Print this page
vm: be careful about empty statements
Split |
Close |
Expand all |
Collapse all |
--- old/usr/src/uts/common/vm/page.h
+++ new/usr/src/uts/common/vm/page.h
1 1 /*
2 2 * CDDL HEADER START
3 3 *
4 4 * The contents of this file are subject to the terms of the
5 5 * Common Development and Distribution License (the "License").
6 6 * You may not use this file except in compliance with the License.
7 7 *
8 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 9 * or http://www.opensolaris.org/os/licensing.
10 10 * See the License for the specific language governing permissions
11 11 * and limitations under the License.
12 12 *
13 13 * When distributing Covered Code, include this CDDL HEADER in each
14 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 15 * If applicable, add the following below this CDDL HEADER, with the
16 16 * fields enclosed by brackets "[]" replaced with your own identifying
17 17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 18 *
19 19 * CDDL HEADER END
20 20 */
21 21 /*
22 22 * Copyright (c) 1986, 2010, Oracle and/or its affiliates. All rights reserved.
23 23 */
24 24
25 25 /* Copyright (c) 1984, 1986, 1987, 1988, 1989 AT&T */
26 26 /* All Rights Reserved */
27 27
28 28 /*
29 29 * University Copyright- Copyright (c) 1982, 1986, 1988
30 30 * The Regents of the University of California
31 31 * All Rights Reserved
32 32 *
33 33 * University Acknowledgment- Portions of this document are derived from
34 34 * software developed by the University of California, Berkeley, and its
35 35 * contributors.
36 36 */
37 37
38 38 #ifndef _VM_PAGE_H
39 39 #define _VM_PAGE_H
40 40
41 41 #include <vm/seg.h>
42 42
43 43 #ifdef __cplusplus
44 44 extern "C" {
45 45 #endif
46 46
47 47 #if defined(_KERNEL) || defined(_KMEMUSER)
48 48
49 49 /*
50 50 * Shared/Exclusive lock.
51 51 */
52 52
53 53 /*
54 54 * Types of page locking supported by page_lock & friends.
55 55 */
56 56 typedef enum {
57 57 SE_SHARED,
58 58 SE_EXCL /* exclusive lock (value == -1) */
59 59 } se_t;
60 60
61 61 /*
62 62 * For requesting that page_lock reclaim the page from the free list.
63 63 */
64 64 typedef enum {
65 65 P_RECLAIM, /* reclaim page from free list */
66 66 P_NO_RECLAIM /* DON`T reclaim the page */
67 67 } reclaim_t;
68 68
69 69 /*
70 70 * Callers of page_try_reclaim_lock and page_lock_es can use this flag
71 71 * to get SE_EXCL access before reader/writers are given access.
72 72 */
73 73 #define SE_EXCL_WANTED 0x02
74 74
75 75 /*
76 76 * All page_*lock() requests will be denied unless this flag is set in
77 77 * the 'es' parameter.
78 78 */
79 79 #define SE_RETIRED 0x04
80 80
81 81 #endif /* _KERNEL | _KMEMUSER */
82 82
83 83 typedef int selock_t;
84 84
85 85 /*
86 86 * Define VM_STATS to turn on all sorts of statistic gathering about
87 87 * the VM layer. By default, it is only turned on when DEBUG is
↓ open down ↓ |
87 lines elided |
↑ open up ↑ |
88 88 * also defined.
89 89 */
90 90 #ifdef DEBUG
91 91 #define VM_STATS
92 92 #endif /* DEBUG */
93 93
94 94 #ifdef VM_STATS
95 95 #define VM_STAT_ADD(stat) (stat)++
96 96 #define VM_STAT_COND_ADD(cond, stat) ((void) (!(cond) || (stat)++))
97 97 #else
98 -#define VM_STAT_ADD(stat)
99 -#define VM_STAT_COND_ADD(cond, stat)
98 +#define VM_STAT_ADD(stat) do { } while (0)
99 +#define VM_STAT_COND_ADD(cond, stat) do { } while (0)
100 100 #endif /* VM_STATS */
101 101
102 102 #ifdef _KERNEL
103 103
104 104 /*
105 105 * PAGE_LLOCK_SIZE is 2 * NCPU, but no smaller than 128.
106 106 * PAGE_LLOCK_SHIFT is log2(PAGE_LLOCK_SIZE).
107 107 *
108 108 * We use ? : instead of #if because <vm/page.h> is included everywhere;
109 109 * NCPU_P2 is only a constant in the "unix" module.
110 110 *
111 111 */
112 112 #define PAGE_LLOCK_SHIFT \
113 113 ((unsigned)(((2*NCPU_P2) > 128) ? NCPU_LOG2 + 1 : 7))
114 114
115 115 #define PAGE_LLOCK_SIZE (1ul << PAGE_LLOCK_SHIFT)
116 116
117 117 /*
118 118 * The number of low order 0 (or less variable) bits in the page_t address.
119 119 */
120 120 #if defined(__sparc)
121 121 #define PP_SHIFT 7
122 122 #else
123 123 #define PP_SHIFT 6
124 124 #endif
125 125
126 126 /*
127 127 * pp may be the root of a large page, and many low order bits will be 0.
128 128 * Shift and XOR multiple times to capture the good bits across the range of
129 129 * possible page sizes.
130 130 */
131 131 #define PAGE_LLOCK_HASH(pp) \
132 132 (((((uintptr_t)(pp) >> PP_SHIFT) ^ \
133 133 ((uintptr_t)(pp) >> (PAGE_LLOCK_SHIFT + PP_SHIFT))) ^ \
134 134 ((uintptr_t)(pp) >> ((PAGE_LLOCK_SHIFT * 2) + PP_SHIFT)) ^ \
135 135 ((uintptr_t)(pp) >> ((PAGE_LLOCK_SHIFT * 3) + PP_SHIFT))) & \
136 136 (PAGE_LLOCK_SIZE - 1))
137 137
138 138 #define page_struct_lock(pp) \
139 139 mutex_enter(&page_llocks[PAGE_LLOCK_HASH(PP_PAGEROOT(pp))].pad_mutex)
140 140 #define page_struct_unlock(pp) \
141 141 mutex_exit(&page_llocks[PAGE_LLOCK_HASH(PP_PAGEROOT(pp))].pad_mutex)
142 142
143 143 #endif /* _KERNEL */
144 144
145 145 #include <sys/t_lock.h>
146 146
147 147 struct as;
148 148
149 149 /*
150 150 * Each physical page has a page structure, which is used to maintain
151 151 * these pages as a cache. A page can be found via a hashed lookup
152 152 * based on the [vp, offset]. If a page has an [vp, offset] identity,
153 153 * then it is entered on a doubly linked circular list off the
154 154 * vnode using the vpnext/vpprev pointers. If the p_free bit
155 155 * is on, then the page is also on a doubly linked circular free
156 156 * list using next/prev pointers. If the "p_selock" and "p_iolock"
157 157 * are held, then the page is currently being read in (exclusive p_selock)
158 158 * or written back (shared p_selock). In this case, the next/prev pointers
159 159 * are used to link the pages together for a consecutive i/o request. If
160 160 * the page is being brought in from its backing store, then other processes
161 161 * will wait for the i/o to complete before attaching to the page since it
162 162 * will have an "exclusive" lock.
163 163 *
164 164 * Each page structure has the locks described below along with
165 165 * the fields they protect:
166 166 *
167 167 * p_selock This is a per-page shared/exclusive lock that is
168 168 * used to implement the logical shared/exclusive
169 169 * lock for each page. The "shared" lock is normally
170 170 * used in most cases while the "exclusive" lock is
171 171 * required to destroy or retain exclusive access to
172 172 * a page (e.g., while reading in pages). The appropriate
173 173 * lock is always held whenever there is any reference
174 174 * to a page structure (e.g., during i/o).
175 175 * (Note that with the addition of the "writer-lock-wanted"
176 176 * semantics (via SE_EWANTED), threads must not acquire
177 177 * multiple reader locks or else a deadly embrace will
178 178 * occur in the following situation: thread 1 obtains a
179 179 * reader lock; next thread 2 fails to get a writer lock
180 180 * but specified SE_EWANTED so it will wait by either
181 181 * blocking (when using page_lock_es) or spinning while
182 182 * retrying (when using page_try_reclaim_lock) until the
183 183 * reader lock is released; then thread 1 attempts to
184 184 * get another reader lock but is denied due to
185 185 * SE_EWANTED being set, and now both threads are in a
186 186 * deadly embrace.)
