%x, offset=0x%x,addr=0x%x)", 1445 map,size,offset,raddr); 1446 1447 size = round_page(size); 1448 if (*raddr < vm_map_min(map)) 1449 *raddr = vm_map_min(map); /* hint */ 1450 1451 /* 1452 * reserve some virtual space. 1453 */ 1454 1455 if (uvm_map(map, raddr, size, NULL, offset, 0, 1456 UVM_MAPFLAG(UVM_PROT_NONE, UVM_PROT_NONE, UVM_INH_NONE, 1457 UVM_ADV_RANDOM, UVM_FLAG_NOMERGE)) != 0) { 1458 UVMHIST_LOG(maphist, "<- done (no VM)", 0,0,0,0); 1459 return (FALSE); 1460 } 1461 1462 UVMHIST_LOG(maphist, "<- done (*raddr=0x%x)", *raddr,0,0,0); 1463 return (TRUE); 1464 } 1465 1466 /* 1467 * uvm_map_replace: replace a reserved (blank) area of memory with 1468 * real mappings. 1469 * 1470 * => caller must WRITE-LOCK the map 1471 * => we return TRUE if replacement was a success 1472 * => we expect the newents chain to have nnewents entrys on it and 1473 * we expect newents->prev to point to the last entry on the list 1474 * => note newents is allowed to be NULL 1475 */ 1476 1477 int 1478 uvm_map_replace(map, start, end, newents, nnewents) 1479 struct vm_map *map; 1480 vaddr_t start, end; 1481 struct vm_map_entry *newents; 1482 int nnewents; 1483 { 1484 struct vm_map_entry *oldent, *last; 1485 1486 /* 1487 * first find the blank map entry at the specified address 1488 */ 1489 1490 if (!uvm_map_lookup_entry(map, start, &oldent)) { 1491 return(FALSE); 1492 } 1493 1494 /* 1495 * check to make sure we have a proper blank entry 1496 */ 1497 1498 if (oldent->start != start || oldent->end != end || 1499 oldent->object.uvm_obj != NULL || oldent->aref.ar_amap != NULL) { 1500 return (FALSE); 1501 } 1502 1503 #ifdef DIAGNOSTIC 1504 1505 /* 1506 * sanity check the newents chain 1507 */ 1508 1509 { 1510 struct vm_map_entry *tmpent = newents; 1511 int nent = 0; 1512 vaddr_t cur = start; 1513 1514 while (tmpent) { 1515 nent++; 1516 if (tmpent->start < cur) 1517 panic("uvm_map_replace1"); 1518 if (tmpent->start > tmpent->end || tmpent->end > end) { 1519 printf("tmpent->start=0x%lx, tmpent->end=0x%lx, end=0x%lx\n", 1520 tmpent->start, tmpent->end, end); 1521 panic("uvm_map_replace2"); 1522 } 1523 cur = tmpent->end; 1524 if (tmpent->next) { 1525 if (tmpent->next->prev != tmpent) 1526 panic("uvm_map_replace3"); 1527 } else { 1528 if (newents->prev != tmpent) 1529 panic("uvm_map_replace4"); 1530 } 1531 tmpent = tmpent->next; 1532 } 1533 if (nent != nnewents) 1534 panic("uvm_map_replace5"); 1535 } 1536 #endif 1537 1538 /* 1539 * map entry is a valid blank! replace it. (this does all the 1540 * work of map entry link/unlink...). 1541 */ 1542 1543 if (newents) { 1544 last = newents->prev; 1545 1546 /* critical: flush stale hints out of map */ 1547 SAVE_HINT(map, map->hint, newents); 1548 if (map->first_free == oldent) 1549 map->first_free = last; 1550 1551 last->next = oldent->next; 1552 last->next->prev = last; 1553 newents->prev = oldent->prev; 1554 newents->prev->next = newents; 1555 map->nentries = map->nentries + (nnewents - 1); 1556 1557 } else { 1558 1559 /* critical: flush stale hints out of map */ 1560 SAVE_HINT(map, map->hint, oldent->prev); 1561 if (map->first_free == oldent) 1562 map->first_free = oldent->prev; 1563 1564 /* NULL list of new entries: just remove the old one */ 1565 uvm_map_entry_unlink(map, oldent); 1566 } 1567 1568 1569 /* 1570 * now we can free the old blank entry, unlock the map and return. 1571 */ 1572 1573 uvm_mapent_free(oldent); 1574 return(TRUE); 1575 } 1576 1577 /* 1578 * uvm_map_extract: extract a mapping from a map and put it somewhere 1579 * (maybe removing the old mapping) 1580 * 1581 * => maps should be unlocked (we will write lock them) 1582 * => returns 0 on success, error code otherwise 1583 * => start must be page aligned 1584 * => len must be page sized 1585 * => flags: 1586 * UVM_EXTRACT_REMOVE: remove mappings from srcmap 1587 * UVM_EXTRACT_CONTIG: abort if unmapped area (advisory only) 1588 * UVM_EXTRACT_QREF: for a temporary extraction do quick obj refs 1589 * UVM_EXTRACT_FIXPROT: set prot to maxprot as we go 1590 * >>>NOTE: if you set REMOVE, you are not allowed to use CONTIG or QREF!<<< 1591 * >>>NOTE: QREF's must be unmapped via the QREF path, thus should only 1592 * be used from within the kernel in a kernel level map <<< 1593 */ 1594 1595 int 1596 uvm_map_extract(srcmap, start, len, dstmap, dstaddrp, flags) 1597 struct vm_map *srcmap, *dstmap; 1598 vaddr_t start, *dstaddrp; 1599 vsize_t len; 1600 int flags; 1601 { 1602 vaddr_t dstaddr, end, newend, oldoffset, fudge, orig_fudge, 1603 oldstart; 1604 struct vm_map_entry *chain, *endchain, *entry, *orig_entry, *newentry, 1605 *deadentry, *oldentry; 1606 vsize_t elen; 1607 int nchain, error, copy_ok; 1608 UVMHIST_FUNC("uvm_map_extract"); UVMHIST_CALLED(maphist); 1609 1610 UVMHIST_LOG(maphist,"(srcmap=0x%x,start=0x%x, len=0x%x", srcmap, start, 1611 len,0); 1612 UVMHIST_LOG(maphist," ...,dstmap=0x%x, flags=0x%x)", dstmap,flags,0,0); 1613 1614 /* 1615 * step 0: sanity check: start must be on a page boundary, length 1616 * must be page sized. can't ask for CONTIG/QREF if you asked for 1617 * REMOVE. 1618 */ 1619 1620 KASSERT((start & PAGE_MASK) == 0 && (len & PAGE_MASK) == 0); 1621 KASSERT((flags & UVM_EXTRACT_REMOVE) == 0 || 1622 (flags & (UVM_EXTRACT_CONTIG|UVM_EXTRACT_QREF)) == 0); 1623 1624 /* 1625 * step 1: reserve space in the target map for the extracted area 1626 */ 1627 1628 dstaddr = vm_map_min(dstmap); 1629 if (uvm_map_reserve(dstmap, len, start, 0, &dstaddr) == FALSE) 1630 return(ENOMEM); 1631 *dstaddrp = dstaddr; /* pass address back to caller */ 1632 UVMHIST_LOG(maphist, " dstaddr=0x%x", dstaddr,0,0,0); 1633 1634 /* 1635 * step 2: setup for the extraction process loop by init'ing the 1636 * map entry chain, locking src map, and looking up the first useful 1637 * entry in the map. 1638 */ 1639 1640 end = start + len; 1641 newend = dstaddr + len; 1642 chain = endchain = NULL; 1643 nchain = 0; 1644 vm_map_lock(srcmap); 1645 1646 if (uvm_map_lookup_entry(srcmap, start, &entry)) { 1647 1648 /* "start" is within an entry */ 1649 if (flags & UVM_EXTRACT_QREF) { 1650 1651 /* 1652 * for quick references we don't clip the entry, so 1653 * the entry may map space "before" the starting 1654 * virtual address... this is the "fudge" factor 1655 * (which can be non-zero only the first time 1656 * through the "while" loop in step 3). 