xref: /illumos-gate/usr/src/uts/common/vm/page.h (revision 3df2e8b2fd61f45437285750d2880d6416a9200c)
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 (c) 1986, 2010, Oracle and/or its affiliates. All rights reserved.
23  * Copyright 2021 Oxide Computer Company
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_PAGE_H
40 #define	_VM_PAGE_H
41 
42 #include <vm/seg.h>
43 
44 #ifdef	__cplusplus
45 extern "C" {
46 #endif
47 
48 #if defined(_KERNEL) || defined(_KMEMUSER)
49 
50 /*
51  * Shared/Exclusive lock.
52  */
53 
54 /*
55  * Types of page locking supported by page_lock & friends.
56  */
57 typedef enum {
58 	SE_SHARED,
59 	SE_EXCL			/* exclusive lock (value == -1) */
60 } se_t;
61 
62 /*
63  * For requesting that page_lock reclaim the page from the free list.
64  */
65 typedef enum {
66 	P_RECLAIM,		/* reclaim page from free list */
67 	P_NO_RECLAIM		/* DON`T reclaim the page	*/
68 } reclaim_t;
69 
70 /*
71  * Callers of page_try_reclaim_lock and page_lock_es can use this flag
72  * to get SE_EXCL access before reader/writers are given access.
73  */
74 #define	SE_EXCL_WANTED	0x02
75 
76 /*
77  * All page_*lock() requests will be denied unless this flag is set in
78  * the 'es' parameter.
79  */
80 #define	SE_RETIRED	0x04
81 
82 #endif	/* _KERNEL | _KMEMUSER */
83 
84 typedef int	selock_t;
85 
86 /*
87  * Define VM_STATS to turn on all sorts of statistic gathering about
88  * the VM layer.  By default, it is only turned on when DEBUG is
89  * also defined.
90  */
91 #ifdef DEBUG
92 #define	VM_STATS
93 #endif	/* DEBUG */
94 
95 #ifdef VM_STATS
96 #define	VM_STAT_ADD(stat)			(stat)++
97 #define	VM_STAT_COND_ADD(cond, stat)		((void) (!(cond) || (stat)++))
98 #else
99 #define	VM_STAT_ADD(stat)
100 #define	VM_STAT_COND_ADD(cond, stat)
101 #endif	/* VM_STATS */
102 
103 #ifdef _KERNEL
104 
105 /*
106  * PAGE_LLOCK_SIZE is 2 * NCPU, but no smaller than 128.
107  * PAGE_LLOCK_SHIFT is log2(PAGE_LLOCK_SIZE).
108  *
109  * We use ? : instead of #if because <vm/page.h> is included everywhere;
110  * NCPU_P2 is only a constant in the "unix" module.
111  *
112  */
113 #define	PAGE_LLOCK_SHIFT \
114 	    ((unsigned)(((2*NCPU_P2) > 128) ? NCPU_LOG2 + 1 : 7))
115 
116 #define	PAGE_LLOCK_SIZE (1ul << PAGE_LLOCK_SHIFT)
117 
118 /*
119  * The number of low order 0 (or less variable) bits in the page_t address.
120  */
121 #if defined(__sparc)
122 #define	PP_SHIFT		7
123 #else
124 #define	PP_SHIFT		6
125 #endif
126 
127 /*
128  * pp may be the root of a large page, and many low order bits will be 0.
129  * Shift and XOR multiple times to capture the good bits across the range of
130  * possible page sizes.
131  */
132 #define	PAGE_LLOCK_HASH(pp)	\
133 	(((((uintptr_t)(pp) >> PP_SHIFT) ^ \
134 	((uintptr_t)(pp) >> (PAGE_LLOCK_SHIFT + PP_SHIFT))) ^ \
135 	((uintptr_t)(pp) >> ((PAGE_LLOCK_SHIFT * 2) + PP_SHIFT)) ^ \
136 	((uintptr_t)(pp) >> ((PAGE_LLOCK_SHIFT * 3) + PP_SHIFT))) & \
137 	(PAGE_LLOCK_SIZE - 1))
138 
139 #define	page_struct_lock(pp)	\
140 	mutex_enter(&page_llocks[PAGE_LLOCK_HASH(PP_PAGEROOT(pp))].pad_mutex)
141 #define	page_struct_unlock(pp)	\
142 	mutex_exit(&page_llocks[PAGE_LLOCK_HASH(PP_PAGEROOT(pp))].pad_mutex)
143 
144 #endif	/* _KERNEL */
145 
146 #include <sys/t_lock.h>
147 
148 struct as;
149 
150 /*
151  * Each physical page has a page structure, which is used to maintain
152  * these pages as a cache.  A page can be found via a hashed lookup
153  * based on the [vp, offset].  If a page has an [vp, offset] identity,
154  * then it is entered on a doubly linked circular list off the
155  * vnode using the vpnext/vpprev pointers.   If the p_free bit
156  * is on, then the page is also on a doubly linked circular free
157  * list using next/prev pointers.  If the "p_selock" and "p_iolock"
158  * are held, then the page is currently being read in (exclusive p_selock)
159  * or written back (shared p_selock).  In this case, the next/prev pointers
160  * are used to link the pages together for a consecutive i/o request.  If
161  * the page is being brought in from its backing store, then other processes
162  * will wait for the i/o to complete before attaching to the page since it
163  * will have an "exclusive" lock.
164  *
165  * Each page structure has the locks described below along with
166  * the fields they protect:
167  *
168  *	p_selock	This is a per-page shared/exclusive lock that is
169  *			used to implement the logical shared/exclusive
170  *			lock for each page.  The "shared" lock is normally
171  *			used in most cases while the "exclusive" lock is
172  *			required to destroy or retain exclusive access to
173  *			a page (e.g., while reading in pages).  The appropriate
174  *			lock is always held whenever there is any reference
175  *			to a page structure (e.g., during i/o).
176  *			(Note that with the addition of the "writer-lock-wanted"
177  *			semantics (via SE_EWANTED), threads must not acquire
178  *			multiple reader locks or else a deadly embrace will
179  *			occur in the following situation: thread 1 obtains a
180  *			reader lock; next thread 2 fails to get a writer lock
181  *			but specified SE_EWANTED so it will wait by either
182  *			blocking (when using page_lock_es) or spinning while
183  *			retrying (when using page_try_reclaim_lock) until the
184  *			reader lock is released; then thread 1 attempts to
185  *			get another reader lock but is denied due to
186  *			SE_EWANTED being set, and now both threads are in a
187  *			deadly embrace.)
188  *
189  *				p_hash
190  *				p_vnode
191  *				p_offset
192  *
193  *				p_free
194  *				p_age
195  *
196  *	p_iolock	This is a binary semaphore lock that provides
197  *			exclusive access to the i/o list links in each
198  *			page structure.  It is always held while the page
199  *			is on an i/o list (i.e., involved in i/o).  That is,
200  *			even though a page may be only `shared' locked
201  *			while it is doing a write, the following fields may
202  *			change anyway.  Normally, the page must be
203  *			`exclusively' locked to change anything in it.