187 187 *
188 188 * p_hash
189 189 * p_vnode
190 190 * p_offset
191 191 *
192 192 * p_free
193 193 * p_age
194 194 *
195 195 * p_iolock This is a binary semaphore lock that provides
196 196 * exclusive access to the i/o list links in each
197 197 * page structure. It is always held while the page
198 198 * is on an i/o list (i.e., involved in i/o). That is,
199 199 * even though a page may be only `shared' locked
200 200 * while it is doing a write, the following fields may
201 201 * change anyway. Normally, the page must be
202 202 * `exclusively' locked to change anything in it.
203 203 *
204 204 * p_next
205 205 * p_prev
206 206 *
207 207 * The following fields are protected by the global page_llocks[]:
208 208 *
209 209 * p_lckcnt
210 210 * p_cowcnt
211 211 *
212 212 * The following lists are protected by the global page_freelock:
213 213 *
214 214 * page_cachelist
215 215 * page_freelist
216 216 *
217 217 * The following, for our purposes, are protected by
218 218 * the global freemem_lock:
219 219 *
220 220 * freemem
221 221 * freemem_wait
222 222 * freemem_cv
223 223 *
224 224 * The following fields are protected by hat layer lock(s). When a page
225 225 * structure is not mapped and is not associated with a vnode (after a call
226 226 * to page_hashout() for example) the p_nrm field may be modified with out
227 227 * holding the hat layer lock:
228 228 *
229 229 * p_nrm
230 230 * p_mapping
231 231 * p_share
232 232 *
233 233 * The following field is file system dependent. How it is used and
234 234 * the locking strategies applied are up to the individual file system
235 235 * implementation.
236 236 *
237 237 * p_fsdata
238 238 *
239 239 * The page structure is used to represent and control the system's
240 240 * physical pages. There is one instance of the structure for each
241 241 * page that is not permenately allocated. For example, the pages that
242 242 * hold the page structures are permanently held by the kernel
243 243 * and hence do not need page structures to track them. The array
244 244 * of page structures is allocated early on in the kernel's life and
245 245 * is based on the amount of available physical memory.
246 246 *
247 247 * Each page structure may simultaneously appear on several linked lists.
248 248 * The lists are: hash list, free or in i/o list, and a vnode's page list.
249 249 * Each type of list is protected by a different group of mutexes as described
250 250 * below:
251 251 *
252 252 * The hash list is used to quickly find a page when the page's vnode and
253 253 * offset within the vnode are known. Each page that is hashed is
254 254 * connected via the `p_hash' field. The anchor for each hash is in the
255 255 * array `page_hash'. An array of mutexes, `ph_mutex', protects the
256 256 * lists anchored by page_hash[]. To either search or modify a given hash
257 257 * list, the appropriate mutex in the ph_mutex array must be held.
258 258 *
259 259 * The free list contains pages that are `free to be given away'. For
260 260 * efficiency reasons, pages on this list are placed in two catagories:
261 261 * pages that are still associated with a vnode, and pages that are not
262 262 * associated with a vnode. Free pages always have their `p_free' bit set,
263 263 * free pages that are still associated with a vnode also have their
264 264 * `p_age' bit set. Pages on the free list are connected via their
265 265 * `p_next' and `p_prev' fields. When a page is involved in some sort
266 266 * of i/o, it is not free and these fields may be used to link associated
267 267 * pages together. At the moment, the free list is protected by a
268 268 * single mutex `page_freelock'. The list of free pages still associated
269 269 * with a vnode is anchored by `page_cachelist' while other free pages
270 270 * are anchored in architecture dependent ways (to handle page coloring etc.).
271 271 *
272 272 * Pages associated with a given vnode appear on a list anchored in the
273 273 * vnode by the `v_pages' field. They are linked together with
274 274 * `p_vpnext' and `p_vpprev'. The field `p_offset' contains a page's
275 275 * offset within the vnode. The pages on this list are not kept in
276 276 * offset order. These lists, in a manner similar to the hash lists,
277 277 * are protected by an array of mutexes called `vph_hash'. Before
278 278 * searching or modifying this chain the appropriate mutex in the
279 279 * vph_hash[] array must be held.
280 280 *
281 281 * Again, each of the lists that a page can appear on is protected by a
282 282 * mutex. Before reading or writing any of the fields comprising the
283 283 * list, the appropriate lock must be held. These list locks should only
284 284 * be held for very short intervals.
285 285 *
286 286 * In addition to the list locks, each page structure contains a
287 287 * shared/exclusive lock that protects various fields within it.
288 288 * To modify one of these fields, the `p_selock' must be exclusively held.
289 289 * To read a field with a degree of certainty, the lock must be at least
290 290 * held shared.
291 291 *
292 292 * Removing a page structure from one of the lists requires holding
293 293 * the appropriate list lock and the page's p_selock. A page may be
294 294 * prevented from changing identity, being freed, or otherwise modified
295 295 * by acquiring p_selock shared.
296 296 *
297 297 * To avoid deadlocks, a strict locking protocol must be followed. Basically
298 298 * there are two cases: In the first case, the page structure in question
299 299 * is known ahead of time (e.g., when the page is to be added or removed
300 300 * from a list). In the second case, the page structure is not known and
301 301 * must be found by searching one of the lists.
302 302 *
303 303 * When adding or removing a known page to one of the lists, first the
304 304 * page must be exclusively locked (since at least one of its fields
305 305 * will be modified), second the lock protecting the list must be acquired,
306 306 * third the page inserted or deleted, and finally the list lock dropped.
307 307 *
308 308 * The more interesting case occures when the particular page structure
309 309 * is not known ahead of time. For example, when a call is made to
310 310 * page_lookup(), it is not known if a page with the desired (vnode and
311 311 * offset pair) identity exists. So the appropriate mutex in ph_mutex is
312 312 * acquired, the hash list searched, and if the desired page is found
313 313 * an attempt is made to lock it. The attempt to acquire p_selock must
314 314 * not block while the hash list lock is held. A deadlock could occure
315 315 * if some other process was trying to remove the page from the list.
316 316 * The removing process (following the above protocol) would have exclusively
317 317 * locked the page, and be spinning waiting to acquire the lock protecting
318 318 * the hash list. Since the searching process holds the hash list lock
319 319 * and is waiting to acquire the page lock, a deadlock occurs.
320 320 *
321 321 * The proper scheme to follow is: first, lock the appropriate list,
322 322 * search the list, and if the desired page is found either use
323 323 * page_trylock() (which will not block) or pass the address of the
324 324 * list lock to page_lock(). If page_lock() can not acquire the page's
325 325 * lock, it will drop the list lock before going to sleep. page_lock()
326 326 * returns a value to indicate if the list lock was dropped allowing the
327 327 * calling program to react appropriately (i.e., retry the operation).
328 328 *
329 329 * If the list lock was dropped before the attempt at locking the page
330 330 * was made, checks would have to be made to ensure that the page had
331 331 * not changed identity before its lock was obtained. This is because
332 332 * the interval between dropping the list lock and acquiring the page
333 333 * lock is indeterminate.
334 334 *
335 335 * In addition, when both a hash list lock (ph_mutex[]) and a vnode list
336 336 * lock (vph_mutex[]) are needed, the hash list lock must be acquired first.
337 337 * The routine page_hashin() is a good example of this sequence.
338 338 * This sequence is ASSERTed by checking that the vph_mutex[] is not held
339 339 * just before each acquisition of one of the mutexs in ph_mutex[].
340 340 *
341 341 * So, as a quick summary:
342 342 *
343 343 * pse_mutex[]'s protect the p_selock and p_cv fields.
344 344 *
345 345 * p_selock protects the p_free, p_age, p_vnode, p_offset and p_hash,
346 346 *
347 347 * ph_mutex[]'s protect the page_hash[] array and its chains.
