xref: /titanic_41/usr/src/uts/common/vm/page.h (revision 5203bc321053fb87d7073c7640548fab73634793)
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  */
24 
25 /*	Copyright (c) 1984, 1986, 1987, 1988, 1989 AT&T	*/
26 /*	  All Rights Reserved  	*/
27 
28 /*
29  * University Copyright- Copyright (c) 1982, 1986, 1988
30  * The Regents of the University of California
31  * All Rights Reserved
32  *
33  * University Acknowledgment- Portions of this document are derived from
34  * software developed by the University of California, Berkeley, and its
35  * contributors.
36  */
37 
38 #ifndef	_VM_PAGE_H
39 #define	_VM_PAGE_H
40 
41 #include <vm/seg.h>
42 
43 #ifdef	__cplusplus
44 extern "C" {
45 #endif
46 
47 #if defined(_KERNEL) || defined(_KMEMUSER)
48 
49 /*
50  * Shared/Exclusive lock.
51  */
52 
53 /*
54  * Types of page locking supported by page_lock & friends.
55  */
56 typedef enum {
57 	SE_SHARED,
58 	SE_EXCL			/* exclusive lock (value == -1) */
59 } se_t;
60 
61 /*
62  * For requesting that page_lock reclaim the page from the free list.
63  */
64 typedef enum {
65 	P_RECLAIM,		/* reclaim page from free list */
66 	P_NO_RECLAIM		/* DON`T reclaim the page	*/
67 } reclaim_t;
68 
69 /*
70  * Callers of page_try_reclaim_lock and page_lock_es can use this flag
71  * to get SE_EXCL access before reader/writers are given access.
72  */
73 #define	SE_EXCL_WANTED	0x02
74 
75 /*
76  * All page_*lock() requests will be denied unless this flag is set in
77  * the 'es' parameter.
78  */
79 #define	SE_RETIRED	0x04
80 
81 #endif	/* _KERNEL | _KMEMUSER */
82 
83 typedef int	selock_t;
84 
85 /*
86  * Define VM_STATS to turn on all sorts of statistic gathering about
87  * the VM layer.  By default, it is only turned on when DEBUG is
88  * also defined.
89  */
90 #ifdef DEBUG
91 #define	VM_STATS
92 #endif	/* DEBUG */
93 
94 #ifdef VM_STATS
95 #define	VM_STAT_ADD(stat)			(stat)++
96 #define	VM_STAT_COND_ADD(cond, stat)		((void) (!(cond) || (stat)++))
97 #else
98 #define	VM_STAT_ADD(stat)
99 #define	VM_STAT_COND_ADD(cond, stat)
100 #endif	/* VM_STATS */
101 
102 #ifdef _KERNEL
103 
104 /*
105  * PAGE_LLOCK_SIZE is 2 * NCPU, but no smaller than 128.
106  * PAGE_LLOCK_SHIFT is log2(PAGE_LLOCK_SIZE).
107  */
108 #if ((2*NCPU_P2) > 128)
109 #define	PAGE_LLOCK_SHIFT	((unsigned)(NCPU_LOG2 + 1))
110 #else
111 #define	PAGE_LLOCK_SHIFT	7U
112 #endif
113 #define	PAGE_LLOCK_SIZE (1 << PAGE_LLOCK_SHIFT)
114 
115 /*
116  * The number of low order 0 (or less variable) bits in the page_t address.
117  */
118 #if defined(__sparc)
119 #define	PP_SHIFT		7
120 #else
121 #define	PP_SHIFT		6
122 #endif
123 
124 /*
125  * pp may be the root of a large page, and many low order bits will be 0.
126  * Shift and XOR multiple times to capture the good bits across the range of
127  * possible page sizes.
128  */
129 #define	PAGE_LLOCK_HASH(pp)	\
130 	(((((uintptr_t)(pp) >> PP_SHIFT) ^ \
131 	((uintptr_t)(pp) >> (PAGE_LLOCK_SHIFT + PP_SHIFT))) ^ \
132 	((uintptr_t)(pp) >> ((PAGE_LLOCK_SHIFT * 2) + PP_SHIFT)) ^ \
133 	((uintptr_t)(pp) >> ((PAGE_LLOCK_SHIFT * 3) + PP_SHIFT))) & \
134 	(PAGE_LLOCK_SIZE - 1))
135 
136 #define	page_struct_lock(pp)	\
137 	mutex_enter(&page_llocks[PAGE_LLOCK_HASH(PP_PAGEROOT(pp))].pad_mutex)
138 #define	page_struct_unlock(pp)	\
139 	mutex_exit(&page_llocks[PAGE_LLOCK_HASH(PP_PAGEROOT(pp))].pad_mutex)
140 
141 #endif	/* _KERNEL */
142 
143 #include <sys/t_lock.h>
144 
145 struct as;
146 
147 /*
148  * Each physical page has a page structure, which is used to maintain
149  * these pages as a cache.  A page can be found via a hashed lookup
150  * based on the [vp, offset].  If a page has an [vp, offset] identity,
151  * then it is entered on a doubly linked circular list off the
152  * vnode using the vpnext/vpprev pointers.   If the p_free bit
153  * is on, then the page is also on a doubly linked circular free
154  * list using next/prev pointers.  If the "p_selock" and "p_iolock"
155  * are held, then the page is currently being read in (exclusive p_selock)
156  * or written back (shared p_selock).  In this case, the next/prev pointers
157  * are used to link the pages together for a consecutive i/o request.  If
158  * the page is being brought in from its backing store, then other processes
159  * will wait for the i/o to complete before attaching to the page since it
160  * will have an "exclusive" lock.
161  *
162  * Each page structure has the locks described below along with
163  * the fields they protect:
164  *
165  *	p_selock	This is a per-page shared/exclusive lock that is
166  *			used to implement the logical shared/exclusive
167  *			lock for each page.  The "shared" lock is normally
168  *			used in most cases while the "exclusive" lock is
169  *			required to destroy or retain exclusive access to
170  *			a page (e.g., while reading in pages).  The appropriate
171  *			lock is always held whenever there is any reference
172  *			to a page structure (e.g., during i/o).
173  *			(Note that with the addition of the "writer-lock-wanted"
174  *			semantics (via SE_EWANTED), threads must not acquire
175  *			multiple reader locks or else a deadly embrace will
176  *			occur in the following situation: thread 1 obtains a
177  *			reader lock; next thread 2 fails to get a writer lock
178  *			but specified SE_EWANTED so it will wait by either
179  *			blocking (when using page_lock_es) or spinning while
180  *			retrying (when using page_try_reclaim_lock) until the
181  *			reader lock is released; then thread 1 attempts to
182  *			get another reader lock but is denied due to
183  *			SE_EWANTED being set, and now both threads are in a
184  *			deadly embrace.)
185  *
186  *				p_hash
187  *				p_vnode
188  *				p_offset
189  *
190  *				p_free
191  *				p_age
192  *
193  *	p_iolock	This is a binary semaphore lock that provides
194  *			exclusive access to the i/o list links in each
195  *			page structure.  It is always held while the page
196  *			is on an i/o list (i.e., involved in i/o).  That is,
197  *			even though a page may be only `shared' locked
198  *			while it is doing a write, the following fields may
199  *			change anyway.  Normally, the page must be
200  *			`exclusively' locked to change anything in it.
