1 /* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or http://www.opensolaris.org/os/licensing. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 */ 21 /* 22 * Copyright 2007 Sun Microsystems, Inc. All rights reserved. 23 * Use is subject to license terms. 24 */ 25 26 /* Copyright (c) 1984, 1986, 1987, 1988, 1989 AT&T */ 27 /* All Rights Reserved */ 28 29 /* 30 * University Copyright- Copyright (c) 1982, 1986, 1988 31 * The Regents of the University of California 32 * All Rights Reserved 33 * 34 * University Acknowledgment- Portions of this document are derived from 35 * software developed by the University of California, Berkeley, and its 36 * contributors. 37 */ 38 39 #ifndef _VM_ANON_H 40 #define _VM_ANON_H 41 42 #pragma ident "%Z%%M% %I% %E% SMI" 43 44 #include <sys/cred.h> 45 #include <sys/zone.h> 46 #include <vm/seg.h> 47 #include <vm/vpage.h> 48 49 #ifdef __cplusplus 50 extern "C" { 51 #endif 52 53 /* 54 * VM - Anonymous pages. 55 */ 56 57 typedef unsigned long anoff_t; /* anon offsets */ 58 59 /* 60 * Each anonymous page, either in memory or in swap, has an anon structure. 61 * The structure (slot) provides a level of indirection between anonymous pages 62 * and their backing store. 63 * 64 * (an_vp, an_off) names the vnode of the anonymous page for this slot. 65 * 66 * (an_pvp, an_poff) names the location of the physical backing store 67 * for the page this slot represents. If the name is null there is no 68 * associated physical store. The physical backing store location can 69 * change while the slot is in use. 70 * 71 * an_hash is a hash list of anon slots. The list is hashed by 72 * (an_vp, an_off) of the associated anonymous page and provides a 73 * method of going from the name of an anonymous page to its 74 * associated anon slot. 75 * 76 * an_refcnt holds a reference count which is the number of separate 77 * copies that will need to be created in case of copy-on-write. 78 * A refcnt > 0 protects the existence of the slot. The refcnt is 79 * initialized to 1 when the anon slot is created in anon_alloc(). 80 * If a client obtains an anon slot and allows multiple threads to 81 * share it, then it is the client's responsibility to insure that 82 * it does not allow one thread to try to reference the slot at the 83 * same time as another is trying to decrement the last count and 84 * destroy the anon slot. E.g., the seg_vn segment type protects 85 * against this with higher level locks. 86 */ 87 88 struct anon { 89 struct vnode *an_vp; /* vnode of anon page */ 90 struct vnode *an_pvp; /* vnode of physical backing store */ 91 anoff_t an_off; /* offset of anon page */ 92 anoff_t an_poff; /* offset in vnode */ 93 struct anon *an_hash; /* hash table of anon slots */ 94 int an_refcnt; /* # of people sharing slot */ 95 }; 96 97 #ifdef _KERNEL 98 /* 99 * The swapinfo_lock protects: 100 * swapinfo list 101 * individual swapinfo structures 102 * 103 * The anoninfo_lock protects: 104 * anoninfo counters 105 * 106 * The anonhash_lock protects: 107 * anon hash lists 108 * anon slot fields 109 * 110 * Fields in the anon slot which are read-only for the life of the slot 111 * (an_vp, an_off) do not require the anonhash_lock be held to access them. 112 * If you access a field without the anonhash_lock held you must be holding 113 * the slot with an_refcnt to make sure it isn't destroyed. 