1 /* 2 * Copyright (c) 1991, 1993 3 * The Regents of the University of California. All rights reserved. 4 * 5 * This code is derived from software contributed to Berkeley by 6 * The Mach Operating System project at Carnegie-Mellon University. 7 * 8 * Redistribution and use in source and binary forms, with or without 9 * modification, are permitted provided that the following conditions 10 * are met: 11 * 1. Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in the 15 * documentation and/or other materials provided with the distribution. 16 * 3. All advertising materials mentioning features or use of this software 17 * must display the following acknowledgement: 18 * This product includes software developed by the University of 19 * California, Berkeley and its contributors. 20 * 4. Neither the name of the University nor the names of its contributors 21 * may be used to endorse or promote products derived from this software 22 * without specific prior written permission. 23 * 24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 34 * SUCH DAMAGE. 35 * 36 * from: @(#)vm_object.c 8.5 (Berkeley) 3/22/94 37 * 38 * 39 * Copyright (c) 1987, 1990 Carnegie-Mellon University. 40 * All rights reserved. 41 * 42 * Authors: Avadis Tevanian, Jr., Michael Wayne Young 43 * 44 * Permission to use, copy, modify and distribute this software and 45 * its documentation is hereby granted, provided that both the copyright 46 * notice and this permission notice appear in all copies of the 47 * software, derivative works or modified versions, and any portions 48 * thereof, and that both notices appear in supporting documentation. 49 * 50 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS" 51 * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND 52 * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE. 53 * 54 * Carnegie Mellon requests users of this software to return to 55 * 56 * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU 57 * School of Computer Science 58 * Carnegie Mellon University 59 * Pittsburgh PA 15213-3890 60 * 61 * any improvements or extensions that they make and grant Carnegie the 62 * rights to redistribute these changes. 63 * 64 * $FreeBSD$ 65 */ 66 67 /* 68 * Virtual memory object module. 69 */ 70 71 #include <sys/param.h> 72 #include <sys/systm.h> 73 #include <sys/proc.h> /* for curproc, pageproc */ 74 #include <sys/vnode.h> 75 #include <sys/vmmeter.h> 76 #include <sys/mman.h> 77 #include <sys/mount.h> 78 #include <sys/mutex.h> 79 80 #include <vm/vm.h> 81 #include <vm/vm_param.h> 82 #include <vm/pmap.h> 83 #include <vm/vm_map.h> 84 #include <vm/vm_object.h> 85 #include <vm/vm_page.h> 86 #include <vm/vm_pageout.h> 87 #include <vm/vm_pager.h> 88 #include <vm/vm_zone.h> 89 #include <vm/swap_pager.h> 90 #include <vm/vm_kern.h> 91 #include <vm/vm_extern.h> 92 93 static void vm_object_qcollapse __P((vm_object_t object)); 94 95 /* 96 * Virtual memory objects maintain the actual data 97 * associated with allocated virtual memory. A given 98 * page of memory exists within exactly one object. 99 * 100 * An object is only deallocated when all "references" 101 * are given up. Only one "reference" to a given 102 * region of an object should be writeable. 103 * 104 * Associated with each object is a list of all resident 105 * memory pages belonging to that object; this list is 106 * maintained by the "vm_page" module, and locked by the object's 107 * lock. 108 * 109 * Each object also records a "pager" routine which is 110 * used to retrieve (and store) pages to the proper backing 111 * storage. In addition, objects may be backed by other 112 * objects from which they were virtual-copied. 113 * 114 * The only items within the object structure which are 115 * modified after time of creation are: 116 * reference count locked by object's lock 117 * pager routine locked by object's lock 118 * 119 */ 120 121 struct object_q vm_object_list; 122 static struct mtx vm_object_list_mtx; 123 static long vm_object_count; /* count of all objects */ 124 vm_object_t kernel_object; 125 vm_object_t kmem_object; 126 static struct vm_object kernel_object_store; 127 static struct vm_object kmem_object_store; 128 extern int vm_pageout_page_count; 129 130 static long object_collapses; 131 static long object_bypasses; 132 static int next_index; 133 static vm_zone_t obj_zone; 134 static struct vm_zone obj_zone_store; 135 static int object_hash_rand; 136 #define VM_OBJECTS_INIT 256 137 static struct vm_object vm_objects_init[VM_OBJECTS_INIT]; 138 139 void 140 _vm_object_allocate(type, size, object) 141 objtype_t type; 142 vm_size_t size; 143 vm_object_t object; 144 { 145 int incr; 146 TAILQ_INIT(&object->memq); 147 TAILQ_INIT(&object->shadow_head); 148 149 object->type = type; 150 object->size = size; 151 object->ref_count = 1; 152 object->flags = 0; 153 if ((object->type == OBJT_DEFAULT) || (object->type == OBJT_SWAP)) 154 vm_object_set_flag(object, OBJ_ONEMAPPING); 155 object->paging_in_progress = 0; 156 object->resident_page_count = 0; 157 object->shadow_count = 0; 158 object->pg_color = next_index; 159 if ( size > (PQ_L2_SIZE / 3 + PQ_PRIME1)) 160 incr = PQ_L2_SIZE / 3 + PQ_PRIME1; 161 else 162 incr = size; 163 next_index = (next_index + incr) & PQ_L2_MASK; 164 object->handle = NULL; 165 object->backing_object = NULL; 166 object->backing_object_offset = (vm_ooffset_t) 0; 167 /* 168 * Try to generate a number that will spread objects out in the 169 * hash table. We 'wipe' new objects across the hash in 128 page 170 * increments plus 1 more to offset it a little more by the time 171 * it wraps around. 172 */ 173 object->hash_rand = object_hash_rand - 129; 174 175 object->generation++; 176 177 TAILQ_INSERT_TAIL(&vm_object_list, object, object_list); 178 vm_object_count++; 179 object_hash_rand = object->hash_rand; 180 } 181 182 /* 183 * vm_object_init: 184 * 185 * Initialize the VM objects module. 