201  *
202  *				p_next
203  *				p_prev
204  *
205  * The following fields are protected by the global page_llocks[]:
206  *
207  *				p_lckcnt
208  *				p_cowcnt
209  *
210  * The following lists are protected by the global page_freelock:
211  *
212  *				page_cachelist
213  *				page_freelist
214  *
215  * The following, for our purposes, are protected by
216  * the global freemem_lock:
217  *
218  *				freemem
219  *				freemem_wait
220  *				freemem_cv
221  *
222  * The following fields are protected by hat layer lock(s).  When a page
223  * structure is not mapped and is not associated with a vnode (after a call
224  * to page_hashout() for example) the p_nrm field may be modified with out
225  * holding the hat layer lock:
226  *
227  *				p_nrm
228  *				p_mapping
229  *				p_share
230  *
231  * The following field is file system dependent.  How it is used and
232  * the locking strategies applied are up to the individual file system
233  * implementation.
234  *
235  *				p_fsdata
236  *
237  * The page structure is used to represent and control the system's
238  * physical pages.  There is one instance of the structure for each
239  * page that is not permenately allocated.  For example, the pages that
240  * hold the page structures are permanently held by the kernel
241  * and hence do not need page structures to track them.  The array
242  * of page structures is allocated early on in the kernel's life and
243  * is based on the amount of available physical memory.
244  *
245  * Each page structure may simultaneously appear on several linked lists.
246  * The lists are:  hash list, free or in i/o list, and a vnode's page list.
247  * Each type of list is protected by a different group of mutexes as described
248  * below:
249  *
250  * The hash list is used to quickly find a page when the page's vnode and
251  * offset within the vnode are known.  Each page that is hashed is
252  * connected via the `p_hash' field.  The anchor for each hash is in the
253  * array `page_hash'.  An array of mutexes, `ph_mutex', protects the
254  * lists anchored by page_hash[].  To either search or modify a given hash
255  * list, the appropriate mutex in the ph_mutex array must be held.
256  *
257  * The free list contains pages that are `free to be given away'.  For
258  * efficiency reasons, pages on this list are placed in two catagories:
259  * pages that are still associated with a vnode, and pages that are not
260  * associated with a vnode.  Free pages always have their `p_free' bit set,
261  * free pages that are still associated with a vnode also have their
262  * `p_age' bit set.  Pages on the free list are connected via their
263  * `p_next' and `p_prev' fields.  When a page is involved in some sort
264  * of i/o, it is not free and these fields may be used to link associated
265  * pages together.  At the moment, the free list is protected by a
266  * single mutex `page_freelock'.  The list of free pages still associated
267  * with a vnode is anchored by `page_cachelist' while other free pages
268  * are anchored in architecture dependent ways (to handle page coloring etc.).
269  *
270  * Pages associated with a given vnode appear on a list anchored in the
271  * vnode by the `v_pages' field.  They are linked together with
272  * `p_vpnext' and `p_vpprev'.  The field `p_offset' contains a page's
273  * offset within the vnode.  The pages on this list are not kept in
274  * offset order.  These lists, in a manner similar to the hash lists,
275  * are protected by an array of mutexes called `vph_hash'.  Before
276  * searching or modifying this chain the appropriate mutex in the
277  * vph_hash[] array must be held.
278  *
279  * Again, each of the lists that a page can appear on is protected by a
280  * mutex.  Before reading or writing any of the fields comprising the
281  * list, the appropriate lock must be held.  These list locks should only
282  * be held for very short intervals.
283  *
284  * In addition to the list locks, each page structure contains a
285  * shared/exclusive lock that protects various fields within it.
286  * To modify one of these fields, the `p_selock' must be exclusively held.
287  * To read a field with a degree of certainty, the lock must be at least
288  * held shared.
289  *
290  * Removing a page structure from one of the lists requires holding
291  * the appropriate list lock and the page's p_selock.  A page may be
292  * prevented from changing identity, being freed, or otherwise modified
293  * by acquiring p_selock shared.
294  *
295  * To avoid deadlocks, a strict locking protocol must be followed.  Basically
296  * there are two cases:  In the first case, the page structure in question
297  * is known ahead of time (e.g., when the page is to be added or removed
298  * from a list).  In the second case, the page structure is not known and
299  * must be found by searching one of the lists.
300  *
301  * When adding or removing a known page to one of the lists, first the
302  * page must be exclusively locked (since at least one of its fields
303  * will be modified), second the lock protecting the list must be acquired,
304  * third the page inserted or deleted, and finally the list lock dropped.
305  *
306  * The more interesting case occures when the particular page structure
307  * is not known ahead of time.  For example, when a call is made to
308  * page_lookup(), it is not known if a page with the desired (vnode and
309  * offset pair) identity exists.  So the appropriate mutex in ph_mutex is
310  * acquired, the hash list searched, and if the desired page is found
311  * an attempt is made to lock it.  The attempt to acquire p_selock must
312  * not block while the hash list lock is held.  A deadlock could occure
313  * if some other process was trying to remove the page from the list.
314  * The removing process (following the above protocol) would have exclusively
315  * locked the page, and be spinning waiting to acquire the lock protecting
316  * the hash list.  Since the searching process holds the hash list lock
317  * and is waiting to acquire the page lock, a deadlock occurs.
318  *
319  * The proper scheme to follow is: first, lock the appropriate list,
320  * search the list, and if the desired page is found either use
321  * page_trylock() (which will not block) or pass the address of the
322  * list lock to page_lock().  If page_lock() can not acquire the page's
323  * lock, it will drop the list lock before going to sleep.  page_lock()
324  * returns a value to indicate if the list lock was dropped allowing the
325  * calling program to react appropriately (i.e., retry the operation).
326  *
327  * If the list lock was dropped before the attempt at locking the page
328  * was made, checks would have to be made to ensure that the page had
329  * not changed identity before its lock was obtained.  This is because
330  * the interval between dropping the list lock and acquiring the page
331  * lock is indeterminate.
332  *
333  * In addition, when both a hash list lock (ph_mutex[]) and a vnode list
334  * lock (vph_mutex[]) are needed, the hash list lock must be acquired first.
335  * The routine page_hashin() is a good example of this sequence.
336  * This sequence is ASSERTed by checking that the vph_mutex[] is not held
337  * just before each acquisition of one of the mutexs in ph_mutex[].
338  *
339  * So, as a quick summary:
340  *
341  * 	pse_mutex[]'s protect the p_selock and p_cv fields.
342  *
343  * 	p_selock protects the p_free, p_age, p_vnode, p_offset and p_hash,
344  *
345  * 	ph_mutex[]'s protect the page_hash[] array and its chains.
346  *
347  * 	vph_mutex[]'s protect the v_pages field and the vp page chains.
348  *
349  *	First lock the page, then the hash chain, then the vnode chain.  When
350  *	this is not possible `trylocks' must be used.  Sleeping while holding
351  *	any of these mutexes (p_selock is not a mutex) is not allowed.
352  *
353  *
354  *	field		reading		writing		    ordering
355  *	======================================================================
356  *	p_vnode		p_selock(E,S)	p_selock(E)
357  *	p_offset
358  *	p_free
359  *	p_age
360  *	=====================================================================
361  *	p_hash		p_selock(E,S)	p_selock(E) &&	    p_selock, ph_mutex
362  *					ph_mutex[]
363  *	=====================================================================
364  *	p_vpnext	p_selock(E,S)	p_selock(E) &&	    p_selock, vph_mutex
365  *	p_vpprev			vph_mutex[]
366  *	=====================================================================
367  *	When the p_free bit is set:
368  *
369  *	p_next		p_selock(E,S)	p_selock(E) &&	    p_selock,
370  *	p_prev				page_freelock	    page_freelock
371  *
372  *	When the p_free bit is not set:
373  *
374  *	p_next		p_selock(E,S)	p_selock(E) &&	    p_selock, p_iolock
375  *	p_prev				p_iolock
376  *	=====================================================================
377  *	p_selock	pse_mutex[]	pse_mutex[]	    can`t acquire any
378  *	p_cv						    other mutexes or
379  *							    sleep while holding
380  *							    this lock.