114 * To write (an_pvp, an_poff) in a given slot you must also hold the 115 * p_iolock of the anonymous page for slot. 116 */ 117 extern kmutex_t anoninfo_lock; 118 extern kmutex_t swapinfo_lock; 119 extern kmutex_t anonhash_lock[]; 120 extern pad_mutex_t anon_array_lock[]; 121 extern kcondvar_t anon_array_cv[]; 122 123 /* 124 * Global hash table to provide a function from (vp, off) -> ap 125 */ 126 extern size_t anon_hash_size; 127 extern struct anon **anon_hash; 128 #define ANON_HASH_SIZE anon_hash_size 129 #define ANON_HASHAVELEN 4 130 #define ANON_HASH(VP, OFF) \ 131 ((((uintptr_t)(VP) >> 7) ^ ((OFF) >> PAGESHIFT)) & (ANON_HASH_SIZE - 1)) 132 133 #define AH_LOCK_SIZE 64 134 #define AH_LOCK(vp, off) (ANON_HASH((vp), (off)) & (AH_LOCK_SIZE -1)) 135 136 #endif /* _KERNEL */ 137 138 /* 139 * Declaration for the Global counters to accurately 140 * track the kernel foot print in memory. 141 */ 142 extern pgcnt_t segvn_pages_locked; 143 extern pgcnt_t pages_locked; 144 extern pgcnt_t pages_claimed; 145 extern pgcnt_t pages_useclaim; 146 extern pgcnt_t obp_pages; 147 148 /* 149 * Anonymous backing store accounting structure for swapctl. 150 * 151 * ani_max = maximum amount of swap space 152 * (including potentially available physical memory) 153 * ani_free = amount of unallocated anonymous memory 154 * (some of which might be reserved and including 155 * potentially available physical memory) 156 * ani_resv = amount of claimed (reserved) anonymous memory 157 * 158 * The swap data can be aquired more efficiently through the 159 * kstats interface. 160 * Total slots currently available for reservation = 161 * MAX(ani_max - ani_resv, 0) + (availrmem - swapfs_minfree) 162 */ 163 struct anoninfo { 164 pgcnt_t ani_max; 165 pgcnt_t ani_free; 166 pgcnt_t ani_resv; 167 }; 168 169 #ifdef _SYSCALL32 170 struct anoninfo32 { 171 size32_t ani_max; 172 size32_t ani_free; 173 size32_t ani_resv; 174 }; 175 #endif /* _SYSCALL32 */ 176 177 /* 178 * Define the NCPU pool of the ani_free counters. Update the counter 179 * of the cpu on which the thread is running and in every clock intr 180 * sync anoninfo.ani_free with the current total off all the NCPU entries. 181 */ 182 183 typedef struct ani_free { 184 kmutex_t ani_lock; 185 pgcnt_t ani_count; 186 uchar_t pad[64 - sizeof (kmutex_t) - sizeof (pgcnt_t)]; 187 /* XXX 64 = cacheline size */ 188 } ani_free_t; 189 190 #define ANI_MAX_POOL 128 191 extern ani_free_t ani_free_pool[]; 192 193 #define ANI_ADD(inc) { \ 194 ani_free_t *anifp; \ 195 int index; \ 196 index = (CPU->cpu_id & (ANI_MAX_POOL - 1)); \ 197 anifp = &ani_free_pool[index]; \ 198 mutex_enter(&anifp->ani_lock); \ 199 anifp->ani_count += inc; \ 200 mutex_exit(&anifp->ani_lock); \ 201 } 202 203 /* 204 * Anon array pointers are allocated in chunks. Each chunk 205 * has PAGESIZE/sizeof(u_long *) of anon pointers. 206 * There are two levels of arrays for anon array pointers larger 207 * than a chunk. The first level points to anon array chunks. 208 * The second level consists of chunks of anon pointers. 209 * 210 * If anon array is smaller than a chunk then the whole anon array 211 * is created (memory is allocated for whole anon array). 212 * If anon array is larger than a chunk only first level array is 213 * allocated. Then other arrays (chunks) are allocated only when 214 * they are initialized with anon pointers. 