uvm_page.h revision 1.23
1/* $OpenBSD: uvm_page.h,v 1.23 2007/04/04 17:44:45 art Exp $ */ 2/* $NetBSD: uvm_page.h,v 1.19 2000/12/28 08:24:55 chs 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.h 7.3 (Berkeley) 4/21/91 43 * from: Id: uvm_page.h,v 1.1.2.6 1998/02/04 02:31:42 chuck 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#ifndef _UVM_UVM_PAGE_H_ 71#define _UVM_UVM_PAGE_H_ 72 73/* 74 * uvm_page.h 75 */ 76 77/* 78 * Resident memory system definitions. 79 */ 80 81/* 82 * Management of resident (logical) pages. 83 * 84 * A small structure is kept for each resident 85 * page, indexed by page number. Each structure 86 * is an element of several lists: 87 * 88 * A hash table bucket used to quickly 89 * perform object/offset lookups 90 * 91 * A list of all pages for a given object, 92 * so they can be quickly deactivated at 93 * time of deallocation. 94 * 95 * An ordered list of pages due for pageout. 96 * 97 * In addition, the structure contains the object 98 * and offset to which this page belongs (for pageout), 99 * and sundry status bits. 100 * 101 * Fields in this structure are locked either by the lock on the 102 * object that the page belongs to (O) or by the lock on the page 103 * queues (P) [or both]. 104 */ 105 106/* 107 * locking note: the mach version of this data structure had bit 108 * fields for the flags, and the bit fields were divided into two 109 * items (depending on who locked what). some time, in BSD, the bit 110 * fields were dumped and all the flags were lumped into one short. 111 * that is fine for a single threaded uniprocessor OS, but bad if you 112 * want to actual make use of locking (simple_lock's). so, we've 113 * separated things back out again. 114 * 115 * note the page structure has no lock of its own. 116 */ 117 118#include <uvm/uvm_extern.h> 119#include <uvm/uvm_pglist.h> 120 121struct vm_page { 122 TAILQ_ENTRY(vm_page) pageq; /* queue info for FIFO 123 * queue or free list (P) */ 124 TAILQ_ENTRY(vm_page) hashq; /* hash table links (O)*/ 125 TAILQ_ENTRY(vm_page) listq; /* pages in same object (O)*/ 126 127 struct vm_anon *uanon; /* anon (O,P) */ 128 struct uvm_object *uobject; /* object (O,P) */ 129 voff_t offset; /* offset into object (O,P) */ 130 131 u_short pg_flags; /* object flags [O] */ 132 u_short pg_version; /* version count [O] */ 133 u_short wire_count; /* wired down map refs [P] */ 134 u_short pqflags; /* page queue flags [P] */ 135 u_int loan_count; /* number of active loans 136 * to read: [O or P] 137 * to modify: [O _and_ P] */ 138 paddr_t phys_addr; /* physical address of page */ 139 140#ifdef __HAVE_VM_PAGE_MD 141 struct vm_page_md mdpage; /* pmap-specific data */ 142#endif 143#if defined(UVM_PAGE_TRKOWN) 144 /* debugging fields to track page ownership */ 145 pid_t owner; /* proc that set PG_BUSY */ 146 char *owner_tag; /* why it was set busy */ 147#endif 148}; 149 150/* 151 * These are the flags defined for vm_page. 152 * 153 * Note: PG_FILLED and PG_DIRTY are added for the filesystems. 154 */ 155 156/* 157 * locking rules: 158 * PG_ ==> locked by object lock 159 * PQ_ ==> lock by page queue lock 160 * PQ_FREE is locked by free queue lock and is mutex with all other PQs 161 * 162 * PG_ZERO is used to indicate that a page has been pre-zero'd. This flag 163 * is only set when the page is on no queues, and is cleared when the page 164 * is placed on the free list. 165 */ 166 167#define PG_BUSY 0x0001 /* page is locked */ 168#define PG_WANTED 0x0002 /* someone is waiting for page */ 169#define PG_TABLED 0x0004 /* page is in VP table */ 170#define PG_CLEAN 0x0008 /* page has not been modified */ 171#define PG_CLEANCHK 0x0010 /* clean bit has been checked */ 172#define