1 /* $NetBSD: uvm_page.h,v 1.32 2002/11/08 02:05:16 enami 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.h 7.3 (Berkeley) 4/21/91 42 * from: Id: uvm_page.h,v 1.1.2.6 1998/02/04 02:31:42 chuck 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 #ifndef _UVM_UVM_PAGE_H_ 70 #define _UVM_UVM_PAGE_H_ 71 72 /* 73 * uvm_page.h 74 */ 75 76 /* 77 * Resident memory system definitions. 78 */ 79 80 /* 81 * Management of resident (logical) pages. 82 * 83 * A small structure is kept for each resident 84 * page, indexed by page number. Each structure 85 * is an element of several lists: 86 * 87 * A hash table bucket used to quickly 88 * perform object/offset lookups 89 * 90 * A list of all pages for a given object, 91 * so they can be quickly deactivated at 92 * time of deallocation. 93 * 94 * An ordered list of pages due for pageout. 95 * 96 * In addition, the structure contains the object 97 * and offset to which this page belongs (for pageout), 98 * and sundry status bits. 99 * 100 * Fields in this structure are locked either by the lock on the 101 * object that the page belongs to (O) or by the lock on the page 102 * queues (P) [or both]. 103 */ 104 105 /* 106 * locking note: the mach version of this data structure had bit 107 * fields for the flags, and the bit fields were divided into two 108 * items (depending on who locked what). some time, in BSD, the bit 109 * fields were dumped and all the flags were lumped into one short. 110 * that is fine for a single threaded uniprocessor OS, but bad if you 111 * want to actual make use of locking (simple_lock's). so, we've 112 * separated things back out again. 113 * 114 * note the page structure has no lock of its own. 115 */ 116 117 #include <uvm/uvm_extern.h> 118 #include <uvm/uvm_pglist.h> 119 120 struct vm_page { 121 TAILQ_ENTRY(vm_page) pageq; /* queue info for FIFO 122 * queue or free list (P) */ 123 TAILQ_ENTRY(vm_page) hashq; /* hash table links (O)*/ 124 TAILQ_ENTRY(vm_page) listq; /* pages in same object (O)*/ 125 126 struct vm_anon *uanon; /* anon (O,P) */ 127 struct uvm_object *uobject; /* object (O,P) */ 128 voff_t offset; /* offset into object (O,P) */ 129 uint16_t flags; /* object flags [O] */ 130 uint16_t loan_count; /* number of active loans 131 * to read: [O or P] 132 * to modify: [O _and_ P] */ 133 uint16_t wire_count; /* wired down map refs [P] */ 134 uint16_t pqflags; /* page queue flags [P] */ 135 paddr_t phys_addr; /* physical address of page */ 136 137 #ifdef __HAVE_VM_PAGE_MD 138 struct vm_page_md mdpage; /* pmap-specific data */ 139 #endif 140 141 #if defined(UVM_PAGE_TRKOWN) 142 /* debugging fields to track page ownership */ 143 pid_t owner; /* proc that set PG_BUSY */ 144 char *owner_tag; /* why it was set busy */ 145 #endif 146 }; 147 148 /* 149 * These are the flags defined for vm_page. 150 */ 151 152 /* 153 * locking rules: 154 * PG_ ==> locked by object lock 155 * PQ_ ==> lock by page queue lock 156 * PQ_FREE is locked by free queue lock and is mutex with all other PQs 157 * 158 * PG_ZERO is used to indicate that a page has been pre-zero'd. This flag 159 * is only set when the page is on no queues, and is cleared when the page 160 * is placed on the free list. 161 */ 162 163 #define PG_BUSY 0x0001 /* page is locked */ 164 #define PG_WANTED 0x0002 /* someone is waiting for page */ 165 #define PG_TABLED 0x0004 /* page is in VP table */ 166 #define PG_CLEAN 0x0008 /* page has not been modified */ 167 #define PG_PAGEOUT 0x0010 /* page to be freed for pagedaemon */ 168 #define PG_RELEASED 0x0020 /* page to be freed when unbusied */ 169 #define PG_FAKE 0x0040 /* page is not yet initialized */ 170 #define PG_RDONLY 0x0080 /* page must be mapped read-only */ 171 #define PG_ZERO 0x0100 /* page is pre-zero'd */ 172 173 #define PG_PAGER1 0x1000 /* pager-specific flag */ 174 175 #define PQ_FREE 0x01 /* page