204  *
205  *				p_next
206  *				p_prev
207  *
208  * The following fields are protected by the global page_llocks[]:
209  *
210  *				p_lckcnt
211  *				p_cowcnt
212  *
213  * The following lists are protected by the global page_freelock:
214  *
215  *				page_cachelist
216  *				page_freelist
217  *
218  * The following, for our purposes, are protected by
219  * the global freemem_lock:
220  *
221  *				freemem
222  *				freemem_wait
223  *				freemem_cv
224  *
225  * The following fields are protected by hat layer lock(s).  When a page
226  * structure is not mapped and is not associated with a vnode (after a call
227  * to page_hashout() for example) the p_nrm field may be modified with out
228  * holding the hat layer lock:
229  *
230  *				p_nrm
231  *				p_mapping
232  *				p_share
233  *
234  * The following field is file system dependent.  How it is used and
235  * the locking strategies applied are up to the individual file system
236  * implementation.
237  *
238  *				p_fsdata
239  *
240  * The page structure is used to represent and control the system's
241  * physical pages.  There is one instance of the structure for each
242  * page that is not permenately allocated.  For example, the pages that
243  * hold the page structures are permanently held by the kernel
244  * and hence do not need page structures to track them.  The array
245  * of page structures is allocated early on in the kernel's life and
246  * is based on the amount of available physical memory.
247  *
248  * Each page structure may simultaneously appear on several linked lists.
249  * The lists are:  hash list, free or in i/o list, and a vnode's page list.
250  * Each type of list is protected by a different group of mutexes as described
251  * below:
252  *
253  * The hash list is used to quickly find a page when the page's vnode and
254  * offset within the vnode are known.  Each page that is hashed is
255  * connected via the `p_hash' field.  The anchor for each hash is in the
256  * array `page_hash'.  An array of mutexes, `ph_mutex', protects the
257  * lists anchored by page_hash[].  To either search or modify a given hash
258  * list, the appropriate mutex in the ph_mutex array must be held.
259  *
260  * The free list contains pages that are `free to be given away'.  For
261  * efficiency reasons, pages on this list are placed in two catagories:
262  * pages that are still associated with a vnode, and pages that are not
263  * associated with a vnode.  Free pages always have their `p_free' bit set,
264  * free pages that are still associated with a vnode also have their
265  * `p_age' bit set.  Pages on the free list are connected via their
266  * `p_next' and `p_prev' fields.  When a page is involved in some sort
267  * of i/o, it is not free and these fields may be used to link associated
268  * pages together.  At the moment, the free list is protected by a
269  * single mutex `page_freelock'.  The list of free pages still associated
270  * with a vnode is anchored by `page_cachelist' while other free pages
271  * are anchored in architecture dependent ways (to handle page coloring etc.).
272  *
273  * Pages associated with a given vnode appear on a list anchored in the
274  * vnode by the `v_pages' field.  They are linked together with
275  * `p_vpnext' and `p_vpprev'.  The field `p_offset' contains a page's
276  * offset within the vnode.  The pages on this list are not kept in
277  * offset order.  These lists, in a manner similar to the hash lists,
278  * are protected by an array of mutexes called `vph_hash'.  Before
279  * searching or modifying this chain the appropriate mutex in the
280  * vph_hash[] array must be held.
281  *
282  * Again, each of the lists that a page can appear on is protected by a
283  * mutex.  Before reading or writing any of the fields comprising the
284  * list, the appropriate lock must be held.  These list locks should only
285  * be held for very short intervals.
286  *
287  * In addition to the list locks, each page structure contains a
288  * shared/exclusive lock that protects various fields within it.
289  * To modify one of these fields, the `p_selock' must be exclusively held.
290  * To read a field with a degree of certainty, the lock must be at least
291  * held shared.
292  *
293  * Removing a page structure from one of the lists requires holding
294  * the appropriate list lock and the page's p_selock.  A page may be
295  * prevented from changing identity, being freed, or otherwise modified
296  * by acquiring p_selock shared.
297  *
298  * To avoid deadlocks, a strict locking protocol must be followed.  Basically
299  * there are two cases:  In the first case, the page structure in question
300  * is known ahead of time (e.g., when the page is to be added or removed
301  * from a list).  In the second case, the page structure is not known and
302  * must be found by searching one of the lists.
303  *
304  * When adding or removing a known page to one of the lists, first the
305  * page must be exclusively locked (since at least one of its fields
306  * will be modified), second the lock protecting the list must be acquired,
307  * third the page inserted or deleted, and finally the list lock dropped.
308  *
309  * The more interesting case occures when the particular page structure
310  * is not known ahead of time.  For example, when a call is made to
311  * page_lookup(), it is not known if a page with the desired (vnode and
312  * offset pair) identity exists.  So the appropriate mutex in ph_mutex is
313  * acquired, the hash list searched, and if the desired page is found
314  * an attempt is made to lock it.  The attempt to acquire p_selock must
315  * not block while the hash list lock is held.  A deadlock could occure
316  * if some other process was trying to remove the page from the list.
317  * The removing process (following the above protocol) would have exclusively
318  * locked the page, and be spinning waiting to acquire the lock protecting
319  * the hash list.  Since the searching process holds the hash list lock
320  * and is waiting to acquire the page lock, a deadlock occurs.
321  *
322  * The proper scheme to follow is: first, lock the appropriate list,
323  * search the list, and if the desired page is found either use
324  * page_trylock() (which will not block) or pass the address of the
325  * list lock to page_lock().  If page_lock() can not acquire the page's
326  * lock, it will drop the list lock before going to sleep.  page_lock()
327  * returns a value to indicate if the list lock was dropped allowing the
328  * calling program to react appropriately (i.e., retry the operation).
329  *
330  * If the list lock was dropped before the attempt at locking the page
331  * was made, checks would have to be made to ensure that the page had
332  * not changed identity before its lock was obtained.  This is because
333  * the interval between dropping the list lock and acquiring the page
334  * lock is indeterminate.
335  *
336  * In addition, when both a hash list lock (ph_mutex[]) and a vnode list
337  * lock (vph_mutex[]) are needed, the hash list lock must be acquired first.
338  * The routine page_hashin() is a good example of this sequence.
339  * This sequence is ASSERTed by checking that the vph_mutex[] is not held
340  * just before each acquisition of one of the mutexs in ph_mutex[].
341  *
342  * So, as a quick summary:
343  *
344  *	pse_mutex[]'s protect the p_selock and p_cv fields.
345  *
346  *	p_selock protects the p_free, p_age, p_vnode, p_offset and p_hash,
347  *
348  *	ph_mutex[]'s protect the page_hash[] array and its chains.
349  *
350  *	vph_mutex[]'s protect the v_pages field and the vp page chains.
351  *
352  *	First lock the page, then the hash chain, then the vnode chain.  When
353  *	this is not possible `trylocks' must be used.  Sleeping while holding
354  *	any of these mutexes (p_selock is not a mutex) is not allowed.
355  *
356  *
357  *	field		reading		writing		    ordering
358  *	======================================================================
359  *	p_vnode		p_selock(E,S)	p_selock(E)
360  *	p_offset
361  *	p_free
362  *	p_age
363  *	=====================================================================
364  *	p_hash		p_selock(E,S)	p_selock(E) &&	    p_selock, ph_mutex
365  *					ph_mutex[]
366  *	=====================================================================
367  *	p_vpnext	p_selock(E,S)	p_selock(E) &&	    p_selock, vph_mutex
368  *	p_vpprev			vph_mutex[]
369  *	=====================================================================
370  *	When the p_free bit is set:
371  *
372  *	p_next		p_selock(E,S)	p_selock(E) &&	    p_selock,
373  *	p_prev				page_freelock	    page_freelock
374  *
375  *	When the p_free bit is not set:
376  *
377  *	p_next		p_selock(E,S)	p_selock(E) &&	    p_selock, p_iolock
378  *	p_prev				p_iolock
379  *	=====================================================================
380  *	p_selock	pse_mutex[]	pse_mutex[]	    can`t acquire any
381  *	p_cv						    other mutexes or
382  *							    sleep while holding
383  *							    this lock.
