xref: /titanic_44/usr/src/uts/common/vm/page.h (revision 5c46c6a053defc1bf25f2e08df2653bc7717ed0b)
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 2006 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 
26 /*	Copyright (c) 1984, 1986, 1987, 1988, 1989 AT&T	*/
27 /*	  All Rights Reserved  	*/
28 
29 /*
30  * University Copyright- Copyright (c) 1982, 1986, 1988
31  * The Regents of the University of California
32  * All Rights Reserved
33  *
34  * University Acknowledgment- Portions of this document are derived from
35  * software developed by the University of California, Berkeley, and its
36  * contributors.
37  */
38 
39 #ifndef	_VM_PAGE_H
40 #define	_VM_PAGE_H
41 
42 #pragma ident	"%Z%%M%	%I%	%E% SMI"
43 
44 #include <vm/seg.h>
45 
46 #ifdef	__cplusplus
47 extern "C" {
48 #endif
49 
50 #if defined(_KERNEL) || defined(_KMEMUSER)
51 
52 /*
53  * Shared/Exclusive lock.
54  */
55 
56 /*
57  * Types of page locking supported by page_lock & friends.
58  */
59 typedef enum {
60 	SE_SHARED,
61 	SE_EXCL			/* exclusive lock (value == -1) */
62 } se_t;
63 
64 /*
65  * For requesting that page_lock reclaim the page from the free list.
66  */
67 typedef enum {
68 	P_RECLAIM,		/* reclaim page from free list */
69 	P_NO_RECLAIM		/* DON`T reclaim the page	*/
70 } reclaim_t;
71 
72 /*
73  * Callers of page_try_reclaim_lock and page_lock_es can use this flag
74  * to get SE_EXCL access before reader/writers are given access.
75  */
76 #define	SE_EXCL_WANTED	0x02
77 
78 /*
79  * All page_*lock() requests will be denied unless this flag is set in
80  * the 'es' parameter.
81  */
82 #define	SE_RETIRED	0x04
83 
84 #endif	/* _KERNEL | _KMEMUSER */
85 
86 typedef int	selock_t;
87 
88 /*
89  * Define VM_STATS to turn on all sorts of statistic gathering about
90  * the VM layer.  By default, it is only turned on when DEBUG is
91  * also defined.
92  */
93 #ifdef DEBUG
94 #define	VM_STATS
95 #endif	/* DEBUG */
96 
97 #ifdef VM_STATS
98 #define	VM_STAT_ADD(stat)			(stat)++
99 #define	VM_STAT_COND_ADD(cond, stat)		((void) (!(cond) || (stat)++))
100 #else
101 #define	VM_STAT_ADD(stat)
102 #define	VM_STAT_COND_ADD(cond, stat)
103 #endif	/* VM_STATS */
104 
105 #ifdef _KERNEL
106 
107 /*
108  * Macros to acquire and release the page logical lock.
109  */
110 #define	page_struct_lock(pp)	mutex_enter(&page_llock)
111 #define	page_struct_unlock(pp)	mutex_exit(&page_llock)
112 
113 #endif	/* _KERNEL */
114 
115 #include <sys/t_lock.h>
116 
117 struct as;
118 
119 /*
120  * Each physical page has a page structure, which is used to maintain
121  * these pages as a cache.  A page can be found via a hashed lookup
122  * based on the [vp, offset].  If a page has an [vp, offset] identity,
123  * then it is entered on a doubly linked circular list off the
124  * vnode using the vpnext/vpprev pointers.   If the p_free bit
125  * is on, then the page is also on a doubly linked circular free
126  * list using next/prev pointers.  If the "p_selock" and "p_iolock"
127  * are held, then the page is currently being read in (exclusive p_selock)
128  * or written back (shared p_selock).  In this case, the next/prev pointers
129  * are used to link the pages together for a consecutive i/o request.  If
130  * the page is being brought in from its backing store, then other processes
131  * will wait for the i/o to complete before attaching to the page since it
132  * will have an "exclusive" lock.
133  *
134  * Each page structure has the locks described below along with
135  * the fields they protect:
136  *
137  *	p_selock	This is a per-page shared/exclusive lock that is
138  *			used to implement the logical shared/exclusive
139  *			lock for each page.  The "shared" lock is normally
140  *			used in most cases while the "exclusive" lock is
141  *			required to destroy or retain exclusive access to
142  *			a page (e.g., while reading in pages).  The appropriate
143  *			lock is always held whenever there is any reference
144  *			to a page structure (e.g., during i/o).
145  *			(Note that with the addition of the "writer-lock-wanted"
146  *			semantics (via SE_EWANTED), threads must not acquire
147  *			multiple reader locks or else a deadly embrace will
148  *			occur in the following situation: thread 1 obtains a
149  *			reader lock; next thread 2 fails to get a writer lock
150  *			but specified SE_EWANTED so it will wait by either
151  *			blocking (when using page_lock_es) or spinning while
152  *			retrying (when using page_try_reclaim_lock) until the
153  *			reader lock is released; then thread 1 attempts to
154  *			get another reader lock but is denied due to
155  *			SE_EWANTED being set, and now both threads are in a
156  *			deadly embrace.)
157  *
158  *				p_hash
159  *				p_vnode
160  *				p_offset
161  *
162  *				p_free
163  *				p_age
164  *
165  *	p_iolock	This is a binary semaphore lock that provides
166  *			exclusive access to the i/o list links in each
167  *			page structure.  It is always held while the page
168  *			is on an i/o list (i.e., involved in i/o).  That is,
169  *			even though a page may be only `shared' locked
170  *			while it is doing a write, the following fields may
171  *			change anyway.  Normally, the page must be
172  *			`exclusively' locked to change anything in it.