381  *	=====================================================================
382  *	p_lckcnt	p_selock(E,S)	p_selock(E)
383  *					    OR
384  *					p_selock(S) &&
385  *					page_llocks[]
386  *	p_cowcnt
387  *	=====================================================================
388  *	p_nrm		hat layer lock	hat layer lock
389  *	p_mapping
390  *	p_pagenum
391  *	=====================================================================
392  *
393  *	where:
394  *		E----> exclusive version of p_selock.
395  *		S----> shared version of p_selock.
396  *
397  *
398  *	Global data structures and variable:
399  *
400  *	field		reading		writing		    ordering
401  *	=====================================================================
402  *	page_hash[]	ph_mutex[]	ph_mutex[]	    can hold this lock
403  *							    before acquiring
404  *							    a vph_mutex or
405  *							    pse_mutex.
406  *	=====================================================================
407  *	vp->v_pages	vph_mutex[]	vph_mutex[]	    can only acquire
408  *							    a pse_mutex while
409  *							    holding this lock.
410  *	=====================================================================
411  *	page_cachelist	page_freelock	page_freelock	    can't acquire any
412  *	page_freelist	page_freelock	page_freelock
413  *	=====================================================================
414  *	freemem		freemem_lock	freemem_lock	    can't acquire any
415  *	freemem_wait					    other mutexes while
416  *	freemem_cv					    holding this mutex.
417  *	=====================================================================
418  *
419  * Page relocation, PG_NORELOC and P_NORELOC.
420  *
421  * Pages may be relocated using the page_relocate() interface. Relocation
422  * involves moving the contents and identity of a page to another, free page.
423  * To relocate a page, the SE_EXCL lock must be obtained. The way to prevent
424  * a page from being relocated is to hold the SE_SHARED lock (the SE_EXCL
425  * lock must not be held indefinitely). If the page is going to be held
426  * SE_SHARED indefinitely, then the PG_NORELOC hint should be passed
427  * to page_create_va so that pages that are prevented from being relocated
428  * can be managed differently by the platform specific layer.
429  *
430  * Pages locked in memory using page_pp_lock (p_lckcnt/p_cowcnt != 0)
431  * are guaranteed to be held in memory, but can still be relocated
432  * providing the SE_EXCL lock can be obtained.
433  *
434  * The P_NORELOC bit in the page_t.p_state field is provided for use by
435  * the platform specific code in managing pages when the PG_NORELOC
436  * hint is used.
437  *
438  * Memory delete and page locking.
439  *
440  * The set of all usable pages is managed using the global page list as
441  * implemented by the memseg structure defined below. When memory is added
442  * or deleted this list changes. Additions to this list guarantee that the
443  * list is never corrupt.  In order to avoid the necessity of an additional
444  * lock to protect against failed accesses to the memseg being deleted and,
445  * more importantly, the page_ts, the memseg structure is never freed and the
446  * page_t virtual address space is remapped to a page (or pages) of
447  * zeros.  If a page_t is manipulated while it is p_selock'd, or if it is
448  * locked indirectly via a hash or freelist lock, it is not possible for
449  * memory delete to collect the page and so that part of the page list is
450  * prevented from being deleted. If the page is referenced outside of one
451  * of these locks, it is possible for the page_t being referenced to be
452  * deleted.  Examples of this are page_t pointers returned by
453  * page_numtopp_nolock, page_first and page_next.  Providing the page_t
454  * is re-checked after taking the p_selock (for p_vnode != NULL), the
455  * remapping to the zero pages will be detected.
456  *
457  *
458  * Page size (p_szc field) and page locking.
459  *
460  * p_szc field of free pages is changed by free list manager under freelist
461  * locks and is of no concern to the rest of VM subsystem.
462  *
463  * p_szc changes of allocated anonymous (swapfs) can only be done only after
464  * exclusively locking all constituent pages and calling hat_pageunload() on
465  * each of them. To prevent p_szc changes of non free anonymous (swapfs) large
466  * pages it's enough to either lock SHARED any of constituent pages or prevent
467  * hat_pageunload() by holding hat level lock that protects mapping lists (this
468  * method is for hat code only)
469  *
470  * To increase (promote) p_szc of allocated non anonymous file system pages
471  * one has to first lock exclusively all involved constituent pages and call
472  * hat_pageunload() on each of them. To prevent p_szc promote it's enough to
473  * either lock SHARED any of constituent pages that will be needed to make a
474  * large page or prevent hat_pageunload() by holding hat level lock that
475  * protects mapping lists (this method is for hat code only).
476  *
477  * To decrease (demote) p_szc of an allocated non anonymous file system large
478  * page one can either use the same method as used for changeing p_szc of
479  * anonymous large pages or if it's not possible to lock all constituent pages
480  * exclusively a different method can be used. In the second method one only
481  * has to exclusively lock one of constituent pages but then one has to
482  * acquire further locks by calling page_szc_lock() and
483  * hat_page_demote(). hat_page_demote() acquires hat level locks and then
484  * demotes the page. This mechanism relies on the fact that any code that
485  * needs to prevent p_szc of a file system large page from changeing either
486  * locks all constituent large pages at least SHARED or locks some pages at
487  * least SHARED and calls page_szc_lock() or uses hat level page locks.
488  * Demotion using this method is implemented by page_demote_vp_pages().
489  * Please see comments in front of page_demote_vp_pages(), hat_page_demote()
490  * and page_szc_lock() for more details.
491  *
492  * Lock order: p_selock, page_szc_lock, ph_mutex/vph_mutex/freelist,
493  * hat level locks.
