uvm_page.c revision 1.3
1/* $OpenBSD: uvm_page.c,v 1.3 1999/07/23 14:47:06 ho Exp $ */ 2/* $NetBSD: uvm_page.c,v 1.15 1998/10/18 23:50:00 chs Exp $ */ 3 4/* 5 * XXXCDC: "ROUGH DRAFT" QUALITY UVM PRE-RELEASE FILE! 6 * >>>USE AT YOUR OWN RISK, WORK IS NOT FINISHED<<< 7 */ 8/* 9 * Copyright (c) 1997 Charles D. Cranor and Washington University. 10 * Copyright (c) 1991, 1993, The Regents of the University of California. 11 * 12 * All rights reserved. 13 * 14 * This code is derived from software contributed to Berkeley by 15 * The Mach Operating System project at Carnegie-Mellon University. 16 * 17 * Redistribution and use in source and binary forms, with or without 18 * modification, are permitted provided that the following conditions 19 * are met: 20 * 1. Redistributions of source code must retain the above copyright 21 * notice, this list of conditions and the following disclaimer. 22 * 2. Redistributions in binary form must reproduce the above copyright 23 * notice, this list of conditions and the following disclaimer in the 24 * documentation and/or other materials provided with the distribution. 25 * 3. All advertising materials mentioning features or use of this software 26 * must display the following acknowledgement: 27 * This product includes software developed by Charles D. Cranor, 28 * Washington University, the University of California, Berkeley and 29 * its contributors. 30 * 4. Neither the name of the University nor the names of its contributors 31 * may be used to endorse or promote products derived from this software 32 * without specific prior written permission. 33 * 34 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 35 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 36 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 37 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 38 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 39 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 40 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 41 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 42 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 43 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 44 * SUCH DAMAGE. 45 * 46 * @(#)vm_page.c 8.3 (Berkeley) 3/21/94 47 * from: Id: uvm_page.c,v 1.1.2.18 1998/02/06 05:24:42 chs Exp 48 * 49 * 50 * Copyright (c) 1987, 1990 Carnegie-Mellon University. 51 * All rights reserved. 52 * 53 * Permission to use, copy, modify and distribute this software and 54 * its documentation is hereby granted, provided that both the copyright 55 * notice and this permission notice appear in all copies of the 56 * software, derivative works or modified versions, and any portions 57 * thereof, and that both notices appear in supporting documentation. 58 * 59 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS" 60 * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND 61 * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE. 62 * 63 * Carnegie Mellon requests users of this software to return to 64 * 65 * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU 66 * School of Computer Science 67 * Carnegie Mellon University 68 * Pittsburgh PA 15213-3890 69 * 70 * any improvements or extensions that they make and grant Carnegie the 71 * rights to redistribute these changes. 72 */ 73 74/* 75 * uvm_page.c: page ops. 76 */ 77 78#include <sys/param.h> 79#include <sys/systm.h> 80#include <sys/malloc.h> 81#include <sys/proc.h> 82 83#include <vm/vm.h> 84#include <vm/vm_page.h> 85#include <vm/vm_kern.h> 86 87#define UVM_PAGE /* pull in uvm_page.h functions */ 88#include <uvm/uvm.h> 89 90/* 91 * global vars... XXXCDC: move to uvm. structure. 92 */ 93 94/* 95 * physical memory config is stored in vm_physmem. 96 */ 97 98struct vm_physseg vm_physmem[VM_PHYSSEG_MAX]; /* XXXCDC: uvm.physmem */ 99int vm_nphysseg = 0; /* XXXCDC: uvm.nphysseg */ 100 101/* 102 * local variables 103 */ 104 105/* 106 * these variables record the values returned by vm_page_bootstrap, 107 * for debugging purposes. The implementation of uvm_pageboot_alloc 108 * and pmap_startup here also uses them internally. 