173  *
174  *				p_next
175  *				p_prev
176  *
177  * The following fields are protected by the global page_llock:
178  *
179  *				p_lckcnt
180  *				p_cowcnt
181  *
182  * The following lists are protected by the global page_freelock:
183  *
184  *				page_cachelist
185  *				page_freelist
186  *
187  * The following, for our purposes, are protected by
188  * the global freemem_lock:
189  *
190  *				freemem
191  *				freemem_wait
192  *				freemem_cv
193  *
194  * The following fields are protected by hat layer lock(s).  When a page
195  * structure is not mapped and is not associated with a vnode (after a call
196  * to page_hashout() for example) the p_nrm field may be modified with out
197  * holding the hat layer lock:
198  *
199  *				p_nrm
200  *				p_mapping
201  *				p_share
202  *
203  * The following field is file system dependent.  How it is used and
204  * the locking strategies applied are up to the individual file system
205  * implementation.
206  *
207  *				p_fsdata
208  *
209  * The page structure is used to represent and control the system's
210  * physical pages.  There is one instance of the structure for each
211  * page that is not permenately allocated.  For example, the pages that
212  * hold the page structures are permanently held by the kernel
213  * and hence do not need page structures to track them.  The array
214  * of page structures is allocated early on in the kernel's life and
215  * is based on the amount of available physical memory.
216  *
217  * Each page structure may simultaneously appear on several linked lists.
218  * The lists are:  hash list, free or in i/o list, and a vnode's page list.
219  * Each type of list is protected by a different group of mutexes as described
220  * below:
221  *
222  * The hash list is used to quickly find a page when the page's vnode and
223  * offset within the vnode are known.  Each page that is hashed is
224  * connected via the `p_hash' field.  The anchor for each hash is in the
225  * array `page_hash'.  An array of mutexes, `ph_mutex', protects the
226  * lists anchored by page_hash[].  To either search or modify a given hash
227  * list, the appropriate mutex in the ph_mutex array must be held.
228  *
229  * The free list contains pages that are `free to be given away'.  For
230  * efficiency reasons, pages on this list are placed in two catagories:
231  * pages that are still associated with a vnode, and pages that are not
232  * associated with a vnode.  Free pages always have their `p_free' bit set,
233  * free pages that are still associated with a vnode also have their
234  * `p_age' bit set.  Pages on the free list are connected via their
235  * `p_next' and `p_prev' fields.  When a page is involved in some sort
236  * of i/o, it is not free and these fields may be used to link associated
237  * pages together.  At the moment, the free list is protected by a
238  * single mutex `page_freelock'.  The list of free pages still associated
239  * with a vnode is anchored by `page_cachelist' while other free pages
240  * are anchored in architecture dependent ways (to handle page coloring etc.).
241  *
242  * Pages associated with a given vnode appear on a list anchored in the
243  * vnode by the `v_pages' field.  They are linked together with
244  * `p_vpnext' and `p_vpprev'.  The field `p_offset' contains a page's
245  * offset within the vnode.  The pages on this list are not kept in
246  * offset order.  These lists, in a manner similar to the hash lists,
247  * are protected by an array of mutexes called `vph_hash'.  Before
248  * searching or modifying this chain the appropriate mutex in the
249  * vph_hash[] array must be held.
250  *
251  * Again, each of the lists that a page can appear on is protected by a
252  * mutex.  Before reading or writing any of the fields comprising the
253  * list, the appropriate lock must be held.  These list locks should only
254  * be held for very short intervals.
255  *
256  * In addition to the list locks, each page structure contains a
257  * shared/exclusive lock that protects various fields within it.
258  * To modify one of these fields, the `p_selock' must be exclusively held.
259  * To read a field with a degree of certainty, the lock must be at least
260  * held shared.
261  *
262  * Removing a page structure from one of the lists requires holding
263  * the appropriate list lock and the page's p_selock.  A page may be
264  * prevented from changing identity, being freed, or otherwise modified
265  * by acquiring p_selock shared.
266  *
267  * To avoid deadlocks, a strict locking protocol must be followed.  Basically
268  * there are two cases:  In the first case, the page structure in question
269  * is known ahead of time (e.g., when the page is to be added or removed
270  * from a list).  In the second case, the page structure is not known and
271  * must be found by searching one of the lists.
272  *
273  * When adding or removing a known page to one of the lists, first the
274  * page must be exclusively locked (since at least one of its fields
275  * will be modified), second the lock protecting the list must be acquired,
276  * third the page inserted or deleted, and finally the list lock dropped.
277  *
278  * The more interesting case occures when the particular page structure
279  * is not known ahead of time.  For example, when a call is made to
280  * page_lookup(), it is not known if a page with the desired (vnode and
281  * offset pair) identity exists.  So the appropriate mutex in ph_mutex is
282  * acquired, the hash list searched, and if the desired page is found
283  * an attempt is made to lock it.  The attempt to acquire p_selock must
284  * not block while the hash list lock is held.  A deadlock could occure
285  * if some other process was trying to remove the page from the list.
286  * The removing process (following the above protocol) would have exclusively
287  * locked the page, and be spinning waiting to acquire the lock protecting
288  * the hash list.  Since the searching process holds the hash list lock
289  * and is waiting to acquire the page lock, a deadlock occurs.
290  *
291  * The proper scheme to follow is: first, lock the appropriate list,
292  * search the list, and if the desired page is found either use
293  * page_trylock() (which will not block) or pass the address of the
294  * list lock to page_lock().  If page_lock() can not acquire the page's
295  * lock, it will drop the list lock before going to sleep.  page_lock()
296  * returns a value to indicate if the list lock was dropped allowing the
297  * calling program to react appropriately (i.e., retry the operation).
298  *
299  * If the list lock was dropped before the attempt at locking the page
300  * was made, checks would have to be made to ensure that the page had
301  * not changed identity before its lock was obtained.  This is because
302  * the interval between dropping the list lock and acquiring the page
303  * lock is indeterminate.
304  *
305  * In addition, when both a hash list lock (ph_mutex[]) and a vnode list
306  * lock (vph_mutex[]) are needed, the hash list lock must be acquired first.
307  * The routine page_hashin() is a good example of this sequence.
308  * This sequence is ASSERTed by checking that the vph_mutex[] is not held
309  * just before each acquisition of one of the mutexs in ph_mutex[].
310  *
311  * So, as a quick summary:
312  *
313  * 	pse_mutex[]'s protect the p_selock and p_cv fields.
314  *
315  * 	p_selock protects the p_free, p_age, p_vnode, p_offset and p_hash,
316  *
317  * 	ph_mutex[]'s protect the page_hash[] array and its chains.