1657 */ 1658 1659 fudge = start - entry->start; 1660 } else { 1661 1662 /* 1663 * normal reference: we clip the map to fit (thus 1664 * fudge is zero) 1665 */ 1666 1667 UVM_MAP_CLIP_START(srcmap, entry, start); 1668 SAVE_HINT(srcmap, srcmap->hint, entry->prev); 1669 fudge = 0; 1670 } 1671 } else { 1672 1673 /* "start" is not within an entry ... skip to next entry */ 1674 if (flags & UVM_EXTRACT_CONTIG) { 1675 error = EINVAL; 1676 goto bad; /* definite hole here ... */ 1677 } 1678 1679 entry = entry->next; 1680 fudge = 0; 1681 } 1682 1683 /* save values from srcmap for step 6 */ 1684 orig_entry = entry; 1685 orig_fudge = fudge; 1686 1687 /* 1688 * step 3: now start looping through the map entries, extracting 1689 * as we go. 1690 */ 1691 1692 while (entry->start < end && entry != &srcmap->header) { 1693 1694 /* if we are not doing a quick reference, clip it */ 1695 if ((flags & UVM_EXTRACT_QREF) == 0) 1696 UVM_MAP_CLIP_END(srcmap, entry, end); 1697 1698 /* clear needs_copy (allow chunking) */ 1699 if (UVM_ET_ISNEEDSCOPY(entry)) { 1700 if (fudge) 1701 oldstart = entry->start; 1702 else 1703 oldstart = 0; /* XXX: gcc */ 1704 amap_copy(srcmap, entry, M_NOWAIT, TRUE, start, end); 1705 if (UVM_ET_ISNEEDSCOPY(entry)) { /* failed? */ 1706 error = ENOMEM; 1707 goto bad; 1708 } 1709 1710 /* amap_copy could clip (during chunk)! update fudge */ 1711 if (fudge) { 1712 fudge = fudge - (entry->start - oldstart); 1713 orig_fudge = fudge; 1714 } 1715 } 1716 1717 /* calculate the offset of this from "start" */ 1718 oldoffset = (entry->start + fudge) - start; 1719 1720 /* allocate a new map entry */ 1721 newentry = uvm_mapent_alloc(dstmap, 0); 1722 if (newentry == NULL) { 1723 error = ENOMEM; 1724 goto bad; 1725 } 1726 1727 /* set up new map entry */ 1728 newentry->next = NULL; 1729 newentry->prev = endchain; 1730 newentry->start = dstaddr + oldoffset; 1731 newentry->end = 1732 newentry->start + (entry->end - (entry->start + fudge)); 1733 if (newentry->end > newend || newentry->end < newentry->start) 1734 newentry->end = newend; 1735 newentry->object.uvm_obj = entry->object.uvm_obj; 1736 if (newentry->object.uvm_obj) { 1737 if (newentry->object.uvm_obj->pgops->pgo_reference) 1738 newentry->object.uvm_obj->pgops-> 1739 pgo_reference(newentry->object.uvm_obj); 1740 newentry->offset = entry->offset + fudge; 1741 } else { 1742 newentry->offset = 0; 1743 } 1744 newentry->etype = entry->etype; 1745 newentry->protection = (flags & UVM_EXTRACT_FIXPROT) ? 1746 entry->max_protection : entry->protection; 1747 newentry->max_protection = entry->max_protection; 1748 newentry->inheritance = entry->inheritance; 1749 newentry->wired_count = 0; 1750 newentry->aref.ar_amap = entry->aref.ar_amap; 1751 if (newentry->aref.ar_amap) { 1752 newentry->aref.ar_pageoff = 1753 entry->aref.ar_pageoff + (fudge >> PAGE_SHIFT); 1754 uvm_map_reference_amap(newentry, AMAP_SHARED | 1755 ((flags & UVM_EXTRACT_QREF) ? AMAP_REFALL : 0)); 1756 } else { 1757 newentry->aref.ar_pageoff = 0; 1758 } 1759 newentry->advice = entry->advice; 1760 1761 /* now link it on the chain */ 1762 nchain++; 1763 if (endchain == NULL) { 1764 chain = endchain = newentry; 1765 } else { 1766 endchain->next = newentry; 1767 endchain = newentry; 1768 } 1769 1770 /* end of 'while' loop! */ 1771 if ((flags & UVM_EXTRACT_CONTIG) && entry->end < end && 1772 (entry->next == &srcmap->header || 1773 entry->next->start != entry->end)) { 1774 error = EINVAL; 1775 goto bad; 1776 } 1777 entry = entry->next; 1778 fudge = 0; 1779 } 1780 1781 /* 1782 * step 4: close off chain (in format expected by uvm_map_replace) 1783 */ 1784 1785 if (chain) 1786 chain->prev = endchain; 1787 1788 /* 1789 * step 5: attempt to lock the dest map so we can pmap_copy. 1790 * note usage of copy_ok: 1791 * 1 => dstmap locked, pmap_copy ok, and we "replace" here (step 5) 1792 * 0 => dstmap unlocked, NO pmap_copy, and we will "replace" in step 7 1793 */ 1794 1795 if (srcmap == dstmap || vm_map_lock_try(dstmap) == TRUE) { 1796 copy_ok = 1; 1797 if (!uvm_map_replace(dstmap, dstaddr, dstaddr+len, chain, 1798 nchain)) { 1799 if (srcmap != dstmap) 1800 vm_map_unlock(dstmap); 1801 error = EIO; 1802 goto bad; 1803 } 1804 } else { 1805 copy_ok = 0; 1806 /* replace defered until step 7 */ 1807 } 1808 1809 /* 1810 * step 6: traverse the srcmap a second time to do the following: 1811 * - if we got a lock on the dstmap do pmap_copy 1812 * - if UVM_EXTRACT_REMOVE remove the entries 1813 * we make use of orig_entry and orig_fudge (saved in step 2) 1814 */ 1815 1816 if (copy_ok || (flags & UVM_EXTRACT_REMOVE)) { 1817 1818 /* purge possible stale hints from srcmap */ 1819 if (flags & UVM_EXTRACT_REMOVE) { 1820 SAVE_HINT(srcmap, srcmap->hint, orig_entry->prev); 1821 if (srcmap->first_free->start >= start) 1822 srcmap->first_free = orig_entry->prev; 1823 } 1824 1825 entry = orig_entry; 1826 fudge = orig_fudge; 1827 deadentry = NULL; /* for UVM_EXTRACT_REMOVE */ 1828 1829 while (entry->start < end && entry != &srcmap->header) { 1830 if (copy_ok) { 1831 oldoffset = (entry->start + fudge) - start; 1832 elen = MIN(end, entry->end) - 1833 (entry->start + fudge); 1834 pmap_copy(dstmap->pmap, srcmap->pmap, 1835 dstaddr + oldoffset, elen, 1836 entry->start + fudge); 1837 } 1838 1839 /* we advance "entry" in the following if statement */ 1840 if (flags & UVM_EXTRACT_REMOVE) { 1841 pmap_remove(srcmap->pmap, entry->start, 1842 entry->end); 1843 oldentry = entry; /* save entry */ 1844 entry = entry->next; /* advance */ 1845 uvm_map_entry_unlink(srcmap, oldentry); 1846 /* add to dead list */ 1847 oldentry->next = deadentry; 1848 deadentry = oldentry; 1849 } else { 1850 entry = entry->next; /* advance */ 1851 } 1852 1853 /* end of 'while' loop */ 1854 fudge = 0; 1855 } 1856 pmap_update(srcmap->pmap); 1857 1858 /* 1859 * unlock dstmap. we will dispose of deadentry in 1860 * step 7 if needed 1861 */ 1862 1863 if (copy_ok && srcmap != dstmap) 1864 vm_map_unlock(dstmap); 1865 1866 } else { 1867 deadentry = NULL; 1868 } 1869 1870 /* 1871 * step 7: we are done with the source map, unlock. if copy_ok 1872 * is 0 then we have not replaced the dummy mapping in dstmap yet 1873 * and we need to do so now. 1874 */ 1875 1876 vm_map_unlock(srcmap); 1877 if ((flags & UVM_EXTRACT_REMOVE) && deadentry) 1878 uvm_unmap_detach(deadentry, 0); /* dispose of old entries */ 1879 1880 /* now do the replacement if we didn't do it in step 5 */ 1881 if (copy_ok == 0) { 1882 vm_map_lock(dstmap); 1883 error = uvm_map_replace(dstmap, dstaddr, dstaddr+len, chain, 1884 nchain); 1885 vm_map_unlock(dstmap); 1886 1887 if (error == FALSE) { 1888 error = EIO; 1889 goto bad2; 1890 } 1891 } 1892 return(0); 1893 1894 /* 1895 * bad: failure recovery 1896 */ 1897 bad: 1898 vm_map_unlock(srcmap); 1899 bad2: /* src already unlocked */ 1900 if (chain) 1901 uvm_unmap_detach(chain, 1902 (flags & UVM_EXTRACT_QREF) ? AMAP_REFALL : 0); 1903 uvm_unmap(dstmap, dstaddr, dstaddr+len); /* ??? */ 1904 return(error); 1905 } 1906 1907 /* end of extraction functions */ 1908 1909 /* 1910 * uvm_map_submap: punch down part of a map into a submap 1911 * 1912 * => only the kernel_map is allowed to be submapped 1913 * => the purpose of submapping is to break up the locking granularity 1914 * of a larger map 1915 * => the range specified must have been mapped previously with a uvm_map() 1916 * call [with uobj==NULL] to create a blank map entry in the main map. 1917 * [And it had better still be blank!] 1918 * => maps which contain submaps should never be copied or forked. 1919 * => to remove a submap, use uvm_unmap() on the main map 1920 * and then uvm_map_deallocate() the submap. 1921 * => main map must be unlocked. 1922 * => submap must have been init'd and have a zero reference count. 