uvm_page.c revision 1.14
1/* $OpenBSD: uvm_page.c,v 1.14 2001/03/22 03:05:56 smart Exp $ */ 2/* $NetBSD: uvm_page.c,v 1.24 1999/07/22 22:58:38 thorpej Exp $ */ 3 4/* 5 * Copyright (c) 1997 Charles D. Cranor and Washington University. 6 * Copyright (c) 1991, 1993, The Regents of the University of California. 7 * 8 * All rights reserved. 9 * 10 * This code is derived from software contributed to Berkeley by 11 * The Mach Operating System project at Carnegie-Mellon University. 12 * 13 * Redistribution and use in source and binary forms, with or without 14 * modification, are permitted provided that the following conditions 15 * are met: 16 * 1. Redistributions of source code must retain the above copyright 17 * notice, this list of conditions and the following disclaimer. 18 * 2. Redistributions in binary form must reproduce the above copyright 19 * notice, this list of conditions and the following disclaimer in the 20 * documentation and/or other materials provided with the distribution. 21 * 3. All advertising materials mentioning features or use of this software 22 * must display the following acknowledgement: 23 * This product includes software developed by Charles D. Cranor, 24 * Washington University, the University of California, Berkeley and 25 * its contributors. 26 * 4. Neither the name of the University nor the names of its contributors 27 * may be used to endorse or promote products derived from this software 28 * without specific prior written permission. 29 * 30 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 31 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 32 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 33 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 34 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 35 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 36 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 37 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 38 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 39 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 40 * SUCH DAMAGE. 41 * 42 * @(#)vm_page.c 8.3 (Berkeley) 3/21/94 43 * from: Id: uvm_page.c,v 1.1.2.18 1998/02/06 05:24:42 chs Exp 44 * 45 * 46 * Copyright (c) 1987, 1990 Carnegie-Mellon University. 47 * All rights reserved. 48 * 49 * Permission to use, copy, modify and distribute this software and 50 * its documentation is hereby granted, provided that both the copyright 51 * notice and this permission notice appear in all copies of the 52 * software, derivative works or modified versions, and any portions 53 * thereof, and that both notices appear in supporting documentation. 54 * 55 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS" 56 * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND 57 * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE. 58 * 59 * Carnegie Mellon requests users of this software to return to 60 * 61 * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU 62 * School of Computer Science 63 * Carnegie Mellon University 64 * Pittsburgh PA 15213-3890 65 * 66 * any improvements or extensions that they make and grant Carnegie the 67 * rights to redistribute these changes. 68 */ 69 70/* 71 * uvm_page.c: page ops. 72 */ 73 74#include <sys/param.h> 75#include <sys/systm.h> 76#include <sys/malloc.h> 77#include <sys/proc.h> 78 79#include <vm/vm.h> 80#include <vm/vm_page.h> 81#include <vm/vm_kern.h> 82 83#define UVM_PAGE /* pull in uvm_page.h functions */ 84#include <uvm/uvm.h> 85 86/* 87 * global vars... XXXCDC: move to uvm. structure. 88 */ 89 90/* 91 * physical memory config is stored in vm_physmem. 92 */ 93 94struct vm_physseg vm_physmem[VM_PHYSSEG_MAX]; /* XXXCDC: uvm.physmem */ 95int vm_nphysseg = 0; /* XXXCDC: uvm.nphysseg */ 96 97/* 98 * local variables 99 */ 100 101/* 102 * these variables record the values returned by vm_page_bootstrap, 103 * for debugging purposes. The implementation of uvm_pageboot_alloc 104 * and pmap_startup here also uses them internally. 