384  *	=====================================================================
385  *	p_lckcnt	p_selock(E,S)	p_selock(E)
386  *					    OR
387  *					p_selock(S) &&
388  *					page_llocks[]
389  *	p_cowcnt
390  *	=====================================================================
391  *	p_nrm		hat layer lock	hat layer lock
392  *	p_mapping
393  *	p_pagenum
394  *	=====================================================================
395  *
396  *	where:
397  *		E----> exclusive version of p_selock.
398  *		S----> shared version of p_selock.
399  *
400  *
401  *	Global data structures and variable:
402  *
403  *	field		reading		writing		    ordering
404  *	=====================================================================
405  *	page_hash[]	ph_mutex[]	ph_mutex[]	    can hold this lock
406  *							    before acquiring
407  *							    a vph_mutex or
408  *							    pse_mutex.
409  *	=====================================================================
410  *	vp->v_pages	vph_mutex[]	vph_mutex[]	    can only acquire
411  *							    a pse_mutex while
412  *							    holding this lock.
413  *	=====================================================================
414  *	page_cachelist	page_freelock	page_freelock	    can't acquire any
415  *	page_freelist	page_freelock	page_freelock
416  *	=====================================================================
417  *	freemem		freemem_lock	freemem_lock	    can't acquire any
418  *	freemem_wait					    other mutexes while
419  *	freemem_cv					    holding this mutex.
420  *	=====================================================================
421  *
422  * Page relocation, PG_NORELOC and P_NORELOC.
423  *
424  * Pages may be relocated using the page_relocate() interface. Relocation
425  * involves moving the contents and identity of a page to another, free page.
426  * To relocate a page, the SE_EXCL lock must be obtained. The way to prevent
427  * a page from being relocated is to hold the SE_SHARED lock (the SE_EXCL
428  * lock must not be held indefinitely). If the page is going to be held
429  * SE_SHARED indefinitely, then the PG_NORELOC hint should be passed
430  * to page_create_va so that pages that are prevented from being relocated
431  * can be managed differently by the platform specific layer.
432  *
433  * Pages locked in memory using page_pp_lock (p_lckcnt/p_cowcnt != 0)
434  * are guaranteed to be held in memory, but can still be relocated
435  * providing the SE_EXCL lock can be obtained.
436  *
437  * The P_NORELOC bit in the page_t.p_state field is provided for use by
438  * the platform specific code in managing pages when the PG_NORELOC
439  * hint is used.
440  *
441  * Memory delete and page locking.
442  *
443  * The set of all usable pages is managed using the global page list as
444  * implemented by the memseg structure defined below. When memory is added
445  * or deleted this list changes. Additions to this list guarantee that the
446  * list is never corrupt.  In order to avoid the necessity of an additional
447  * lock to protect against failed accesses to the memseg being deleted and,
448  * more importantly, the page_ts, the memseg structure is never freed and the
449  * page_t virtual address space is remapped to a page (or pages) of
450  * zeros.  If a page_t is manipulated while it is p_selock'd, or if it is
451  * locked indirectly via a hash or freelist lock, it is not possible for
452  * memory delete to collect the page and so that part of the page list is
453  * prevented from being deleted. If the page is referenced outside of one
454  * of these locks, it is possible for the page_t being referenced to be
455  * deleted.  Examples of this are page_t pointers returned by
456  * page_numtopp_nolock, page_first and page_next.  Providing the page_t
457  * is re-checked after taking the p_selock (for p_vnode != NULL), the
458  * remapping to the zero pages will be detected.
459  *
460  *
461  * Page size (p_szc field) and page locking.
462  *
463  * p_szc field of free pages is changed by free list manager under freelist
464  * locks and is of no concern to the rest of VM subsystem.
465  *
466  * p_szc changes of allocated anonymous (swapfs) can only be done only after
467  * exclusively locking all constituent pages and calling hat_pageunload() on
468  * each of them. To prevent p_szc changes of non free anonymous (swapfs) large
469  * pages it's enough to either lock SHARED any of constituent pages or prevent
470  * hat_pageunload() by holding hat level lock that protects mapping lists (this
471  * method is for hat code only)
472  *
473  * To increase (promote) p_szc of allocated non anonymous file system pages
474  * one has to first lock exclusively all involved constituent pages and call
475  * hat_pageunload() on each of them. To prevent p_szc promote it's enough to
476  * either lock SHARED any of constituent pages that will be needed to make a
477  * large page or prevent hat_pageunload() by holding hat level lock that
478  * protects mapping lists (this method is for hat code only).
479  *
480  * To decrease (demote) p_szc of an allocated non anonymous file system large
481  * page one can either use the same method as used for changeing p_szc of
482  * anonymous large pages or if it's not possible to lock all constituent pages
483  * exclusively a different method can be used. In the second method one only
484  * has to exclusively lock one of constituent pages but then one has to
485  * acquire further locks by calling page_szc_lock() and
486  * hat_page_demote(). hat_page_demote() acquires hat level locks and then
487  * demotes the page. This mechanism relies on the fact that any code that
488  * needs to prevent p_szc of a file system large page from changeing either
489  * locks all constituent large pages at least SHARED or locks some pages at
490  * least SHARED and calls page_szc_lock() or uses hat level page locks.
491  * Demotion using this method is implemented by page_demote_vp_pages().
492  * Please see comments in front of page_demote_vp_pages(), hat_page_demote()
493  * and page_szc_lock() for more details.
494  *
495  * Lock order: p_selock, page_szc_lock, ph_mutex/vph_mutex/freelist,
496  * hat level locks.