186 */ 187 void 188 vm_object_init() 189 { 190 TAILQ_INIT(&vm_object_list); 191 mtx_init(&vm_object_list_mtx, "vm object_list", MTX_DEF); 192 vm_object_count = 0; 193 194 kernel_object = &kernel_object_store; 195 _vm_object_allocate(OBJT_DEFAULT, OFF_TO_IDX(VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS), 196 kernel_object); 197 198 kmem_object = &kmem_object_store; 199 _vm_object_allocate(OBJT_DEFAULT, OFF_TO_IDX(VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS), 200 kmem_object); 201 202 obj_zone = &obj_zone_store; 203 zbootinit(obj_zone, "VM OBJECT", sizeof (struct vm_object), 204 vm_objects_init, VM_OBJECTS_INIT); 205 } 206 207 void 208 vm_object_init2() { 209 zinitna(obj_zone, NULL, NULL, 0, 0, 0, 1); 210 } 211 212 /* 213 * vm_object_allocate: 214 * 215 * Returns a new object with the given size. 216 */ 217 218 vm_object_t 219 vm_object_allocate(type, size) 220 objtype_t type; 221 vm_size_t size; 222 { 223 vm_object_t result; 224 225 result = (vm_object_t) zalloc(obj_zone); 226 227 _vm_object_allocate(type, size, result); 228 229 return (result); 230 } 231 232 233 /* 234 * vm_object_reference: 235 * 236 * Gets another reference to the given object. 237 */ 238 void 239 vm_object_reference(object) 240 vm_object_t object; 241 { 242 if (object == NULL) 243 return; 244 245 KASSERT(!(object->flags & OBJ_DEAD), 246 ("vm_object_reference: attempting to reference dead obj")); 247 248 object->ref_count++; 249 if (object->type == OBJT_VNODE) { 250 while (vget((struct vnode *) object->handle, LK_RETRY|LK_NOOBJ, curproc)) { 251 printf("vm_object_reference: delay in getting object\n"); 252 } 253 } 254 } 255 256 void 257 vm_object_vndeallocate(object) 258 vm_object_t object; 259 { 260 struct vnode *vp = (struct vnode *) object->handle; 261 262 KASSERT(object->type == OBJT_VNODE, 263 ("vm_object_vndeallocate: not a vnode object")); 264 KASSERT(vp != NULL, ("vm_object_vndeallocate: missing vp")); 265 #ifdef INVARIANTS 266 if (object->ref_count == 0) { 267 vprint("vm_object_vndeallocate", vp); 268 panic("vm_object_vndeallocate: bad object reference count"); 269 } 270 #endif 271 272 object->ref_count--; 273 if (object->ref_count == 0) { 274 vp->v_flag &= ~VTEXT; 275 vm_object_clear_flag(object, OBJ_OPT); 276 } 277 vrele(vp); 278 } 279 280 /* 281 * vm_object_deallocate: 282 * 283 * Release a reference to the specified object, 284 * gained either through a vm_object_allocate 285 * or a vm_object_reference call. When all references 286 * are gone, storage associated with this object 287 * may be relinquished. 288 * 289 * No object may be locked. 290 */ 291 void 292 vm_object_deallocate(object) 293 vm_object_t object; 294 { 295 vm_object_t temp; 296 297 while (object != NULL) { 298 299 if (object->type == OBJT_VNODE) { 300 vm_object_vndeallocate(object); 301 return; 302 } 303 304 if (object->ref_count == 0) { 305 panic("vm_object_deallocate: object deallocated too many times: %d", object->type); 306 } else if (object->ref_count > 2) { 307 object->ref_count--; 308 return; 309 } 310 311 /* 312 * Here on ref_count of one or two, which are special cases for 313 * objects. 314 */ 315 if ((object->ref_count == 2) && (object->shadow_count == 0)) { 316 vm_object_set_flag(object, OBJ_ONEMAPPING); 317 object->ref_count--; 318 return; 319 } else if ((object->ref_count == 2) && (object->shadow_count == 1)) { 320 object->ref_count--; 321 if ((object->handle == NULL) && 322 (object->type == OBJT_DEFAULT || 323 object->type == OBJT_SWAP)) { 324 vm_object_t robject; 325 326 robject = TAILQ_FIRST(&object->shadow_head); 327 KASSERT(robject != NULL, 328 ("vm_object_deallocate: ref_count: %d, shadow_count: %d", 329 object->ref_count, 330 object->shadow_count)); 331 if ((robject->handle == NULL) && 332 (robject->type == OBJT_DEFAULT || 333 robject->type == OBJT_SWAP)) { 334 335 robject->ref_count++; 336 337 while ( 338 robject->paging_in_progress || 339 object->paging_in_progress 340 ) { 341 vm_object_pip_sleep(robject, "objde1"); 342 vm_object_pip_sleep(object, "objde2"); 343 } 344 345 if (robject->ref_count == 1) { 346 robject->ref_count--; 347 object = robject; 348 goto doterm; 349 } 350 351 object = robject; 352 vm_object_collapse(object); 353 continue; 354 } 355 } 356 357 return; 358 359 } else { 360 object->ref_count--; 361 if (object->ref_count != 0) 362 return; 363 } 364 365 doterm: 366 367 temp = object->backing_object; 368 if (temp) { 369 TAILQ_REMOVE(&temp->shadow_head, object, shadow_list); 370 temp->shadow_count--; 371 if (temp->ref_count == 0) 372 vm_object_clear_flag(temp, OBJ_OPT); 373 temp->generation++; 374 object->backing_object = NULL; 375 } 376 vm_object_terminate(object); 377 /* unlocks and deallocates object */ 378 object = temp; 379 } 380 } 381 382 /* 383 * vm_object_terminate actually destroys the specified object, freeing 384 * up all previously used resources. 385 * 386 * The object must be locked. 387 * This routine may block. 388 */ 389 void 390 vm_object_terminate(object) 391 vm_object_t object; 392 { 393 vm_page_t p; 394 int s; 395 396 /* 397 * Make sure no one uses us. 398 */ 399 vm_object_set_flag(object, OBJ_DEAD); 400 401 /* 402 * wait for the pageout daemon to be done with the object 403 */ 404 vm_object_pip_wait(object, "objtrm"); 405 406 KASSERT(!object->paging_in_progress, 407 ("vm_object_terminate: pageout in progress")); 408 409 /* 410 * Clean and free the pages, as appropriate. All references to the 411 * object are gone, so we don't need to lock it. 412 */ 413 if (object->type == OBJT_VNODE) { 414 struct vnode *vp; 415 416 /* 417 * Freeze optimized copies. 418 */ 419 vm_freeze_copyopts(object, 0, object->size); 420 421 /* 422 * Clean pages and flush buffers. 