494  */
495 
496 typedef struct page {
497 	u_offset_t	p_offset;	/* offset into vnode for this page */
498 	struct vnode	*p_vnode;	/* vnode that this page is named by */
499 	selock_t	p_selock;	/* shared/exclusive lock on the page */
500 #if defined(_LP64)
501 	uint_t		p_vpmref;	/* vpm ref - index of the vpmap_t */
502 #endif
503 	struct page	*p_hash;	/* hash by [vnode, offset] */
504 	struct page	*p_vpnext;	/* next page in vnode list */
505 	struct page	*p_vpprev;	/* prev page in vnode list */
506 	struct page	*p_next;	/* next page in free/intrans lists */
507 	struct page	*p_prev;	/* prev page in free/intrans lists */
508 	ushort_t	p_lckcnt;	/* number of locks on page data */
509 	ushort_t	p_cowcnt;	/* number of copy on write lock */
510 	kcondvar_t	p_cv;		/* page struct's condition var */
511 	kcondvar_t	p_io_cv;	/* for iolock */
512 	uchar_t		p_iolock_state;	/* replaces p_iolock */
513 	volatile uchar_t p_szc;		/* page size code */
514 	uchar_t		p_fsdata;	/* file system dependent byte */
515 	uchar_t		p_state;	/* p_free, p_noreloc */
516 	uchar_t		p_nrm;		/* non-cache, ref, mod readonly bits */
517 #if defined(__sparc)
518 	uchar_t		p_vcolor;	/* virtual color */
519 #else
520 	uchar_t		p_embed;	/* x86 - changes p_mapping & p_index */
521 #endif
522 	uchar_t		p_index;	/* MPSS mapping info. Not used on x86 */
523 	uchar_t		p_toxic;	/* page has an unrecoverable error */
524 	void		*p_mapping;	/* hat specific translation info */
525 	pfn_t		p_pagenum;	/* physical page number */
526 
527 	uint_t		p_share;	/* number of translations */
528 #if defined(_LP64)
529 	uint_t		p_sharepad;	/* pad for growing p_share */
530 #endif
531 	uint_t		p_slckcnt;	/* number of softlocks */
532 #if defined(__sparc)
533 	uint_t		p_kpmref;	/* number of kpm mapping sharers */
534 	struct kpme	*p_kpmelist;	/* kpm specific mapping info */
535 #else
536 	/* index of entry in p_map when p_embed is set */
537 	uint_t		p_mlentry;
538 #endif
539 #if defined(_LP64)
540 	kmutex_t	p_ilock;	/* protects p_vpmref */
541 #else
542 	uint64_t	p_msresv_2;	/* page allocation debugging */
543 #endif
544 } page_t;
545 
546 
547 typedef	page_t	devpage_t;
548 #define	devpage	page
549 
550 #define	PAGE_LOCK_MAXIMUM \
551 	((1 << (sizeof (((page_t *)0)->p_lckcnt) * NBBY)) - 1)
552 
553 #define	PAGE_SLOCK_MAXIMUM UINT_MAX
554 
555 /*
556  * Page hash table is a power-of-two in size, externally chained
557  * through the hash field.  PAGE_HASHAVELEN is the average length
558  * desired for this chain, from which the size of the page_hash
559  * table is derived at boot time and stored in the kernel variable
560  * page_hashsz.  In the hash function it is given by PAGE_HASHSZ.
561  *
562  * PAGE_HASH_FUNC returns an index into the page_hash[] array.  This
563  * index is also used to derive the mutex that protects the chain.
564  *
565  * In constructing the hash function, first we dispose of unimportant bits
566  * (page offset from "off" and the low 3 bits of "vp" which are zero for
567  * struct alignment). Then shift and sum the remaining bits a couple times
568  * in order to get as many source bits from the two source values into the
569  * resulting hashed value.  Note that this will perform quickly, since the
570  * shifting/summing are fast register to register operations with no additional
571  * memory references).
572  *
573  * PH_SHIFT_SIZE is the amount to use for the successive shifts in the hash
574  * function below.  The actual value is LOG2(PH_TABLE_SIZE), so that as many
575  * bits as possible will filter thru PAGE_HASH_FUNC() and PAGE_HASH_MUTEX().
576  */
577 #if defined(_LP64)
578 
579 #if NCPU < 4
580 #define	PH_TABLE_SIZE	128
581 #define	PH_SHIFT_SIZE	7
582 #else
583 #define	PH_TABLE_SIZE	(2 * NCPU_P2)
584 #define	PH_SHIFT_SIZE	(NCPU_LOG2 + 1)
585 #endif
586 
587 #else	/* 32 bits */
588 
589 #if NCPU < 4
590 #define	PH_TABLE_SIZE	16
591 #define	PH_SHIFT_SIZE	4
592 #else
593 #define	PH_TABLE_SIZE	128
594 #define	PH_SHIFT_SIZE	7
595 #endif
596 
597 #endif	/* _LP64 */
598 
599 /*
600  *
601  * We take care to get as much randomness as possible from both the vp and
602  * the offset.  Workloads can have few vnodes with many offsets, many vnodes
603  * with few offsets or a moderate mix of both.  This hash should perform
604  * equally well for each of these possibilities and for all types of memory
605  * allocations.
606  *
607  * vnodes representing files are created over a long period of time and
608  * have good variation in the upper vp bits, and the right shifts below
609  * capture these bits.  However, swap vnodes are created quickly in a
610  * narrow vp* range.  Refer to comments at swap_alloc: vnum has exactly
611  * AN_VPSHIFT bits, so the kmem_alloc'd vnode addresses have approximately
612  * AN_VPSHIFT bits of variation above their VNODE_ALIGN low order 0 bits.
613  * Spread swap vnodes widely in the hash table by XOR'ing a term with the
614  * vp bits of variation left shifted to the top of the range.
615  */
616 
617 #define	PAGE_HASHSZ	page_hashsz
618 #define	PAGE_HASHAVELEN		4
619 #define	PAGE_HASH_FUNC(vp, off) \
620 	(((((uintptr_t)(off) >> PAGESHIFT) ^ \
621 		((uintptr_t)(off) >> (PAGESHIFT + PH_SHIFT_SIZE))) ^ \
622 		(((uintptr_t)(vp) >> 3) ^ \
623 		((uintptr_t)(vp) >> (3 + PH_SHIFT_SIZE)) ^ \
624 		((uintptr_t)(vp) >> (3 + 2 * PH_SHIFT_SIZE)) ^ \
625 		((uintptr_t)(vp) << \
626 		(page_hashsz_shift - AN_VPSHIFT - VNODE_ALIGN_LOG2)))) & \
627 		(PAGE_HASHSZ - 1))
628 #ifdef _KERNEL
629 
630 /*
631  * The page hash value is re-hashed to an index for the ph_mutex array.
632  *
633  * For 64 bit kernels, the mutex array is padded out to prevent false
634  * sharing of cache sub-blocks (64 bytes) of adjacent mutexes.
635  *
636  * For 32 bit kernels, we don't want to waste kernel address space with
637  * padding, so instead we rely on the hash function to introduce skew of
638  * adjacent vnode/offset indexes (the left shift part of the hash function).
639  * Since sizeof (kmutex_t) is 8, we shift an additional 3 to skew to a different
640  * 64 byte sub-block.
641  */
642 extern pad_mutex_t ph_mutex[];
643 
644 #define	PAGE_HASH_MUTEX(x) \
645 	&(ph_mutex[((x) ^ ((x) >> PH_SHIFT_SIZE) + ((x) << 3)) & \
646 		(PH_TABLE_SIZE - 1)].pad_mutex)
647 
648 /*
649  * Flags used while creating pages.
650  */
651 #define	PG_EXCL		0x0001
652 #define	PG_WAIT		0x0002		/* Blocking memory allocations */
653 #define	PG_PHYSCONTIG	0x0004		/* NOT SUPPORTED */
654 #define	PG_MATCH_COLOR	0x0008		/* SUPPORTED by free list routines */
655 #define	PG_NORELOC	0x0010		/* Non-relocatable alloc hint. */
656 					/* Page must be PP_ISNORELOC */
657 #define	PG_PANIC	0x0020		/* system will panic if alloc fails */
658 #define	PG_PUSHPAGE	0x0040		/* alloc may use reserve */
659 #define	PG_LOCAL	0x0080		/* alloc from given lgrp only */
660 #define	PG_NORMALPRI	0x0100		/* PG_WAIT like priority, but */
661 					/* non-blocking */
662 /*
663  * When p_selock has the SE_EWANTED bit set, threads waiting for SE_EXCL
664  * access are given priority over all other waiting threads.