348 348 *
349 349 * vph_mutex[]'s protect the v_pages field and the vp page chains.
350 350 *
351 351 * First lock the page, then the hash chain, then the vnode chain. When
352 352 * this is not possible `trylocks' must be used. Sleeping while holding
353 353 * any of these mutexes (p_selock is not a mutex) is not allowed.
354 354 *
355 355 *
356 356 * field reading writing ordering
357 357 * ======================================================================
358 358 * p_vnode p_selock(E,S) p_selock(E)
359 359 * p_offset
360 360 * p_free
361 361 * p_age
362 362 * =====================================================================
363 363 * p_hash p_selock(E,S) p_selock(E) && p_selock, ph_mutex
364 364 * ph_mutex[]
365 365 * =====================================================================
366 366 * p_vpnext p_selock(E,S) p_selock(E) && p_selock, vph_mutex
367 367 * p_vpprev vph_mutex[]
368 368 * =====================================================================
369 369 * When the p_free bit is set:
370 370 *
371 371 * p_next p_selock(E,S) p_selock(E) && p_selock,
372 372 * p_prev page_freelock page_freelock
373 373 *
374 374 * When the p_free bit is not set:
375 375 *
376 376 * p_next p_selock(E,S) p_selock(E) && p_selock, p_iolock
377 377 * p_prev p_iolock
378 378 * =====================================================================
379 379 * p_selock pse_mutex[] pse_mutex[] can`t acquire any
380 380 * p_cv other mutexes or
381 381 * sleep while holding
382 382 * this lock.
383 383 * =====================================================================
384 384 * p_lckcnt p_selock(E,S) p_selock(E)
385 385 * OR
386 386 * p_selock(S) &&
387 387 * page_llocks[]
388 388 * p_cowcnt
389 389 * =====================================================================
390 390 * p_nrm hat layer lock hat layer lock
391 391 * p_mapping
392 392 * p_pagenum
393 393 * =====================================================================
394 394 *
395 395 * where:
396 396 * E----> exclusive version of p_selock.
397 397 * S----> shared version of p_selock.
398 398 *
399 399 *
400 400 * Global data structures and variable:
401 401 *
402 402 * field reading writing ordering
403 403 * =====================================================================
404 404 * page_hash[] ph_mutex[] ph_mutex[] can hold this lock
405 405 * before acquiring
406 406 * a vph_mutex or
407 407 * pse_mutex.
408 408 * =====================================================================
409 409 * vp->v_pages vph_mutex[] vph_mutex[] can only acquire
410 410 * a pse_mutex while
411 411 * holding this lock.
412 412 * =====================================================================
413 413 * page_cachelist page_freelock page_freelock can't acquire any
414 414 * page_freelist page_freelock page_freelock
415 415 * =====================================================================
416 416 * freemem freemem_lock freemem_lock can't acquire any
417 417 * freemem_wait other mutexes while
418 418 * freemem_cv holding this mutex.
419 419 * =====================================================================
420 420 *
421 421 * Page relocation, PG_NORELOC and P_NORELOC.
422 422 *
423 423 * Pages may be relocated using the page_relocate() interface. Relocation
424 424 * involves moving the contents and identity of a page to another, free page.
425 425 * To relocate a page, the SE_EXCL lock must be obtained. The way to prevent
426 426 * a page from being relocated is to hold the SE_SHARED lock (the SE_EXCL
427 427 * lock must not be held indefinitely). If the page is going to be held
428 428 * SE_SHARED indefinitely, then the PG_NORELOC hint should be passed
429 429 * to page_create_va so that pages that are prevented from being relocated
430 430 * can be managed differently by the platform specific layer.
431 431 *
432 432 * Pages locked in memory using page_pp_lock (p_lckcnt/p_cowcnt != 0)
433 433 * are guaranteed to be held in memory, but can still be relocated
434 434 * providing the SE_EXCL lock can be obtained.
435 435 *
436 436 * The P_NORELOC bit in the page_t.p_state field is provided for use by
437 437 * the platform specific code in managing pages when the PG_NORELOC
438 438 * hint is used.
439 439 *
440 440 * Memory delete and page locking.
441 441 *
442 442 * The set of all usable pages is managed using the global page list as
443 443 * implemented by the memseg structure defined below. When memory is added
444 444 * or deleted this list changes. Additions to this list guarantee that the
445 445 * list is never corrupt. In order to avoid the necessity of an additional
446 446 * lock to protect against failed accesses to the memseg being deleted and,
447 447 * more importantly, the page_ts, the memseg structure is never freed and the
448 448 * page_t virtual address space is remapped to a page (or pages) of
449 449 * zeros. If a page_t is manipulated while it is p_selock'd, or if it is
450 450 * locked indirectly via a hash or freelist lock, it is not possible for
451 451 * memory delete to collect the page and so that part of the page list is
452 452 * prevented from being deleted. If the page is referenced outside of one
453 453 * of these locks, it is possible for the page_t being referenced to be
454 454 * deleted. Examples of this are page_t pointers returned by
455 455 * page_numtopp_nolock, page_first and page_next. Providing the page_t
456 456 * is re-checked after taking the p_selock (for p_vnode != NULL), the
457 457 * remapping to the zero pages will be detected.
458 458 *
459 459 *
460 460 * Page size (p_szc field) and page locking.
461 461 *
462 462 * p_szc field of free pages is changed by free list manager under freelist
463 463 * locks and is of no concern to the rest of VM subsystem.
464 464 *
465 465 * p_szc changes of allocated anonymous (swapfs) can only be done only after
466 466 * exclusively locking all constituent pages and calling hat_pageunload() on
467 467 * each of them. To prevent p_szc changes of non free anonymous (swapfs) large
468 468 * pages it's enough to either lock SHARED any of constituent pages or prevent
469 469 * hat_pageunload() by holding hat level lock that protects mapping lists (this
470 470 * method is for hat code only)
471 471 *
472 472 * To increase (promote) p_szc of allocated non anonymous file system pages
473 473 * one has to first lock exclusively all involved constituent pages and call
474 474 * hat_pageunload() on each of them. To prevent p_szc promote it's enough to
475 475 * either lock SHARED any of constituent pages that will be needed to make a
476 476 * large page or prevent hat_pageunload() by holding hat level lock that
477 477 * protects mapping lists (this method is for hat code only).
478 478 *
479 479 * To decrease (demote) p_szc of an allocated non anonymous file system large
480 480 * page one can either use the same method as used for changeing p_szc of
481 481 * anonymous large pages or if it's not possible to lock all constituent pages
482 482 * exclusively a different method can be used. In the second method one only
483 483 * has to exclusively lock one of constituent pages but then one has to
484 484 * acquire further locks by calling page_szc_lock() and
485 485 * hat_page_demote(). hat_page_demote() acquires hat level locks and then
486 486 * demotes the page. This mechanism relies on the fact that any code that
487 487 * needs to prevent p_szc of a file system large page from changeing either
488 488 * locks all constituent large pages at least SHARED or locks some pages at
489 489 * least SHARED and calls page_szc_lock() or uses hat level page locks.
490 490 * Demotion using this method is implemented by page_demote_vp_pages().
491 491 * Please see comments in front of page_demote_vp_pages(), hat_page_demote()
492 492 * and page_szc_lock() for more details.
493 493 *
494 494 * Lock order: p_selock, page_szc_lock, ph_mutex/vph_mutex/freelist,
495 495 * hat level locks.