1923 * [need not be locked as we don't actually reference it] 1924 */ 1925 1926 int 1927 uvm_map_submap(map, start, end, submap) 1928 struct vm_map *map, *submap; 1929 vaddr_t start, end; 1930 { 1931 struct vm_map_entry *entry; 1932 int error; 1933 1934 vm_map_lock(map); 1935 VM_MAP_RANGE_CHECK(map, start, end); 1936 1937 if (uvm_map_lookup_entry(map, start, &entry)) { 1938 UVM_MAP_CLIP_START(map, entry, start); 1939 UVM_MAP_CLIP_END(map, entry, end); /* to be safe */ 1940 } else { 1941 entry = NULL; 1942 } 1943 1944 if (entry != NULL && 1945 entry->start == start && entry->end == end && 1946 entry->object.uvm_obj == NULL && entry->aref.ar_amap == NULL && 1947 !UVM_ET_ISCOPYONWRITE(entry) && !UVM_ET_ISNEEDSCOPY(entry)) { 1948 entry->etype |= UVM_ET_SUBMAP; 1949 entry->object.sub_map = submap; 1950 entry->offset = 0; 1951 uvm_map_reference(submap); 1952 error = 0; 1953 } else { 1954 error = EINVAL; 1955 } 1956 vm_map_unlock(map); 1957 return error; 1958 } 1959 1960 1961 /* 1962 * uvm_map_protect: change map protection 1963 * 1964 * => set_max means set max_protection. 1965 * => map must be unlocked. 1966 */ 1967 1968 #define MASK(entry) (UVM_ET_ISCOPYONWRITE(entry) ? \ 1969 ~VM_PROT_WRITE : VM_PROT_ALL) 1970 1971 int 1972 uvm_map_protect(map, start, end, new_prot, set_max) 1973 struct vm_map *map; 1974 vaddr_t start, end; 1975 vm_prot_t new_prot; 1976 boolean_t set_max; 1977 { 1978 struct vm_map_entry *current, *entry; 1979 int error = 0; 1980 UVMHIST_FUNC("uvm_map_protect"); UVMHIST_CALLED(maphist); 1981 UVMHIST_LOG(maphist,"(map=0x%x,start=0x%x,end=0x%x,new_prot=0x%x)", 1982 map, start, end, new_prot); 1983 1984 vm_map_lock(map); 1985 VM_MAP_RANGE_CHECK(map, start, end); 1986 if (uvm_map_lookup_entry(map, start, &entry)) { 1987 UVM_MAP_CLIP_START(map, entry, start); 1988 } else { 1989 entry = entry->next; 1990 } 1991 1992 /* 1993 * make a first pass to check for protection violations. 1994 */ 1995 1996 current = entry; 1997 while ((current != &map->header) && (current->start < end)) { 1998 if (UVM_ET_ISSUBMAP(current)) { 1999 error = EINVAL; 2000 goto out; 2001 } 2002 if ((new_prot & current->max_protection) != new_prot) { 2003 error = EACCES; 2004 goto out; 2005 } 2006 /* 2007 * Don't allow VM_PROT_EXECUTE to be set on entries that 2008 * point to vnodes that are associated with a NOEXEC file 2009 * system. 2010 */ 2011 if (UVM_ET_ISOBJ(current) && 2012 UVM_OBJ_IS_VNODE(current->object.uvm_obj)) { 2013 struct vnode *vp = 2014 (struct vnode *) current->object.uvm_obj; 2015 2016 if ((new_prot & VM_PROT_EXECUTE) != 0 && 2017 (vp->v_mount->mnt_flag & MNT_NOEXEC) != 0) { 2018 error = EACCES; 2019 goto out; 2020 } 2021 } 2022 current = current->next; 2023 } 2024 2025 /* go back and fix up protections (no need to clip this time). */ 2026 2027 current = entry; 2028 while ((current != &map->header) && (current->start < end)) { 2029 vm_prot_t old_prot; 2030 2031 UVM_MAP_CLIP_END(map, current, end); 2032 old_prot = current->protection; 2033 if (set_max) 2034 current->protection = 2035 (current->max_protection = new_prot) & old_prot; 2036 else 2037 current->protection = new_prot; 2038 2039 /* 2040 * update physical map if necessary. worry about copy-on-write 2041 * here -- CHECK THIS XXX 2042 */ 2043 2044 if (current->protection != old_prot) { 2045 /* update pmap! */ 2046 pmap_protect(map->pmap, current->start, current->end, 2047 current->protection & MASK(entry)); 2048 2049 /* 2050 * If this entry points at a vnode, and the 2051 * protection includes VM_PROT_EXECUTE, mark 2052 * the vnode as VEXECMAP. 2053 */ 2054 if (UVM_ET_ISOBJ(current)) { 2055 struct uvm_object *uobj = 2056 current->object.uvm_obj; 2057 2058 if (UVM_OBJ_IS_VNODE(uobj) && 2059 (current->protection & VM_PROT_EXECUTE)) 2060 vn_markexec((struct vnode *) uobj); 2061 } 2062 } 2063 2064 /* 2065 * If the map is configured to lock any future mappings, 2066 * wire this entry now if the old protection was VM_PROT_NONE 2067 * and the new protection is not VM_PROT_NONE. 2068 */ 2069 2070 if ((map->flags & VM_MAP_WIREFUTURE) != 0 && 2071 VM_MAPENT_ISWIRED(entry) == 0 && 2072 old_prot == VM_PROT_NONE && 2073 new_prot != VM_PROT_NONE) { 2074 if (uvm_map_pageable(map, entry->start, 2075 entry->end, FALSE, 2076 UVM_LK_ENTER|UVM_LK_EXIT) != 0) { 2077 2078 /* 2079 * If locking the entry fails, remember the 2080 * error if it's the first one. Note we 2081 * still continue setting the protection in 2082 * the map, but will return the error 2083 * condition regardless. 2084 * 2085 * XXX Ignore what the actual error is, 2086 * XXX just call it a resource shortage 2087 * XXX so that it doesn't get confused 2088 * XXX what uvm_map_protect() itself would 2089 * XXX normally return. 2090 */ 2091 2092 error = ENOMEM; 2093 } 2094 } 2095 current = current->next; 2096 } 2097 pmap_update(map->pmap); 2098 2099 out: 2100 vm_map_unlock(map); 2101 UVMHIST_LOG(maphist, "<- done, error=%d",error,0,0,0); 2102 return error; 2103 } 2104 2105 #undef MASK 2106 2107 /* 2108 * uvm_map_inherit: set inheritance code for range of addrs in map. 2109 * 2110 * => map must be unlocked 2111 * => note that the inherit code is used during a "fork". see fork 2112 * code for details. 2113 */ 2114 2115 int 2116 uvm_map_inherit(map, start, end, new_inheritance) 2117 struct vm_map *map; 2118 vaddr_t start; 2119 vaddr_t end; 2120 vm_inherit_t new_inheritance; 2121 { 2122 struct vm_map_entry *entry, *temp_entry; 2123 UVMHIST_FUNC("uvm_map_inherit"); UVMHIST_CALLED(maphist); 2124 UVMHIST_LOG(maphist,"(map=0x%x,start=0x%x,end=0x%x,new_inh=0x%x)", 2125 map, start, end, new_inheritance); 2126 2127 switch (new_inheritance) { 2128 case MAP_INHERIT_NONE: 2129 case MAP_INHERIT_COPY: 2130 case MAP_INHERIT_SHARE: 2131 break; 2132 default: 2133 UVMHIST_LOG(maphist,"<- done (INVALID ARG)",0,0,0,0); 2134 return EINVAL; 2135 } 2136 2137 vm_map_lock(map); 2138 VM_MAP_RANGE_CHECK(map, start, end); 2139 if (uvm_map_lookup_entry(map, start, &temp_entry)) { 2140 entry = temp_entry; 2141 UVM_MAP_CLIP_START(map, entry, start); 2142 } else { 2143 entry = temp_entry->next; 2144 } 2145 while ((entry != &map->header) && (entry->start < end)) { 2146 UVM_MAP_CLIP_END(map, entry, end); 2147 entry->inheritance = new_inheritance; 2148 entry = entry->next; 2149 } 2150 vm_map_unlock(map); 2151 UVMHIST_LOG(maphist,"<- done (OK)",0,0,0,0); 2152 return 0; 2153 } 2154 2155 /* 2156 * uvm_map_advice: set advice code for range of addrs in map. 2157 * 2158 * => map must be unlocked 2159 */ 2160 2161 int 2162 uvm_map_advice(map, start, end, new_advice) 2163 struct vm_map *map; 2164 vaddr_t start; 2165 vaddr_t end; 2166 int new_advice; 2167 { 2168 struct vm_map_entry *entry, *temp_entry; 2169 UVMHIST_FUNC("uvm_map_advice"); UVMHIST_CALLED(maphist); 2170 UVMHIST_LOG(maphist,"(map=0x%x,start=0x%x,end=0x%x,new_adv=0x%x)", 2171 map, start, end, new_advice); 2172 2173 vm_map_lock(map); 2174 VM_MAP_RANGE_CHECK(map, start, end); 2175 if (uvm_map_lookup_entry(map, start, &temp_entry)) { 2176 entry = temp_entry; 2177 UVM_MAP_CLIP_START(map, entry, start); 2178 } else { 2179 entry = temp_entry->next; 2180 } 2181 2182 /* 2183 * XXXJRT: disallow holes? 2184 */ 2185 2186 while ((entry != &map->header) && (entry->start < end)) { 2187 UVM_MAP_CLIP_END(map, entry, end); 2188 2189 switch (new_advice) { 2190 case MADV_NORMAL: 2191 case MADV_RANDOM: 2192 case MADV_SEQUENTIAL: 2193 /* nothing special here */ 2194 break; 2195 2196 default: 2197 vm_map_unlock(map); 2198 UVMHIST_LOG(maphist,"<- done (INVALID ARG)",0,0,0,0); 2199 return EINVAL; 2200 } 2201 entry->advice = new_advice; 2202 entry = entry->next; 2203 } 2204 2205 vm_map_unlock(map); 2206 UVMHIST_LOG(maphist,"<- done (OK)",0,0,0,0); 2207 return 0; 2208 } 2209 2210 /* 2211 * uvm_map_pageable: sets the pageability of a range in a map. 2212 * 2213 * => wires map entries. should not be used for transient page locking. 