665  */
666 #define	SE_EWANTED	0x40000000
667 #define	PAGE_LOCKED(pp)		(((pp)->p_selock & ~SE_EWANTED) != 0)
668 #define	PAGE_SHARED(pp)		(((pp)->p_selock & ~SE_EWANTED) > 0)
669 #define	PAGE_EXCL(pp)		((pp)->p_selock < 0)
670 #define	PAGE_LOCKED_SE(pp, se)	\
671 	((se) == SE_EXCL ? PAGE_EXCL(pp) : PAGE_SHARED(pp))
672 
673 extern	long page_hashsz;
674 extern	unsigned int page_hashsz_shift;
675 extern	page_t **page_hash;
676 
677 extern	pad_mutex_t page_llocks[];	/* page logical lock mutex */
678 extern	kmutex_t freemem_lock;		/* freemem lock */
679 
680 extern	pgcnt_t	total_pages;		/* total pages in the system */
681 
682 /*
683  * Variables controlling locking of physical memory.
684  */
685 extern	pgcnt_t	pages_pp_maximum;	/* tuning: lock + claim <= max */
686 extern	void init_pages_pp_maximum(void);
687 
688 struct lgrp;
689 
690 /* page_list_{add,sub} flags */
691 
692 /* which list */
693 #define	PG_FREE_LIST	0x0001
694 #define	PG_CACHE_LIST	0x0002
695 
696 /* where on list */
697 #define	PG_LIST_TAIL	0x0010
698 #define	PG_LIST_HEAD	0x0020
699 
700 /* called from */
701 #define	PG_LIST_ISINIT	0x1000
702 
703 /*
704  * Page frame operations.
705  */
706 page_t	*page_lookup(struct vnode *, u_offset_t, se_t);
707 page_t	*page_lookup_create(struct vnode *, u_offset_t, se_t, page_t *,
708 	spgcnt_t *, int);
709 page_t	*page_lookup_nowait(struct vnode *, u_offset_t, se_t);
710 page_t	*page_find(struct vnode *, u_offset_t);
711 page_t	*page_exists(struct vnode *, u_offset_t);
712 int	page_exists_physcontig(vnode_t *, u_offset_t, uint_t, page_t *[]);
713 int	page_exists_forreal(struct vnode *, u_offset_t, uint_t *);
714 void	page_needfree(spgcnt_t);
715 page_t	*page_create(struct vnode *, u_offset_t, size_t, uint_t);
716 int	page_alloc_pages(struct vnode *, struct seg *, caddr_t, page_t **,
717 	page_t **, uint_t, int, int);
718 page_t  *page_create_va_large(vnode_t *vp, u_offset_t off, size_t bytes,
719 	uint_t flags, struct seg *seg, caddr_t vaddr, void *arg);
720 page_t	*page_create_va(struct vnode *, u_offset_t, size_t, uint_t,
721 	struct seg *, caddr_t);
722 int	page_create_wait(pgcnt_t npages, uint_t flags);
723 void    page_create_putback(spgcnt_t npages);
724 void	page_free(page_t *, int);
725 void	page_free_at_startup(page_t *);
726 void	page_free_pages(page_t *);
727 void	free_vp_pages(struct vnode *, u_offset_t, size_t);
728 int	page_reclaim(page_t *, kmutex_t *);
729 int	page_reclaim_pages(page_t *, kmutex_t *, uint_t);
730 void	page_destroy(page_t *, int);
731 void	page_destroy_pages(page_t *);
732 void	page_destroy_free(page_t *);
733 void	page_rename(page_t *, struct vnode *, u_offset_t);
734 int	page_hashin(page_t *, struct vnode *, u_offset_t, kmutex_t *);
735 void	page_hashout(page_t *, kmutex_t *);
736 int	page_num_hashin(pfn_t, struct vnode *, u_offset_t);
737 void	page_add(page_t **, page_t *);
738 void	page_add_common(page_t **, page_t *);
739 void	page_sub(page_t **, page_t *);
740 void	page_sub_common(page_t **, page_t *);
741 page_t	*page_get_freelist(struct vnode *, u_offset_t, struct seg *,
742 		caddr_t, size_t, uint_t, struct lgrp *);
743 
744 page_t	*page_get_cachelist(struct vnode *, u_offset_t, struct seg *,
745 		caddr_t, uint_t, struct lgrp *);
746 #if defined(__i386) || defined(__amd64)
747 int	page_chk_freelist(uint_t);
748 #endif
749 void	page_list_add(page_t *, int);
750 void	page_boot_demote(page_t *);
751 void	page_promote_size(page_t *, uint_t);
752 void	page_list_add_pages(page_t *, int);
753 void	page_list_sub(page_t *, int);
754 void	page_list_sub_pages(page_t *, uint_t);
755 void	page_list_xfer(page_t *, int, int);
756 void	page_list_break(page_t **, page_t **, size_t);
757 void	page_list_concat(page_t **, page_t **);
758 void	page_vpadd(page_t **, page_t *);
759 void	page_vpsub(page_t **, page_t *);
760 int	page_lock(page_t *, se_t, kmutex_t *, reclaim_t);
761 int	page_lock_es(page_t *, se_t, kmutex_t *, reclaim_t, int);
762 void page_lock_clr_exclwanted(page_t *);
763 int	page_trylock(page_t *, se_t);
764 int	page_try_reclaim_lock(page_t *, se_t, int);
765 int	page_tryupgrade(page_t *);
766 void	page_downgrade(page_t *);
767 void	page_unlock(page_t *);
768 void	page_unlock_nocapture(page_t *);
769 void	page_lock_delete(page_t *);
770 int	page_deleted(page_t *);
771 int	page_pp_lock(page_t *, int, int);
772 void	page_pp_unlock(page_t *, int, int);
773 int	page_resv(pgcnt_t, uint_t);
774 void	page_unresv(pgcnt_t);
775 void	page_pp_useclaim(page_t *, page_t *, uint_t);
776 int	page_addclaim(page_t *);
777 int	page_subclaim(page_t *);
778 int	page_addclaim_pages(page_t **);
779 int	page_subclaim_pages(page_t **);
780 pfn_t	page_pptonum(page_t *);
781 page_t	*page_numtopp(pfn_t, se_t);
782 page_t	*page_numtopp_noreclaim(pfn_t, se_t);
783 page_t	*page_numtopp_nolock(pfn_t);
784 page_t	*page_numtopp_nowait(pfn_t, se_t);
785 page_t  *page_first();
786 page_t  *page_next(page_t *);
787 page_t  *page_list_next(page_t *);
788 page_t	*page_nextn(page_t *, ulong_t);
789 page_t	*page_next_scan_init(void **);
790 page_t	*page_next_scan_large(page_t *, ulong_t *, void **);
791 void    prefetch_page_r(void *);
792 int	ppcopy(page_t *, page_t *);
793 void	page_relocate_hash(page_t *, page_t *);
794 void	pagezero(page_t *, uint_t, uint_t);
795 void	pagescrub(page_t *, uint_t, uint_t);
796 void	page_io_lock(page_t *);
797 void	page_io_unlock(page_t *);
798 int	page_io_trylock(page_t *);
799 int	page_iolock_assert(page_t *);
800 void	page_iolock_init(page_t *);
801 void	page_io_wait(page_t *);
802 int	page_io_locked(page_t *);
803 pgcnt_t	page_busy(int);
804 void	page_lock_init(void);
805 ulong_t	page_share_cnt(page_t *);
806 int	page_isshared(page_t *);
807 int	page_isfree(page_t *);
808 int	page_isref(page_t *);
809 int	page_ismod(page_t *);
810 int	page_release(page_t *, int);
811 void	page_retire_init(void);
812 int	page_retire(uint64_t, uchar_t);
813 int	page_retire_check(uint64_t, uint64_t *);
814 int	page_unretire(uint64_t);
815 int	page_unretire_pp(page_t *, int);
816 void	page_tryretire(page_t *);
817 void	page_retire_mdboot();
818 uint64_t	page_retire_pend_count(void);
819 uint64_t	page_retire_pend_kas_count(void);
820 void	page_retire_incr_pend_count(void *);
821 void	page_retire_decr_pend_count(void *);
822 void	page_clrtoxic(page_t *, uchar_t);
823 void	page_settoxic(page_t *, uchar_t);
824 
825 int	page_mem_avail(pgcnt_t);
826 int	page_reclaim_mem(pgcnt_t, pgcnt_t, int);
827 
828 void page_set_props(page_t *, uint_t);
829 void page_clr_all_props(page_t *);
830 int page_clear_lck_cow(page_t *, int);
831 
832 kmutex_t	*page_vnode_mutex(struct vnode *);
833 kmutex_t	*page_se_mutex(struct page *);
834 kmutex_t	*page_szc_lock(struct page *);
835 int		page_szc_lock_assert(struct page *pp);
836 
837 /*
838  * Page relocation interfaces. page_relocate() is generic.
