1 /* $NetBSD: uvm_page.h,v 1.42 2005/12/24 20:45:10 perry 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 const 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 #define PG_SPECULATIVE 0x0200 /* page has been read speculatively */ 173 174 #define PG_PAGER1 0x1000 /* pager-specific flag */ 175 176 #define PQ_FREE 0x01 /* page is on free list */ 177 #define PQ_INACTIVE 0x02 /* page is in inactive list */ 178 #define PQ_ACTIVE 0x04 /* page is in active list */ 179 #define PQ_ANON 0x10 /* page is part of an anon, rather 180 than an uvm_object */ 181 #define PQ_AOBJ 0x20 /* page is part of an anonymous 182 uvm_object */ 183 #define PQ_SWAPBACKED (PQ_ANON|PQ_AOBJ) 184 185 /* 186 * physical memory layout structure 187 * 188 * MD vmparam.h must #define: 189 * VM_PHYSEG_MAX = max number of physical memory segments we support 190 * (if this is "1" then we revert to a "contig" case) 191 * VM_PHYSSEG_STRAT: memory sort/search options (for VM_PHYSEG_MAX > 1) 192 * - VM_PSTRAT_RANDOM: linear search (random order) 193 * - VM_PSTRAT_BSEARCH: binary search (sorted by address) 194 * - VM_PSTRAT_BIGFIRST: linear search (sorted by largest segment first) 195 * - others? 196 * XXXCDC: eventually we should purge all left-over global variables... 197 */ 198 #define VM_PSTRAT_RANDOM 1 199 #define VM_PSTRAT_BSEARCH 2 200 #define VM_PSTRAT_BIGFIRST 3 201 202 /* 203 * vm_physmemseg: describes one segment of physical memory 204 */ 205 struct vm_physseg { 206 paddr_t start; /* PF# of first page in segment */ 207 paddr_t end; /* (PF# of last page in segment) + 1 */ 208 paddr_t avail_start; /* PF# of first free page in segment */ 209 paddr_t avail_end; /* (PF# of last free page in segment) +1 */ 210 int free_list; /* which free list they belong on */ 211 struct vm_page *pgs; /* vm_page structures (from start) */ 212 struct vm_page *lastpg; /* vm_page structure for end */ 213 #ifdef __HAVE_PMAP_PHYSSEG 214 struct pmap_physseg pmseg; /* pmap specific (MD) data */ 215 #endif 216 }; 217 218 #ifdef _KERNEL 219 220 /* 221 * globals 222 */ 223 224 extern boolean_t vm_page_zero_enable; 225 226 /* 227 * physical memory config is stored in vm_physmem. 228 */ 229 230 extern struct vm_physseg vm_physmem[VM_PHYSSEG_MAX]; 231 extern int vm_nphysseg; 232 233 /* 234 * handle inline options 235 */ 236 237 #ifdef UVM_PAGE_INLINE 238 #define PAGE_INLINE static inline 239 #else 240 #define PAGE_INLINE /* nothing */ 241 #endif /* UVM_PAGE_INLINE */ 242 243 /* 244 * prototypes: the following prototypes define the interface to pages 245 */ 246 247 void uvm_page_init(vaddr_t *, vaddr_t *); 248 #if defined(UVM_PAGE_TRKOWN) 249 void uvm_page_own(struct vm_page *, const char *); 250 #endif 251 #if !defined(PMAP_STEAL_MEMORY) 252 boolean_t uvm_page_physget(paddr_t *); 253 #endif 254 void uvm_page_rehash(void); 255 void uvm_page_recolor(int); 256 void uvm_pageidlezero(void); 257 258 PAGE_INLINE int uvm_lock_fpageq(void); 259 PAGE_INLINE void uvm_unlock_fpageq(int); 260 261 PAGE_INLINE void uvm_pageactivate(struct vm_page *); 262 vaddr_t uvm_pageboot_alloc(vsize_t); 263 PAGE_INLINE void uvm_pagecopy(struct vm_page *, struct vm_page *); 264 PAGE_INLINE void uvm_pagedeactivate(struct vm_page *); 265 PAGE_INLINE void uvm_pagedequeue(struct vm_page *); 266 void uvm_pagefree(struct vm_page *); 267 void uvm_page_unbusy(struct vm_page **, int); 268 PAGE_INLINE struct vm_page *uvm_pagelookup(struct uvm_object *, voff_t); 269 PAGE_INLINE void uvm_pageunwire(struct vm_page *); 270 PAGE_INLINE void uvm_pagewait(struct vm_page *, int); 271 PAGE_INLINE void uvm_pagewake(struct vm_page *); 272 PAGE_INLINE void uvm_pagewire(struct vm_page *); 273 PAGE_INLINE void uvm_pagezero(struct