215 */ 216 struct anon_hdr { 217 kmutex_t serial_lock; /* serialize array chunk allocation */ 218 pgcnt_t size; /* number of pointers to (anon) pages */ 219 void **array_chunk; /* pointers to anon pointers or chunks of */ 220 /* anon pointers */ 221 int flags; /* ANON_ALLOC_FORCE force preallocation of */ 222 /* whole anon array */ 223 }; 224 225 #ifdef _LP64 226 #define ANON_PTRSHIFT 3 227 #define ANON_PTRMASK ~7 228 #else 229 #define ANON_PTRSHIFT 2 230 #define ANON_PTRMASK ~3 231 #endif 232 233 #define ANON_CHUNK_SIZE (PAGESIZE >> ANON_PTRSHIFT) 234 #define ANON_CHUNK_SHIFT (PAGESHIFT - ANON_PTRSHIFT) 235 #define ANON_CHUNK_OFF (ANON_CHUNK_SIZE - 1) 236 237 /* 238 * Anon flags. 239 */ 240 #define ANON_SLEEP 0x0 /* ok to block */ 241 #define ANON_NOSLEEP 0x1 /* non-blocking call */ 242 #define ANON_ALLOC_FORCE 0x2 /* force single level anon array */ 243 #define ANON_GROWDOWN 0x4 /* anon array should grow downward */ 244 245 struct kshmid; 246 247 /* 248 * The anon_map structure is used by various clients of the anon layer to 249 * manage anonymous memory. When anonymous memory is shared, 250 * then the different clients sharing it will point to the 251 * same anon_map structure. Also, if a segment is unmapped 252 * in the middle where an anon_map structure exists, the 253 * newly created segment will also share the anon_map structure, 254 * although the two segments will use different ranges of the 255 * anon array. When mappings are private (or shared with 256 * a reference count of 1), an unmap operation will free up 257 * a range of anon slots in the array given by the anon_map 258 * structure. Because of fragmentation due to this unmapping, 259 * we have to store the size of the anon array in the anon_map 260 * structure so that we can free everything when the referernce 261 * count goes to zero. 262 * 263 * A new rangelock scheme is introduced to make the anon layer scale. 264 * A reader/writer lock per anon_amp and an array of system-wide hash 265 * locks, anon_array_lock[] are introduced to replace serial_lock and 266 * anonmap lock. The writer lock is held when we want to singlethreaD 267 * the reference to the anon array pointers or when references to 268 * anon_map's members, whereas reader lock and anon_array_lock are 269 * held to allows multiple threads to reference different part of 270 * anon array. A global set of condition variables, anon_array_cv, 271 * are used with anon_array_lock[] to make the hold time of the locks 272 * short. 273 * 274 * szc is used to calculate the index of hash locks and cv's. We 275 * could've just used seg->s_szc if not for the possible sharing of 276 * anon_amp between SYSV shared memory and ISM, so now we introduce 277 * szc in the anon_map structure. For MAP_SHARED, the amp->szc is either 278 * 0 (base page size) or page_num_pagesizes() - 1, while MAP_PRIVATE 279 * the amp->szc could be anything in [0, page_num_pagesizes() - 1]. 280 */ 281 struct anon_map { 282 krwlock_t a_rwlock; /* protect anon_map and anon array */ 283 size_t size; /* size in bytes mapped by the anon array */ 284 struct anon_hdr *ahp; /* anon array header pointer, containing */ 285 /* anon pointer array(s) */ 286 size_t swresv; /* swap space reserved for this anon_map */ 287 ulong_t refcnt; /* reference count on this structure */ 288 ushort_t a_szc; /* max szc among shared processes */ 289 void *locality; /* lgroup locality info */ 290 struct kshmid *a_sp; /* kshmid if amp backs sysV, or NULL */ 291 }; 292 293 #ifdef _KERNEL 294 295 #define ANON_BUSY 0x1 296 #define ANON_ISBUSY(slot) (*(slot) & ANON_BUSY) 297 #define ANON_SETBUSY(slot) (*(slot) |= ANON_BUSY) 298 #define ANON_CLRBUSY(slot) (*(slot) &= ~ANON_BUSY) 299 300 #define ANON_MAP_SHIFT 6 /* log2(sizeof (struct anon_map)) */ 301 #define ANON_ARRAY_SHIFT 7 /* log2(ANON_LOCKSIZE) */ 302 #define