839  * page_get_replacement_page() is provided by the PSM.
840  * page_free_replacement_page() is generic.
841  */
842 int group_page_trylock(page_t *, se_t);
843 void group_page_unlock(page_t *);
844 int page_relocate(page_t **, page_t **, int, int, spgcnt_t *, struct lgrp *);
845 int do_page_relocate(page_t **, page_t **, int, spgcnt_t *, struct lgrp *);
846 page_t *page_get_replacement_page(page_t *, struct lgrp *, uint_t);
847 void page_free_replacement_page(page_t *);
848 int page_relocate_cage(page_t **, page_t **);
849 
850 int page_try_demote_pages(page_t *);
851 int page_try_demote_free_pages(page_t *);
852 void page_demote_free_pages(page_t *);
853 
854 struct anon_map;
855 
856 void page_mark_migrate(struct seg *, caddr_t, size_t, struct anon_map *,
857     ulong_t, vnode_t *, u_offset_t, int);
858 void page_migrate(struct seg *, caddr_t, page_t **, pgcnt_t);
859 
860 /*
861  * Tell the PIM we are adding physical memory
862  */
863 void add_physmem(page_t *, size_t, pfn_t);
864 void add_physmem_cb(page_t *, pfn_t);	/* callback for page_t part */
865 
866 /*
867  * hw_page_array[] is configured with hardware supported page sizes by
868  * platform specific code.
869  */
870 typedef struct {
871 	size_t	hp_size;
872 	uint_t	hp_shift;
873 	uint_t  hp_colors;
874 	pgcnt_t	hp_pgcnt;	/* base pagesize cnt */
875 } hw_pagesize_t;
876 
877 extern hw_pagesize_t	hw_page_array[];
878 extern uint_t		page_coloring_shift;
879 extern uint_t		page_colors_mask;
880 extern int		cpu_page_colors;
881 extern uint_t		colorequiv;
882 extern uchar_t		colorequivszc[];
883 
884 uint_t	page_num_pagesizes(void);
885 uint_t	page_num_user_pagesizes(int);
886 size_t	page_get_pagesize(uint_t);
887 size_t	page_get_user_pagesize(uint_t n);
888 pgcnt_t	page_get_pagecnt(uint_t);
889 uint_t	page_get_shift(uint_t);
890 int	page_szc(size_t);
891 int	page_szc_user_filtered(size_t);
892 
893 /* page_get_replacement page flags */
894 #define	PGR_SAMESZC	0x1	/* only look for page size same as orig */
895 #define	PGR_NORELOC	0x2	/* allocate a P_NORELOC page */
896 
897 /*
898  * macros for "masked arithmetic"
899  * The purpose is to step through all combinations of a set of bits while
900  * keeping some other bits fixed. Fixed bits need not be contiguous. The
901  * variable bits need not be contiguous either, or even right aligned. The
902  * trick is to set all fixed bits to 1, then increment, then restore the
903  * fixed bits. If incrementing causes a carry from a low bit position, the
904  * carry propagates thru the fixed bits, because they are temporarily set to 1.
905  *	v is the value
906  *	i is the increment
907  *	eq_mask defines the fixed bits
908  *	mask limits the size of the result
909  */
910 #define	ADD_MASKED(v, i, eq_mask, mask) \
911 	(((((v) | (eq_mask)) + (i)) & (mask) & ~(eq_mask)) | ((v) & (eq_mask)))
912 
913 /*
914  * convenience macro which increments by 1
915  */
916 #define	INC_MASKED(v, eq_mask, mask) ADD_MASKED(v, 1, eq_mask, mask)
917 
918 #endif	/* _KERNEL */
919 
920 /*
921  * Constants used for the p_iolock_state
922  */
923 #define	PAGE_IO_INUSE	0x1
924 #define	PAGE_IO_WANTED	0x2
925 
926 /*
927  * Constants used for page_release status
928  */
929 #define	PGREL_NOTREL    0x1
930 #define	PGREL_CLEAN	0x2
931 #define	PGREL_MOD	0x3
932 
933 /*
934  * The p_state field holds what used to be the p_age and p_free
935  * bits.  These fields are protected by p_selock (see above).
936  */
937 #define	P_FREE		0x80		/* Page on free list */
938 #define	P_NORELOC	0x40		/* Page is non-relocatable */
939 #define	P_MIGRATE	0x20		/* Migrate page on next touch */
940 #define	P_SWAP		0x10		/* belongs to vnode that is V_ISSWAP */
941 #define	P_BOOTPAGES	0x08		/* member of bootpages list */
942 #define	P_RAF		0x04		/* page retired at free */
943 
944 #define	PP_ISFREE(pp)		((pp)->p_state & P_FREE)
945 #define	PP_ISAGED(pp)		(((pp)->p_state & P_FREE) && \
946 					((pp)->p_vnode == NULL))
947 #define	PP_ISNORELOC(pp)	((pp)->p_state & P_NORELOC)
948 #define	PP_ISKAS(pp)		(VN_ISKAS((pp)->p_vnode))
949 #define	PP_ISNORELOCKERNEL(pp)	(PP_ISNORELOC(pp) && PP_ISKAS(pp))
950 #define	PP_ISMIGRATE(pp)	((pp)->p_state & P_MIGRATE)
951 #define	PP_ISSWAP(pp)		((pp)->p_state & P_SWAP)
952 #define	PP_ISBOOTPAGES(pp)	((pp)->p_state & P_BOOTPAGES)
953 #define	PP_ISRAF(pp)		((pp)->p_state & P_RAF)
954 
955 #define	PP_SETFREE(pp)		((pp)->p_state = ((pp)->p_state & ~P_MIGRATE) \
956 				| P_FREE)
957 #define	PP_SETAGED(pp)		ASSERT(PP_ISAGED(pp))
958 #define	PP_SETNORELOC(pp)	((pp)->p_state |= P_NORELOC)
959 #define	PP_SETMIGRATE(pp)	((pp)->p_state |= P_MIGRATE)
960 #define	PP_SETSWAP(pp)		((pp)->p_state |= P_SWAP)
961 #define	PP_SETBOOTPAGES(pp)	((pp)->p_state |= P_BOOTPAGES)
962 #define	PP_SETRAF(pp)		((pp)->p_state |= P_RAF)
963 
964 #define	PP_CLRFREE(pp)		((pp)->p_state &= ~P_FREE)
965 #define	PP_CLRAGED(pp)		ASSERT(!PP_ISAGED(pp))
966 #define	PP_CLRNORELOC(pp)	((pp)->p_state &= ~P_NORELOC)
967 #define	PP_CLRMIGRATE(pp)	((pp)->p_state &= ~P_MIGRATE)
968 #define	PP_CLRSWAP(pp)		((pp)->p_state &= ~P_SWAP)
969 #define	PP_CLRBOOTPAGES(pp)	((pp)->p_state &= ~P_BOOTPAGES)
970 #define	PP_CLRRAF(pp)		((pp)->p_state &= ~P_RAF)
971 
972 /*
973  * Flags for page_t p_toxic, for tracking memory hardware errors.