109 */ 110 111static vaddr_t virtual_space_start; 112static vaddr_t virtual_space_end; 113 114/* 115 * we use a hash table with only one bucket during bootup. we will 116 * later rehash (resize) the hash table once malloc() is ready. 117 * we static allocate the bootstrap bucket below... 118 */ 119 120static struct pglist uvm_bootbucket; 121 122/* 123 * local prototypes 124 */ 125 126static void uvm_pageinsert __P((struct vm_page *)); 127 128 129/* 130 * inline functions 131 */ 132 133/* 134 * uvm_pageinsert: insert a page in the object and the hash table 135 * 136 * => caller must lock object 137 * => caller must lock page queues 138 * => call should have already set pg's object and offset pointers 139 * and bumped the version counter 140 */ 141 142__inline static void 143uvm_pageinsert(pg) 144 struct vm_page *pg; 145{ 146 struct pglist *buck; 147 int s; 148 149#ifdef DIAGNOSTIC 150 if (pg->flags & PG_TABLED) 151 panic("uvm_pageinsert: already inserted"); 152#endif 153 154 buck = &uvm.page_hash[uvm_pagehash(pg->uobject,pg->offset)]; 155 s = splimp(); 156 simple_lock(&uvm.hashlock); 157 TAILQ_INSERT_TAIL(buck, pg, hashq); /* put in hash */ 158 simple_unlock(&uvm.hashlock); 159 splx(s); 160 161 TAILQ_INSERT_TAIL(&pg->uobject->memq, pg, listq); /* put in object */ 162 pg->flags |= PG_TABLED; 163 pg->uobject->uo_npages++; 164 165} 166 167/* 168 * uvm_page_remove: remove page from object and hash 169 * 170 * => caller must lock object 171 * => caller must lock page queues 172 */ 173 174void __inline 175uvm_pageremove(pg) 176 struct vm_page *pg; 177{ 178 struct pglist *buck; 179 int s; 180 181#ifdef DIAGNOSTIC 182 if ((pg->flags & (PG_FAULTING)) != 0) 183 panic("uvm_pageremove: page is faulting"); 184#endif 185 186 if ((pg->flags & PG_TABLED) == 0) 187 return; /* XXX: log */ 188 189 buck = &uvm.page_hash[uvm_pagehash(pg->uobject,pg->offset)]; 190 s = splimp(); 191 simple_lock(&uvm.hashlock); 192 TAILQ_REMOVE(buck, pg, hashq); 193 simple_unlock(&uvm.hashlock); 194 splx(s); 195 196 /* object should be locked */ 197 TAILQ_REMOVE(&pg->uobject->memq, pg, listq); 198 199 pg->flags &= ~PG_TABLED; 200 pg->uobject->uo_npages--; 201 pg->uobject = NULL; 202 pg->version++; 203 204} 205 206/* 207 * uvm_page_init: init the page system. called from uvm_init(). 208 * 209 * => we return the range of kernel virtual memory in kvm_startp/kvm_endp 210 */ 211 212void 213uvm_page_init(kvm_startp, kvm_endp) 214 vaddr_t *kvm_startp, *kvm_endp; 215{ 216 int freepages, pagecount; 217 vm_page_t pagearray; 218 int lcv, n, i; 219 paddr_t paddr; 220 221 222 /* 223 * step 1: init the page queues and page queue locks 224 */ 225 for (lcv = 0; lcv < VM_NFREELIST; lcv++) 226 TAILQ_INIT(&uvm.page_free[lcv]); 227 TAILQ_INIT(&uvm.page_active); 228 TAILQ_INIT(&uvm.page_inactive_swp); 229 TAILQ_INIT(&uvm.page_inactive_obj); 230 simple_lock_init(&uvm.pageqlock); 231 simple_lock_init(&uvm.fpageqlock); 232 233 /* 234 * step 2: init the <obj,offset> => <page> hash table. for now 235 * we just have one bucket (the bootstrap bucket). later on we 236 * will malloc() new buckets as we dynamically resize the hash table. 237 */ 238 239 uvm.page_nhash = 1; /* 1 bucket */ 240 uvm.page_hashmask = 0; /* mask for hash function */ 241 uvm.page_hash = &uvm_bootbucket; /* install bootstrap bucket */ 242 TAILQ_INIT(uvm.page_hash); /* init hash table */ 243 simple_lock_init(&uvm.hashlock); /* init hash table lock */ 244 245 /* 246 * step 3: allocate vm_page structures. 247 */ 248 249 /* 250 * sanity check: 251 * before calling this function the MD code is expected to register 252 * some free RAM with the uvm_page_physload() function. our job 253 * now is to allocate vm_page structures for this memory. 254 */ 255 256 if (vm_nphysseg == 0) 257 panic("vm_page_bootstrap: no memory pre-allocated"); 258 259 /* 260 * first calculate the number of free pages... 