uvm_page.c revision 1.7
1/* $NetBSD: uvm_page.c,v 1.23 1999/05/25 01:34:13 thorpej Exp $ */ 2 3/* 4 * Copyright (c) 1997 Charles D. Cranor and Washington University. 5 * Copyright (c) 1991, 1993, The Regents of the University of California. 6 * 7 * All rights reserved. 8 * 9 * This code is derived from software contributed to Berkeley by 10 * The Mach Operating System project at Carnegie-Mellon University. 11 * 12 * Redistribution and use in source and binary forms, with or without 13 * modification, are permitted provided that the following conditions 14 * are met: 15 * 1. Redistributions of source code must retain the above copyright 16 * notice, this list of conditions and the following disclaimer. 17 * 2. Redistributions in binary form must reproduce the above copyright 18 * notice, this list of conditions and the following disclaimer in the 19 * documentation and/or other materials provided with the distribution. 20 * 3. All advertising materials mentioning features or use of this software 21 * must display the following acknowledgement: 22 * This product includes software developed by Charles D. Cranor, 23 * Washington University, the University of California, Berkeley and 24 * its contributors. 25 * 4. Neither the name of the University nor the names of its contributors 26 * may be used to endorse or promote products derived from this software 27 * without specific prior written permission. 28 * 29 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 30 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 31 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 32 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 33 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 34 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 35 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 36 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 37 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 38 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 39 * SUCH DAMAGE. 40 * 41 * @(#)vm_page.c 8.3 (Berkeley) 3/21/94 42 * from: Id: uvm_page.c,v 1.1.2.18 1998/02/06 05:24:42 chs Exp 43 * 44 * 45 * Copyright (c) 1987, 1990 Carnegie-Mellon University. 46 * All rights reserved. 47 * 48 * Permission to use, copy, modify and distribute this software and 49 * its documentation is hereby granted, provided that both the copyright 50 * notice and this permission notice appear in all copies of the 51 * software, derivative works or modified versions, and any portions 52 * thereof, and that both notices appear in supporting documentation. 53 * 54 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS" 55 * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND 56 * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE. 57 * 58 * Carnegie Mellon requests users of this software to return to 59 * 60 * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU 61 * School of Computer Science 62 * Carnegie Mellon University 63 * Pittsburgh PA 15213-3890 64 * 65 * any improvements or extensions that they make and grant Carnegie the 66 * rights to redistribute these changes. 67 */ 68 69/* 70 * uvm_page.c: page ops. 71 */ 72 73#include <sys/param.h> 74#include <sys/systm.h> 75#include <sys/malloc.h> 76#include <sys/proc.h> 77 78#include <vm/vm.h> 79#include <vm/vm_page.h> 80#include <vm/vm_kern.h> 81 82#define UVM_PAGE /* pull in uvm_page.h functions */ 83#include <uvm/uvm.h> 84 85/* 86 * global vars... XXXCDC: move to uvm. structure. 87 */ 88 89/* 90 * physical memory config is stored in vm_physmem. 91 */ 92 93struct vm_physseg vm_physmem[VM_PHYSSEG_MAX]; /* XXXCDC: uvm.physmem */ 94int vm_nphysseg = 0; /* XXXCDC: uvm.nphysseg */ 95 96/* 97 * local variables 98 */ 99 100/* 101 * these variables record the values returned by vm_page_bootstrap, 102 * for debugging purposes. The implementation of uvm_pageboot_alloc 103 * and pmap_startup here also uses them internally. 104 */ 105 106static vaddr_t virtual_space_start; 107static vaddr_t virtual_space_end; 108 109/* 110 * we use a hash table with only one bucket during bootup. we will 111 * later rehash (resize) the hash table once malloc() is ready. 