318  *
319  * 	vph_mutex[]'s protect the v_pages field and the vp page chains.
320  *
321  *	First lock the page, then the hash chain, then the vnode chain.  When
322  *	this is not possible `trylocks' must be used.  Sleeping while holding
323  *	any of these mutexes (p_selock is not a mutex) is not allowed.
324  *
325  *
326  *	field		reading		writing		    ordering
327  *	======================================================================
328  *	p_vnode		p_selock(E,S)	p_selock(E)
329  *	p_offset
330  *	p_free
331  *	p_age
332  *	=====================================================================
333  *	p_hash		p_selock(E,S)	p_selock(E) &&	    p_selock, ph_mutex
334  *					ph_mutex[]
335  *	=====================================================================
336  *	p_vpnext	p_selock(E,S)	p_selock(E) &&	    p_selock, vph_mutex
337  *	p_vpprev			vph_mutex[]
338  *	=====================================================================
339  *	When the p_free bit is set:
340  *
341  *	p_next		p_selock(E,S)	p_selock(E) &&	    p_selock,
342  *	p_prev				page_freelock	    page_freelock
343  *
344  *	When the p_free bit is not set:
345  *
346  *	p_next		p_selock(E,S)	p_selock(E) &&	    p_selock, p_iolock
347  *	p_prev				p_iolock
348  *	=====================================================================
349  *	p_selock	pse_mutex[]	pse_mutex[]	    can`t acquire any
350  *	p_cv						    other mutexes or
351  *							    sleep while holding
352  *							    this lock.
353  *	=====================================================================
354  *	p_lckcnt	p_selock(E,S)	p_selock(E) &&
355  *	p_cowcnt			page_llock
356  *	=====================================================================
357  *	p_nrm		hat layer lock	hat layer lock
358  *	p_mapping
359  *	p_pagenum
360  *	=====================================================================
361  *
362  *	where:
363  *		E----> exclusive version of p_selock.
364  *		S----> shared version of p_selock.
365  *
366  *
367  *	Global data structures and variable:
368  *
369  *	field		reading		writing		    ordering
370  *	=====================================================================
371  *	page_hash[]	ph_mutex[]	ph_mutex[]	    can hold this lock
372  *							    before acquiring
373  *							    a vph_mutex or
374  *							    pse_mutex.
375  *	=====================================================================
376  *	vp->v_pages	vph_mutex[]	vph_mutex[]	    can only acquire
377  *							    a pse_mutex while
378  *							    holding this lock.
379  *	=====================================================================
380  *	page_cachelist	page_freelock	page_freelock	    can't acquire any
381  *	page_freelist	page_freelock	page_freelock
382  *	=====================================================================
383  *	freemem		freemem_lock	freemem_lock	    can't acquire any
384  *	freemem_wait					    other mutexes while
385  *	freemem_cv					    holding this mutex.
386  *	=====================================================================
387  *
388  * Page relocation, PG_NORELOC and P_NORELOC.
389  *
390  * Pages may be relocated using the page_relocate() interface. Relocation
391  * involves moving the contents and identity of a page to another, free page.
392  * To relocate a page, the SE_EXCL lock must be obtained. The way to prevent
393  * a page from being relocated is to hold the SE_SHARED lock (the SE_EXCL
394  * lock must not be held indefinitely). If the page is going to be held
395  * SE_SHARED indefinitely, then the PG_NORELOC hint should be passed
396  * to page_create_va so that pages that are prevented from being relocated
397  * can be managed differently by the platform specific layer.
398  *
399  * Pages locked in memory using page_pp_lock (p_lckcnt/p_cowcnt != 0)
400  * are guaranteed to be held in memory, but can still be relocated
401  * providing the SE_EXCL lock can be obtained.
402  *
403  * The P_NORELOC bit in the page_t.p_state field is provided for use by
404  * the platform specific code in managing pages when the PG_NORELOC
405  * hint is used.
406  *
407  * Memory delete and page locking.
408  *
409  * The set of all usable pages is managed using the global page list as
410  * implemented by the memseg structure defined below. When memory is added
411  * or deleted this list changes. Additions to this list guarantee that the
412  * list is never corrupt.  In order to avoid the necessity of an additional
413  * lock to protect against failed accesses to the memseg being deleted and,
414  * more importantly, the page_ts, the memseg structure is never freed and the
415  * page_t virtual address space is remapped to a page (or pages) of
416  * zeros.  If a page_t is manipulated while it is p_selock'd, or if it is
417  * locked indirectly via a hash or freelist lock, it is not possible for
418  * memory delete to collect the page and so that part of the page list is
419  * prevented from being deleted. If the page is referenced outside of one
420  * of these locks, it is possible for the page_t being referenced to be
421  * deleted.  Examples of this are page_t pointers returned by
422  * page_numtopp_nolock, page_first and page_next.  Providing the page_t
423  * is re-checked after taking the p_selock (for p_vnode != NULL), the
424  * remapping to the zero pages will be detected.
425  *
426  *
427  * Page size (p_szc field) and page locking.
428  *
429  * p_szc field of free pages is changed by free list manager under freelist
430  * locks and is of no concern to the rest of VM subsystem.
431  *
432  * p_szc changes of allocated anonymous (swapfs) can only be done only after
433  * exclusively locking all constituent pages and calling hat_pageunload() on
434  * each of them. To prevent p_szc changes of non free anonymous (swapfs) large
435  * pages it's enough to either lock SHARED any of constituent pages or prevent
436  * hat_pageunload() by holding hat level lock that protects mapping lists (this
437  * method is for hat code only)
438  *
439  * To increase (promote) p_szc of allocated non anonymous file system pages
440  * one has to first lock exclusively all involved constituent pages and call
441  * hat_pageunload() on each of them. To prevent p_szc promote it's enough to
442  * either lock SHARED any of constituent pages that will be needed to make a
443  * large page or prevent hat_pageunload() by holding hat level lock that
444  * protects mapping lists (this method is for hat code only).