497  */
498 
499 typedef struct page {
500 	u_offset_t	p_offset;	/* offset into vnode for this page */
501 	struct vnode	*p_vnode;	/* vnode that this page is named by */
502 	selock_t	p_selock;	/* shared/exclusive lock on the page */
503 #if defined(_LP64)
504 	uint_t		p_vpmref;	/* vpm ref - index of the vpmap_t */
505 #endif
506 	struct page	*p_hash;	/* hash by [vnode, offset] */
507 	struct page	*p_vpnext;	/* next page in vnode list */
508 	struct page	*p_vpprev;	/* prev page in vnode list */
509 	struct page	*p_next;	/* next page in free/intrans lists */
510 	struct page	*p_prev;	/* prev page in free/intrans lists */
511 	ushort_t	p_lckcnt;	/* number of locks on page data */
512 	ushort_t	p_cowcnt;	/* number of copy on write lock */
513 	kcondvar_t	p_cv;		/* page struct's condition var */
514 	kcondvar_t	p_io_cv;	/* for iolock */
515 	uchar_t		p_iolock_state;	/* replaces p_iolock */
516 	volatile uchar_t p_szc;		/* page size code */
517 	uchar_t		p_fsdata;	/* file system dependent byte */
518 	uchar_t		p_state;	/* p_free, p_noreloc */
519 	uchar_t		p_nrm;		/* non-cache, ref, mod readonly bits */
520 #if defined(__sparc)
521 	uchar_t		p_vcolor;	/* virtual color */
522 #else
523 	uchar_t		p_embed;	/* x86 - changes p_mapping & p_index */
524 #endif
525 	uchar_t		p_index;	/* MPSS mapping info. Not used on x86 */
526 	uchar_t		p_toxic;	/* page has an unrecoverable error */
527 	void		*p_mapping;	/* hat specific translation info */
528 	pfn_t		p_pagenum;	/* physical page number */
529 
530 	uint_t		p_share;	/* number of translations */
531 #if defined(_LP64)
532 	uint_t		p_sharepad;	/* pad for growing p_share */
533 #endif
534 	uint_t		p_slckcnt;	/* number of softlocks */
535 #if defined(__sparc)
536 	uint_t		p_kpmref;	/* number of kpm mapping sharers */
537 	struct kpme	*p_kpmelist;	/* kpm specific mapping info */
538 #else
539 	/* index of entry in p_map when p_embed is set */
540 	uint_t		p_mlentry;
541 #endif
542 #if defined(_LP64)
543 	kmutex_t	p_ilock;	/* protects p_vpmref */
544 #else
545 	uint64_t	p_msresv_2;	/* page allocation debugging */
546 #endif
547 } page_t;
548 
549 
550 typedef	page_t	devpage_t;
551 #define	devpage	page
552 
553 #define	PAGE_LOCK_MAXIMUM \
554 	((1 << (sizeof (((page_t *)0)->p_lckcnt) * NBBY)) - 1)
555 
556 #define	PAGE_SLOCK_MAXIMUM UINT_MAX
557 
558 /*
559  * Page hash table is a power-of-two in size, externally chained
560  * through the hash field.  PAGE_HASHAVELEN is the average length
561  * desired for this chain, from which the size of the page_hash
562  * table is derived at boot time and stored in the kernel variable
563  * page_hashsz.  In the hash function it is given by PAGE_HASHSZ.
564  *
565  * PAGE_HASH_FUNC returns an index into the page_hash[] array.  This
566  * index is also used to derive the mutex that protects the chain.
567  *
568  * In constructing the hash function, first we dispose of unimportant bits
569  * (page offset from "off" and the low 3 bits of "vp" which are zero for
570  * struct alignment). Then shift and sum the remaining bits a couple times
571  * in order to get as many source bits from the two source values into the
572  * resulting hashed value.  Note that this will perform quickly, since the
573  * shifting/summing are fast register to register operations with no additional
574  * memory references).
575  *
576  * PH_SHIFT_SIZE is the amount to use for the successive shifts in the hash
577  * function below.  The actual value is LOG2(PH_TABLE_SIZE), so that as many
578  * bits as possible will filter thru PAGE_HASH_FUNC() and PAGE_HASH_MUTEX().
579  *
580  * We use ? : instead of #if because <vm/page.h> is included everywhere;
581  * NCPU maps to a global variable outside of the "unix" module.
582  */
583 #if defined(_LP64)
584 #define	PH_SHIFT_SIZE	((NCPU < 4) ? 7		: (NCPU_LOG2 + 1))
585 #else	/* 32 bits */
586 #define	PH_SHIFT_SIZE	((NCPU < 4) ? 4		: 7)
587 #endif	/* _LP64 */
588 
589 #define	PH_TABLE_SIZE	(1ul << PH_SHIFT_SIZE)
590 
591 /*
592  *
593  * We take care to get as much randomness as possible from both the vp and
594  * the offset.  Workloads can have few vnodes with many offsets, many vnodes
595  * with few offsets or a moderate mix of both.  This hash should perform
596  * equally well for each of these possibilities and for all types of memory
597  * allocations.
598  *
599  * vnodes representing files are created over a long period of time and
600  * have good variation in the upper vp bits, and the right shifts below
601  * capture these bits.  However, swap vnodes are created quickly in a
602  * narrow vp* range.  Refer to comments at swap_alloc: vnum has exactly
603  * AN_VPSHIFT bits, so the kmem_alloc'd vnode addresses have approximately
604  * AN_VPSHIFT bits of variation above their VNODE_ALIGN low order 0 bits.
605  * Spread swap vnodes widely in the hash table by XOR'ing a term with the
606  * vp bits of variation left shifted to the top of the range.
607  */
608 
609 #define	PAGE_HASHSZ	page_hashsz
610 #define	PAGE_HASHAVELEN		4
611 #define	PAGE_HASH_FUNC(vp, off) \
612 	(((((uintptr_t)(off) >> PAGESHIFT) ^ \
613 	    ((uintptr_t)(off) >> (PAGESHIFT + PH_SHIFT_SIZE))) ^ \
614 	    (((uintptr_t)(vp) >> 3) ^ \
615 	    ((uintptr_t)(vp) >> (3 + PH_SHIFT_SIZE)) ^ \
616 	    ((uintptr_t)(vp) >> (3 + 2 * PH_SHIFT_SIZE)) ^ \
617 	    ((uintptr_t)(vp) << \
618 	    (page_hashsz_shift - AN_VPSHIFT - VNODE_ALIGN_LOG2)))) & \
619 	    (PAGE_HASHSZ - 1))
620 
621 #ifdef _KERNEL
622 
623 /*
624  * The page hash value is re-hashed to an index for the ph_mutex array.
625  *
626  * For 64 bit kernels, the mutex array is padded out to prevent false
627  * sharing of cache sub-blocks (64 bytes) of adjacent mutexes.
628  *
629  * For 32 bit kernels, we don't want to waste kernel address space with
630  * padding, so instead we rely on the hash function to introduce skew of
631  * adjacent vnode/offset indexes (the left shift part of the hash function).
632  * Since sizeof (kmutex_t) is 8, we shift an additional 3 to skew to a different
633  * 64 byte sub-block.
634  */
635 extern pad_mutex_t ph_mutex[];
636 
637 #define	PAGE_HASH_MUTEX(x) \
638 	&(ph_mutex[((x) ^ (((x) >> PH_SHIFT_SIZE) + ((x) << 3))) & \
639 		(PH_TABLE_SIZE - 1)].pad_mutex)
640 
641 /*
642  * Flags used while creating pages.
643  */
644 #define	PG_EXCL		0x0001
645 #define	PG_WAIT		0x0002		/* Blocking memory allocations */
646 #define	PG_PHYSCONTIG	0x0004		/* NOT SUPPORTED */
647 #define	PG_MATCH_COLOR	0x0008		/* SUPPORTED by free list routines */
648 #define	PG_NORELOC	0x0010		/* Non-relocatable alloc hint. */
649 					/* Page must be PP_ISNORELOC */
650 #define	PG_PANIC	0x0020		/* system will panic if alloc fails */
651 #define	PG_PUSHPAGE	0x0040		/* alloc may use reserve */
652 #define	PG_LOCAL	0x0080		/* alloc from given lgrp only */
653 #define	PG_NORMALPRI	0x0100		/* PG_WAIT like priority, but */
654 					/* non-blocking */
655 /*
656  * When p_selock has the SE_EWANTED bit set, threads waiting for SE_EXCL
657  * access are given priority over all other waiting threads.
658  */
659 #define	SE_EWANTED	0x40000000
660 #define	PAGE_LOCKED(pp)		(((pp)->p_selock & ~SE_EWANTED) != 0)
661 #define	PAGE_SHARED(pp)		(((pp)->p_selock & ~SE_EWANTED) > 0)
662 #define	PAGE_EXCL(pp)		((pp)->p_selock < 0)
663 #define	PAGE_LOCKED_SE(pp, se)	\
664 	((se) == SE_EXCL ? PAGE_EXCL(pp) : PAGE_SHARED(pp))
665 
666 extern	long page_hashsz;
667 extern	unsigned int page_hashsz_shift;
668 extern	page_t **page_hash;
669 
670 extern	pad_mutex_t page_llocks[];	/* page logical lock mutex */
671 extern	kmutex_t freemem_lock;		/* freemem lock */
672 
673 extern	pgcnt_t	total_pages;		/* total pages in the system */
674 
675 /*
676  * Variables controlling locking of physical memory.