496 496 */
497 497
498 498 typedef struct page {
499 499 u_offset_t p_offset; /* offset into vnode for this page */
500 500 struct vnode *p_vnode; /* vnode that this page is named by */
501 501 selock_t p_selock; /* shared/exclusive lock on the page */
502 502 #if defined(_LP64)
503 503 uint_t p_vpmref; /* vpm ref - index of the vpmap_t */
504 504 #endif
505 505 struct page *p_hash; /* hash by [vnode, offset] */
506 506 struct page *p_vpnext; /* next page in vnode list */
507 507 struct page *p_vpprev; /* prev page in vnode list */
508 508 struct page *p_next; /* next page in free/intrans lists */
509 509 struct page *p_prev; /* prev page in free/intrans lists */
510 510 ushort_t p_lckcnt; /* number of locks on page data */
511 511 ushort_t p_cowcnt; /* number of copy on write lock */
512 512 kcondvar_t p_cv; /* page struct's condition var */
513 513 kcondvar_t p_io_cv; /* for iolock */
514 514 uchar_t p_iolock_state; /* replaces p_iolock */
515 515 volatile uchar_t p_szc; /* page size code */
516 516 uchar_t p_fsdata; /* file system dependent byte */
517 517 uchar_t p_state; /* p_free, p_noreloc */
518 518 uchar_t p_nrm; /* non-cache, ref, mod readonly bits */
519 519 #if defined(__sparc)
520 520 uchar_t p_vcolor; /* virtual color */
521 521 #else
522 522 uchar_t p_embed; /* x86 - changes p_mapping & p_index */
523 523 #endif
524 524 uchar_t p_index; /* MPSS mapping info. Not used on x86 */
525 525 uchar_t p_toxic; /* page has an unrecoverable error */
526 526 void *p_mapping; /* hat specific translation info */
527 527 pfn_t p_pagenum; /* physical page number */
528 528
529 529 uint_t p_share; /* number of translations */
530 530 #if defined(_LP64)
531 531 uint_t p_sharepad; /* pad for growing p_share */
532 532 #endif
533 533 uint_t p_slckcnt; /* number of softlocks */
534 534 #if defined(__sparc)
535 535 uint_t p_kpmref; /* number of kpm mapping sharers */
536 536 struct kpme *p_kpmelist; /* kpm specific mapping info */
537 537 #else
538 538 /* index of entry in p_map when p_embed is set */
539 539 uint_t p_mlentry;
540 540 #endif
541 541 #if defined(_LP64)
542 542 kmutex_t p_ilock; /* protects p_vpmref */
543 543 #else
544 544 uint64_t p_msresv_2; /* page allocation debugging */
545 545 #endif
546 546 } page_t;
547 547
548 548
549 549 typedef page_t devpage_t;
550 550 #define devpage page
551 551
552 552 #define PAGE_LOCK_MAXIMUM \
553 553 ((1 << (sizeof (((page_t *)0)->p_lckcnt) * NBBY)) - 1)
554 554
555 555 #define PAGE_SLOCK_MAXIMUM UINT_MAX
556 556
557 557 /*
558 558 * Page hash table is a power-of-two in size, externally chained
559 559 * through the hash field. PAGE_HASHAVELEN is the average length
560 560 * desired for this chain, from which the size of the page_hash
561 561 * table is derived at boot time and stored in the kernel variable
562 562 * page_hashsz. In the hash function it is given by PAGE_HASHSZ.
563 563 *
564 564 * PAGE_HASH_FUNC returns an index into the page_hash[] array. This
565 565 * index is also used to derive the mutex that protects the chain.
566 566 *
567 567 * In constructing the hash function, first we dispose of unimportant bits
568 568 * (page offset from "off" and the low 3 bits of "vp" which are zero for
569 569 * struct alignment). Then shift and sum the remaining bits a couple times
570 570 * in order to get as many source bits from the two source values into the
571 571 * resulting hashed value. Note that this will perform quickly, since the
572 572 * shifting/summing are fast register to register operations with no additional
573 573 * memory references).
574 574 *
575 575 * PH_SHIFT_SIZE is the amount to use for the successive shifts in the hash
576 576 * function below. The actual value is LOG2(PH_TABLE_SIZE), so that as many
577 577 * bits as possible will filter thru PAGE_HASH_FUNC() and PAGE_HASH_MUTEX().
578 578 *
579 579 * We use ? : instead of #if because <vm/page.h> is included everywhere;
580 580 * NCPU maps to a global variable outside of the "unix" module.
581 581 */
582 582 #if defined(_LP64)
583 583 #define PH_SHIFT_SIZE ((NCPU < 4) ? 7 : (NCPU_LOG2 + 1))
584 584 #else /* 32 bits */
585 585 #define PH_SHIFT_SIZE ((NCPU < 4) ? 4 : 7)
586 586 #endif /* _LP64 */
587 587
588 588 #define PH_TABLE_SIZE (1ul << PH_SHIFT_SIZE)
589 589
590 590 /*
591 591 *
592 592 * We take care to get as much randomness as possible from both the vp and
593 593 * the offset. Workloads can have few vnodes with many offsets, many vnodes
594 594 * with few offsets or a moderate mix of both. This hash should perform
595 595 * equally well for each of these possibilities and for all types of memory
596 596 * allocations.
597 597 *
598 598 * vnodes representing files are created over a long period of time and
599 599 * have good variation in the upper vp bits, and the right shifts below
600 600 * capture these bits. However, swap vnodes are created quickly in a
601 601 * narrow vp* range. Refer to comments at swap_alloc: vnum has exactly
602 602 * AN_VPSHIFT bits, so the kmem_alloc'd vnode addresses have approximately
603 603 * AN_VPSHIFT bits of variation above their VNODE_ALIGN low order 0 bits.
604 604 * Spread swap vnodes widely in the hash table by XOR'ing a term with the
605 605 * vp bits of variation left shifted to the top of the range.
606 606 */
607 607
608 608 #define PAGE_HASHSZ page_hashsz
609 609 #define PAGE_HASHAVELEN 4
610 610 #define PAGE_HASH_FUNC(vp, off) \
611 611 (((((uintptr_t)(off) >> PAGESHIFT) ^ \
612 612 ((uintptr_t)(off) >> (PAGESHIFT + PH_SHIFT_SIZE))) ^ \
613 613 (((uintptr_t)(vp) >> 3) ^ \
614 614 ((uintptr_t)(vp) >> (3 + PH_SHIFT_SIZE)) ^ \
615 615 ((uintptr_t)(vp) >> (3 + 2 * PH_SHIFT_SIZE)) ^ \
616 616 ((uintptr_t)(vp) << \
617 617 (page_hashsz_shift - AN_VPSHIFT - VNODE_ALIGN_LOG2)))) & \
618 618 (PAGE_HASHSZ - 1))
619 619
620 620 #ifdef _KERNEL
621 621
622 622 /*
623 623 * The page hash value is re-hashed to an index for the ph_mutex array.
624 624 *
625 625 * For 64 bit kernels, the mutex array is padded out to prevent false
626 626 * sharing of cache sub-blocks (64 bytes) of adjacent mutexes.
627 627 *
628 628 * For 32 bit kernels, we don't want to waste kernel address space with
629 629 * padding, so instead we rely on the hash function to introduce skew of
630 630 * adjacent vnode/offset indexes (the left shift part of the hash function).
631 631 * Since sizeof (kmutex_t) is 8, we shift an additional 3 to skew to a different
632 632 * 64 byte sub-block.
633 633 */
634 634 extern pad_mutex_t ph_mutex[];
635 635
636 636 #define PAGE_HASH_MUTEX(x) \
637 637 &(ph_mutex[((x) ^ ((x) >> PH_SHIFT_SIZE) + ((x) << 3)) & \
638 638 (PH_TABLE_SIZE - 1)].pad_mutex)
639 639
640 640 /*
641 641 * Flags used while creating pages.
642 642 */
643 643 #define PG_EXCL 0x0001
644 644 #define PG_WAIT 0x0002 /* Blocking memory allocations */
645 645 #define PG_PHYSCONTIG 0x0004 /* NOT SUPPORTED */
646 646 #define PG_MATCH_COLOR 0x0008 /* SUPPORTED by free list routines */
647 647 #define PG_NORELOC 0x0010 /* Non-relocatable alloc hint. */
648 648 /* Page must be PP_ISNORELOC */
649 649 #define PG_PANIC 0x0020 /* system will panic if alloc fails */
650 650 #define PG_PUSHPAGE 0x0040 /* alloc may use reserve */
651 651 #define PG_LOCAL 0x0080 /* alloc from given lgrp only */
652 652 #define PG_NORMALPRI 0x0100 /* PG_WAIT like priority, but */
653 653 /* non-blocking */
654 654 /*
655 655 * When p_selock has the SE_EWANTED bit set, threads waiting for SE_EXCL
656 656 * access are given priority over all other waiting threads.