2214 * for that, use uvm_fault_wire()/uvm_fault_unwire() (see uvm_vslock()). 2215 * => regions sepcified as not pageable require lock-down (wired) memory 2216 * and page tables. 2217 * => map must never be read-locked 2218 * => if islocked is TRUE, map is already write-locked 2219 * => we always unlock the map, since we must downgrade to a read-lock 2220 * to call uvm_fault_wire() 2221 * => XXXCDC: check this and try and clean it up. 2222 */ 2223 2224 int 2225 uvm_map_pageable(map, start, end, new_pageable, lockflags) 2226 struct vm_map *map; 2227 vaddr_t start, end; 2228 boolean_t new_pageable; 2229 int lockflags; 2230 { 2231 struct vm_map_entry *entry, *start_entry, *failed_entry; 2232 int rv; 2233 #ifdef DIAGNOSTIC 2234 u_int timestamp_save; 2235 #endif 2236 UVMHIST_FUNC("uvm_map_pageable"); UVMHIST_CALLED(maphist); 2237 UVMHIST_LOG(maphist,"(map=0x%x,start=0x%x,end=0x%x,new_pageable=0x%x)", 2238 map, start, end, new_pageable); 2239 KASSERT(map->flags & VM_MAP_PAGEABLE); 2240 2241 if ((lockflags & UVM_LK_ENTER) == 0) 2242 vm_map_lock(map); 2243 VM_MAP_RANGE_CHECK(map, start, end); 2244 2245 /* 2246 * only one pageability change may take place at one time, since 2247 * uvm_fault_wire assumes it will be called only once for each 2248 * wiring/unwiring. therefore, we have to make sure we're actually 2249 * changing the pageability for the entire region. we do so before 2250 * making any changes. 2251 */ 2252 2253 if (uvm_map_lookup_entry(map, start, &start_entry) == FALSE) { 2254 if ((lockflags & UVM_LK_EXIT) == 0) 2255 vm_map_unlock(map); 2256 2257 UVMHIST_LOG(maphist,"<- done (fault)",0,0,0,0); 2258 return EFAULT; 2259 } 2260 entry = start_entry; 2261 2262 /* 2263 * handle wiring and unwiring separately. 2264 */ 2265 2266 if (new_pageable) { /* unwire */ 2267 UVM_MAP_CLIP_START(map, entry, start); 2268 2269 /* 2270 * unwiring. first ensure that the range to be unwired is 2271 * really wired down and that there are no holes. 2272 */ 2273 2274 while ((entry != &map->header) && (entry->start < end)) { 2275 if (entry->wired_count == 0 || 2276 (entry->end < end && 2277 (entry->next == &map->header || 2278 entry->next->start > entry->end))) { 2279 if ((lockflags & UVM_LK_EXIT) == 0) 2280 vm_map_unlock(map); 2281 UVMHIST_LOG(maphist, "<- done (INVAL)",0,0,0,0); 2282 return EINVAL; 2283 } 2284 entry = entry->next; 2285 } 2286 2287 /* 2288 * POSIX 1003.1b - a single munlock call unlocks a region, 2289 * regardless of the number of mlock calls made on that 2290 * region. 2291 */ 2292 2293 entry = start_entry; 2294 while ((entry != &map->header) && (entry->start < end)) { 2295 UVM_MAP_CLIP_END(map, entry, end); 2296 if (VM_MAPENT_ISWIRED(entry)) 2297 uvm_map_entry_unwire(map, entry); 2298 entry = entry->next; 2299 } 2300 if ((lockflags & UVM_LK_EXIT) == 0) 2301 vm_map_unlock(map); 2302 UVMHIST_LOG(maphist,"<- done (OK UNWIRE)",0,0,0,0); 2303 return 0; 2304 } 2305 2306 /* 2307 * wire case: in two passes [XXXCDC: ugly block of code here] 2308 * 2309 * 1: holding the write lock, we create any anonymous maps that need 2310 * to be created. then we clip each map entry to the region to 2311 * be wired and increment its wiring count. 2312 * 2313 * 2: we downgrade to a read lock, and call uvm_fault_wire to fault 2314 * in the pages for any newly wired area (wired_count == 1). 2315 * 2316 * downgrading to a read lock for uvm_fault_wire avoids a possible 2317 * deadlock with another thread that may have faulted on one of 2318 * the pages to be wired (it would mark the page busy, blocking 2319 * us, then in turn block on the map lock that we hold). because 2320 * of problems in the recursive lock package, we cannot upgrade 2321 * to a write lock in vm_map_lookup. thus, any actions that 2322 * require the write lock must be done beforehand. because we 2323 * keep the read lock on the map, the copy-on-write status of the 2324 * entries we modify here cannot change. 2325 */ 2326 2327 while ((entry != &map->header) && (entry->start < end)) { 2328 if (VM_MAPENT_ISWIRED(entry) == 0) { /* not already wired? */ 2329 2330 /* 2331 * perform actions of vm_map_lookup that need the 2332 * write lock on the map: create an anonymous map 2333 * for a copy-on-write region, or an anonymous map 2334 * for a zero-fill region. (XXXCDC: submap case 2335 * ok?) 2336 */ 2337 2338 if (!UVM_ET_ISSUBMAP(entry)) { /* not submap */ 2339 if (UVM_ET_ISNEEDSCOPY(entry) && 2340 ((entry->max_protection & VM_PROT_WRITE) || 2341 (entry->object.uvm_obj == NULL))) { 2342 amap_copy(map, entry, M_WAITOK, TRUE, 2343 start, end); 2344 /* XXXCDC: wait OK? */ 2345 } 2346 } 2347 } 2348 UVM_MAP_CLIP_START(map, entry, start); 2349 UVM_MAP_CLIP_END(map, entry, end); 2350 entry->wired_count++; 2351 2352 /* 2353 * Check for holes 2354 */ 2355 2356 if (entry->protection == VM_PROT_NONE || 2357 (entry->end < end && 2358 (entry->next == &map->header || 2359 entry->next->start > entry->end))) { 2360 2361 /* 2362 * found one. amap creation actions do not need to 2363 * be undone, but the wired counts need to be restored. 2364 */ 2365 2366 while (entry != &map->header && entry->end > start) { 2367 entry->wired_count--; 2368 entry = entry->prev; 2369 } 2370 if ((lockflags & UVM_LK_EXIT) == 0) 2371 vm_map_unlock(map); 2372 UVMHIST_LOG(maphist,"<- done (INVALID WIRE)",0,0,0,0); 2373 return EINVAL; 2374 } 2375 entry = entry->next; 2376 } 2377 2378 /* 2379 * Pass 2. 2380 */ 2381 2382 #ifdef DIAGNOSTIC 2383 timestamp_save = map->timestamp; 2384 #endif 2385 vm_map_busy(map); 2386 vm_map_downgrade(map); 2387 2388 rv = 0; 2389 entry = start_entry; 2390 while (entry != &map->header && entry->start < end) { 2391 if (entry->wired_count == 1) { 2392 rv = uvm_fault_wire(map, entry->start, entry->end, 2393 VM_FAULT_WIREMAX, entry->max_protection); 2394 if (rv) { 2395 2396 /* 2397 * wiring failed. break out of the loop. 2398 * we'll clean up the map below, once we 2399 * have a write lock again. 2400 */ 2401 2402 break; 2403 } 2404 } 2405 entry = entry->next; 2406 } 2407 2408 if (rv) { /* failed? */ 2409 2410 /* 2411 * Get back to an exclusive (write) lock. 2412 */ 2413 2414 vm_map_upgrade(map); 2415 vm_map_unbusy(map); 2416 2417 #ifdef DIAGNOSTIC 2418 if (timestamp_save != map->timestamp) 2419 panic("uvm_map_pageable: stale map"); 2420 #endif 2421 2422 /* 2423 * first drop the wiring count on all the entries 2424 * which haven't actually been wired yet. 2425 */ 2426 2427 failed_entry = entry; 2428 while (entry != &map->header && entry->start < end) { 2429 entry->wired_count--; 2430 entry = entry->next; 2431 } 2432 2433 /* 2434 * now, unwire all the entries that were successfully 2435 * wired above. 2436 */ 2437 2438 entry = start_entry; 2439 while (entry != failed_entry) { 2440 entry->wired_count--; 2441 if (VM_MAPENT_ISWIRED(entry) == 0) 2442 uvm_map_entry_unwire(map, entry); 2443 entry = entry->next; 2444 } 2445 if ((lockflags & UVM_LK_EXIT) == 0) 2446 vm_map_unlock(map); 2447 UVMHIST_LOG(maphist, "<- done (RV=%d)", rv,0,0,0); 2448 return(rv); 2449 } 2450 2451 /* We are holding a read lock here. */ 2452 if ((lockflags & UVM_LK_EXIT) == 0) { 2453 vm_map_unbusy(map); 2454 vm_map_unlock_read(map); 2455 } else { 2456 2457 /* 2458 * Get back to an exclusive (write) lock. 2459 */ 2460 2461 vm_map_upgrade(map); 2462 vm_map_unbusy(map); 2463 } 2464 2465 UVMHIST_LOG(maphist,"<- done (OK WIRE)",0,0,0,0); 2466 return 0; 2467 } 2468 2469 /* 2470 * uvm_map_pageable_all: special case of uvm_map_pageable - affects 2471 * all mapped regions. 