974  *
975  * These flags are OR'ed into p_toxic with page_settoxic() to track which
976  * error(s) have occurred on a given page. The flags are cleared with
977  * page_clrtoxic(). Both page_settoxic() and page_cleartoxic use atomic
978  * primitives to manipulate the p_toxic field so no other locking is needed.
979  *
980  * When an error occurs on a page, p_toxic is set to record the error. The
981  * error could be a memory error or something else (i.e. a datapath). The Page
982  * Retire mechanism does not try to determine the exact cause of the error;
983  * Page Retire rightly leaves that sort of determination to FMA's Diagnostic
984  * Engine (DE).
985  *
986  * Note that, while p_toxic bits can be set without holding any locks, they
987  * should only be cleared while holding the page exclusively locked.
988  * There is one exception to this, the PR_CAPTURE bit is protected by a mutex
989  * within the page capture logic and thus to set or clear the bit, that mutex
990  * needs to be held.  The page does not need to be locked but the page_clrtoxic
991  * function must be used as we need an atomic operation.
992  * Also note that there is what amounts to a hack to prevent recursion with
993  * large pages such that if we are unlocking a page and the PR_CAPTURE bit is
994  * set, we will only try to capture the page if the current threads T_CAPTURING
995  * flag is not set.  If the flag is set, the unlock will not try to capture
996  * the page even though the PR_CAPTURE bit is set.
997  *
998  * Pages with PR_UE or PR_FMA flags are retired unconditionally, while pages
999  * with PR_MCE are retired if the system has not retired too many of them.
1000  *
1001  * A page must be exclusively locked to be retired. Pages can be retired if
1002  * they are mapped, modified, or both, as long as they are not marked PR_UE,
1003  * since pages with uncorrectable errors cannot be relocated in memory.
1004  * Once a page has been successfully retired it is zeroed, attached to the
1005  * retired_pages vnode and, finally, PR_RETIRED is set in p_toxic. The other
1006  * p_toxic bits are NOT cleared. Pages are not left locked after retiring them
1007  * to avoid special case code throughout the kernel; rather, page_*lock() will
1008  * fail to lock the page, unless SE_RETIRED is passed as an argument.
1009  *
1010  * While we have your attention, go take a look at the comments at the
1011  * beginning of page_retire.c too.
1012  */
1013 #define	PR_OK		0x00	/* no problem */
1014 #define	PR_MCE		0x01	/* page has seen two or more CEs */
1015 #define	PR_UE		0x02	/* page has an unhandled UE */
1016 #define	PR_UE_SCRUBBED	0x04	/* page has seen a UE but was cleaned */
1017 #define	PR_FMA		0x08	/* A DE wants this page retired */
1018 #define	PR_CAPTURE	0x10	/* page is hashed on page_capture_hash[] */
1019 #define	PR_RESV		0x20	/* Reserved for future use */
1020 #define	PR_MSG		0x40	/* message(s) already printed for this page */
1021 #define	PR_RETIRED	0x80	/* This page has been retired */
1022 
1023 #define	PR_REASONS	(PR_UE | PR_MCE | PR_FMA)
1024 #define	PR_TOXIC	(PR_UE)
1025 #define	PR_ERRMASK	(PR_UE | PR_UE_SCRUBBED | PR_MCE | PR_FMA)
1026 #define	PR_TOXICFLAGS	(0xCF)
1027 
1028 #define	PP_RETIRED(pp)	((pp)->p_toxic & PR_RETIRED)
1029 #define	PP_TOXIC(pp)	((pp)->p_toxic & PR_TOXIC)
1030 #define	PP_PR_REQ(pp)	(((pp)->p_toxic & PR_REASONS) && !PP_RETIRED(pp))
1031 #define	PP_PR_NOSHARE(pp)						\
1032 	((((pp)->p_toxic & (PR_RETIRED | PR_FMA | PR_UE)) == PR_FMA) &&	\
1033 	!PP_ISKAS(pp))
1034 
1035 /*
1036  * Flags for page_unretire_pp
1037  */
1038 #define	PR_UNR_FREE	0x1
1039 #define	PR_UNR_CLEAN	0x2
1040 #define	PR_UNR_TEMP	0x4
1041 
1042 /*
1043  * kpm large page description.
1044  * The virtual address range of segkpm is divided into chunks of
1045  * kpm_pgsz. Each chunk is controlled by a kpm_page_t. The ushort
1046  * is sufficient for 2^^15 * PAGESIZE, so e.g. the maximum kpm_pgsz
1047  * for 8K is 256M and 2G for 64K pages. It it kept as small as
1048  * possible to save physical memory space.
1049  *
1050  * There are 2 segkpm mapping windows within in the virtual address
1051  * space when we have to prevent VAC alias conflicts. The so called
1052  * Alias window (mappings are always by PAGESIZE) is controlled by
1053  * kp_refcnta. The regular window is controlled by kp_refcnt for the
1054  * normal operation, which is to use the largest available pagesize.
1055  * When VAC alias conflicts are present within a chunk in the regular
1056  * window the large page mapping is broken up into smaller PAGESIZE
1057  * mappings. kp_refcntc is used to control the pages that are invoked
1058  * in the conflict and kp_refcnts holds the active mappings done
1059  * with the small page size. In non vac conflict mode kp_refcntc is
1060  * also used as "go" indication (-1) for the trap level tsbmiss
1061  * handler.
1062  */
1063 typedef struct kpm_page {
1064 	short kp_refcnt;	/* pages mapped large */
1065 	short kp_refcnta;	/* pages mapped in Alias window */
1066 	short kp_refcntc;	/* TL-tsbmiss flag; #vac alias conflict pages */
1067 	short kp_refcnts;	/* vac alias: pages mapped small */
1068 } kpm_page_t;
1069 
1070 /*
1071  * Note: khl_lock offset changes must be reflected in sfmmu_asm.s
1072  */
1073 typedef struct kpm_hlk {
1074 	kmutex_t khl_mutex;	/* kpm_page mutex */
1075 	uint_t   khl_lock;	/* trap level tsbmiss handling */
1076 } kpm_hlk_t;
1077 
1078 /*
1079  * kpm small page description.
1080  * When kpm_pgsz is equal to PAGESIZE a smaller representation is used
1081  * to save memory space. Alias range mappings and regular segkpm
1082  * mappings are done in units of PAGESIZE and can share the mapping
1083  * information and the mappings are always distinguishable by their
1084  * virtual address. Other information needed for VAC conflict prevention
1085  * is already available on a per page basis.
1086  *
1087  * The state about how a kpm page is mapped and whether it is ready to go
1088  * is indicated by the following 1 byte kpm_spage structure. This byte is
1089  * split into two 4-bit parts - kp_mapped and kp_mapped_go.
1090  * 	- kp_mapped == 1	the page is mapped cacheable
1091  *	- kp_mapped == 2	the page is mapped non-cacheable
1092  *	- kp_mapped_go == 1	the mapping is ready to be dropped in
1093  *	- kp_mapped_go == 0	the mapping is not ready to be dropped in.