112 * we static allocate the bootstrap bucket below... 113 */ 114 115static struct pglist uvm_bootbucket; 116 117/* 118 * local prototypes 119 */ 120 121static void uvm_pageinsert __P((struct vm_page *)); 122 123 124/* 125 * inline functions 126 */ 127 128/* 129 * uvm_pageinsert: insert a page in the object and the hash table 130 * 131 * => caller must lock object 132 * => caller must lock page queues 133 * => call should have already set pg's object and offset pointers 134 * and bumped the version counter 135 */ 136 137__inline static void 138uvm_pageinsert(pg) 139 struct vm_page *pg; 140{ 141 struct pglist *buck; 142 int s; 143 144#ifdef DIAGNOSTIC 145 if (pg->flags & PG_TABLED) 146 panic("uvm_pageinsert: already inserted"); 147#endif 148 149 buck = &uvm.page_hash[uvm_pagehash(pg->uobject,pg->offset)]; 150 s = splimp(); 151 simple_lock(&uvm.hashlock); 152 TAILQ_INSERT_TAIL(buck, pg, hashq); /* put in hash */ 153 simple_unlock(&uvm.hashlock); 154 splx(s); 155 156 TAILQ_INSERT_TAIL(&pg->uobject->memq, pg, listq); /* put in object */ 157 pg->flags |= PG_TABLED; 158 pg->uobject->uo_npages++; 159 160} 161 162/* 163 * uvm_page_remove: remove page from object and hash 164 * 165 * => caller must lock object 166 * => caller must lock page queues 167 */ 168 169void __inline 170uvm_pageremove(pg) 171 struct vm_page *pg; 172{ 173 struct pglist *buck; 174 int s; 175 176#ifdef DIAGNOSTIC 177 if ((pg->flags & (PG_FAULTING)) != 0) 178 panic("uvm_pageremove: page is faulting"); 179#endif 180 181 if ((pg->flags & PG_TABLED) == 0) 182 return; /* XXX: log */ 183 184 buck = &uvm.page_hash[uvm_pagehash(pg->uobject,pg->offset)]; 185 s = splimp(); 186 simple_lock(&uvm.hashlock); 187 TAILQ_REMOVE(buck, pg, hashq); 188 simple_unlock(&uvm.hashlock); 189 splx(s); 190 191 /* object should be locked */ 192 TAILQ_REMOVE(&pg->uobject->memq, pg, listq); 193 194 pg->flags &= ~PG_TABLED; 195 pg->uobject->uo_npages--; 196 pg->uobject = NULL; 197 pg->version++; 198 199} 200 201/* 202 * uvm_page_init: init the page system. called from uvm_init(). 203 * 204 * => we return the range of kernel virtual memory in kvm_startp/kvm_endp 205 */ 206 207void 208uvm_page_init(kvm_startp, kvm_endp) 209 vaddr_t *kvm_startp, *kvm_endp; 210{ 211 int freepages, pagecount; 212 vm_page_t pagearray; 213 int lcv, n, i; 214 paddr_t paddr; 215 216 217 /* 218 * step 1: init the page queues and page queue locks 219 */ 220 for (lcv = 0; lcv < VM_NFREELIST; lcv++) 221 TAILQ_INIT(&uvm.page_free[lcv]); 222 TAILQ_INIT(&uvm.page_active); 223 TAILQ_INIT(&uvm.page_inactive_swp); 224 TAILQ_INIT(&uvm.page_inactive_obj); 225 simple_lock_init(&uvm.pageqlock); 226 simple_lock_init(&uvm.fpageqlock); 227 228 /* 229 * step 2: init the <obj,offset> => <page> hash table. for now 230 * we just have one bucket (the bootstrap bucket). later on we 231 * will malloc() new buckets as we dynamically resize the hash table. 232 */ 233 234 uvm.page_nhash = 1; /* 1 bucket */ 235 uvm.page_hashmask = 0; /* mask for hash function */ 236 uvm.page_hash = &uvm_bootbucket; /* install bootstrap bucket */ 237 TAILQ_INIT(uvm.page_hash); /* init hash table */ 238 simple_lock_init(&uvm.hashlock); /* init hash table lock */ 239 240 /* 241 * step 3: allocate vm_page structures. 242 */ 243 244 /* 245 * sanity check: 246 * before calling this function the MD code is expected to register 247 * some free RAM with the uvm_page_physload() function. our job 248 * now is to allocate vm_page structures for this memory. 249 */ 250 251 if (vm_nphysseg == 0) 252 panic("vm_page_bootstrap: no memory pre-allocated"); 253 254 /* 255 * first calculate the number of free pages... 256 * 257 * note that we use start/end rather than avail_start/avail_end. 258 * this allows us to allocate extra vm_page structures in case we 259 * want to return some memory to the pool after booting. 260 */ 261 262 freepages = 0; 263 for (lcv = 0 ; lcv < vm_nphysseg ; lcv++) 264 freepages += (vm_physmem[lcv].end - vm_physmem[lcv].start); 265 266 /* 267 * we now know we have (PAGE_SIZE * freepages) bytes of memory we can 268 * use. for each page of memory we use we need a vm_page structure. 