2472 * 2473 * => map must not be locked. 2474 * => if no flags are specified, all regions are unwired. 2475 * => XXXJRT: has some of the same problems as uvm_map_pageable() above. 2476 */ 2477 2478 int 2479 uvm_map_pageable_all(map, flags, limit) 2480 struct vm_map *map; 2481 int flags; 2482 vsize_t limit; 2483 { 2484 struct vm_map_entry *entry, *failed_entry; 2485 vsize_t size; 2486 int rv; 2487 #ifdef DIAGNOSTIC 2488 u_int timestamp_save; 2489 #endif 2490 UVMHIST_FUNC("uvm_map_pageable_all"); UVMHIST_CALLED(maphist); 2491 UVMHIST_LOG(maphist,"(map=0x%x,flags=0x%x)", map, flags, 0, 0); 2492 2493 KASSERT(map->flags & VM_MAP_PAGEABLE); 2494 2495 vm_map_lock(map); 2496 2497 /* 2498 * handle wiring and unwiring separately. 2499 */ 2500 2501 if (flags == 0) { /* unwire */ 2502 2503 /* 2504 * POSIX 1003.1b -- munlockall unlocks all regions, 2505 * regardless of how many times mlockall has been called. 2506 */ 2507 2508 for (entry = map->header.next; entry != &map->header; 2509 entry = entry->next) { 2510 if (VM_MAPENT_ISWIRED(entry)) 2511 uvm_map_entry_unwire(map, entry); 2512 } 2513 vm_map_modflags(map, 0, VM_MAP_WIREFUTURE); 2514 vm_map_unlock(map); 2515 UVMHIST_LOG(maphist,"<- done (OK UNWIRE)",0,0,0,0); 2516 return 0; 2517 } 2518 2519 if (flags & MCL_FUTURE) { 2520 2521 /* 2522 * must wire all future mappings; remember this. 2523 */ 2524 2525 vm_map_modflags(map, VM_MAP_WIREFUTURE, 0); 2526 } 2527 2528 if ((flags & MCL_CURRENT) == 0) { 2529 2530 /* 2531 * no more work to do! 2532 */ 2533 2534 UVMHIST_LOG(maphist,"<- done (OK no wire)",0,0,0,0); 2535 vm_map_unlock(map); 2536 return 0; 2537 } 2538 2539 /* 2540 * wire case: in three passes [XXXCDC: ugly block of code here] 2541 * 2542 * 1: holding the write lock, count all pages mapped by non-wired 2543 * entries. if this would cause us to go over our limit, we fail. 2544 * 2545 * 2: still holding the write lock, we create any anonymous maps that 2546 * need to be created. then we increment its wiring count. 2547 * 2548 * 3: we downgrade to a read lock, and call uvm_fault_wire to fault 2549 * in the pages for any newly wired area (wired_count == 1). 2550 * 2551 * downgrading to a read lock for uvm_fault_wire avoids a possible 2552 * deadlock with another thread that may have faulted on one of 2553 * the pages to be wired (it would mark the page busy, blocking 2554 * us, then in turn block on the map lock that we hold). because 2555 * of problems in the recursive lock package, we cannot upgrade 2556 * to a write lock in vm_map_lookup. thus, any actions that 2557 * require the write lock must be done beforehand. because we 2558 * keep the read lock on the map, the copy-on-write status of the 2559 * entries we modify here cannot change. 2560 */ 2561 2562 for (size = 0, entry = map->header.next; entry != &map->header; 2563 entry = entry->next) { 2564 if (entry->protection != VM_PROT_NONE && 2565 VM_MAPENT_ISWIRED(entry) == 0) { /* not already wired? */ 2566 size += entry->end - entry->start; 2567 } 2568 } 2569 2570 if (atop(size) + uvmexp.wired > uvmexp.wiredmax) { 2571 vm_map_unlock(map); 2572 return ENOMEM; 2573 } 2574 2575 /* XXX non-pmap_wired_count case must be handled by caller */ 2576 #ifdef pmap_wired_count 2577 if (limit != 0 && 2578 (size + ptoa(pmap_wired_count(vm_map_pmap(map))) > limit)) { 2579 vm_map_unlock(map); 2580 return ENOMEM; 2581 } 2582 #endif 2583 2584 /* 2585 * Pass 2. 2586 */ 2587 2588 for (entry = map->header.next; entry != &map->header; 2589 entry = entry->next) { 2590 if (entry->protection == VM_PROT_NONE) 2591 continue; 2592 if (VM_MAPENT_ISWIRED(entry) == 0) { /* not already wired? */ 2593 2594 /* 2595 * perform actions of vm_map_lookup that need the 2596 * write lock on the map: create an anonymous map 2597 * for a copy-on-write region, or an anonymous map 2598 * for a zero-fill region. (XXXCDC: submap case 2599 * ok?) 2600 */ 2601 2602 if (!UVM_ET_ISSUBMAP(entry)) { /* not submap */ 2603 if (UVM_ET_ISNEEDSCOPY(entry) && 2604 ((entry->max_protection & VM_PROT_WRITE) || 2605 (entry->object.uvm_obj == NULL))) { 2606 amap_copy(map, entry, M_WAITOK, TRUE, 2607 entry->start, entry->end); 2608 /* XXXCDC: wait OK? */ 2609 } 2610 } 2611 } 2612 entry->wired_count++; 2613 } 2614 2615 /* 2616 * Pass 3. 2617 */ 2618 2619 #ifdef DIAGNOSTIC 2620 timestamp_save = map->timestamp; 2621 #endif 2622 vm_map_busy(map); 2623 vm_map_downgrade(map); 2624 2625 rv = 0; 2626 for (entry = map->header.next; entry != &map->header; 2627 entry = entry->next) { 2628 if (entry->wired_count == 1) { 2629 rv = uvm_fault_wire(map, entry->start, entry->end, 2630 VM_FAULT_WIREMAX, entry->max_protection); 2631 if (rv) { 2632 2633 /* 2634 * wiring failed. break out of the loop. 2635 * we'll clean up the map below, once we 2636 * have a write lock again. 2637 */ 2638 2639 break; 2640 } 2641 } 2642 } 2643 2644 if (rv) { 2645 2646 /* 2647 * Get back an exclusive (write) lock. 2648 */ 2649 2650 vm_map_upgrade(map); 2651 vm_map_unbusy(map); 2652 2653 #ifdef DIAGNOSTIC 2654 if (timestamp_save != map->timestamp) 2655 panic("uvm_map_pageable_all: stale map"); 2656 #endif 2657 2658 /* 2659 * first drop the wiring count on all the entries 2660 * which haven't actually been wired yet. 2661 * 2662 * Skip VM_PROT_NONE entries like we did above. 2663 */ 2664 2665 failed_entry = entry; 2666 for (/* nothing */; entry != &map->header; 2667 entry = entry->next) { 2668 if (entry->protection == VM_PROT_NONE) 2669 continue; 2670 entry->wired_count--; 2671 } 2672 2673 /* 2674 * now, unwire all the entries that were successfully 2675 * wired above. 2676 * 2677 * Skip VM_PROT_NONE entries like we did above. 2678 */ 2679 2680 for (entry = map->header.next; entry != failed_entry; 2681 entry = entry->next) { 2682 if (entry->protection == VM_PROT_NONE) 2683 continue; 2684 entry->wired_count--; 2685 if (VM_MAPENT_ISWIRED(entry)) 2686 uvm_map_entry_unwire(map, entry); 2687 } 2688 vm_map_unlock(map); 2689 UVMHIST_LOG(maphist,"<- done (RV=%d)", rv,0,0,0); 2690 return (rv); 2691 } 2692 2693 /* We are holding a read lock here. */ 2694 vm_map_unbusy(map); 2695 vm_map_unlock_read(map); 2696 2697 UVMHIST_LOG(maphist,"<- done (OK WIRE)",0,0,0,0); 2698 return 0; 2699 } 2700 2701 /* 2702 * uvm_map_clean: clean out a map range 2703 * 2704 * => valid flags: 2705 * if (flags & PGO_CLEANIT): dirty pages are cleaned first 2706 * if (flags & PGO_SYNCIO): dirty pages are written synchronously 2707 * if (flags & PGO_DEACTIVATE): any cached pages are deactivated after clean 2708 * if (flags & PGO_FREE): any cached pages are freed after clean 2709 * => returns an error if any part of the specified range isn't mapped 2710 * => never a need to flush amap layer since the anonymous memory has 2711 * no permanent home, but may deactivate pages there 2712 * => called from sys_msync() and sys_madvise() 2713 * => caller must not write-lock map (read OK). 