677  */
678 extern	pgcnt_t	pages_pp_maximum;	/* tuning: lock + claim <= max */
679 extern	void init_pages_pp_maximum(void);
680 
681 struct lgrp;
682 
683 /* page_list_{add,sub} flags */
684 
685 /* which list */
686 #define	PG_FREE_LIST	0x0001
687 #define	PG_CACHE_LIST	0x0002
688 
689 /* where on list */
690 #define	PG_LIST_TAIL	0x0010
691 #define	PG_LIST_HEAD	0x0020
692 
693 /* called from */
694 #define	PG_LIST_ISINIT	0x1000
695 
696 /*
697  * Page frame operations.
698  */
699 page_t	*page_lookup(struct vnode *, u_offset_t, se_t);
700 page_t	*page_lookup_create(struct vnode *, u_offset_t, se_t, page_t *,
701 	spgcnt_t *, int);
702 page_t	*page_lookup_nowait(struct vnode *, u_offset_t, se_t);
703 page_t	*page_find(struct vnode *, u_offset_t);
704 page_t	*page_exists(struct vnode *, u_offset_t);
705 int	page_exists_physcontig(vnode_t *, u_offset_t, uint_t, page_t *[]);
706 int	page_exists_forreal(struct vnode *, u_offset_t, uint_t *);
707 void	page_needfree(spgcnt_t);
708 page_t	*page_create(struct vnode *, u_offset_t, size_t, uint_t);
709 int	page_alloc_pages(struct vnode *, struct seg *, caddr_t, page_t **,
710 	page_t **, uint_t, int, int);
711 page_t  *page_create_va_large(vnode_t *vp, u_offset_t off, size_t bytes,
712 	uint_t flags, struct seg *seg, caddr_t vaddr, void *arg);
713 page_t	*page_create_va(struct vnode *, u_offset_t, size_t, uint_t,
714 	struct seg *, caddr_t);
715 int	page_create_wait(pgcnt_t npages, uint_t flags);
716 void    page_create_putback(spgcnt_t npages);
717 void	page_free(page_t *, int);
718 void	page_free_at_startup(page_t *);
719 void	page_free_pages(page_t *);
720 void	free_vp_pages(struct vnode *, u_offset_t, size_t);
721 int	page_reclaim(page_t *, kmutex_t *);
722 int	page_reclaim_pages(page_t *, kmutex_t *, uint_t);
723 void	page_destroy(page_t *, int);
724 void	page_destroy_pages(page_t *);
725 void	page_destroy_free(page_t *);
726 void	page_rename(page_t *, struct vnode *, u_offset_t);
727 int	page_hashin(page_t *, struct vnode *, u_offset_t, kmutex_t *);
728 void	page_hashout(page_t *, kmutex_t *);
729 int	page_num_hashin(pfn_t, struct vnode *, u_offset_t);
730 void	page_add(page_t **, page_t *);
731 void	page_add_common(page_t **, page_t *);
732 void	page_sub(page_t **, page_t *);
733 void	page_sub_common(page_t **, page_t *);
734 page_t	*page_get_freelist(struct vnode *, u_offset_t, struct seg *,
735 		caddr_t, size_t, uint_t, struct lgrp *);
736 
737 page_t	*page_get_cachelist(struct vnode *, u_offset_t, struct seg *,
738 		caddr_t, uint_t, struct lgrp *);
739 #if defined(__i386) || defined(__amd64)
740 int	page_chk_freelist(uint_t);
741 #endif
742 void	page_list_add(page_t *, int);
743 void	page_boot_demote(page_t *);
744 void	page_promote_size(page_t *, uint_t);
745 void	page_list_add_pages(page_t *, int);
746 void	page_list_sub(page_t *, int);
747 void	page_list_sub_pages(page_t *, uint_t);
748 void	page_list_xfer(page_t *, int, int);
749 void	page_list_break(page_t **, page_t **, size_t);
750 void	page_list_concat(page_t **, page_t **);
751 void	page_vpadd(page_t **, page_t *);
752 void	page_vpsub(page_t **, page_t *);
753 int	page_lock(page_t *, se_t, kmutex_t *, reclaim_t);
754 int	page_lock_es(page_t *, se_t, kmutex_t *, reclaim_t, int);
755 void page_lock_clr_exclwanted(page_t *);
756 int	page_trylock(page_t *, se_t);
757 int	page_try_reclaim_lock(page_t *, se_t, int);
758 int	page_tryupgrade(page_t *);
759 void	page_downgrade(page_t *);
760 void	page_unlock(page_t *);
761 void	page_unlock_nocapture(page_t *);
762 void	page_lock_delete(page_t *);
763 int	page_deleted(page_t *);
764 int	page_pp_lock(page_t *, int, int);
765 void	page_pp_unlock(page_t *, int, int);
766 int	page_xresv(pgcnt_t, uint_t, int (*)(void));
767 int	page_resv(pgcnt_t, uint_t);
768 void	page_unresv(pgcnt_t);
769 void	page_pp_useclaim(page_t *, page_t *, uint_t);
770 int	page_addclaim(page_t *);
771 int	page_subclaim(page_t *);
772 int	page_addclaim_pages(page_t **);
773 int	page_subclaim_pages(page_t **);
774 pfn_t	page_pptonum(page_t *);
775 page_t	*page_numtopp(pfn_t, se_t);
776 page_t	*page_numtopp_noreclaim(pfn_t, se_t);
777 page_t	*page_numtopp_nolock(pfn_t);
778 page_t	*page_numtopp_nowait(pfn_t, se_t);
779 page_t  *page_first();
780 page_t  *page_next(page_t *);
781 page_t  *page_list_next(page_t *);
782 page_t	*page_nextn(page_t *, ulong_t);
783 page_t	*page_next_scan_init(void **);
784 page_t	*page_next_scan_large(page_t *, ulong_t *, void **);
785 void    prefetch_page_r(void *);
786 int	ppcopy(page_t *, page_t *);
787 void	page_relocate_hash(page_t *, page_t *);
788 void	pagezero(page_t *, uint_t, uint_t);
789 void	pagescrub(page_t *, uint_t, uint_t);
790 void	page_io_lock(page_t *);
791 void	page_io_unlock(page_t *);
792 int	page_io_trylock(page_t *);
793 int	page_iolock_assert(page_t *);
794 void	page_iolock_init(page_t *);
795 void	page_io_wait(page_t *);
796 int	page_io_locked(page_t *);
797 pgcnt_t	page_busy(int);
798 void	page_lock_init(void);
799 ulong_t	page_share_cnt(page_t *);
800 int	page_isshared(page_t *);
801 int	page_isfree(page_t *);
802 int	page_isref(page_t *);
803 int	page_ismod(page_t *);
804 int	page_release(page_t *, int);
805 void	page_retire_init(void);
806 int	page_retire(uint64_t, uchar_t);
807 int	page_retire_check(uint64_t, uint64_t *);
808 int	page_unretire(uint64_t);
809 int	page_unretire_pp(page_t *, int);
810 void	page_tryretire(page_t *);
811 void	page_retire_mdboot();
812 uint64_t	page_retire_pend_count(void);
813 uint64_t	page_retire_pend_kas_count(void);
814 void	page_retire_incr_pend_count(void *);
815 void	page_retire_decr_pend_count(void *);
816 void	page_clrtoxic(page_t *, uchar_t);
817 void	page_settoxic(page_t *, uchar_t);
818 
819 int	page_reclaim_mem(pgcnt_t, pgcnt_t, int);
820 
821 void page_set_props(page_t *, uint_t);
822 void page_clr_all_props(page_t *);
823 int page_clear_lck_cow(page_t *, int);
824 
825 kmutex_t	*page_vnode_mutex(struct vnode *);
826 kmutex_t	*page_se_mutex(struct page *);
827 kmutex_t	*page_szc_lock(struct page *);
828 int		page_szc_lock_assert(struct page *pp);
829 
830 /*
831  * Page relocation interfaces. page_relocate() is generic.