ANON_LOCKSIZE 128 303 304 #define ANON_LOCK_ENTER(lock, type) rw_enter((lock), (type)) 305 #define ANON_LOCK_EXIT(lock) rw_exit((lock)) 306 307 #define ANON_ARRAY_HASH(amp, idx)\ 308 ((((idx) + ((idx) >> ANON_ARRAY_SHIFT) +\ 309 ((idx) >> (ANON_ARRAY_SHIFT << 1)) +\ 310 ((idx) >> (ANON_ARRAY_SHIFT + (ANON_ARRAY_SHIFT << 1)))) ^\ 311 ((uintptr_t)(amp) >> ANON_MAP_SHIFT)) & (ANON_LOCKSIZE - 1)) 312 313 typedef struct anon_sync_obj { 314 kmutex_t *sync_mutex; 315 kcondvar_t *sync_cv; 316 ulong_t *sync_data; 317 } anon_sync_obj_t; 318 319 /* 320 * Anonymous backing store accounting structure for kernel. 321 * ani_max = total reservable slots on physical (disk-backed) swap 322 * ani_phys_resv = total phys slots reserved for use by clients 323 * ani_mem_resv = total mem slots reserved for use by clients 324 * ani_free = # unallocated physical slots + # of reserved unallocated 325 * memory slots 326 */ 327 328 /* 329 * Initial total swap slots available for reservation 330 */ 331 #define TOTAL_AVAILABLE_SWAP \ 332 (k_anoninfo.ani_max + MAX((spgcnt_t)(availrmem - swapfs_minfree), 0)) 333 334 /* 335 * Swap slots currently available for reservation 336 */ 337 #define CURRENT_TOTAL_AVAILABLE_SWAP \ 338 ((k_anoninfo.ani_max - k_anoninfo.ani_phys_resv) + \ 339 MAX((spgcnt_t)(availrmem - swapfs_minfree), 0)) 340 341 struct k_anoninfo { 342 pgcnt_t ani_max; /* total reservable slots on phys */ 343 /* (disk) swap */ 344 pgcnt_t ani_free; /* # of unallocated phys and mem slots */ 345 pgcnt_t ani_phys_resv; /* # of reserved phys (disk) slots */ 346 pgcnt_t ani_mem_resv; /* # of reserved mem slots */ 347 pgcnt_t ani_locked_swap; /* # of swap slots locked in reserved */ 348 /* mem swap */ 349 }; 350 351 extern struct k_anoninfo k_anoninfo; 352 353 extern void anon_init(void); 354 extern struct anon *anon_alloc(struct vnode *, anoff_t); 355 extern void anon_dup(struct anon_hdr *, ulong_t, 356 struct anon_hdr *, ulong_t, size_t); 357 extern void anon_dup_fill_holes(struct anon_hdr *, ulong_t, 358 struct anon_hdr *, ulong_t, size_t, uint_t, int); 359 extern int anon_fill_cow_holes(struct seg *, caddr_t, struct anon_hdr *, 360 ulong_t, struct vnode *, u_offset_t, size_t, uint_t, 361 uint_t, struct vpage [], struct cred *); 362 extern void anon_free(struct anon_hdr *, ulong_t, size_t); 363 extern void anon_free_pages(struct anon_hdr *, ulong_t, size_t, uint_t); 364 extern void anon_disclaim(struct anon_map *, ulong_t, size_t, int); 365 extern int anon_getpage(struct anon **, uint_t *, struct page **, 366 size_t, struct seg *, caddr_t, enum seg_rw, struct cred *); 367 extern int swap_getconpage(struct vnode *, u_offset_t, size_t, 368 uint_t *, page_t *[], size_t, page_t *, uint_t *, 369 spgcnt_t *, struct seg *, caddr_t, 370 enum seg_rw, struct cred *); 371 extern int anon_map_getpages(struct anon_map *, ulong_t, 372 uint_t, struct seg *, caddr_t, uint_t, 373 uint_t *, page_t *[], uint_t *, 374 struct vpage [], enum seg_rw, int, int, int, struct cred *); 375 extern int anon_map_privatepages(struct anon_map *, ulong_t, 376 uint_t, struct seg *, caddr_t, uint_t, 377 page_t *[], struct vpage [], int, int, struct cred *); 378 extern struct page *anon_private(struct anon **, struct seg *, 379 caddr_t, uint_t, struct page *, 380 int, struct cred *); 381 extern struct page *anon_zero(struct seg *, caddr_t, 382 struct anon **, struct cred *); 383 extern int anon_map_createpages(struct anon_map *, ulong_t, 384 size_t, struct page **, 385 struct seg *, caddr_t, 386 enum seg_rw, struct cred *); 387 extern int anon_map_demotepages(struct anon_map *, ulong_t, 388 struct seg *, caddr_t, uint_t, 389 struct vpage [], struct cred *); 390 extern void anon_shmap_free_pages(struct anon_map *, ulong_t, size_t); 391 extern int anon_resvmem(size_t, boolean_t, zone_t *, int); 392 extern void anon_unresvmem(size_t, zone_t *); 393 extern struct anon_map *anonmap_alloc(size_t, size_t, int); 394 extern void anonmap_free(struct anon_map *); 395 extern void anon_decref(struct anon *); 396 extern int non_anon(struct anon_hdr *, ulong_t, u_offset_t *, size_t *); 397 extern pgcnt_t anon_pages(struct anon_hdr *, ulong_t, pgcnt_t); 398 extern int anon_swap_adjust(pgcnt_t); 399 extern void anon_swap_restore(pgcnt_t); 400 extern struct anon_hdr *anon_create(pgcnt_t, int); 401 extern void anon_release(struct anon_hdr *, pgcnt_t); 402 extern struct anon *anon_get_ptr(struct anon_hdr *, ulong_t); 403 extern ulong_t *anon_get_slot(struct anon_hdr *, ulong_t); 404 extern struct anon *anon_get_next_ptr(struct anon_hdr *, ulong_t *); 405 extern int anon_set_ptr(struct anon_hdr *, ulong_t, struct anon *, int); 406 extern int anon_copy_ptr(struct anon_hdr *, ulong_t, 407 struct anon_hdr *, ulong_t, pgcnt_t, int); 408 extern pgcnt_t anon_grow(struct anon_hdr *, ulong_t *, pgcnt_t, pgcnt_t, int); 409 extern void anon_array_enter(struct anon_map *, ulong_t, 410 anon_sync_obj_t *); 411 extern int anon_array_try_enter(struct anon_map *, ulong_t, 412 anon_sync_obj_t *); 413 extern void anon_array_exit(anon_sync_obj_t *); 414 415 /* 416 * anon_resv checks to see if there is enough swap space to fulfill a 417 * request and if so, reserves the appropriate anonymous memory resources. 418 * anon_checkspace just checks to see if there is space to fulfill the request, 419 * without taking any resources. Both return 1 if successful and 0 if not. 420 * 421 * Macros are provided as anon reservation is usually charged to the zone of 422 * the current process. In some cases (such as anon reserved by tmpfs), a 423 * zone pointer is needed to charge the appropriate zone. 424 */ 425 #define anon_unresv(size) anon_unresvmem(size, curproc->p_zone) 426 #define anon_unresv_zone(size, zone) anon_unresvmem(size, zone) 427 #define anon_resv(size) \ 428 anon_resvmem((size), 1, curproc->p_zone, 1) 429 #define anon_resv_zone(size, zone) anon_resvmem((size), 1, zone, 1) 430 #define anon_checkspace(size, zone) anon_resvmem((size), 0, zone, 0) 431 #define anon_try_resv_zone(size, zone) anon_resvmem((size), 1, zone, 0) 432 433 /* 434 * Flags to anon_private 435 */ 436 #define STEAL_PAGE 0x1 /* page can be stolen */ 437 #define LOCK_PAGE 0x2 /* page must be ``logically'' locked */ 438 439 /* 440 * Flags to anon_disclaim 441 */ 442 #define ANON_PGLOOKUP_BLK 0x1 /* block on locked pages */ 443 444 /* 445 * SEGKP ANON pages that are locked are assumed to be LWP stack pages 446 * and thus count towards the user pages locked count. 447 * This value is protected by the same lock as availrmem. 448 */ 449 extern pgcnt_t anon_segkp_pages_locked; 450 451 extern int anon_debug; 452 453 #ifdef ANON_DEBUG 454 455 #define A_ANON 0x01 456 #define A_RESV 0x02 457 #define A_MRESV 0x04 458 459 /* vararg-like debugging macro. */ 460 #define ANON_PRINT(f, printf_args) \ 461 if (anon_debug & f) \ 462 printf printf_args 463 464 #else /* ANON_DEBUG */ 465 466 #define ANON_PRINT(f, printf_args) 467 468 #endif /* ANON_DEBUG */ 469 470 #endif /* _KERNEL */ 471 472 #ifdef __cplusplus 473 } 474 #endif 475 476 #endif /* _VM_ANON_H */ 477