105 */ 106 107static vaddr_t virtual_space_start; 108static vaddr_t virtual_space_end; 109 110/* 111 * we use a hash table with only one bucket during bootup. we will 112 * later rehash (resize) the hash table once the allocator is ready. 113 * we static allocate the one bootstrap bucket below... 114 */ 115 116static struct pglist uvm_bootbucket; 117 118/* 119 * local prototypes 120 */ 121 122static void uvm_pageinsert __P((struct vm_page *)); 123 124 125/* 126 * inline functions 127 */ 128 129/* 130 * uvm_pageinsert: insert a page in the object and the hash table 131 * 132 * => caller must lock object 133 * => caller must lock page queues 134 * => call should have already set pg's object and offset pointers 135 * and bumped the version counter 136 */ 137 138__inline static void 139uvm_pageinsert(pg) 140 struct vm_page *pg; 141{ 142 struct pglist *buck; 143 int s; 144 145#ifdef DIAGNOSTIC 146 if (pg->flags & PG_TABLED) 147 panic("uvm_pageinsert: already inserted"); 148#endif 149 150 buck = &uvm.page_hash[uvm_pagehash(pg->uobject,pg->offset)]; 151 s = splimp(); 152 simple_lock(&uvm.hashlock); 153 TAILQ_INSERT_TAIL(buck, pg, hashq); /* put in hash */ 154 simple_unlock(&uvm.hashlock); 155 splx(s); 156 157 TAILQ_INSERT_TAIL(&pg->uobject->memq, pg, listq); /* put in object */ 158 pg->flags |= PG_TABLED; 159 pg->uobject->uo_npages++; 160 161} 162 163/* 164 * uvm_page_remove: remove page from object and hash 165 * 166 * => caller must lock object 167 * => caller must lock page queues 168 */ 169 170void __inline 171uvm_pageremove(pg) 172 struct vm_page *pg; 173{ 174 struct pglist *buck; 175 int s; 176 177#ifdef DIAGNOSTIC 178 if ((pg->flags & (PG_FAULTING)) != 0) 179 panic("uvm_pageremove: page is faulting"); 180#endif 181 182 if ((pg->flags & PG_TABLED) == 0) 183 return; /* XXX: log */ 184 185 buck = &uvm.page_hash[uvm_pagehash(pg->uobject,pg->offset)]; 186 s = splimp(); 187 simple_lock(&uvm.hashlock); 188 TAILQ_REMOVE(buck, pg, hashq); 189 simple_unlock(&uvm.hashlock); 190 splx(s); 191 192 /* object should be locked */ 193 TAILQ_REMOVE(&pg->uobject->memq, pg, listq); 194 195 pg->flags &= ~PG_TABLED; 196 pg->uobject->uo_npages--; 197 pg->uobject = NULL; 198 pg->version++; 199 200} 201 202/* 203 * uvm_page_init: init the page system. called from uvm_init(). 204 * 205 * => we return the range of kernel virtual memory in kvm_startp/kvm_endp 206 */ 207 208void 209uvm_page_init(kvm_startp, kvm_endp) 210 vaddr_t *kvm_startp, *kvm_endp; 211{ 212 int freepages, pagecount; 213 vm_page_t pagearray; 214 int lcv, n, i; 215 paddr_t paddr; 216 217 218 /* 219 * step 1: init the page queues and page queue locks 220 */ 221 for (lcv = 0; lcv < VM_NFREELIST; lcv++) 222 TAILQ_INIT(&uvm.page_free[lcv]); 223 TAILQ_INIT(&uvm.page_active); 224 TAILQ_INIT(&uvm.page_inactive_swp); 225 TAILQ_INIT(&uvm.page_inactive_obj); 226 simple_lock_init(&uvm.pageqlock); 227 simple_lock_init(&uvm.fpageqlock); 228 229 /* 230 * step 2: init the <obj,offset> => <page> hash table. for now 231 * we just have one bucket (the bootstrap bucket). later on we 232 * will allocate new buckets as we dynamically resize the hash table. 233 */ 234 235 uvm.page_nhash = 1; /* 1 bucket */ 236 uvm.page_hashmask = 0; /* mask for hash function */ 237 uvm.page_hash = &uvm_bootbucket; /* install bootstrap bucket */ 238 TAILQ_INIT(uvm.page_hash); /* init hash table */ 239 simple_lock_init(&uvm.hashlock); /* init hash table lock */ 240 241 /* 242 * step 3: allocate vm_page structures. 243 */ 244 245 /* 246 * sanity check: 247 * before calling this function the MD code is expected to register 248 * some free RAM with the uvm_page_physload() function. our job 249 * now is to allocate vm_page structures for this memory. 250 */ 251 252 if (vm_nphysseg == 0) 253 panic("vm_page_bootstrap: no memory pre-allocated"); 254 255 /* 256 * first calculate the number of free pages... 257 * 258 * note that we use start/end rather than avail_start/avail_end. 259 * this allows us to allocate extra vm_page structures in case we 260 * want to return some memory to the pool after booting. 