657 657 */
658 658 #define SE_EWANTED 0x40000000
659 659 #define PAGE_LOCKED(pp) (((pp)->p_selock & ~SE_EWANTED) != 0)
660 660 #define PAGE_SHARED(pp) (((pp)->p_selock & ~SE_EWANTED) > 0)
661 661 #define PAGE_EXCL(pp) ((pp)->p_selock < 0)
662 662 #define PAGE_LOCKED_SE(pp, se) \
663 663 ((se) == SE_EXCL ? PAGE_EXCL(pp) : PAGE_SHARED(pp))
664 664
665 665 extern long page_hashsz;
666 666 extern unsigned int page_hashsz_shift;
667 667 extern page_t **page_hash;
668 668
669 669 extern pad_mutex_t page_llocks[]; /* page logical lock mutex */
670 670 extern kmutex_t freemem_lock; /* freemem lock */
671 671
672 672 extern pgcnt_t total_pages; /* total pages in the system */
673 673
674 674 /*
675 675 * Variables controlling locking of physical memory.
676 676 */
677 677 extern pgcnt_t pages_pp_maximum; /* tuning: lock + claim <= max */
678 678 extern void init_pages_pp_maximum(void);
679 679
680 680 struct lgrp;
681 681
682 682 /* page_list_{add,sub} flags */
683 683
684 684 /* which list */
685 685 #define PG_FREE_LIST 0x0001
686 686 #define PG_CACHE_LIST 0x0002
687 687
688 688 /* where on list */
689 689 #define PG_LIST_TAIL 0x0010
690 690 #define PG_LIST_HEAD 0x0020
691 691
692 692 /* called from */
693 693 #define PG_LIST_ISINIT 0x1000
694 694
695 695 /*
696 696 * Page frame operations.
697 697 */
698 698 page_t *page_lookup(struct vnode *, u_offset_t, se_t);
699 699 page_t *page_lookup_create(struct vnode *, u_offset_t, se_t, page_t *,
700 700 spgcnt_t *, int);
701 701 page_t *page_lookup_nowait(struct vnode *, u_offset_t, se_t);
702 702 page_t *page_find(struct vnode *, u_offset_t);
703 703 page_t *page_exists(struct vnode *, u_offset_t);
704 704 int page_exists_physcontig(vnode_t *, u_offset_t, uint_t, page_t *[]);
705 705 int page_exists_forreal(struct vnode *, u_offset_t, uint_t *);
706 706 void page_needfree(spgcnt_t);
707 707 page_t *page_create(struct vnode *, u_offset_t, size_t, uint_t);
708 708 int page_alloc_pages(struct vnode *, struct seg *, caddr_t, page_t **,
709 709 page_t **, uint_t, int, int);
710 710 page_t *page_create_va_large(vnode_t *vp, u_offset_t off, size_t bytes,
711 711 uint_t flags, struct seg *seg, caddr_t vaddr, void *arg);
712 712 page_t *page_create_va(struct vnode *, u_offset_t, size_t, uint_t,
713 713 struct seg *, caddr_t);
714 714 int page_create_wait(pgcnt_t npages, uint_t flags);
715 715 void page_create_putback(spgcnt_t npages);
716 716 void page_free(page_t *, int);
717 717 void page_free_at_startup(page_t *);
718 718 void page_free_pages(page_t *);
719 719 void free_vp_pages(struct vnode *, u_offset_t, size_t);
720 720 int page_reclaim(page_t *, kmutex_t *);
721 721 int page_reclaim_pages(page_t *, kmutex_t *, uint_t);
722 722 void page_destroy(page_t *, int);
723 723 void page_destroy_pages(page_t *);
724 724 void page_destroy_free(page_t *);
725 725 void page_rename(page_t *, struct vnode *, u_offset_t);
726 726 int page_hashin(page_t *, struct vnode *, u_offset_t, kmutex_t *);
727 727 void page_hashout(page_t *, kmutex_t *);
728 728 int page_num_hashin(pfn_t, struct vnode *, u_offset_t);
729 729 void page_add(page_t **, page_t *);
730 730 void page_add_common(page_t **, page_t *);
731 731 void page_sub(page_t **, page_t *);
732 732 void page_sub_common(page_t **, page_t *);
733 733 page_t *page_get_freelist(struct vnode *, u_offset_t, struct seg *,
734 734 caddr_t, size_t, uint_t, struct lgrp *);
735 735
736 736 page_t *page_get_cachelist(struct vnode *, u_offset_t, struct seg *,
737 737 caddr_t, uint_t, struct lgrp *);
738 738 #if defined(__i386) || defined(__amd64)
739 739 int page_chk_freelist(uint_t);
740 740 #endif
741 741 void page_list_add(page_t *, int);
742 742 void page_boot_demote(page_t *);
743 743 void page_promote_size(page_t *, uint_t);
744 744 void page_list_add_pages(page_t *, int);
745 745 void page_list_sub(page_t *, int);
746 746 void page_list_sub_pages(page_t *, uint_t);
747 747 void page_list_xfer(page_t *, int, int);
748 748 void page_list_break(page_t **, page_t **, size_t);
749 749 void page_list_concat(page_t **, page_t **);
750 750 void page_vpadd(page_t **, page_t *);
751 751 void page_vpsub(page_t **, page_t *);
752 752 int page_lock(page_t *, se_t, kmutex_t *, reclaim_t);
753 753 int page_lock_es(page_t *, se_t, kmutex_t *, reclaim_t, int);
754 754 void page_lock_clr_exclwanted(page_t *);
755 755 int page_trylock(page_t *, se_t);
756 756 int page_try_reclaim_lock(page_t *, se_t, int);
757 757 int page_tryupgrade(page_t *);
758 758 void page_downgrade(page_t *);
759 759 void page_unlock(page_t *);
760 760 void page_unlock_nocapture(page_t *);
761 761 void page_lock_delete(page_t *);
762 762 int page_deleted(page_t *);
763 763 int page_pp_lock(page_t *, int, int);
764 764 void page_pp_unlock(page_t *, int, int);
765 765 int page_resv(pgcnt_t, uint_t);
766 766 void page_unresv(pgcnt_t);
767 767 void page_pp_useclaim(page_t *, page_t *, uint_t);
768 768 int page_addclaim(page_t *);
769 769 int page_subclaim(page_t *);
770 770 int page_addclaim_pages(page_t **);
771 771 int page_subclaim_pages(page_t **);
772 772 pfn_t page_pptonum(page_t *);
773 773 page_t *page_numtopp(pfn_t, se_t);
774 774 page_t *page_numtopp_noreclaim(pfn_t, se_t);
775 775 page_t *page_numtopp_nolock(pfn_t);
776 776 page_t *page_numtopp_nowait(pfn_t, se_t);
777 777 page_t *page_first();
778 778 page_t *page_next(page_t *);
779 779 page_t *page_list_next(page_t *);
780 780 page_t *page_nextn(page_t *, ulong_t);
781 781 page_t *page_next_scan_init(void **);
782 782 page_t *page_next_scan_large(page_t *, ulong_t *, void **);
783 783 void prefetch_page_r(void *);
784 784 int ppcopy(page_t *, page_t *);
785 785 void page_relocate_hash(page_t *, page_t *);
786 786 void pagezero(page_t *, uint_t, uint_t);
787 787 void pagescrub(page_t *, uint_t, uint_t);
788 788 void page_io_lock(page_t *);
789 789 void page_io_unlock(page_t *);
790 790 int page_io_trylock(page_t *);
791 791 int page_iolock_assert(page_t *);
792 792 void page_iolock_init(page_t *);
793 793 void page_io_wait(page_t *);
794 794 int page_io_locked(page_t *);
795 795 pgcnt_t page_busy(int);
796 796 void page_lock_init(void);
797 797 ulong_t page_share_cnt(page_t *);
798 798 int page_isshared(page_t *);
799 799 int page_isfree(page_t *);
800 800 int page_isref(page_t *);
801 801 int page_ismod(page_t *);
802 802 int page_release(page_t *, int);
803 803 void page_retire_init(void);
804 804 int page_retire(uint64_t, uchar_t);
805 805 int page_retire_check(uint64_t, uint64_t *);
806 806 int page_unretire(uint64_t);
807 807 int page_unretire_pp(page_t *, int);
808 808 void page_tryretire(page_t *);
809 809 void page_retire_mdboot();
810 810 uint64_t page_retire_pend_count(void);
811 811 uint64_t page_retire_pend_kas_count(void);
812 812 void page_retire_incr_pend_count(void *);
813 813 void page_retire_decr_pend_count(void *);
814 814 void page_clrtoxic(page_t *, uchar_t);
815 815 void page_settoxic(page_t *, uchar_t);
816 816
817 817 int page_reclaim_mem(pgcnt_t, pgcnt_t, int);
818 818
819 819 void page_set_props(page_t *, uint_t);
820 820 void page_clr_all_props(page_t *);
821 821 int page_clear_lck_cow(page_t *, int);
822 822
823 823 kmutex_t *page_vnode_mutex(struct vnode *);
824 824 kmutex_t *page_se_mutex(struct page *);
825 825 kmutex_t *page_szc_lock(struct page *);
826 826 int page_szc_lock_assert(struct page *pp);
827 827
828 828 /*
829 829 * Page relocation interfaces. page_relocate() is generic.