445  *
446  * To decrease (demote) p_szc of an allocated non anonymous file system large
447  * page one can either use the same method as used for changeing p_szc of
448  * anonymous large pages or if it's not possible to lock all constituent pages
449  * exclusively a different method can be used. In the second method one only
450  * has to exclusively lock one of constituent pages but then one has to
451  * acquire further locks by calling page_szc_lock() and
452  * hat_page_demote(). hat_page_demote() acquires hat level locks and then
453  * demotes the page. This mechanism relies on the fact that any code that
454  * needs to prevent p_szc of a file system large page from changeing either
455  * locks all constituent large pages at least SHARED or locks some pages at
456  * least SHARED and calls page_szc_lock() or uses hat level page locks.
457  * Demotion using this method is implemented by page_demote_vp_pages().
458  * Please see comments in front of page_demote_vp_pages(), hat_page_demote()
459  * and page_szc_lock() for more details.
460  *
461  * Lock order: p_selock, page_szc_lock, ph_mutex/vph_mutex/freelist,
462  * hat level locks.
463  */
464 
465 typedef struct page {
466 	u_offset_t	p_offset;	/* offset into vnode for this page */
467 	struct vnode	*p_vnode;	/* vnode that this page is named by */
468 	selock_t	p_selock;	/* shared/exclusive lock on the page */
469 #if defined(_LP64)
470 	uint_t		p_vpmref;	/* vpm ref - index of the vpmap_t */
471 #endif
472 	struct page	*p_hash;	/* hash by [vnode, offset] */
473 	struct page	*p_vpnext;	/* next page in vnode list */
474 	struct page	*p_vpprev;	/* prev page in vnode list */
475 	struct page	*p_next;	/* next page in free/intrans lists */
476 	struct page	*p_prev;	/* prev page in free/intrans lists */
477 	ushort_t	p_lckcnt;	/* number of locks on page data */
478 	ushort_t	p_cowcnt;	/* number of copy on write lock */
479 	kcondvar_t	p_cv;		/* page struct's condition var */
480 	kcondvar_t	p_io_cv;	/* for iolock */
481 	uchar_t		p_iolock_state;	/* replaces p_iolock */
482 	volatile uchar_t p_szc;		/* page size code */
483 	uchar_t		p_fsdata;	/* file system dependent byte */
484 	uchar_t		p_state;	/* p_free, p_noreloc */
485 	uchar_t		p_nrm;		/* non-cache, ref, mod readonly bits */
486 #if defined(__sparc)
487 	uchar_t		p_vcolor;	/* virtual color */
488 #else
489 	uchar_t		p_embed;	/* x86 - changes p_mapping & p_index */
490 #endif
491 	uchar_t		p_index;	/* MPSS mapping info. Not used on x86 */
492 	uchar_t		p_toxic;	/* page has an unrecoverable error */
493 	void		*p_mapping;	/* hat specific translation info */
494 	pfn_t		p_pagenum;	/* physical page number */
495 
496 	uint_t		p_share;	/* number of translations */
497 #if defined(_LP64)
498 	uint_t		p_sharepad;	/* pad for growing p_share */
499 #endif
500 	uint_t		p_msresv_1;	/* reserved for future use */
501 #if defined(__sparc)
502 	uint_t		p_kpmref;	/* number of kpm mapping sharers */
503 	struct kpme	*p_kpmelist;	/* kpm specific mapping info */
504 #else
505 	/* index of entry in p_map when p_embed is set */
506 	uint_t		p_mlentry;
507 #endif
508 #if defined(_LP64)
509 	kmutex_t	p_ilock;	/* protects p_vpmref */
510 #else
511 	uint64_t	p_msresv_2;	/* page allocation debugging */
512 #endif
513 } page_t;
514 
515 
516 typedef	page_t	devpage_t;
517 #define	devpage	page
518 
519 
520 /*
521  * Page hash table is a power-of-two in size, externally chained
522  * through the hash field.  PAGE_HASHAVELEN is the average length
523  * desired for this chain, from which the size of the page_hash
524  * table is derived at boot time and stored in the kernel variable
525  * page_hashsz.  In the hash function it is given by PAGE_HASHSZ.
526  *
527  * PAGE_HASH_FUNC returns an index into the page_hash[] array.  This
528  * index is also used to derive the mutex that protects the chain.
529  *
530  * In constructing the hash function, first we dispose of unimportant bits
531  * (page offset from "off" and the low 3 bits of "vp" which are zero for
532  * struct alignment). Then shift and sum the remaining bits a couple times
533  * in order to get as many source bits from the two source values into the
534  * resulting hashed value.  Note that this will perform quickly, since the
535  * shifting/summing are fast register to register operations with no additional
536  * memory references).
537  */
538 #if NCPU < 4
539 #define	PH_TABLE_SIZE	16
540 #define	VP_SHIFT	7
541 #else
542 #define	PH_TABLE_SIZE	128
543 #define	VP_SHIFT	9
544 #endif
545 
546 /*
547  * The amount to use for the successive shifts in the hash function below.
548  * The actual value is LOG2(PH_TABLE_SIZE), so that as many bits as
549  * possible will filter thru PAGE_HASH_FUNC() and PAGE_HASH_MUTEX().
550  */
551 #define	PH_SHIFT_SIZE   (7)
552 
553 #define	PAGE_HASHSZ	page_hashsz
554 #define	PAGE_HASHAVELEN		4
555 #define	PAGE_HASH_FUNC(vp, off) \
556 	((((uintptr_t)(off) >> PAGESHIFT) + \
557 		((uintptr_t)(off) >> (PAGESHIFT + PH_SHIFT_SIZE)) + \
558 		((uintptr_t)(vp) >> 3) + \
559 		((uintptr_t)(vp) >> (3 + PH_SHIFT_SIZE)) + \
560 		((uintptr_t)(vp) >> (3 + 2 * PH_SHIFT_SIZE))) & \
561 		(PAGE_HASHSZ - 1))
562 #ifdef _KERNEL
563 
564 /*
565  * The page hash value is re-hashed to an index for the ph_mutex array.
566  *
567  * For 64 bit kernels, the mutex array is padded out to prevent false
568  * sharing of cache sub-blocks (64 bytes) of adjacent mutexes.