261 */ 262 263 freepages = 0; 264 for (lcv = 0 ; lcv < vm_nphysseg ; lcv++) 265 freepages += (vm_physmem[lcv].end - vm_physmem[lcv].start); 266 267 /* 268 * we now know we have (PAGE_SIZE * freepages) bytes of memory we can 269 * use. for each page of memory we use we need a vm_page structure. 270 * thus, the total number of pages we can use is the total size of 271 * the memory divided by the PAGE_SIZE plus the size of the vm_page 272 * structure. we add one to freepages as a fudge factor to avoid 273 * truncation errors (since we can only allocate in terms of whole 274 * pages). 275 */ 276 277 pagecount = ((freepages + 1) << PAGE_SHIFT) / 278 (PAGE_SIZE + sizeof(struct vm_page)); 279 pagearray = (vm_page_t)uvm_pageboot_alloc(pagecount * 280 sizeof(struct vm_page)); 281 memset(pagearray, 0, pagecount * sizeof(struct vm_page)); 282 283 /* 284 * step 4: init the vm_page structures and put them in the correct 285 * place... 286 */ 287 288 for (lcv = 0 ; lcv < vm_nphysseg ; lcv++) { 289 290 n = vm_physmem[lcv].end - vm_physmem[lcv].start; 291 if (n > pagecount) { 292 printf("uvm_page_init: lost %d page(s) in init\n", 293 n - pagecount); 294 panic("uvm_page_init"); /* XXXCDC: shouldn't happen? */ 295 /* n = pagecount; */ 296 } 297 /* set up page array pointers */ 298 vm_physmem[lcv].pgs = pagearray; 299 pagearray += n; 300 pagecount -= n; 301 vm_physmem[lcv].lastpg = vm_physmem[lcv].pgs + (n - 1); 302 303 /* init and free vm_pages (we've already zeroed them) */ 304 paddr = ptoa(vm_physmem[lcv].start); 305 for (i = 0 ; i < n ; i++, paddr += PAGE_SIZE) { 306 vm_physmem[lcv].pgs[i].phys_addr = paddr; 307 if (atop(paddr) >= vm_physmem[lcv].avail_start && 308 atop(paddr) <= vm_physmem[lcv].avail_end) { 309 uvmexp.npages++; 310 /* add page to free pool */ 311 uvm_pagefree(&vm_physmem[lcv].pgs[i]); 312 } 313 } 314 } 315 /* 316 * step 5: pass up the values of virtual_space_start and 317 * virtual_space_end (obtained by uvm_pageboot_alloc) to the upper 318 * layers of the VM. 319 */ 320 321 *kvm_startp = round_page(virtual_space_start); 322 *kvm_endp = trunc_page(virtual_space_end); 323 324 /* 325 * step 6: init pagedaemon lock 326 */ 327 328 simple_lock_init(&uvm.pagedaemon_lock); 329 330 /* 331 * step 7: init reserve thresholds 332 * XXXCDC - values may need adjusting 333 */ 334 uvmexp.reserve_pagedaemon = 4; 335 uvmexp.reserve_kernel = 6; 336 337 /* 338 * done! 339 */ 340 341} 342 343/* 344 * uvm_setpagesize: set the page size 345 * 346 * => sets page_shift and page_mask from uvmexp.pagesize. 347 * => XXXCDC: move global vars. 348 */ 349 350void 351uvm_setpagesize() 352{ 353 if (uvmexp.pagesize == 0) 354 uvmexp.pagesize = DEFAULT_PAGE_SIZE; 355 uvmexp.pagemask = uvmexp.pagesize - 1; 356 if ((uvmexp.pagemask & uvmexp.pagesize) != 0) 357 panic("uvm_setpagesize: page size not a power of two"); 358 for (uvmexp.pageshift = 0; ; uvmexp.pageshift++) 359 if ((1 << uvmexp.pageshift) == uvmexp.pagesize) 360 break; 361} 362 363/* 364 * uvm_pageboot_alloc: steal memory from physmem for bootstrapping 365 */ 366 367vaddr_t 368uvm_pageboot_alloc(size) 369 vsize_t size; 370{ 371#if defined(PMAP_STEAL_MEMORY) 372 vaddr_t addr; 373 374 /* 375 * defer bootstrap allocation to MD code (it may want to allocate 376 * from a direct-mapped segment). pmap_steal_memory should round 377 * off virtual_space_start/virtual_space_end. 378 */ 379 380 addr = pmap_steal_memory(size, &virtual_space_start, 381 &virtual_space_end); 382 383 return(addr); 384 385#else /* !PMAP_STEAL_MEMORY */ 386 387 static boolean_t initialized = FALSE; 388 vaddr_t addr, vaddr; 389 paddr_t paddr; 390 391 /* round to page size */ 392 size = round_page(size); 393 394 /* 395 * on first call to this function, initialize ourselves. 