832  * page_get_replacement_page() is provided by the PSM.
833  * page_free_replacement_page() is generic.
834  */
835 int group_page_trylock(page_t *, se_t);
836 void group_page_unlock(page_t *);
837 int page_relocate(page_t **, page_t **, int, int, spgcnt_t *, struct lgrp *);
838 int do_page_relocate(page_t **, page_t **, int, spgcnt_t *, struct lgrp *);
839 page_t *page_get_replacement_page(page_t *, struct lgrp *, uint_t);
840 void page_free_replacement_page(page_t *);
841 int page_relocate_cage(page_t **, page_t **);
842 
843 int page_try_demote_pages(page_t *);
844 int page_try_demote_free_pages(page_t *);
845 void page_demote_free_pages(page_t *);
846 
847 struct anon_map;
848 
849 void page_mark_migrate(struct seg *, caddr_t, size_t, struct anon_map *,
850     ulong_t, vnode_t *, u_offset_t, int);
851 void page_migrate(struct seg *, caddr_t, page_t **, pgcnt_t);
852 
853 /*
854  * Tell the PIM we are adding physical memory
855  */
856 void add_physmem(page_t *, size_t, pfn_t);
857 void add_physmem_cb(page_t *, pfn_t);	/* callback for page_t part */
858 
859 /*
860  * hw_page_array[] is configured with hardware supported page sizes by
861  * platform specific code.
862  */
863 typedef struct {
864 	size_t	hp_size;
865 	uint_t	hp_shift;
866 	uint_t  hp_colors;
867 	pgcnt_t	hp_pgcnt;	/* base pagesize cnt */
868 } hw_pagesize_t;
869 
870 extern hw_pagesize_t	hw_page_array[];
871 extern uint_t		page_coloring_shift;
872 extern uint_t		page_colors_mask;
873 extern int		cpu_page_colors;
874 extern uint_t		colorequiv;
875 extern uchar_t		colorequivszc[];
876 
877 uint_t	page_num_pagesizes(void);
878 uint_t	page_num_user_pagesizes(int);
879 size_t	page_get_pagesize(uint_t);
880 size_t	page_get_user_pagesize(uint_t n);
881 pgcnt_t	page_get_pagecnt(uint_t);
882 uint_t	page_get_shift(uint_t);
883 int	page_szc(size_t);
884 int	page_szc_user_filtered(size_t);
885 
886 /* page_get_replacement page flags */
887 #define	PGR_SAMESZC	0x1	/* only look for page size same as orig */
888 #define	PGR_NORELOC	0x2	/* allocate a P_NORELOC page */
889 
890 /*
891  * macros for "masked arithmetic"
892  * The purpose is to step through all combinations of a set of bits while
893  * keeping some other bits fixed. Fixed bits need not be contiguous. The
894  * variable bits need not be contiguous either, or even right aligned. The
895  * trick is to set all fixed bits to 1, then increment, then restore the
896  * fixed bits. If incrementing causes a carry from a low bit position, the
897  * carry propagates thru the fixed bits, because they are temporarily set to 1.
898  *	v is the value
899  *	i is the increment
900  *	eq_mask defines the fixed bits
901  *	mask limits the size of the result
902  */
903 #define	ADD_MASKED(v, i, eq_mask, mask) \
904 	(((((v) | (eq_mask)) + (i)) & (mask) & ~(eq_mask)) | ((v) & (eq_mask)))
905 
906 /*
907  * convenience macro which increments by 1
908  */
909 #define	INC_MASKED(v, eq_mask, mask) ADD_MASKED(v, 1, eq_mask, mask)
910 
911 #endif	/* _KERNEL */
912 
913 /*
914  * Constants used for the p_iolock_state
915  */
916 #define	PAGE_IO_INUSE	0x1
917 #define	PAGE_IO_WANTED	0x2
918 
919 /*
920  * Constants used for page_release status
921  */
922 #define	PGREL_NOTREL    0x1
923 #define	PGREL_CLEAN	0x2
924 #define	PGREL_MOD	0x3
925 
926 /*
927  * The p_state field holds what used to be the p_age and p_free
928  * bits.  These fields are protected by p_selock (see above).
929  */
930 #define	P_FREE		0x80		/* Page on free list */
931 #define	P_NORELOC	0x40		/* Page is non-relocatable */
932 #define	P_MIGRATE	0x20		/* Migrate page on next touch */
933 #define	P_SWAP		0x10		/* belongs to vnode that is V_ISSWAP */
934 #define	P_BOOTPAGES	0x08		/* member of bootpages list */
935 #define	P_RAF		0x04		/* page retired at free */
936 
937 #define	PP_ISFREE(pp)		((pp)->p_state & P_FREE)
938 #define	PP_ISAGED(pp)		(((pp)->p_state & P_FREE) && \
939 					((pp)->p_vnode == NULL))
940 #define	PP_ISNORELOC(pp)	((pp)->p_state & P_NORELOC)
941 #define	PP_ISKAS(pp)		(VN_ISKAS((pp)->p_vnode))
942 #define	PP_ISNORELOCKERNEL(pp)	(PP_ISNORELOC(pp) && PP_ISKAS(pp))
943 #define	PP_ISMIGRATE(pp)	((pp)->p_state & P_MIGRATE)
944 #define	PP_ISSWAP(pp)		((pp)->p_state & P_SWAP)
945 #define	PP_ISBOOTPAGES(pp)	((pp)->p_state & P_BOOTPAGES)
946 #define	PP_ISRAF(pp)		((pp)->p_state & P_RAF)
947 
948 #define	PP_SETFREE(pp)		((pp)->p_state = ((pp)->p_state & ~P_MIGRATE) \
949 				| P_FREE)
950 #define	PP_SETAGED(pp)		ASSERT(PP_ISAGED(pp))
951 #define	PP_SETNORELOC(pp)	((pp)->p_state |= P_NORELOC)
952 #define	PP_SETMIGRATE(pp)	((pp)->p_state |= P_MIGRATE)
953 #define	PP_SETSWAP(pp)		((pp)->p_state |= P_SWAP)
954 #define	PP_SETBOOTPAGES(pp)	((pp)->p_state |= P_BOOTPAGES)
955 #define	PP_SETRAF(pp)		((pp)->p_state |= P_RAF)
956 
957 #define	PP_CLRFREE(pp)		((pp)->p_state &= ~P_FREE)
958 #define	PP_CLRAGED(pp)		ASSERT(!PP_ISAGED(pp))
959 #define	PP_CLRNORELOC(pp)	((pp)->p_state &= ~P_NORELOC)
960 #define	PP_CLRMIGRATE(pp)	((pp)->p_state &= ~P_MIGRATE)
961 #define	PP_CLRSWAP(pp)		((pp)->p_state &= ~P_SWAP)
962 #define	PP_CLRBOOTPAGES(pp)	((pp)->p_state &= ~P_BOOTPAGES)
963 #define	PP_CLRRAF(pp)		((pp)->p_state &= ~P_RAF)
964 
965 /*
966  * Flags for page_t p_toxic, for tracking memory hardware errors.