569  *
570  * For 32 bit kernels, we don't want to waste kernel address space with
571  * padding, so instead we rely on the hash function to introduce skew of
572  * adjacent vnode/offset indexes (the left shift part of the hash function).
573  * Since sizeof (kmutex_t) is 8, we shift an additional 3 to skew to a different
574  * 64 byte sub-block.
575  */
576 typedef struct pad_mutex {
577 	kmutex_t	pad_mutex;
578 #ifdef _LP64
579 	char		pad_pad[64 - sizeof (kmutex_t)];
580 #endif
581 } pad_mutex_t;
582 extern pad_mutex_t ph_mutex[];
583 
584 #define	PAGE_HASH_MUTEX(x) \
585 	&(ph_mutex[((x) + ((x) >> VP_SHIFT) + ((x) << 3)) & \
586 		(PH_TABLE_SIZE - 1)].pad_mutex)
587 
588 /*
589  * Flags used while creating pages.
590  */
591 #define	PG_EXCL		0x0001
592 #define	PG_WAIT		0x0002
593 #define	PG_PHYSCONTIG	0x0004		/* NOT SUPPORTED */
594 #define	PG_MATCH_COLOR	0x0008		/* SUPPORTED by free list routines */
595 #define	PG_NORELOC	0x0010		/* Non-relocatable alloc hint. */
596 					/* Page must be PP_ISNORELOC */
597 #define	PG_PANIC	0x0020		/* system will panic if alloc fails */
598 #define	PG_PUSHPAGE	0x0040		/* alloc may use reserve */
599 
600 /*
601  * When p_selock has the SE_EWANTED bit set, threads waiting for SE_EXCL
602  * access are given priority over all other waiting threads.
603  */
604 #define	SE_EWANTED	0x40000000
605 #define	PAGE_LOCKED(pp)		(((pp)->p_selock & ~SE_EWANTED) != 0)
606 #define	PAGE_SHARED(pp)		(((pp)->p_selock & ~SE_EWANTED) > 0)
607 #define	PAGE_EXCL(pp)		((pp)->p_selock < 0)
608 #define	PAGE_LOCKED_SE(pp, se)	\
609 	((se) == SE_EXCL ? PAGE_EXCL(pp) : PAGE_SHARED(pp))
610 
611 extern	long page_hashsz;
612 extern	page_t **page_hash;
613 
614 extern	kmutex_t page_llock;		/* page logical lock mutex */
615 extern	kmutex_t freemem_lock;		/* freemem lock */
616 
617 extern	pgcnt_t	total_pages;		/* total pages in the system */
618 
619 /*
620  * Variables controlling locking of physical memory.
621  */
622 extern	pgcnt_t	pages_pp_maximum;	/* tuning: lock + claim <= max */
623 extern	void init_pages_pp_maximum(void);
624 
625 struct lgrp;
626 
627 /* page_list_{add,sub} flags */
628 
629 /* which list */
630 #define	PG_FREE_LIST	0x0001
631 #define	PG_CACHE_LIST	0x0002
632 
633 /* where on list */
634 #define	PG_LIST_TAIL	0x0010
635 #define	PG_LIST_HEAD	0x0020
636 
637 /* called from */
638 #define	PG_LIST_ISINIT	0x1000
639 
640 /*
641  * Page frame operations.
642  */
643 page_t	*page_lookup(struct vnode *, u_offset_t, se_t);
644 page_t	*page_lookup_create(struct vnode *, u_offset_t, se_t, page_t *,
645 	spgcnt_t *, int);
646 page_t	*page_lookup_nowait(struct vnode *, u_offset_t, se_t);
647 page_t	*page_find(struct vnode *, u_offset_t);
648 page_t	*page_exists(struct vnode *, u_offset_t);
649 int	page_exists_physcontig(vnode_t *, u_offset_t, uint_t, page_t *[]);
650 int	page_exists_forreal(struct vnode *, u_offset_t, uint_t *);
651 void	page_needfree(spgcnt_t);
652 page_t	*page_create(struct vnode *, u_offset_t, size_t, uint_t);
653 int	page_alloc_pages(struct vnode *, struct seg *, caddr_t, page_t **,
654 	page_t **, uint_t, int);
655 page_t  *page_create_va_large(vnode_t *vp, u_offset_t off, size_t bytes,
656 	uint_t flags, struct seg *seg, caddr_t vaddr, void *arg);
657 page_t	*page_create_va(struct vnode *, u_offset_t, size_t, uint_t,
658 	struct seg *, caddr_t);
659 int	page_create_wait(size_t npages, uint_t flags);
660 void    page_create_putback(ssize_t npages);
661 void	page_free(page_t *, int);
662 void	page_free_at_startup(page_t *);
663 void	page_free_pages(page_t *);
664 void	free_vp_pages(struct vnode *, u_offset_t, size_t);
665 int	page_reclaim(page_t *, kmutex_t *);
666 void	page_destroy(page_t *, int);
667 void	page_destroy_pages(page_t *);
668 void	page_destroy_free(page_t *);
669 void	page_rename(page_t *, struct vnode *, u_offset_t);
670 int	page_hashin(page_t *, struct vnode *, u_offset_t, kmutex_t *);
671 void	page_hashout(page_t *, kmutex_t *);
672 int	page_num_hashin(pfn_t, struct vnode *, u_offset_t);
673 void	page_add(page_t **, page_t *);
674 void	page_add_common(page_t **, page_t *);
675 void	page_sub(page_t **, page_t *);
676 void	page_sub_common(page_t **, page_t *);
677 page_t	*page_get_freelist(struct vnode *, u_offset_t, struct seg *,
678 		caddr_t, size_t, uint_t, struct lgrp *);
679 
680 page_t	*page_get_cachelist(struct vnode *, u_offset_t, struct seg *,
681 		caddr_t, uint_t, struct lgrp *);