396 */ 397 if (initialized == FALSE) { 398 pmap_virtual_space(&virtual_space_start, &virtual_space_end); 399 400 /* round it the way we like it */ 401 virtual_space_start = round_page(virtual_space_start); 402 virtual_space_end = trunc_page(virtual_space_end); 403 404 initialized = TRUE; 405 } 406 407 /* 408 * allocate virtual memory for this request 409 */ 410 if (virtual_space_start == virtual_space_end || 411 (virtual_space_end - virtual_space_start) < size) 412 panic("uvm_pageboot_alloc: out of virtual space"); 413 414 addr = virtual_space_start; 415 416#ifdef PMAP_GROWKERNEL 417 /* 418 * If the kernel pmap can't map the requested space, 419 * then allocate more resources for it. 420 */ 421 if (uvm_maxkaddr < (addr + size)) { 422 uvm_maxkaddr = pmap_growkernel(addr + size); 423 if (uvm_maxkaddr < (addr + size)) 424 panic("uvm_pageboot_alloc: pmap_growkernel() failed"); 425 } 426#endif 427 428 virtual_space_start += size; 429 430 /* 431 * allocate and mapin physical pages to back new virtual pages 432 */ 433 434 for (vaddr = round_page(addr) ; vaddr < addr + size ; 435 vaddr += PAGE_SIZE) { 436 437 if (!uvm_page_physget(&paddr)) 438 panic("uvm_pageboot_alloc: out of memory"); 439 440 /* XXX: should be wired, but some pmaps don't like that ... */ 441#if defined(PMAP_NEW) 442 /* 443 * Note this memory is no longer managed, so using 444 * pmap_kenter is safe. 445 */ 446 pmap_kenter_pa(vaddr, paddr, VM_PROT_READ|VM_PROT_WRITE); 447#else 448 pmap_enter(pmap_kernel(), vaddr, paddr, 449 VM_PROT_READ|VM_PROT_WRITE, FALSE, 450 VM_PROT_READ|VM_PROT_WRITE); 451#endif 452 453 } 454 return(addr); 455#endif /* PMAP_STEAL_MEMORY */ 456} 457 458#if !defined(PMAP_STEAL_MEMORY) 459/* 460 * uvm_page_physget: "steal" one page from the vm_physmem structure. 461 * 462 * => attempt to allocate it off the end of a segment in which the "avail" 463 * values match the start/end values. if we can't do that, then we 464 * will advance both values (making them equal, and removing some 465 * vm_page structures from the non-avail area). 466 * => return false if out of memory. 467 */ 468 469boolean_t 470uvm_page_physget(paddrp) 471 paddr_t *paddrp; 472{ 473 int lcv, x; 474 475 /* pass 1: try allocating from a matching end */ 476#if (VM_PHYSSEG_STRAT == VM_PSTRAT_BIGFIRST) || \ 477 (VM_PHYSSEG_STRAT == VM_PSTRAT_BSEARCH) 478 for (lcv = vm_nphysseg - 1 ; lcv >= 0 ; lcv--) 479#else 480 for (lcv = 0 ; lcv < vm_nphysseg ; lcv++) 481#endif 482 { 483 484 if (vm_physmem[lcv].pgs) 485 panic("vm_page_physget: called _after_ bootstrap"); 486 487 /* try from front */ 488 if (vm_physmem[lcv].avail_start == vm_physmem[lcv].start && 489 vm_physmem[lcv].avail_start < vm_physmem[lcv].avail_end) { 490 *paddrp = ptoa(vm_physmem[lcv].avail_start); 491 vm_physmem[lcv].avail_start++; 492 vm_physmem[lcv].start++; 493 /* nothing left? nuke it */ 494 if (vm_physmem[lcv].avail_start == 495 vm_physmem[lcv].end) { 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 /* try from rear */ 507 if (vm_physmem[lcv].avail_end == vm_physmem[lcv].end && 508 vm_physmem[lcv].avail_start < vm_physmem[lcv].avail_end) { 509 *paddrp = ptoa(vm_physmem[lcv].avail_end - 1); 510 vm_physmem[lcv].avail_end--; 511 vm_physmem[lcv].end--; 512 /* nothing left? nuke it */ 513 if (vm_physmem[lcv].avail_end == 514 vm_physmem[lcv].start) { 515 if (vm_nphysseg == 1) 516 panic("vm_page_physget: out of memory!"); 517 vm_nphysseg--; 518 for (x = lcv ; x < vm_nphysseg ; x++) 519 /* structure copy */ 520 vm_physmem[x] = vm_physmem[x+1]; 521 } 522 return (TRUE); 523 } 524 } 525 526 /* pass2: forget about matching ends, just allocate something */ 527#if (VM_PHYSSEG_STRAT == VM_PSTRAT_BIGFIRST) || \ 528 (VM_PHYSSEG_STRAT == VM_PSTRAT_BSEARCH) 529 for (lcv = vm_nphysseg - 1 ; lcv >= 0 ; lcv--) 530#else 531 for (lcv = 0 ; lcv < vm_nphysseg ; lcv++) 532#endif 533 { 534 535 /* any room in this bank? */ 536 if (vm_physmem[lcv].avail_start >= vm_physmem[lcv].avail_end) 537 continue; /* nope */ 538 539 *paddrp = ptoa(vm_physmem[lcv].avail_start); 540 vm_physmem[lcv].avail_start++; 541 /* truncate! */ 542 vm_physmem[lcv].start = vm_physmem[lcv].avail_start; 543 544 /* nothing left? nuke it */ 545 if (vm_physmem[lcv].avail_start == vm_physmem[lcv].end) { 546 if (vm_nphysseg == 1) 547 panic("vm_page_physget: out of memory!"); 548 vm_nphysseg--; 549 for (x = lcv ; x < vm_nphysseg ; x++) 550 /* structure copy */ 551 vm_physmem[x] = vm_physmem[x+1]; 552 } 553 return (TRUE); 554 } 555 556 return (FALSE); /* whoops! */ 557} 558#endif /* PMAP_STEAL_MEMORY */ 559 560/* 561 * uvm_page_physload: load physical memory into VM