261 * 262 * note that we use start/end rather than avail_start/avail_end. 263 * this allows us to allocate extra vm_page structures in case we 264 * want to return some memory to the pool after booting. 265 */ 266 267 freepages = 0; 268 for (lcv = 0 ; lcv < vm_nphysseg ; lcv++) 269 freepages += (vm_physmem[lcv].end - vm_physmem[lcv].start); 270 271 /* 272 * we now know we have (PAGE_SIZE * freepages) bytes of memory we can 273 * use. for each page of memory we use we need a vm_page structure. 274 * thus, the total number of pages we can use is the total size of 275 * the memory divided by the PAGE_SIZE plus the size of the vm_page 276 * structure. we add one to freepages as a fudge factor to avoid 277 * truncation errors (since we can only allocate in terms of whole 278 * pages). 279 */ 280 281 pagecount = ((freepages + 1) << PAGE_SHIFT) / 282 (PAGE_SIZE + sizeof(struct vm_page)); 283 pagearray = (vm_page_t)uvm_pageboot_alloc(pagecount * 284 sizeof(struct vm_page)); 285 bzero(pagearray, pagecount * sizeof(struct vm_page)); 286 287 /* 288 * step 4: init the vm_page structures and put them in the correct 289 * place... 290 */ 291 292 for (lcv = 0 ; lcv < vm_nphysseg ; lcv++) { 293 294 n = vm_physmem[lcv].end - vm_physmem[lcv].start; 295 if (n > pagecount) { 296 printf("uvm_page_init: lost %d page(s) in init\n", 297 n - pagecount); 298 panic("uvm_page_init"); /* XXXCDC: shouldn't happen? */ 299 /* n = pagecount; */ 300 } 301 /* set up page array pointers */ 302 vm_physmem[lcv].pgs = pagearray; 303 pagearray += n; 304 pagecount -= n; 305 vm_physmem[lcv].lastpg = vm_physmem[lcv].pgs + (n - 1); 306 307 /* init and free vm_pages (we've already zeroed them) */ 308 paddr = ptoa(vm_physmem[lcv].start); 309 for (i = 0 ; i < n ; i++, paddr += PAGE_SIZE) { 310 vm_physmem[lcv].pgs[i].phys_addr = paddr; 311 if (atop(paddr) >= vm_physmem[lcv].avail_start && 312 atop(paddr) <= vm_physmem[lcv].avail_end) { 313 uvmexp.npages++; 314 /* add page to free pool */ 315 uvm_pagefree(&vm_physmem[lcv].pgs[i]); 316 } 317 } 318 } 319 /* 320 * step 5: pass up the values of virtual_space_start and 321 * virtual_space_end (obtained by uvm_pageboot_alloc) to the upper 322 * layers of the VM. 323 */ 324 325 *kvm_startp = round_page(virtual_space_start); 326 *kvm_endp = trunc_page(virtual_space_end); 327 328 /* 329 * step 6: init pagedaemon lock 330 */ 331 332 simple_lock_init(&uvm.pagedaemon_lock); 333 334 /* 335 * step 7: init reserve thresholds 336 * XXXCDC - values may need adjusting 337 */ 338 uvmexp.reserve_pagedaemon = 1; 339 uvmexp.reserve_kernel = 5; 340 341 /* 342 * done! 343 */ 344 345} 346 347/* 348 * uvm_setpagesize: set the page size 349 * 350 * => sets page_shift and page_mask from uvmexp.pagesize. 351 * => XXXCDC: move global vars. 352 */ 353 354void 355uvm_setpagesize() 356{ 357 if (uvmexp.pagesize == 0) 358 uvmexp.pagesize = DEFAULT_PAGE_SIZE; 359 uvmexp.pagemask = uvmexp.pagesize - 1; 360 if ((uvmexp.pagemask & uvmexp.pagesize) != 0) 361 panic("uvm_setpagesize: page size not a power of two"); 362 for (uvmexp.pageshift = 0; ; uvmexp.pageshift++) 363 if ((1 << uvmexp.pageshift) == uvmexp.pagesize) 364 break; 365} 366 367/* 368 * uvm_pageboot_alloc: steal memory from physmem for bootstrapping 369 */ 370 371vaddr_t 372uvm_pageboot_alloc(size) 373 vsize_t size; 374{ 375#if defined(PMAP_STEAL_MEMORY) 376 vaddr_t addr; 377 378 /* 379 * defer bootstrap allocation to MD code (it may want to allocate 380 * from a direct-mapped segment). pmap_steal_memory should round 381 * off virtual_space_start/virtual_space_end. 382 */ 383 384 addr = pmap_steal_memory(size, &virtual_space_start, 385 &virtual_space_end); 386 387 return(addr); 388 389#else /* !PMAP_STEAL_MEMORY */ 390 391 vaddr_t addr, vaddr; 392 paddr_t paddr; 393 394 /* round to page size */ 395 size = round_page(size); 396 397 /* 398 * on first call to this function init ourselves. we detect this 399 * by checking virtual_space_start/end which are in the zero'd BSS area. 400 */ 401 402 if (virtual_space_start == virtual_space_end) { 403 pmap_virtual_space(&virtual_space_start, &virtual_space_end); 