423 */ 424 vm_object_page_clean(object, 0, 0, OBJPC_SYNC); 425 426 vp = (struct vnode *) object->handle; 427 vinvalbuf(vp, V_SAVE, NOCRED, NULL, 0, 0); 428 } 429 430 if (object->ref_count != 0) 431 panic("vm_object_terminate: object with references, ref_count=%d", object->ref_count); 432 433 /* 434 * Now free any remaining pages. For internal objects, this also 435 * removes them from paging queues. Don't free wired pages, just 436 * remove them from the object. 437 */ 438 s = splvm(); 439 while ((p = TAILQ_FIRST(&object->memq)) != NULL) { 440 if (p->busy || (p->flags & PG_BUSY)) 441 panic("vm_object_terminate: freeing busy page %p\n", p); 442 if (p->wire_count == 0) { 443 vm_page_busy(p); 444 vm_page_free(p); 445 cnt.v_pfree++; 446 } else { 447 vm_page_busy(p); 448 vm_page_remove(p); 449 } 450 } 451 splx(s); 452 453 /* 454 * Let the pager know object is dead. 455 */ 456 vm_pager_deallocate(object); 457 458 /* 459 * Remove the object from the global object list. 460 */ 461 mtx_lock(&vm_object_list_mtx); 462 TAILQ_REMOVE(&vm_object_list, object, object_list); 463 mtx_unlock(&vm_object_list_mtx); 464 465 wakeup(object); 466 467 /* 468 * Free the space for the object. 469 */ 470 zfree(obj_zone, object); 471 } 472 473 /* 474 * vm_object_page_clean 475 * 476 * Clean all dirty pages in the specified range of object. Leaves page 477 * on whatever queue it is currently on. If NOSYNC is set then do not 478 * write out pages with PG_NOSYNC set (originally comes from MAP_NOSYNC), 479 * leaving the object dirty. 480 * 481 * Odd semantics: if start == end, we clean everything. 482 * 483 * The object must be locked. 484 */ 485 486 void 487 vm_object_page_clean(object, start, end, flags) 488 vm_object_t object; 489 vm_pindex_t start; 490 vm_pindex_t end; 491 int flags; 492 { 493 vm_page_t p, np, tp; 494 vm_offset_t tstart, tend; 495 vm_pindex_t pi; 496 int s; 497 struct vnode *vp; 498 int runlen; 499 int maxf; 500 int chkb; 501 int maxb; 502 int i; 503 int clearobjflags; 504 int pagerflags; 505 vm_page_t maf[vm_pageout_page_count]; 506 vm_page_t mab[vm_pageout_page_count]; 507 vm_page_t ma[vm_pageout_page_count]; 508 int curgeneration; 509 510 if (object->type != OBJT_VNODE || 511 (object->flags & OBJ_MIGHTBEDIRTY) == 0) 512 return; 513 514 pagerflags = (flags & (OBJPC_SYNC | OBJPC_INVAL)) ? VM_PAGER_PUT_SYNC : 0; 515 pagerflags |= (flags & OBJPC_INVAL) ? VM_PAGER_PUT_INVAL : 0; 516 517 vp = object->handle; 518 519 vm_object_set_flag(object, OBJ_CLEANING); 520 521 tstart = start; 522 if (end == 0) { 523 tend = object->size; 524 } else { 525 tend = end; 526 } 527 528 /* 529 * Generally set CLEANCHK interlock and make the page read-only so 530 * we can then clear the object flags. 531 * 532 * However, if this is a nosync mmap then the object is likely to 533 * stay dirty so do not mess with the page and do not clear the 534 * object flags. 535 */ 536 537 clearobjflags = 1; 538 539 TAILQ_FOREACH(p, &object->memq, listq) { 540 vm_page_flag_set(p, PG_CLEANCHK); 541 if ((flags & OBJPC_NOSYNC) && (p->flags & PG_NOSYNC)) 542 clearobjflags = 0; 543 else 544 vm_page_protect(p, VM_PROT_READ); 545 } 546 547 if (clearobjflags && (tstart == 0) && (tend == object->size)) { 548 vm_object_clear_flag(object, OBJ_WRITEABLE|OBJ_MIGHTBEDIRTY); 549 } 550 551 rescan: 552 curgeneration = object->generation; 553 554 for(p = TAILQ_FIRST(&object->memq); p; p = np) { 555 np = TAILQ_NEXT(p, listq); 556 557 pi = p->pindex; 558 if (((p->flags & PG_CLEANCHK) == 0) || 559 (pi < tstart) || (pi >= tend) || 560 (p->valid == 0) || 561 ((p->queue - p->pc) == PQ_CACHE)) { 562 vm_page_flag_clear(p, PG_CLEANCHK); 563 continue; 564 } 565 566 vm_page_test_dirty(p); 567 if ((p->dirty & p->valid) == 0) { 568 vm_page_flag_clear(p, PG_CLEANCHK); 569 continue; 570 } 571 572 /* 573 * If we have been asked to skip nosync pages and this is a 574 * nosync page, skip it. Note that the object flags were 575 * not cleared in this case so we do not have to set them. 576 */ 577 if ((flags & OBJPC_NOSYNC) && (p->flags & PG_NOSYNC)) { 578 vm_page_flag_clear(p, PG_CLEANCHK); 579 continue; 580 } 581 582 s = splvm(); 583 while (vm_page_sleep_busy(p, TRUE, "vpcwai")) { 584 if (object->generation != curgeneration) { 585 splx(s); 586 goto rescan; 587 } 588 } 589 590 maxf = 0; 591 for(i=1;i<vm_pageout_page_count;i++) { 592 if ((tp = vm_page_lookup(object, pi + i)) != NULL) { 593 if ((tp->flags & PG_BUSY) || 594 (tp->flags & PG_CLEANCHK) == 0 || 595 (tp->busy != 0)) 596 break; 597 if((tp->queue - tp->pc) == PQ_CACHE) { 598 vm_page_flag_clear(tp, PG_CLEANCHK); 599 break; 600 } 601 vm_page_test_dirty(tp); 602 if ((tp->dirty & tp->valid) == 0) { 603 vm_page_flag_clear(tp, PG_CLEANCHK); 604 break; 605 } 606 maf[ i - 1 ] = tp; 607 maxf++; 608 continue; 609 } 610 break; 611 } 612 613 maxb = 0; 614 chkb = vm_pageout_page_count - maxf; 615 if (chkb) { 616 for(i = 1; i < chkb;i++) { 617 if ((tp = vm_page_lookup(object, pi - i)) != NULL) { 618 if ((tp->flags & PG_BUSY) || 619 (tp->flags & PG_CLEANCHK) == 0 || 620 (tp->busy != 0)) 621 break; 622 if((tp->queue - tp->pc) == PQ_CACHE) { 623 vm_page_flag_clear(tp, PG_CLEANCHK); 624 break; 625 } 626 vm_page_test_dirty(tp); 627 if ((tp->dirty & tp->valid) == 0) { 628 vm_page_flag_clear(tp, PG_CLEANCHK); 629 break; 630 } 631 mab[ i - 1 ] = tp; 632 maxb++; 633 continue; 634 } 635 break; 636 } 637 } 638 639 for(i=0;i<maxb;i++) { 640 int index = (maxb - i) - 1; 641 ma[index] = mab[i]; 642 vm_page_flag_clear(ma[index], PG_CLEANCHK); 643 } 644 vm_page_flag_clear(p, PG_CLEANCHK); 645 ma[maxb] = p; 646 for(i=0;i<maxf;i++) { 647 int index = (maxb + i) + 1; 648 ma[index] = maf[i]; 649 vm_page_flag_clear(ma[index], PG_CLEANCHK); 650 } 651 runlen = maxb + maxf + 1; 652 653 splx(s); 654 vm_pageout_flush(ma, runlen, pagerflags); 655 for (i = 0; i<runlen; i++) { 656 if (ma[i]->valid & ma[i]->dirty) { 657 vm_page_protect(ma[i], VM_PROT_READ); 658 vm_page_flag_set(ma[i], PG_CLEANCHK); 659 } 660 } 661 if (object->generation != curgeneration) 662 goto rescan; 663 } 664 665 #if 0 666 VOP_FSYNC(vp, NULL, (pagerflags & VM_PAGER_PUT_SYNC)?MNT_WAIT:0, curproc); 667 #endif 668 669 vm_object_clear_flag(object, OBJ_CLEANING); 670 return; 671 } 672 673 #ifdef not_used 674 /* XXX I cannot tell if this should be an exported symbol */ 675 /* 676 * vm_object_deactivate_pages 677 * 678 * Deactivate all pages in the specified object. (Keep its pages 679 * in memory even though it is no longer referenced.) 