2714 * => we may sleep while cleaning if SYNCIO [with map read-locked] 2715 */ 2716 2717 int 2718 uvm_map_clean(map, start, end, flags) 2719 struct vm_map *map; 2720 vaddr_t start, end; 2721 int flags; 2722 { 2723 struct vm_map_entry *current, *entry; 2724 struct uvm_object *uobj; 2725 struct vm_amap *amap; 2726 struct vm_anon *anon; 2727 struct vm_page *pg; 2728 vaddr_t offset; 2729 vsize_t size; 2730 int error, refs; 2731 UVMHIST_FUNC("uvm_map_clean"); UVMHIST_CALLED(maphist); 2732 2733 UVMHIST_LOG(maphist,"(map=0x%x,start=0x%x,end=0x%x,flags=0x%x)", 2734 map, start, end, flags); 2735 KASSERT((flags & (PGO_FREE|PGO_DEACTIVATE)) != 2736 (PGO_FREE|PGO_DEACTIVATE)); 2737 2738 vm_map_lock_read(map); 2739 VM_MAP_RANGE_CHECK(map, start, end); 2740 if (uvm_map_lookup_entry(map, start, &entry) == FALSE) { 2741 vm_map_unlock_read(map); 2742 return EFAULT; 2743 } 2744 2745 /* 2746 * Make a first pass to check for holes. 2747 */ 2748 2749 for (current = entry; current->start < end; current = current->next) { 2750 if (UVM_ET_ISSUBMAP(current)) { 2751 vm_map_unlock_read(map); 2752 return EINVAL; 2753 } 2754 if (end <= current->end) { 2755 break; 2756 } 2757 if (current->end != current->next->start) { 2758 vm_map_unlock_read(map); 2759 return EFAULT; 2760 } 2761 } 2762 2763 error = 0; 2764 for (current = entry; start < end; current = current->next) { 2765 amap = current->aref.ar_amap; /* top layer */ 2766 uobj = current->object.uvm_obj; /* bottom layer */ 2767 KASSERT(start >= current->start); 2768 2769 /* 2770 * No amap cleaning necessary if: 2771 * 2772 * (1) There's no amap. 2773 * 2774 * (2) We're not deactivating or freeing pages. 2775 */ 2776 2777 if (amap == NULL || (flags & (PGO_DEACTIVATE|PGO_FREE)) == 0) 2778 goto flush_object; 2779 2780 amap_lock(amap); 2781 offset = start - current->start; 2782 size = MIN(end, current->end) - start; 2783 for ( ; size != 0; size -= PAGE_SIZE, offset += PAGE_SIZE) { 2784 anon = amap_lookup(¤t->aref, offset); 2785 if (anon == NULL) 2786 continue; 2787 2788 simple_lock(&anon->an_lock); 2789 pg = anon->u.an_page; 2790 if (pg == NULL) { 2791 simple_unlock(&anon->an_lock); 2792 continue; 2793 } 2794 2795 switch (flags & (PGO_CLEANIT|PGO_FREE|PGO_DEACTIVATE)) { 2796 2797 /* 2798 * In these first 3 cases, we just deactivate the page. 2799 */ 2800 2801 case PGO_CLEANIT|PGO_FREE: 2802 case PGO_CLEANIT|PGO_DEACTIVATE: 2803 case PGO_DEACTIVATE: 2804 deactivate_it: 2805 /* 2806 * skip the page if it's loaned or wired, 2807 * since it shouldn't be on a paging queue 2808 * at all in these cases. 2809 */ 2810 2811 uvm_lock_pageq(); 2812 if (pg->loan_count != 0 || 2813 pg->wire_count != 0) { 2814 uvm_unlock_pageq(); 2815 simple_unlock(&anon->an_lock); 2816 continue; 2817 } 2818 KASSERT(pg->uanon == anon); 2819 pmap_clear_reference(pg); 2820 uvm_pagedeactivate(pg); 2821 uvm_unlock_pageq(); 2822 simple_unlock(&anon->an_lock); 2823 continue; 2824 2825 case PGO_FREE: 2826 2827 /* 2828 * If there are multiple references to 2829 * the amap, just deactivate the page. 2830 */ 2831 2832 if (amap_refs(amap) > 1) 2833 goto deactivate_it; 2834 2835 /* skip the page if it's wired */ 2836 if (pg->wire_count != 0) { 2837 simple_unlock(&anon->an_lock); 2838 continue; 2839 } 2840 amap_unadd(¤t->aref, offset); 2841 refs = --anon->an_ref; 2842 simple_unlock(&anon->an_lock); 2843 if (refs == 0) 2844 uvm_anfree(anon); 2845 continue; 2846 } 2847 } 2848 amap_unlock(amap); 2849 2850 flush_object: 2851 /* 2852 * flush pages if we've got a valid backing object. 2853 * note that we must always clean object pages before 2854 * freeing them since otherwise we could reveal stale 2855 * data from files. 2856 */ 2857 2858 offset = current->offset + (start - current->start); 2859 size = MIN(end, current->end) - start; 2860 if (uobj != NULL) { 2861 simple_lock(&uobj->vmobjlock); 2862 error = (uobj->pgops->pgo_put)(uobj, offset, 2863 offset + size, flags | PGO_CLEANIT); 2864 } 2865 start += size; 2866 } 2867 vm_map_unlock_read(map); 2868 return (error); 2869 } 2870 2871 2872 /* 2873 * uvm_map_checkprot: check protection in map 2874 * 2875 * => must allow specified protection in a fully allocated region. 2876 * => map must be read or write locked by caller. 2877 */ 2878 2879 boolean_t 2880 uvm_map_checkprot(map, start, end, protection) 2881 struct vm_map * map; 2882 vaddr_t start, end; 2883 vm_prot_t protection; 2884 { 2885 struct vm_map_entry *entry; 2886 struct vm_map_entry *tmp_entry; 2887 2888 if (!uvm_map_lookup_entry(map, start, &tmp_entry)) { 2889 return(FALSE); 2890 } 2891 entry = tmp_entry; 2892 while (start < end) { 2893 if (entry == &map->header) { 2894 return(FALSE); 2895 } 2896 2897 /* 2898 * no holes allowed 2899 */ 2900 2901 if (start < entry->start) { 2902 return(FALSE); 2903 } 2904 2905 /* 2906 * check protection associated with entry 2907 */ 2908 2909 if ((entry->protection & protection) != protection) { 2910 return(FALSE); 2911 } 2912 start = entry->end; 2913 entry = entry->next; 2914 } 2915 return(TRUE); 2916 } 2917 2918 /* 2919 * uvmspace_alloc: allocate a vmspace structure. 2920 * 2921 * - structure includes vm_map and pmap 2922 * - XXX: no locking on this structure 2923 * - refcnt set to 1, rest must be init'd by caller 2924 */ 2925 struct vmspace * 2926 uvmspace_alloc(min, max) 2927 vaddr_t min, max; 2928 { 2929 struct vmspace *vm; 2930 UVMHIST_FUNC("uvmspace_alloc"); UVMHIST_CALLED(maphist); 2931 2932 vm = pool_get(&uvm_vmspace_pool, PR_WAITOK); 2933 uvmspace_init(vm, NULL, min, max); 2934 UVMHIST_LOG(maphist,"<- done (vm=0x%x)", vm,0,0,0); 2935 return (vm); 2936 } 2937 2938 /* 2939 * uvmspace_init: initialize a vmspace structure. 2940 * 2941 * - XXX: no locking on this structure 2942 * - refcnt set to 1, rest must me init'd by caller 2943 */ 2944 void 2945 uvmspace_init(vm, pmap, min, max) 2946 struct vmspace *vm; 2947 struct pmap *pmap; 2948 vaddr_t min, max; 2949 { 2950 UVMHIST_FUNC("uvmspace_init"); UVMHIST_CALLED(maphist); 2951 2952 memset(vm, 0, sizeof(*vm)); 2953 uvm_map_setup(&vm->vm_map, min, max, VM_MAP_PAGEABLE); 2954 if (pmap) 2955 pmap_reference(pmap); 2956 else 2957 pmap = pmap_create(); 2958 vm->vm_map.pmap = pmap; 2959 vm->vm_refcnt = 1; 2960 UVMHIST_LOG(maphist,"<- done",0,0,0,0); 2961 } 2962 2963 /* 2964 * uvmspace_share: share a vmspace between two proceses 2965 * 2966 * - XXX: no locking on vmspace 2967 * - used for vfork, threads(?) 2968 */ 2969 2970 void 2971 uvmspace_share(p1, p2) 2972 struct proc *p1, *p2; 2973 { 2974 p2->p_vmspace = p1->p_vmspace; 2975 p1->p_vmspace->vm_refcnt++; 2976 } 2977 2978 /* 2979 * uvmspace_unshare: ensure that process "p" has its own, unshared, vmspace 2980 * 2981 * - XXX: no locking on vmspace 2982 */ 2983 2984 void 2985 uvmspace_unshare(p) 2986 struct proc *p; 2987 { 2988 struct vmspace *nvm, *ovm = p->p_vmspace; 2989 2990 if (ovm->vm_refcnt == 1) 2991 /* nothing to do: vmspace isn't shared in the first place */ 2992 return; 2993 2994 /* make a new vmspace, still holding old one */ 2995 nvm = uvmspace_fork(ovm); 2996 2997 pmap_deactivate(p); /* unbind old vmspace */ 2998 p->p_vmspace = nvm; 2999 pmap_activate(p); /* switch to new vmspace */ 3000 3001 uvmspace_free(ovm); /* drop reference to old vmspace */ 3002 } 3003 3004 /* 3005 * uvmspace_exec: the process wants to exec a new program 3006 * 3007 * - XXX: no locking on vmspace 3008 */ 3009 3010 void 3011 uvmspace_exec(p, start, end) 3012 struct proc *p; 3013 vaddr_t start, end; 3014 { 3015 struct vmspace *nvm, *ovm = p->p_vmspace; 3016 struct vm_map *map = &ovm->vm_map; 3017 3018 #ifdef __sparc__ 3019 /* XXX cgd 960926: the sparc #ifdef should be a MD hook */ 3020 kill_user_windows(p); /* before stack addresses go away */ 3021 #endif 3022 3023 /* 3024 * see if more than one process is using this vmspace... 