967  *
968  * These flags are OR'ed into p_toxic with page_settoxic() to track which
969  * error(s) have occurred on a given page. The flags are cleared with
970  * page_clrtoxic(). Both page_settoxic() and page_cleartoxic use atomic
971  * primitives to manipulate the p_toxic field so no other locking is needed.
972  *
973  * When an error occurs on a page, p_toxic is set to record the error. The
974  * error could be a memory error or something else (i.e. a datapath). The Page
975  * Retire mechanism does not try to determine the exact cause of the error;
976  * Page Retire rightly leaves that sort of determination to FMA's Diagnostic
977  * Engine (DE).
978  *
979  * Note that, while p_toxic bits can be set without holding any locks, they
980  * should only be cleared while holding the page exclusively locked.
981  * There is one exception to this, the PR_CAPTURE bit is protected by a mutex
982  * within the page capture logic and thus to set or clear the bit, that mutex
983  * needs to be held.  The page does not need to be locked but the page_clrtoxic
984  * function must be used as we need an atomic operation.
985  * Also note that there is what amounts to a hack to prevent recursion with
986  * large pages such that if we are unlocking a page and the PR_CAPTURE bit is
987  * set, we will only try to capture the page if the current threads T_CAPTURING
988  * flag is not set.  If the flag is set, the unlock will not try to capture
989  * the page even though the PR_CAPTURE bit is set.
990  *
991  * Pages with PR_UE or PR_FMA flags are retired unconditionally, while pages
992  * with PR_MCE are retired if the system has not retired too many of them.
993  *
994  * A page must be exclusively locked to be retired. Pages can be retired if
995  * they are mapped, modified, or both, as long as they are not marked PR_UE,
996  * since pages with uncorrectable errors cannot be relocated in memory.
997  * Once a page has been successfully retired it is zeroed, attached to the
998  * retired_pages vnode and, finally, PR_RETIRED is set in p_toxic. The other
999  * p_toxic bits are NOT cleared. Pages are not left locked after retiring them
1000  * to avoid special case code throughout the kernel; rather, page_*lock() will
1001  * fail to lock the page, unless SE_RETIRED is passed as an argument.
1002  *
1003  * While we have your attention, go take a look at the comments at the
1004  * beginning of page_retire.c too.
1005  */
1006 #define	PR_OK		0x00	/* no problem */
1007 #define	PR_MCE		0x01	/* page has seen two or more CEs */
1008 #define	PR_UE		0x02	/* page has an unhandled UE */
1009 #define	PR_UE_SCRUBBED	0x04	/* page has seen a UE but was cleaned */
1010 #define	PR_FMA		0x08	/* A DE wants this page retired */
1011 #define	PR_CAPTURE	0x10	/* page is hashed on page_capture_hash[] */
1012 #define	PR_RESV		0x20	/* Reserved for future use */
1013 #define	PR_MSG		0x40	/* message(s) already printed for this page */
1014 #define	PR_RETIRED	0x80	/* This page has been retired */
1015 
1016 #define	PR_REASONS	(PR_UE | PR_MCE | PR_FMA)
1017 #define	PR_TOXIC	(PR_UE)
1018 #define	PR_ERRMASK	(PR_UE | PR_UE_SCRUBBED | PR_MCE | PR_FMA)
1019 #define	PR_TOXICFLAGS	(0xCF)
1020 
1021 #define	PP_RETIRED(pp)	((pp)->p_toxic & PR_RETIRED)
1022 #define	PP_TOXIC(pp)	((pp)->p_toxic & PR_TOXIC)
1023 #define	PP_PR_REQ(pp)	(((pp)->p_toxic & PR_REASONS) && !PP_RETIRED(pp))
1024 #define	PP_PR_NOSHARE(pp)						\
1025 	((((pp)->p_toxic & (PR_RETIRED | PR_FMA | PR_UE)) == PR_FMA) &&	\
1026 	!PP_ISKAS(pp))
1027 
1028 /*
1029  * Flags for page_unretire_pp
1030  */
1031 #define	PR_UNR_FREE	0x1
1032 #define	PR_UNR_CLEAN	0x2
1033 #define	PR_UNR_TEMP	0x4
1034 
1035 /*
1036  * kpm large page description.
1037  * The virtual address range of segkpm is divided into chunks of
1038  * kpm_pgsz. Each chunk is controlled by a kpm_page_t. The ushort
1039  * is sufficient for 2^^15 * PAGESIZE, so e.g. the maximum kpm_pgsz
1040  * for 8K is 256M and 2G for 64K pages. It it kept as small as
1041  * possible to save physical memory space.
1042  *
1043  * There are 2 segkpm mapping windows within in the virtual address
1044  * space when we have to prevent VAC alias conflicts. The so called
1045  * Alias window (mappings are always by PAGESIZE) is controlled by
1046  * kp_refcnta. The regular window is controlled by kp_refcnt for the
1047  * normal operation, which is to use the largest available pagesize.
1048  * When VAC alias conflicts are present within a chunk in the regular
1049  * window the large page mapping is broken up into smaller PAGESIZE
1050  * mappings. kp_refcntc is used to control the pages that are invoked
1051  * in the conflict and kp_refcnts holds the active mappings done
1052  * with the small page size. In non vac conflict mode kp_refcntc is
1053  * also used as "go" indication (-1) for the trap level tsbmiss
1054  * handler.
1055  */
1056 typedef struct kpm_page {
1057 	short kp_refcnt;	/* pages mapped large */
1058 	short kp_refcnta;	/* pages mapped in Alias window */
1059 	short kp_refcntc;	/* TL-tsbmiss flag; #vac alias conflict pages */
1060 	short kp_refcnts;	/* vac alias: pages mapped small */
1061 } kpm_page_t;
1062 
1063 /*
1064  * Note: khl_lock offset changes must be reflected in sfmmu_asm.s
1065  */
1066 typedef struct kpm_hlk {
1067 	kmutex_t khl_mutex;	/* kpm_page mutex */
1068 	uint_t   khl_lock;	/* trap level tsbmiss handling */
1069 } kpm_hlk_t;
1070 
1071 /*
1072  * kpm small page description.
1073  * When kpm_pgsz is equal to PAGESIZE a smaller representation is used
1074  * to save memory space. Alias range mappings and regular segkpm
1075  * mappings are done in units of PAGESIZE and can share the mapping
1076  * information and the mappings are always distinguishable by their
1077  * virtual address. Other information needed for VAC conflict prevention
1078  * is already available on a per page basis.
1079  *
1080  * The state about how a kpm page is mapped and whether it is ready to go
1081  * is indicated by the following 1 byte kpm_spage structure. This byte is
1082  * split into two 4-bit parts - kp_mapped and kp_mapped_go.
1083  *	- kp_mapped == 1	the page is mapped cacheable
1084  *	- kp_mapped == 2	the page is mapped non-cacheable
1085  *	- kp_mapped_go == 1	the mapping is ready to be dropped in
1086  *	- kp_mapped_go == 0	the mapping is not ready to be dropped in.