680 * 681 * The object must be locked. 682 */ 683 static void 684 vm_object_deactivate_pages(object) 685 vm_object_t object; 686 { 687 vm_page_t p, next; 688 689 for (p = TAILQ_FIRST(&object->memq); p != NULL; p = next) { 690 next = TAILQ_NEXT(p, listq); 691 vm_page_deactivate(p); 692 } 693 } 694 #endif 695 696 /* 697 * Same as vm_object_pmap_copy, except range checking really 698 * works, and is meant for small sections of an object. 699 * 700 * This code protects resident pages by making them read-only 701 * and is typically called on a fork or split when a page 702 * is converted to copy-on-write. 703 * 704 * NOTE: If the page is already at VM_PROT_NONE, calling 705 * vm_page_protect will have no effect. 706 */ 707 708 void 709 vm_object_pmap_copy_1(object, start, end) 710 vm_object_t object; 711 vm_pindex_t start; 712 vm_pindex_t end; 713 { 714 vm_pindex_t idx; 715 vm_page_t p; 716 717 if (object == NULL || (object->flags & OBJ_WRITEABLE) == 0) 718 return; 719 720 for (idx = start; idx < end; idx++) { 721 p = vm_page_lookup(object, idx); 722 if (p == NULL) 723 continue; 724 vm_page_protect(p, VM_PROT_READ); 725 } 726 } 727 728 /* 729 * vm_object_pmap_remove: 730 * 731 * Removes all physical pages in the specified 732 * object range from all physical maps. 733 * 734 * The object must *not* be locked. 735 */ 736 void 737 vm_object_pmap_remove(object, start, end) 738 vm_object_t object; 739 vm_pindex_t start; 740 vm_pindex_t end; 741 { 742 vm_page_t p; 743 744 if (object == NULL) 745 return; 746 for (p = TAILQ_FIRST(&object->memq); 747 p != NULL; 748 p = TAILQ_NEXT(p, listq)) { 749 if (p->pindex >= start && p->pindex < end) 750 vm_page_protect(p, VM_PROT_NONE); 751 } 752 if ((start == 0) && (object->size == end)) 753 vm_object_clear_flag(object, OBJ_WRITEABLE); 754 } 755 756 /* 757 * vm_object_madvise: 758 * 759 * Implements the madvise function at the object/page level. 760 * 761 * MADV_WILLNEED (any object) 762 * 763 * Activate the specified pages if they are resident. 764 * 765 * MADV_DONTNEED (any object) 766 * 767 * Deactivate the specified pages if they are resident. 768 * 769 * MADV_FREE (OBJT_DEFAULT/OBJT_SWAP objects, 770 * OBJ_ONEMAPPING only) 771 * 772 * Deactivate and clean the specified pages if they are 773 * resident. This permits the process to reuse the pages 774 * without faulting or the kernel to reclaim the pages 775 * without I/O. 776 */ 777 void 778 vm_object_madvise(object, pindex, count, advise) 779 vm_object_t object; 780 vm_pindex_t pindex; 781 int count; 782 int advise; 783 { 784 vm_pindex_t end, tpindex; 785 vm_object_t tobject; 786 vm_page_t m; 787 788 if (object == NULL) 789 return; 790 791 end = pindex + count; 792 793 /* 794 * Locate and adjust resident pages 795 */ 796 797 for (; pindex < end; pindex += 1) { 798 relookup: 799 tobject = object; 800 tpindex = pindex; 801 shadowlookup: 802 /* 803 * MADV_FREE only operates on OBJT_DEFAULT or OBJT_SWAP pages 804 * and those pages must be OBJ_ONEMAPPING. 805 */ 806 if (advise == MADV_FREE) { 807 if ((tobject->type != OBJT_DEFAULT && 808 tobject->type != OBJT_SWAP) || 809 (tobject->flags & OBJ_ONEMAPPING) == 0) { 810 continue; 811 } 812 } 813 814 m = vm_page_lookup(tobject, tpindex); 815 816 if (m == NULL) { 817 /* 818 * There may be swap even if there is no backing page 819 */ 820 if (advise == MADV_FREE && tobject->type == OBJT_SWAP) 821 swap_pager_freespace(tobject, tpindex, 1); 822 823 /* 824 * next object 825 */ 826 tobject = tobject->backing_object; 827 if (tobject == NULL) 828 continue; 829 tpindex += OFF_TO_IDX(tobject->backing_object_offset); 830 goto shadowlookup; 831 } 832 833 /* 834 * If the page is busy or not in a normal active state, 835 * we skip it. If the page is not managed there are no 836 * page queues to mess with. Things can break if we mess 837 * with pages in any of the below states. 838 */ 839 if ( 840 m->hold_count || 841 m->wire_count || 842 (m->flags & PG_UNMANAGED) || 843 m->valid != VM_PAGE_BITS_ALL 844 ) { 845 continue; 846 } 847 848 if (vm_page_sleep_busy(m, TRUE, "madvpo")) 849 goto relookup; 850 851 if (advise == MADV_WILLNEED) { 852 vm_page_activate(m); 853 } else if (advise == MADV_DONTNEED) { 854 vm_page_dontneed(m); 855 } else if (advise == MADV_FREE) { 856 /* 857 * Mark the page clean. This will allow the page 858 * to be freed up by the system. However, such pages 859 * are often reused quickly by malloc()/free() 860 * so we do not do anything that would cause 861 * a page fault if we can help it. 862 * 863 * Specifically, we do not try to actually free 864 * the page now nor do we try to put it in the 865 * cache (which would cause a page fault on reuse). 866 * 867 * But we do make the page is freeable as we 868 * can without actually taking the step of unmapping 869 * it. 870 */ 871 pmap_clear_modify(m); 872 m->dirty = 0; 873 m->act_count = 0; 874 vm_page_dontneed(m); 875 if (tobject->type == OBJT_SWAP) 876 swap_pager_freespace(tobject, tpindex, 1); 877 } 878 } 879 } 880 881 /* 882 * vm_object_shadow: 883 * 884 * Create a new object which is backed by the 885 * specified existing object range. The source 886 * object reference is deallocated. 887 * 888 * The new object and offset into that object 889 * are returned in the source parameters. 890 */ 891 892 void 893 vm_object_shadow(object, offset, length) 894 vm_object_t *object; /* IN/OUT */ 895 vm_ooffset_t *offset; /* IN/OUT */ 896 vm_size_t length; 897 { 898 vm_object_t source; 899 vm_object_t result; 900 901 source = *object; 902 903 /* 904 * Don't create the new object if the old object isn't shared. 