830 830 * page_get_replacement_page() is provided by the PSM.
831 831 * page_free_replacement_page() is generic.
832 832 */
833 833 int group_page_trylock(page_t *, se_t);
834 834 void group_page_unlock(page_t *);
835 835 int page_relocate(page_t **, page_t **, int, int, spgcnt_t *, struct lgrp *);
836 836 int do_page_relocate(page_t **, page_t **, int, spgcnt_t *, struct lgrp *);
837 837 page_t *page_get_replacement_page(page_t *, struct lgrp *, uint_t);
838 838 void page_free_replacement_page(page_t *);
839 839 int page_relocate_cage(page_t **, page_t **);
840 840
841 841 int page_try_demote_pages(page_t *);
842 842 int page_try_demote_free_pages(page_t *);
843 843 void page_demote_free_pages(page_t *);
844 844
845 845 struct anon_map;
846 846
847 847 void page_mark_migrate(struct seg *, caddr_t, size_t, struct anon_map *,
848 848 ulong_t, vnode_t *, u_offset_t, int);
849 849 void page_migrate(struct seg *, caddr_t, page_t **, pgcnt_t);
850 850
851 851 /*
852 852 * Tell the PIM we are adding physical memory
853 853 */
854 854 void add_physmem(page_t *, size_t, pfn_t);
855 855 void add_physmem_cb(page_t *, pfn_t); /* callback for page_t part */
856 856
857 857 /*
858 858 * hw_page_array[] is configured with hardware supported page sizes by
859 859 * platform specific code.
860 860 */
861 861 typedef struct {
862 862 size_t hp_size;
863 863 uint_t hp_shift;
864 864 uint_t hp_colors;
865 865 pgcnt_t hp_pgcnt; /* base pagesize cnt */
866 866 } hw_pagesize_t;
867 867
868 868 extern hw_pagesize_t hw_page_array[];
869 869 extern uint_t page_coloring_shift;
870 870 extern uint_t page_colors_mask;
871 871 extern int cpu_page_colors;
872 872 extern uint_t colorequiv;
873 873 extern uchar_t colorequivszc[];
874 874
875 875 uint_t page_num_pagesizes(void);
876 876 uint_t page_num_user_pagesizes(int);
877 877 size_t page_get_pagesize(uint_t);
878 878 size_t page_get_user_pagesize(uint_t n);
879 879 pgcnt_t page_get_pagecnt(uint_t);
880 880 uint_t page_get_shift(uint_t);
881 881 int page_szc(size_t);
882 882 int page_szc_user_filtered(size_t);
883 883
884 884 /* page_get_replacement page flags */
885 885 #define PGR_SAMESZC 0x1 /* only look for page size same as orig */
886 886 #define PGR_NORELOC 0x2 /* allocate a P_NORELOC page */
887 887
888 888 /*
889 889 * macros for "masked arithmetic"
890 890 * The purpose is to step through all combinations of a set of bits while
891 891 * keeping some other bits fixed. Fixed bits need not be contiguous. The
892 892 * variable bits need not be contiguous either, or even right aligned. The
893 893 * trick is to set all fixed bits to 1, then increment, then restore the
894 894 * fixed bits. If incrementing causes a carry from a low bit position, the
895 895 * carry propagates thru the fixed bits, because they are temporarily set to 1.
896 896 * v is the value
897 897 * i is the increment
898 898 * eq_mask defines the fixed bits
899 899 * mask limits the size of the result
900 900 */
901 901 #define ADD_MASKED(v, i, eq_mask, mask) \
902 902 (((((v) | (eq_mask)) + (i)) & (mask) & ~(eq_mask)) | ((v) & (eq_mask)))
903 903
904 904 /*
905 905 * convenience macro which increments by 1
906 906 */
907 907 #define INC_MASKED(v, eq_mask, mask) ADD_MASKED(v, 1, eq_mask, mask)
908 908
909 909 #endif /* _KERNEL */
910 910
911 911 /*
912 912 * Constants used for the p_iolock_state
913 913 */
914 914 #define PAGE_IO_INUSE 0x1
915 915 #define PAGE_IO_WANTED 0x2
916 916
917 917 /*
918 918 * Constants used for page_release status
919 919 */
920 920 #define PGREL_NOTREL 0x1
921 921 #define PGREL_CLEAN 0x2
922 922 #define PGREL_MOD 0x3
923 923
924 924 /*
925 925 * The p_state field holds what used to be the p_age and p_free
926 926 * bits. These fields are protected by p_selock (see above).
927 927 */
928 928 #define P_FREE 0x80 /* Page on free list */
929 929 #define P_NORELOC 0x40 /* Page is non-relocatable */
930 930 #define P_MIGRATE 0x20 /* Migrate page on next touch */
931 931 #define P_SWAP 0x10 /* belongs to vnode that is V_ISSWAP */
932 932 #define P_BOOTPAGES 0x08 /* member of bootpages list */
933 933 #define P_RAF 0x04 /* page retired at free */
934 934
935 935 #define PP_ISFREE(pp) ((pp)->p_state & P_FREE)
936 936 #define PP_ISAGED(pp) (((pp)->p_state & P_FREE) && \
937 937 ((pp)->p_vnode == NULL))
938 938 #define PP_ISNORELOC(pp) ((pp)->p_state & P_NORELOC)
939 939 #define PP_ISKAS(pp) (VN_ISKAS((pp)->p_vnode))
940 940 #define PP_ISNORELOCKERNEL(pp) (PP_ISNORELOC(pp) && PP_ISKAS(pp))
941 941 #define PP_ISMIGRATE(pp) ((pp)->p_state & P_MIGRATE)
942 942 #define PP_ISSWAP(pp) ((pp)->p_state & P_SWAP)
943 943 #define PP_ISBOOTPAGES(pp) ((pp)->p_state & P_BOOTPAGES)
944 944 #define PP_ISRAF(pp) ((pp)->p_state & P_RAF)
945 945
946 946 #define PP_SETFREE(pp) ((pp)->p_state = ((pp)->p_state & ~P_MIGRATE) \
947 947 | P_FREE)
948 948 #define PP_SETAGED(pp) ASSERT(PP_ISAGED(pp))
949 949 #define PP_SETNORELOC(pp) ((pp)->p_state |= P_NORELOC)
950 950 #define PP_SETMIGRATE(pp) ((pp)->p_state |= P_MIGRATE)
951 951 #define PP_SETSWAP(pp) ((pp)->p_state |= P_SWAP)
952 952 #define PP_SETBOOTPAGES(pp) ((pp)->p_state |= P_BOOTPAGES)
953 953 #define PP_SETRAF(pp) ((pp)->p_state |= P_RAF)
954 954
955 955 #define PP_CLRFREE(pp) ((pp)->p_state &= ~P_FREE)
956 956 #define PP_CLRAGED(pp) ASSERT(!PP_ISAGED(pp))
957 957 #define PP_CLRNORELOC(pp) ((pp)->p_state &= ~P_NORELOC)
958 958 #define PP_CLRMIGRATE(pp) ((pp)->p_state &= ~P_MIGRATE)
959 959 #define PP_CLRSWAP(pp) ((pp)->p_state &= ~P_SWAP)
960 960 #define PP_CLRBOOTPAGES(pp) ((pp)->p_state &= ~P_BOOTPAGES)
961 961 #define PP_CLRRAF(pp) ((pp)->p_state &= ~P_RAF)
962 962
963 963 /*
964 964 * Flags for page_t p_toxic, for tracking memory hardware errors.