269 * thus, the total number of pages we can use is the total size of 270 * the memory divided by the PAGE_SIZE plus the size of the vm_page 271 * structure. we add one to freepages as a fudge factor to avoid 272 * truncation errors (since we can only allocate in terms of whole 273 * pages). 274 */ 275 276 pagecount = ((freepages + 1) << PAGE_SHIFT) / 277 (PAGE_SIZE + sizeof(struct vm_page)); 278 pagearray = (vm_page_t)uvm_pageboot_alloc(pagecount * 279 sizeof(struct vm_page)); 280 bzero(pagearray, pagecount * sizeof(struct vm_page)); 281 282 /* 283 * step 4: init the vm_page structures and put them in the correct 284 * place... 285 */ 286 287 for (lcv = 0 ; lcv < vm_nphysseg ; lcv++) { 288 289 n = vm_physmem[lcv].end - vm_physmem[lcv].start; 290 if (n > pagecount) { 291 printf("uvm_page_init: lost %d page(s) in init\n", 292 n - pagecount); 293 panic("uvm_page_init"); /* XXXCDC: shouldn't happen? */ 294 /* n = pagecount; */ 295 } 296 /* set up page array pointers */ 297 vm_physmem[lcv].pgs = pagearray; 298 pagearray += n; 299 pagecount -= n; 300 vm_physmem[lcv].lastpg = vm_physmem[lcv].pgs + (n - 1); 301 302 /* init and free vm_pages (we've already zeroed them) */ 303 paddr = ptoa(vm_physmem[lcv].start); 304 for (i = 0 ; i < n ; i++, paddr += PAGE_SIZE) { 305 vm_physmem[lcv].pgs[i].phys_addr = paddr; 306 if (atop(paddr) >= vm_physmem[lcv].avail_start && 307 atop(paddr) <= vm_physmem[lcv].avail_end) { 308 uvmexp.npages++; 309 /* add page to free pool */ 310 uvm_pagefree(&vm_physmem[lcv].pgs[i]); 311 } 312 } 313 } 314 /* 315 * step 5: pass up the values of virtual_space_start and 316 * virtual_space_end (obtained by uvm_pageboot_alloc) to the upper 317 * layers of the VM. 318 */ 319 320 *kvm_startp = round_page(virtual_space_start); 321 *kvm_endp = trunc_page(virtual_space_end); 322 323 /* 324 * step 6: init pagedaemon lock 325 */ 326 327 simple_lock_init(&uvm.pagedaemon_lock); 328 329 /* 330 * step 7: init reserve thresholds 331 * XXXCDC - values may need adjusting 332 */ 333 uvmexp.reserve_pagedaemon = 1; 334 uvmexp.reserve_kernel = 5; 335 336 /* 337 * done! 338 */ 339 340} 341 342/* 343 * uvm_setpagesize: set the page size 344 * 345 * => sets page_shift and page_mask from uvmexp.pagesize. 346 * => XXXCDC: move global vars. 347 */ 348 349void 350uvm_setpagesize() 351{ 352 if (uvmexp.pagesize == 0) 353 uvmexp.pagesize = DEFAULT_PAGE_SIZE; 354 uvmexp.pagemask = uvmexp.pagesize - 1; 355 if ((uvmexp.pagemask & uvmexp.pagesize) != 0) 356 panic("uvm_setpagesize: page size not a power of two"); 357 for (uvmexp.pageshift = 0; ; uvmexp.pageshift++) 358 if ((1 << uvmexp.pageshift) == uvmexp.pagesize) 359 break; 360} 361 362/* 363 * uvm_pageboot_alloc: steal memory from physmem for bootstrapping 364 */ 365 366vaddr_t 367uvm_pageboot_alloc(size) 368 vsize_t size; 369{ 370#if defined(PMAP_STEAL_MEMORY) 371 vaddr_t addr; 372 373 /* 374 * defer bootstrap allocation to MD code (it may want to allocate 375 * from a direct-mapped segment). pmap_steal_memory should round 376 * off virtual_space_start/virtual_space_end. 377 */ 378 379 addr = pmap_steal_memory(size, &virtual_space_start, 380 &virtual_space_end); 381 382 return(addr); 383 384#else /* !PMAP_STEAL_MEMORY */ 385 386 static boolean_t initialized = FALSE; 387 vaddr_t addr, vaddr; 388 paddr_t paddr; 389 390 /* round to page size */ 391 size = round_page(size); 392 393 /* 394 * on first call to this function, initialize ourselves. 