vm_page *); 274 275 PAGE_INLINE int uvm_page_lookup_freelist(struct vm_page *); 276 277 static struct vm_page *PHYS_TO_VM_PAGE(paddr_t); 278 static int vm_physseg_find(paddr_t, int *); 279 280 /* 281 * macros 282 */ 283 284 #define UVM_PAGE_HASH_PENALTY 4 /* XXX: a guess */ 285 286 #define uvm_lock_pageq() simple_lock(&uvm.pageqlock) 287 #define uvm_unlock_pageq() simple_unlock(&uvm.pageqlock) 288 #define UVM_LOCK_ASSERT_PAGEQ() LOCK_ASSERT(simple_lock_held(&uvm.pageqlock)) 289 290 #define uvm_pagehash(obj,off) \ 291 (((unsigned long)obj+(unsigned long)atop(off)) & uvm.page_hashmask) 292 293 #define UVM_PAGEZERO_TARGET (uvmexp.free) 294 295 #define VM_PAGE_TO_PHYS(entry) ((entry)->phys_addr) 296 297 /* 298 * Compute the page color bucket for a given page. 299 */ 300 #define VM_PGCOLOR_BUCKET(pg) \ 301 (atop(VM_PAGE_TO_PHYS((pg))) & uvmexp.colormask) 302 303 /* 304 * when VM_PHYSSEG_MAX is 1, we can simplify these functions 305 */ 306 307 /* 308 * vm_physseg_find: find vm_physseg structure that belongs to a PA 309 */ 310 static inline int 311 vm_physseg_find(pframe, offp) 312 paddr_t pframe; 313 int *offp; 314 { 315 #if VM_PHYSSEG_MAX == 1 316 317 /* 'contig' case */ 318 if (pframe >= vm_physmem[0].start && pframe < vm_physmem[0].end) { 319 if (offp) 320 *offp = pframe - vm_physmem[0].start; 321 return(0); 322 } 323 return(-1); 324 325 #elif (VM_PHYSSEG_STRAT == VM_PSTRAT_BSEARCH) 326 /* binary search for it */ 327 u_int start, len, try; 328 329 /* 330 * if try is too large (thus target is less than try) we reduce 331 * the length to trunc(len/2) [i.e. everything smaller than "try"] 332 * 333 * if the try is too small (thus target is greater than try) then 334 * we set the new start to be (try + 1). this means we need to 335 * reduce the length to (round(len/2) - 1). 336 * 337 * note "adjust" below which takes advantage of the fact that 338 * (round(len/2) - 1) == trunc((len - 1) / 2) 339 * for any value of len we may have 340 */ 341 342 for (start = 0, len = vm_nphysseg ; len != 0 ; len = len / 2) { 343 try = start + (len / 2); /* try in the middle */ 344 345 /* start past our try? */ 346 if (pframe >= vm_physmem[try].start) { 347 /* was try correct? */ 348 if (pframe < vm_physmem[try].end) { 349 if (offp) 350 *offp = pframe - vm_physmem[try].start; 351 return(try); /* got it */ 352 } 353 start = try + 1; /* next time, start here */ 354 len--; /* "adjust" */ 355 } else { 356 /* 357 * pframe before try, just reduce length of 358 * region, done in "for" loop 359 */ 360 } 361 } 362 return(-1); 363 364 #else 365 /* linear search for it */ 366 int lcv; 367 368 for (lcv = 0; lcv < vm_nphysseg; lcv++) { 369 if (pframe >= vm_physmem[lcv].start && 370 pframe < vm_physmem[lcv].end) { 371 if (offp) 372 *offp = pframe - vm_physmem[lcv].start; 373 return(lcv); /* got it */ 374 } 375 } 376 return(-1); 377 378 #endif 379 } 380 381 382 /* 383 * IS_VM_PHYSADDR: only used my mips/pmax/pica trap/pmap. 384 */ 385 386 #define IS_VM_PHYSADDR(PA) (vm_physseg_find(atop(PA), NULL) != -1) 387 388 /* 389 * PHYS_TO_VM_PAGE: find vm_page for a PA. used by MI code to get vm_pages 390 * back from an I/O mapping (ugh!). used in some MD code as well. 391 */ 392 static inline struct vm_page * 393 PHYS_TO_VM_PAGE(pa) 394 paddr_t pa; 395 { 396 paddr_t pf = atop(pa); 397 int off; 398 int psi; 399 400 psi = vm_physseg_find(pf, &off); 401 if (psi != -1) 402 return(&vm_physmem[psi].pgs[off]); 403 return(NULL); 404 } 405 406 #define VM_PAGE_IS_FREE(entry) ((entry)->pqflags & PQ_FREE) 407 408 #ifdef DEBUG 409 void uvm_pagezerocheck(struct vm_page *); 410 #endif /* DEBUG */ 411 412 #endif /* _KERNEL */ 413 414 #endif /* _UVM_UVM_PAGE_H_ */ 415