system 562 * 563 * => all args are PFs 564 * => all pages in start/end get vm_page structures 565 * => areas marked by avail_start/avail_end get added to the free page pool 566 * => we are limited to VM_PHYSSEG_MAX physical memory segments 567 */ 568 569void 570uvm_page_physload(start, end, avail_start, avail_end, free_list) 571 vaddr_t start, end, avail_start, avail_end; 572 int free_list; 573{ 574 int preload, lcv; 575 psize_t npages; 576 struct vm_page *pgs; 577 struct vm_physseg *ps; 578 579 if (uvmexp.pagesize == 0) 580 panic("vm_page_physload: page size not set!"); 581 582 if (free_list >= VM_NFREELIST || free_list < VM_FREELIST_DEFAULT) 583 panic("uvm_page_physload: bad free list %d\n", free_list); 584 585 /* 586 * do we have room? 587 */ 588 if (vm_nphysseg == VM_PHYSSEG_MAX) { 589 printf("vm_page_physload: unable to load physical memory " 590 "segment\n"); 591 printf("\t%d segments allocated, ignoring 0x%lx -> 0x%lx\n", 592 VM_PHYSSEG_MAX, start, end); 593 return; 594 } 595 596 /* 597 * check to see if this is a "preload" (i.e. uvm_mem_init hasn't been 598 * called yet, so malloc is not available). 599 */ 600 for (lcv = 0 ; lcv < vm_nphysseg ; lcv++) { 601 if (vm_physmem[lcv].pgs) 602 break; 603 } 604 preload = (lcv == vm_nphysseg); 605 606 /* 607 * if VM is already running, attempt to malloc() vm_page structures 608 */ 609 if (!preload) { 610#if defined(VM_PHYSSEG_NOADD) 611 panic("vm_page_physload: tried to add RAM after vm_mem_init"); 612#else 613 /* XXXCDC: need some sort of lockout for this case */ 614 paddr_t paddr; 615 npages = end - start; /* # of pages */ 616 MALLOC(pgs, struct vm_page *, sizeof(struct vm_page) * npages, 617 M_VMPAGE, M_NOWAIT); 618 if (pgs == NULL) { 619 printf("vm_page_physload: can not malloc vm_page " 620 "structs for segment\n"); 621 printf("\tignoring 0x%lx -> 0x%lx\n", start, end); 622 return; 623 } 624 /* zero data, init phys_addr and free_list, and free pages */ 625 memset(pgs, 0, sizeof(struct vm_page) * npages); 626 for (lcv = 0, paddr = ptoa(start) ; 627 lcv < npages ; lcv++, paddr += PAGE_SIZE) { 628 pgs[lcv].phys_addr = paddr; 629 pgs[lcv].free_list = free_list; 630 if (atop(paddr) >= avail_start && 631 atop(paddr) <= avail_end) 632 uvm_pagefree(&pgs[lcv]); 633 } 634 /* XXXCDC: incomplete: need to update uvmexp.free, what else? */ 635 /* XXXCDC: need hook to tell pmap to rebuild pv_list, etc... */ 636#endif 637 } else { 638 639 /* gcc complains if these don't get init'd */ 640 pgs = NULL; 641 npages = 0; 642 643 } 644 645 /* 646 * now insert us in the proper place in vm_physmem[] 647 */ 648 649#if (VM_PHYSSEG_STRAT == VM_PSTRAT_RANDOM) 650 651 /* random: put it at the end (easy!) */ 652 ps = &vm_physmem[vm_nphysseg]; 653 654#elif (VM_PHYSSEG_STRAT == VM_PSTRAT_BSEARCH) 655 656 { 657 int x; 658 /* sort by address for binary search */ 659 for (lcv = 0 ; lcv < vm_nphysseg ; lcv++) 660 if (start < vm_physmem[lcv].start) 661 break; 662 ps = &vm_physmem[lcv]; 663 /* move back other entries, if necessary ... */ 664 for (x = vm_nphysseg ; x > lcv ; x--) 665 /* structure copy */ 666 vm_physmem[x] = vm_physmem[x - 1]; 667 } 668 669#elif (VM_PHYSSEG_STRAT == VM_PSTRAT_BIGFIRST) 670 671 { 672 int x; 673 /* sort by largest segment first */ 674 for (lcv = 0 ; lcv < vm_nphysseg ; lcv++) 675 if ((end - start) > 676 (vm_physmem[lcv].end - vm_physmem[lcv].start)) 677 break; 678 ps = &vm_physmem[lcv]; 679 /* move back other entries, if necessary ... */ 680 for (x = vm_nphysseg ; x > lcv ; x--) 681 /* structure copy */ 682 vm_physmem[x] = vm_physmem[x - 1]; 683 } 684 685#else 686 687 panic("vm_page_physload: unknown physseg strategy selected!"); 688 689#endif 690 691 ps->start = start; 692 ps->end = end; 693 ps->avail_start = avail_start; 694 ps->avail_end = avail_end; 695 if (preload) { 696 ps->pgs = NULL; 697 } else { 698 ps->pgs = pgs; 699 ps->lastpg = pgs + npages - 1; 700 } 701 ps->free_list = free_list; 702 vm_nphysseg++; 703 704 /* 705 * done! 706 */ 707 708 if (!preload) 709 uvm_page_rehash(); 710 711 return; 712} 713 714/* 715 * uvm_page_rehash: reallocate hash table based on number of free pages. 