682 void	page_list_add(page_t *, int);
683 void	page_boot_demote(page_t *);
684 void	page_promote_size(page_t *, uint_t);
685 void	page_list_add_pages(page_t *, int);
686 void	page_list_sub(page_t *, int);
687 void	page_list_sub_pages(page_t *, uint_t);
688 void	page_list_xfer(page_t *, int, int);
689 void	page_list_break(page_t **, page_t **, size_t);
690 void	page_list_concat(page_t **, page_t **);
691 void	page_vpadd(page_t **, page_t *);
692 void	page_vpsub(page_t **, page_t *);
693 int	page_lock(page_t *, se_t, kmutex_t *, reclaim_t);
694 int	page_lock_es(page_t *, se_t, kmutex_t *, reclaim_t, int);
695 void page_lock_clr_exclwanted(page_t *);
696 int	page_trylock(page_t *, se_t);
697 int	page_try_reclaim_lock(page_t *, se_t, int);
698 int	page_tryupgrade(page_t *);
699 void	page_downgrade(page_t *);
700 void	page_unlock(page_t *);
701 void	page_unlock_noretire(page_t *);
702 void	page_lock_delete(page_t *);
703 int	page_pp_lock(page_t *, int, int);
704 void	page_pp_unlock(page_t *, int, int);
705 int	page_resv(pgcnt_t, uint_t);
706 void	page_unresv(pgcnt_t);
707 void	page_pp_useclaim(page_t *, page_t *, uint_t);
708 int	page_addclaim(page_t *);
709 int	page_subclaim(page_t *);
710 int	page_addclaim_pages(page_t **);
711 int	page_subclaim_pages(page_t **);
712 pfn_t	page_pptonum(page_t *);
713 page_t	*page_numtopp(pfn_t, se_t);
714 page_t	*page_numtopp_noreclaim(pfn_t, se_t);
715 page_t	*page_numtopp_nolock(pfn_t);
716 page_t	*page_numtopp_nowait(pfn_t, se_t);
717 page_t  *page_first();
718 page_t  *page_next(page_t *);
719 page_t  *page_list_next(page_t *);
720 page_t	*page_nextn(page_t *, ulong_t);
721 page_t	*page_next_scan_init(void **);
722 page_t	*page_next_scan_large(page_t *, ulong_t *, void **);
723 void    prefetch_page_r(void *);
724 void	ppcopy(page_t *, page_t *);
725 void	page_relocate_hash(page_t *, page_t *);
726 void	pagezero(page_t *, uint_t, uint_t);
727 void	pagescrub(page_t *, uint_t, uint_t);
728 void	page_io_lock(page_t *);
729 void	page_io_unlock(page_t *);
730 int	page_io_trylock(page_t *);
731 int	page_iolock_assert(page_t *);
732 void	page_iolock_init(page_t *);
733 pgcnt_t	page_busy(int);
734 void	page_lock_init(void);
735 ulong_t	page_share_cnt(page_t *);
736 int	page_isshared(page_t *);
737 int	page_isfree(page_t *);
738 int	page_isref(page_t *);
739 int	page_ismod(page_t *);
740 int	page_release(page_t *, int);
741 void	page_retire_init(void);
742 int	page_retire(uint64_t, uchar_t);
743 int	page_retire_check(uint64_t, uint64_t *);
744 int	page_unretire(uint64_t);
745 int	page_unretire_pp(page_t *, int);
746 void	page_tryretire(page_t *);
747 void	page_retire_hunt(void (*)(page_t *));
748 void	page_retire_mdboot_cb(page_t *);
749 void	page_clrtoxic(page_t *, uchar_t);
750 void	page_settoxic(page_t *, uchar_t);
751 
752 int	page_mem_avail(pgcnt_t);
753 int	page_reclaim_mem(pgcnt_t, pgcnt_t, int);
754 
755 void page_set_props(page_t *, uint_t);
756 void page_clr_all_props(page_t *);
757 int page_clear_lck_cow(page_t *, int);
758 
759 kmutex_t	*page_vnode_mutex(struct vnode *);
760 kmutex_t	*page_se_mutex(struct page *);
761 kmutex_t	*page_szc_lock(struct page *);
762 int		page_szc_lock_assert(struct page *pp);
763 
764 /*
765  * Page relocation interfaces. page_relocate() is generic.
766  * page_get_replacement_page() is provided by the PSM.
767  * page_free_replacement_page() is generic.
768  */
769 int group_page_trylock(page_t *, se_t);
770 void group_page_unlock(page_t *);
771 int page_relocate(page_t **, page_t **, int, int, spgcnt_t *, struct lgrp *);
772 int do_page_relocate(page_t **, page_t **, int, spgcnt_t *, struct lgrp *);
773 page_t *page_get_replacement_page(page_t *, struct lgrp *, uint_t);
774 void page_free_replacement_page(page_t *);
775 int page_relocate_cage(page_t **, page_t **);
776 
777 int page_try_demote_pages(page_t *);
778 int page_try_demote_free_pages(page_t *);
779 void page_demote_free_pages(page_t *);
780 
781 struct anon_map;
782 
783 void page_mark_migrate(struct seg *, caddr_t, size_t, struct anon_map *,
784     ulong_t, vnode_t *, u_offset_t, int);
785 void page_migrate(struct seg *, caddr_t, page_t **, pgcnt_t);
786 
787 /*
788  * Tell the PIM we are adding physical memory
789  */
790 void add_physmem(page_t *, size_t, pfn_t);
791 void add_physmem_cb(page_t *, pfn_t);	/* callback for page_t part */
792 
793 /*
794  * hw_page_array[] is configured with hardware supported page sizes by
795  * platform specific code.
796  */
797 typedef struct {
798 	size_t	hp_size;
799 	uint_t	hp_shift;
800 	pgcnt_t	hp_pgcnt;	/* base pagesize cnt */
801 } hw_pagesize_t;
802 
803 extern hw_pagesize_t	hw_page_array[];
804 extern uint_t		page_colors, page_colors_mask;
805 extern uint_t		page_coloring_shift;
806 extern int		cpu_page_colors;
807 
808 uint_t	page_num_pagesizes(void);
809 uint_t	page_num_user_pagesizes(void);
810 size_t	page_get_pagesize(uint_t);
811 size_t	page_get_user_pagesize(uint_t n);
812 pgcnt_t	page_get_pagecnt(uint_t);
813 uint_t	page_get_shift(uint_t);
814 int	page_szc(size_t);
815 int	page_szc_user_filtered(size_t);
816 
817 
818 /* page_get_replacement page flags */
819 #define	PGR_SAMESZC	0x1	/* only look for page size same as orig */
820 #define	PGR_NORELOC	0x2	/* allocate a P_NORELOC page */
821 
822 #endif	/* _KERNEL */
823 
824 /*
825  * Constants used for the p_iolock_state
826  */
827 #define	PAGE_IO_INUSE	0x1
828 #define	PAGE_IO_WANTED	0x2
829 
830 /*
831  * Constants used for page_release status
832  */
833 #define	PGREL_NOTREL    0x1
834 #define	PGREL_CLEAN	0x2
835 #define	PGREL_MOD	0x3
836 
837 /*
838  * The p_state field holds what used to be the p_age and p_free
839  * bits.  These fields are protected by p_selock (see above).
840  */
841 #define	P_FREE		0x80		/* Page on free list */
842 #define	P_NORELOC	0x40		/* Page is non-relocatable */
843 #define	P_MIGRATE	0x20		/* Migrate page on next touch */
844 #define	P_SWAP		0x10		/* belongs to vnode that is V_ISSWAP */
845 
846 #define	PP_ISFREE(pp)		((pp)->p_state & P_FREE)
847 #define	PP_ISAGED(pp)		(((pp)->p_state & P_FREE) && \
848 					((pp)->p_vnode == NULL))
849 #define	PP_ISNORELOC(pp)	((pp)->p_state & P_NORELOC)
850 #define	PP_ISKVP(pp)		((pp)->p_vnode == &kvp)
851 #define	PP_ISNORELOCKERNEL(pp)	(PP_ISNORELOC(pp) && PP_ISKVP(pp))
852 #define	PP_ISMIGRATE(pp)	((pp)->p_state & P_MIGRATE)
853 #define	PP_ISSWAP(pp)		((pp)->p_state & P_SWAP)
854 
855 #define	PP_SETFREE(pp)		((pp)->p_state = ((pp)->p_state & ~P_MIGRATE) \
856 				| P_FREE)
857 #define	PP_SETAGED(pp)		ASSERT(PP_ISAGED(pp))
858 #define	PP_SETNORELOC(pp)	((pp)->p_state |= P_NORELOC)
859 #define	PP_SETMIGRATE(pp)	((pp)->p_state |= P_MIGRATE)
860 #define	PP_SETSWAP(pp)		((pp)->p_state |= P_SWAP)
861 
862 #define	PP_CLRFREE(pp)		((pp)->p_state &= ~P_FREE)
863 #define	PP_CLRAGED(pp)		ASSERT(!PP_ISAGED(pp))
864 #define	PP_CLRNORELOC(pp)	((pp)->p_state &= ~P_NORELOC)
865 #define	PP_CLRMIGRATE(pp)	((pp)->p_state &= ~P_MIGRATE)
866 #define	PP_CLRSWAP(pp)		((pp)->p_state &= ~P_SWAP)
867 
868 /*
869  * Flags for page_t p_toxic, for tracking memory hardware errors.
870  *
871  * These flags are OR'ed into p_toxic with page_settoxic() to track which
872  * error(s) have occurred on a given page. The flags are cleared with
873  * page_clrtoxic(). Both page_settoxic() and page_cleartoxic use atomic
874  * primitives to manipulate the p_toxic field so no other locking is needed.
875  *
876  * When an error occurs on a page, p_toxic is set to record the error. The
877  * error could be a memory error or something else (i.e. a datapath). The Page
878  * Retire mechanism does not try to determine the exact cause of the error;
879  * Page Retire rightly leaves that sort of determination to FMA's Diagnostic
880  * Engine (DE).
881  *
882  * Note that, while p_toxic bits can be set without holding any locks, they
883  * should only be cleared while holding the page exclusively locked.
884  *
885  * Pages with PR_UE or PR_FMA flags are retired unconditionally, while pages
886  * with PR_MCE are retired if the system has not retired too many of them.
887  *
888  * A page must be exclusively locked to be retired. Pages can be retired if
889  * they are mapped, modified, or both, as long as they are not marked PR_UE,
890  * since pages with uncorrectable errors cannot be relocated in memory.
891  * Once a page has been successfully retired it is zeroed, attached to the
892  * retired_pages vnode and, finally, PR_RETIRED is set in p_toxic. The other
893  * p_toxic bits are NOT cleared. Pages are not left locked after retiring them
894  * to avoid special case code throughout the kernel; rather, page_*lock() will
895  * fail to lock the page, unless SE_RETIRED is passed as an argument.
896  *
897  * While we have your attention, go take a look at the comments at the
898  * beginning of page_retire.c too.
899  */
900 #define	PR_OK		0x00	/* no problem */
901 #define	PR_MCE		0x01	/* page has seen two or more CEs */
902 #define	PR_UE		0x02	/* page has an unhandled UE */
903 #define	PR_UE_SCRUBBED	0x04	/* page has seen a UE but was cleaned */
904 #define	PR_FMA		0x08	/* A DE wants this page retired */
905 #define	PR_RESV		0x10	/* Reserved for future use */
906 #define	PR_BUSY		0x20	/* Page retire is in progress */
907 #define	PR_MSG		0x40	/* message(s) already printed for this page */
908 #define	PR_RETIRED	0x80	/* This page has been retired */
909 
910 #define	PR_REASONS	(PR_UE | PR_MCE | PR_FMA)
911 #define	PR_TOXIC	(PR_UE)
912 #define	PR_ERRMASK	(PR_UE | PR_UE_SCRUBBED | PR_MCE | PR_FMA)
913 #define	PR_ALLFLAGS	(0xFF)
914 
915 #define	PP_RETIRED(pp)	((pp)->p_toxic & PR_RETIRED)
916 #define	PP_TOXIC(pp)	((pp)->p_toxic & PR_TOXIC)
917 #define	PP_PR_REQ(pp)	(((pp)->p_toxic & PR_REASONS) && !PP_RETIRED(pp))
918 #define	PP_PR_NOSHARE(pp)						\
919 	((((pp)->p_toxic & (PR_RETIRED | PR_FMA | PR_UE)) == PR_FMA) &&	\
920 	!PP_ISKVP(pp))
921 
922 /*
923  * kpm large page description.
924  * The virtual address range of segkpm is divided into chunks of
925  * kpm_pgsz. Each chunk is controlled by a kpm_page_t. The ushort
926  * is sufficient for 2^^15 * PAGESIZE, so e.g. the maximum kpm_pgsz
927  * for 8K is 256M and 2G for 64K pages. It it kept as small as
928  * possible to save physical memory space.
929  *
930  * There are 2 segkpm mapping windows within in the virtual address
931  * space when we have to prevent VAC alias conflicts. The so called
932  * Alias window (mappings are always by PAGESIZE) is controlled by
933  * kp_refcnta. The regular window is controlled by kp_refcnt for the
934  * normal operation, which is to use the largest available pagesize.
935  * When VAC alias conflicts are present within a chunk in the regular
936  * window the large page mapping is broken up into smaller PAGESIZE
937  * mappings. kp_refcntc is used to control the pages that are invoked
938  * in the conflict and kp_refcnts holds the active mappings done
939  * with the small page size. In non vac conflict mode kp_refcntc is
940  * also used as "go" indication (-1) for the trap level tsbmiss
941  * handler.
942  */
943 typedef struct kpm_page {
944 	short kp_refcnt;	/* pages mapped large */
945 	short kp_refcnta;	/* pages mapped in Alias window */
946 	short kp_refcntc;	/* TL-tsbmiss flag; #vac alias conflict pages */
947 	short kp_refcnts;	/* vac alias: pages mapped small */
948 } kpm_page_t;
949 
950 /*
951  * Note: khl_lock offset changes must be reflected in sfmmu_asm.s
952  */
953 typedef struct kpm_hlk {
954 	kmutex_t khl_mutex;	/* kpm_page mutex */
955 	uint_t   khl_lock;	/* trap level tsbmiss handling */
956 } kpm_hlk_t;
957 
958 /*
959  * kpm small page description.
960  * When kpm_pgsz is equal to PAGESIZE a smaller representation is used
961  * to save memory space. Alias range mappings and regular segkpm
962  * mappings are done in units of PAGESIZE and can share the mapping
963  * information and the mappings are always distinguishable by their
964  * virtual address. Other information neeeded for VAC conflict prevention
965  * is already available on a per page basis. There are basically 3 states
966  * a kpm_spage can have: not mapped (0), mapped in Alias range or virtually
967  * uncached (1) and mapped in the regular segkpm window (-1). The -1 value
968  * is also used as "go" indication for the segkpm trap level tsbmiss
969  * handler for small pages (value is kept the same as it is used for large
970  * mappings).
971  */
972 typedef struct kpm_spage {
973 	char	kp_mapped;	/* page mapped small */
974 } kpm_spage_t;
975 
976 /*
977  * Note: kshl_lock offset changes must be reflected in sfmmu_asm.s
978  */
979 typedef struct kpm_shlk {
980 	uint_t   kshl_lock;	/* trap level tsbmiss handling */
981 } kpm_shlk_t;
982 
983 /*
984  * Each segment of physical memory is described by a memseg struct.
985  * Within a segment, memory is considered contiguous. The members
986  * can be categorized as follows:
987  * . Platform independent:
988  *         pages, epages, pages_base, pages_end, next, lnext.
989  * . 64bit only but platform independent:
990  *         kpm_pbase, kpm_nkpmpgs, kpm_pages, kpm_spages.
991  * . Really platform or mmu specific:
992  *         pagespa, epagespa, nextpa, kpm_pagespa.
993  * . Mixed:
994  *         msegflags.
995  */
996 struct memseg {
997 	page_t *pages, *epages;		/* [from, to] in page array */
998 	pfn_t pages_base, pages_end;	/* [from, to] in page numbers */
999 	struct memseg *next;		/* next segment in list */
1000 #if defined(__sparc)
1001 	struct memseg *lnext;		/* next segment in deleted list */
1002 	uint64_t pagespa, epagespa;	/* [from, to] page array physical */
1003 	uint64_t nextpa;		/* physical next pointer */
1004 	pfn_t	kpm_pbase;		/* start of kpm range */
1005 	pgcnt_t kpm_nkpmpgs;		/* # of kpm_pgsz pages */
1006 	union _mseg_un {
1007 		kpm_page_t  *kpm_lpgs;	/* ptr to kpm_page array */
1008 		kpm_spage_t *kpm_spgs;	/* ptr to kpm_spage array */
1009 	} mseg_un;
1010 	uint64_t kpm_pagespa;		/* physical ptr to kpm (s)pages array */
1011 	uint_t msegflags;		/* memseg flags */
1012 #endif /* __sparc */
1013 };
1014 
1015 /* memseg union aliases */
1016 #define	kpm_pages	mseg_un.kpm_lpgs
1017 #define	kpm_spages	mseg_un.kpm_spgs
1018 
1019 /* msegflags */
1020 #define	MEMSEG_DYNAMIC		0x1	/* DR: memory was added dynamically */
1021 
1022 /* memseg support macros */
1023 #define	MSEG_NPAGES(SEG)	((SEG)->pages_end - (SEG)->pages_base)
1024 
1025 /* memseg hash */
1026 #define	MEM_HASH_SHIFT		0x9
1027 #define	N_MEM_SLOTS		0x200		/* must be a power of 2 */
1028 #define	MEMSEG_PFN_HASH(pfn)	(((pfn)/mhash_per_slot) & (N_MEM_SLOTS - 1))
1029 
1030 /* memseg  externals */
1031 extern struct memseg *memsegs;		/* list of memory segments */
1032 extern ulong_t mhash_per_slot;
1033 extern uint64_t memsegspa;		/* memsegs as physical address */
1034 
1035 void build_pfn_hash();
1036 extern struct memseg *page_numtomemseg_nolock(pfn_t pfnum);
1037 
1038 
1039 #ifdef	__cplusplus
1040 }
1041 #endif
1042 
1043 #endif	/* _VM_PAGE_H */
1044