is on free list */ 176 #define PQ_INACTIVE 0x02 /* page is in inactive list */ 177 #define PQ_ACTIVE 0x04 /* page is in active list */ 178 #define PQ_ANON 0x10 /* page is part of an anon, rather 179 than an uvm_object */ 180 #define PQ_AOBJ 0x20 /* page is part of an anonymous 181 uvm_object */ 182 #define PQ_SWAPBACKED (PQ_ANON|PQ_AOBJ) 183 184 /* 185 * physical memory layout structure 186 * 187 * MD vmparam.h must #define: 188 * VM_PHYSEG_MAX = max number of physical memory segments we support 189 * (if this is "1" then we revert to a "contig" case) 190 * VM_PHYSSEG_STRAT: memory sort/search options (for VM_PHYSEG_MAX > 1) 191 * - VM_PSTRAT_RANDOM: linear search (random order) 192 * - VM_PSTRAT_BSEARCH: binary search (sorted by address) 193 * - VM_PSTRAT_BIGFIRST: linear search (sorted by largest segment first) 194 * - others? 195 * XXXCDC: eventually we should purge all left-over global variables... 196 */ 197 #define VM_PSTRAT_RANDOM 1 198 #define VM_PSTRAT_BSEARCH 2 199 #define VM_PSTRAT_BIGFIRST 3 200 201 /* 202 * vm_physmemseg: describes one segment of physical memory 203 */ 204 struct vm_physseg { 205 paddr_t start; /* PF# of first page in segment */ 206 paddr_t end; /* (PF# of last page in segment) + 1 */ 207 paddr_t avail_start; /* PF# of first free page in segment */ 208 paddr_t avail_end; /* (PF# of last free page in segment) +1 */ 209 int free_list; /* which free list they belong on */ 210 struct vm_page *pgs; /* vm_page structures (from start) */ 211 struct vm_page *lastpg; /* vm_page structure for end */ 212 #ifdef __HAVE_PMAP_PHYSSEG 213 struct pmap_physseg pmseg; /* pmap specific (MD) data */ 214 #endif 215 }; 216 217 #ifdef _KERNEL 218 219 /* 220 * globals 221 */ 222 223 extern boolean_t vm_page_zero_enable; 224 225 /* 226 * Each pageable resident page falls into one of three lists: 227 * 228 * free 229 * Available for allocation now. 230 * inactive 231 * Not referenced in any map, but still has an 232 * object/offset-page mapping, and may be dirty. 233 * This is the list of pages that should be 234 * paged out next. 235 * active 236 * A list of pages which have been placed in 237 * at least one physical map. This list is 238 * ordered, in LRU-like fashion. 239 */ 240 241 extern struct pglist vm_page_queue_free; /* memory free queue */ 242 extern struct pglist vm_page_queue_active; /* active memory queue */ 243 extern struct pglist vm_page_queue_inactive; /* inactive memory queue */ 244 245 /* 246 * physical memory config is stored in vm_physmem. 247 */ 248 249 extern struct vm_physseg vm_physmem[VM_PHYSSEG_MAX]; 250 extern int vm_nphysseg; 251 252 /* 253 * handle inline options 254 */ 255 256 #ifdef UVM_PAGE_INLINE 257 #define PAGE_INLINE static __inline 258 #else 259 #define PAGE_INLINE /* nothing */ 260 #endif /* UVM_PAGE_INLINE */ 261 262 /* 263 * prototypes: the following prototypes define the interface to pages 264 */ 265 266 void uvm_page_init __P((vaddr_t *, vaddr_t *)); 267 #if defined(UVM_PAGE_TRKOWN) 268 void uvm_page_own __P((struct vm_page *, char *)); 269 #endif 270 #if !defined(PMAP_STEAL_MEMORY) 271 boolean_t uvm_page_physget __P((paddr_t *)); 272 #endif 273 void uvm_page_rehash __P((void)); 274 void uvm_page_recolor __P((int)); 275 void uvm_pageidlezero __P((void)); 276 277 PAGE_INLINE int uvm_lock_fpageq __P((void)); 278 PAGE_INLINE void uvm_unlock_fpageq __P((int)); 279 280 PAGE_INLINE void uvm_pageactivate __P((struct vm_page *)); 281 vaddr_t uvm_pageboot_alloc __P((vsize_t)); 282 PAGE_INLINE void uvm_pagecopy __P((struct vm_page *, struct vm_page *)); 283 PAGE_INLINE void uvm_pagedeactivate __P((struct vm_page *)); 284 PAGE_INLINE void uvm_pagedequeue __P((struct vm_page *)); 285 void uvm_pagefree __P((struct vm_page *)); 286 void uvm_page_unbusy __P((struct vm_page **, int)); 287 PAGE_INLINE struct vm_page *uvm_pagelookup __P((struct uvm_object *, voff_t)); 288 PAGE_INLINE void uvm_pageunwire __P((struct vm_page *)); 289 PAGE_INLINE void uvm_pagewait __P((struct vm_page *, int)); 290 PAGE_INLINE void uvm_pagewake __P((struct vm_page *)); 291 PAGE_INLINE void uvm_pagewire __P((struct vm_page *)); 292 PAGE_INLINE void uvm_pagezero __P((struct vm_page *)); 293 294 PAGE_INLINE int uvm_page_lookup_freelist __P((struct vm_page *)); 295 296 static struct vm_page *PHYS_TO_VM_PAGE __P((paddr_t)); 297 static int vm_physseg_find __P((paddr_t, int *)); 298 299 /* 300 * macros 301 */ 302 303 #define UVM_PAGE_HASH_PENALTY 4 /* XXX: a guess */ 304 305 #define uvm_lock_pageq() simple_lock(&uvm.pageqlock) 306 #define uvm_unlock_pageq() simple_unlock(&uvm.pageqlock) 307 308 #define uvm_pagehash(obj,off) \ 309 (((unsigned long)obj+(unsigned long)atop(off)) & uvm.page_hashmask) 310 311 #define UVM_PAGEZERO_TARGET (uvmexp.free) 312 313 #define VM_PAGE_TO_PHYS(entry) ((entry)->phys_addr) 314 315 /* 316 * Compute the page color bucket for a given page. 317 */ 318 #define VM_PGCOLOR_BUCKET(pg) \ 319 (atop(VM_PAGE_TO_PHYS((pg))) & uvmexp.colormask) 320 321 /* 322 * when VM_PHYSSEG_MAX is 1, we can simplify these functions 323 */ 324 325 /* 326 * vm_physseg_find: find vm_physseg structure that belongs to a PA 327 */ 328 static __inline int 329 vm_physseg_find(pframe, offp) 330 paddr_t pframe; 331 int *offp; 332 { 333 #if VM_PHYSSEG_MAX == 1 334 335 /* 'contig' case */ 336 if (pframe >= vm_physmem[0].start && pframe < vm_physmem[0].end) { 337 if (offp) 338 *offp = pframe - vm_physmem[0].start; 339 return(0); 340 } 341 return(-1); 342 343 #elif (VM_PHYSSEG_STRAT == VM_PSTRAT_BSEARCH) 344 /* binary search for it */ 345 int start, len, try; 346 347 /* 348 * if try is too large (thus target is less than try) we reduce 349 * the length to trunc(len/2) [i.e. everything smaller than "try"] 350 * 351 * if the try is too small (thus target is greater than try) then 352 * we set the new start to be (try + 1). this means we need to 353 * reduce the length to (round(len/2) - 1). 354 * 355 * note "adjust" below which takes advantage of the fact that 356 * (round(len/2) - 1) == trunc((len - 1) / 2) 357 * for any value of len we may have 358 */ 359 360 for (start = 0, len = vm_nphysseg ; len != 0 ; len = len / 2) { 361 try = start + (len / 2); /* try in the middle */ 362 363 /* start past our try? */ 364 if (pframe >= vm_physmem[try].start) { 365 /* was try correct? */ 366 if (pframe < vm_physmem[try].end) { 367 if (offp) 368 *offp = pframe - vm_physmem[try].start; 369 return(try); /* got it */ 370 } 371 start = try + 1; /* next time, start here */ 372 len--; /* "adjust" */ 373 } else { 374 /* 375 * pframe before try, just reduce length of 376 * region, done in "for" loop 377 */ 378 } 379 } 380 return(-1); 381 382 #else 383 /* linear search for it */ 384 int lcv; 385 386 for (lcv = 0; lcv < vm_nphysseg; lcv++) { 387 if (pframe >= vm_physmem[lcv].start && 388 pframe < vm_physmem[lcv].end) { 389 if (offp) 390 *offp = pframe - vm_physmem[lcv].start; 391 return(lcv); /* got it */ 392 } 393 } 394 return(-1); 395 396 #endif 397 } 398 399 400 /* 401 * IS_VM_PHYSADDR: only used my mips/pmax/pica trap/pmap. 402 */ 403 404 #define IS_VM_PHYSADDR(PA) (vm_physseg_find(atop(PA), NULL) != -1) 405 406 /* 407 * PHYS_TO_VM_PAGE: find vm_page for a PA. used by MI code to get vm_pages 408 * back from an I/O mapping (ugh!). used in some MD code as well. 409 */ 410 static __inline struct vm_page * 411 PHYS_TO_VM_PAGE(pa) 412 paddr_t pa; 413 { 414 paddr_t pf = atop(pa); 415 int off; 416 int psi; 417 418 psi = vm_physseg_find(pf, &off); 419 if (psi != -1) 420 return(&vm_physmem[psi].pgs[off]); 421 return(NULL); 422 } 423 424 #define VM_PAGE_IS_FREE(entry) ((entry)->pqflags & PQ_FREE) 425 426 #endif /* _KERNEL */ 427 428 #endif /* _UVM_UVM_PAGE_H_ */ 429