905 */ 906 907 if (source != NULL && 908 source->ref_count == 1 && 909 source->handle == NULL && 910 (source->type == OBJT_DEFAULT || 911 source->type == OBJT_SWAP)) 912 return; 913 914 /* 915 * Allocate a new object with the given length 916 */ 917 918 if ((result = vm_object_allocate(OBJT_DEFAULT, length)) == NULL) 919 panic("vm_object_shadow: no object for shadowing"); 920 921 /* 922 * The new object shadows the source object, adding a reference to it. 923 * Our caller changes his reference to point to the new object, 924 * removing a reference to the source object. Net result: no change 925 * of reference count. 926 * 927 * Try to optimize the result object's page color when shadowing 928 * in order to maintain page coloring consistency in the combined 929 * shadowed object. 930 */ 931 result->backing_object = source; 932 if (source) { 933 TAILQ_INSERT_TAIL(&source->shadow_head, result, shadow_list); 934 source->shadow_count++; 935 source->generation++; 936 result->pg_color = (source->pg_color + OFF_TO_IDX(*offset)) & PQ_L2_MASK; 937 } 938 939 /* 940 * Store the offset into the source object, and fix up the offset into 941 * the new object. 942 */ 943 944 result->backing_object_offset = *offset; 945 946 /* 947 * Return the new things 948 */ 949 950 *offset = 0; 951 *object = result; 952 } 953 954 #define OBSC_TEST_ALL_SHADOWED 0x0001 955 #define OBSC_COLLAPSE_NOWAIT 0x0002 956 #define OBSC_COLLAPSE_WAIT 0x0004 957 958 static __inline int 959 vm_object_backing_scan(vm_object_t object, int op) 960 { 961 int s; 962 int r = 1; 963 vm_page_t p; 964 vm_object_t backing_object; 965 vm_pindex_t backing_offset_index; 966 967 s = splvm(); 968 969 backing_object = object->backing_object; 970 backing_offset_index = OFF_TO_IDX(object->backing_object_offset); 971 972 /* 973 * Initial conditions 974 */ 975 976 if (op & OBSC_TEST_ALL_SHADOWED) { 977 /* 978 * We do not want to have to test for the existence of 979 * swap pages in the backing object. XXX but with the 980 * new swapper this would be pretty easy to do. 981 * 982 * XXX what about anonymous MAP_SHARED memory that hasn't 983 * been ZFOD faulted yet? If we do not test for this, the 984 * shadow test may succeed! XXX 985 */ 986 if (backing_object->type != OBJT_DEFAULT) { 987 splx(s); 988 return(0); 989 } 990 } 991 if (op & OBSC_COLLAPSE_WAIT) { 992 vm_object_set_flag(backing_object, OBJ_DEAD); 993 } 994 995 /* 996 * Our scan 997 */ 998 999 p = TAILQ_FIRST(&backing_object->memq); 1000 while (p) { 1001 vm_page_t next = TAILQ_NEXT(p, listq); 1002 vm_pindex_t new_pindex = p->pindex - backing_offset_index; 1003 1004 if (op & OBSC_TEST_ALL_SHADOWED) { 1005 vm_page_t pp; 1006 1007 /* 1008 * Ignore pages outside the parent object's range 1009 * and outside the parent object's mapping of the 1010 * backing object. 1011 * 1012 * note that we do not busy the backing object's 1013 * page. 1014 */ 1015 1016 if ( 1017 p->pindex < backing_offset_index || 1018 new_pindex >= object->size 1019 ) { 1020 p = next; 1021 continue; 1022 } 1023 1024 /* 1025 * See if the parent has the page or if the parent's 1026 * object pager has the page. If the parent has the 1027 * page but the page is not valid, the parent's 1028 * object pager must have the page. 1029 * 1030 * If this fails, the parent does not completely shadow 1031 * the object and we might as well give up now. 1032 */ 1033 1034 pp = vm_page_lookup(object, new_pindex); 1035 if ( 1036 (pp == NULL || pp->valid == 0) && 1037 !vm_pager_has_page(object, new_pindex, NULL, NULL) 1038 ) { 1039 r = 0; 1040 break; 1041 } 1042 } 1043 1044 /* 1045 * Check for busy page 1046 */ 1047 1048 if (op & (OBSC_COLLAPSE_WAIT | OBSC_COLLAPSE_NOWAIT)) { 1049 vm_page_t pp; 1050 1051 if (op & OBSC_COLLAPSE_NOWAIT) { 1052 if ( 1053 (p->flags & PG_BUSY) || 1054 !p->valid || 1055 p->hold_count || 1056 p->wire_count || 1057 p->busy 1058 ) { 1059 p = next; 1060 continue; 1061 } 1062 } else if (op & OBSC_COLLAPSE_WAIT) { 1063 if (vm_page_sleep_busy(p, TRUE, "vmocol")) { 1064 /* 1065 * If we slept, anything could have 1066 * happened. Since the object is 1067 * marked dead, the backing offset 1068 * should not have changed so we 1069 * just restart our scan. 1070 */ 1071 p = TAILQ_FIRST(&backing_object->memq); 1072 continue; 1073 } 1074 } 1075 1076 /* 1077 * Busy the page 1078 */ 1079 vm_page_busy(p); 1080 1081 KASSERT( 1082 p->object == backing_object, 1083 ("vm_object_qcollapse(): object mismatch") 1084 ); 1085 1086 /* 1087 * Destroy any associated swap 1088 */ 1089 if (backing_object->type == OBJT_SWAP) { 1090 swap_pager_freespace( 1091 backing_object, 1092 p->pindex, 1093 1 1094 ); 1095 } 1096 1097 if ( 1098 p->pindex < backing_offset_index || 1099 new_pindex >= object->size 1100 ) { 1101 /* 1102 * Page is out of the parent object's range, we 1103 * can simply destroy it. 1104 */ 1105 vm_page_protect(p, VM_PROT_NONE); 1106 vm_page_free(p); 1107 p = next; 1108 continue; 1109 } 1110 1111 pp = vm_page_lookup(object, new_pindex); 1112 if ( 1113 pp != NULL || 1114 vm_pager_has_page(object, new_pindex, NULL, NULL) 1115 ) { 1116 /* 1117 * page already exists in parent OR swap exists 1118 * for this location in the parent. Destroy 1119 * the original page from the backing object. 1120 * 1121 * Leave the parent's page alone 1122 */ 1123 vm_page_protect(p, VM_PROT_NONE); 1124 vm_page_free(p); 1125 p = next; 1126 continue; 1127 } 1128 1129 /* 1130 * Page does not exist in parent, rename the 1131 * page from the backing object to the main object. 1132 * 1133 * If the page was mapped to a process, it can remain 1134 * mapped through the rename. 