404 405 /* round it the way we like it */ 406 virtual_space_start = round_page(virtual_space_start); 407 virtual_space_end = trunc_page(virtual_space_end); 408 } 409 410 /* 411 * allocate virtual memory for this request 412 */ 413 414 addr = virtual_space_start; 415 virtual_space_start += size; 416 417 /* 418 * allocate and mapin physical pages to back new virtual pages 419 */ 420 421 for (vaddr = round_page(addr) ; vaddr < addr + size ; 422 vaddr += PAGE_SIZE) { 423 424 if (!uvm_page_physget(&paddr)) 425 panic("uvm_pageboot_alloc: out of memory"); 426 427 /* XXX: should be wired, but some pmaps don't like that ... */ 428#if defined(PMAP_NEW) 429 pmap_kenter_pa(vaddr, paddr, VM_PROT_READ|VM_PROT_WRITE); 430#else 431 pmap_enter(pmap_kernel(), vaddr, paddr, 432 VM_PROT_READ|VM_PROT_WRITE, FALSE); 433#endif 434 435 } 436 return(addr); 437#endif /* PMAP_STEAL_MEMORY */ 438} 439 440#if !defined(PMAP_STEAL_MEMORY) 441/* 442 * uvm_page_physget: "steal" one page from the vm_physmem structure. 443 * 444 * => attempt to allocate it off the end of a segment in which the "avail" 445 * values match the start/end values. if we can't do that, then we 446 * will advance both values (making them equal, and removing some 447 * vm_page structures from the non-avail area). 448 * => return false if out of memory. 449 */ 450 451boolean_t 452uvm_page_physget(paddrp) 453 paddr_t *paddrp; 454{ 455 int lcv, x; 456 457 /* pass 1: try allocating from a matching end */ 458#if (VM_PHYSSEG_STRAT == VM_PSTRAT_BIGFIRST) 459 for (lcv = vm_nphysseg - 1 ; lcv >= 0 ; lcv--) 460#else 461 for (lcv = 0 ; lcv < vm_nphysseg ; lcv++) 462#endif 463 { 464 465 if (vm_physmem[lcv].pgs) 466 panic("vm_page_physget: called _after_ bootstrap"); 467 468 /* try from front */ 469 if (vm_physmem[lcv].avail_start == vm_physmem[lcv].start && 470 vm_physmem[lcv].avail_start < vm_physmem[lcv].avail_end) { 471 *paddrp = ptoa(vm_physmem[lcv].avail_start); 472 vm_physmem[lcv].avail_start++; 473 vm_physmem[lcv].start++; 474 /* nothing left? nuke it */ 475 if (vm_physmem[lcv].avail_start == 476 vm_physmem[lcv].end) { 477 if (vm_nphysseg == 1) 478 panic("vm_page_physget: out of memory!"); 479 vm_nphysseg--; 480 for (x = lcv ; x < vm_nphysseg ; x++) 481 /* structure copy */ 482 vm_physmem[x] = vm_physmem[x+1]; 483 } 484 return (TRUE); 485 } 486 487 /* try from rear */ 488 if (vm_physmem[lcv].avail_end == vm_physmem[lcv].end && 489 vm_physmem[lcv].avail_start < vm_physmem[lcv].avail_end) { 490 *paddrp = ptoa(vm_physmem[lcv].avail_end - 1); 491 vm_physmem[lcv].avail_end--; 492 vm_physmem[lcv].end--; 493 /* nothing left? nuke it */ 494 if (vm_physmem[lcv].avail_end == 495 vm_physmem[lcv].start) { 496 if (vm_nphysseg == 1) 497 panic("vm_page_physget: out of memory!"); 498 vm_nphysseg--; 499 for (x = lcv ; x < vm_nphysseg ; x++) 500 /* structure copy */ 501 vm_physmem[x] = vm_physmem[x+1]; 502 } 503 return (TRUE); 504 } 505 } 506 507 /* pass2: forget about matching ends, just allocate something */ 508#if (VM_PHYSSEG_STRAT == VM_PSTRAT_BIGFIRST) 509 for (lcv = vm_nphysseg - 1 ; lcv >= 0 ; lcv--) 510#else 511 for (lcv = 0 ; lcv < vm_nphysseg ; lcv++) 512#endif 513 { 514 515 /* any room in this bank? */ 516 if (vm_physmem[lcv].avail_start >= vm_physmem[lcv].avail_end) 517 continue; /* nope */ 518 519 *paddrp = ptoa(vm_physmem[lcv].avail_start); 520 vm_physmem[lcv].avail_start++; 521 /* truncate! */ 522 vm_physmem[lcv].start = vm_physmem[lcv].avail_start; 523 524 /* nothing left? nuke it */ 525 if (vm_physmem[lcv].avail_start == vm_physmem[lcv].end) { 526 if (vm_nphysseg == 1) 527 panic("vm_page_physget: out of memory!"); 528 vm_nphysseg--; 529 for (x = lcv ; x < vm_nphysseg ; x++) 530 /* structure copy */ 531 vm_physmem[x] = vm_physmem[x+1]; 532 } 533 return (TRUE); 534 } 535 536 return (FALSE); /* whoops! */ 537} 538#endif /* PMAP_STEAL_MEMORY */ 539 540/* 541 * uvm_page_physload: load physical memory into VM system 542 * 543 * => all args are PFs 544 * => all pages in start/end get vm_page structures 545 * => areas marked by avail_start/avail_end get added to the free page pool 546 * => we are limited to VM_PHYSSEG_MAX physical memory segments 547 */ 548 549void 550uvm_page_physload(start, end, avail_start, avail_end, free_list) 551 vaddr_t start, end, avail_start, avail_end; 552 int free_list; 553{ 554 int preload, lcv; 555 psize_t npages; 556 struct vm_page *pgs; 557 struct vm_physseg *ps; 558 559 if (uvmexp.pagesize == 0) 560 panic("vm_page_physload: page size not set!"); 561 562 if (free_list >= VM_NFREELIST || free_list < VM_FREELIST_DEFAULT) 563 panic("uvm_page_physload: bad free list %d\n", free_list); 564 565 /* 566 * do we have room? 567 */ 568 if (vm_nphysseg == VM_PHYSSEG_MAX) { 569 printf("vm_page_physload: unable to load physical memory " 570 "segment\n"); 571 printf("\t%d segments allocated, ignoring 0x%lx -> 0x%lx\n", 572 VM_PHYSSEG_MAX, start, end); 573 return; 574 } 575 576 /* 577 * check to see if this is a "preload" (i.e. uvm_mem_init hasn't been 578 * called yet, so malloc is not available). 579 */ 580 for (lcv = 0 ; lcv < vm_nphysseg ; lcv++) { 581 if (vm_physmem[lcv].pgs) 582 break; 583 } 584 preload = (lcv == vm_nphysseg); 585 586 /* 587 * if VM is already running, attempt to malloc() vm_page structures 588 */ 589 if (!preload) { 590#if defined(VM_PHYSSEG_NOADD) 591 panic("vm_page_physload: tried to add RAM after vm_mem_init"); 592#else 593 /* XXXCDC: need some sort of lockout for this case */ 594 paddr_t paddr; 595 npages = end - start; /* # of pages */ 596 MALLOC(pgs, struct vm_page *, sizeof(struct vm_page) * npages, 597 M_VMPAGE, M_NOWAIT); 598 if (pgs == NULL) { 599 printf("vm_page_physload: can not malloc vm_page " 600 "structs for segment\n"); 601 printf("\tignoring 0x%lx -> 0x%lx\n", start, end); 602 return; 603 } 604 /* zero data, init phys_addr and free_list, and free pages */ 605 bzero(pgs, sizeof(struct vm_page) * npages); 606 for (lcv = 0, paddr = ptoa(start) ; 607 lcv < npages ; lcv++, paddr += PAGE_SIZE) { 608 pgs[lcv].phys_addr = paddr; 609 pgs[lcv].free_list = free_list; 610 if (atop(paddr) >= avail_start && 611 atop(paddr) <= avail_end) 612 uvm_pagefree(&pgs[lcv]); 613 } 614 /* XXXCDC: incomplete: need to update uvmexp.free, what else? */ 615 /* XXXCDC: need hook to tell pmap to rebuild pv_list, etc... */ 616#endif 617 } else { 618 619 /* gcc complains if these don't get init'd */ 620 pgs = NULL; 621 npages = 0; 622 623 } 624 625 /* 626 * now insert us in the proper place in vm_physmem[] 627 */ 628 629#if (VM_PHYSSEG_STRAT == VM_PSTRAT_RANDOM) 630 631 /* random: put it at the end (easy!) */ 632 ps = &vm_physmem[vm_nphysseg]; 633 634#elif (VM_PHYSSEG_STRAT == VM_PSTRAT_BSEARCH) 635 636 { 637 int x; 638 /* sort by address for binary search */ 639 for (lcv = 0 ; lcv < vm_nphysseg ; lcv++) 640 if (start < vm_physmem[lcv].start) 641 break; 642 ps = &vm_physmem[lcv]; 643 /* move back other entries, if necessary ... */ 644 for (x = vm_nphysseg ; x > lcv ; x--) 645 /* structure copy */ 646 vm_physmem[x] = vm_physmem[x - 1]; 647 } 648 649#elif (VM_PHYSSEG_STRAT == VM_PSTRAT_BIGFIRST) 650 651 { 652 int x; 653 /* sort by largest segment first */ 654 for (lcv = 0 ; lcv < vm_nphysseg ; lcv++) 655 if ((end - start) > 656 (vm_physmem[lcv].end - vm_physmem[lcv].start)) 657 break; 658 ps = &vm_physmem[lcv]; 659 /* move back other entries, if necessary ... */ 660 for (x = vm_nphysseg ; x > lcv ; x--) 661 /* structure copy */ 662 vm_physmem[x] = vm_physmem[x - 1]; 663 } 664 665#else 666 667 panic("vm_page_physload: unknown physseg strategy selected!"); 668 669#endif 670 671 ps->start = start; 672 ps->end = end; 673 ps->avail_start = avail_start; 674 ps->avail_end = avail_end; 675 if (preload) { 676 ps->pgs = NULL; 677 } else { 678 ps->pgs = pgs; 679 ps->lastpg = pgs + npages - 1; 680 } 681 ps->free_list = free_list; 682 vm_nphysseg++; 683 684 /* 685 * done! 686 */ 687 688 if (!preload) 689 uvm_page_rehash(); 690 691 return; 692} 693 694/* 695 * uvm_page_rehash: reallocate hash table based on number of free pages. 