PG_RELEASED 0x0020 /* page released while paging */ 173#define PG_FAKE 0x0040 /* page is not yet initialized */ 174#define PG_RDONLY 0x0080 /* page must be mapped read-only */ 175#define PG_ZERO 0x0100 /* page is pre-zero'd */ 176 177#define PG_PAGER1 0x1000 /* pager-specific flag */ 178 179#define PQ_FREE 0x0001 /* page is on free list */ 180#define PQ_INACTIVE 0x0002 /* page is in inactive list */ 181#define PQ_ACTIVE 0x0004 /* page is in active list */ 182#define PQ_ANON 0x0010 /* page is part of an anon, rather 183 than an uvm_object */ 184#define PQ_AOBJ 0x0020 /* page is part of an anonymous 185 uvm_object */ 186#define PQ_SWAPBACKED (PQ_ANON|PQ_AOBJ) 187#define PQ_ENCRYPT 0x0040 /* page needs {en,de}cryption */ 188 189/* 190 * physical memory layout structure 191 * 192 * MD vmparam.h must #define: 193 * VM_PHYSEG_MAX = max number of physical memory segments we support 194 * (if this is "1" then we revert to a "contig" case) 195 * VM_PHYSSEG_STRAT: memory sort/search options (for VM_PHYSEG_MAX > 1) 196 * - VM_PSTRAT_RANDOM: linear search (random order) 197 * - VM_PSTRAT_BSEARCH: binary search (sorted by address) 198 * - VM_PSTRAT_BIGFIRST: linear search (sorted by largest segment first) 199 * - others? 200 * XXXCDC: eventually we should purge all left-over global variables... 201 */ 202#define VM_PSTRAT_RANDOM 1 203#define VM_PSTRAT_BSEARCH 2 204#define VM_PSTRAT_BIGFIRST 3 205 206/* 207 * vm_physmemseg: describes one segment of physical memory 208 */ 209struct vm_physseg { 210 paddr_t start; /* PF# of first page in segment */ 211 paddr_t end; /* (PF# of last page in segment) + 1 */ 212 paddr_t avail_start; /* PF# of first free page in segment */ 213 paddr_t avail_end; /* (PF# of last free page in segment) +1 */ 214 int free_list; /* which free list they belong on */ 215 struct vm_page *pgs; /* vm_page structures (from start) */ 216 struct vm_page *lastpg; /* vm_page structure for end */ 217#ifdef __HAVE_PMAP_PHYSSEG 218 struct pmap_physseg pmseg; /* pmap specific (MD) data */ 219#endif 220}; 221 222#ifdef _KERNEL 223 224/* 225 * globals 226 */ 227 228extern boolean_t vm_page_zero_enable; 229 230/* 231 * Each pageable resident page falls into one of three lists: 232 * 233 * free 234 * Available for allocation now. 235 * inactive 236 * Not referenced in any map, but still has an 237 * object/offset-page mapping, and may be dirty. 238 * This is the list of pages that should be 239 * paged out next. 240 * active 241 * A list of pages which have been placed in 242 * at least one physical map. This list is 243 * ordered, in LRU-like fashion. 244 */ 245 246extern struct pglist vm_page_queue_free; /* memory free queue */ 247extern struct pglist vm_page_queue_active; /* active memory queue */ 248extern struct pglist vm_page_queue_inactive; /* inactive memory queue */ 249 250/* 251 * physical memory config is stored in vm_physmem. 252 */ 253 254extern struct vm_physseg vm_physmem[VM_PHYSSEG_MAX]; 255extern int vm_nphysseg; 256 257/* 258 * handle inline options 259 */ 260 261#ifdef UVM_PAGE_INLINE 262#define PAGE_INLINE static __inline 263#else 264#define PAGE_INLINE /* nothing */ 265#endif /* UVM_PAGE_INLINE */ 266 267/* 268 * prototypes: the following prototypes define the interface to pages 269 */ 270 271void uvm_page_init(vaddr_t *, vaddr_t *); 272#if defined(UVM_PAGE_TRKOWN) 273void uvm_page_own(struct vm_page *, char *); 274#endif 275#if !defined(PMAP_STEAL_MEMORY) 276boolean_t uvm_page_physget(paddr_t *); 277#endif 278void uvm_page_rehash(void); 279void uvm_pageidlezero(void); 280 281PAGE_INLINE int uvm_lock_fpageq(void); 282PAGE_INLINE void uvm_unlock_fpageq(int); 283 284PAGE_INLINE void uvm_pageactivate(struct vm_page *); 285vaddr_t uvm_pageboot_alloc(vsize_t); 286PAGE_INLINE void uvm_pagecopy(struct vm_page *, struct vm_page *); 287PAGE_INLINE void uvm_pagedeactivate(struct vm_page *); 288void uvm_pagefree(struct vm_page *); 289void uvm_page_unbusy(struct vm_page **, int); 290PAGE_INLINE struct vm_page *uvm_pagelookup(struct uvm_object *, voff_t); 291PAGE_INLINE void uvm_pageunwire(struct vm_page *); 292PAGE_INLINE void uvm_pagewait(struct vm_page *, int); 293PAGE_INLINE void uvm_pagewake(struct vm_page *); 294PAGE_INLINE void uvm_pagewire(struct vm_page *); 295PAGE_INLINE void uvm_pagezero(struct vm_page *); 296 297PAGE_INLINE int uvm_page_lookup_freelist(struct vm_page *); 298 299static struct vm_page *PHYS_TO_VM_PAGE(paddr_t); 300static int vm_physseg_find(paddr_t, int *); 301 302/* 303 * macros 304 */ 305 306#define uvm_lock_pageq() simple_lock(&uvm.pageqlock) 307#define uvm_unlock_pageq() simple_unlock(&uvm.pageqlock) 308 309#define uvm_pagehash(obj,off) \ 310 (((unsigned long)obj+(unsigned long)atop(off)) & uvm.page_hashmask) 311 312#define UVM_PAGEZERO_TARGET (uvmexp.free) 313 314#define VM_PAGE_TO_PHYS(entry) ((entry)->phys_addr) 315 316/* 317 * when VM_PHYSSEG_MAX is 1, we can simplify these functions 318 */ 319 320/* 321 * vm_physseg_find: find vm_physseg structure that belongs to a PA 322 */ 323static __inline int 324vm_physseg_find(pframe, offp) 325 paddr_t pframe; 326 int *offp; 327{ 328#if VM_PHYSSEG_MAX == 1 329 330 /* 'contig' case */ 331 if (pframe >= vm_physmem[0].start && pframe < vm_physmem[0].end) { 332 if (offp) 333 *offp = pframe - vm_physmem[0].start; 334 return(0); 335 } 336 return(-1); 337 338#elif (VM_PHYSSEG_STRAT == VM_PSTRAT_BSEARCH) 339 /* binary search for it */ 340 int start, len, try; 341 342 /* 343 * if try is too large (thus target is less than than try) we reduce 344 * the length to trunc(len/2) [i.e. everything smaller than "try"] 345 * 346 * if the try is too small (thus target is greater than try) then 347 * we set the new start to be (try + 1). this means we need to 348 * reduce the length to (round(len/2) - 1). 349 * 350 * note "adjust" below which takes advantage of the fact that 351 * (round(len/2) - 1) == trunc((len - 1) / 2) 352 * for any value of len we may have 353 */ 354 355 for (start = 0, len = vm_nphysseg ; len != 0 ; len = len / 2) { 356 try = start + (len / 2); /* try in the middle */ 357 358 /* start past our try? */ 359 if (pframe >= vm_physmem[try].start) { 360 /* was try correct? */ 361 if (pframe < vm_physmem[try].end) { 362 if (offp) 363 *offp = pframe - vm_physmem[try].start; 364 return(try); /* got it */ 365 } 366 start = try + 1; /* next time, start here */ 367 len--; /* "adjust" */ 368 } else { 369 /* 370 * pframe before try, just reduce length of 371 * region, done in "for" loop 372 */ 373 } 374 } 375 return(-1); 376 377#else 378 /* linear search for it */ 379 int lcv; 380 381 for (lcv = 0; lcv < vm_nphysseg; lcv++) { 382 if (pframe >= vm_physmem[lcv].start && 383 pframe < vm_physmem[lcv].end) { 384 if (offp) 385 *offp = pframe - vm_physmem[lcv].start; 386 return(lcv); /* got it */ 387 } 388 } 389 return(-1); 390 391#endif 392} 393 394/* 395 * PHYS_TO_VM_PAGE: find vm_page for a PA. used by MI code to get vm_pages 396 * back from an I/O mapping (ugh!). used in some MD code as well. 397 */ 398static __inline struct vm_page * 399PHYS_TO_VM_PAGE(pa) 400 paddr_t pa; 401{ 402 paddr_t pf = atop(pa); 403 int off; 404 int psi; 405 406 psi = vm_physseg_find(pf, &off); 407 if (psi != -1) 408 return(&vm_physmem[psi].pgs[off]); 409 return(NULL); 410} 411 412#define VM_PAGE_IS_FREE(entry) ((entry)->pqflags & PQ_FREE) 413 414#endif /* _KERNEL */ 415 416#endif /* _UVM_UVM_PAGE_H_ */ 417