1135 */ 1136 if ((p->queue - p->pc) == PQ_CACHE) 1137 vm_page_deactivate(p); 1138 1139 vm_page_rename(p, object, new_pindex); 1140 /* page automatically made dirty by rename */ 1141 } 1142 p = next; 1143 } 1144 splx(s); 1145 return(r); 1146 } 1147 1148 1149 /* 1150 * this version of collapse allows the operation to occur earlier and 1151 * when paging_in_progress is true for an object... This is not a complete 1152 * operation, but should plug 99.9% of the rest of the leaks. 1153 */ 1154 static void 1155 vm_object_qcollapse(object) 1156 vm_object_t object; 1157 { 1158 vm_object_t backing_object = object->backing_object; 1159 1160 if (backing_object->ref_count != 1) 1161 return; 1162 1163 backing_object->ref_count += 2; 1164 1165 vm_object_backing_scan(object, OBSC_COLLAPSE_NOWAIT); 1166 1167 backing_object->ref_count -= 2; 1168 } 1169 1170 /* 1171 * vm_object_collapse: 1172 * 1173 * Collapse an object with the object backing it. 1174 * Pages in the backing object are moved into the 1175 * parent, and the backing object is deallocated. 1176 */ 1177 void 1178 vm_object_collapse(object) 1179 vm_object_t object; 1180 { 1181 while (TRUE) { 1182 vm_object_t backing_object; 1183 1184 /* 1185 * Verify that the conditions are right for collapse: 1186 * 1187 * The object exists and the backing object exists. 1188 */ 1189 if (object == NULL) 1190 break; 1191 1192 if ((backing_object = object->backing_object) == NULL) 1193 break; 1194 1195 /* 1196 * we check the backing object first, because it is most likely 1197 * not collapsable. 1198 */ 1199 if (backing_object->handle != NULL || 1200 (backing_object->type != OBJT_DEFAULT && 1201 backing_object->type != OBJT_SWAP) || 1202 (backing_object->flags & OBJ_DEAD) || 1203 object->handle != NULL || 1204 (object->type != OBJT_DEFAULT && 1205 object->type != OBJT_SWAP) || 1206 (object->flags & OBJ_DEAD)) { 1207 break; 1208 } 1209 1210 if ( 1211 object->paging_in_progress != 0 || 1212 backing_object->paging_in_progress != 0 1213 ) { 1214 vm_object_qcollapse(object); 1215 break; 1216 } 1217 1218 /* 1219 * We know that we can either collapse the backing object (if 1220 * the parent is the only reference to it) or (perhaps) have 1221 * the parent bypass the object if the parent happens to shadow 1222 * all the resident pages in the entire backing object. 1223 * 1224 * This is ignoring pager-backed pages such as swap pages. 1225 * vm_object_backing_scan fails the shadowing test in this 1226 * case. 1227 */ 1228 1229 if (backing_object->ref_count == 1) { 1230 /* 1231 * If there is exactly one reference to the backing 1232 * object, we can collapse it into the parent. 1233 */ 1234 1235 vm_object_backing_scan(object, OBSC_COLLAPSE_WAIT); 1236 1237 /* 1238 * Move the pager from backing_object to object. 1239 */ 1240 1241 if (backing_object->type == OBJT_SWAP) { 1242 vm_object_pip_add(backing_object, 1); 1243 1244 /* 1245 * scrap the paging_offset junk and do a 1246 * discrete copy. This also removes major 1247 * assumptions about how the swap-pager 1248 * works from where it doesn't belong. The 1249 * new swapper is able to optimize the 1250 * destroy-source case. 1251 */ 1252 1253 vm_object_pip_add(object, 1); 1254 swap_pager_copy( 1255 backing_object, 1256 object, 1257 OFF_TO_IDX(object->backing_object_offset), TRUE); 1258 vm_object_pip_wakeup(object); 1259 1260 vm_object_pip_wakeup(backing_object); 1261 } 1262 /* 1263 * Object now shadows whatever backing_object did. 1264 * Note that the reference to 1265 * backing_object->backing_object moves from within 1266 * backing_object to within object. 1267 */ 1268 1269 TAILQ_REMOVE( 1270 &object->backing_object->shadow_head, 1271 object, 1272 shadow_list 1273 ); 1274 object->backing_object->shadow_count--; 1275 object->backing_object->generation++; 1276 if (backing_object->backing_object) { 1277 TAILQ_REMOVE( 1278 &backing_object->backing_object->shadow_head, 1279 backing_object, 1280 shadow_list 1281 ); 1282 backing_object->backing_object->shadow_count--; 1283 backing_object->backing_object->generation++; 1284 } 1285 object->backing_object = backing_object->backing_object; 1286 if (object->backing_object) { 1287 TAILQ_INSERT_TAIL( 1288 &object->backing_object->shadow_head, 1289 object, 1290 shadow_list 1291 ); 1292 object->backing_object->shadow_count++; 1293 object->backing_object->generation++; 1294 } 1295 1296 object->backing_object_offset += 1297 backing_object->backing_object_offset; 1298 1299 /* 1300 * Discard backing_object. 1301 * 1302 * Since the backing object has no pages, no pager left, 1303 * and no object references within it, all that is 1304 * necessary is to dispose of it. 1305 */ 1306 1307 TAILQ_REMOVE( 1308 &vm_object_list, 1309 backing_object, 1310 object_list 1311 ); 1312 vm_object_count--; 1313 1314 zfree(obj_zone, backing_object); 1315 1316 object_collapses++; 1317 } else { 1318 vm_object_t new_backing_object; 1319 1320 /* 1321 * If we do not entirely shadow the backing object, 1322 * there is nothing we can do so we give up. 1323 */ 1324 1325 if (vm_object_backing_scan(object, OBSC_TEST_ALL_SHADOWED) == 0) { 1326 break; 1327 } 1328 1329 /* 1330 * Make the parent shadow the next object in the 1331 * chain. Deallocating backing_object will not remove 1332 * it, since its reference count is at least 2. 1333 */ 1334 1335 TAILQ_REMOVE( 1336 &backing_object->shadow_head, 1337 object, 1338 shadow_list 1339 ); 1340 backing_object->shadow_count--; 1341 backing_object->generation++; 1342 1343 new_backing_object = backing_object->backing_object; 1344 if ((object->backing_object = new_backing_object) != NULL) { 1345 vm_object_reference(new_backing_object); 1346 TAILQ_INSERT_TAIL( 1347 &new_backing_object->shadow_head, 1348 object, 1349 shadow_list 1350 ); 1351 new_backing_object->shadow_count++; 1352 new_backing_object->generation++; 1353 object->backing_object_offset += 1354 backing_object->backing_object_offset; 1355 } 1356 1357 /* 1358 * Drop the reference count on backing_object. Since 1359 * its ref_count was at least 2, it will not vanish; 1360 * so we don't need to call vm_object_deallocate, but 1361 * we do anyway. 