3025 */ 3026 3027 if (ovm->vm_refcnt == 1) { 3028 3029 /* 3030 * if p is the only process using its vmspace then we can safely 3031 * recycle that vmspace for the program that is being exec'd. 3032 */ 3033 3034 #ifdef SYSVSHM 3035 /* 3036 * SYSV SHM semantics require us to kill all segments on an exec 3037 */ 3038 3039 if (ovm->vm_shm) 3040 shmexit(ovm); 3041 #endif 3042 3043 /* 3044 * POSIX 1003.1b -- "lock future mappings" is revoked 3045 * when a process execs another program image. 3046 */ 3047 3048 vm_map_modflags(map, 0, VM_MAP_WIREFUTURE); 3049 3050 /* 3051 * now unmap the old program 3052 */ 3053 3054 pmap_remove_all(map->pmap); 3055 uvm_unmap(map, map->min_offset, map->max_offset); 3056 3057 /* 3058 * resize the map 3059 */ 3060 3061 map->min_offset = start; 3062 map->max_offset = end; 3063 } else { 3064 3065 /* 3066 * p's vmspace is being shared, so we can't reuse it for p since 3067 * it is still being used for others. allocate a new vmspace 3068 * for p 3069 */ 3070 3071 nvm = uvmspace_alloc(start, end); 3072 3073 /* 3074 * install new vmspace and drop our ref to the old one. 3075 */ 3076 3077 pmap_deactivate(p); 3078 p->p_vmspace = nvm; 3079 pmap_activate(p); 3080 3081 uvmspace_free(ovm); 3082 } 3083 } 3084 3085 /* 3086 * uvmspace_free: free a vmspace data structure 3087 * 3088 * - XXX: no locking on vmspace 3089 */ 3090 3091 void 3092 uvmspace_free(vm) 3093 struct vmspace *vm; 3094 { 3095 struct vm_map_entry *dead_entries; 3096 struct vm_map *map; 3097 UVMHIST_FUNC("uvmspace_free"); UVMHIST_CALLED(maphist); 3098 3099 UVMHIST_LOG(maphist,"(vm=0x%x) ref=%d", vm, vm->vm_refcnt,0,0); 3100 if (--vm->vm_refcnt > 0) { 3101 return; 3102 } 3103 3104 /* 3105 * at this point, there should be no other references to the map. 3106 * delete all of the mappings, then destroy the pmap. 3107 */ 3108 3109 map = &vm->vm_map; 3110 map->flags |= VM_MAP_DYING; 3111 pmap_remove_all(map->pmap); 3112 #ifdef SYSVSHM 3113 /* Get rid of any SYSV shared memory segments. */ 3114 if (vm->vm_shm != NULL) 3115 shmexit(vm); 3116 #endif 3117 if (map->nentries) { 3118 uvm_unmap_remove(map, map->min_offset, map->max_offset, 3119 &dead_entries); 3120 if (dead_entries != NULL) 3121 uvm_unmap_detach(dead_entries, 0); 3122 } 3123 pmap_destroy(map->pmap); 3124 pool_put(&uvm_vmspace_pool, vm); 3125 } 3126 3127 /* 3128 * F O R K - m a i n e n t r y p o i n t 3129 */ 3130 /* 3131 * uvmspace_fork: fork a process' main map 3132 * 3133 * => create a new vmspace for child process from parent. 3134 * => parent's map must not be locked. 3135 */ 3136 3137 struct vmspace * 3138 uvmspace_fork(vm1) 3139 struct vmspace *vm1; 3140 { 3141 struct vmspace *vm2; 3142 struct vm_map *old_map = &vm1->vm_map; 3143 struct vm_map *new_map; 3144 struct vm_map_entry *old_entry; 3145 struct vm_map_entry *new_entry; 3146 pmap_t new_pmap; 3147 UVMHIST_FUNC("uvmspace_fork"); UVMHIST_CALLED(maphist); 3148 3149 vm_map_lock(old_map); 3150 3151 vm2 = uvmspace_alloc(old_map->min_offset, old_map->max_offset); 3152 memcpy(&vm2->vm_startcopy, &vm1->vm_startcopy, 3153 (caddr_t) (vm1 + 1) - (caddr_t) &vm1->vm_startcopy); 3154 new_map = &vm2->vm_map; /* XXX */ 3155 new_pmap = new_map->pmap; 3156 3157 old_entry = old_map->header.next; 3158 3159 /* 3160 * go entry-by-entry 3161 */ 3162 3163 while (old_entry != &old_map->header) { 3164 3165 /* 3166 * first, some sanity checks on the old entry 3167 */ 3168 3169 KASSERT(!UVM_ET_ISSUBMAP(old_entry)); 3170 KASSERT(UVM_ET_ISCOPYONWRITE(old_entry) || 3171 !UVM_ET_ISNEEDSCOPY(old_entry)); 3172 3173 switch (old_entry->inheritance) { 3174 case MAP_INHERIT_NONE: 3175 3176 /* 3177 * drop the mapping 3178 */ 3179 3180 break; 3181 3182 case MAP_INHERIT_SHARE: 3183 3184 /* 3185 * share the mapping: this means we want the old and 3186 * new entries to share amaps and backing objects. 3187 */ 3188 /* 3189 * if the old_entry needs a new amap (due to prev fork) 3190 * then we need to allocate it now so that we have 3191 * something we own to share with the new_entry. [in 3192 * other words, we need to clear needs_copy] 3193 */ 3194 3195 if (UVM_ET_ISNEEDSCOPY(old_entry)) { 3196 /* get our own amap, clears needs_copy */ 3197 amap_copy(old_map, old_entry, M_WAITOK, FALSE, 3198 0, 0); 3199 /* XXXCDC: WAITOK??? */ 3200 } 3201 3202 new_entry = uvm_mapent_alloc(new_map, 0); 3203 /* old_entry -> new_entry */ 3204 uvm_mapent_copy(old_entry, new_entry); 3205 3206 /* new pmap has nothing wired in it */ 3207 new_entry->wired_count = 0; 3208 3209 /* 3210 * gain reference to object backing the map (can't 3211 * be a submap, already checked this case). 3212 */ 3213 3214 if (new_entry->aref.ar_amap) 3215 uvm_map_reference_amap(new_entry, AMAP_SHARED); 3216 3217 if (new_entry->object.uvm_obj && 3218 new_entry->object.uvm_obj->pgops->pgo_reference) 3219 new_entry->object.uvm_obj-> 3220 pgops->pgo_reference( 3221 new_entry->object.uvm_obj); 3222 3223 /* insert entry at end of new_map's entry list */ 3224 uvm_map_entry_link(new_map, new_map->header.prev, 3225 new_entry); 3226 3227 break; 3228 3229 case MAP_INHERIT_COPY: 3230 3231 /* 3232 * copy-on-write the mapping (using mmap's 3233 * MAP_PRIVATE semantics) 3234 * 3235 * allocate new_entry, adjust reference counts. 3236 * (note that new references are read-only). 3237 */ 3238 3239 new_entry = uvm_mapent_alloc(new_map, 0); 3240 /* old_entry -> new_entry */ 3241 uvm_mapent_copy(old_entry, new_entry); 3242 3243 if (new_entry->aref.ar_amap) 3244 uvm_map_reference_amap(new_entry, 0); 3245 3246 if (new_entry->object.uvm_obj && 3247 new_entry->object.uvm_obj->pgops->pgo_reference) 3248 new_entry->object.uvm_obj->pgops->pgo_reference 3249 (new_entry->object.uvm_obj); 3250 3251 /* new pmap has nothing wired in it */ 3252 new_entry->wired_count = 0; 3253 3254 new_entry->etype |= 3255 (UVM_ET_COPYONWRITE|UVM_ET_NEEDSCOPY); 3256 uvm_map_entry_link(new_map, new_map->header.prev, 3257 new_entry); 3258 3259 /* 3260 * the new entry will need an amap. it will either 3261 * need to be copied from the old entry or created 3262 * from scratch (if the old entry does not have an 3263 * amap). can we defer this process until later 3264 * (by setting "needs_copy") or do we need to copy 3265 * the amap now? 3266 * 3267 * we must copy the amap now if any of the following 3268 * conditions hold: 3269 * 1. the old entry has an amap and that amap is 3270 * being shared. this means that the old (parent) 3271 * process is sharing the amap with another 3272 * process. if we do not clear needs_copy here 3273 * we will end up in a situation where both the 3274 * parent and child process are refering to the 3275 * same amap with "needs_copy" set. if the 3276 * parent write-faults, the fault routine will 3277 * clear "needs_copy" in the parent by allocating 3278 * a new amap. this is wrong because the 3279 * parent is supposed to be sharing the old amap 3280 * and the new amap will break that. 3281 * 3282 * 2. if the old entry has an amap and a non-zero 3283 * wire count then we are going to have to call 3284 * amap_cow_now to avoid page faults in the 3285 * parent process. since amap_cow_now requires 3286 * "needs_copy" to be clear we might as well 3287 * clear it here as well. 3288 * 3289 */ 3290 3291 if (old_entry->aref.ar_amap != NULL) { 3292 if ((amap_flags(old_entry->aref.ar_amap) & 3293 AMAP_SHARED) != 0 || 3294 VM_MAPENT_ISWIRED(old_entry)) { 3295 3296 amap_copy(new_map, new_entry, M_WAITOK, 3297 FALSE, 0, 0); 3298 /* XXXCDC: M_WAITOK ... ok? */ 3299 } 3300 } 3301 3302 /* 3303 * if the parent's entry is wired down, then the 3304 * parent process does not want page faults on 3305 * access to that memory. this means that we 3306 * cannot do copy-on-write because we can't write 3307 * protect the old entry. in this case we 3308 * resolve all copy-on-write faults now, using 3309 * amap_cow_now. note that we have already 3310 * allocated any needed amap (above). 