716 */ 717 718void 719uvm_page_rehash() 720{ 721 int freepages, lcv, bucketcount, s, oldcount; 722 struct pglist *newbuckets, *oldbuckets; 723 struct vm_page *pg; 724 725 /* 726 * compute number of pages that can go in the free pool 727 */ 728 729 freepages = 0; 730 for (lcv = 0 ; lcv < vm_nphysseg ; lcv++) 731 freepages += 732 (vm_physmem[lcv].avail_end - vm_physmem[lcv].avail_start); 733 734 /* 735 * compute number of buckets needed for this number of pages 736 */ 737 738 bucketcount = 1; 739 while (bucketcount < freepages) 740 bucketcount = bucketcount * 2; 741 742 /* 743 * malloc new buckets 744 */ 745 746 MALLOC(newbuckets, struct pglist *, sizeof(struct pglist) * bucketcount, 747 M_VMPBUCKET, M_NOWAIT); 748 if (newbuckets == NULL) { 749 printf("uvm_page_physrehash: WARNING: could not grow page " 750 "hash table\n"); 751 return; 752 } 753 for (lcv = 0 ; lcv < bucketcount ; lcv++) 754 TAILQ_INIT(&newbuckets[lcv]); 755 756 /* 757 * now replace the old buckets with the new ones and rehash everything 758 */ 759 760 s = splimp(); 761 simple_lock(&uvm.hashlock); 762 /* swap old for new ... */ 763 oldbuckets = uvm.page_hash; 764 oldcount = uvm.page_nhash; 765 uvm.page_hash = newbuckets; 766 uvm.page_nhash = bucketcount; 767 uvm.page_hashmask = bucketcount - 1; /* power of 2 */ 768 769 /* ... and rehash */ 770 for (lcv = 0 ; lcv < oldcount ; lcv++) { 771 while ((pg = oldbuckets[lcv].tqh_first) != NULL) { 772 TAILQ_REMOVE(&oldbuckets[lcv], pg, hashq); 773 TAILQ_INSERT_TAIL( 774 &uvm.page_hash[uvm_pagehash(pg->uobject, pg->offset)], 775 pg, hashq); 776 } 777 } 778 simple_unlock(&uvm.hashlock); 779 splx(s); 780 781 /* 782 * free old bucket array if is not the boot-time table 783 */ 784 785 if (oldbuckets != &uvm_bootbucket) 786 FREE(oldbuckets, M_VMPBUCKET); 787 788 /* 789 * done 790 */ 791 return; 792} 793 794 795#if 1 /* XXXCDC: TMP TMP TMP DEBUG DEBUG DEBUG */ 796 797void uvm_page_physdump __P((void)); /* SHUT UP GCC */ 798 799/* call from DDB */ 800void 801uvm_page_physdump() 802{ 803 int lcv; 804 805 printf("rehash: physical memory config [segs=%d of %d]:\n", 806 vm_nphysseg, VM_PHYSSEG_MAX); 807 for (lcv = 0 ; lcv < vm_nphysseg ; lcv++) 808 printf("0x%lx->0x%lx [0x%lx->0x%lx]\n", vm_physmem[lcv].start, 809 vm_physmem[lcv].end, vm_physmem[lcv].avail_start, 810 vm_physmem[lcv].avail_end); 811 printf("STRATEGY = "); 812 switch (VM_PHYSSEG_STRAT) { 813 case VM_PSTRAT_RANDOM: printf("RANDOM\n"); break; 814 case VM_PSTRAT_BSEARCH: printf("BSEARCH\n"); break; 815 case VM_PSTRAT_BIGFIRST: printf("BIGFIRST\n"); break; 816 default: printf("<<UNKNOWN>>!!!!\n"); 817 } 818 printf("number of buckets = %d\n", uvm.page_nhash); 819} 820#endif 821 822/* 823 * uvm_pagealloc_strat: allocate vm_page from a particular free list. 824 * 825 * => return null if no pages free 826 * => wake up pagedaemon if number of free pages drops below low water mark 827 * => if obj != NULL, obj must be locked (to put in hash) 828 * => if anon != NULL, anon must be locked (to put in anon) 829 * => only one of obj or anon can be non-null 830 * => caller must activate/deactivate page if it is not wired. 831 * => free_list is ignored if strat == UVM_PGA_STRAT_NORMAL. 832 */ 833 834struct vm_page * 835uvm_pagealloc_strat(obj, off, anon, flags, strat, free_list) 836 struct uvm_object *obj; 837 vaddr_t off; 838 int flags; 839 struct vm_anon *anon; 840 int strat, free_list; 841{ 842 int lcv, s; 843 struct vm_page *pg; 844 struct pglist *freeq; 845 boolean_t use_reserve; 846 847#ifdef DIAGNOSTIC 848 /* sanity check */ 849 if (obj && anon) 850 panic("uvm_pagealloc: obj and anon != NULL"); 851#endif 852 853 s = uvm_lock_fpageq(); /* lock free page queue */ 854 855 /* 856 * check to see if we need to generate some free pages waking 857 * the pagedaemon. 858 */ 859 860 if (uvmexp.free < uvmexp.freemin || (uvmexp.free < uvmexp.freetarg && 861 uvmexp.inactive < uvmexp.inactarg)) 862 wakeup(&uvm.pagedaemon); 863 864 /* 865 * fail if any of these conditions is true: 866 * [1] there really are no free pages, or 867 * [2] only kernel "reserved" pages remain and 868 * the page isn't being allocated to a kernel object. 