1362 */ 1363 vm_object_deallocate(backing_object); 1364 object_bypasses++; 1365 } 1366 1367 /* 1368 * Try again with this object's new backing object. 1369 */ 1370 } 1371 } 1372 1373 /* 1374 * vm_object_page_remove: [internal] 1375 * 1376 * Removes all physical pages in the specified 1377 * object range from the object's list of pages. 1378 * 1379 * The object must be locked. 1380 */ 1381 void 1382 vm_object_page_remove(object, start, end, clean_only) 1383 vm_object_t object; 1384 vm_pindex_t start; 1385 vm_pindex_t end; 1386 boolean_t clean_only; 1387 { 1388 vm_page_t p, next; 1389 unsigned int size; 1390 int all; 1391 1392 if (object == NULL || 1393 object->resident_page_count == 0) 1394 return; 1395 1396 all = ((end == 0) && (start == 0)); 1397 1398 /* 1399 * Since physically-backed objects do not use managed pages, we can't 1400 * remove pages from the object (we must instead remove the page 1401 * references, and then destroy the object). 1402 */ 1403 KASSERT(object->type != OBJT_PHYS, ("attempt to remove pages from a physical object")); 1404 1405 vm_object_pip_add(object, 1); 1406 again: 1407 size = end - start; 1408 if (all || size > object->resident_page_count / 4) { 1409 for (p = TAILQ_FIRST(&object->memq); p != NULL; p = next) { 1410 next = TAILQ_NEXT(p, listq); 1411 if (all || ((start <= p->pindex) && (p->pindex < end))) { 1412 if (p->wire_count != 0) { 1413 vm_page_protect(p, VM_PROT_NONE); 1414 if (!clean_only) 1415 p->valid = 0; 1416 continue; 1417 } 1418 1419 /* 1420 * The busy flags are only cleared at 1421 * interrupt -- minimize the spl transitions 1422 */ 1423 1424 if (vm_page_sleep_busy(p, TRUE, "vmopar")) 1425 goto again; 1426 1427 if (clean_only && p->valid) { 1428 vm_page_test_dirty(p); 1429 if (p->valid & p->dirty) 1430 continue; 1431 } 1432 1433 vm_page_busy(p); 1434 vm_page_protect(p, VM_PROT_NONE); 1435 vm_page_free(p); 1436 } 1437 } 1438 } else { 1439 while (size > 0) { 1440 if ((p = vm_page_lookup(object, start)) != 0) { 1441 1442 if (p->wire_count != 0) { 1443 vm_page_protect(p, VM_PROT_NONE); 1444 if (!clean_only) 1445 p->valid = 0; 1446 start += 1; 1447 size -= 1; 1448 continue; 1449 } 1450 1451 /* 1452 * The busy flags are only cleared at 1453 * interrupt -- minimize the spl transitions 1454 */ 1455 if (vm_page_sleep_busy(p, TRUE, "vmopar")) 1456 goto again; 1457 1458 if (clean_only && p->valid) { 1459 vm_page_test_dirty(p); 1460 if (p->valid & p->dirty) { 1461 start += 1; 1462 size -= 1; 1463 continue; 1464 } 1465 } 1466 1467 vm_page_busy(p); 1468 vm_page_protect(p, VM_PROT_NONE); 1469 vm_page_free(p); 1470 } 1471 start += 1; 1472 size -= 1; 1473 } 1474 } 1475 vm_object_pip_wakeup(object); 1476 } 1477 1478 /* 1479 * Routine: vm_object_coalesce 1480 * Function: Coalesces two objects backing up adjoining 1481 * regions of memory into a single object. 1482 * 1483 * returns TRUE if objects were combined. 1484 * 1485 * NOTE: Only works at the moment if the second object is NULL - 1486 * if it's not, which object do we lock first? 1487 * 1488 * Parameters: 1489 * prev_object First object to coalesce 1490 * prev_offset Offset into prev_object 1491 * next_object Second object into coalesce 1492 * next_offset Offset into next_object 1493 * 1494 * prev_size Size of reference to prev_object 1495 * next_size Size of reference to next_object 1496 * 1497 * Conditions: 1498 * The object must *not* be locked. 1499 */ 1500 boolean_t 1501 vm_object_coalesce(prev_object, prev_pindex, prev_size, next_size) 1502 vm_object_t prev_object; 1503 vm_pindex_t prev_pindex; 1504 vm_size_t prev_size, next_size; 1505 { 1506 vm_pindex_t next_pindex; 1507 1508 if (prev_object == NULL) { 1509 return (TRUE); 1510 } 1511 1512 if (prev_object->type != OBJT_DEFAULT && 1513 prev_object->type != OBJT_SWAP) { 1514 return (FALSE); 1515 } 1516 1517 /* 1518 * Try to collapse the object first 1519 */ 1520 vm_object_collapse(prev_object); 1521 1522 /* 1523 * Can't coalesce if: . more than one reference . paged out . shadows 1524 * another object . has a copy elsewhere (any of which mean that the 1525 * pages not mapped to prev_entry may be in use anyway) 1526 */ 1527 1528 if (prev_object->backing_object != NULL) { 1529 return (FALSE); 1530 } 1531 1532 prev_size >>= PAGE_SHIFT; 1533 next_size >>= PAGE_SHIFT; 1534 next_pindex = prev_pindex + prev_size; 1535 1536 if ((prev_object->ref_count > 1) && 1537 (prev_object->size != next_pindex)) { 1538 return (FALSE); 1539 } 1540 1541 /* 1542 * Remove any pages that may still be in the object from a previous 1543 * deallocation. 1544 */ 1545 if (next_pindex < prev_object->size) { 1546 vm_object_page_remove(prev_object, 1547 next_pindex, 1548 next_pindex + next_size, FALSE); 1549 if (prev_object->type == OBJT_SWAP) 1550 swap_pager_freespace(prev_object, 1551 next_pindex, next_size); 1552 } 1553 1554 /* 1555 * Extend the object if necessary. 