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