869 * [3] only pagedaemon "reserved" pages remain and 870 * the requestor isn't the pagedaemon. 871 */ 872 873 use_reserve = (flags & UVM_PGA_USERESERVE) || 874 (obj && UVM_OBJ_IS_KERN_OBJECT(obj)); 875 if ((uvmexp.free <= uvmexp.reserve_kernel && !use_reserve) || 876 (uvmexp.free <= uvmexp.reserve_pagedaemon && 877 !(use_reserve && (curproc == uvm.pagedaemon_proc || 878 curproc == syncerproc)))) 879 goto fail; 880 881 again: 882 switch (strat) { 883 case UVM_PGA_STRAT_NORMAL: 884 /* Check all freelists in descending priority order. */ 885 for (lcv = 0; lcv < VM_NFREELIST; lcv++) { 886 freeq = &uvm.page_free[lcv]; 887 if ((pg = freeq->tqh_first) != NULL) 888 goto gotit; 889 } 890 891 /* No pages free! */ 892 goto fail; 893 894 case UVM_PGA_STRAT_ONLY: 895 case UVM_PGA_STRAT_FALLBACK: 896 /* Attempt to allocate from the specified free list. */ 897#ifdef DIAGNOSTIC 898 if (free_list >= VM_NFREELIST || free_list < 0) 899 panic("uvm_pagealloc_strat: bad free list %d", 900 free_list); 901#endif 902 freeq = &uvm.page_free[free_list]; 903 if ((pg = freeq->tqh_first) != NULL) 904 goto gotit; 905 906 /* Fall back, if possible. */ 907 if (strat == UVM_PGA_STRAT_FALLBACK) { 908 strat = UVM_PGA_STRAT_NORMAL; 909 goto again; 910 } 911 912 /* No pages free! */ 913 goto fail; 914 915 default: 916 panic("uvm_pagealloc_strat: bad strat %d", strat); 917 /* NOTREACHED */ 918 } 919 920 gotit: 921 TAILQ_REMOVE(freeq, pg, pageq); 922 uvmexp.free--; 923 924 uvm_unlock_fpageq(s); /* unlock free page queue */ 925 926 pg->offset = off; 927 pg->uobject = obj; 928 pg->uanon = anon; 929 pg->flags = PG_BUSY|PG_CLEAN|PG_FAKE; 930 pg->version++; 931 pg->wire_count = 0; 932 pg->loan_count = 0; 933 if (anon) { 934 anon->u.an_page = pg; 935 pg->pqflags = PQ_ANON; 936 } else { 937 if (obj) 938 uvm_pageinsert(pg); 939 pg->pqflags = 0; 940 } 941#if defined(UVM_PAGE_TRKOWN) 942 pg->owner_tag = NULL; 943#endif 944 UVM_PAGE_OWN(pg, "new alloc"); 945 946 return(pg); 947 948 fail: 949 uvm_unlock_fpageq(s); 950 return (NULL); 951} 952 953/* 954 * uvm_pagealloc_contig: allocate contiguous memory. 955 * 956 * XXX - fix comment. 957 */ 958 959vaddr_t 960uvm_pagealloc_contig(size, low, high, alignment) 961 vaddr_t size; 962 vaddr_t low, high; 963 vaddr_t alignment; 964{ 965 struct pglist pglist; 966 struct vm_page *pg; 967 vaddr_t addr, temp_addr; 968 969 size = round_page(size); 970 971 TAILQ_INIT(&pglist); 972 if (uvm_pglistalloc(size, low, high, alignment, 0, 973 &pglist, 1, FALSE)) 974 return 0; 975 addr = vm_map_min(kernel_map); 976 if (uvm_map(kernel_map, &addr, size, NULL, UVM_UNKNOWN_OFFSET, 977 UVM_MAPFLAG(UVM_PROT_ALL, UVM_PROT_ALL, UVM_INH_NONE, 978 UVM_ADV_RANDOM, 0)) != KERN_SUCCESS) { 979 uvm_pglistfree(&pglist); 980 return 0; 981 } 982 temp_addr = addr; 983 for (pg = TAILQ_FIRST(&pglist); pg != NULL; 984 pg = TAILQ_NEXT(pg, pageq)) { 985 pg->uobject = uvm.kernel_object; 986 pg->offset = temp_addr - vm_map_min(kernel_map); 987 uvm_pageinsert(pg); 988 uvm_pagewire(pg); 989#if defined(PMAP_NEW) 990 pmap_kenter_pa(temp_addr, VM_PAGE_TO_PHYS(pg), 991 VM_PROT_READ|VM_PROT_WRITE); 992#else 993 pmap_enter(pmap_kernel(), temp_addr, VM_PAGE_TO_PHYS(pg), 994 VM_PROT_READ|VM_PROT_WRITE, TRUE, 995 VM_PROT_READ|VM_PROT_WRITE); 996#endif 997 temp_addr += PAGE_SIZE; 998 } 999 return addr; 1000} 1001 1002/* 1003 * uvm_pagerealloc: reallocate a page from one object to another 1004 * 1005 * => both objects must be locked 1006 */ 1007 1008void 1009uvm_pagerealloc(pg, newobj, newoff) 1010 struct vm_page *pg; 1011 struct uvm_object *newobj; 1012 vaddr_t newoff; 1013{ 1014 /* 1015 * remove it from the old object 1016 */ 1017 1018 if (pg->uobject) { 1019 uvm_pageremove(pg); 1020 } 1021 1022 /* 1023 * put it in the new object 1024 */ 1025 1026 if (newobj) { 1027 pg->uobject = newobj; 1028 pg->offset = newoff; 1029 pg->version++; 1030 uvm_pageinsert(pg); 1031 } 1032 1033 return; 1034} 1035 1036 1037/* 1038 * uvm_pagefree: free page 1039 * 1040 * => erase page's identity (i.e. remove from hash/object) 1041 * => put page on free list 1042 * => caller must lock owning object (either anon or uvm_object) 1043 * => caller must lock page queues 1044 * => assumes all valid mappings of pg are gone 1045 */ 1046 1047void uvm_pagefree(pg) 1048 1049struct vm_page *pg; 1050 1051{ 1052 int s; 1053 int saved_loan_count = pg->loan_count; 1054 1055 /* 1056 * if the page was an object page (and thus "TABLED"), remove it 1057 * from the object. 