1556 */ 1557 if (next_pindex + next_size > prev_object->size) 1558 prev_object->size = next_pindex + next_size; 1559 1560 return (TRUE); 1561 } 1562 1563 #include "opt_ddb.h" 1564 #ifdef DDB 1565 #include <sys/kernel.h> 1566 1567 #include <sys/cons.h> 1568 1569 #include <ddb/ddb.h> 1570 1571 static int _vm_object_in_map __P((vm_map_t map, vm_object_t object, 1572 vm_map_entry_t entry)); 1573 static int vm_object_in_map __P((vm_object_t object)); 1574 1575 static int 1576 _vm_object_in_map(map, object, entry) 1577 vm_map_t map; 1578 vm_object_t object; 1579 vm_map_entry_t entry; 1580 { 1581 vm_map_t tmpm; 1582 vm_map_entry_t tmpe; 1583 vm_object_t obj; 1584 int entcount; 1585 1586 if (map == 0) 1587 return 0; 1588 1589 if (entry == 0) { 1590 tmpe = map->header.next; 1591 entcount = map->nentries; 1592 while (entcount-- && (tmpe != &map->header)) { 1593 if( _vm_object_in_map(map, object, tmpe)) { 1594 return 1; 1595 } 1596 tmpe = tmpe->next; 1597 } 1598 } else if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) { 1599 tmpm = entry->object.sub_map; 1600 tmpe = tmpm->header.next; 1601 entcount = tmpm->nentries; 1602 while (entcount-- && tmpe != &tmpm->header) { 1603 if( _vm_object_in_map(tmpm, object, tmpe)) { 1604 return 1; 1605 } 1606 tmpe = tmpe->next; 1607 } 1608 } else if ((obj = entry->object.vm_object) != NULL) { 1609 for(; obj; obj=obj->backing_object) 1610 if( obj == object) { 1611 return 1; 1612 } 1613 } 1614 return 0; 1615 } 1616 1617 static int 1618 vm_object_in_map( object) 1619 vm_object_t object; 1620 { 1621 struct proc *p; 1622 ALLPROC_LOCK(AP_SHARED); 1623 LIST_FOREACH(p, &allproc, p_list) { 1624 if( !p->p_vmspace /* || (p->p_flag & (P_SYSTEM|P_WEXIT)) */) 1625 continue; 1626 if( _vm_object_in_map(&p->p_vmspace->vm_map, object, 0)) { 1627 ALLPROC_LOCK(AP_RELEASE); 1628 return 1; 1629 } 1630 } 1631 ALLPROC_LOCK(AP_RELEASE); 1632 if( _vm_object_in_map( kernel_map, object, 0)) 1633 return 1; 1634 if( _vm_object_in_map( kmem_map, object, 0)) 1635 return 1; 1636 if( _vm_object_in_map( pager_map, object, 0)) 1637 return 1; 1638 if( _vm_object_in_map( buffer_map, object, 0)) 1639 return 1; 1640 if( _vm_object_in_map( mb_map, object, 0)) 1641 return 1; 1642 return 0; 1643 } 1644 1645 DB_SHOW_COMMAND(vmochk, vm_object_check) 1646 { 1647 vm_object_t object; 1648 1649 /* 1650 * make sure that internal objs are in a map somewhere 1651 * and none have zero ref counts. 1652 */ 1653 for (object = TAILQ_FIRST(&vm_object_list); 1654 object != NULL; 1655 object = TAILQ_NEXT(object, object_list)) { 1656 if (object->handle == NULL && 1657 (object->type == OBJT_DEFAULT || object->type == OBJT_SWAP)) { 1658 if (object->ref_count == 0) { 1659 db_printf("vmochk: internal obj has zero ref count: %ld\n", 1660 (long)object->size); 1661 } 1662 if (!vm_object_in_map(object)) { 1663 db_printf( 1664 "vmochk: internal obj is not in a map: " 1665 "ref: %d, size: %lu: 0x%lx, backing_object: %p\n", 1666 object->ref_count, (u_long)object->size, 1667 (u_long)object->size, 1668 (void *)object->backing_object); 1669 } 1670 } 1671 } 1672 } 1673 1674 /* 1675 * vm_object_print: [ debug ] 1676 */ 1677 DB_SHOW_COMMAND(object, vm_object_print_static) 1678 { 1679 /* XXX convert args. */ 1680 vm_object_t object = (vm_object_t)addr; 1681 boolean_t full = have_addr; 1682 1683 vm_page_t p; 1684 1685 /* XXX count is an (unused) arg. Avoid shadowing it. */ 1686 #define count was_count 1687 1688 int count; 1689 1690 if (object == NULL) 1691 return; 1692 1693 db_iprintf( 1694 "Object %p: type=%d, size=0x%lx, res=%d, ref=%d, flags=0x%x\n", 1695 object, (int)object->type, (u_long)object->size, 1696 object->resident_page_count, object->ref_count, object->flags); 1697 /* 1698 * XXX no %qd in kernel. Truncate object->backing_object_offset. 1699 */ 1700 db_iprintf(" sref=%d, backing_object(%d)=(%p)+0x%lx\n", 1701 object->shadow_count, 1702 object->backing_object ? object->backing_object->ref_count : 0, 1703 object->backing_object, (long)object->backing_object_offset); 1704 1705 if (!full) 1706 return; 1707 1708 db_indent += 2; 1709 count = 0; 1710 TAILQ_FOREACH(p, &object->memq, listq) { 1711 if (count == 0) 1712 db_iprintf("memory:="); 1713 else if (count == 6) { 1714 db_printf("\n"); 1715 db_iprintf(" ..."); 1716 count = 0; 1717 } else 1718 db_printf(","); 1719 count++; 1720 1721 db_printf("(off=0x%lx,page=0x%lx)", 1722 (u_long) p->pindex, (u_long) VM_PAGE_TO_PHYS(p)); 1723 } 1724 if (count != 0) 1725 db_printf("\n"); 1726 db_indent -= 2; 1727 } 1728 1729 /* XXX. */ 1730 #undef count 1731 1732 /* XXX need this non-static entry for calling from vm_map_print. */ 1733 void 1734 vm_object_print(addr, have_addr, count, modif) 1735 /* db_expr_t */ long addr; 1736 boolean_t have_addr; 1737 /* db_expr_t */ long count; 1738 char *modif; 1739 { 1740 vm_object_print_static(addr, have_addr, count, modif); 1741 } 1742 1743 DB_SHOW_COMMAND(vmopag, vm_object_print_pages) 1744 { 1745 vm_object_t object; 1746 int nl = 0; 1747 int c; 1748 for (object = TAILQ_FIRST(&vm_object_list); 1749 object != NULL; 1750 object = TAILQ_NEXT(object, object_list)) { 1751 vm_pindex_t idx, fidx; 1752 vm_pindex_t osize; 1753 vm_offset_t pa = -1, padiff; 1754 int rcount; 1755 vm_page_t m; 1756 1757 db_printf("new object: %p\n", (void *)object); 1758 if ( nl > 18) { 1759 c = cngetc(); 1760 if (c != ' ') 1761 return; 1762 nl = 0; 1763 } 1764 nl++; 1765 rcount = 0; 1766 fidx = 0; 1767 osize = object->size; 1768 if (osize > 128) 1769 osize = 128; 1770 for(idx=0;idx<osize;idx++) { 1771 m = vm_page_lookup(object, idx); 1772 if (m == NULL) { 1773 if (rcount) { 1774 db_printf(" index(%ld)run(%d)pa(0x%lx)\n", 1775 (long)fidx, rcount, (long)pa); 1776 if ( nl > 18) { 1777 c = cngetc(); 1778 if (c != ' ') 1779 return; 1780 nl = 0; 1781 } 1782 nl++; 1783 rcount = 0; 1784 } 1785 continue; 1786 } 1787 1788 1789 if (rcount && 1790 (VM_PAGE_TO_PHYS(m) == pa + rcount * PAGE_SIZE)) { 1791 ++rcount; 1792 continue; 1793 } 1794 if (rcount) { 1795 padiff = pa + rcount * PAGE_SIZE - VM_PAGE_TO_PHYS(m); 1796 padiff >>= PAGE_SHIFT; 1797 padiff &= PQ_L2_MASK; 1798 if (padiff == 0) { 1799 pa = VM_PAGE_TO_PHYS(m) - rcount * PAGE_SIZE; 1800 ++rcount; 1801 continue; 1802 } 1803 db_printf(" index(%ld)run(%d)pa(0x%lx)", 1804 (long)fidx, rcount, (long)pa); 1805 db_printf("pd(%ld)\n", (long)padiff); 1806 if ( nl > 18) { 1807 c = cngetc(); 1808 if (c != ' ') 1809 return; 1810 nl = 0; 1811 } 1812 nl++; 1813 } 1814 fidx = idx; 1815 pa = VM_PAGE_TO_PHYS(m); 1816 rcount = 1; 1817 } 1818 if (rcount) { 1819 db_printf(" index(%ld)run(%d)pa(0x%lx)\n", 1820 (long)fidx, rcount, (long)pa); 1821 if ( nl > 18) { 1822 c = cngetc(); 1823 if (c != ' ') 1824 return; 1825 nl = 0; 1826 } 1827 nl++; 1828 } 1829 } 1830 } 1831 #endif /* DDB */ 1832