965 965 *
966 966 * These flags are OR'ed into p_toxic with page_settoxic() to track which
967 967 * error(s) have occurred on a given page. The flags are cleared with
968 968 * page_clrtoxic(). Both page_settoxic() and page_cleartoxic use atomic
969 969 * primitives to manipulate the p_toxic field so no other locking is needed.
970 970 *
971 971 * When an error occurs on a page, p_toxic is set to record the error. The
972 972 * error could be a memory error or something else (i.e. a datapath). The Page
973 973 * Retire mechanism does not try to determine the exact cause of the error;
974 974 * Page Retire rightly leaves that sort of determination to FMA's Diagnostic
975 975 * Engine (DE).
976 976 *
977 977 * Note that, while p_toxic bits can be set without holding any locks, they
978 978 * should only be cleared while holding the page exclusively locked.
979 979 * There is one exception to this, the PR_CAPTURE bit is protected by a mutex
980 980 * within the page capture logic and thus to set or clear the bit, that mutex
981 981 * needs to be held. The page does not need to be locked but the page_clrtoxic
982 982 * function must be used as we need an atomic operation.
983 983 * Also note that there is what amounts to a hack to prevent recursion with
984 984 * large pages such that if we are unlocking a page and the PR_CAPTURE bit is
985 985 * set, we will only try to capture the page if the current threads T_CAPTURING
986 986 * flag is not set. If the flag is set, the unlock will not try to capture
987 987 * the page even though the PR_CAPTURE bit is set.
988 988 *
989 989 * Pages with PR_UE or PR_FMA flags are retired unconditionally, while pages
990 990 * with PR_MCE are retired if the system has not retired too many of them.
991 991 *
992 992 * A page must be exclusively locked to be retired. Pages can be retired if
993 993 * they are mapped, modified, or both, as long as they are not marked PR_UE,
994 994 * since pages with uncorrectable errors cannot be relocated in memory.
995 995 * Once a page has been successfully retired it is zeroed, attached to the
996 996 * retired_pages vnode and, finally, PR_RETIRED is set in p_toxic. The other
997 997 * p_toxic bits are NOT cleared. Pages are not left locked after retiring them
998 998 * to avoid special case code throughout the kernel; rather, page_*lock() will
999 999 * fail to lock the page, unless SE_RETIRED is passed as an argument.
1000 1000 *
1001 1001 * While we have your attention, go take a look at the comments at the
1002 1002 * beginning of page_retire.c too.
1003 1003 */
1004 1004 #define PR_OK 0x00 /* no problem */
1005 1005 #define PR_MCE 0x01 /* page has seen two or more CEs */
1006 1006 #define PR_UE 0x02 /* page has an unhandled UE */
1007 1007 #define PR_UE_SCRUBBED 0x04 /* page has seen a UE but was cleaned */
1008 1008 #define PR_FMA 0x08 /* A DE wants this page retired */
1009 1009 #define PR_CAPTURE 0x10 /* page is hashed on page_capture_hash[] */
1010 1010 #define PR_RESV 0x20 /* Reserved for future use */
1011 1011 #define PR_MSG 0x40 /* message(s) already printed for this page */
1012 1012 #define PR_RETIRED 0x80 /* This page has been retired */
1013 1013
1014 1014 #define PR_REASONS (PR_UE | PR_MCE | PR_FMA)
1015 1015 #define PR_TOXIC (PR_UE)
1016 1016 #define PR_ERRMASK (PR_UE | PR_UE_SCRUBBED | PR_MCE | PR_FMA)
1017 1017 #define PR_TOXICFLAGS (0xCF)
1018 1018
1019 1019 #define PP_RETIRED(pp) ((pp)->p_toxic & PR_RETIRED)
1020 1020 #define PP_TOXIC(pp) ((pp)->p_toxic & PR_TOXIC)
1021 1021 #define PP_PR_REQ(pp) (((pp)->p_toxic & PR_REASONS) && !PP_RETIRED(pp))
1022 1022 #define PP_PR_NOSHARE(pp) \
1023 1023 ((((pp)->p_toxic & (PR_RETIRED | PR_FMA | PR_UE)) == PR_FMA) && \
1024 1024 !PP_ISKAS(pp))
1025 1025
1026 1026 /*
1027 1027 * Flags for page_unretire_pp
1028 1028 */
1029 1029 #define PR_UNR_FREE 0x1
1030 1030 #define PR_UNR_CLEAN 0x2
1031 1031 #define PR_UNR_TEMP 0x4
1032 1032
1033 1033 /*
1034 1034 * kpm large page description.
1035 1035 * The virtual address range of segkpm is divided into chunks of
1036 1036 * kpm_pgsz. Each chunk is controlled by a kpm_page_t. The ushort
1037 1037 * is sufficient for 2^^15 * PAGESIZE, so e.g. the maximum kpm_pgsz
1038 1038 * for 8K is 256M and 2G for 64K pages. It it kept as small as
1039 1039 * possible to save physical memory space.
1040 1040 *
1041 1041 * There are 2 segkpm mapping windows within in the virtual address
1042 1042 * space when we have to prevent VAC alias conflicts. The so called
1043 1043 * Alias window (mappings are always by PAGESIZE) is controlled by
1044 1044 * kp_refcnta. The regular window is controlled by kp_refcnt for the
1045 1045 * normal operation, which is to use the largest available pagesize.
1046 1046 * When VAC alias conflicts are present within a chunk in the regular
1047 1047 * window the large page mapping is broken up into smaller PAGESIZE
1048 1048 * mappings. kp_refcntc is used to control the pages that are invoked
1049 1049 * in the conflict and kp_refcnts holds the active mappings done
1050 1050 * with the small page size. In non vac conflict mode kp_refcntc is
1051 1051 * also used as "go" indication (-1) for the trap level tsbmiss
1052 1052 * handler.
1053 1053 */
1054 1054 typedef struct kpm_page {
1055 1055 short kp_refcnt; /* pages mapped large */
1056 1056 short kp_refcnta; /* pages mapped in Alias window */
1057 1057 short kp_refcntc; /* TL-tsbmiss flag; #vac alias conflict pages */
1058 1058 short kp_refcnts; /* vac alias: pages mapped small */
1059 1059 } kpm_page_t;
1060 1060
1061 1061 /*
1062 1062 * Note: khl_lock offset changes must be reflected in sfmmu_asm.s
1063 1063 */
1064 1064 typedef struct kpm_hlk {
1065 1065 kmutex_t khl_mutex; /* kpm_page mutex */
1066 1066 uint_t khl_lock; /* trap level tsbmiss handling */
1067 1067 } kpm_hlk_t;
1068 1068
1069 1069 /*
1070 1070 * kpm small page description.
1071 1071 * When kpm_pgsz is equal to PAGESIZE a smaller representation is used
1072 1072 * to save memory space. Alias range mappings and regular segkpm
1073 1073 * mappings are done in units of PAGESIZE and can share the mapping
1074 1074 * information and the mappings are always distinguishable by their
1075 1075 * virtual address. Other information needed for VAC conflict prevention
1076 1076 * is already available on a per page basis.
1077 1077 *
1078 1078 * The state about how a kpm page is mapped and whether it is ready to go
1079 1079 * is indicated by the following 1 byte kpm_spage structure. This byte is
1080 1080 * split into two 4-bit parts - kp_mapped and kp_mapped_go.
1081 1081 * - kp_mapped == 1 the page is mapped cacheable
1082 1082 * - kp_mapped == 2 the page is mapped non-cacheable
1083 1083 * - kp_mapped_go == 1 the mapping is ready to be dropped in
1084 1084 * - kp_mapped_go == 0 the mapping is not ready to be dropped in.
1085 1085 * When kp_mapped_go == 0, we will have C handler resolve the VAC conflict.