1087  * When kp_mapped_go == 0, we will have C handler resolve the VAC conflict.
1088  * Otherwise, the assembly tsb miss handler can simply drop in the mapping
1089  * when a tsb miss occurs.
1090  */
1091 typedef union kpm_spage {
1092 	struct {
1093 #ifdef  _BIG_ENDIAN
1094 		uchar_t mapped_go: 4;	/* go or nogo flag */
1095 		uchar_t mapped: 4;	/* page mapped small */
1096 #else
1097 		uchar_t mapped: 4;	/* page mapped small */
1098 		uchar_t mapped_go: 4;	/* go or nogo flag */
1099 #endif
1100 	} kpm_spage_un;
1101 	uchar_t kp_mapped_flag;
1102 } kpm_spage_t;
1103 
1104 #define	kp_mapped	kpm_spage_un.mapped
1105 #define	kp_mapped_go	kpm_spage_un.mapped_go
1106 
1107 /*
1108  * Note: kshl_lock offset changes must be reflected in sfmmu_asm.s
1109  */
1110 typedef struct kpm_shlk {
1111 	uint_t   kshl_lock;	/* trap level tsbmiss handling */
1112 } kpm_shlk_t;
1113 
1114 /*
1115  * Each segment of physical memory is described by a memseg struct.
1116  * Within a segment, memory is considered contiguous. The members
1117  * can be categorized as follows:
1118  * . Platform independent:
1119  *         pages, epages, pages_base, pages_end, next, lnext.
1120  * . 64bit only but platform independent:
1121  *         kpm_pbase, kpm_nkpmpgs, kpm_pages, kpm_spages.
1122  * . Really platform or mmu specific:
1123  *         pagespa, epagespa, nextpa, kpm_pagespa.
1124  * . Mixed:
1125  *         msegflags.
1126  */
1127 struct memseg {
1128 	page_t *pages, *epages;		/* [from, to] in page array */
1129 	pfn_t pages_base, pages_end;	/* [from, to] in page numbers */
1130 	struct memseg *next;		/* next segment in list */
1131 	struct memseg *lnext;		/* next segment in deleted list */
1132 #if defined(__sparc)
1133 	uint64_t pagespa, epagespa;	/* [from, to] page array physical */
1134 	uint64_t nextpa;		/* physical next pointer */
1135 	pfn_t	kpm_pbase;		/* start of kpm range */
1136 	pgcnt_t kpm_nkpmpgs;		/* # of kpm_pgsz pages */
1137 	union _mseg_un {
1138 		kpm_page_t  *kpm_lpgs;	/* ptr to kpm_page array */
1139 		kpm_spage_t *kpm_spgs;	/* ptr to kpm_spage array */
1140 	} mseg_un;
1141 	uint64_t kpm_pagespa;		/* physical ptr to kpm (s)pages array */
1142 #endif /* __sparc */
1143 	uint_t msegflags;		/* memseg flags */
1144 };
1145 
1146 /* memseg union aliases */
1147 #define	kpm_pages	mseg_un.kpm_lpgs
1148 #define	kpm_spages	mseg_un.kpm_spgs
1149 
1150 /* msegflags */
1151 #define	MEMSEG_DYNAMIC		0x1	/* DR: memory was added dynamically */
1152 #define	MEMSEG_META_INCL	0x2	/* DR: memseg includes it's metadata */
1153 #define	MEMSEG_META_ALLOC	0x4	/* DR: memseg allocated it's metadata */
1154 
1155 /* memseg support macros */
1156 #define	MSEG_NPAGES(SEG)	((SEG)->pages_end - (SEG)->pages_base)
1157 
1158 /* memseg hash */
1159 #define	MEM_HASH_SHIFT		0x9
1160 #define	N_MEM_SLOTS		0x200		/* must be a power of 2 */
1161 #define	MEMSEG_PFN_HASH(pfn)	(((pfn)/mhash_per_slot) & (N_MEM_SLOTS - 1))
1162 
1163 /* memseg  externals */
1164 extern struct memseg *memsegs;		/* list of memory segments */
1165 extern ulong_t mhash_per_slot;
1166 extern uint64_t memsegspa;		/* memsegs as physical address */
1167 
1168 void build_pfn_hash();
1169 extern struct memseg *page_numtomemseg_nolock(pfn_t pfnum);
1170 
1171 /*
1172  * page capture related info:
1173  * The page capture routines allow us to asynchronously capture given pages
1174  * for the explicit use of the requestor.  New requestors can be added by
1175  * explicitly adding themselves to the PC_* flags below and incrementing
1176  * PC_NUM_CALLBACKS as necessary.
1177  *
1178  * Subsystems using page capture must register a callback before attempting
1179  * to capture a page.  A duration of -1 will indicate that we will never give
1180  * up while trying to capture a page and will only stop trying to capture the
1181  * given page once we have successfully captured it.  Thus the user needs to be
1182  * aware of the behavior of all callers who have a duration of -1.
1183  *
1184  * For now, only /dev/physmem and page retire use the page capture interface
1185  * and only a single request can be outstanding for a given page.  Thus, if
1186  * /dev/phsymem wants a page and page retire also wants the same page, only
1187  * the page retire request will be honored until the point in time that the
1188  * page is actually retired, at which point in time, subsequent requests by
1189  * /dev/physmem will succeed if the CAPTURE_GET_RETIRED flag was set.
1190  */
1191 
1192 #define	PC_RETIRE		(0)
1193 #define	PC_PHYSMEM		(1)
1194 #define	PC_NUM_CALLBACKS	(2)
1195 #define	PC_MASK			((1 << PC_NUM_CALLBACKS) - 1)
1196 
1197 #define	CAPTURE_RETIRE		(1 << PC_RETIRE)
1198 #define	CAPTURE_PHYSMEM		(1 << PC_PHYSMEM)
1199 
1200 #define	CAPTURE_ASYNC		(0x0200)
1201 
1202 #define	CAPTURE_GET_RETIRED	(0x1000)
1203 #define	CAPTURE_GET_CAGE	(0x2000)
1204 
1205 struct page_capture_callback {
1206 	int cb_active;		/* 1 means active, 0 means inactive */
1207 	clock_t duration;	/* the length in time that we'll attempt to */
1208 				/* capture this page asynchronously. (in HZ) */
1209 	krwlock_t cb_rwlock;
1210 	int (*cb_func)(page_t *, void *, uint_t); /* callback function */
1211 };
1212 
1213 extern kcondvar_t pc_cv;
1214 
1215 void page_capture_register_callback(uint_t index, clock_t duration,
1216     int (*cb_func)(page_t *, void *, uint_t));
1217 void page_capture_unregister_callback(uint_t index);
1218 int page_trycapture(page_t *pp, uint_t szc, uint_t flags, void *datap);
1219 void page_unlock_capture(page_t *pp);
1220 int page_capture_unretire_pp(page_t *);
1221 
1222 extern int memsegs_trylock(int);
1223 extern void memsegs_lock(int);
1224 extern void memsegs_unlock(int);
1225 extern int memsegs_lock_held(void);
1226 extern void memlist_read_lock(void);
1227 extern void memlist_read_unlock(void);
1228 extern void memlist_write_lock(void);
1229 extern void memlist_write_unlock(void);
1230 
1231 #ifdef	__cplusplus
1232 }
1233 #endif
1234 
1235 #endif	/* _VM_PAGE_H */
1236