1058 */ 1059 1060 if (pg->flags & PG_TABLED) { 1061 1062 /* 1063 * if the object page is on loan we are going to drop ownership. 1064 * it is possible that an anon will take over as owner for this 1065 * page later on. the anon will want a !PG_CLEAN page so that 1066 * it knows it needs to allocate swap if it wants to page the 1067 * page out. 1068 */ 1069 1070 if (saved_loan_count) 1071 pg->flags &= ~PG_CLEAN; /* in case an anon takes over */ 1072 1073 uvm_pageremove(pg); 1074 1075 /* 1076 * if our page was on loan, then we just lost control over it 1077 * (in fact, if it was loaned to an anon, the anon may have 1078 * already taken over ownership of the page by now and thus 1079 * changed the loan_count [e.g. in uvmfault_anonget()]) we just 1080 * return (when the last loan is dropped, then the page can be 1081 * freed by whatever was holding the last loan). 1082 */ 1083 if (saved_loan_count) 1084 return; 1085 1086 } else if (saved_loan_count && (pg->pqflags & PQ_ANON)) { 1087 1088 /* 1089 * if our page is owned by an anon and is loaned out to the 1090 * kernel then we just want to drop ownership and return. 1091 * the kernel must free the page when all its loans clear ... 1092 * note that the kernel can't change the loan status of our 1093 * page as long as we are holding PQ lock. 1094 */ 1095 pg->pqflags &= ~PQ_ANON; 1096 pg->uanon = NULL; 1097 return; 1098 } 1099 1100#ifdef DIAGNOSTIC 1101 if (saved_loan_count) { 1102 printf("uvm_pagefree: warning: freeing page with a loan " 1103 "count of %d\n", saved_loan_count); 1104 panic("uvm_pagefree: loan count"); 1105 } 1106#endif 1107 1108 1109 /* 1110 * now remove the page from the queues 1111 */ 1112 1113 if (pg->pqflags & PQ_ACTIVE) { 1114 TAILQ_REMOVE(&uvm.page_active, pg, pageq); 1115 pg->pqflags &= ~PQ_ACTIVE; 1116 uvmexp.active--; 1117 } 1118 if (pg->pqflags & PQ_INACTIVE) { 1119 if (pg->pqflags & PQ_SWAPBACKED) 1120 TAILQ_REMOVE(&uvm.page_inactive_swp, pg, pageq); 1121 else 1122 TAILQ_REMOVE(&uvm.page_inactive_obj, pg, pageq); 1123 pg->pqflags &= ~PQ_INACTIVE; 1124 uvmexp.inactive--; 1125 } 1126 1127 /* 1128 * if the page was wired, unwire it now. 1129 */ 1130 if (pg->wire_count) 1131 { 1132 pg->wire_count = 0; 1133 uvmexp.wired--; 1134 } 1135 1136 /* 1137 * and put on free queue 1138 */ 1139 1140 s = uvm_lock_fpageq(); 1141 TAILQ_INSERT_TAIL(&uvm.page_free[uvm_page_lookup_freelist(pg)], 1142 pg, pageq); 1143 pg->pqflags = PQ_FREE; 1144#ifdef DEBUG 1145 pg->uobject = (void *)0xdeadbeef; 1146 pg->offset = 0xdeadbeef; 1147 pg->uanon = (void *)0xdeadbeef; 1148#endif 1149 uvmexp.free++; 1150 uvm_unlock_fpageq(s); 1151} 1152 1153#if defined(UVM_PAGE_TRKOWN) 1154/* 1155 * uvm_page_own: set or release page ownership 1156 * 1157 * => this is a debugging function that keeps track of who sets PG_BUSY 1158 * and where they do it. it can be used to track down problems 1159 * such a process setting "PG_BUSY" and never releasing it. 1160 * => page's object [if any] must be locked 1161 * => if "tag" is NULL then we are releasing page ownership 1162 */ 1163void 1164uvm_page_own(pg, tag) 1165 struct vm_page *pg; 1166 char *tag; 1167{ 1168 /* gain ownership? */ 1169 if (tag) { 1170 if (pg->owner_tag) { 1171 printf("uvm_page_own: page %p already owned " 1172 "by proc %d [%s]\n", pg, 1173 pg->owner, pg->owner_tag); 1174 panic("uvm_page_own"); 1175 } 1176 pg->owner = (curproc) ? curproc->p_pid : (pid_t) -1; 1177 pg->owner_tag = tag; 1178 return; 1179 } 1180 1181 /* drop ownership */ 1182 if (pg->owner_tag == NULL) { 1183 printf("uvm_page_own: dropping ownership of an non-owned " 1184 "page (%p)\n", pg); 1185 panic("uvm_page_own"); 1186 } 1187 pg->owner_tag = NULL; 1188 return; 1189} 1190#endif 1191