3311 */ 3312 3313 if (VM_MAPENT_ISWIRED(old_entry)) { 3314 3315 /* 3316 * resolve all copy-on-write faults now 3317 * (note that there is nothing to do if 3318 * the old mapping does not have an amap). 3319 */ 3320 if (old_entry->aref.ar_amap) 3321 amap_cow_now(new_map, new_entry); 3322 3323 } else { 3324 3325 /* 3326 * setup mappings to trigger copy-on-write faults 3327 * we must write-protect the parent if it has 3328 * an amap and it is not already "needs_copy"... 3329 * if it is already "needs_copy" then the parent 3330 * has already been write-protected by a previous 3331 * fork operation. 3332 */ 3333 3334 if (old_entry->aref.ar_amap && 3335 !UVM_ET_ISNEEDSCOPY(old_entry)) { 3336 if (old_entry->max_protection & VM_PROT_WRITE) { 3337 pmap_protect(old_map->pmap, 3338 old_entry->start, 3339 old_entry->end, 3340 old_entry->protection & 3341 ~VM_PROT_WRITE); 3342 pmap_update(old_map->pmap); 3343 } 3344 old_entry->etype |= UVM_ET_NEEDSCOPY; 3345 } 3346 } 3347 break; 3348 } /* end of switch statement */ 3349 old_entry = old_entry->next; 3350 } 3351 3352 new_map->size = old_map->size; 3353 vm_map_unlock(old_map); 3354 3355 #ifdef SYSVSHM 3356 if (vm1->vm_shm) 3357 shmfork(vm1, vm2); 3358 #endif 3359 3360 #ifdef PMAP_FORK 3361 pmap_fork(vm1->vm_map.pmap, vm2->vm_map.pmap); 3362 #endif 3363 3364 UVMHIST_LOG(maphist,"<- done",0,0,0,0); 3365 return(vm2); 3366 } 3367 3368 3369 #if defined(DDB) 3370 3371 /* 3372 * DDB hooks 3373 */ 3374 3375 /* 3376 * uvm_map_printit: actually prints the map 3377 */ 3378 3379 void 3380 uvm_map_printit(map, full, pr) 3381 struct vm_map *map; 3382 boolean_t full; 3383 void (*pr) __P((const char *, ...)); 3384 { 3385 struct vm_map_entry *entry; 3386 3387 (*pr)("MAP %p: [0x%lx->0x%lx]\n", map, map->min_offset,map->max_offset); 3388 (*pr)("\t#ent=%d, sz=%d, ref=%d, version=%d, flags=0x%x\n", 3389 map->nentries, map->size, map->ref_count, map->timestamp, 3390 map->flags); 3391 (*pr)("\tpmap=%p(resident=%d)\n", map->pmap, 3392 pmap_resident_count(map->pmap)); 3393 if (!full) 3394 return; 3395 for (entry = map->header.next; entry != &map->header; 3396 entry = entry->next) { 3397 (*pr)(" - %p: 0x%lx->0x%lx: obj=%p/0x%llx, amap=%p/%d\n", 3398 entry, entry->start, entry->end, entry->object.uvm_obj, 3399 (long long)entry->offset, entry->aref.ar_amap, 3400 entry->aref.ar_pageoff); 3401 (*pr)( 3402 "\tsubmap=%c, cow=%c, nc=%c, prot(max)=%d/%d, inh=%d, " 3403 "wc=%d, adv=%d\n", 3404 (entry->etype & UVM_ET_SUBMAP) ? 'T' : 'F', 3405 (entry->etype & UVM_ET_COPYONWRITE) ? 'T' : 'F', 3406 (entry->etype & UVM_ET_NEEDSCOPY) ? 'T' : 'F', 3407 entry->protection, entry->max_protection, 3408 entry->inheritance, entry->wired_count, entry->advice); 3409 } 3410 } 3411 3412 /* 3413 * uvm_object_printit: actually prints the object 3414 */ 3415 3416 void 3417 uvm_object_printit(uobj, full, pr) 3418 struct uvm_object *uobj; 3419 boolean_t full; 3420 void (*pr) __P((const char *, ...)); 3421 { 3422 struct vm_page *pg; 3423 int cnt = 0; 3424 3425 (*pr)("OBJECT %p: locked=%d, pgops=%p, npages=%d, ", 3426 uobj, uobj->vmobjlock.lock_data, uobj->pgops, uobj->uo_npages); 3427 if (UVM_OBJ_IS_KERN_OBJECT(uobj)) 3428 (*pr)("refs=<SYSTEM>\n"); 3429 else 3430 (*pr)("refs=%d\n", uobj->uo_refs); 3431 3432 if (!full) { 3433 return; 3434 } 3435 (*pr)(" PAGES <pg,offset>:\n "); 3436 TAILQ_FOREACH(pg, &uobj->memq, listq) { 3437 cnt++; 3438 (*pr)("<%p,0x%llx> ", pg, (long long)pg->offset); 3439 if ((cnt % 3) == 0) { 3440 (*pr)("\n "); 3441 } 3442 } 3443 if ((cnt % 3) != 0) { 3444 (*pr)("\n"); 3445 } 3446 } 3447 3448 /* 3449 * uvm_page_printit: actually print the page 3450 */ 3451 3452 static const char page_flagbits[] = 3453 "\20\1BUSY\2WANTED\3TABLED\4CLEAN\5PAGEOUT\6RELEASED\7FAKE\10RDONLY" 3454 "\11ZERO\15PAGER1"; 3455 static const char page_pqflagbits[] = 3456 "\20\1FREE\2INACTIVE\3ACTIVE\5ANON\6AOBJ"; 3457 3458 void 3459 uvm_page_printit(pg, full, pr) 3460 struct vm_page *pg; 3461 boolean_t full; 3462 void (*pr) __P((const char *, ...)); 3463 { 3464 struct vm_page *tpg; 3465 struct uvm_object *uobj; 3466 struct pglist *pgl; 3467 char pgbuf[128]; 3468 char pqbuf[128]; 3469 3470 (*pr)("PAGE %p:\n", pg); 3471 bitmask_snprintf(pg->flags, page_flagbits, pgbuf, sizeof(pgbuf)); 3472 bitmask_snprintf(pg->pqflags, page_pqflagbits, pqbuf, sizeof(pqbuf)); 3473 (*pr)(" flags=%s, pqflags=%s, wire_count=%d, pa=0x%lx\n", 3474 pgbuf, pqbuf, pg->wire_count, (long)pg->phys_addr); 3475 (*pr)(" uobject=%p, uanon=%p, offset=0x%llx loan_count=%d\n", 3476 pg->uobject, pg->uanon, (long long)pg->offset, pg->loan_count); 3477 #if defined(UVM_PAGE_TRKOWN) 3478 if (pg->flags & PG_BUSY) 3479 (*pr)(" owning process = %d, tag=%s\n", 3480 pg->owner, pg->owner_tag); 3481 else 3482 (*pr)(" page not busy, no owner\n"); 3483 #else 3484 (*pr)(" [page ownership tracking disabled]\n"); 3485 #endif 3486 3487 if (!full) 3488 return; 3489 3490 /* cross-verify object/anon */ 3491 if ((pg->pqflags & PQ_FREE) == 0) { 3492 if (pg->pqflags & PQ_ANON) { 3493 if (pg->uanon == NULL || pg->uanon->u.an_page != pg) 3494 (*pr)(" >>> ANON DOES NOT POINT HERE <<< (%p)\n", 3495 (pg->uanon) ? pg->uanon->u.an_page : NULL); 3496 else 3497 (*pr)(" anon backpointer is OK\n"); 3498 } else { 3499 uobj = pg->uobject; 3500 if (uobj) { 3501 (*pr)(" checking object list\n"); 3502 TAILQ_FOREACH(tpg, &uobj->memq, listq) { 3503 if (tpg == pg) { 3504 break; 3505 } 3506 } 3507 if (tpg) 3508 (*pr)(" page found on object list\n"); 3509 else 3510 (*pr)(" >>> PAGE NOT FOUND ON OBJECT LIST! <<<\n"); 3511 } 3512 } 3513 } 3514 3515 /* cross-verify page queue */ 3516 if (pg->pqflags & PQ_FREE) { 3517 int fl = uvm_page_lookup_freelist(pg); 3518 int color = VM_PGCOLOR_BUCKET(pg); 3519 pgl = &uvm.page_free[fl].pgfl_buckets[color].pgfl_queues[ 3520 ((pg)->flags & PG_ZERO) ? PGFL_ZEROS : PGFL_UNKNOWN]; 3521 } else if (pg->pqflags & PQ_INACTIVE) { 3522 pgl = &uvm.page_inactive; 3523 } else if (pg->pqflags & PQ_ACTIVE) { 3524 pgl = &uvm.page_active; 3525 } else { 3526 pgl = NULL; 3527 } 3528 3529 if (pgl) { 3530 (*pr)(" checking pageq list\n"); 3531 TAILQ_FOREACH(tpg, pgl, pageq) { 3532 if (tpg == pg) { 3533 break; 3534 } 3535 } 3536 if (tpg) 3537 (*pr)(" page found on pageq list\n"); 3538 else 3539 (*pr)(" >>> PAGE NOT FOUND ON PAGEQ LIST! <<<\n"); 3540 } 3541 } 3542 #endif 3543