696 */ 697 698void 699uvm_page_rehash() 700{ 701 int freepages, lcv, bucketcount, s, oldcount; 702 struct pglist *newbuckets, *oldbuckets; 703 struct vm_page *pg; 704 705 /* 706 * compute number of pages that can go in the free pool 707 */ 708 709 freepages = 0; 710 for (lcv = 0 ; lcv < vm_nphysseg ; lcv++) 711 freepages += 712 (vm_physmem[lcv].avail_end - vm_physmem[lcv].avail_start); 713 714 /* 715 * compute number of buckets needed for this number of pages 716 */ 717 718 bucketcount = 1; 719 while (bucketcount < freepages) 720 bucketcount = bucketcount * 2; 721 722 /* 723 * malloc new buckets 724 */ 725 726 MALLOC(newbuckets, struct pglist *, sizeof(struct pglist) * bucketcount, 727 M_VMPBUCKET, M_NOWAIT); 728 if (newbuckets == NULL) { 729 printf("vm_page_physrehash: WARNING: could not grow page " 730 "hash table\n"); 731 return; 732 } 733 for (lcv = 0 ; lcv < bucketcount ; lcv++) 734 TAILQ_INIT(&newbuckets[lcv]); 735 736 /* 737 * now replace the old buckets with the new ones and rehash everything 738 */ 739 740 s = splimp(); 741 simple_lock(&uvm.hashlock); 742 /* swap old for new ... */ 743 oldbuckets = uvm.page_hash; 744 oldcount = uvm.page_nhash; 745 uvm.page_hash = newbuckets; 746 uvm.page_nhash = bucketcount; 747 uvm.page_hashmask = bucketcount - 1; /* power of 2 */ 748 749 /* ... and rehash */ 750 for (lcv = 0 ; lcv < oldcount ; lcv++) { 751 while ((pg = oldbuckets[lcv].tqh_first) != NULL) { 752 TAILQ_REMOVE(&oldbuckets[lcv], pg, hashq); 753 TAILQ_INSERT_TAIL( 754 &uvm.page_hash[uvm_pagehash(pg->uobject, pg->offset)], 755 pg, hashq); 756 } 757 } 758 simple_unlock(&uvm.hashlock); 759 splx(s); 760 761 /* 762 * free old bucket array if we malloc'd it previously 763 */ 764 765 if (oldbuckets != &uvm_bootbucket) 766 FREE(oldbuckets, M_VMPBUCKET); 767 768 /* 769 * done 770 */ 771 return; 772} 773 774 775#if 1 /* XXXCDC: TMP TMP TMP DEBUG DEBUG DEBUG */ 776 777void uvm_page_physdump __P((void)); /* SHUT UP GCC */ 778 779/* call from DDB */ 780void 781uvm_page_physdump() 782{ 783 int lcv; 784 785 printf("rehash: physical memory config [segs=%d of %d]:\n", 786 vm_nphysseg, VM_PHYSSEG_MAX); 787 for (lcv = 0 ; lcv < vm_nphysseg ; lcv++) 788 printf("0x%lx->0x%lx [0x%lx->0x%lx]\n", vm_physmem[lcv].start, 789 vm_physmem[lcv].end, vm_physmem[lcv].avail_start, 790 vm_physmem[lcv].avail_end); 791 printf("STRATEGY = "); 792 switch (VM_PHYSSEG_STRAT) { 793 case VM_PSTRAT_RANDOM: printf("RANDOM\n"); break; 794 case VM_PSTRAT_BSEARCH: printf("BSEARCH\n"); break; 795 case VM_PSTRAT_BIGFIRST: printf("BIGFIRST\n"); break; 796 default: printf("<<UNKNOWN>>!!!!\n"); 797 } 798 printf("number of buckets = %d\n", uvm.page_nhash); 799} 800#endif 801 802/* 803 * uvm_pagealloc_strat: allocate vm_page from a particular free list. 804 * 805 * => return null if no pages free 806 * => wake up pagedaemon if number of free pages drops below low water mark 807 * => if obj != NULL, obj must be locked (to put in hash) 808 * => if anon != NULL, anon must be locked (to put in anon) 809 * => only one of obj or anon can be non-null 810 * => caller must activate/deactivate page if it is not wired. 811 * => free_list is ignored if strat == UVM_PGA_STRAT_NORMAL. 812 */ 813 814struct vm_page * 815uvm_pagealloc_strat(obj, off, anon, strat, free_list) 816 struct uvm_object *obj; 817 vaddr_t off; 818 struct vm_anon *anon; 819 int strat, free_list; 820{ 821 int lcv, s; 822 struct vm_page *pg; 823 struct pglist *freeq; 824 825#ifdef DIAGNOSTIC 826 /* sanity check */ 827 if (obj && anon) 828 panic("uvm_pagealloc: obj and anon != NULL"); 829#endif 830 831 s = splimp(); 832 833 uvm_lock_fpageq(); /* lock free page queue */ 834 835 /* 836 * check to see if we need to generate some free pages waking 837 * the pagedaemon. 838 */ 839 840 if (uvmexp.free < uvmexp.freemin || (uvmexp.free < uvmexp.freetarg && 841 uvmexp.inactive < uvmexp.inactarg)) 842 thread_wakeup(&uvm.pagedaemon); 843 844 /* 845 * fail if any of these conditions is true: 846 * [1] there really are no free pages, or 847 * [2] only kernel "reserved" pages remain and 848 * the page isn't being allocated to a kernel object. 