395 */ 396 if (initialized == FALSE) { 397 pmap_virtual_space(&virtual_space_start, &virtual_space_end); 398 399 /* round it the way we like it */ 400 virtual_space_start = round_page(virtual_space_start); 401 virtual_space_end = trunc_page(virtual_space_end); 402 403 initialized = TRUE; 404 } 405 406 /* 407 * allocate virtual memory for this request 408 */ 409 if (virtual_space_start == virtual_space_end || 410 (virtual_space_end - virtual_space_start) < size) 411 panic("uvm_pageboot_alloc: out of virtual space"); 412 413 addr = virtual_space_start; 414 415#ifdef PMAP_GROWKERNEL 416 /* 417 * If the kernel pmap can't map the requested space, 418 * then allocate more resources for it. 419 */ 420 if (uvm_maxkaddr < (addr + size)) { 421 uvm_maxkaddr = pmap_growkernel(addr + size); 422 if (uvm_maxkaddr < (addr + size)) 423 panic("uvm_pageboot_alloc: pmap_growkernel() failed"); 424 } 425#endif 426 427 virtual_space_start += size; 428 429 /* 430 * allocate and mapin physical pages to back new virtual pages 431 */ 432 433 for (vaddr = round_page(addr) ; vaddr < addr + size ; 434 vaddr += PAGE_SIZE) { 435 436 if (!uvm_page_physget(&paddr)) 437 panic("uvm_pageboot_alloc: out of memory"); 438 439 /* XXX: should be wired, but some pmaps don't like that ... */ 440#if defined(PMAP_NEW) 441 /* 442 * Note this memory is no longer managed, so using 443 * pmap_kenter is safe. 444 */ 445 pmap_kenter_pa(vaddr, paddr, VM_PROT_READ|VM_PROT_WRITE); 446#else 447 pmap_enter(pmap_kernel(), vaddr, paddr, 448 VM_PROT_READ|VM_PROT_WRITE, FALSE, 449 VM_PROT_READ|VM_PROT_WRITE); 450#endif 451 452 } 453 return(addr); 454#endif /* PMAP_STEAL_MEMORY */ 455} 456 457#if !defined(PMAP_STEAL_MEMORY) 458/* 459 * uvm_page_physget: "steal" one page from the vm_physmem structure. 460 * 461 * => attempt to allocate it off the end of a segment in which the "avail" 462 * values match the start/end values. if we can't do that, then we 463 * will advance both values (making them equal, and removing some 464 * vm_page structures from the non-avail area). 465 * => return false if out of memory. 466 */ 467 468boolean_t 469uvm_page_physget(paddrp) 470 paddr_t *paddrp; 471{ 472 int lcv, x; 473 474 /* pass 1: try allocating from a matching end */ 475#if (VM_PHYSSEG_STRAT == VM_PSTRAT_BIGFIRST) 476 for (lcv = vm_nphysseg - 1 ; lcv >= 0 ; lcv--) 477#else 478 for (lcv = 0 ; lcv < vm_nphysseg ; lcv++) 479#endif 480 { 481 482 if (vm_physmem[lcv].pgs) 483 panic("vm_page_physget: called _after_ bootstrap"); 484 485 /* try from front */ 486 if (vm_physmem[lcv].avail_start == vm_physmem[lcv].start && 487 vm_physmem[lcv].avail_start < vm_physmem[lcv].avail_end) { 488 *paddrp = ptoa(vm_physmem[lcv].avail_start); 489 vm_physmem[lcv].avail_start++; 490 vm_physmem[lcv].start++; 491 /* nothing left? nuke it */ 492 if (vm_physmem[lcv].avail_start == 493 vm_physmem[lcv].end) { 494 if (vm_nphysseg == 1) 495 panic("vm_page_physget: out of memory!"); 496 vm_nphysseg--; 497 for (x = lcv ; x < vm_nphysseg ; x++) 498 /* structure copy */ 499 vm_physmem[x] = vm_physmem[x+1]; 500 } 501 return (TRUE); 502 } 503 504 /* try from rear */ 505 if (vm_physmem[lcv].avail_end == vm_physmem[lcv].end && 506 vm_physmem[lcv].avail_start < vm_physmem[lcv].avail_end) { 507 *paddrp = ptoa(vm_physmem[lcv].avail_end - 1); 508 vm_physmem[lcv].avail_end--; 509 vm_physmem[lcv].end--; 510 /* nothing left? nuke it */ 511 if (vm_physmem[lcv].avail_end == 512 vm_physmem[lcv].start) { 513 if (vm_nphysseg == 1) 514 panic("vm_page_physget: out of memory!"); 515 vm_nphysseg--; 516 for (x = lcv ; x < vm_nphysseg ; x++) 517 /* structure copy */ 518 vm_physmem[x] = vm_physmem[x+1]; 519 } 520 return (TRUE); 521 } 522 } 523 524 /* pass2: forget about matching ends, just allocate something */ 525#if (VM_PHYSSEG_STRAT == VM_PSTRAT_BIGFIRST) 526 for (lcv = vm_nphysseg - 1 ; lcv >= 0 ; lcv--) 527#else 528 for (lcv = 0 ; lcv < vm_nphysseg ; lcv++) 529#endif 530 { 531 532 /* any room in this bank? */ 533 if (vm_physmem[lcv].avail_start >= vm_physmem[lcv].avail_end) 534 continue; /* nope */ 535 536 *paddrp = ptoa(vm_physmem[lcv].avail_start); 537 vm_physmem[lcv].avail_start++; 538 /* truncate! */ 539 vm_physmem[lcv].start = vm_physmem[lcv].avail_start; 540 541 /* nothing left? nuke it */ 542 if (vm_physmem[lcv].avail_start == vm_physmem[lcv].end) { 543 if (vm_nphysseg == 1) 544 panic("vm_page_physget: out of memory!"); 545 vm_nphysseg--; 546 for (x = lcv ; x < vm_nphysseg ; x++) 547 /* structure copy */ 548 vm_physmem[x] = vm_physmem[x+1]; 549 } 550 return (TRUE); 551 } 552 553 return (FALSE); /* whoops! */ 554} 555#endif /* PMAP_STEAL_MEMORY */ 556 557/* 558 * uvm_page_physload: load physical memory into VM system 559 * 560 * => all args are PFs 561 * => all pages in start/end get vm_page structures 562 * => areas marked by avail_start/avail_end get added to the free page pool 563 * => we are limited to VM_PHYSSEG_MAX physical memory segments 564 */ 565 566void 567uvm_page_physload(start, end, avail_start, avail_end, free_list) 568 vaddr_t start, end, avail_start, avail_end; 569 int free_list; 570{ 571 int preload, lcv; 572 psize_t npages; 573 struct vm_page *pgs; 574 struct vm_physseg *ps; 575 576 if (uvmexp.pagesize == 0) 577 panic("vm_page_physload: page size not set!"); 578 579 if (free_list >= VM_NFREELIST || free_list < VM_FREELIST_DEFAULT) 580 panic("uvm_page_physload: bad free list %d\n", free_list); 581 582 /* 583 * do we have room? 584 */ 585 if (vm_nphysseg == VM_PHYSSEG_MAX) { 586 printf("vm_page_physload: unable to load physical memory " 587 "segment\n"); 588 printf("\t%d segments allocated, ignoring 0x%lx -> 0x%lx\n", 589 VM_PHYSSEG_MAX, start, end); 590 return; 591 } 592 593 /* 594 * check to see if this is a "preload" (i.e. uvm_mem_init hasn't been 595 * called yet, so malloc is not available). 596 */ 597 for (lcv = 0 ; lcv < vm_nphysseg ; lcv++) { 598 if (vm_physmem[lcv].pgs) 599 break; 600 } 601 preload = (lcv == vm_nphysseg); 602 603 /* 604 * if VM is already running, attempt to malloc() vm_page structures 605 */ 606 if (!preload) { 607#if defined(VM_PHYSSEG_NOADD) 608 panic("vm_page_physload: tried to add RAM after vm_mem_init"); 609#else 610 /* XXXCDC: need some sort of lockout for this case */ 611 paddr_t paddr; 612 npages = end - start; /* # of pages */ 613 MALLOC(pgs, struct vm_page *, sizeof(struct vm_page) * npages, 614 M_VMPAGE, M_NOWAIT); 615 if (pgs == NULL) { 616 printf("vm_page_physload: can not malloc vm_page " 617 "structs for segment\n"); 618 printf("\tignoring 0x%lx -> 0x%lx\n", start, end); 619 return; 620 } 621 /* zero data, init phys_addr and free_list, and free pages */ 622 bzero(pgs, sizeof(struct vm_page) * npages); 623 for (lcv = 0, paddr = ptoa(start) ; 624 lcv < npages ; lcv++, paddr += PAGE_SIZE) { 625 pgs[lcv].phys_addr = paddr; 626 pgs[lcv].free_list = free_list; 627 if (atop(paddr) >= avail_start && 628 atop(paddr) <= avail_end) 629 uvm_pagefree(&pgs[lcv]); 630 } 631 /* XXXCDC: incomplete: need to update uvmexp.free, what else? */ 632 /* XXXCDC: need hook to tell pmap to rebuild pv_list, etc... */ 633#endif 634 } else { 635 636 /* gcc complains if these don't get init'd */ 637 pgs = NULL; 638 npages = 0; 639 640 } 641 642 /* 643 * now insert us in the proper place in vm_physmem[] 644 */ 645 646#if (VM_PHYSSEG_STRAT == VM_PSTRAT_RANDOM) 647 648 /* random: put it at the end (easy!) */ 649 ps = &vm_physmem[vm_nphysseg]; 650 651#elif (VM_PHYSSEG_STRAT == VM_PSTRAT_BSEARCH) 652 653 { 654 int x; 655 /* sort by address for binary search */ 656 for (lcv = 0 ; lcv < vm_nphysseg ; lcv++) 657 if (start < vm_physmem[lcv].start) 658 break; 659 ps = &vm_physmem[lcv]; 660 /* move back other entries, if necessary ... */ 661 for (x = vm_nphysseg ; x > lcv ; x--) 662 /* structure copy */ 663 vm_physmem[x] = vm_physmem[x - 1]; 664 } 665 666#elif (VM_PHYSSEG_STRAT == VM_PSTRAT_BIGFIRST) 667 668 { 669 int x; 670 /* sort by largest segment first */ 671 for (lcv = 0 ; lcv < vm_nphysseg ; lcv++) 672 if ((end - start) > 673 (vm_physmem[lcv].end - vm_physmem[lcv].start)) 674 break; 675 ps = &vm_physmem[lcv]; 676 /* move back other entries, if necessary ... */ 677 for (x = vm_nphysseg ; x > lcv ; x--) 678 /* structure copy */ 679 vm_physmem[x] = vm_physmem[x - 1]; 680 } 681 682#else 683 684 panic("vm_page_physload: unknown physseg strategy selected!"); 685 686#endif 687 688 ps->start = start; 689 ps->end = end; 690 ps->avail_start = avail_start; 691 ps->avail_end = avail_end; 692 if (preload) { 693 ps->pgs = NULL; 694 } else { 695 ps->pgs = pgs; 696 ps->lastpg = pgs + npages - 1; 697 } 698 ps->free_list = free_list; 699 vm_nphysseg++; 700 701 /* 702 * done! 703 */ 704 705 if (!preload) 706 uvm_page_rehash(); 707 708 return; 709} 710 711/* 712 * uvm_page_rehash: reallocate hash table based on number of free pages. 713 */ 714 715void 716uvm_page_rehash() 717{ 718 int freepages, lcv, bucketcount, s, oldcount; 719 struct pglist *newbuckets, *oldbuckets; 720 struct vm_page *pg; 721 722 /* 723 * compute number of pages that can go in the free pool 724 */ 725 726 freepages = 0; 727 for (lcv = 0 ; lcv < vm_nphysseg ; lcv++) 728 freepages += 729 (vm_physmem[lcv].avail_end - vm_physmem[lcv].avail_start); 730 731 /* 732 * compute number of buckets needed for this number of pages 733 */ 734 735 bucketcount = 1; 736 while (bucketcount < freepages) 737 bucketcount = bucketcount * 2; 738 739 /* 740 * malloc new buckets 741 */ 742 743 MALLOC(newbuckets, struct pglist *, sizeof(struct pglist) * bucketcount, 744 M_VMPBUCKET, M_NOWAIT); 745 if (newbuckets == NULL) { 746 printf("vm_page_physrehash: WARNING: could not grow page " 747 "hash table\n"); 748 return; 749 } 750 for (lcv = 0 ; lcv < bucketcount ; lcv++) 751 TAILQ_INIT(&newbuckets[lcv]); 752 753 /* 754 * now replace the old buckets with the new ones and rehash everything 755 */ 756 757 s = splimp(); 758 simple_lock(&uvm.hashlock); 759 /* swap old for new ... */ 760 oldbuckets = uvm.page_hash; 761 oldcount = uvm.page_nhash; 762 uvm.page_hash = newbuckets; 763 uvm.page_nhash = bucketcount; 764 uvm.page_hashmask = bucketcount - 1; /* power of 2 */ 765 766 /* ... and rehash */ 767 for (lcv = 0 ; lcv < oldcount ; lcv++) { 768 while ((pg = oldbuckets[lcv].tqh_first) != NULL) { 769 TAILQ_REMOVE(&oldbuckets[lcv], pg, hashq); 770 TAILQ_INSERT_TAIL( 771 &uvm.page_hash[uvm_pagehash(pg->uobject, pg->offset)], 772 pg, hashq); 773 } 774 } 775 simple_unlock(&uvm.hashlock); 776 splx(s); 777 778 /* 779 * free old bucket array if we malloc'd it previously 780 */ 781 782 if (oldbuckets != &uvm_bootbucket) 783 FREE(oldbuckets, M_VMPBUCKET); 784 785 /* 786 * done 787 */ 788 return; 789} 790 791 792#if 1 /* XXXCDC: TMP TMP TMP DEBUG DEBUG DEBUG */ 793 794void uvm_page_physdump __P((void)); /* SHUT UP GCC */ 795 796/* call from DDB */ 797void 798uvm_page_physdump() 799{ 800 int lcv; 801 802 printf("rehash: physical memory config [segs=%d of %d]:\n", 803 vm_nphysseg, VM_PHYSSEG_MAX); 804 for (lcv = 0 ; lcv < vm_nphysseg ; lcv++) 805 printf("0x%lx->0x%lx [0x%lx->0x%lx]\n", vm_physmem[lcv].start, 806 vm_physmem[lcv].end, vm_physmem[lcv].avail_start, 807 vm_physmem[lcv].avail_end); 808 printf("STRATEGY = "); 809 switch (VM_PHYSSEG_STRAT) { 810 case VM_PSTRAT_RANDOM: printf("RANDOM\n"); break; 811 case VM_PSTRAT_BSEARCH: printf("BSEARCH\n"); break; 812 case VM_PSTRAT_BIGFIRST: printf("BIGFIRST\n"); break; 813 default: printf("<<UNKNOWN>>!!!!\n"); 814 } 815 printf("number of buckets = %d\n", uvm.page_nhash); 816} 817#endif 818 819/* 820 * uvm_pagealloc_strat: allocate vm_page from a particular free list. 821 * 822 * => return null if no pages free 823 * => wake up pagedaemon if number of free pages drops below low water mark 824 * => if obj != NULL, obj must be locked (to put in hash) 825 * => if anon != NULL, anon must be locked (to put in anon) 826 * => only one of obj or anon can be non-null 827 * => caller must activate/deactivate page if it is not wired. 828 * => free_list is ignored if strat == UVM_PGA_STRAT_NORMAL. 829 */ 830 831struct vm_page * 832uvm_pagealloc_strat(obj, off, anon, flags, strat, free_list) 833 struct uvm_object *obj; 834 vaddr_t off; 835 int flags; 836 struct vm_anon *anon; 837 int strat, free_list; 838{ 839 int lcv, s; 840 struct vm_page *pg; 841 struct pglist *freeq; 842 boolean_t use_reserve; 843 844#ifdef DIAGNOSTIC 845 /* sanity check */ 846 if (obj && anon) 847 panic("uvm_pagealloc: obj and anon != NULL"); 848#endif 849 850 s = uvm_lock_fpageq(); /* lock free page queue */ 851 852 /* 853 * check to see if we need to generate some free pages waking 854 * the pagedaemon. 855 */ 856 857 if (uvmexp.free < uvmexp.freemin || (uvmexp.free < uvmexp.freetarg && 858 uvmexp.inactive < uvmexp.inactarg)) 859 thread_wakeup(&uvm.pagedaemon); 860 861 /* 862 * fail if any of these conditions is true: 863 * [1] there really are no free pages, or 864 * [2] only kernel "reserved" pages remain and 865 * the page isn't being allocated to a kernel object. 866 * [3] only pagedaemon "reserved" pages remain and 867 * the requestor isn't the pagedaemon. 868 */ 869 870 use_reserve = (flags & UVM_PGA_USERESERVE) || 871 (obj && UVM_OBJ_IS_KERN_OBJECT(obj)); 872 if ((uvmexp.free <= uvmexp.reserve_kernel && !use_reserve) || 873 (uvmexp.free <= uvmexp.reserve_pagedaemon && 874 !(use_reserve && curproc == uvm.pagedaemon_proc))) 875 goto fail; 876 877 again: 878 switch (strat) { 879 case UVM_PGA_STRAT_NORMAL: 880 /* Check all freelists in descending priority order. */ 881 for (lcv = 0; lcv < VM_NFREELIST; lcv++) { 882 freeq = &uvm.page_free[lcv]; 883 if ((pg = freeq->tqh_first) != NULL) 884 goto gotit; 885 } 886 887 /* No pages free! */ 888 goto fail; 889 890 case UVM_PGA_STRAT_ONLY: 891 case UVM_PGA_STRAT_FALLBACK: 892 /* Attempt to allocate from the specified free list. */ 893#ifdef DIAGNOSTIC 894 if (free_list >= VM_NFREELIST || free_list < 0) 895 panic("uvm_pagealloc_strat: bad free list %d", 896 free_list); 897#endif 898 freeq = &uvm.page_free[free_list]; 899 if ((pg = freeq->tqh_first) != NULL) 900 goto gotit; 901 902 /* Fall back, if possible. */ 903 if (strat == UVM_PGA_STRAT_FALLBACK) { 904 strat = UVM_PGA_STRAT_NORMAL; 905 goto again; 906 } 907 908 /* No pages free! */ 909 goto fail; 910 911 default: 912 panic("uvm_pagealloc_strat: bad strat %d", strat); 913 /* NOTREACHED */ 914 } 915 916 gotit: 917 TAILQ_REMOVE(freeq, pg, pageq); 918 uvmexp.free--; 919 920 uvm_unlock_fpageq(s); /* unlock free page queue */ 921 922 pg->offset = off; 923 pg->uobject = obj; 924 pg->uanon = anon; 925 pg->flags = PG_BUSY|PG_CLEAN|PG_FAKE; 926 pg->version++; 927 pg->wire_count = 0; 928 pg->loan_count = 0; 929 if (anon) { 930 anon->u.an_page = pg; 931 pg->pqflags = PQ_ANON; 932 } else { 933 if (obj) 934 uvm_pageinsert(pg); 935 pg->pqflags = 0; 936 } 937#if defined(UVM_PAGE_TRKOWN) 938 pg->owner_tag = NULL; 939#endif 940 UVM_PAGE_OWN(pg, "new alloc"); 941 942 return(pg); 943 944 fail: 945 uvm_unlock_fpageq(s); 946 return (NULL); 947} 948 949/* 950 * uvm_pagealloc_contig: allocate contiguous memory. 951 * 952 * XXX - fix comment. 953 */ 954 955vaddr_t 956uvm_pagealloc_contig(size, low, high, alignment) 957 vaddr_t size; 958 vaddr_t low, high; 959 vaddr_t alignment; 960{ 961 struct pglist pglist; 962 struct vm_page *pg; 963 vaddr_t addr, temp_addr; 964 965 size = round_page(size); 966 967 TAILQ_INIT(&pglist); 968 if (uvm_pglistalloc(size, low, high, alignment, 0, 969 &pglist, 1, FALSE)) 970 return 0; 971 addr = vm_map_min(kernel_map); 972 if (uvm_map(kernel_map, &addr, size, NULL, UVM_UNKNOWN_OFFSET, 973 UVM_MAPFLAG(UVM_PROT_ALL, UVM_PROT_ALL, UVM_INH_NONE, 974 UVM_ADV_RANDOM, 0)) != KERN_SUCCESS) { 975 uvm_pglistfree(&pglist); 976 return 0; 977 } 978 temp_addr = addr; 979 for (pg = TAILQ_FIRST(&pglist); pg != NULL; 980 pg = TAILQ_NEXT(pg, pageq)) { 981 pg->uobject = uvm.kernel_object; 982 pg->offset = temp_addr - vm_map_min(kernel_map); 983 uvm_pageinsert(pg); 984 uvm_pagewire(pg); 985#if defined(PMAP_NEW) 986 pmap_kenter_pa(temp_addr, VM_PAGE_TO_PHYS(pg), 987 VM_PROT_READ|VM_PROT_WRITE); 988#else 989 pmap_enter(pmap_kernel(), temp_addr, VM_PAGE_TO_PHYS(pg), 990 VM_PROT_READ|VM_PROT_WRITE, TRUE, 991 VM_PROT_READ|VM_PROT_WRITE); 992#endif 993 temp_addr += PAGE_SIZE; 994 } 995 return addr; 996} 997 998/* 999 * uvm_pagerealloc: reallocate a page from one object to another 1000 * 1001 * => both objects must be locked 1002 */ 1003 1004void 1005uvm_pagerealloc(pg, newobj, newoff) 1006 struct vm_page *pg; 1007 struct uvm_object *newobj; 1008 vaddr_t newoff; 1009{ 1010 /* 1011 * remove it from the old object 1012 */ 1013 1014 if (pg->uobject) { 1015 uvm_pageremove(pg); 1016 } 1017 1018 /* 1019 * put it in the new object 1020 */ 1021 1022 if (newobj) { 1023 pg->uobject = newobj; 1024 pg->offset = newoff; 1025 pg->version++; 1026 uvm_pageinsert(pg); 1027 } 1028 1029 return; 1030} 1031 1032 1033/* 1034 * uvm_pagefree: free page 1035 * 1036 * => erase page's identity (i.e. remove from hash/object) 1037 * => put page on free list 1038 * => caller must lock owning object (either anon or uvm_object) 1039 * => caller must lock page queues 1040 * => assumes all valid mappings of pg are gone 1041 */ 1042 1043void uvm_pagefree(pg) 1044 1045struct vm_page *pg; 1046 1047{ 1048 int s; 1049 int saved_loan_count = pg->loan_count; 1050 1051 /* 1052 * if the page was an object page (and thus "TABLED"), remove it 1053 * from the object. 1054 */ 1055 1056 if (pg->flags & PG_TABLED) { 1057 1058 /* 1059 * if the object page is on loan we are going to drop ownership. 1060 * it is possible that an anon will take over as owner for this 1061 * page later on. the anon will want a !PG_CLEAN page so that 1062 * it knows it needs to allocate swap if it wants to page the 1063 * page out. 1064 */ 1065 1066 if (saved_loan_count) 1067 pg->flags &= ~PG_CLEAN; /* in case an anon takes over */ 1068 1069 uvm_pageremove(pg); 1070 1071 /* 1072 * if our page was on loan, then we just lost control over it 1073 * (in fact, if it was loaned to an anon, the anon may have 1074 * already taken over ownership of the page by now and thus 1075 * changed the loan_count [e.g. in uvmfault_anonget()]) we just 1076 * return (when the last loan is dropped, then the page can be 1077 * freed by whatever was holding the last loan). 1078 */ 1079 if (saved_loan_count) 1080 return; 1081 1082 } else if (saved_loan_count && (pg->pqflags & PQ_ANON)) { 1083 1084 /* 1085 * if our page is owned by an anon and is loaned out to the 1086 * kernel then we just want to drop ownership and return. 1087 * the kernel must free the page when all its loans clear ... 1088 * note that the kernel can't change the loan status of our 1089 * page as long as we are holding PQ lock. 1090 */ 1091 pg->pqflags &= ~PQ_ANON; 1092 pg->uanon = NULL; 1093 return; 1094 } 1095 1096#ifdef DIAGNOSTIC 1097 if (saved_loan_count) { 1098 printf("uvm_pagefree: warning: freeing page with a loan " 1099 "count of %d\n", saved_loan_count); 1100 panic("uvm_pagefree: loan count"); 1101 } 1102#endif 1103 1104 1105 /* 1106 * now remove the page from the queues 1107 */ 1108 1109 if (pg->pqflags & PQ_ACTIVE) { 1110 TAILQ_REMOVE(&uvm.page_active, pg, pageq); 1111 pg->pqflags &= ~PQ_ACTIVE; 1112 uvmexp.active--; 1113 } 1114 if (pg->pqflags & PQ_INACTIVE) { 1115 if (pg->pqflags & PQ_SWAPBACKED) 1116 TAILQ_REMOVE(&uvm.page_inactive_swp, pg, pageq); 1117 else 1118 TAILQ_REMOVE(&uvm.page_inactive_obj, pg, pageq); 1119 pg->pqflags &= ~PQ_INACTIVE; 1120 uvmexp.inactive--; 1121 } 1122 1123 /* 1124 * if the page was wired, unwire it now. 1125 */ 1126 if (pg->wire_count) 1127 { 1128 pg->wire_count = 0; 1129 uvmexp.wired--; 1130 } 1131 1132 /* 1133 * and put on free queue 1134 */ 1135 1136 s = uvm_lock_fpageq(); 1137 TAILQ_INSERT_TAIL(&uvm.page_free[uvm_page_lookup_freelist(pg)], 1138 pg, pageq); 1139 pg->pqflags = PQ_FREE; 1140#ifdef DEBUG 1141 pg->uobject = (void *)0xdeadbeef; 1142 pg->offset = 0xdeadbeef; 1143 pg->uanon = (void *)0xdeadbeef; 1144#endif 1145 uvmexp.free++; 1146 uvm_unlock_fpageq(s); 1147} 1148 1149#if defined(UVM_PAGE_TRKOWN) 1150/* 1151 * uvm_page_own: set or release page ownership 1152 * 1153 * => this is a debugging function that keeps track of who sets PG_BUSY 1154 * and where they do it. it can be used to track down problems 1155 * such a process setting "PG_BUSY" and never releasing it. 1156 * => page's object [if any] must be locked 1157 * => if "tag" is NULL then we are releasing page ownership 1158 */ 1159void 1160uvm_page_own(pg, tag) 1161 struct vm_page *pg; 1162 char *tag; 1163{ 1164 /* gain ownership? */ 1165 if (tag) { 1166 if (pg->owner_tag) { 1167 printf("uvm_page_own: page %p already owned " 1168 "by proc %d [%s]\n", pg, 1169 pg->owner, pg->owner_tag); 1170 panic("uvm_page_own"); 1171 } 1172 pg->owner = (curproc) ? curproc->p_pid : (pid_t) -1; 1173 pg->owner_tag = tag; 1174 return; 1175 } 1176 1177 /* drop ownership */ 1178 if (pg->owner_tag == NULL) { 1179 printf("uvm_page_own: dropping ownership of an non-owned " 1180 "page (%p)\n", pg); 1181 panic("uvm_page_own"); 1182 } 1183 pg->owner_tag = NULL; 1184 return; 1185} 1186#endif 1187