1094  * When kp_mapped_go == 0, we will have C handler resolve the VAC conflict.
1095  * Otherwise, the assembly tsb miss handler can simply drop in the mapping
1096  * when a tsb miss occurs.
1097  */
1098 typedef union kpm_spage {
1099 	struct {
1100 #ifdef  _BIG_ENDIAN
1101 		uchar_t mapped_go: 4;	/* go or nogo flag */
1102 		uchar_t mapped: 4;	/* page mapped small */
1103 #else
1104 		uchar_t mapped: 4;	/* page mapped small */
1105 		uchar_t mapped_go: 4;	/* go or nogo flag */
1106 #endif
1107 	} kpm_spage_un;
1108 	uchar_t kp_mapped_flag;
1109 } kpm_spage_t;
1110 
1111 #define	kp_mapped	kpm_spage_un.mapped
1112 #define	kp_mapped_go	kpm_spage_un.mapped_go
1113 
1114 /*
1115  * Note: kshl_lock offset changes must be reflected in sfmmu_asm.s
1116  */
1117 typedef struct kpm_shlk {
1118 	uint_t   kshl_lock;	/* trap level tsbmiss handling */
1119 } kpm_shlk_t;
1120 
1121 /*
1122  * Each segment of physical memory is described by a memseg struct.
1123  * Within a segment, memory is considered contiguous. The members
1124  * can be categorized as follows:
1125  * . Platform independent:
1126  *         pages, epages, pages_base, pages_end, next, lnext.
1127  * . 64bit only but platform independent:
1128  *         kpm_pbase, kpm_nkpmpgs, kpm_pages, kpm_spages.
1129  * . Really platform or mmu specific:
1130  *         pagespa, epagespa, nextpa, kpm_pagespa.
1131  * . Mixed:
1132  *         msegflags.
1133  */
1134 struct memseg {
1135 	page_t *pages, *epages;		/* [from, to] in page array */
1136 	pfn_t pages_base, pages_end;	/* [from, to] in page numbers */
1137 	struct memseg *next;		/* next segment in list */
1138 	struct memseg *lnext;		/* next segment in deleted list */
1139 #if defined(__sparc)
1140 	uint64_t pagespa, epagespa;	/* [from, to] page array physical */
1141 	uint64_t nextpa;		/* physical next pointer */
1142 	pfn_t	kpm_pbase;		/* start of kpm range */
1143 	pgcnt_t kpm_nkpmpgs;		/* # of kpm_pgsz pages */
1144 	union _mseg_un {
1145 		kpm_page_t  *kpm_lpgs;	/* ptr to kpm_page array */
1146 		kpm_spage_t *kpm_spgs;	/* ptr to kpm_spage array */
1147 	} mseg_un;
1148 	uint64_t kpm_pagespa;		/* physical ptr to kpm (s)pages array */
1149 #endif /* __sparc */
1150 	uint_t msegflags;		/* memseg flags */
1151 };
1152 
1153 /* memseg union aliases */
1154 #define	kpm_pages	mseg_un.kpm_lpgs
1155 #define	kpm_spages	mseg_un.kpm_spgs
1156 
1157 /* msegflags */
1158 #define	MEMSEG_DYNAMIC		0x1	/* DR: memory was added dynamically */
1159 #define	MEMSEG_META_INCL	0x2	/* DR: memseg includes it's metadata */
1160 #define	MEMSEG_META_ALLOC	0x4	/* DR: memseg allocated it's metadata */
1161 
1162 /* memseg support macros */
1163 #define	MSEG_NPAGES(SEG)	((SEG)->pages_end - (SEG)->pages_base)
1164 
1165 /* memseg hash */
1166 #define	MEM_HASH_SHIFT		0x9
1167 #define	N_MEM_SLOTS		0x200		/* must be a power of 2 */
1168 #define	MEMSEG_PFN_HASH(pfn)	(((pfn)/mhash_per_slot) & (N_MEM_SLOTS - 1))
1169 
1170 /* memseg  externals */
1171 extern struct memseg *memsegs;		/* list of memory segments */
1172 extern ulong_t mhash_per_slot;
1173 extern uint64_t memsegspa;		/* memsegs as physical address */
1174 
1175 void build_pfn_hash();
1176 extern struct memseg *page_numtomemseg_nolock(pfn_t pfnum);
1177 
1178 /*
1179  * page capture related info:
1180  * The page capture routines allow us to asynchronously capture given pages
1181  * for the explicit use of the requestor.  New requestors can be added by
1182  * explicitly adding themselves to the PC_* flags below and incrementing
1183  * PC_NUM_CALLBACKS as necessary.
1184  *
1185  * Subsystems using page capture must register a callback before attempting
1186  * to capture a page.  A duration of -1 will indicate that we will never give
1187  * up while trying to capture a page and will only stop trying to capture the
1188  * given page once we have successfully captured it.  Thus the user needs to be
1189  * aware of the behavior of all callers who have a duration of -1.
1190  *
1191  * For now, only /dev/physmem and page retire use the page capture interface
1192  * and only a single request can be outstanding for a given page.  Thus, if
1193  * /dev/phsymem wants a page and page retire also wants the same page, only
1194  * the page retire request will be honored until the point in time that the
1195  * page is actually retired, at which point in time, subsequent requests by
1196  * /dev/physmem will succeed if the CAPTURE_GET_RETIRED flag was set.
1197  */
1198 
1199 #define	PC_RETIRE		(0)
1200 #define	PC_PHYSMEM		(1)
1201 #define	PC_NUM_CALLBACKS	(2)
1202 #define	PC_MASK			((1 << PC_NUM_CALLBACKS) - 1)
1203 
1204 #define	CAPTURE_RETIRE		(1 << PC_RETIRE)
1205 #define	CAPTURE_PHYSMEM		(1 << PC_PHYSMEM)
1206 
1207 #define	CAPTURE_ASYNC		(0x0200)
1208 
1209 #define	CAPTURE_GET_RETIRED	(0x1000)
1210 #define	CAPTURE_GET_CAGE	(0x2000)
1211 
1212 struct page_capture_callback {
1213 	int cb_active;		/* 1 means active, 0 means inactive */
1214 	clock_t duration;	/* the length in time that we'll attempt to */
1215 				/* capture this page asynchronously. (in HZ) */
1216 	krwlock_t cb_rwlock;
1217 	int (*cb_func)(page_t *, void *, uint_t); /* callback function */
1218 };
1219 
1220 extern kcondvar_t pc_cv;
1221 
1222 void page_capture_register_callback(uint_t index, clock_t duration,
1223     int (*cb_func)(page_t *, void *, uint_t));
1224 void page_capture_unregister_callback(uint_t index);
1225 int page_trycapture(page_t *pp, uint_t szc, uint_t flags, void *datap);
1226 void page_unlock_capture(page_t *pp);
1227 int page_capture_unretire_pp(page_t *);
1228 
1229 extern int memsegs_trylock(int);
1230 extern void memsegs_lock(int);
1231 extern void memsegs_unlock(int);
1232 extern int memsegs_lock_held(void);
1233 extern void memlist_read_lock(void);
1234 extern void memlist_read_unlock(void);
1235 extern void memlist_write_lock(void);
1236 extern void memlist_write_unlock(void);
1237 
1238 #ifdef	__cplusplus
1239 }
1240 #endif
1241 
1242 #endif	/* _VM_PAGE_H */
1243