849 * [3] only pagedaemon "reserved" pages remain and 850 * the requestor isn't the pagedaemon. 851 */ 852 853 if ((uvmexp.free <= uvmexp.reserve_kernel && 854 !(obj && obj->uo_refs == UVM_OBJ_KERN)) || 855 (uvmexp.free <= uvmexp.reserve_pagedaemon && 856 !(obj == uvmexp.kmem_object && curproc == uvm.pagedaemon_proc))) 857 goto fail; 858 859 again: 860 switch (strat) { 861 case UVM_PGA_STRAT_NORMAL: 862 /* Check all freelists in descending priority order. */ 863 for (lcv = 0; lcv < VM_NFREELIST; lcv++) { 864 freeq = &uvm.page_free[lcv]; 865 if ((pg = freeq->tqh_first) != NULL) 866 goto gotit; 867 } 868 869 /* No pages free! */ 870 goto fail; 871 872 case UVM_PGA_STRAT_ONLY: 873 case UVM_PGA_STRAT_FALLBACK: 874 /* Attempt to allocate from the specified free list. */ 875#ifdef DIAGNOSTIC 876 if (free_list >= VM_NFREELIST || free_list < 0) 877 panic("uvm_pagealloc_strat: bad free list %d", 878 free_list); 879#endif 880 freeq = &uvm.page_free[free_list]; 881 if ((pg = freeq->tqh_first) != NULL) 882 goto gotit; 883 884 /* Fall back, if possible. */ 885 if (strat == UVM_PGA_STRAT_FALLBACK) { 886 strat = UVM_PGA_STRAT_NORMAL; 887 goto again; 888 } 889 890 /* No pages free! */ 891 goto fail; 892 893 default: 894 panic("uvm_pagealloc_strat: bad strat %d", strat); 895 /* NOTREACHED */ 896 } 897 898 gotit: 899 TAILQ_REMOVE(freeq, pg, pageq); 900 uvmexp.free--; 901 902 uvm_unlock_fpageq(); /* unlock free page queue */ 903 splx(s); 904 905 pg->offset = off; 906 pg->uobject = obj; 907 pg->uanon = anon; 908 pg->flags = PG_BUSY|PG_CLEAN|PG_FAKE; 909 pg->version++; 910 pg->wire_count = 0; 911 pg->loan_count = 0; 912 if (anon) { 913 anon->u.an_page = pg; 914 pg->pqflags = PQ_ANON; 915 } else { 916 if (obj) 917 uvm_pageinsert(pg); 918 pg->pqflags = 0; 919 } 920#if defined(UVM_PAGE_TRKOWN) 921 pg->owner_tag = NULL; 922#endif 923 UVM_PAGE_OWN(pg, "new alloc"); 924 925 return(pg); 926 927 fail: 928 uvm_unlock_fpageq(); 929 splx(s); 930 return (NULL); 931} 932 933/* 934 * uvm_pagealloc_contig: allocate contiguous memory. 935 * 936 * XXX - fix comment. 937 */ 938 939vaddr_t 940uvm_pagealloc_contig(size, low, high, alignment) 941 vaddr_t size; 942 vaddr_t low, high; 943 vaddr_t alignment; 944{ 945 struct pglist pglist; 946 struct vm_page *pg; 947 vaddr_t addr, temp_addr; 948 949 size = round_page(size); 950 951 TAILQ_INIT(&pglist); 952 if (uvm_pglistalloc(size, low, high, alignment, 0, 953 &pglist, 1, FALSE)) 954 return 0; 955 addr = vm_map_min(kernel_map); 956 if (uvm_map(kernel_map, &addr, size, NULL, 957 UVM_UNKNOWN_OFFSET, TRUE) != KERN_SUCCESS) 958 addr = 0; 959 temp_addr = addr; 960 for (pg = TAILQ_FIRST(&pglist); pg != NULL; 961 pg = TAILQ_NEXT(pg, pageq)) { 962 pg->uobject = uvm.kernel_object; 963 pg->offset = temp_addr - vm_map_min(kernel_map); 964 uvm_pageinsert(pg); 965 uvm_pagewire(pg); 966#if defined(PMAP_NEW) 967 pmap_kenter_pa(temp_addr, VM_PAGE_TO_PHYS(pg), 968 VM_PROT_READ|VM_PROT_WRITE); 969#else 970 pmap_enter(kernel_map->pmap, temp_addr, VM_PAGE_TO_PHYS(pg), 971 UVM_PROT_READ|UVM_PROT_WRITE, TRUE); 972#endif 973 temp_addr += PAGE_SIZE; 974 } 975 return addr; 976} 977 978/* 979 * uvm_pagerealloc: reallocate a page from one object to another 980 * 981 * => both objects must be locked 982 */ 983 984void 985uvm_pagerealloc(pg, newobj, newoff) 986 struct vm_page *pg; 987 struct uvm_object *newobj; 988 vaddr_t newoff; 989{ 990 /* 991 * remove it from the old object 992 */ 993 994 if (pg->uobject) { 995 uvm_pageremove(pg); 996 } 997 998 /* 999 * put it in the new object 1000 */ 1001 1002 if (newobj) { 1003 pg->uobject = newobj; 1004 pg->offset = newoff; 1005 pg->version++; 1006 uvm_pageinsert(pg); 1007 } 1008 1009 return; 1010} 1011 1012 1013/* 1014 * uvm_pagefree: free page 1015 * 1016 * => erase page's identity (i.e. remove from hash/object) 1017 * => put page on free list 1018 * => caller must lock owning object (either anon or uvm_object) 1019 * => caller must lock page queues 1020 * => assumes all valid mappings of pg are gone 1021 */ 1022 1023void uvm_pagefree(pg) 1024 1025struct vm_page *pg; 1026 1027{ 1028 int s; 1029 int saved_loan_count = pg->loan_count; 1030 1031 /* 1032 * if the page was an object page (and thus "TABLED"), remove it 1033 * from the object. 1034 */ 1035 1036 if (pg->flags & PG_TABLED) { 1037 1038 /* 1039 * if the object page is on loan we are going to drop ownership. 1040 * it is possible that an anon will take over as owner for this 1041 * page later on. the anon will want a !PG_CLEAN page so that 1042 * it knows it needs to allocate swap if it wants to page the 1043 * page out. 1044 */ 1045 1046 if (saved_loan_count) 1047 pg->flags &= ~PG_CLEAN; /* in case an anon takes over */ 1048 1049 uvm_pageremove(pg); 1050 1051 /* 1052 * if our page was on loan, then we just lost control over it 1053 * (in fact, if it was loaned to an anon, the anon may have 1054 * already taken over ownership of the page by now and thus 1055 * changed the loan_count [e.g. in uvmfault_anonget()]) we just 1056 * return (when the last loan is dropped, then the page can be 1057 * freed by whatever was holding the last loan). 1058 */ 1059 if (saved_loan_count) 1060 return; 1061 1062 } else if (saved_loan_count && (pg->pqflags & PQ_ANON)) { 1063 1064 /* 1065 * if our page is owned by an anon and is loaned out to the 1066 * kernel then we just want to drop ownership and return. 1067 * the kernel must free the page when all its loans clear ... 1068 * note that the kernel can't change the loan status of our 1069 * page as long as we are holding PQ lock. 1070 */ 1071 pg->pqflags &= ~PQ_ANON; 1072 pg->uanon = NULL; 1073 return; 1074 } 1075 1076#ifdef DIAGNOSTIC 1077 if (saved_loan_count) { 1078 printf("uvm_pagefree: warning: freeing page with a loan " 1079 "count of %d\n", saved_loan_count); 1080 panic("uvm_pagefree: loan count"); 1081 } 1082#endif 1083 1084 1085 /* 1086 * now remove the page from the queues 1087 */ 1088 1089 if (pg->pqflags & PQ_ACTIVE) { 1090 TAILQ_REMOVE(&uvm.page_active, pg, pageq); 1091 pg->pqflags &= ~PQ_ACTIVE; 1092 uvmexp.active--; 1093 } 1094 if (pg->pqflags & PQ_INACTIVE) { 1095 if (pg->pqflags & PQ_SWAPBACKED) 1096 TAILQ_REMOVE(&uvm.page_inactive_swp, pg, pageq); 1097 else 1098 TAILQ_REMOVE(&uvm.page_inactive_obj, pg, pageq); 1099 pg->pqflags &= ~PQ_INACTIVE; 1100 uvmexp.inactive--; 1101 } 1102 1103 /* 1104 * if the page was wired, unwire it now. 1105 */ 1106 if (pg->wire_count) 1107 { 1108 pg->wire_count = 0; 1109 uvmexp.wired--; 1110 } 1111 1112 /* 1113 * and put on free queue 1114 */ 1115 1116 s = splimp(); 1117 uvm_lock_fpageq(); 1118 TAILQ_INSERT_TAIL(&uvm.page_free[uvm_page_lookup_freelist(pg)], 1119 pg, pageq); 1120 pg->pqflags = PQ_FREE; 1121#ifdef DEBUG 1122 pg->uobject = (void *)0xdeadbeef; 1123 pg->offset = 0xdeadbeef; 1124 pg->uanon = (void *)0xdeadbeef; 1125#endif 1126 uvmexp.free++; 1127 uvm_unlock_fpageq(); 1128 splx(s); 1129} 1130 1131#if defined(UVM_PAGE_TRKOWN) 1132/* 1133 * uvm_page_own: set or release page ownership 1134 * 1135 * => this is a debugging function that keeps track of who sets PG_BUSY 1136 * and where they do it. it can be used to track down problems 1137 * such a process setting "PG_BUSY" and never releasing it. 1138 * => page's object [if any] must be locked 1139 * => if "tag" is NULL then we are releasing page ownership 1140 */ 1141void 1142uvm_page_own(pg, tag) 1143 struct vm_page *pg; 1144 char *tag; 1145{ 1146 /* gain ownership? */ 1147 if (tag) { 1148 if (pg->owner_tag) { 1149 printf("uvm_page_own: page %p already owned " 1150 "by proc %d [%s]\n", pg, 1151 pg->owner, pg->owner_tag); 1152 panic("uvm_page_own"); 1153 } 1154 pg->owner = (curproc) ? curproc->p_pid : (pid_t) -1; 1155 pg->owner_tag = tag; 1156 return; 1157 } 1158 1159 /* drop ownership */ 1160 if (pg->owner_tag == NULL) { 1161 printf("uvm_page_own: dropping ownership of an non-owned " 1162 "page (%p)\n", pg); 1163 panic("uvm_page_own"); 1164 } 1165 pg->owner_tag = NULL; 1166 return; 1167} 1168#endif 1169