1086 1086 * Otherwise, the assembly tsb miss handler can simply drop in the mapping
1087 1087 * when a tsb miss occurs.
1088 1088 */
1089 1089 typedef union kpm_spage {
1090 1090 struct {
1091 1091 #ifdef _BIG_ENDIAN
1092 1092 uchar_t mapped_go: 4; /* go or nogo flag */
1093 1093 uchar_t mapped: 4; /* page mapped small */
1094 1094 #else
1095 1095 uchar_t mapped: 4; /* page mapped small */
1096 1096 uchar_t mapped_go: 4; /* go or nogo flag */
1097 1097 #endif
1098 1098 } kpm_spage_un;
1099 1099 uchar_t kp_mapped_flag;
1100 1100 } kpm_spage_t;
1101 1101
1102 1102 #define kp_mapped kpm_spage_un.mapped
1103 1103 #define kp_mapped_go kpm_spage_un.mapped_go
1104 1104
1105 1105 /*
1106 1106 * Note: kshl_lock offset changes must be reflected in sfmmu_asm.s
1107 1107 */
1108 1108 typedef struct kpm_shlk {
1109 1109 uint_t kshl_lock; /* trap level tsbmiss handling */
1110 1110 } kpm_shlk_t;
1111 1111
1112 1112 /*
1113 1113 * Each segment of physical memory is described by a memseg struct.
1114 1114 * Within a segment, memory is considered contiguous. The members
1115 1115 * can be categorized as follows:
1116 1116 * . Platform independent:
1117 1117 * pages, epages, pages_base, pages_end, next, lnext.
1118 1118 * . 64bit only but platform independent:
1119 1119 * kpm_pbase, kpm_nkpmpgs, kpm_pages, kpm_spages.
1120 1120 * . Really platform or mmu specific:
1121 1121 * pagespa, epagespa, nextpa, kpm_pagespa.
1122 1122 * . Mixed:
1123 1123 * msegflags.
1124 1124 */
1125 1125 struct memseg {
1126 1126 page_t *pages, *epages; /* [from, to] in page array */
1127 1127 pfn_t pages_base, pages_end; /* [from, to] in page numbers */
1128 1128 struct memseg *next; /* next segment in list */
1129 1129 struct memseg *lnext; /* next segment in deleted list */
1130 1130 #if defined(__sparc)
1131 1131 uint64_t pagespa, epagespa; /* [from, to] page array physical */
1132 1132 uint64_t nextpa; /* physical next pointer */
1133 1133 pfn_t kpm_pbase; /* start of kpm range */
1134 1134 pgcnt_t kpm_nkpmpgs; /* # of kpm_pgsz pages */
1135 1135 union _mseg_un {
1136 1136 kpm_page_t *kpm_lpgs; /* ptr to kpm_page array */
1137 1137 kpm_spage_t *kpm_spgs; /* ptr to kpm_spage array */
1138 1138 } mseg_un;
1139 1139 uint64_t kpm_pagespa; /* physical ptr to kpm (s)pages array */
1140 1140 #endif /* __sparc */
1141 1141 uint_t msegflags; /* memseg flags */
1142 1142 };
1143 1143
1144 1144 /* memseg union aliases */
1145 1145 #define kpm_pages mseg_un.kpm_lpgs
1146 1146 #define kpm_spages mseg_un.kpm_spgs
1147 1147
1148 1148 /* msegflags */
1149 1149 #define MEMSEG_DYNAMIC 0x1 /* DR: memory was added dynamically */
1150 1150 #define MEMSEG_META_INCL 0x2 /* DR: memseg includes it's metadata */
1151 1151 #define MEMSEG_META_ALLOC 0x4 /* DR: memseg allocated it's metadata */
1152 1152
1153 1153 /* memseg support macros */
1154 1154 #define MSEG_NPAGES(SEG) ((SEG)->pages_end - (SEG)->pages_base)
1155 1155
1156 1156 /* memseg hash */
1157 1157 #define MEM_HASH_SHIFT 0x9
1158 1158 #define N_MEM_SLOTS 0x200 /* must be a power of 2 */
1159 1159 #define MEMSEG_PFN_HASH(pfn) (((pfn)/mhash_per_slot) & (N_MEM_SLOTS - 1))
1160 1160
1161 1161 /* memseg externals */
1162 1162 extern struct memseg *memsegs; /* list of memory segments */
1163 1163 extern ulong_t mhash_per_slot;
1164 1164 extern uint64_t memsegspa; /* memsegs as physical address */
1165 1165
1166 1166 void build_pfn_hash();
1167 1167 extern struct memseg *page_numtomemseg_nolock(pfn_t pfnum);
1168 1168
1169 1169 /*
1170 1170 * page capture related info:
1171 1171 * The page capture routines allow us to asynchronously capture given pages
1172 1172 * for the explicit use of the requestor. New requestors can be added by
1173 1173 * explicitly adding themselves to the PC_* flags below and incrementing
1174 1174 * PC_NUM_CALLBACKS as necessary.
1175 1175 *
1176 1176 * Subsystems using page capture must register a callback before attempting
1177 1177 * to capture a page. A duration of -1 will indicate that we will never give
1178 1178 * up while trying to capture a page and will only stop trying to capture the
1179 1179 * given page once we have successfully captured it. Thus the user needs to be
1180 1180 * aware of the behavior of all callers who have a duration of -1.
1181 1181 *
1182 1182 * For now, only /dev/physmem and page retire use the page capture interface
1183 1183 * and only a single request can be outstanding for a given page. Thus, if
1184 1184 * /dev/phsymem wants a page and page retire also wants the same page, only
1185 1185 * the page retire request will be honored until the point in time that the
1186 1186 * page is actually retired, at which point in time, subsequent requests by
1187 1187 * /dev/physmem will succeed if the CAPTURE_GET_RETIRED flag was set.
1188 1188 */
1189 1189
1190 1190 #define PC_RETIRE (0)
1191 1191 #define PC_PHYSMEM (1)
1192 1192 #define PC_NUM_CALLBACKS (2)
1193 1193 #define PC_MASK ((1 << PC_NUM_CALLBACKS) - 1)
1194 1194
1195 1195 #define CAPTURE_RETIRE (1 << PC_RETIRE)
1196 1196 #define CAPTURE_PHYSMEM (1 << PC_PHYSMEM)
1197 1197
1198 1198 #define CAPTURE_ASYNC (0x0200)
1199 1199
1200 1200 #define CAPTURE_GET_RETIRED (0x1000)
1201 1201 #define CAPTURE_GET_CAGE (0x2000)
1202 1202
1203 1203 struct page_capture_callback {
1204 1204 int cb_active; /* 1 means active, 0 means inactive */
1205 1205 clock_t duration; /* the length in time that we'll attempt to */
1206 1206 /* capture this page asynchronously. (in HZ) */
1207 1207 krwlock_t cb_rwlock;
1208 1208 int (*cb_func)(page_t *, void *, uint_t); /* callback function */
1209 1209 };
1210 1210
1211 1211 extern kcondvar_t pc_cv;
1212 1212
1213 1213 void page_capture_register_callback(uint_t index, clock_t duration,
1214 1214 int (*cb_func)(page_t *, void *, uint_t));
1215 1215 void page_capture_unregister_callback(uint_t index);
1216 1216 int page_trycapture(page_t *pp, uint_t szc, uint_t flags, void *datap);
1217 1217 void page_unlock_capture(page_t *pp);
1218 1218 int page_capture_unretire_pp(page_t *);
1219 1219
1220 1220 extern int memsegs_trylock(int);
1221 1221 extern void memsegs_lock(int);
1222 1222 extern void memsegs_unlock(int);
1223 1223 extern int memsegs_lock_held(void);
1224 1224 extern void memlist_read_lock(void);
1225 1225 extern void memlist_read_unlock(void);
1226 1226 extern void memlist_write_lock(void);
1227 1227 extern void memlist_write_unlock(void);
1228 1228
1229 1229 #ifdef __cplusplus
1230 1230 }
1231 1231 #endif
1232 1232
1233 1233 #endif /* _VM_PAGE_H */
↓ open down ↓ |
1124 lines elided |
↑ open up ↑ |
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX