xref: /illumos-gate/usr/src/uts/common/vm/vm_page.c (revision 0209230b)
17c478bd9Sstevel@tonic-gate /*
27c478bd9Sstevel@tonic-gate  * CDDL HEADER START
37c478bd9Sstevel@tonic-gate  *
47c478bd9Sstevel@tonic-gate  * The contents of this file are subject to the terms of the
53cff2f43Sstans  * Common Development and Distribution License (the "License").
63cff2f43Sstans  * You may not use this file except in compliance with the License.
77c478bd9Sstevel@tonic-gate  *
87c478bd9Sstevel@tonic-gate  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
97c478bd9Sstevel@tonic-gate  * or http://www.opensolaris.org/os/licensing.
107c478bd9Sstevel@tonic-gate  * See the License for the specific language governing permissions
117c478bd9Sstevel@tonic-gate  * and limitations under the License.
127c478bd9Sstevel@tonic-gate  *
137c478bd9Sstevel@tonic-gate  * When distributing Covered Code, include this CDDL HEADER in each
147c478bd9Sstevel@tonic-gate  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
157c478bd9Sstevel@tonic-gate  * If applicable, add the following below this CDDL HEADER, with the
167c478bd9Sstevel@tonic-gate  * fields enclosed by brackets "[]" replaced with your own identifying
177c478bd9Sstevel@tonic-gate  * information: Portions Copyright [yyyy] [name of copyright owner]
187c478bd9Sstevel@tonic-gate  *
197c478bd9Sstevel@tonic-gate  * CDDL HEADER END
207c478bd9Sstevel@tonic-gate  */
217c478bd9Sstevel@tonic-gate /*
224fc2445aSelowe  * Copyright 2006 Sun Microsystems, Inc.  All rights reserved.
237c478bd9Sstevel@tonic-gate  * Use is subject to license terms.
247c478bd9Sstevel@tonic-gate  */
257c478bd9Sstevel@tonic-gate 
267c478bd9Sstevel@tonic-gate /*	Copyright (c) 1983, 1984, 1985, 1986, 1987, 1988, 1989  AT&T	*/
277c478bd9Sstevel@tonic-gate /*	  All Rights Reserved  	*/
287c478bd9Sstevel@tonic-gate 
297c478bd9Sstevel@tonic-gate /*
307c478bd9Sstevel@tonic-gate  * University Copyright- Copyright (c) 1982, 1986, 1988
317c478bd9Sstevel@tonic-gate  * The Regents of the University of California
327c478bd9Sstevel@tonic-gate  * All Rights Reserved
337c478bd9Sstevel@tonic-gate  *
347c478bd9Sstevel@tonic-gate  * University Acknowledgment- Portions of this document are derived from
357c478bd9Sstevel@tonic-gate  * software developed by the University of California, Berkeley, and its
367c478bd9Sstevel@tonic-gate  * contributors.
377c478bd9Sstevel@tonic-gate  */
387c478bd9Sstevel@tonic-gate 
397c478bd9Sstevel@tonic-gate #pragma ident	"%Z%%M%	%I%	%E% SMI"
407c478bd9Sstevel@tonic-gate 
417c478bd9Sstevel@tonic-gate /*
427c478bd9Sstevel@tonic-gate  * VM - physical page management.
437c478bd9Sstevel@tonic-gate  */
447c478bd9Sstevel@tonic-gate 
457c478bd9Sstevel@tonic-gate #include <sys/types.h>
467c478bd9Sstevel@tonic-gate #include <sys/t_lock.h>
477c478bd9Sstevel@tonic-gate #include <sys/param.h>
487c478bd9Sstevel@tonic-gate #include <sys/systm.h>
497c478bd9Sstevel@tonic-gate #include <sys/errno.h>
507c478bd9Sstevel@tonic-gate #include <sys/time.h>
517c478bd9Sstevel@tonic-gate #include <sys/vnode.h>
527c478bd9Sstevel@tonic-gate #include <sys/vm.h>
537c478bd9Sstevel@tonic-gate #include <sys/vtrace.h>
547c478bd9Sstevel@tonic-gate #include <sys/swap.h>
557c478bd9Sstevel@tonic-gate #include <sys/cmn_err.h>
567c478bd9Sstevel@tonic-gate #include <sys/tuneable.h>
577c478bd9Sstevel@tonic-gate #include <sys/sysmacros.h>
587c478bd9Sstevel@tonic-gate #include <sys/cpuvar.h>
597c478bd9Sstevel@tonic-gate #include <sys/callb.h>
607c478bd9Sstevel@tonic-gate #include <sys/debug.h>
617c478bd9Sstevel@tonic-gate #include <sys/tnf_probe.h>
627c478bd9Sstevel@tonic-gate #include <sys/condvar_impl.h>
637c478bd9Sstevel@tonic-gate #include <sys/mem_config.h>
647c478bd9Sstevel@tonic-gate #include <sys/mem_cage.h>
657c478bd9Sstevel@tonic-gate #include <sys/kmem.h>
667c478bd9Sstevel@tonic-gate #include <sys/atomic.h>
677c478bd9Sstevel@tonic-gate #include <sys/strlog.h>
687c478bd9Sstevel@tonic-gate #include <sys/mman.h>
697c478bd9Sstevel@tonic-gate #include <sys/ontrap.h>
707c478bd9Sstevel@tonic-gate #include <sys/lgrp.h>
717c478bd9Sstevel@tonic-gate #include <sys/vfs.h>
727c478bd9Sstevel@tonic-gate 
737c478bd9Sstevel@tonic-gate #include <vm/hat.h>
747c478bd9Sstevel@tonic-gate #include <vm/anon.h>
757c478bd9Sstevel@tonic-gate #include <vm/page.h>
767c478bd9Sstevel@tonic-gate #include <vm/seg.h>
777c478bd9Sstevel@tonic-gate #include <vm/pvn.h>
787c478bd9Sstevel@tonic-gate #include <vm/seg_kmem.h>
797c478bd9Sstevel@tonic-gate #include <vm/vm_dep.h>
80*0209230bSgjelinek #include <sys/vm_usage.h>
817c478bd9Sstevel@tonic-gate #include <fs/fs_subr.h>
827c478bd9Sstevel@tonic-gate 
837c478bd9Sstevel@tonic-gate static int nopageage = 0;
847c478bd9Sstevel@tonic-gate 
857c478bd9Sstevel@tonic-gate static pgcnt_t max_page_get;	/* max page_get request size in pages */
867c478bd9Sstevel@tonic-gate pgcnt_t total_pages = 0;	/* total number of pages (used by /proc) */
877c478bd9Sstevel@tonic-gate 
887c478bd9Sstevel@tonic-gate /*
897c478bd9Sstevel@tonic-gate  * freemem_lock protects all freemem variables:
907c478bd9Sstevel@tonic-gate  * availrmem. Also this lock protects the globals which track the
917c478bd9Sstevel@tonic-gate  * availrmem changes for accurate kernel footprint calculation.
927c478bd9Sstevel@tonic-gate  * See below for an explanation of these
937c478bd9Sstevel@tonic-gate  * globals.
947c478bd9Sstevel@tonic-gate  */
957c478bd9Sstevel@tonic-gate kmutex_t freemem_lock;
967c478bd9Sstevel@tonic-gate pgcnt_t availrmem;
977c478bd9Sstevel@tonic-gate pgcnt_t availrmem_initial;
987c478bd9Sstevel@tonic-gate 
997c478bd9Sstevel@tonic-gate /*
1007c478bd9Sstevel@tonic-gate  * These globals track availrmem changes to get a more accurate
1017c478bd9Sstevel@tonic-gate  * estimate of tke kernel size. Historically pp_kernel is used for
1027c478bd9Sstevel@tonic-gate  * kernel size and is based on availrmem. But availrmem is adjusted for
1037c478bd9Sstevel@tonic-gate  * locked pages in the system not just for kernel locked pages.
1047c478bd9Sstevel@tonic-gate  * These new counters will track the pages locked through segvn and
1057c478bd9Sstevel@tonic-gate  * by explicit user locking.
1067c478bd9Sstevel@tonic-gate  *
1077c478bd9Sstevel@tonic-gate  * segvn_pages_locked : This keeps track on a global basis how many pages
1087c478bd9Sstevel@tonic-gate  * are currently locked because of I/O.
1097c478bd9Sstevel@tonic-gate  *
1107c478bd9Sstevel@tonic-gate  * pages_locked : How many pages are locked becuase of user specified
1117c478bd9Sstevel@tonic-gate  * locking through mlock or plock.
1127c478bd9Sstevel@tonic-gate  *
1137c478bd9Sstevel@tonic-gate  * pages_useclaim,pages_claimed : These two variables track the
1147c478bd9Sstevel@tonic-gate  * cliam adjustments because of the protection changes on a segvn segment.
1157c478bd9Sstevel@tonic-gate  *
1167c478bd9Sstevel@tonic-gate  * All these globals are protected by the same lock which protects availrmem.
1177c478bd9Sstevel@tonic-gate  */
1187c478bd9Sstevel@tonic-gate pgcnt_t segvn_pages_locked;
1197c478bd9Sstevel@tonic-gate pgcnt_t pages_locked;
1207c478bd9Sstevel@tonic-gate pgcnt_t pages_useclaim;
1217c478bd9Sstevel@tonic-gate pgcnt_t pages_claimed;
1227c478bd9Sstevel@tonic-gate 
1237c478bd9Sstevel@tonic-gate 
1247c478bd9Sstevel@tonic-gate /*
1257c478bd9Sstevel@tonic-gate  * new_freemem_lock protects freemem, freemem_wait & freemem_cv.
1267c478bd9Sstevel@tonic-gate  */
1277c478bd9Sstevel@tonic-gate static kmutex_t	new_freemem_lock;
1287c478bd9Sstevel@tonic-gate static uint_t	freemem_wait;	/* someone waiting for freemem */
1297c478bd9Sstevel@tonic-gate static kcondvar_t freemem_cv;
1307c478bd9Sstevel@tonic-gate 
1317c478bd9Sstevel@tonic-gate /*
1327c478bd9Sstevel@tonic-gate  * The logical page free list is maintained as two lists, the 'free'
1337c478bd9Sstevel@tonic-gate  * and the 'cache' lists.
1347c478bd9Sstevel@tonic-gate  * The free list contains those pages that should be reused first.
1357c478bd9Sstevel@tonic-gate  *
1367c478bd9Sstevel@tonic-gate  * The implementation of the lists is machine dependent.
1377c478bd9Sstevel@tonic-gate  * page_get_freelist(), page_get_cachelist(),
1387c478bd9Sstevel@tonic-gate  * page_list_sub(), and page_list_add()
1397c478bd9Sstevel@tonic-gate  * form the interface to the machine dependent implementation.
1407c478bd9Sstevel@tonic-gate  *
1417c478bd9Sstevel@tonic-gate  * Pages with p_free set are on the cache list.
1427c478bd9Sstevel@tonic-gate  * Pages with p_free and p_age set are on the free list,
1437c478bd9Sstevel@tonic-gate  *
1447c478bd9Sstevel@tonic-gate  * A page may be locked while on either list.
1457c478bd9Sstevel@tonic-gate  */
1467c478bd9Sstevel@tonic-gate 
1477c478bd9Sstevel@tonic-gate /*
1487c478bd9Sstevel@tonic-gate  * free list accounting stuff.
1497c478bd9Sstevel@tonic-gate  *
1507c478bd9Sstevel@tonic-gate  *
1517c478bd9Sstevel@tonic-gate  * Spread out the value for the number of pages on the
1527c478bd9Sstevel@tonic-gate  * page free and page cache lists.  If there is just one
1537c478bd9Sstevel@tonic-gate  * value, then it must be under just one lock.
1547c478bd9Sstevel@tonic-gate  * The lock contention and cache traffic are a real bother.
1557c478bd9Sstevel@tonic-gate  *
1567c478bd9Sstevel@tonic-gate  * When we acquire and then drop a single pcf lock
1577c478bd9Sstevel@tonic-gate  * we can start in the middle of the array of pcf structures.
1587c478bd9Sstevel@tonic-gate  * If we acquire more than one pcf lock at a time, we need to
1597c478bd9Sstevel@tonic-gate  * start at the front to avoid deadlocking.
1607c478bd9Sstevel@tonic-gate  *
1617c478bd9Sstevel@tonic-gate  * pcf_count holds the number of pages in each pool.
1627c478bd9Sstevel@tonic-gate  *
1637c478bd9Sstevel@tonic-gate  * pcf_block is set when page_create_get_something() has asked the
1647c478bd9Sstevel@tonic-gate  * PSM page freelist and page cachelist routines without specifying
1657c478bd9Sstevel@tonic-gate  * a color and nothing came back.  This is used to block anything
1667c478bd9Sstevel@tonic-gate  * else from moving pages from one list to the other while the
1677c478bd9Sstevel@tonic-gate  * lists are searched again.  If a page is freeed while pcf_block is
1687c478bd9Sstevel@tonic-gate  * set, then pcf_reserve is incremented.  pcgs_unblock() takes care
1697c478bd9Sstevel@tonic-gate  * of clearning pcf_block, doing the wakeups, etc.
1707c478bd9Sstevel@tonic-gate  */
1717c478bd9Sstevel@tonic-gate 
1727c478bd9Sstevel@tonic-gate #if NCPU <= 4
173f2b37d75Sfr #define	PAD	2
1747c478bd9Sstevel@tonic-gate #define	PCF_FANOUT	4
1757c478bd9Sstevel@tonic-gate static	uint_t	pcf_mask = PCF_FANOUT - 1;
1767c478bd9Sstevel@tonic-gate #else
177f2b37d75Sfr #define	PAD	10
1787c478bd9Sstevel@tonic-gate #ifdef sun4v
1797c478bd9Sstevel@tonic-gate #define	PCF_FANOUT	32
1807c478bd9Sstevel@tonic-gate #else
1817c478bd9Sstevel@tonic-gate #define	PCF_FANOUT	128
1827c478bd9Sstevel@tonic-gate #endif
1837c478bd9Sstevel@tonic-gate static	uint_t	pcf_mask = PCF_FANOUT - 1;
1847c478bd9Sstevel@tonic-gate #endif
1857c478bd9Sstevel@tonic-gate 
1867c478bd9Sstevel@tonic-gate struct pcf {
1877c478bd9Sstevel@tonic-gate 	kmutex_t	pcf_lock;	/* protects the structure */
188f2b37d75Sfr 	uint_t		pcf_count;	/* page count */
1897c478bd9Sstevel@tonic-gate 	uint_t		pcf_wait;	/* number of waiters */
1907c478bd9Sstevel@tonic-gate 	uint_t		pcf_block; 	/* pcgs flag to page_free() */
1917c478bd9Sstevel@tonic-gate 	uint_t		pcf_reserve; 	/* pages freed after pcf_block set */
1927c478bd9Sstevel@tonic-gate 	uint_t		pcf_fill[PAD];	/* to line up on the caches */
1937c478bd9Sstevel@tonic-gate };
1947c478bd9Sstevel@tonic-gate 
1957c478bd9Sstevel@tonic-gate static struct	pcf	pcf[PCF_FANOUT];
1967c478bd9Sstevel@tonic-gate #define	PCF_INDEX()	((CPU->cpu_id) & (pcf_mask))
1977c478bd9Sstevel@tonic-gate 
1987c478bd9Sstevel@tonic-gate kmutex_t	pcgs_lock;		/* serializes page_create_get_ */
1997c478bd9Sstevel@tonic-gate kmutex_t	pcgs_cagelock;		/* serializes NOSLEEP cage allocs */
2007c478bd9Sstevel@tonic-gate kmutex_t	pcgs_wait_lock;		/* used for delay in pcgs */
2017c478bd9Sstevel@tonic-gate static kcondvar_t	pcgs_cv;	/* cv for delay in pcgs */
2027c478bd9Sstevel@tonic-gate 
2037c478bd9Sstevel@tonic-gate #ifdef VM_STATS
2047c478bd9Sstevel@tonic-gate 
2057c478bd9Sstevel@tonic-gate /*
2067c478bd9Sstevel@tonic-gate  * No locks, but so what, they are only statistics.
2077c478bd9Sstevel@tonic-gate  */
2087c478bd9Sstevel@tonic-gate 
2097c478bd9Sstevel@tonic-gate static struct page_tcnt {
2107c478bd9Sstevel@tonic-gate 	int	pc_free_cache;		/* free's into cache list */
2117c478bd9Sstevel@tonic-gate 	int	pc_free_dontneed;	/* free's with dontneed */
2127c478bd9Sstevel@tonic-gate 	int	pc_free_pageout;	/* free's from pageout */
2137c478bd9Sstevel@tonic-gate 	int	pc_free_free;		/* free's into free list */
2147c478bd9Sstevel@tonic-gate 	int	pc_free_pages;		/* free's into large page free list */
2157c478bd9Sstevel@tonic-gate 	int	pc_destroy_pages;	/* large page destroy's */
2167c478bd9Sstevel@tonic-gate 	int	pc_get_cache;		/* get's from cache list */
2177c478bd9Sstevel@tonic-gate 	int	pc_get_free;		/* get's from free list */
2187c478bd9Sstevel@tonic-gate 	int	pc_reclaim;		/* reclaim's */
2197c478bd9Sstevel@tonic-gate 	int	pc_abortfree;		/* abort's of free pages */
2207c478bd9Sstevel@tonic-gate 	int	pc_find_hit;		/* find's that find page */
2217c478bd9Sstevel@tonic-gate 	int	pc_find_miss;		/* find's that don't find page */
2227c478bd9Sstevel@tonic-gate 	int	pc_destroy_free;	/* # of free pages destroyed */
2237c478bd9Sstevel@tonic-gate #define	PC_HASH_CNT	(4*PAGE_HASHAVELEN)
2247c478bd9Sstevel@tonic-gate 	int	pc_find_hashlen[PC_HASH_CNT+1];
2257c478bd9Sstevel@tonic-gate 	int	pc_addclaim_pages;
2267c478bd9Sstevel@tonic-gate 	int	pc_subclaim_pages;
2277c478bd9Sstevel@tonic-gate 	int	pc_free_replacement_page[2];
2287c478bd9Sstevel@tonic-gate 	int	pc_try_demote_pages[6];
2297c478bd9Sstevel@tonic-gate 	int	pc_demote_pages[2];
2307c478bd9Sstevel@tonic-gate } pagecnt;
2317c478bd9Sstevel@tonic-gate 
2327c478bd9Sstevel@tonic-gate uint_t	hashin_count;
2337c478bd9Sstevel@tonic-gate uint_t	hashin_not_held;
2347c478bd9Sstevel@tonic-gate uint_t	hashin_already;
2357c478bd9Sstevel@tonic-gate 
2367c478bd9Sstevel@tonic-gate uint_t	hashout_count;
2377c478bd9Sstevel@tonic-gate uint_t	hashout_not_held;
2387c478bd9Sstevel@tonic-gate 
2397c478bd9Sstevel@tonic-gate uint_t	page_create_count;
2407c478bd9Sstevel@tonic-gate uint_t	page_create_not_enough;
2417c478bd9Sstevel@tonic-gate uint_t	page_create_not_enough_again;
2427c478bd9Sstevel@tonic-gate uint_t	page_create_zero;
2437c478bd9Sstevel@tonic-gate uint_t	page_create_hashout;
2447c478bd9Sstevel@tonic-gate uint_t	page_create_page_lock_failed;
2457c478bd9Sstevel@tonic-gate uint_t	page_create_trylock_failed;
2467c478bd9Sstevel@tonic-gate uint_t	page_create_found_one;
2477c478bd9Sstevel@tonic-gate uint_t	page_create_hashin_failed;
2487c478bd9Sstevel@tonic-gate uint_t	page_create_dropped_phm;
2497c478bd9Sstevel@tonic-gate 
2507c478bd9Sstevel@tonic-gate uint_t	page_create_new;
2517c478bd9Sstevel@tonic-gate uint_t	page_create_exists;
2527c478bd9Sstevel@tonic-gate uint_t	page_create_putbacks;
2537c478bd9Sstevel@tonic-gate uint_t	page_create_overshoot;
2547c478bd9Sstevel@tonic-gate 
2557c478bd9Sstevel@tonic-gate uint_t	page_reclaim_zero;
2567c478bd9Sstevel@tonic-gate uint_t	page_reclaim_zero_locked;
2577c478bd9Sstevel@tonic-gate 
2587c478bd9Sstevel@tonic-gate uint_t	page_rename_exists;
2597c478bd9Sstevel@tonic-gate uint_t	page_rename_count;
2607c478bd9Sstevel@tonic-gate 
2617c478bd9Sstevel@tonic-gate uint_t	page_lookup_cnt[20];
2627c478bd9Sstevel@tonic-gate uint_t	page_lookup_nowait_cnt[10];
2637c478bd9Sstevel@tonic-gate uint_t	page_find_cnt;
2647c478bd9Sstevel@tonic-gate uint_t	page_exists_cnt;
2657c478bd9Sstevel@tonic-gate uint_t	page_exists_forreal_cnt;
2667c478bd9Sstevel@tonic-gate uint_t	page_lookup_dev_cnt;
2677c478bd9Sstevel@tonic-gate uint_t	get_cachelist_cnt;
2687c478bd9Sstevel@tonic-gate uint_t	page_create_cnt[10];
2697c478bd9Sstevel@tonic-gate uint_t	alloc_pages[8];
2707c478bd9Sstevel@tonic-gate uint_t	page_exphcontg[19];
2717c478bd9Sstevel@tonic-gate uint_t  page_create_large_cnt[10];
2727c478bd9Sstevel@tonic-gate 
2737c478bd9Sstevel@tonic-gate /*
2747c478bd9Sstevel@tonic-gate  * Collects statistics.
2757c478bd9Sstevel@tonic-gate  */
2767c478bd9Sstevel@tonic-gate #define	PAGE_HASH_SEARCH(index, pp, vp, off) { \
2777c478bd9Sstevel@tonic-gate 	uint_t	mylen = 0; \
2787c478bd9Sstevel@tonic-gate 			\
2797c478bd9Sstevel@tonic-gate 	for ((pp) = page_hash[(index)]; (pp); (pp) = (pp)->p_hash, mylen++) { \
2807c478bd9Sstevel@tonic-gate 		if ((pp)->p_vnode == (vp) && (pp)->p_offset == (off)) \
2817c478bd9Sstevel@tonic-gate 			break; \
2827c478bd9Sstevel@tonic-gate 	} \
2837c478bd9Sstevel@tonic-gate 	if ((pp) != NULL) \
2847c478bd9Sstevel@tonic-gate 		pagecnt.pc_find_hit++; \
2857c478bd9Sstevel@tonic-gate 	else \
2867c478bd9Sstevel@tonic-gate 		pagecnt.pc_find_miss++; \
2877c478bd9Sstevel@tonic-gate 	if (mylen > PC_HASH_CNT) \
2887c478bd9Sstevel@tonic-gate 		mylen = PC_HASH_CNT; \
2897c478bd9Sstevel@tonic-gate 	pagecnt.pc_find_hashlen[mylen]++; \
2907c478bd9Sstevel@tonic-gate }
2917c478bd9Sstevel@tonic-gate 
2927c478bd9Sstevel@tonic-gate #else	/* VM_STATS */
2937c478bd9Sstevel@tonic-gate 
2947c478bd9Sstevel@tonic-gate /*
2957c478bd9Sstevel@tonic-gate  * Don't collect statistics
2967c478bd9Sstevel@tonic-gate  */
2977c478bd9Sstevel@tonic-gate #define	PAGE_HASH_SEARCH(index, pp, vp, off) { \
2987c478bd9Sstevel@tonic-gate 	for ((pp) = page_hash[(index)]; (pp); (pp) = (pp)->p_hash) { \
2997c478bd9Sstevel@tonic-gate 		if ((pp)->p_vnode == (vp) && (pp)->p_offset == (off)) \
3007c478bd9Sstevel@tonic-gate 			break; \
3017c478bd9Sstevel@tonic-gate 	} \
3027c478bd9Sstevel@tonic-gate }
3037c478bd9Sstevel@tonic-gate 
3047c478bd9Sstevel@tonic-gate #endif	/* VM_STATS */
3057c478bd9Sstevel@tonic-gate 
3067c478bd9Sstevel@tonic-gate 
3077c478bd9Sstevel@tonic-gate 
3087c478bd9Sstevel@tonic-gate #ifdef DEBUG
3097c478bd9Sstevel@tonic-gate #define	MEMSEG_SEARCH_STATS
3107c478bd9Sstevel@tonic-gate #endif
3117c478bd9Sstevel@tonic-gate 
3127c478bd9Sstevel@tonic-gate #ifdef MEMSEG_SEARCH_STATS
3137c478bd9Sstevel@tonic-gate struct memseg_stats {
3147c478bd9Sstevel@tonic-gate     uint_t nsearch;
3157c478bd9Sstevel@tonic-gate     uint_t nlastwon;
3167c478bd9Sstevel@tonic-gate     uint_t nhashwon;
3177c478bd9Sstevel@tonic-gate     uint_t nnotfound;
3187c478bd9Sstevel@tonic-gate } memseg_stats;
3197c478bd9Sstevel@tonic-gate 
3207c478bd9Sstevel@tonic-gate #define	MEMSEG_STAT_INCR(v) \
3217c478bd9Sstevel@tonic-gate 	atomic_add_32(&memseg_stats.v, 1)
3227c478bd9Sstevel@tonic-gate #else
3237c478bd9Sstevel@tonic-gate #define	MEMSEG_STAT_INCR(x)
3247c478bd9Sstevel@tonic-gate #endif
3257c478bd9Sstevel@tonic-gate 
3267c478bd9Sstevel@tonic-gate struct memseg *memsegs;		/* list of memory segments */
3277c478bd9Sstevel@tonic-gate 
3287c478bd9Sstevel@tonic-gate 
3297c478bd9Sstevel@tonic-gate static void page_init_mem_config(void);
3307c478bd9Sstevel@tonic-gate static int page_do_hashin(page_t *, vnode_t *, u_offset_t);
3317c478bd9Sstevel@tonic-gate static void page_do_hashout(page_t *);
3327c478bd9Sstevel@tonic-gate 
3337c478bd9Sstevel@tonic-gate static void page_demote_vp_pages(page_t *);
3347c478bd9Sstevel@tonic-gate 
3357c478bd9Sstevel@tonic-gate /*
3367c478bd9Sstevel@tonic-gate  * vm subsystem related initialization
3377c478bd9Sstevel@tonic-gate  */
3387c478bd9Sstevel@tonic-gate void
3397c478bd9Sstevel@tonic-gate vm_init(void)
3407c478bd9Sstevel@tonic-gate {
3417c478bd9Sstevel@tonic-gate 	boolean_t callb_vm_cpr(void *, int);
3427c478bd9Sstevel@tonic-gate 
3437c478bd9Sstevel@tonic-gate 	(void) callb_add(callb_vm_cpr, 0, CB_CL_CPR_VM, "vm");
3447c478bd9Sstevel@tonic-gate 	page_init_mem_config();
345db874c57Selowe 	page_retire_init();
346*0209230bSgjelinek 	vm_usage_init();
3477c478bd9Sstevel@tonic-gate }
3487c478bd9Sstevel@tonic-gate 
3497c478bd9Sstevel@tonic-gate /*
3507c478bd9Sstevel@tonic-gate  * This function is called at startup and when memory is added or deleted.
3517c478bd9Sstevel@tonic-gate  */
3527c478bd9Sstevel@tonic-gate void
3537c478bd9Sstevel@tonic-gate init_pages_pp_maximum()
3547c478bd9Sstevel@tonic-gate {
3557c478bd9Sstevel@tonic-gate 	static pgcnt_t p_min;
3567c478bd9Sstevel@tonic-gate 	static pgcnt_t pages_pp_maximum_startup;
3577c478bd9Sstevel@tonic-gate 	static pgcnt_t avrmem_delta;
3587c478bd9Sstevel@tonic-gate 	static int init_done;
3597c478bd9Sstevel@tonic-gate 	static int user_set;	/* true if set in /etc/system */
3607c478bd9Sstevel@tonic-gate 
3617c478bd9Sstevel@tonic-gate 	if (init_done == 0) {
3627c478bd9Sstevel@tonic-gate 
3637c478bd9Sstevel@tonic-gate 		/* If the user specified a value, save it */
3647c478bd9Sstevel@tonic-gate 		if (pages_pp_maximum != 0) {
3657c478bd9Sstevel@tonic-gate 			user_set = 1;
3667c478bd9Sstevel@tonic-gate 			pages_pp_maximum_startup = pages_pp_maximum;
3677c478bd9Sstevel@tonic-gate 		}
3687c478bd9Sstevel@tonic-gate 
3697c478bd9Sstevel@tonic-gate 		/*
3707c478bd9Sstevel@tonic-gate 		 * Setting of pages_pp_maximum is based first time
3717c478bd9Sstevel@tonic-gate 		 * on the value of availrmem just after the start-up
3727c478bd9Sstevel@tonic-gate 		 * allocations. To preserve this relationship at run
3737c478bd9Sstevel@tonic-gate 		 * time, use a delta from availrmem_initial.
3747c478bd9Sstevel@tonic-gate 		 */
3757c478bd9Sstevel@tonic-gate 		ASSERT(availrmem_initial >= availrmem);
3767c478bd9Sstevel@tonic-gate 		avrmem_delta = availrmem_initial - availrmem;
3777c478bd9Sstevel@tonic-gate 
3787c478bd9Sstevel@tonic-gate 		/* The allowable floor of pages_pp_maximum */
3797c478bd9Sstevel@tonic-gate 		p_min = tune.t_minarmem + 100;
3807c478bd9Sstevel@tonic-gate 
3817c478bd9Sstevel@tonic-gate 		/* Make sure we don't come through here again. */
3827c478bd9Sstevel@tonic-gate 		init_done = 1;
3837c478bd9Sstevel@tonic-gate 	}
3847c478bd9Sstevel@tonic-gate 	/*
3857c478bd9Sstevel@tonic-gate 	 * Determine pages_pp_maximum, the number of currently available
3867c478bd9Sstevel@tonic-gate 	 * pages (availrmem) that can't be `locked'. If not set by
3877c478bd9Sstevel@tonic-gate 	 * the user, we set it to 4% of the currently available memory
3887c478bd9Sstevel@tonic-gate 	 * plus 4MB.
3897c478bd9Sstevel@tonic-gate 	 * But we also insist that it be greater than tune.t_minarmem;
3907c478bd9Sstevel@tonic-gate 	 * otherwise a process could lock down a lot of memory, get swapped
3917c478bd9Sstevel@tonic-gate 	 * out, and never have enough to get swapped back in.
3927c478bd9Sstevel@tonic-gate 	 */
3937c478bd9Sstevel@tonic-gate 	if (user_set)
3947c478bd9Sstevel@tonic-gate 		pages_pp_maximum = pages_pp_maximum_startup;
3957c478bd9Sstevel@tonic-gate 	else
3967c478bd9Sstevel@tonic-gate 		pages_pp_maximum = ((availrmem_initial - avrmem_delta) / 25)
3977c478bd9Sstevel@tonic-gate 		    + btop(4 * 1024 * 1024);
3987c478bd9Sstevel@tonic-gate 
3997c478bd9Sstevel@tonic-gate 	if (pages_pp_maximum <= p_min) {
4007c478bd9Sstevel@tonic-gate 		pages_pp_maximum = p_min;
4017c478bd9Sstevel@tonic-gate 	}
4027c478bd9Sstevel@tonic-gate }
4037c478bd9Sstevel@tonic-gate 
4047c478bd9Sstevel@tonic-gate void
4057c478bd9Sstevel@tonic-gate set_max_page_get(pgcnt_t target_total_pages)
4067c478bd9Sstevel@tonic-gate {
4077c478bd9Sstevel@tonic-gate 	max_page_get = target_total_pages / 2;
4087c478bd9Sstevel@tonic-gate }
4097c478bd9Sstevel@tonic-gate 
4107c478bd9Sstevel@tonic-gate static pgcnt_t pending_delete;
4117c478bd9Sstevel@tonic-gate 
4127c478bd9Sstevel@tonic-gate /*ARGSUSED*/
4137c478bd9Sstevel@tonic-gate static void
4147c478bd9Sstevel@tonic-gate page_mem_config_post_add(
4157c478bd9Sstevel@tonic-gate 	void *arg,
4167c478bd9Sstevel@tonic-gate 	pgcnt_t delta_pages)
4177c478bd9Sstevel@tonic-gate {
4187c478bd9Sstevel@tonic-gate 	set_max_page_get(total_pages - pending_delete);
4197c478bd9Sstevel@tonic-gate 	init_pages_pp_maximum();
4207c478bd9Sstevel@tonic-gate }
4217c478bd9Sstevel@tonic-gate 
4227c478bd9Sstevel@tonic-gate /*ARGSUSED*/
4237c478bd9Sstevel@tonic-gate static int
4247c478bd9Sstevel@tonic-gate page_mem_config_pre_del(
4257c478bd9Sstevel@tonic-gate 	void *arg,
4267c478bd9Sstevel@tonic-gate 	pgcnt_t delta_pages)
4277c478bd9Sstevel@tonic-gate {
4287c478bd9Sstevel@tonic-gate 	pgcnt_t nv;
4297c478bd9Sstevel@tonic-gate 
4307c478bd9Sstevel@tonic-gate 	nv = atomic_add_long_nv(&pending_delete, (spgcnt_t)delta_pages);
4317c478bd9Sstevel@tonic-gate 	set_max_page_get(total_pages - nv);
4327c478bd9Sstevel@tonic-gate 	return (0);
4337c478bd9Sstevel@tonic-gate }
4347c478bd9Sstevel@tonic-gate 
4357c478bd9Sstevel@tonic-gate /*ARGSUSED*/
4367c478bd9Sstevel@tonic-gate static void
4377c478bd9Sstevel@tonic-gate page_mem_config_post_del(
4387c478bd9Sstevel@tonic-gate 	void *arg,
4397c478bd9Sstevel@tonic-gate 	pgcnt_t delta_pages,
4407c478bd9Sstevel@tonic-gate 	int cancelled)
4417c478bd9Sstevel@tonic-gate {
4427c478bd9Sstevel@tonic-gate 	pgcnt_t nv;
4437c478bd9Sstevel@tonic-gate 
4447c478bd9Sstevel@tonic-gate 	nv = atomic_add_long_nv(&pending_delete, -(spgcnt_t)delta_pages);
4457c478bd9Sstevel@tonic-gate 	set_max_page_get(total_pages - nv);
4467c478bd9Sstevel@tonic-gate 	if (!cancelled)
4477c478bd9Sstevel@tonic-gate 		init_pages_pp_maximum();
4487c478bd9Sstevel@tonic-gate }
4497c478bd9Sstevel@tonic-gate 
4507c478bd9Sstevel@tonic-gate static kphysm_setup_vector_t page_mem_config_vec = {
4517c478bd9Sstevel@tonic-gate 	KPHYSM_SETUP_VECTOR_VERSION,
4527c478bd9Sstevel@tonic-gate 	page_mem_config_post_add,
4537c478bd9Sstevel@tonic-gate 	page_mem_config_pre_del,
4547c478bd9Sstevel@tonic-gate 	page_mem_config_post_del,
4557c478bd9Sstevel@tonic-gate };
4567c478bd9Sstevel@tonic-gate 
4577c478bd9Sstevel@tonic-gate static void
4587c478bd9Sstevel@tonic-gate page_init_mem_config(void)
4597c478bd9Sstevel@tonic-gate {
4607c478bd9Sstevel@tonic-gate 	int ret;
4617c478bd9Sstevel@tonic-gate 
4627c478bd9Sstevel@tonic-gate 	ret = kphysm_setup_func_register(&page_mem_config_vec, (void *)NULL);
4637c478bd9Sstevel@tonic-gate 	ASSERT(ret == 0);
4647c478bd9Sstevel@tonic-gate }
4657c478bd9Sstevel@tonic-gate 
4667c478bd9Sstevel@tonic-gate /*
4677c478bd9Sstevel@tonic-gate  * Evenly spread out the PCF counters for large free pages
4687c478bd9Sstevel@tonic-gate  */
4697c478bd9Sstevel@tonic-gate static void
4707c478bd9Sstevel@tonic-gate page_free_large_ctr(pgcnt_t npages)
4717c478bd9Sstevel@tonic-gate {
4727c478bd9Sstevel@tonic-gate 	static struct pcf	*p = pcf;
4737c478bd9Sstevel@tonic-gate 	pgcnt_t			lump;
4747c478bd9Sstevel@tonic-gate 
4757c478bd9Sstevel@tonic-gate 	freemem += npages;
4767c478bd9Sstevel@tonic-gate 
4777c478bd9Sstevel@tonic-gate 	lump = roundup(npages, PCF_FANOUT) / PCF_FANOUT;
4787c478bd9Sstevel@tonic-gate 
4797c478bd9Sstevel@tonic-gate 	while (npages > 0) {
4807c478bd9Sstevel@tonic-gate 
4817c478bd9Sstevel@tonic-gate 		ASSERT(!p->pcf_block);
4827c478bd9Sstevel@tonic-gate 
4837c478bd9Sstevel@tonic-gate 		if (lump < npages) {
4847c478bd9Sstevel@tonic-gate 			p->pcf_count += (uint_t)lump;
4857c478bd9Sstevel@tonic-gate 			npages -= lump;
4867c478bd9Sstevel@tonic-gate 		} else {
4877c478bd9Sstevel@tonic-gate 			p->pcf_count += (uint_t)npages;
4887c478bd9Sstevel@tonic-gate 			npages = 0;
4897c478bd9Sstevel@tonic-gate 		}
4907c478bd9Sstevel@tonic-gate 
4917c478bd9Sstevel@tonic-gate 		ASSERT(!p->pcf_wait);
4927c478bd9Sstevel@tonic-gate 
4937c478bd9Sstevel@tonic-gate 		if (++p > &pcf[PCF_FANOUT - 1])
4947c478bd9Sstevel@tonic-gate 			p = pcf;
4957c478bd9Sstevel@tonic-gate 	}
4967c478bd9Sstevel@tonic-gate 
4977c478bd9Sstevel@tonic-gate 	ASSERT(npages == 0);
4987c478bd9Sstevel@tonic-gate }
4997c478bd9Sstevel@tonic-gate 
5007c478bd9Sstevel@tonic-gate /*
5017c478bd9Sstevel@tonic-gate  * Add a physical chunk of memory to the system freee lists during startup.
5027c478bd9Sstevel@tonic-gate  * Platform specific startup() allocates the memory for the page structs.
5037c478bd9Sstevel@tonic-gate  *
5047c478bd9Sstevel@tonic-gate  * num	- number of page structures
5057c478bd9Sstevel@tonic-gate  * base - page number (pfn) to be associated with the first page.
5067c478bd9Sstevel@tonic-gate  *
5077c478bd9Sstevel@tonic-gate  * Since we are doing this during startup (ie. single threaded), we will
5087c478bd9Sstevel@tonic-gate  * use shortcut routines to avoid any locking overhead while putting all
5097c478bd9Sstevel@tonic-gate  * these pages on the freelists.
5107c478bd9Sstevel@tonic-gate  *
5117c478bd9Sstevel@tonic-gate  * NOTE: Any changes performed to page_free(), must also be performed to
5127c478bd9Sstevel@tonic-gate  *	 add_physmem() since this is how we initialize all page_t's at
5137c478bd9Sstevel@tonic-gate  *	 boot time.
5147c478bd9Sstevel@tonic-gate  */
5157c478bd9Sstevel@tonic-gate void
5167c478bd9Sstevel@tonic-gate add_physmem(
5177c478bd9Sstevel@tonic-gate 	page_t	*pp,
5187c478bd9Sstevel@tonic-gate 	pgcnt_t	num,
5197c478bd9Sstevel@tonic-gate 	pfn_t	pnum)
5207c478bd9Sstevel@tonic-gate {
5217c478bd9Sstevel@tonic-gate 	page_t	*root = NULL;
5227c478bd9Sstevel@tonic-gate 	uint_t	szc = page_num_pagesizes() - 1;
5237c478bd9Sstevel@tonic-gate 	pgcnt_t	large = page_get_pagecnt(szc);
5247c478bd9Sstevel@tonic-gate 	pgcnt_t	cnt = 0;
5257c478bd9Sstevel@tonic-gate 
5267c478bd9Sstevel@tonic-gate 	TRACE_2(TR_FAC_VM, TR_PAGE_INIT,
5277c478bd9Sstevel@tonic-gate 		"add_physmem:pp %p num %lu", pp, num);
5287c478bd9Sstevel@tonic-gate 
5297c478bd9Sstevel@tonic-gate 	/*
5307c478bd9Sstevel@tonic-gate 	 * Arbitrarily limit the max page_get request
5317c478bd9Sstevel@tonic-gate 	 * to 1/2 of the page structs we have.
5327c478bd9Sstevel@tonic-gate 	 */
5337c478bd9Sstevel@tonic-gate 	total_pages += num;
5347c478bd9Sstevel@tonic-gate 	set_max_page_get(total_pages);
5357c478bd9Sstevel@tonic-gate 
536e21bae1bSkchow 	PLCNT_MODIFY_MAX(pnum, (long)num);
537e21bae1bSkchow 
5387c478bd9Sstevel@tonic-gate 	/*
5397c478bd9Sstevel@tonic-gate 	 * The physical space for the pages array
5407c478bd9Sstevel@tonic-gate 	 * representing ram pages has already been
5417c478bd9Sstevel@tonic-gate 	 * allocated.  Here we initialize each lock
5427c478bd9Sstevel@tonic-gate 	 * in the page structure, and put each on
5437c478bd9Sstevel@tonic-gate 	 * the free list
5447c478bd9Sstevel@tonic-gate 	 */
545affbd3ccSkchow 	for (; num; pp++, pnum++, num--) {
5467c478bd9Sstevel@tonic-gate 
5477c478bd9Sstevel@tonic-gate 		/*
5487c478bd9Sstevel@tonic-gate 		 * this needs to fill in the page number
5497c478bd9Sstevel@tonic-gate 		 * and do any other arch specific initialization
5507c478bd9Sstevel@tonic-gate 		 */
5517c478bd9Sstevel@tonic-gate 		add_physmem_cb(pp, pnum);
5527c478bd9Sstevel@tonic-gate 
55307b65a64Saguzovsk 		pp->p_lckcnt = 0;
55407b65a64Saguzovsk 		pp->p_cowcnt = 0;
55507b65a64Saguzovsk 		pp->p_slckcnt = 0;
55607b65a64Saguzovsk 
5577c478bd9Sstevel@tonic-gate 		/*
5587c478bd9Sstevel@tonic-gate 		 * Initialize the page lock as unlocked, since nobody
5597c478bd9Sstevel@tonic-gate 		 * can see or access this page yet.
5607c478bd9Sstevel@tonic-gate 		 */
5617c478bd9Sstevel@tonic-gate 		pp->p_selock = 0;
5627c478bd9Sstevel@tonic-gate 
5637c478bd9Sstevel@tonic-gate 		/*
5647c478bd9Sstevel@tonic-gate 		 * Initialize IO lock
5657c478bd9Sstevel@tonic-gate 		 */
5667c478bd9Sstevel@tonic-gate 		page_iolock_init(pp);
5677c478bd9Sstevel@tonic-gate 
5687c478bd9Sstevel@tonic-gate 		/*
5697c478bd9Sstevel@tonic-gate 		 * initialize other fields in the page_t
5707c478bd9Sstevel@tonic-gate 		 */
5717c478bd9Sstevel@tonic-gate 		PP_SETFREE(pp);
5727c478bd9Sstevel@tonic-gate 		page_clr_all_props(pp);
5737c478bd9Sstevel@tonic-gate 		PP_SETAGED(pp);
5747c478bd9Sstevel@tonic-gate 		pp->p_offset = (u_offset_t)-1;
5757c478bd9Sstevel@tonic-gate 		pp->p_next = pp;
5767c478bd9Sstevel@tonic-gate 		pp->p_prev = pp;
5777c478bd9Sstevel@tonic-gate 
5787c478bd9Sstevel@tonic-gate 		/*
5797c478bd9Sstevel@tonic-gate 		 * Simple case: System doesn't support large pages.
5807c478bd9Sstevel@tonic-gate 		 */
5817c478bd9Sstevel@tonic-gate 		if (szc == 0) {
5827c478bd9Sstevel@tonic-gate 			pp->p_szc = 0;
5837c478bd9Sstevel@tonic-gate 			page_free_at_startup(pp);
5847c478bd9Sstevel@tonic-gate 			continue;
5857c478bd9Sstevel@tonic-gate 		}
5867c478bd9Sstevel@tonic-gate 
5877c478bd9Sstevel@tonic-gate 		/*
5887c478bd9Sstevel@tonic-gate 		 * Handle unaligned pages, we collect them up onto
5897c478bd9Sstevel@tonic-gate 		 * the root page until we have a full large page.
5907c478bd9Sstevel@tonic-gate 		 */
5917c478bd9Sstevel@tonic-gate 		if (!IS_P2ALIGNED(pnum, large)) {
5927c478bd9Sstevel@tonic-gate 
5937c478bd9Sstevel@tonic-gate 			/*
5947c478bd9Sstevel@tonic-gate 			 * If not in a large page,
5957c478bd9Sstevel@tonic-gate 			 * just free as small page.
5967c478bd9Sstevel@tonic-gate 			 */
5977c478bd9Sstevel@tonic-gate 			if (root == NULL) {
5987c478bd9Sstevel@tonic-gate 				pp->p_szc = 0;
5997c478bd9Sstevel@tonic-gate 				page_free_at_startup(pp);
6007c478bd9Sstevel@tonic-gate 				continue;
6017c478bd9Sstevel@tonic-gate 			}
6027c478bd9Sstevel@tonic-gate 
6037c478bd9Sstevel@tonic-gate 			/*
6047c478bd9Sstevel@tonic-gate 			 * Link a constituent page into the large page.
6057c478bd9Sstevel@tonic-gate 			 */
6067c478bd9Sstevel@tonic-gate 			pp->p_szc = szc;
6077c478bd9Sstevel@tonic-gate 			page_list_concat(&root, &pp);
6087c478bd9Sstevel@tonic-gate 
6097c478bd9Sstevel@tonic-gate 			/*
6107c478bd9Sstevel@tonic-gate 			 * When large page is fully formed, free it.
6117c478bd9Sstevel@tonic-gate 			 */
6127c478bd9Sstevel@tonic-gate 			if (++cnt == large) {
6137c478bd9Sstevel@tonic-gate 				page_free_large_ctr(cnt);
6147c478bd9Sstevel@tonic-gate 				page_list_add_pages(root, PG_LIST_ISINIT);
6157c478bd9Sstevel@tonic-gate 				root = NULL;
6167c478bd9Sstevel@tonic-gate 				cnt = 0;
6177c478bd9Sstevel@tonic-gate 			}
6187c478bd9Sstevel@tonic-gate 			continue;
6197c478bd9Sstevel@tonic-gate 		}
6207c478bd9Sstevel@tonic-gate 
6217c478bd9Sstevel@tonic-gate 		/*
6227c478bd9Sstevel@tonic-gate 		 * At this point we have a page number which
6237c478bd9Sstevel@tonic-gate 		 * is aligned. We assert that we aren't already
6247c478bd9Sstevel@tonic-gate 		 * in a different large page.
6257c478bd9Sstevel@tonic-gate 		 */
6267c478bd9Sstevel@tonic-gate 		ASSERT(IS_P2ALIGNED(pnum, large));
6277c478bd9Sstevel@tonic-gate 		ASSERT(root == NULL && cnt == 0);
6287c478bd9Sstevel@tonic-gate 
6297c478bd9Sstevel@tonic-gate 		/*
6307c478bd9Sstevel@tonic-gate 		 * If insufficient number of pages left to form
6317c478bd9Sstevel@tonic-gate 		 * a large page, just free the small page.
6327c478bd9Sstevel@tonic-gate 		 */
6337c478bd9Sstevel@tonic-gate 		if (num < large) {
6347c478bd9Sstevel@tonic-gate 			pp->p_szc = 0;
6357c478bd9Sstevel@tonic-gate 			page_free_at_startup(pp);
6367c478bd9Sstevel@tonic-gate 			continue;
6377c478bd9Sstevel@tonic-gate 		}
6387c478bd9Sstevel@tonic-gate 
6397c478bd9Sstevel@tonic-gate 		/*
6407c478bd9Sstevel@tonic-gate 		 * Otherwise start a new large page.
6417c478bd9Sstevel@tonic-gate 		 */
6427c478bd9Sstevel@tonic-gate 		pp->p_szc = szc;
6437c478bd9Sstevel@tonic-gate 		cnt++;
6447c478bd9Sstevel@tonic-gate 		root = pp;
6457c478bd9Sstevel@tonic-gate 	}
6467c478bd9Sstevel@tonic-gate 	ASSERT(root == NULL && cnt == 0);
6477c478bd9Sstevel@tonic-gate }
6487c478bd9Sstevel@tonic-gate 
6497c478bd9Sstevel@tonic-gate /*
6507c478bd9Sstevel@tonic-gate  * Find a page representing the specified [vp, offset].
6517c478bd9Sstevel@tonic-gate  * If we find the page but it is intransit coming in,
6527c478bd9Sstevel@tonic-gate  * it will have an "exclusive" lock and we wait for
6537c478bd9Sstevel@tonic-gate  * the i/o to complete.  A page found on the free list
6547c478bd9Sstevel@tonic-gate  * is always reclaimed and then locked.  On success, the page
6557c478bd9Sstevel@tonic-gate  * is locked, its data is valid and it isn't on the free
6567c478bd9Sstevel@tonic-gate  * list, while a NULL is returned if the page doesn't exist.
6577c478bd9Sstevel@tonic-gate  */
6587c478bd9Sstevel@tonic-gate page_t *
6597c478bd9Sstevel@tonic-gate page_lookup(vnode_t *vp, u_offset_t off, se_t se)
6607c478bd9Sstevel@tonic-gate {
6617c478bd9Sstevel@tonic-gate 	return (page_lookup_create(vp, off, se, NULL, NULL, 0));
6627c478bd9Sstevel@tonic-gate }
6637c478bd9Sstevel@tonic-gate 
6647c478bd9Sstevel@tonic-gate /*
6657c478bd9Sstevel@tonic-gate  * Find a page representing the specified [vp, offset].
6667c478bd9Sstevel@tonic-gate  * We either return the one we found or, if passed in,
6677c478bd9Sstevel@tonic-gate  * create one with identity of [vp, offset] of the
6687c478bd9Sstevel@tonic-gate  * pre-allocated page. If we find exsisting page but it is
6697c478bd9Sstevel@tonic-gate  * intransit coming in, it will have an "exclusive" lock
6707c478bd9Sstevel@tonic-gate  * and we wait for the i/o to complete.  A page found on
6717c478bd9Sstevel@tonic-gate  * the free list is always reclaimed and then locked.
6727c478bd9Sstevel@tonic-gate  * On success, the page is locked, its data is valid and
6737c478bd9Sstevel@tonic-gate  * it isn't on the free list, while a NULL is returned
6747c478bd9Sstevel@tonic-gate  * if the page doesn't exist and newpp is NULL;
6757c478bd9Sstevel@tonic-gate  */
6767c478bd9Sstevel@tonic-gate page_t *
6777c478bd9Sstevel@tonic-gate page_lookup_create(
6787c478bd9Sstevel@tonic-gate 	vnode_t *vp,
6797c478bd9Sstevel@tonic-gate 	u_offset_t off,
6807c478bd9Sstevel@tonic-gate 	se_t se,
6817c478bd9Sstevel@tonic-gate 	page_t *newpp,
6827c478bd9Sstevel@tonic-gate 	spgcnt_t *nrelocp,
6837c478bd9Sstevel@tonic-gate 	int flags)
6847c478bd9Sstevel@tonic-gate {
6857c478bd9Sstevel@tonic-gate 	page_t		*pp;
6867c478bd9Sstevel@tonic-gate 	kmutex_t	*phm;
6877c478bd9Sstevel@tonic-gate 	ulong_t		index;
6887c478bd9Sstevel@tonic-gate 	uint_t		hash_locked;
6897c478bd9Sstevel@tonic-gate 	uint_t		es;
6907c478bd9Sstevel@tonic-gate 
6917c478bd9Sstevel@tonic-gate 	ASSERT(MUTEX_NOT_HELD(page_vnode_mutex(vp)));
6927c478bd9Sstevel@tonic-gate 	VM_STAT_ADD(page_lookup_cnt[0]);
6937c478bd9Sstevel@tonic-gate 	ASSERT(newpp ? PAGE_EXCL(newpp) : 1);
6947c478bd9Sstevel@tonic-gate 
6957c478bd9Sstevel@tonic-gate 	/*
6967c478bd9Sstevel@tonic-gate 	 * Acquire the appropriate page hash lock since
6977c478bd9Sstevel@tonic-gate 	 * we have to search the hash list.  Pages that
6987c478bd9Sstevel@tonic-gate 	 * hash to this list can't change identity while
6997c478bd9Sstevel@tonic-gate 	 * this lock is held.
7007c478bd9Sstevel@tonic-gate 	 */
7017c478bd9Sstevel@tonic-gate 	hash_locked = 0;
7027c478bd9Sstevel@tonic-gate 	index = PAGE_HASH_FUNC(vp, off);
7037c478bd9Sstevel@tonic-gate 	phm = NULL;
7047c478bd9Sstevel@tonic-gate top:
7057c478bd9Sstevel@tonic-gate 	PAGE_HASH_SEARCH(index, pp, vp, off);
7067c478bd9Sstevel@tonic-gate 	if (pp != NULL) {
7077c478bd9Sstevel@tonic-gate 		VM_STAT_ADD(page_lookup_cnt[1]);
7087c478bd9Sstevel@tonic-gate 		es = (newpp != NULL) ? 1 : 0;
7097c478bd9Sstevel@tonic-gate 		es |= flags;
7107c478bd9Sstevel@tonic-gate 		if (!hash_locked) {
7117c478bd9Sstevel@tonic-gate 			VM_STAT_ADD(page_lookup_cnt[2]);
7127c478bd9Sstevel@tonic-gate 			if (!page_try_reclaim_lock(pp, se, es)) {
7137c478bd9Sstevel@tonic-gate 				/*
7147c478bd9Sstevel@tonic-gate 				 * On a miss, acquire the phm.  Then
7157c478bd9Sstevel@tonic-gate 				 * next time, page_lock() will be called,
7167c478bd9Sstevel@tonic-gate 				 * causing a wait if the page is busy.
7177c478bd9Sstevel@tonic-gate 				 * just looping with page_trylock() would
7187c478bd9Sstevel@tonic-gate 				 * get pretty boring.
7197c478bd9Sstevel@tonic-gate 				 */
7207c478bd9Sstevel@tonic-gate 				VM_STAT_ADD(page_lookup_cnt[3]);
7217c478bd9Sstevel@tonic-gate 				phm = PAGE_HASH_MUTEX(index);
7227c478bd9Sstevel@tonic-gate 				mutex_enter(phm);
7237c478bd9Sstevel@tonic-gate 				hash_locked = 1;
7247c478bd9Sstevel@tonic-gate 				goto top;
7257c478bd9Sstevel@tonic-gate 			}
7267c478bd9Sstevel@tonic-gate 		} else {
7277c478bd9Sstevel@tonic-gate 			VM_STAT_ADD(page_lookup_cnt[4]);
7287c478bd9Sstevel@tonic-gate 			if (!page_lock_es(pp, se, phm, P_RECLAIM, es)) {
7297c478bd9Sstevel@tonic-gate 				VM_STAT_ADD(page_lookup_cnt[5]);
7307c478bd9Sstevel@tonic-gate 				goto top;
7317c478bd9Sstevel@tonic-gate 			}
7327c478bd9Sstevel@tonic-gate 		}
7337c478bd9Sstevel@tonic-gate 
7347c478bd9Sstevel@tonic-gate 		/*
7357c478bd9Sstevel@tonic-gate 		 * Since `pp' is locked it can not change identity now.
7367c478bd9Sstevel@tonic-gate 		 * Reconfirm we locked the correct page.
7377c478bd9Sstevel@tonic-gate 		 *
7387c478bd9Sstevel@tonic-gate 		 * Both the p_vnode and p_offset *must* be cast volatile
7397c478bd9Sstevel@tonic-gate 		 * to force a reload of their values: The PAGE_HASH_SEARCH
7407c478bd9Sstevel@tonic-gate 		 * macro will have stuffed p_vnode and p_offset into
7417c478bd9Sstevel@tonic-gate 		 * registers before calling page_trylock(); another thread,
7427c478bd9Sstevel@tonic-gate 		 * actually holding the hash lock, could have changed the
7437c478bd9Sstevel@tonic-gate 		 * page's identity in memory, but our registers would not
7447c478bd9Sstevel@tonic-gate 		 * be changed, fooling the reconfirmation.  If the hash
7457c478bd9Sstevel@tonic-gate 		 * lock was held during the search, the casting would
7467c478bd9Sstevel@tonic-gate 		 * not be needed.
7477c478bd9Sstevel@tonic-gate 		 */
7487c478bd9Sstevel@tonic-gate 		VM_STAT_ADD(page_lookup_cnt[6]);
7497c478bd9Sstevel@tonic-gate 		if (((volatile struct vnode *)(pp->p_vnode) != vp) ||
7507c478bd9Sstevel@tonic-gate 		    ((volatile u_offset_t)(pp->p_offset) != off)) {
7517c478bd9Sstevel@tonic-gate 			VM_STAT_ADD(page_lookup_cnt[7]);
7527c478bd9Sstevel@tonic-gate 			if (hash_locked) {
7537c478bd9Sstevel@tonic-gate 				panic("page_lookup_create: lost page %p",
7547c478bd9Sstevel@tonic-gate 				    (void *)pp);
7557c478bd9Sstevel@tonic-gate 				/*NOTREACHED*/
7567c478bd9Sstevel@tonic-gate 			}
7577c478bd9Sstevel@tonic-gate 			page_unlock(pp);
7587c478bd9Sstevel@tonic-gate 			phm = PAGE_HASH_MUTEX(index);
7597c478bd9Sstevel@tonic-gate 			mutex_enter(phm);
7607c478bd9Sstevel@tonic-gate 			hash_locked = 1;
7617c478bd9Sstevel@tonic-gate 			goto top;
7627c478bd9Sstevel@tonic-gate 		}
7637c478bd9Sstevel@tonic-gate 
7647c478bd9Sstevel@tonic-gate 		/*
7657c478bd9Sstevel@tonic-gate 		 * If page_trylock() was called, then pp may still be on
7667c478bd9Sstevel@tonic-gate 		 * the cachelist (can't be on the free list, it would not
7677c478bd9Sstevel@tonic-gate 		 * have been found in the search).  If it is on the
7687c478bd9Sstevel@tonic-gate 		 * cachelist it must be pulled now. To pull the page from
7697c478bd9Sstevel@tonic-gate 		 * the cachelist, it must be exclusively locked.
7707c478bd9Sstevel@tonic-gate 		 *
7717c478bd9Sstevel@tonic-gate 		 * The other big difference between page_trylock() and
7727c478bd9Sstevel@tonic-gate 		 * page_lock(), is that page_lock() will pull the
7737c478bd9Sstevel@tonic-gate 		 * page from whatever free list (the cache list in this
7747c478bd9Sstevel@tonic-gate 		 * case) the page is on.  If page_trylock() was used
7757c478bd9Sstevel@tonic-gate 		 * above, then we have to do the reclaim ourselves.
7767c478bd9Sstevel@tonic-gate 		 */
7777c478bd9Sstevel@tonic-gate 		if ((!hash_locked) && (PP_ISFREE(pp))) {
7787c478bd9Sstevel@tonic-gate 			ASSERT(PP_ISAGED(pp) == 0);
7797c478bd9Sstevel@tonic-gate 			VM_STAT_ADD(page_lookup_cnt[8]);
7807c478bd9Sstevel@tonic-gate 
7817c478bd9Sstevel@tonic-gate 			/*
7827c478bd9Sstevel@tonic-gate 			 * page_relcaim will insure that we
7837c478bd9Sstevel@tonic-gate 			 * have this page exclusively
7847c478bd9Sstevel@tonic-gate 			 */
7857c478bd9Sstevel@tonic-gate 
7867c478bd9Sstevel@tonic-gate 			if (!page_reclaim(pp, NULL)) {
7877c478bd9Sstevel@tonic-gate 				/*
7887c478bd9Sstevel@tonic-gate 				 * Page_reclaim dropped whatever lock
7897c478bd9Sstevel@tonic-gate 				 * we held.
7907c478bd9Sstevel@tonic-gate 				 */
7917c478bd9Sstevel@tonic-gate 				VM_STAT_ADD(page_lookup_cnt[9]);
7927c478bd9Sstevel@tonic-gate 				phm = PAGE_HASH_MUTEX(index);
7937c478bd9Sstevel@tonic-gate 				mutex_enter(phm);
7947c478bd9Sstevel@tonic-gate 				hash_locked = 1;
7957c478bd9Sstevel@tonic-gate 				goto top;
7967c478bd9Sstevel@tonic-gate 			} else if (se == SE_SHARED && newpp == NULL) {
7977c478bd9Sstevel@tonic-gate 				VM_STAT_ADD(page_lookup_cnt[10]);
7987c478bd9Sstevel@tonic-gate 				page_downgrade(pp);
7997c478bd9Sstevel@tonic-gate 			}
8007c478bd9Sstevel@tonic-gate 		}
8017c478bd9Sstevel@tonic-gate 
8027c478bd9Sstevel@tonic-gate 		if (hash_locked) {
8037c478bd9Sstevel@tonic-gate 			mutex_exit(phm);
8047c478bd9Sstevel@tonic-gate 		}
8057c478bd9Sstevel@tonic-gate 
8067c478bd9Sstevel@tonic-gate 		if (newpp != NULL && pp->p_szc < newpp->p_szc &&
8077c478bd9Sstevel@tonic-gate 		    PAGE_EXCL(pp) && nrelocp != NULL) {
8087c478bd9Sstevel@tonic-gate 			ASSERT(nrelocp != NULL);
8097c478bd9Sstevel@tonic-gate 			(void) page_relocate(&pp, &newpp, 1, 1, nrelocp,
8107c478bd9Sstevel@tonic-gate 			    NULL);
8117c478bd9Sstevel@tonic-gate 			if (*nrelocp > 0) {
8127c478bd9Sstevel@tonic-gate 				VM_STAT_COND_ADD(*nrelocp == 1,
8137c478bd9Sstevel@tonic-gate 				    page_lookup_cnt[11]);
8147c478bd9Sstevel@tonic-gate 				VM_STAT_COND_ADD(*nrelocp > 1,
8157c478bd9Sstevel@tonic-gate 				    page_lookup_cnt[12]);
8167c478bd9Sstevel@tonic-gate 				pp = newpp;
8177c478bd9Sstevel@tonic-gate 				se = SE_EXCL;
8187c478bd9Sstevel@tonic-gate 			} else {
8197c478bd9Sstevel@tonic-gate 				if (se == SE_SHARED) {
8207c478bd9Sstevel@tonic-gate 					page_downgrade(pp);
8217c478bd9Sstevel@tonic-gate 				}
8227c478bd9Sstevel@tonic-gate 				VM_STAT_ADD(page_lookup_cnt[13]);
8237c478bd9Sstevel@tonic-gate 			}
8247c478bd9Sstevel@tonic-gate 		} else if (newpp != NULL && nrelocp != NULL) {
8257c478bd9Sstevel@tonic-gate 			if (PAGE_EXCL(pp) && se == SE_SHARED) {
8267c478bd9Sstevel@tonic-gate 				page_downgrade(pp);
8277c478bd9Sstevel@tonic-gate 			}
8287c478bd9Sstevel@tonic-gate 			VM_STAT_COND_ADD(pp->p_szc < newpp->p_szc,
8297c478bd9Sstevel@tonic-gate 			    page_lookup_cnt[14]);
8307c478bd9Sstevel@tonic-gate 			VM_STAT_COND_ADD(pp->p_szc == newpp->p_szc,
8317c478bd9Sstevel@tonic-gate 			    page_lookup_cnt[15]);
8327c478bd9Sstevel@tonic-gate 			VM_STAT_COND_ADD(pp->p_szc > newpp->p_szc,
8337c478bd9Sstevel@tonic-gate 			    page_lookup_cnt[16]);
8347c478bd9Sstevel@tonic-gate 		} else if (newpp != NULL && PAGE_EXCL(pp)) {
8357c478bd9Sstevel@tonic-gate 			se = SE_EXCL;
8367c478bd9Sstevel@tonic-gate 		}
8377c478bd9Sstevel@tonic-gate 	} else if (!hash_locked) {
8387c478bd9Sstevel@tonic-gate 		VM_STAT_ADD(page_lookup_cnt[17]);
8397c478bd9Sstevel@tonic-gate 		phm = PAGE_HASH_MUTEX(index);
8407c478bd9Sstevel@tonic-gate 		mutex_enter(phm);
8417c478bd9Sstevel@tonic-gate 		hash_locked = 1;
8427c478bd9Sstevel@tonic-gate 		goto top;
8437c478bd9Sstevel@tonic-gate 	} else if (newpp != NULL) {
8447c478bd9Sstevel@tonic-gate 		/*
8457c478bd9Sstevel@tonic-gate 		 * If we have a preallocated page then
8467c478bd9Sstevel@tonic-gate 		 * insert it now and basically behave like
8477c478bd9Sstevel@tonic-gate 		 * page_create.
8487c478bd9Sstevel@tonic-gate 		 */
8497c478bd9Sstevel@tonic-gate 		VM_STAT_ADD(page_lookup_cnt[18]);
8507c478bd9Sstevel@tonic-gate 		/*
8517c478bd9Sstevel@tonic-gate 		 * Since we hold the page hash mutex and
8527c478bd9Sstevel@tonic-gate 		 * just searched for this page, page_hashin
8537c478bd9Sstevel@tonic-gate 		 * had better not fail.  If it does, that
8547c478bd9Sstevel@tonic-gate 		 * means some thread did not follow the
8557c478bd9Sstevel@tonic-gate 		 * page hash mutex rules.  Panic now and
8567c478bd9Sstevel@tonic-gate 		 * get it over with.  As usual, go down
8577c478bd9Sstevel@tonic-gate 		 * holding all the locks.
8587c478bd9Sstevel@tonic-gate 		 */
8597c478bd9Sstevel@tonic-gate 		ASSERT(MUTEX_HELD(phm));
8607c478bd9Sstevel@tonic-gate 		if (!page_hashin(newpp, vp, off, phm)) {
8617c478bd9Sstevel@tonic-gate 			ASSERT(MUTEX_HELD(phm));
8627c478bd9Sstevel@tonic-gate 			panic("page_lookup_create: hashin failed %p %p %llx %p",
8637c478bd9Sstevel@tonic-gate 			    (void *)newpp, (void *)vp, off, (void *)phm);
8647c478bd9Sstevel@tonic-gate 			/*NOTREACHED*/
8657c478bd9Sstevel@tonic-gate 		}
8667c478bd9Sstevel@tonic-gate 		ASSERT(MUTEX_HELD(phm));
8677c478bd9Sstevel@tonic-gate 		mutex_exit(phm);
8687c478bd9Sstevel@tonic-gate 		phm = NULL;
8697c478bd9Sstevel@tonic-gate 		page_set_props(newpp, P_REF);
8707c478bd9Sstevel@tonic-gate 		page_io_lock(newpp);
8717c478bd9Sstevel@tonic-gate 		pp = newpp;
8727c478bd9Sstevel@tonic-gate 		se = SE_EXCL;
8737c478bd9Sstevel@tonic-gate 	} else {
8747c478bd9Sstevel@tonic-gate 		VM_STAT_ADD(page_lookup_cnt[19]);
8757c478bd9Sstevel@tonic-gate 		mutex_exit(phm);
8767c478bd9Sstevel@tonic-gate 	}
8777c478bd9Sstevel@tonic-gate 
8787c478bd9Sstevel@tonic-gate 	ASSERT(pp ? PAGE_LOCKED_SE(pp, se) : 1);
8797c478bd9Sstevel@tonic-gate 
8807c478bd9Sstevel@tonic-gate 	ASSERT(pp ? ((PP_ISFREE(pp) == 0) && (PP_ISAGED(pp) == 0)) : 1);
8817c478bd9Sstevel@tonic-gate 
8827c478bd9Sstevel@tonic-gate 	return (pp);
8837c478bd9Sstevel@tonic-gate }
8847c478bd9Sstevel@tonic-gate 
8857c478bd9Sstevel@tonic-gate /*
8867c478bd9Sstevel@tonic-gate  * Search the hash list for the page representing the
8877c478bd9Sstevel@tonic-gate  * specified [vp, offset] and return it locked.  Skip
8887c478bd9Sstevel@tonic-gate  * free pages and pages that cannot be locked as requested.
8897c478bd9Sstevel@tonic-gate  * Used while attempting to kluster pages.
8907c478bd9Sstevel@tonic-gate  */
8917c478bd9Sstevel@tonic-gate page_t *
8927c478bd9Sstevel@tonic-gate page_lookup_nowait(vnode_t *vp, u_offset_t off, se_t se)
8937c478bd9Sstevel@tonic-gate {
8947c478bd9Sstevel@tonic-gate 	page_t		*pp;
8957c478bd9Sstevel@tonic-gate 	kmutex_t	*phm;
8967c478bd9Sstevel@tonic-gate 	ulong_t		index;
8977c478bd9Sstevel@tonic-gate 	uint_t		locked;
8987c478bd9Sstevel@tonic-gate 
8997c478bd9Sstevel@tonic-gate 	ASSERT(MUTEX_NOT_HELD(page_vnode_mutex(vp)));
9007c478bd9Sstevel@tonic-gate 	VM_STAT_ADD(page_lookup_nowait_cnt[0]);
9017c478bd9Sstevel@tonic-gate 
9027c478bd9Sstevel@tonic-gate 	index = PAGE_HASH_FUNC(vp, off);
9037c478bd9Sstevel@tonic-gate 	PAGE_HASH_SEARCH(index, pp, vp, off);
9047c478bd9Sstevel@tonic-gate 	locked = 0;
9057c478bd9Sstevel@tonic-gate 	if (pp == NULL) {
9067c478bd9Sstevel@tonic-gate top:
9077c478bd9Sstevel@tonic-gate 		VM_STAT_ADD(page_lookup_nowait_cnt[1]);
9087c478bd9Sstevel@tonic-gate 		locked = 1;
9097c478bd9Sstevel@tonic-gate 		phm = PAGE_HASH_MUTEX(index);
9107c478bd9Sstevel@tonic-gate 		mutex_enter(phm);
9117c478bd9Sstevel@tonic-gate 		PAGE_HASH_SEARCH(index, pp, vp, off);
9127c478bd9Sstevel@tonic-gate 	}
9137c478bd9Sstevel@tonic-gate 
9147c478bd9Sstevel@tonic-gate 	if (pp == NULL || PP_ISFREE(pp)) {
9157c478bd9Sstevel@tonic-gate 		VM_STAT_ADD(page_lookup_nowait_cnt[2]);
9167c478bd9Sstevel@tonic-gate 		pp = NULL;
9177c478bd9Sstevel@tonic-gate 	} else {
9187c478bd9Sstevel@tonic-gate 		if (!page_trylock(pp, se)) {
9197c478bd9Sstevel@tonic-gate 			VM_STAT_ADD(page_lookup_nowait_cnt[3]);
9207c478bd9Sstevel@tonic-gate 			pp = NULL;
9217c478bd9Sstevel@tonic-gate 		} else {
9227c478bd9Sstevel@tonic-gate 			VM_STAT_ADD(page_lookup_nowait_cnt[4]);
9237c478bd9Sstevel@tonic-gate 			/*
9247c478bd9Sstevel@tonic-gate 			 * See the comment in page_lookup()
9257c478bd9Sstevel@tonic-gate 			 */
9267c478bd9Sstevel@tonic-gate 			if (((volatile struct vnode *)(pp->p_vnode) != vp) ||
9277c478bd9Sstevel@tonic-gate 			    ((u_offset_t)(pp->p_offset) != off)) {
9287c478bd9Sstevel@tonic-gate 				VM_STAT_ADD(page_lookup_nowait_cnt[5]);
9297c478bd9Sstevel@tonic-gate 				if (locked) {
9307c478bd9Sstevel@tonic-gate 					panic("page_lookup_nowait %p",
9317c478bd9Sstevel@tonic-gate 					    (void *)pp);
9327c478bd9Sstevel@tonic-gate 					/*NOTREACHED*/
9337c478bd9Sstevel@tonic-gate 				}
9347c478bd9Sstevel@tonic-gate 				page_unlock(pp);
9357c478bd9Sstevel@tonic-gate 				goto top;
9367c478bd9Sstevel@tonic-gate 			}
9377c478bd9Sstevel@tonic-gate 			if (PP_ISFREE(pp)) {
9387c478bd9Sstevel@tonic-gate 				VM_STAT_ADD(page_lookup_nowait_cnt[6]);
9397c478bd9Sstevel@tonic-gate 				page_unlock(pp);
9407c478bd9Sstevel@tonic-gate 				pp = NULL;
9417c478bd9Sstevel@tonic-gate 			}
9427c478bd9Sstevel@tonic-gate 		}
9437c478bd9Sstevel@tonic-gate 	}
9447c478bd9Sstevel@tonic-gate 	if (locked) {
9457c478bd9Sstevel@tonic-gate 		VM_STAT_ADD(page_lookup_nowait_cnt[7]);
9467c478bd9Sstevel@tonic-gate 		mutex_exit(phm);
9477c478bd9Sstevel@tonic-gate 	}
9487c478bd9Sstevel@tonic-gate 
9497c478bd9Sstevel@tonic-gate 	ASSERT(pp ? PAGE_LOCKED_SE(pp, se) : 1);
9507c478bd9Sstevel@tonic-gate 
9517c478bd9Sstevel@tonic-gate 	return (pp);
9527c478bd9Sstevel@tonic-gate }
9537c478bd9Sstevel@tonic-gate 
9547c478bd9Sstevel@tonic-gate /*
9557c478bd9Sstevel@tonic-gate  * Search the hash list for a page with the specified [vp, off]
9567c478bd9Sstevel@tonic-gate  * that is known to exist and is already locked.  This routine
9577c478bd9Sstevel@tonic-gate  * is typically used by segment SOFTUNLOCK routines.
9587c478bd9Sstevel@tonic-gate  */
9597c478bd9Sstevel@tonic-gate page_t *
9607c478bd9Sstevel@tonic-gate page_find(vnode_t *vp, u_offset_t off)
9617c478bd9Sstevel@tonic-gate {
9627c478bd9Sstevel@tonic-gate 	page_t		*pp;
9637c478bd9Sstevel@tonic-gate 	kmutex_t	*phm;
9647c478bd9Sstevel@tonic-gate 	ulong_t		index;
9657c478bd9Sstevel@tonic-gate 
9667c478bd9Sstevel@tonic-gate 	ASSERT(MUTEX_NOT_HELD(page_vnode_mutex(vp)));
9677c478bd9Sstevel@tonic-gate 	VM_STAT_ADD(page_find_cnt);
9687c478bd9Sstevel@tonic-gate 
9697c478bd9Sstevel@tonic-gate 	index = PAGE_HASH_FUNC(vp, off);
9707c478bd9Sstevel@tonic-gate 	phm = PAGE_HASH_MUTEX(index);
9717c478bd9Sstevel@tonic-gate 
9727c478bd9Sstevel@tonic-gate 	mutex_enter(phm);
9737c478bd9Sstevel@tonic-gate 	PAGE_HASH_SEARCH(index, pp, vp, off);
9747c478bd9Sstevel@tonic-gate 	mutex_exit(phm);
9757c478bd9Sstevel@tonic-gate 
9764fc2445aSelowe 	ASSERT(pp == NULL || PAGE_LOCKED(pp) || panicstr);
9777c478bd9Sstevel@tonic-gate 	return (pp);
9787c478bd9Sstevel@tonic-gate }
9797c478bd9Sstevel@tonic-gate 
9807c478bd9Sstevel@tonic-gate /*
9817c478bd9Sstevel@tonic-gate  * Determine whether a page with the specified [vp, off]
9827c478bd9Sstevel@tonic-gate  * currently exists in the system.  Obviously this should
9837c478bd9Sstevel@tonic-gate  * only be considered as a hint since nothing prevents the
9847c478bd9Sstevel@tonic-gate  * page from disappearing or appearing immediately after
9857c478bd9Sstevel@tonic-gate  * the return from this routine. Subsequently, we don't
9867c478bd9Sstevel@tonic-gate  * even bother to lock the list.
9877c478bd9Sstevel@tonic-gate  */
9887c478bd9Sstevel@tonic-gate page_t *
9897c478bd9Sstevel@tonic-gate page_exists(vnode_t *vp, u_offset_t off)
9907c478bd9Sstevel@tonic-gate {
9917c478bd9Sstevel@tonic-gate 	page_t	*pp;
9927c478bd9Sstevel@tonic-gate 	ulong_t		index;
9937c478bd9Sstevel@tonic-gate 
9947c478bd9Sstevel@tonic-gate 	ASSERT(MUTEX_NOT_HELD(page_vnode_mutex(vp)));
9957c478bd9Sstevel@tonic-gate 	VM_STAT_ADD(page_exists_cnt);
9967c478bd9Sstevel@tonic-gate 
9977c478bd9Sstevel@tonic-gate 	index = PAGE_HASH_FUNC(vp, off);
9987c478bd9Sstevel@tonic-gate 	PAGE_HASH_SEARCH(index, pp, vp, off);
9997c478bd9Sstevel@tonic-gate 
10007c478bd9Sstevel@tonic-gate 	return (pp);
10017c478bd9Sstevel@tonic-gate }
10027c478bd9Sstevel@tonic-gate 
10037c478bd9Sstevel@tonic-gate /*
10047c478bd9Sstevel@tonic-gate  * Determine if physically contiguous pages exist for [vp, off] - [vp, off +
10057c478bd9Sstevel@tonic-gate  * page_size(szc)) range.  if they exist and ppa is not NULL fill ppa array
10067c478bd9Sstevel@tonic-gate  * with these pages locked SHARED. If necessary reclaim pages from
10077c478bd9Sstevel@tonic-gate  * freelist. Return 1 if contiguous pages exist and 0 otherwise.
10087c478bd9Sstevel@tonic-gate  *
10097c478bd9Sstevel@tonic-gate  * If we fail to lock pages still return 1 if pages exist and contiguous.
10107c478bd9Sstevel@tonic-gate  * But in this case return value is just a hint. ppa array won't be filled.
10117c478bd9Sstevel@tonic-gate  * Caller should initialize ppa[0] as NULL to distinguish return value.
10127c478bd9Sstevel@tonic-gate  *
10137c478bd9Sstevel@tonic-gate  * Returns 0 if pages don't exist or not physically contiguous.
10147c478bd9Sstevel@tonic-gate  *
10157c478bd9Sstevel@tonic-gate  * This routine doesn't work for anonymous(swapfs) pages.
10167c478bd9Sstevel@tonic-gate  */
10177c478bd9Sstevel@tonic-gate int
10187c478bd9Sstevel@tonic-gate page_exists_physcontig(vnode_t *vp, u_offset_t off, uint_t szc, page_t *ppa[])
10197c478bd9Sstevel@tonic-gate {
10207c478bd9Sstevel@tonic-gate 	pgcnt_t pages;
10217c478bd9Sstevel@tonic-gate 	pfn_t pfn;
10227c478bd9Sstevel@tonic-gate 	page_t *rootpp;
10237c478bd9Sstevel@tonic-gate 	pgcnt_t i;
10247c478bd9Sstevel@tonic-gate 	pgcnt_t j;
10257c478bd9Sstevel@tonic-gate 	u_offset_t save_off = off;
10267c478bd9Sstevel@tonic-gate 	ulong_t index;
10277c478bd9Sstevel@tonic-gate 	kmutex_t *phm;
10287c478bd9Sstevel@tonic-gate 	page_t *pp;
10297c478bd9Sstevel@tonic-gate 	uint_t pszc;
10307c478bd9Sstevel@tonic-gate 	int loopcnt = 0;
10317c478bd9Sstevel@tonic-gate 
10327c478bd9Sstevel@tonic-gate 	ASSERT(szc != 0);
10337c478bd9Sstevel@tonic-gate 	ASSERT(vp != NULL);
10347c478bd9Sstevel@tonic-gate 	ASSERT(!IS_SWAPFSVP(vp));
10357c478bd9Sstevel@tonic-gate 	ASSERT(vp != &kvp);
10367c478bd9Sstevel@tonic-gate 
10377c478bd9Sstevel@tonic-gate again:
10387c478bd9Sstevel@tonic-gate 	if (++loopcnt > 3) {
10397c478bd9Sstevel@tonic-gate 		VM_STAT_ADD(page_exphcontg[0]);
10407c478bd9Sstevel@tonic-gate 		return (0);
10417c478bd9Sstevel@tonic-gate 	}
10427c478bd9Sstevel@tonic-gate 
10437c478bd9Sstevel@tonic-gate 	index = PAGE_HASH_FUNC(vp, off);
10447c478bd9Sstevel@tonic-gate 	phm = PAGE_HASH_MUTEX(index);
10457c478bd9Sstevel@tonic-gate 
10467c478bd9Sstevel@tonic-gate 	mutex_enter(phm);
10477c478bd9Sstevel@tonic-gate 	PAGE_HASH_SEARCH(index, pp, vp, off);
10487c478bd9Sstevel@tonic-gate 	mutex_exit(phm);
10497c478bd9Sstevel@tonic-gate 
10507c478bd9Sstevel@tonic-gate 	VM_STAT_ADD(page_exphcontg[1]);
10517c478bd9Sstevel@tonic-gate 
10527c478bd9Sstevel@tonic-gate 	if (pp == NULL) {
10537c478bd9Sstevel@tonic-gate 		VM_STAT_ADD(page_exphcontg[2]);
10547c478bd9Sstevel@tonic-gate 		return (0);
10557c478bd9Sstevel@tonic-gate 	}
10567c478bd9Sstevel@tonic-gate 
10577c478bd9Sstevel@tonic-gate 	pages = page_get_pagecnt(szc);
10587c478bd9Sstevel@tonic-gate 	rootpp = pp;
10597c478bd9Sstevel@tonic-gate 	pfn = rootpp->p_pagenum;
10607c478bd9Sstevel@tonic-gate 
10617c478bd9Sstevel@tonic-gate 	if ((pszc = pp->p_szc) >= szc && ppa != NULL) {
10627c478bd9Sstevel@tonic-gate 		VM_STAT_ADD(page_exphcontg[3]);
10637c478bd9Sstevel@tonic-gate 		if (!page_trylock(pp, SE_SHARED)) {
10647c478bd9Sstevel@tonic-gate 			VM_STAT_ADD(page_exphcontg[4]);
10657c478bd9Sstevel@tonic-gate 			return (1);
10667c478bd9Sstevel@tonic-gate 		}
10677c478bd9Sstevel@tonic-gate 		if (pp->p_szc != pszc || pp->p_vnode != vp ||
10687c478bd9Sstevel@tonic-gate 		    pp->p_offset != off) {
10697c478bd9Sstevel@tonic-gate 			VM_STAT_ADD(page_exphcontg[5]);
10707c478bd9Sstevel@tonic-gate 			page_unlock(pp);
10717c478bd9Sstevel@tonic-gate 			off = save_off;
10727c478bd9Sstevel@tonic-gate 			goto again;
10737c478bd9Sstevel@tonic-gate 		}
10747c478bd9Sstevel@tonic-gate 		/*
10757c478bd9Sstevel@tonic-gate 		 * szc was non zero and vnode and offset matched after we
10767c478bd9Sstevel@tonic-gate 		 * locked the page it means it can't become free on us.
10777c478bd9Sstevel@tonic-gate 		 */
10787c478bd9Sstevel@tonic-gate 		ASSERT(!PP_ISFREE(pp));
10797c478bd9Sstevel@tonic-gate 		if (!IS_P2ALIGNED(pfn, pages)) {
10807c478bd9Sstevel@tonic-gate 			page_unlock(pp);
10817c478bd9Sstevel@tonic-gate 			return (0);
10827c478bd9Sstevel@tonic-gate 		}
10837c478bd9Sstevel@tonic-gate 		ppa[0] = pp;
10847c478bd9Sstevel@tonic-gate 		pp++;
10857c478bd9Sstevel@tonic-gate 		off += PAGESIZE;
10867c478bd9Sstevel@tonic-gate 		pfn++;
10877c478bd9Sstevel@tonic-gate 		for (i = 1; i < pages; i++, pp++, off += PAGESIZE, pfn++) {
10887c478bd9Sstevel@tonic-gate 			if (!page_trylock(pp, SE_SHARED)) {
10897c478bd9Sstevel@tonic-gate 				VM_STAT_ADD(page_exphcontg[6]);
10907c478bd9Sstevel@tonic-gate 				pp--;
10917c478bd9Sstevel@tonic-gate 				while (i-- > 0) {
10927c478bd9Sstevel@tonic-gate 					page_unlock(pp);
10937c478bd9Sstevel@tonic-gate 					pp--;
10947c478bd9Sstevel@tonic-gate 				}
10957c478bd9Sstevel@tonic-gate 				ppa[0] = NULL;
10967c478bd9Sstevel@tonic-gate 				return (1);
10977c478bd9Sstevel@tonic-gate 			}
10987c478bd9Sstevel@tonic-gate 			if (pp->p_szc != pszc) {
10997c478bd9Sstevel@tonic-gate 				VM_STAT_ADD(page_exphcontg[7]);
11007c478bd9Sstevel@tonic-gate 				page_unlock(pp);
11017c478bd9Sstevel@tonic-gate 				pp--;
11027c478bd9Sstevel@tonic-gate 				while (i-- > 0) {
11037c478bd9Sstevel@tonic-gate 					page_unlock(pp);
11047c478bd9Sstevel@tonic-gate 					pp--;
11057c478bd9Sstevel@tonic-gate 				}
11067c478bd9Sstevel@tonic-gate 				ppa[0] = NULL;
11077c478bd9Sstevel@tonic-gate 				off = save_off;
11087c478bd9Sstevel@tonic-gate 				goto again;
11097c478bd9Sstevel@tonic-gate 			}
11107c478bd9Sstevel@tonic-gate 			/*
11117c478bd9Sstevel@tonic-gate 			 * szc the same as for previous already locked pages
11127c478bd9Sstevel@tonic-gate 			 * with right identity. Since this page had correct
11137c478bd9Sstevel@tonic-gate 			 * szc after we locked it can't get freed or destroyed
11147c478bd9Sstevel@tonic-gate 			 * and therefore must have the expected identity.
11157c478bd9Sstevel@tonic-gate 			 */
11167c478bd9Sstevel@tonic-gate 			ASSERT(!PP_ISFREE(pp));
11177c478bd9Sstevel@tonic-gate 			if (pp->p_vnode != vp ||
11187c478bd9Sstevel@tonic-gate 			    pp->p_offset != off) {
11197c478bd9Sstevel@tonic-gate 				panic("page_exists_physcontig: "
11207c478bd9Sstevel@tonic-gate 				    "large page identity doesn't match");
11217c478bd9Sstevel@tonic-gate 			}
11227c478bd9Sstevel@tonic-gate 			ppa[i] = pp;
11237c478bd9Sstevel@tonic-gate 			ASSERT(pp->p_pagenum == pfn);
11247c478bd9Sstevel@tonic-gate 		}
11257c478bd9Sstevel@tonic-gate 		VM_STAT_ADD(page_exphcontg[8]);
11267c478bd9Sstevel@tonic-gate 		ppa[pages] = NULL;
11277c478bd9Sstevel@tonic-gate 		return (1);
11287c478bd9Sstevel@tonic-gate 	} else if (pszc >= szc) {
11297c478bd9Sstevel@tonic-gate 		VM_STAT_ADD(page_exphcontg[9]);
11307c478bd9Sstevel@tonic-gate 		if (!IS_P2ALIGNED(pfn, pages)) {
11317c478bd9Sstevel@tonic-gate 			return (0);
11327c478bd9Sstevel@tonic-gate 		}
11337c478bd9Sstevel@tonic-gate 		return (1);
11347c478bd9Sstevel@tonic-gate 	}
11357c478bd9Sstevel@tonic-gate 
11367c478bd9Sstevel@tonic-gate 	if (!IS_P2ALIGNED(pfn, pages)) {
11377c478bd9Sstevel@tonic-gate 		VM_STAT_ADD(page_exphcontg[10]);
11387c478bd9Sstevel@tonic-gate 		return (0);
11397c478bd9Sstevel@tonic-gate 	}
11407c478bd9Sstevel@tonic-gate 
11417c478bd9Sstevel@tonic-gate 	if (page_numtomemseg_nolock(pfn) !=
11427c478bd9Sstevel@tonic-gate 	    page_numtomemseg_nolock(pfn + pages - 1)) {
11437c478bd9Sstevel@tonic-gate 		VM_STAT_ADD(page_exphcontg[11]);
11447c478bd9Sstevel@tonic-gate 		return (0);
11457c478bd9Sstevel@tonic-gate 	}
11467c478bd9Sstevel@tonic-gate 
11477c478bd9Sstevel@tonic-gate 	/*
11487c478bd9Sstevel@tonic-gate 	 * We loop up 4 times across pages to promote page size.
11497c478bd9Sstevel@tonic-gate 	 * We're extra cautious to promote page size atomically with respect
11507c478bd9Sstevel@tonic-gate 	 * to everybody else.  But we can probably optimize into 1 loop if
11517c478bd9Sstevel@tonic-gate 	 * this becomes an issue.
11527c478bd9Sstevel@tonic-gate 	 */
11537c478bd9Sstevel@tonic-gate 
11547c478bd9Sstevel@tonic-gate 	for (i = 0; i < pages; i++, pp++, off += PAGESIZE, pfn++) {
11557c478bd9Sstevel@tonic-gate 		ASSERT(pp->p_pagenum == pfn);
11567c478bd9Sstevel@tonic-gate 		if (!page_trylock(pp, SE_EXCL)) {
11577c478bd9Sstevel@tonic-gate 			VM_STAT_ADD(page_exphcontg[12]);
11587c478bd9Sstevel@tonic-gate 			break;
11597c478bd9Sstevel@tonic-gate 		}
11607c478bd9Sstevel@tonic-gate 		if (pp->p_vnode != vp ||
11617c478bd9Sstevel@tonic-gate 		    pp->p_offset != off) {
11627c478bd9Sstevel@tonic-gate 			VM_STAT_ADD(page_exphcontg[13]);
11637c478bd9Sstevel@tonic-gate 			page_unlock(pp);
11647c478bd9Sstevel@tonic-gate 			break;
11657c478bd9Sstevel@tonic-gate 		}
11667c478bd9Sstevel@tonic-gate 		if (pp->p_szc >= szc) {
11677c478bd9Sstevel@tonic-gate 			ASSERT(i == 0);
11687c478bd9Sstevel@tonic-gate 			page_unlock(pp);
11697c478bd9Sstevel@tonic-gate 			off = save_off;
11707c478bd9Sstevel@tonic-gate 			goto again;
11717c478bd9Sstevel@tonic-gate 		}
11727c478bd9Sstevel@tonic-gate 	}
11737c478bd9Sstevel@tonic-gate 
11747c478bd9Sstevel@tonic-gate 	if (i != pages) {
11757c478bd9Sstevel@tonic-gate 		VM_STAT_ADD(page_exphcontg[14]);
11767c478bd9Sstevel@tonic-gate 		--pp;
11777c478bd9Sstevel@tonic-gate 		while (i-- > 0) {
11787c478bd9Sstevel@tonic-gate 			page_unlock(pp);
11797c478bd9Sstevel@tonic-gate 			--pp;
11807c478bd9Sstevel@tonic-gate 		}
11817c478bd9Sstevel@tonic-gate 		return (0);
11827c478bd9Sstevel@tonic-gate 	}
11837c478bd9Sstevel@tonic-gate 
11847c478bd9Sstevel@tonic-gate 	pp = rootpp;
11857c478bd9Sstevel@tonic-gate 	for (i = 0; i < pages; i++, pp++) {
11867c478bd9Sstevel@tonic-gate 		if (PP_ISFREE(pp)) {
11877c478bd9Sstevel@tonic-gate 			VM_STAT_ADD(page_exphcontg[15]);
11887c478bd9Sstevel@tonic-gate 			ASSERT(!PP_ISAGED(pp));
11897c478bd9Sstevel@tonic-gate 			ASSERT(pp->p_szc == 0);
11907c478bd9Sstevel@tonic-gate 			if (!page_reclaim(pp, NULL)) {
11917c478bd9Sstevel@tonic-gate 				break;
11927c478bd9Sstevel@tonic-gate 			}
11937c478bd9Sstevel@tonic-gate 		} else {
11947c478bd9Sstevel@tonic-gate 			ASSERT(pp->p_szc < szc);
11957c478bd9Sstevel@tonic-gate 			VM_STAT_ADD(page_exphcontg[16]);
11967c478bd9Sstevel@tonic-gate 			(void) hat_pageunload(pp, HAT_FORCE_PGUNLOAD);
11977c478bd9Sstevel@tonic-gate 		}
11987c478bd9Sstevel@tonic-gate 	}
11997c478bd9Sstevel@tonic-gate 	if (i < pages) {
12007c478bd9Sstevel@tonic-gate 		VM_STAT_ADD(page_exphcontg[17]);
12017c478bd9Sstevel@tonic-gate 		/*
12027c478bd9Sstevel@tonic-gate 		 * page_reclaim failed because we were out of memory.
12037c478bd9Sstevel@tonic-gate 		 * drop the rest of the locks and return because this page
12047c478bd9Sstevel@tonic-gate 		 * must be already reallocated anyway.
12057c478bd9Sstevel@tonic-gate 		 */
12067c478bd9Sstevel@tonic-gate 		pp = rootpp;
12077c478bd9Sstevel@tonic-gate 		for (j = 0; j < pages; j++, pp++) {
12087c478bd9Sstevel@tonic-gate 			if (j != i) {
12097c478bd9Sstevel@tonic-gate 				page_unlock(pp);
12107c478bd9Sstevel@tonic-gate 			}
12117c478bd9Sstevel@tonic-gate 		}
12127c478bd9Sstevel@tonic-gate 		return (0);
12137c478bd9Sstevel@tonic-gate 	}
12147c478bd9Sstevel@tonic-gate 
12157c478bd9Sstevel@tonic-gate 	off = save_off;
12167c478bd9Sstevel@tonic-gate 	pp = rootpp;
12177c478bd9Sstevel@tonic-gate 	for (i = 0; i < pages; i++, pp++, off += PAGESIZE) {
12187c478bd9Sstevel@tonic-gate 		ASSERT(PAGE_EXCL(pp));
12197c478bd9Sstevel@tonic-gate 		ASSERT(!PP_ISFREE(pp));
12207c478bd9Sstevel@tonic-gate 		ASSERT(!hat_page_is_mapped(pp));
12217c478bd9Sstevel@tonic-gate 		ASSERT(pp->p_vnode == vp);
12227c478bd9Sstevel@tonic-gate 		ASSERT(pp->p_offset == off);
12237c478bd9Sstevel@tonic-gate 		pp->p_szc = szc;
12247c478bd9Sstevel@tonic-gate 	}
12257c478bd9Sstevel@tonic-gate 	pp = rootpp;
12267c478bd9Sstevel@tonic-gate 	for (i = 0; i < pages; i++, pp++) {
12277c478bd9Sstevel@tonic-gate 		if (ppa == NULL) {
12287c478bd9Sstevel@tonic-gate 			page_unlock(pp);
12297c478bd9Sstevel@tonic-gate 		} else {
12307c478bd9Sstevel@tonic-gate 			ppa[i] = pp;
12317c478bd9Sstevel@tonic-gate 			page_downgrade(ppa[i]);
12327c478bd9Sstevel@tonic-gate 		}
12337c478bd9Sstevel@tonic-gate 	}
12347c478bd9Sstevel@tonic-gate 	if (ppa != NULL) {
12357c478bd9Sstevel@tonic-gate 		ppa[pages] = NULL;
12367c478bd9Sstevel@tonic-gate 	}
12377c478bd9Sstevel@tonic-gate 	VM_STAT_ADD(page_exphcontg[18]);
12387c478bd9Sstevel@tonic-gate 	ASSERT(vp->v_pages != NULL);
12397c478bd9Sstevel@tonic-gate 	return (1);
12407c478bd9Sstevel@tonic-gate }
12417c478bd9Sstevel@tonic-gate 
12427c478bd9Sstevel@tonic-gate /*
12437c478bd9Sstevel@tonic-gate  * Determine whether a page with the specified [vp, off]
12447c478bd9Sstevel@tonic-gate  * currently exists in the system and if so return its
12457c478bd9Sstevel@tonic-gate  * size code. Obviously this should only be considered as
12467c478bd9Sstevel@tonic-gate  * a hint since nothing prevents the page from disappearing
12477c478bd9Sstevel@tonic-gate  * or appearing immediately after the return from this routine.
12487c478bd9Sstevel@tonic-gate  */
12497c478bd9Sstevel@tonic-gate int
12507c478bd9Sstevel@tonic-gate page_exists_forreal(vnode_t *vp, u_offset_t off, uint_t *szc)
12517c478bd9Sstevel@tonic-gate {
12527c478bd9Sstevel@tonic-gate 	page_t		*pp;
12537c478bd9Sstevel@tonic-gate 	kmutex_t	*phm;
12547c478bd9Sstevel@tonic-gate 	ulong_t		index;
12557c478bd9Sstevel@tonic-gate 	int		rc = 0;
12567c478bd9Sstevel@tonic-gate 
12577c478bd9Sstevel@tonic-gate 	ASSERT(MUTEX_NOT_HELD(page_vnode_mutex(vp)));
12587c478bd9Sstevel@tonic-gate 	ASSERT(szc != NULL);
12597c478bd9Sstevel@tonic-gate 	VM_STAT_ADD(page_exists_forreal_cnt);
12607c478bd9Sstevel@tonic-gate 
12617c478bd9Sstevel@tonic-gate 	index = PAGE_HASH_FUNC(vp, off);
12627c478bd9Sstevel@tonic-gate 	phm = PAGE_HASH_MUTEX(index);
12637c478bd9Sstevel@tonic-gate 
12647c478bd9Sstevel@tonic-gate 	mutex_enter(phm);
12657c478bd9Sstevel@tonic-gate 	PAGE_HASH_SEARCH(index, pp, vp, off);
12667c478bd9Sstevel@tonic-gate 	if (pp != NULL) {
12677c478bd9Sstevel@tonic-gate 		*szc = pp->p_szc;
12687c478bd9Sstevel@tonic-gate 		rc = 1;
12697c478bd9Sstevel@tonic-gate 	}
12707c478bd9Sstevel@tonic-gate 	mutex_exit(phm);
12717c478bd9Sstevel@tonic-gate 	return (rc);
12727c478bd9Sstevel@tonic-gate }
12737c478bd9Sstevel@tonic-gate 
12747c478bd9Sstevel@tonic-gate /* wakeup threads waiting for pages in page_create_get_something() */
12757c478bd9Sstevel@tonic-gate void
12767c478bd9Sstevel@tonic-gate wakeup_pcgs(void)
12777c478bd9Sstevel@tonic-gate {
12787c478bd9Sstevel@tonic-gate 	if (!CV_HAS_WAITERS(&pcgs_cv))
12797c478bd9Sstevel@tonic-gate 		return;
12807c478bd9Sstevel@tonic-gate 	cv_broadcast(&pcgs_cv);
12817c478bd9Sstevel@tonic-gate }
12827c478bd9Sstevel@tonic-gate 
12837c478bd9Sstevel@tonic-gate /*
12847c478bd9Sstevel@tonic-gate  * 'freemem' is used all over the kernel as an indication of how many
12857c478bd9Sstevel@tonic-gate  * pages are free (either on the cache list or on the free page list)
12867c478bd9Sstevel@tonic-gate  * in the system.  In very few places is a really accurate 'freemem'
12877c478bd9Sstevel@tonic-gate  * needed.  To avoid contention of the lock protecting a the
12887c478bd9Sstevel@tonic-gate  * single freemem, it was spread out into NCPU buckets.  Set_freemem
12897c478bd9Sstevel@tonic-gate  * sets freemem to the total of all NCPU buckets.  It is called from
12907c478bd9Sstevel@tonic-gate  * clock() on each TICK.
12917c478bd9Sstevel@tonic-gate  */
12927c478bd9Sstevel@tonic-gate void
12937c478bd9Sstevel@tonic-gate set_freemem()
12947c478bd9Sstevel@tonic-gate {
12957c478bd9Sstevel@tonic-gate 	struct pcf	*p;
12967c478bd9Sstevel@tonic-gate 	ulong_t		t;
12977c478bd9Sstevel@tonic-gate 	uint_t		i;
12987c478bd9Sstevel@tonic-gate 
12997c478bd9Sstevel@tonic-gate 	t = 0;
13007c478bd9Sstevel@tonic-gate 	p = pcf;
13017c478bd9Sstevel@tonic-gate 	for (i = 0;  i < PCF_FANOUT; i++) {
13027c478bd9Sstevel@tonic-gate 		t += p->pcf_count;
13037c478bd9Sstevel@tonic-gate 		p++;
13047c478bd9Sstevel@tonic-gate 	}
13057c478bd9Sstevel@tonic-gate 	freemem = t;
13067c478bd9Sstevel@tonic-gate 
13077c478bd9Sstevel@tonic-gate 	/*
13087c478bd9Sstevel@tonic-gate 	 * Don't worry about grabbing mutex.  It's not that
13097c478bd9Sstevel@tonic-gate 	 * critical if we miss a tick or two.  This is
13107c478bd9Sstevel@tonic-gate 	 * where we wakeup possible delayers in
13117c478bd9Sstevel@tonic-gate 	 * page_create_get_something().
13127c478bd9Sstevel@tonic-gate 	 */
13137c478bd9Sstevel@tonic-gate 	wakeup_pcgs();
13147c478bd9Sstevel@tonic-gate }
13157c478bd9Sstevel@tonic-gate 
13167c478bd9Sstevel@tonic-gate ulong_t
13177c478bd9Sstevel@tonic-gate get_freemem()
13187c478bd9Sstevel@tonic-gate {
13197c478bd9Sstevel@tonic-gate 	struct pcf	*p;
13207c478bd9Sstevel@tonic-gate 	ulong_t		t;
13217c478bd9Sstevel@tonic-gate 	uint_t		i;
13227c478bd9Sstevel@tonic-gate 
13237c478bd9Sstevel@tonic-gate 	t = 0;
13247c478bd9Sstevel@tonic-gate 	p = pcf;
13257c478bd9Sstevel@tonic-gate 	for (i = 0; i < PCF_FANOUT; i++) {
13267c478bd9Sstevel@tonic-gate 		t += p->pcf_count;
13277c478bd9Sstevel@tonic-gate 		p++;
13287c478bd9Sstevel@tonic-gate 	}
13297c478bd9Sstevel@tonic-gate 	/*
13307c478bd9Sstevel@tonic-gate 	 * We just calculated it, might as well set it.
13317c478bd9Sstevel@tonic-gate 	 */
13327c478bd9Sstevel@tonic-gate 	freemem = t;
13337c478bd9Sstevel@tonic-gate 	return (t);
13347c478bd9Sstevel@tonic-gate }
13357c478bd9Sstevel@tonic-gate 
13367c478bd9Sstevel@tonic-gate /*
13377c478bd9Sstevel@tonic-gate  * Acquire all of the page cache & free (pcf) locks.
13387c478bd9Sstevel@tonic-gate  */
13397c478bd9Sstevel@tonic-gate void
13407c478bd9Sstevel@tonic-gate pcf_acquire_all()
13417c478bd9Sstevel@tonic-gate {
13427c478bd9Sstevel@tonic-gate 	struct pcf	*p;
13437c478bd9Sstevel@tonic-gate 	uint_t		i;
13447c478bd9Sstevel@tonic-gate 
13457c478bd9Sstevel@tonic-gate 	p = pcf;
13467c478bd9Sstevel@tonic-gate 	for (i = 0; i < PCF_FANOUT; i++) {
13477c478bd9Sstevel@tonic-gate 		mutex_enter(&p->pcf_lock);
13487c478bd9Sstevel@tonic-gate 		p++;
13497c478bd9Sstevel@tonic-gate 	}
13507c478bd9Sstevel@tonic-gate }
13517c478bd9Sstevel@tonic-gate 
13527c478bd9Sstevel@tonic-gate /*
13537c478bd9Sstevel@tonic-gate  * Release all the pcf_locks.
13547c478bd9Sstevel@tonic-gate  */
13557c478bd9Sstevel@tonic-gate void
13567c478bd9Sstevel@tonic-gate pcf_release_all()
13577c478bd9Sstevel@tonic-gate {
13587c478bd9Sstevel@tonic-gate 	struct pcf	*p;
13597c478bd9Sstevel@tonic-gate 	uint_t		i;
13607c478bd9Sstevel@tonic-gate 
13617c478bd9Sstevel@tonic-gate 	p = pcf;
13627c478bd9Sstevel@tonic-gate 	for (i = 0; i < PCF_FANOUT; i++) {
13637c478bd9Sstevel@tonic-gate 		mutex_exit(&p->pcf_lock);
13647c478bd9Sstevel@tonic-gate 		p++;
13657c478bd9Sstevel@tonic-gate 	}
13667c478bd9Sstevel@tonic-gate }
13677c478bd9Sstevel@tonic-gate 
13687c478bd9Sstevel@tonic-gate /*
13697c478bd9Sstevel@tonic-gate  * Inform the VM system that we need some pages freed up.
13707c478bd9Sstevel@tonic-gate  * Calls must be symmetric, e.g.:
13717c478bd9Sstevel@tonic-gate  *
13727c478bd9Sstevel@tonic-gate  *	page_needfree(100);
13737c478bd9Sstevel@tonic-gate  *	wait a bit;
13747c478bd9Sstevel@tonic-gate  *	page_needfree(-100);
13757c478bd9Sstevel@tonic-gate  */
13767c478bd9Sstevel@tonic-gate void
13777c478bd9Sstevel@tonic-gate page_needfree(spgcnt_t npages)
13787c478bd9Sstevel@tonic-gate {
13797c478bd9Sstevel@tonic-gate 	mutex_enter(&new_freemem_lock);
13807c478bd9Sstevel@tonic-gate 	needfree += npages;
13817c478bd9Sstevel@tonic-gate 	mutex_exit(&new_freemem_lock);
13827c478bd9Sstevel@tonic-gate }
13837c478bd9Sstevel@tonic-gate 
13847c478bd9Sstevel@tonic-gate /*
13857c478bd9Sstevel@tonic-gate  * Throttle for page_create(): try to prevent freemem from dropping
13867c478bd9Sstevel@tonic-gate  * below throttlefree.  We can't provide a 100% guarantee because
13877c478bd9Sstevel@tonic-gate  * KM_NOSLEEP allocations, page_reclaim(), and various other things
13887c478bd9Sstevel@tonic-gate  * nibble away at the freelist.  However, we can block all PG_WAIT
13897c478bd9Sstevel@tonic-gate  * allocations until memory becomes available.  The motivation is
13907c478bd9Sstevel@tonic-gate  * that several things can fall apart when there's no free memory:
13917c478bd9Sstevel@tonic-gate  *
13927c478bd9Sstevel@tonic-gate  * (1) If pageout() needs memory to push a page, the system deadlocks.
13937c478bd9Sstevel@tonic-gate  *
13947c478bd9Sstevel@tonic-gate  * (2) By (broken) specification, timeout(9F) can neither fail nor
13957c478bd9Sstevel@tonic-gate  *     block, so it has no choice but to panic the system if it
13967c478bd9Sstevel@tonic-gate  *     cannot allocate a callout structure.
13977c478bd9Sstevel@tonic-gate  *
13987c478bd9Sstevel@tonic-gate  * (3) Like timeout(), ddi_set_callback() cannot fail and cannot block;
13997c478bd9Sstevel@tonic-gate  *     it panics if it cannot allocate a callback structure.
14007c478bd9Sstevel@tonic-gate  *
14017c478bd9Sstevel@tonic-gate  * (4) Untold numbers of third-party drivers have not yet been hardened
14027c478bd9Sstevel@tonic-gate  *     against KM_NOSLEEP and/or allocb() failures; they simply assume
14037c478bd9Sstevel@tonic-gate  *     success and panic the system with a data fault on failure.
14047c478bd9Sstevel@tonic-gate  *     (The long-term solution to this particular problem is to ship
14057c478bd9Sstevel@tonic-gate  *     hostile fault-injecting DEBUG kernels with the DDK.)
14067c478bd9Sstevel@tonic-gate  *
14077c478bd9Sstevel@tonic-gate  * It is theoretically impossible to guarantee success of non-blocking
14087c478bd9Sstevel@tonic-gate  * allocations, but in practice, this throttle is very hard to break.
14097c478bd9Sstevel@tonic-gate  */
14107c478bd9Sstevel@tonic-gate static int
14117c478bd9Sstevel@tonic-gate page_create_throttle(pgcnt_t npages, int flags)
14127c478bd9Sstevel@tonic-gate {
14137c478bd9Sstevel@tonic-gate 	ulong_t	fm;
14147c478bd9Sstevel@tonic-gate 	uint_t	i;
14157c478bd9Sstevel@tonic-gate 	pgcnt_t tf;	/* effective value of throttlefree */
14167c478bd9Sstevel@tonic-gate 
14177c478bd9Sstevel@tonic-gate 	/*
14187c478bd9Sstevel@tonic-gate 	 * Never deny pages when:
14197c478bd9Sstevel@tonic-gate 	 * - it's a thread that cannot block [NOMEMWAIT()]
14207c478bd9Sstevel@tonic-gate 	 * - the allocation cannot block and must not fail
14217c478bd9Sstevel@tonic-gate 	 * - the allocation cannot block and is pageout dispensated
14227c478bd9Sstevel@tonic-gate 	 */
14237c478bd9Sstevel@tonic-gate 	if (NOMEMWAIT() ||
14247c478bd9Sstevel@tonic-gate 	    ((flags & (PG_WAIT | PG_PANIC)) == PG_PANIC) ||
14257c478bd9Sstevel@tonic-gate 	    ((flags & (PG_WAIT | PG_PUSHPAGE)) == PG_PUSHPAGE))
14267c478bd9Sstevel@tonic-gate 		return (1);
14277c478bd9Sstevel@tonic-gate 
14287c478bd9Sstevel@tonic-gate 	/*
14297c478bd9Sstevel@tonic-gate 	 * If the allocation can't block, we look favorably upon it
14307c478bd9Sstevel@tonic-gate 	 * unless we're below pageout_reserve.  In that case we fail
14317c478bd9Sstevel@tonic-gate 	 * the allocation because we want to make sure there are a few
14327c478bd9Sstevel@tonic-gate 	 * pages available for pageout.
14337c478bd9Sstevel@tonic-gate 	 */
14347c478bd9Sstevel@tonic-gate 	if ((flags & PG_WAIT) == 0)
14357c478bd9Sstevel@tonic-gate 		return (freemem >= npages + pageout_reserve);
14367c478bd9Sstevel@tonic-gate 
14377c478bd9Sstevel@tonic-gate 	/* Calculate the effective throttlefree value */
14387c478bd9Sstevel@tonic-gate 	tf = throttlefree -
14397c478bd9Sstevel@tonic-gate 	    ((flags & PG_PUSHPAGE) ? pageout_reserve : 0);
14407c478bd9Sstevel@tonic-gate 
14417c478bd9Sstevel@tonic-gate 	cv_signal(&proc_pageout->p_cv);
14427c478bd9Sstevel@tonic-gate 
14437c478bd9Sstevel@tonic-gate 	while (freemem < npages + tf) {
14447c478bd9Sstevel@tonic-gate 		pcf_acquire_all();
14457c478bd9Sstevel@tonic-gate 		mutex_enter(&new_freemem_lock);
14467c478bd9Sstevel@tonic-gate 		fm = 0;
14477c478bd9Sstevel@tonic-gate 		for (i = 0; i < PCF_FANOUT; i++) {
14487c478bd9Sstevel@tonic-gate 			fm += pcf[i].pcf_count;
14497c478bd9Sstevel@tonic-gate 			pcf[i].pcf_wait++;
14507c478bd9Sstevel@tonic-gate 			mutex_exit(&pcf[i].pcf_lock);
14517c478bd9Sstevel@tonic-gate 		}
14527c478bd9Sstevel@tonic-gate 		freemem = fm;
14537c478bd9Sstevel@tonic-gate 		needfree += npages;
14547c478bd9Sstevel@tonic-gate 		freemem_wait++;
14557c478bd9Sstevel@tonic-gate 		cv_wait(&freemem_cv, &new_freemem_lock);
14567c478bd9Sstevel@tonic-gate 		freemem_wait--;
14577c478bd9Sstevel@tonic-gate 		needfree -= npages;
14587c478bd9Sstevel@tonic-gate 		mutex_exit(&new_freemem_lock);
14597c478bd9Sstevel@tonic-gate 	}
14607c478bd9Sstevel@tonic-gate 	return (1);
14617c478bd9Sstevel@tonic-gate }
14627c478bd9Sstevel@tonic-gate 
14637c478bd9Sstevel@tonic-gate /*
14647c478bd9Sstevel@tonic-gate  * page_create_wait() is called to either coalecse pages from the
14657c478bd9Sstevel@tonic-gate  * different pcf buckets or to wait because there simply are not
14667c478bd9Sstevel@tonic-gate  * enough pages to satisfy the caller's request.
14677c478bd9Sstevel@tonic-gate  *
14687c478bd9Sstevel@tonic-gate  * Sadly, this is called from platform/vm/vm_machdep.c
14697c478bd9Sstevel@tonic-gate  */
14707c478bd9Sstevel@tonic-gate int
14717c478bd9Sstevel@tonic-gate page_create_wait(size_t npages, uint_t flags)
14727c478bd9Sstevel@tonic-gate {
14737c478bd9Sstevel@tonic-gate 	pgcnt_t		total;
14747c478bd9Sstevel@tonic-gate 	uint_t		i;
14757c478bd9Sstevel@tonic-gate 	struct pcf	*p;
14767c478bd9Sstevel@tonic-gate 
14777c478bd9Sstevel@tonic-gate 	/*
14787c478bd9Sstevel@tonic-gate 	 * Wait until there are enough free pages to satisfy our
14797c478bd9Sstevel@tonic-gate 	 * entire request.
14807c478bd9Sstevel@tonic-gate 	 * We set needfree += npages before prodding pageout, to make sure
14817c478bd9Sstevel@tonic-gate 	 * it does real work when npages > lotsfree > freemem.
14827c478bd9Sstevel@tonic-gate 	 */
14837c478bd9Sstevel@tonic-gate 	VM_STAT_ADD(page_create_not_enough);
14847c478bd9Sstevel@tonic-gate 
14857c478bd9Sstevel@tonic-gate 	ASSERT(!kcage_on ? !(flags & PG_NORELOC) : 1);
14867c478bd9Sstevel@tonic-gate checkagain:
14877c478bd9Sstevel@tonic-gate 	if ((flags & PG_NORELOC) &&
14887c478bd9Sstevel@tonic-gate 	    kcage_freemem < kcage_throttlefree + npages)
14897c478bd9Sstevel@tonic-gate 		(void) kcage_create_throttle(npages, flags);
14907c478bd9Sstevel@tonic-gate 
14917c478bd9Sstevel@tonic-gate 	if (freemem < npages + throttlefree)
14927c478bd9Sstevel@tonic-gate 		if (!page_create_throttle(npages, flags))
14937c478bd9Sstevel@tonic-gate 			return (0);
14947c478bd9Sstevel@tonic-gate 
14957c478bd9Sstevel@tonic-gate 	/*
14967c478bd9Sstevel@tonic-gate 	 * Since page_create_va() looked at every
14977c478bd9Sstevel@tonic-gate 	 * bucket, assume we are going to have to wait.
14987c478bd9Sstevel@tonic-gate 	 * Get all of the pcf locks.
14997c478bd9Sstevel@tonic-gate 	 */
15007c478bd9Sstevel@tonic-gate 	total = 0;
15017c478bd9Sstevel@tonic-gate 	p = pcf;
15027c478bd9Sstevel@tonic-gate 	for (i = 0; i < PCF_FANOUT; i++) {
15037c478bd9Sstevel@tonic-gate 		mutex_enter(&p->pcf_lock);
15047c478bd9Sstevel@tonic-gate 		total += p->pcf_count;
15057c478bd9Sstevel@tonic-gate 		if (total >= npages) {
15067c478bd9Sstevel@tonic-gate 			/*
15077c478bd9Sstevel@tonic-gate 			 * Wow!  There are enough pages laying around
15087c478bd9Sstevel@tonic-gate 			 * to satisfy the request.  Do the accounting,
15097c478bd9Sstevel@tonic-gate 			 * drop the locks we acquired, and go back.
15107c478bd9Sstevel@tonic-gate 			 *
15117c478bd9Sstevel@tonic-gate 			 * freemem is not protected by any lock. So,
15127c478bd9Sstevel@tonic-gate 			 * we cannot have any assertion containing
15137c478bd9Sstevel@tonic-gate 			 * freemem.
15147c478bd9Sstevel@tonic-gate 			 */
15157c478bd9Sstevel@tonic-gate 			freemem -= npages;
15167c478bd9Sstevel@tonic-gate 
15177c478bd9Sstevel@tonic-gate 			while (p >= pcf) {
15187c478bd9Sstevel@tonic-gate 				if (p->pcf_count <= npages) {
15197c478bd9Sstevel@tonic-gate 					npages -= p->pcf_count;
15207c478bd9Sstevel@tonic-gate 					p->pcf_count = 0;
15217c478bd9Sstevel@tonic-gate 				} else {
15227c478bd9Sstevel@tonic-gate 					p->pcf_count -= (uint_t)npages;
15237c478bd9Sstevel@tonic-gate 					npages = 0;
15247c478bd9Sstevel@tonic-gate 				}
15257c478bd9Sstevel@tonic-gate 				mutex_exit(&p->pcf_lock);
15267c478bd9Sstevel@tonic-gate 				p--;
15277c478bd9Sstevel@tonic-gate 			}
15287c478bd9Sstevel@tonic-gate 			ASSERT(npages == 0);
15297c478bd9Sstevel@tonic-gate 			return (1);
15307c478bd9Sstevel@tonic-gate 		}
15317c478bd9Sstevel@tonic-gate 		p++;
15327c478bd9Sstevel@tonic-gate 	}
15337c478bd9Sstevel@tonic-gate 
15347c478bd9Sstevel@tonic-gate 	/*
15357c478bd9Sstevel@tonic-gate 	 * All of the pcf locks are held, there are not enough pages
15367c478bd9Sstevel@tonic-gate 	 * to satisfy the request (npages < total).
15377c478bd9Sstevel@tonic-gate 	 * Be sure to acquire the new_freemem_lock before dropping
15387c478bd9Sstevel@tonic-gate 	 * the pcf locks.  This prevents dropping wakeups in page_free().
15397c478bd9Sstevel@tonic-gate 	 * The order is always pcf_lock then new_freemem_lock.
15407c478bd9Sstevel@tonic-gate 	 *
15417c478bd9Sstevel@tonic-gate 	 * Since we hold all the pcf locks, it is a good time to set freemem.
15427c478bd9Sstevel@tonic-gate 	 *
15437c478bd9Sstevel@tonic-gate 	 * If the caller does not want to wait, return now.
15447c478bd9Sstevel@tonic-gate 	 * Else turn the pageout daemon loose to find something
15457c478bd9Sstevel@tonic-gate 	 * and wait till it does.
15467c478bd9Sstevel@tonic-gate 	 *
15477c478bd9Sstevel@tonic-gate 	 */
15487c478bd9Sstevel@tonic-gate 	freemem = total;
15497c478bd9Sstevel@tonic-gate 
15507c478bd9Sstevel@tonic-gate 	if ((flags & PG_WAIT) == 0) {
15517c478bd9Sstevel@tonic-gate 		pcf_release_all();
15527c478bd9Sstevel@tonic-gate 
15537c478bd9Sstevel@tonic-gate 		TRACE_2(TR_FAC_VM, TR_PAGE_CREATE_NOMEM,
15547c478bd9Sstevel@tonic-gate 		"page_create_nomem:npages %ld freemem %ld", npages, freemem);
15557c478bd9Sstevel@tonic-gate 		return (0);
15567c478bd9Sstevel@tonic-gate 	}
15577c478bd9Sstevel@tonic-gate 
15587c478bd9Sstevel@tonic-gate 	ASSERT(proc_pageout != NULL);
15597c478bd9Sstevel@tonic-gate 	cv_signal(&proc_pageout->p_cv);
15607c478bd9Sstevel@tonic-gate 
15617c478bd9Sstevel@tonic-gate 	TRACE_2(TR_FAC_VM, TR_PAGE_CREATE_SLEEP_START,
15627c478bd9Sstevel@tonic-gate 	    "page_create_sleep_start: freemem %ld needfree %ld",
15637c478bd9Sstevel@tonic-gate 	    freemem, needfree);
15647c478bd9Sstevel@tonic-gate 
15657c478bd9Sstevel@tonic-gate 	/*
15667c478bd9Sstevel@tonic-gate 	 * We are going to wait.
15677c478bd9Sstevel@tonic-gate 	 * We currently hold all of the pcf_locks,
15687c478bd9Sstevel@tonic-gate 	 * get the new_freemem_lock (it protects freemem_wait),
15697c478bd9Sstevel@tonic-gate 	 * before dropping the pcf_locks.
15707c478bd9Sstevel@tonic-gate 	 */
15717c478bd9Sstevel@tonic-gate 	mutex_enter(&new_freemem_lock);
15727c478bd9Sstevel@tonic-gate 
15737c478bd9Sstevel@tonic-gate 	p = pcf;
15747c478bd9Sstevel@tonic-gate 	for (i = 0; i < PCF_FANOUT; i++) {
15757c478bd9Sstevel@tonic-gate 		p->pcf_wait++;
15767c478bd9Sstevel@tonic-gate 		mutex_exit(&p->pcf_lock);
15777c478bd9Sstevel@tonic-gate 		p++;
15787c478bd9Sstevel@tonic-gate 	}
15797c478bd9Sstevel@tonic-gate 
15807c478bd9Sstevel@tonic-gate 	needfree += npages;
15817c478bd9Sstevel@tonic-gate 	freemem_wait++;
15827c478bd9Sstevel@tonic-gate 
15837c478bd9Sstevel@tonic-gate 	cv_wait(&freemem_cv, &new_freemem_lock);
15847c478bd9Sstevel@tonic-gate 
15857c478bd9Sstevel@tonic-gate 	freemem_wait--;
15867c478bd9Sstevel@tonic-gate 	needfree -= npages;
15877c478bd9Sstevel@tonic-gate 
15887c478bd9Sstevel@tonic-gate 	mutex_exit(&new_freemem_lock);
15897c478bd9Sstevel@tonic-gate 
15907c478bd9Sstevel@tonic-gate 	TRACE_2(TR_FAC_VM, TR_PAGE_CREATE_SLEEP_END,
15917c478bd9Sstevel@tonic-gate 	    "page_create_sleep_end: freemem %ld needfree %ld",
15927c478bd9Sstevel@tonic-gate 	    freemem, needfree);
15937c478bd9Sstevel@tonic-gate 
15947c478bd9Sstevel@tonic-gate 	VM_STAT_ADD(page_create_not_enough_again);
15957c478bd9Sstevel@tonic-gate 	goto checkagain;
15967c478bd9Sstevel@tonic-gate }
15977c478bd9Sstevel@tonic-gate 
15987c478bd9Sstevel@tonic-gate /*
15997c478bd9Sstevel@tonic-gate  * A routine to do the opposite of page_create_wait().
16007c478bd9Sstevel@tonic-gate  */
16017c478bd9Sstevel@tonic-gate void
16027c478bd9Sstevel@tonic-gate page_create_putback(spgcnt_t npages)
16037c478bd9Sstevel@tonic-gate {
16047c478bd9Sstevel@tonic-gate 	struct pcf	*p;
16057c478bd9Sstevel@tonic-gate 	pgcnt_t		lump;
16067c478bd9Sstevel@tonic-gate 	uint_t		*which;
16077c478bd9Sstevel@tonic-gate 
16087c478bd9Sstevel@tonic-gate 	/*
16097c478bd9Sstevel@tonic-gate 	 * When a contiguous lump is broken up, we have to
16107c478bd9Sstevel@tonic-gate 	 * deal with lots of pages (min 64) so lets spread
16117c478bd9Sstevel@tonic-gate 	 * the wealth around.
16127c478bd9Sstevel@tonic-gate 	 */
16137c478bd9Sstevel@tonic-gate 	lump = roundup(npages, PCF_FANOUT) / PCF_FANOUT;
16147c478bd9Sstevel@tonic-gate 	freemem += npages;
16157c478bd9Sstevel@tonic-gate 
16167c478bd9Sstevel@tonic-gate 	for (p = pcf; (npages > 0) && (p < &pcf[PCF_FANOUT]); p++) {
16177c478bd9Sstevel@tonic-gate 		which = &p->pcf_count;
16187c478bd9Sstevel@tonic-gate 
16197c478bd9Sstevel@tonic-gate 		mutex_enter(&p->pcf_lock);
16207c478bd9Sstevel@tonic-gate 
16217c478bd9Sstevel@tonic-gate 		if (p->pcf_block) {
16227c478bd9Sstevel@tonic-gate 			which = &p->pcf_reserve;
16237c478bd9Sstevel@tonic-gate 		}
16247c478bd9Sstevel@tonic-gate 
16257c478bd9Sstevel@tonic-gate 		if (lump < npages) {
16267c478bd9Sstevel@tonic-gate 			*which += (uint_t)lump;
16277c478bd9Sstevel@tonic-gate 			npages -= lump;
16287c478bd9Sstevel@tonic-gate 		} else {
16297c478bd9Sstevel@tonic-gate 			*which += (uint_t)npages;
16307c478bd9Sstevel@tonic-gate 			npages = 0;
16317c478bd9Sstevel@tonic-gate 		}
16327c478bd9Sstevel@tonic-gate 
16337c478bd9Sstevel@tonic-gate 		if (p->pcf_wait) {
16347c478bd9Sstevel@tonic-gate 			mutex_enter(&new_freemem_lock);
16357c478bd9Sstevel@tonic-gate 			/*
16367c478bd9Sstevel@tonic-gate 			 * Check to see if some other thread
16377c478bd9Sstevel@tonic-gate 			 * is actually waiting.  Another bucket
16387c478bd9Sstevel@tonic-gate 			 * may have woken it up by now.  If there
16397c478bd9Sstevel@tonic-gate 			 * are no waiters, then set our pcf_wait
16407c478bd9Sstevel@tonic-gate 			 * count to zero to avoid coming in here
16417c478bd9Sstevel@tonic-gate 			 * next time.
16427c478bd9Sstevel@tonic-gate 			 */
16437c478bd9Sstevel@tonic-gate 			if (freemem_wait) {
16447c478bd9Sstevel@tonic-gate 				if (npages > 1) {
16457c478bd9Sstevel@tonic-gate 					cv_broadcast(&freemem_cv);
16467c478bd9Sstevel@tonic-gate 				} else {
16477c478bd9Sstevel@tonic-gate 					cv_signal(&freemem_cv);
16487c478bd9Sstevel@tonic-gate 				}
16497c478bd9Sstevel@tonic-gate 				p->pcf_wait--;
16507c478bd9Sstevel@tonic-gate 			} else {
16517c478bd9Sstevel@tonic-gate 				p->pcf_wait = 0;
16527c478bd9Sstevel@tonic-gate 			}
16537c478bd9Sstevel@tonic-gate 			mutex_exit(&new_freemem_lock);
16547c478bd9Sstevel@tonic-gate 		}
16557c478bd9Sstevel@tonic-gate 		mutex_exit(&p->pcf_lock);
16567c478bd9Sstevel@tonic-gate 	}
16577c478bd9Sstevel@tonic-gate 	ASSERT(npages == 0);
16587c478bd9Sstevel@tonic-gate }
16597c478bd9Sstevel@tonic-gate 
16607c478bd9Sstevel@tonic-gate /*
16617c478bd9Sstevel@tonic-gate  * A helper routine for page_create_get_something.
16627c478bd9Sstevel@tonic-gate  * The indenting got to deep down there.
16637c478bd9Sstevel@tonic-gate  * Unblock the pcf counters.  Any pages freed after
16647c478bd9Sstevel@tonic-gate  * pcf_block got set are moved to pcf_count and
16657c478bd9Sstevel@tonic-gate  * wakeups (cv_broadcast() or cv_signal()) are done as needed.
16667c478bd9Sstevel@tonic-gate  */
16677c478bd9Sstevel@tonic-gate static void
16687c478bd9Sstevel@tonic-gate pcgs_unblock(void)
16697c478bd9Sstevel@tonic-gate {
16707c478bd9Sstevel@tonic-gate 	int		i;
16717c478bd9Sstevel@tonic-gate 	struct pcf	*p;
16727c478bd9Sstevel@tonic-gate 
16737c478bd9Sstevel@tonic-gate 	/* Update freemem while we're here. */
16747c478bd9Sstevel@tonic-gate 	freemem = 0;
16757c478bd9Sstevel@tonic-gate 	p = pcf;
16767c478bd9Sstevel@tonic-gate 	for (i = 0; i < PCF_FANOUT; i++) {
16777c478bd9Sstevel@tonic-gate 		mutex_enter(&p->pcf_lock);
16787c478bd9Sstevel@tonic-gate 		ASSERT(p->pcf_count == 0);
16797c478bd9Sstevel@tonic-gate 		p->pcf_count = p->pcf_reserve;
16807c478bd9Sstevel@tonic-gate 		p->pcf_block = 0;
16817c478bd9Sstevel@tonic-gate 		freemem += p->pcf_count;
16827c478bd9Sstevel@tonic-gate 		if (p->pcf_wait) {
16837c478bd9Sstevel@tonic-gate 			mutex_enter(&new_freemem_lock);
16847c478bd9Sstevel@tonic-gate 			if (freemem_wait) {
16857c478bd9Sstevel@tonic-gate 				if (p->pcf_reserve > 1) {
16867c478bd9Sstevel@tonic-gate 					cv_broadcast(&freemem_cv);
16877c478bd9Sstevel@tonic-gate 					p->pcf_wait = 0;
16887c478bd9Sstevel@tonic-gate 				} else {
16897c478bd9Sstevel@tonic-gate 					cv_signal(&freemem_cv);
16907c478bd9Sstevel@tonic-gate 					p->pcf_wait--;
16917c478bd9Sstevel@tonic-gate 				}
16927c478bd9Sstevel@tonic-gate 			} else {
16937c478bd9Sstevel@tonic-gate 				p->pcf_wait = 0;
16947c478bd9Sstevel@tonic-gate 			}
16957c478bd9Sstevel@tonic-gate 			mutex_exit(&new_freemem_lock);
16967c478bd9Sstevel@tonic-gate 		}
16977c478bd9Sstevel@tonic-gate 		p->pcf_reserve = 0;
16987c478bd9Sstevel@tonic-gate 		mutex_exit(&p->pcf_lock);
16997c478bd9Sstevel@tonic-gate 		p++;
17007c478bd9Sstevel@tonic-gate 	}
17017c478bd9Sstevel@tonic-gate }
17027c478bd9Sstevel@tonic-gate 
17037c478bd9Sstevel@tonic-gate /*
17047c478bd9Sstevel@tonic-gate  * Called from page_create_va() when both the cache and free lists
17057c478bd9Sstevel@tonic-gate  * have been checked once.
17067c478bd9Sstevel@tonic-gate  *
17077c478bd9Sstevel@tonic-gate  * Either returns a page or panics since the accounting was done
17087c478bd9Sstevel@tonic-gate  * way before we got here.
17097c478bd9Sstevel@tonic-gate  *
17107c478bd9Sstevel@tonic-gate  * We don't come here often, so leave the accounting on permanently.
17117c478bd9Sstevel@tonic-gate  */
17127c478bd9Sstevel@tonic-gate 
17137c478bd9Sstevel@tonic-gate #define	MAX_PCGS	100
17147c478bd9Sstevel@tonic-gate 
17157c478bd9Sstevel@tonic-gate #ifdef	DEBUG
17167c478bd9Sstevel@tonic-gate #define	PCGS_TRIES	100
17177c478bd9Sstevel@tonic-gate #else	/* DEBUG */
17187c478bd9Sstevel@tonic-gate #define	PCGS_TRIES	10
17197c478bd9Sstevel@tonic-gate #endif	/* DEBUG */
17207c478bd9Sstevel@tonic-gate 
17217c478bd9Sstevel@tonic-gate #ifdef	VM_STATS
17227c478bd9Sstevel@tonic-gate uint_t	pcgs_counts[PCGS_TRIES];
17237c478bd9Sstevel@tonic-gate uint_t	pcgs_too_many;
17247c478bd9Sstevel@tonic-gate uint_t	pcgs_entered;
17257c478bd9Sstevel@tonic-gate uint_t	pcgs_entered_noreloc;
17267c478bd9Sstevel@tonic-gate uint_t	pcgs_locked;
17277c478bd9Sstevel@tonic-gate uint_t	pcgs_cagelocked;
17287c478bd9Sstevel@tonic-gate #endif	/* VM_STATS */
17297c478bd9Sstevel@tonic-gate 
17307c478bd9Sstevel@tonic-gate static page_t *
17317c478bd9Sstevel@tonic-gate page_create_get_something(vnode_t *vp, u_offset_t off, struct seg *seg,
17327c478bd9Sstevel@tonic-gate     caddr_t vaddr, uint_t flags)
17337c478bd9Sstevel@tonic-gate {
17347c478bd9Sstevel@tonic-gate 	uint_t		count;
17357c478bd9Sstevel@tonic-gate 	page_t		*pp;
17367c478bd9Sstevel@tonic-gate 	uint_t		locked, i;
17377c478bd9Sstevel@tonic-gate 	struct	pcf	*p;
17387c478bd9Sstevel@tonic-gate 	lgrp_t		*lgrp;
17397c478bd9Sstevel@tonic-gate 	int		cagelocked = 0;
17407c478bd9Sstevel@tonic-gate 
17417c478bd9Sstevel@tonic-gate 	VM_STAT_ADD(pcgs_entered);
17427c478bd9Sstevel@tonic-gate 
17437c478bd9Sstevel@tonic-gate 	/*
17447c478bd9Sstevel@tonic-gate 	 * Tap any reserve freelists: if we fail now, we'll die
17457c478bd9Sstevel@tonic-gate 	 * since the page(s) we're looking for have already been
17467c478bd9Sstevel@tonic-gate 	 * accounted for.
17477c478bd9Sstevel@tonic-gate 	 */
17487c478bd9Sstevel@tonic-gate 	flags |= PG_PANIC;
17497c478bd9Sstevel@tonic-gate 
17507c478bd9Sstevel@tonic-gate 	if ((flags & PG_NORELOC) != 0) {
17517c478bd9Sstevel@tonic-gate 		VM_STAT_ADD(pcgs_entered_noreloc);
17527c478bd9Sstevel@tonic-gate 		/*
17537c478bd9Sstevel@tonic-gate 		 * Requests for free pages from critical threads
17547c478bd9Sstevel@tonic-gate 		 * such as pageout still won't throttle here, but
17557c478bd9Sstevel@tonic-gate 		 * we must try again, to give the cageout thread
17567c478bd9Sstevel@tonic-gate 		 * another chance to catch up. Since we already
17577c478bd9Sstevel@tonic-gate 		 * accounted for the pages, we had better get them
17587c478bd9Sstevel@tonic-gate 		 * this time.
17597c478bd9Sstevel@tonic-gate 		 *
17607c478bd9Sstevel@tonic-gate 		 * N.B. All non-critical threads acquire the pcgs_cagelock
17617c478bd9Sstevel@tonic-gate 		 * to serialize access to the freelists. This implements a
17627c478bd9Sstevel@tonic-gate 		 * turnstile-type synchornization to avoid starvation of
17637c478bd9Sstevel@tonic-gate 		 * critical requests for PG_NORELOC memory by non-critical
17647c478bd9Sstevel@tonic-gate 		 * threads: all non-critical threads must acquire a 'ticket'
17657c478bd9Sstevel@tonic-gate 		 * before passing through, which entails making sure
17667c478bd9Sstevel@tonic-gate 		 * kcage_freemem won't fall below minfree prior to grabbing
17677c478bd9Sstevel@tonic-gate 		 * pages from the freelists.
17687c478bd9Sstevel@tonic-gate 		 */
17697c478bd9Sstevel@tonic-gate 		if (kcage_create_throttle(1, flags) == KCT_NONCRIT) {
17707c478bd9Sstevel@tonic-gate 			mutex_enter(&pcgs_cagelock);
17717c478bd9Sstevel@tonic-gate 			cagelocked = 1;
17727c478bd9Sstevel@tonic-gate 			VM_STAT_ADD(pcgs_cagelocked);
17737c478bd9Sstevel@tonic-gate 		}
17747c478bd9Sstevel@tonic-gate 	}
17757c478bd9Sstevel@tonic-gate 
17767c478bd9Sstevel@tonic-gate 	/*
17777c478bd9Sstevel@tonic-gate 	 * Time to get serious.
17787c478bd9Sstevel@tonic-gate 	 * We failed to get a `correctly colored' page from both the
17797c478bd9Sstevel@tonic-gate 	 * free and cache lists.
17807c478bd9Sstevel@tonic-gate 	 * We escalate in stage.
17817c478bd9Sstevel@tonic-gate 	 *
17827c478bd9Sstevel@tonic-gate 	 * First try both lists without worring about color.
17837c478bd9Sstevel@tonic-gate 	 *
17847c478bd9Sstevel@tonic-gate 	 * Then, grab all page accounting locks (ie. pcf[]) and
17857c478bd9Sstevel@tonic-gate 	 * steal any pages that they have and set the pcf_block flag to
17867c478bd9Sstevel@tonic-gate 	 * stop deletions from the lists.  This will help because
17877c478bd9Sstevel@tonic-gate 	 * a page can get added to the free list while we are looking
17887c478bd9Sstevel@tonic-gate 	 * at the cache list, then another page could be added to the cache
17897c478bd9Sstevel@tonic-gate 	 * list allowing the page on the free list to be removed as we
17907c478bd9Sstevel@tonic-gate 	 * move from looking at the cache list to the free list. This
17917c478bd9Sstevel@tonic-gate 	 * could happen over and over. We would never find the page
17927c478bd9Sstevel@tonic-gate 	 * we have accounted for.
17937c478bd9Sstevel@tonic-gate 	 *
17947c478bd9Sstevel@tonic-gate 	 * Noreloc pages are a subset of the global (relocatable) page pool.
17957c478bd9Sstevel@tonic-gate 	 * They are not tracked separately in the pcf bins, so it is
17967c478bd9Sstevel@tonic-gate 	 * impossible to know when doing pcf accounting if the available
17977c478bd9Sstevel@tonic-gate 	 * page(s) are noreloc pages or not. When looking for a noreloc page
17987c478bd9Sstevel@tonic-gate 	 * it is quite easy to end up here even if the global (relocatable)
17997c478bd9Sstevel@tonic-gate 	 * page pool has plenty of free pages but the noreloc pool is empty.
18007c478bd9Sstevel@tonic-gate 	 *
18017c478bd9Sstevel@tonic-gate 	 * When the noreloc pool is empty (or low), additional noreloc pages
18027c478bd9Sstevel@tonic-gate 	 * are created by converting pages from the global page pool. This
18037c478bd9Sstevel@tonic-gate 	 * process will stall during pcf accounting if the pcf bins are
18047c478bd9Sstevel@tonic-gate 	 * already locked. Such is the case when a noreloc allocation is
18057c478bd9Sstevel@tonic-gate 	 * looping here in page_create_get_something waiting for more noreloc
18067c478bd9Sstevel@tonic-gate 	 * pages to appear.
18077c478bd9Sstevel@tonic-gate 	 *
18087c478bd9Sstevel@tonic-gate 	 * Short of adding a new field to the pcf bins to accurately track
18097c478bd9Sstevel@tonic-gate 	 * the number of free noreloc pages, we instead do not grab the
18107c478bd9Sstevel@tonic-gate 	 * pcgs_lock, do not set the pcf blocks and do not timeout when
18117c478bd9Sstevel@tonic-gate 	 * allocating a noreloc page. This allows noreloc allocations to
18127c478bd9Sstevel@tonic-gate 	 * loop without blocking global page pool allocations.
18137c478bd9Sstevel@tonic-gate 	 *
18147c478bd9Sstevel@tonic-gate 	 * NOTE: the behaviour of page_create_get_something has not changed
18157c478bd9Sstevel@tonic-gate 	 * for the case of global page pool allocations.
18167c478bd9Sstevel@tonic-gate 	 */
18177c478bd9Sstevel@tonic-gate 
18187c478bd9Sstevel@tonic-gate 	flags &= ~PG_MATCH_COLOR;
18197c478bd9Sstevel@tonic-gate 	locked = 0;
182007ad560dSkchow #if defined(__i386) || defined(__amd64)
18217c478bd9Sstevel@tonic-gate 	/*
18227c478bd9Sstevel@tonic-gate 	 * page_create_get_something may be called because 4g memory may be
18237c478bd9Sstevel@tonic-gate 	 * depleted. Set flags to allow for relocation of base page below
18247c478bd9Sstevel@tonic-gate 	 * 4g if necessary.
18257c478bd9Sstevel@tonic-gate 	 */
18267c478bd9Sstevel@tonic-gate 	if (physmax4g)
18277c478bd9Sstevel@tonic-gate 		flags |= (PGI_PGCPSZC0 | PGI_PGCPHIPRI);
18287c478bd9Sstevel@tonic-gate #endif
18297c478bd9Sstevel@tonic-gate 
18307c478bd9Sstevel@tonic-gate 	lgrp = lgrp_mem_choose(seg, vaddr, PAGESIZE);
18317c478bd9Sstevel@tonic-gate 
18327c478bd9Sstevel@tonic-gate 	for (count = 0; kcage_on || count < MAX_PCGS; count++) {
18337c478bd9Sstevel@tonic-gate 		pp = page_get_freelist(vp, off, seg, vaddr, PAGESIZE,
18347c478bd9Sstevel@tonic-gate 		    flags, lgrp);
18357c478bd9Sstevel@tonic-gate 		if (pp == NULL) {
18367c478bd9Sstevel@tonic-gate 			pp = page_get_cachelist(vp, off, seg, vaddr,
18377c478bd9Sstevel@tonic-gate 				flags, lgrp);
18387c478bd9Sstevel@tonic-gate 		}
18397c478bd9Sstevel@tonic-gate 		if (pp == NULL) {
18407c478bd9Sstevel@tonic-gate 			/*
18417c478bd9Sstevel@tonic-gate 			 * Serialize.  Don't fight with other pcgs().
18427c478bd9Sstevel@tonic-gate 			 */
18437c478bd9Sstevel@tonic-gate 			if (!locked && (!kcage_on || !(flags & PG_NORELOC))) {
18447c478bd9Sstevel@tonic-gate 				mutex_enter(&pcgs_lock);
18457c478bd9Sstevel@tonic-gate 				VM_STAT_ADD(pcgs_locked);
18467c478bd9Sstevel@tonic-gate 				locked = 1;
18477c478bd9Sstevel@tonic-gate 				p = pcf;
18487c478bd9Sstevel@tonic-gate 				for (i = 0; i < PCF_FANOUT; i++) {
18497c478bd9Sstevel@tonic-gate 					mutex_enter(&p->pcf_lock);
18507c478bd9Sstevel@tonic-gate 					ASSERT(p->pcf_block == 0);
18517c478bd9Sstevel@tonic-gate 					p->pcf_block = 1;
18527c478bd9Sstevel@tonic-gate 					p->pcf_reserve = p->pcf_count;
18537c478bd9Sstevel@tonic-gate 					p->pcf_count = 0;
18547c478bd9Sstevel@tonic-gate 					mutex_exit(&p->pcf_lock);
18557c478bd9Sstevel@tonic-gate 					p++;
18567c478bd9Sstevel@tonic-gate 				}
18577c478bd9Sstevel@tonic-gate 				freemem = 0;
18587c478bd9Sstevel@tonic-gate 			}
18597c478bd9Sstevel@tonic-gate 
18607c478bd9Sstevel@tonic-gate 			if (count) {
18617c478bd9Sstevel@tonic-gate 				/*
18627c478bd9Sstevel@tonic-gate 				 * Since page_free() puts pages on
18637c478bd9Sstevel@tonic-gate 				 * a list then accounts for it, we
18647c478bd9Sstevel@tonic-gate 				 * just have to wait for page_free()
18657c478bd9Sstevel@tonic-gate 				 * to unlock any page it was working
18667c478bd9Sstevel@tonic-gate 				 * with. The page_lock()-page_reclaim()
18677c478bd9Sstevel@tonic-gate 				 * path falls in the same boat.
18687c478bd9Sstevel@tonic-gate 				 *
18697c478bd9Sstevel@tonic-gate 				 * We don't need to check on the
18707c478bd9Sstevel@tonic-gate 				 * PG_WAIT flag, we have already
18717c478bd9Sstevel@tonic-gate 				 * accounted for the page we are
18727c478bd9Sstevel@tonic-gate 				 * looking for in page_create_va().
18737c478bd9Sstevel@tonic-gate 				 *
18747c478bd9Sstevel@tonic-gate 				 * We just wait a moment to let any
18757c478bd9Sstevel@tonic-gate 				 * locked pages on the lists free up,
18767c478bd9Sstevel@tonic-gate 				 * then continue around and try again.
18777c478bd9Sstevel@tonic-gate 				 *
18787c478bd9Sstevel@tonic-gate 				 * Will be awakened by set_freemem().
18797c478bd9Sstevel@tonic-gate 				 */
18807c478bd9Sstevel@tonic-gate 				mutex_enter(&pcgs_wait_lock);
18817c478bd9Sstevel@tonic-gate 				cv_wait(&pcgs_cv, &pcgs_wait_lock);
18827c478bd9Sstevel@tonic-gate 				mutex_exit(&pcgs_wait_lock);
18837c478bd9Sstevel@tonic-gate 			}
18847c478bd9Sstevel@tonic-gate 		} else {
18857c478bd9Sstevel@tonic-gate #ifdef VM_STATS
18867c478bd9Sstevel@tonic-gate 			if (count >= PCGS_TRIES) {
18877c478bd9Sstevel@tonic-gate 				VM_STAT_ADD(pcgs_too_many);
18887c478bd9Sstevel@tonic-gate 			} else {
18897c478bd9Sstevel@tonic-gate 				VM_STAT_ADD(pcgs_counts[count]);
18907c478bd9Sstevel@tonic-gate 			}
18917c478bd9Sstevel@tonic-gate #endif
18927c478bd9Sstevel@tonic-gate 			if (locked) {
18937c478bd9Sstevel@tonic-gate 				pcgs_unblock();
18947c478bd9Sstevel@tonic-gate 				mutex_exit(&pcgs_lock);
18957c478bd9Sstevel@tonic-gate 			}
18967c478bd9Sstevel@tonic-gate 			if (cagelocked)
18977c478bd9Sstevel@tonic-gate 				mutex_exit(&pcgs_cagelock);
18987c478bd9Sstevel@tonic-gate 			return (pp);
18997c478bd9Sstevel@tonic-gate 		}
19007c478bd9Sstevel@tonic-gate 	}
19017c478bd9Sstevel@tonic-gate 	/*
19027c478bd9Sstevel@tonic-gate 	 * we go down holding the pcf locks.
19037c478bd9Sstevel@tonic-gate 	 */
19047c478bd9Sstevel@tonic-gate 	panic("no %spage found %d",
19057c478bd9Sstevel@tonic-gate 	    ((flags & PG_NORELOC) ? "non-reloc " : ""), count);
19067c478bd9Sstevel@tonic-gate 	/*NOTREACHED*/
19077c478bd9Sstevel@tonic-gate }
19087c478bd9Sstevel@tonic-gate 
19097c478bd9Sstevel@tonic-gate /*
19107c478bd9Sstevel@tonic-gate  * Create enough pages for "bytes" worth of data starting at
19117c478bd9Sstevel@tonic-gate  * "off" in "vp".
19127c478bd9Sstevel@tonic-gate  *
19137c478bd9Sstevel@tonic-gate  *	Where flag must be one of:
19147c478bd9Sstevel@tonic-gate  *
19157c478bd9Sstevel@tonic-gate  *		PG_EXCL:	Exclusive create (fail if any page already
19167c478bd9Sstevel@tonic-gate  *				exists in the page cache) which does not
19177c478bd9Sstevel@tonic-gate  *				wait for memory to become available.
19187c478bd9Sstevel@tonic-gate  *
19197c478bd9Sstevel@tonic-gate  *		PG_WAIT:	Non-exclusive create which can wait for
19207c478bd9Sstevel@tonic-gate  *				memory to become available.
19217c478bd9Sstevel@tonic-gate  *
19227c478bd9Sstevel@tonic-gate  *		PG_PHYSCONTIG:	Allocate physically contiguous pages.
19237c478bd9Sstevel@tonic-gate  *				(Not Supported)
19247c478bd9Sstevel@tonic-gate  *
19257c478bd9Sstevel@tonic-gate  * A doubly linked list of pages is returned to the caller.  Each page
19267c478bd9Sstevel@tonic-gate  * on the list has the "exclusive" (p_selock) lock and "iolock" (p_iolock)
19277c478bd9Sstevel@tonic-gate  * lock.
19287c478bd9Sstevel@tonic-gate  *
19297c478bd9Sstevel@tonic-gate  * Unable to change the parameters to page_create() in a minor release,
19307c478bd9Sstevel@tonic-gate  * we renamed page_create() to page_create_va(), changed all known calls
19317c478bd9Sstevel@tonic-gate  * from page_create() to page_create_va(), and created this wrapper.
19327c478bd9Sstevel@tonic-gate  *
19337c478bd9Sstevel@tonic-gate  * Upon a major release, we should break compatibility by deleting this
19347c478bd9Sstevel@tonic-gate  * wrapper, and replacing all the strings "page_create_va", with "page_create".
19357c478bd9Sstevel@tonic-gate  *
19367c478bd9Sstevel@tonic-gate  * NOTE: There is a copy of this interface as page_create_io() in
19377c478bd9Sstevel@tonic-gate  *	 i86/vm/vm_machdep.c. Any bugs fixed here should be applied
19387c478bd9Sstevel@tonic-gate  *	 there.
19397c478bd9Sstevel@tonic-gate  */
19407c478bd9Sstevel@tonic-gate page_t *
19417c478bd9Sstevel@tonic-gate page_create(vnode_t *vp, u_offset_t off, size_t bytes, uint_t flags)
19427c478bd9Sstevel@tonic-gate {
19437c478bd9Sstevel@tonic-gate 	caddr_t random_vaddr;
19447c478bd9Sstevel@tonic-gate 	struct seg kseg;
19457c478bd9Sstevel@tonic-gate 
19467c478bd9Sstevel@tonic-gate #ifdef DEBUG
19477c478bd9Sstevel@tonic-gate 	cmn_err(CE_WARN, "Using deprecated interface page_create: caller %p",
19487c478bd9Sstevel@tonic-gate 	    (void *)caller());
19497c478bd9Sstevel@tonic-gate #endif
19507c478bd9Sstevel@tonic-gate 
19517c478bd9Sstevel@tonic-gate 	random_vaddr = (caddr_t)(((uintptr_t)vp >> 7) ^
19527c478bd9Sstevel@tonic-gate 	    (uintptr_t)(off >> PAGESHIFT));
19537c478bd9Sstevel@tonic-gate 	kseg.s_as = &kas;
19547c478bd9Sstevel@tonic-gate 
19557c478bd9Sstevel@tonic-gate 	return (page_create_va(vp, off, bytes, flags, &kseg, random_vaddr));
19567c478bd9Sstevel@tonic-gate }
19577c478bd9Sstevel@tonic-gate 
19587c478bd9Sstevel@tonic-gate #ifdef DEBUG
19597c478bd9Sstevel@tonic-gate uint32_t pg_alloc_pgs_mtbf = 0;
19607c478bd9Sstevel@tonic-gate #endif
19617c478bd9Sstevel@tonic-gate 
19627c478bd9Sstevel@tonic-gate /*
19637c478bd9Sstevel@tonic-gate  * Used for large page support. It will attempt to allocate
19647c478bd9Sstevel@tonic-gate  * a large page(s) off the freelist.
19657c478bd9Sstevel@tonic-gate  *
19667c478bd9Sstevel@tonic-gate  * Returns non zero on failure.
19677c478bd9Sstevel@tonic-gate  */
19687c478bd9Sstevel@tonic-gate int
1969e44bd21cSsusans page_alloc_pages(struct vnode *vp, struct seg *seg, caddr_t addr,
1970e44bd21cSsusans     page_t **basepp, page_t *ppa[], uint_t szc, int anypgsz)
19717c478bd9Sstevel@tonic-gate {
19727c478bd9Sstevel@tonic-gate 	pgcnt_t		npgs, curnpgs, totpgs;
19737c478bd9Sstevel@tonic-gate 	size_t		pgsz;
19747c478bd9Sstevel@tonic-gate 	page_t		*pplist = NULL, *pp;
19757c478bd9Sstevel@tonic-gate 	int		err = 0;
19767c478bd9Sstevel@tonic-gate 	lgrp_t		*lgrp;
19777c478bd9Sstevel@tonic-gate 
19787c478bd9Sstevel@tonic-gate 	ASSERT(szc != 0 && szc <= (page_num_pagesizes() - 1));
19797c478bd9Sstevel@tonic-gate 
19807c478bd9Sstevel@tonic-gate 	VM_STAT_ADD(alloc_pages[0]);
19817c478bd9Sstevel@tonic-gate 
19827c478bd9Sstevel@tonic-gate #ifdef DEBUG
19837c478bd9Sstevel@tonic-gate 	if (pg_alloc_pgs_mtbf && !(gethrtime() % pg_alloc_pgs_mtbf)) {
19847c478bd9Sstevel@tonic-gate 		return (ENOMEM);
19857c478bd9Sstevel@tonic-gate 	}
19867c478bd9Sstevel@tonic-gate #endif
19877c478bd9Sstevel@tonic-gate 
19887c478bd9Sstevel@tonic-gate 	pgsz = page_get_pagesize(szc);
19897c478bd9Sstevel@tonic-gate 	totpgs = curnpgs = npgs = pgsz >> PAGESHIFT;
19907c478bd9Sstevel@tonic-gate 
19917c478bd9Sstevel@tonic-gate 	ASSERT(((uintptr_t)addr & (pgsz - 1)) == 0);
19927c478bd9Sstevel@tonic-gate 	/*
19937c478bd9Sstevel@tonic-gate 	 * One must be NULL but not both.
19947c478bd9Sstevel@tonic-gate 	 * And one must be non NULL but not both.
19957c478bd9Sstevel@tonic-gate 	 */
19967c478bd9Sstevel@tonic-gate 	ASSERT(basepp != NULL || ppa != NULL);
19977c478bd9Sstevel@tonic-gate 	ASSERT(basepp == NULL || ppa == NULL);
19987c478bd9Sstevel@tonic-gate 
19997c478bd9Sstevel@tonic-gate 	(void) page_create_wait(npgs, PG_WAIT);
20007c478bd9Sstevel@tonic-gate 
20017c478bd9Sstevel@tonic-gate 	while (npgs && szc) {
20027c478bd9Sstevel@tonic-gate 		lgrp = lgrp_mem_choose(seg, addr, pgsz);
2003e44bd21cSsusans 		pp = page_get_freelist(vp, 0, seg, addr, pgsz, 0, lgrp);
20047c478bd9Sstevel@tonic-gate 		if (pp != NULL) {
20057c478bd9Sstevel@tonic-gate 			VM_STAT_ADD(alloc_pages[1]);
20067c478bd9Sstevel@tonic-gate 			page_list_concat(&pplist, &pp);
20077c478bd9Sstevel@tonic-gate 			ASSERT(npgs >= curnpgs);
20087c478bd9Sstevel@tonic-gate 			npgs -= curnpgs;
20097c478bd9Sstevel@tonic-gate 		} else if (anypgsz) {
20107c478bd9Sstevel@tonic-gate 			VM_STAT_ADD(alloc_pages[2]);
20117c478bd9Sstevel@tonic-gate 			szc--;
20127c478bd9Sstevel@tonic-gate 			pgsz = page_get_pagesize(szc);
20137c478bd9Sstevel@tonic-gate 			curnpgs = pgsz >> PAGESHIFT;
20147c478bd9Sstevel@tonic-gate 		} else {
20157c478bd9Sstevel@tonic-gate 			VM_STAT_ADD(alloc_pages[3]);
20167c478bd9Sstevel@tonic-gate 			ASSERT(npgs == totpgs);
20177c478bd9Sstevel@tonic-gate 			page_create_putback(npgs);
20187c478bd9Sstevel@tonic-gate 			return (ENOMEM);
20197c478bd9Sstevel@tonic-gate 		}
20207c478bd9Sstevel@tonic-gate 	}
20217c478bd9Sstevel@tonic-gate 	if (szc == 0) {
20227c478bd9Sstevel@tonic-gate 		VM_STAT_ADD(alloc_pages[4]);
20237c478bd9Sstevel@tonic-gate 		ASSERT(npgs != 0);
20247c478bd9Sstevel@tonic-gate 		page_create_putback(npgs);
20257c478bd9Sstevel@tonic-gate 		err = ENOMEM;
20267c478bd9Sstevel@tonic-gate 	} else if (basepp != NULL) {
20277c478bd9Sstevel@tonic-gate 		ASSERT(npgs == 0);
20287c478bd9Sstevel@tonic-gate 		ASSERT(ppa == NULL);
20297c478bd9Sstevel@tonic-gate 		*basepp = pplist;
20307c478bd9Sstevel@tonic-gate 	}
20317c478bd9Sstevel@tonic-gate 
20327c478bd9Sstevel@tonic-gate 	npgs = totpgs - npgs;
20337c478bd9Sstevel@tonic-gate 	pp = pplist;
20347c478bd9Sstevel@tonic-gate 
20357c478bd9Sstevel@tonic-gate 	/*
20367c478bd9Sstevel@tonic-gate 	 * Clear the free and age bits. Also if we were passed in a ppa then
20377c478bd9Sstevel@tonic-gate 	 * fill it in with all the constituent pages from the large page. But
20387c478bd9Sstevel@tonic-gate 	 * if we failed to allocate all the pages just free what we got.
20397c478bd9Sstevel@tonic-gate 	 */
20407c478bd9Sstevel@tonic-gate 	while (npgs != 0) {
20417c478bd9Sstevel@tonic-gate 		ASSERT(PP_ISFREE(pp));
20427c478bd9Sstevel@tonic-gate 		ASSERT(PP_ISAGED(pp));
20437c478bd9Sstevel@tonic-gate 		if (ppa != NULL || err != 0) {
20447c478bd9Sstevel@tonic-gate 			if (err == 0) {
20457c478bd9Sstevel@tonic-gate 				VM_STAT_ADD(alloc_pages[5]);
20467c478bd9Sstevel@tonic-gate 				PP_CLRFREE(pp);
20477c478bd9Sstevel@tonic-gate 				PP_CLRAGED(pp);
20487c478bd9Sstevel@tonic-gate 				page_sub(&pplist, pp);
20497c478bd9Sstevel@tonic-gate 				*ppa++ = pp;
20507c478bd9Sstevel@tonic-gate 				npgs--;
20517c478bd9Sstevel@tonic-gate 			} else {
20527c478bd9Sstevel@tonic-gate 				VM_STAT_ADD(alloc_pages[6]);
20537c478bd9Sstevel@tonic-gate 				ASSERT(pp->p_szc != 0);
20547c478bd9Sstevel@tonic-gate 				curnpgs = page_get_pagecnt(pp->p_szc);
20557c478bd9Sstevel@tonic-gate 				page_list_break(&pp, &pplist, curnpgs);
20567c478bd9Sstevel@tonic-gate 				page_list_add_pages(pp, 0);
20577c478bd9Sstevel@tonic-gate 				page_create_putback(curnpgs);
20587c478bd9Sstevel@tonic-gate 				ASSERT(npgs >= curnpgs);
20597c478bd9Sstevel@tonic-gate 				npgs -= curnpgs;
20607c478bd9Sstevel@tonic-gate 			}
20617c478bd9Sstevel@tonic-gate 			pp = pplist;
20627c478bd9Sstevel@tonic-gate 		} else {
20637c478bd9Sstevel@tonic-gate 			VM_STAT_ADD(alloc_pages[7]);
20647c478bd9Sstevel@tonic-gate 			PP_CLRFREE(pp);
20657c478bd9Sstevel@tonic-gate 			PP_CLRAGED(pp);
20667c478bd9Sstevel@tonic-gate 			pp = pp->p_next;
20677c478bd9Sstevel@tonic-gate 			npgs--;
20687c478bd9Sstevel@tonic-gate 		}
20697c478bd9Sstevel@tonic-gate 	}
20707c478bd9Sstevel@tonic-gate 	return (err);
20717c478bd9Sstevel@tonic-gate }
20727c478bd9Sstevel@tonic-gate 
20737c478bd9Sstevel@tonic-gate /*
20747c478bd9Sstevel@tonic-gate  * Get a single large page off of the freelists, and set it up for use.
20757c478bd9Sstevel@tonic-gate  * Number of bytes requested must be a supported page size.
20767c478bd9Sstevel@tonic-gate  *
20777c478bd9Sstevel@tonic-gate  * Note that this call may fail even if there is sufficient
20787c478bd9Sstevel@tonic-gate  * memory available or PG_WAIT is set, so the caller must
20797c478bd9Sstevel@tonic-gate  * be willing to fallback on page_create_va(), block and retry,
20807c478bd9Sstevel@tonic-gate  * or fail the requester.
20817c478bd9Sstevel@tonic-gate  */
20827c478bd9Sstevel@tonic-gate page_t *
20837c478bd9Sstevel@tonic-gate page_create_va_large(vnode_t *vp, u_offset_t off, size_t bytes, uint_t flags,
20847c478bd9Sstevel@tonic-gate     struct seg *seg, caddr_t vaddr, void *arg)
20857c478bd9Sstevel@tonic-gate {
20867c478bd9Sstevel@tonic-gate 	pgcnt_t		npages, pcftotal;
20877c478bd9Sstevel@tonic-gate 	page_t		*pp;
20887c478bd9Sstevel@tonic-gate 	page_t		*rootpp;
20897c478bd9Sstevel@tonic-gate 	lgrp_t		*lgrp;
20907c478bd9Sstevel@tonic-gate 	uint_t		enough;
20917c478bd9Sstevel@tonic-gate 	uint_t		pcf_index;
20927c478bd9Sstevel@tonic-gate 	uint_t		i;
20937c478bd9Sstevel@tonic-gate 	struct pcf	*p;
20947c478bd9Sstevel@tonic-gate 	struct pcf	*q;
20957c478bd9Sstevel@tonic-gate 	lgrp_id_t	*lgrpid = (lgrp_id_t *)arg;
20967c478bd9Sstevel@tonic-gate 
20977c478bd9Sstevel@tonic-gate 	ASSERT(vp != NULL);
20987c478bd9Sstevel@tonic-gate 
20997c478bd9Sstevel@tonic-gate 	ASSERT((flags & ~(PG_EXCL | PG_WAIT |
21007c478bd9Sstevel@tonic-gate 		    PG_NORELOC | PG_PANIC | PG_PUSHPAGE)) == 0);
21017c478bd9Sstevel@tonic-gate 	/* but no others */
21027c478bd9Sstevel@tonic-gate 
21037c478bd9Sstevel@tonic-gate 	ASSERT((flags & PG_EXCL) == PG_EXCL);
21047c478bd9Sstevel@tonic-gate 
21057c478bd9Sstevel@tonic-gate 	npages = btop(bytes);
21067c478bd9Sstevel@tonic-gate 
21077c478bd9Sstevel@tonic-gate 	if (!kcage_on || panicstr) {
21087c478bd9Sstevel@tonic-gate 		/*
21097c478bd9Sstevel@tonic-gate 		 * Cage is OFF, or we are single threaded in
21107c478bd9Sstevel@tonic-gate 		 * panic, so make everything a RELOC request.
21117c478bd9Sstevel@tonic-gate 		 */
21127c478bd9Sstevel@tonic-gate 		flags &= ~PG_NORELOC;
21137c478bd9Sstevel@tonic-gate 	}
21147c478bd9Sstevel@tonic-gate 
21157c478bd9Sstevel@tonic-gate 	/*
21167c478bd9Sstevel@tonic-gate 	 * Make sure there's adequate physical memory available.
21177c478bd9Sstevel@tonic-gate 	 * Note: PG_WAIT is ignored here.
21187c478bd9Sstevel@tonic-gate 	 */
21197c478bd9Sstevel@tonic-gate 	if (freemem <= throttlefree + npages) {
21207c478bd9Sstevel@tonic-gate 		VM_STAT_ADD(page_create_large_cnt[1]);
21217c478bd9Sstevel@tonic-gate 		return (NULL);
21227c478bd9Sstevel@tonic-gate 	}
21237c478bd9Sstevel@tonic-gate 
21247c478bd9Sstevel@tonic-gate 	/*
21257c478bd9Sstevel@tonic-gate 	 * If cage is on, dampen draw from cage when available
21267c478bd9Sstevel@tonic-gate 	 * cage space is low.
21277c478bd9Sstevel@tonic-gate 	 */
21287c478bd9Sstevel@tonic-gate 	if ((flags & (PG_NORELOC | PG_WAIT)) ==  (PG_NORELOC | PG_WAIT) &&
21297c478bd9Sstevel@tonic-gate 	    kcage_freemem < kcage_throttlefree + npages) {
21307c478bd9Sstevel@tonic-gate 
21317c478bd9Sstevel@tonic-gate 		/*
21327c478bd9Sstevel@tonic-gate 		 * The cage is on, the caller wants PG_NORELOC
21337c478bd9Sstevel@tonic-gate 		 * pages and available cage memory is very low.
21347c478bd9Sstevel@tonic-gate 		 * Call kcage_create_throttle() to attempt to
21357c478bd9Sstevel@tonic-gate 		 * control demand on the cage.
21367c478bd9Sstevel@tonic-gate 		 */
21377c478bd9Sstevel@tonic-gate 		if (kcage_create_throttle(npages, flags) == KCT_FAILURE) {
21387c478bd9Sstevel@tonic-gate 			VM_STAT_ADD(page_create_large_cnt[2]);
21397c478bd9Sstevel@tonic-gate 			return (NULL);
21407c478bd9Sstevel@tonic-gate 		}
21417c478bd9Sstevel@tonic-gate 	}
21427c478bd9Sstevel@tonic-gate 
21437c478bd9Sstevel@tonic-gate 	enough = 0;
21447c478bd9Sstevel@tonic-gate 	pcf_index = PCF_INDEX();
21457c478bd9Sstevel@tonic-gate 	p = &pcf[pcf_index];
21467c478bd9Sstevel@tonic-gate 	q = &pcf[PCF_FANOUT];
21477c478bd9Sstevel@tonic-gate 	for (pcftotal = 0, i = 0; i < PCF_FANOUT; i++) {
21487c478bd9Sstevel@tonic-gate 		if (p->pcf_count > npages) {
21497c478bd9Sstevel@tonic-gate 			/*
21507c478bd9Sstevel@tonic-gate 			 * a good one to try.
21517c478bd9Sstevel@tonic-gate 			 */
21527c478bd9Sstevel@tonic-gate 			mutex_enter(&p->pcf_lock);
21537c478bd9Sstevel@tonic-gate 			if (p->pcf_count > npages) {
21547c478bd9Sstevel@tonic-gate 				p->pcf_count -= (uint_t)npages;
21557c478bd9Sstevel@tonic-gate 				/*
21567c478bd9Sstevel@tonic-gate 				 * freemem is not protected by any lock.
21577c478bd9Sstevel@tonic-gate 				 * Thus, we cannot have any assertion
21587c478bd9Sstevel@tonic-gate 				 * containing freemem here.
21597c478bd9Sstevel@tonic-gate 				 */
21607c478bd9Sstevel@tonic-gate 				freemem -= npages;
21617c478bd9Sstevel@tonic-gate 				enough = 1;
21627c478bd9Sstevel@tonic-gate 				mutex_exit(&p->pcf_lock);
21637c478bd9Sstevel@tonic-gate 				break;
21647c478bd9Sstevel@tonic-gate 			}
21657c478bd9Sstevel@tonic-gate 			mutex_exit(&p->pcf_lock);
21667c478bd9Sstevel@tonic-gate 		}
21677c478bd9Sstevel@tonic-gate 		pcftotal += p->pcf_count;
21687c478bd9Sstevel@tonic-gate 		p++;
21697c478bd9Sstevel@tonic-gate 		if (p >= q) {
21707c478bd9Sstevel@tonic-gate 			p = pcf;
21717c478bd9Sstevel@tonic-gate 		}
21727c478bd9Sstevel@tonic-gate 	}
21737c478bd9Sstevel@tonic-gate 
21747c478bd9Sstevel@tonic-gate 	if (!enough) {
21757c478bd9Sstevel@tonic-gate 		/* If there isn't enough memory available, give up. */
21767c478bd9Sstevel@tonic-gate 		if (pcftotal < npages) {
21777c478bd9Sstevel@tonic-gate 			VM_STAT_ADD(page_create_large_cnt[3]);
21787c478bd9Sstevel@tonic-gate 			return (NULL);
21797c478bd9Sstevel@tonic-gate 		}
21807c478bd9Sstevel@tonic-gate 
21817c478bd9Sstevel@tonic-gate 		/* try to collect pages from several pcf bins */
21827c478bd9Sstevel@tonic-gate 		for (p = pcf, pcftotal = 0, i = 0; i < PCF_FANOUT; i++) {
21837c478bd9Sstevel@tonic-gate 			mutex_enter(&p->pcf_lock);
21847c478bd9Sstevel@tonic-gate 			pcftotal += p->pcf_count;
21857c478bd9Sstevel@tonic-gate 			if (pcftotal >= npages) {
21867c478bd9Sstevel@tonic-gate 				/*
21877c478bd9Sstevel@tonic-gate 				 * Wow!  There are enough pages laying around
21887c478bd9Sstevel@tonic-gate 				 * to satisfy the request.  Do the accounting,
21897c478bd9Sstevel@tonic-gate 				 * drop the locks we acquired, and go back.
21907c478bd9Sstevel@tonic-gate 				 *
21917c478bd9Sstevel@tonic-gate 				 * freemem is not protected by any lock. So,
21927c478bd9Sstevel@tonic-gate 				 * we cannot have any assertion containing
21937c478bd9Sstevel@tonic-gate 				 * freemem.
21947c478bd9Sstevel@tonic-gate 				 */
21957c478bd9Sstevel@tonic-gate 				pgcnt_t	tpages = npages;
21967c478bd9Sstevel@tonic-gate 				freemem -= npages;
21977c478bd9Sstevel@tonic-gate 				while (p >= pcf) {
21987c478bd9Sstevel@tonic-gate 					if (p->pcf_count <= tpages) {
21997c478bd9Sstevel@tonic-gate 						tpages -= p->pcf_count;
22007c478bd9Sstevel@tonic-gate 						p->pcf_count = 0;
22017c478bd9Sstevel@tonic-gate 					} else {
22027c478bd9Sstevel@tonic-gate 						p->pcf_count -= (uint_t)tpages;
22037c478bd9Sstevel@tonic-gate 						tpages = 0;
22047c478bd9Sstevel@tonic-gate 					}
22057c478bd9Sstevel@tonic-gate 					mutex_exit(&p->pcf_lock);
22067c478bd9Sstevel@tonic-gate 					p--;
22077c478bd9Sstevel@tonic-gate 				}
22087c478bd9Sstevel@tonic-gate 				ASSERT(tpages == 0);
22097c478bd9Sstevel@tonic-gate 				break;
22107c478bd9Sstevel@tonic-gate 			}
22117c478bd9Sstevel@tonic-gate 			p++;
22127c478bd9Sstevel@tonic-gate 		}
22137c478bd9Sstevel@tonic-gate 		if (i == PCF_FANOUT) {
22147c478bd9Sstevel@tonic-gate 			/* failed to collect pages - release the locks */
22157c478bd9Sstevel@tonic-gate 			while (--p >= pcf) {
22167c478bd9Sstevel@tonic-gate 				mutex_exit(&p->pcf_lock);
22177c478bd9Sstevel@tonic-gate 			}
22187c478bd9Sstevel@tonic-gate 			VM_STAT_ADD(page_create_large_cnt[4]);
22197c478bd9Sstevel@tonic-gate 			return (NULL);
22207c478bd9Sstevel@tonic-gate 		}
22217c478bd9Sstevel@tonic-gate 	}
22227c478bd9Sstevel@tonic-gate 
22237c478bd9Sstevel@tonic-gate 	/*
22247c478bd9Sstevel@tonic-gate 	 * This is where this function behaves fundamentally differently
22257c478bd9Sstevel@tonic-gate 	 * than page_create_va(); since we're intending to map the page
22267c478bd9Sstevel@tonic-gate 	 * with a single TTE, we have to get it as a physically contiguous
22277c478bd9Sstevel@tonic-gate 	 * hardware pagesize chunk.  If we can't, we fail.
22287c478bd9Sstevel@tonic-gate 	 */
22297c478bd9Sstevel@tonic-gate 	if (lgrpid != NULL && *lgrpid >= 0 && *lgrpid <= lgrp_alloc_max &&
22307c478bd9Sstevel@tonic-gate 		LGRP_EXISTS(lgrp_table[*lgrpid]))
22317c478bd9Sstevel@tonic-gate 		lgrp = lgrp_table[*lgrpid];
22327c478bd9Sstevel@tonic-gate 	else
22337c478bd9Sstevel@tonic-gate 		lgrp = lgrp_mem_choose(seg, vaddr, bytes);
22347c478bd9Sstevel@tonic-gate 
22357c478bd9Sstevel@tonic-gate 	if ((rootpp = page_get_freelist(&kvp, off, seg, vaddr,
22367c478bd9Sstevel@tonic-gate 	    bytes, flags & ~PG_MATCH_COLOR, lgrp)) == NULL) {
22377c478bd9Sstevel@tonic-gate 		page_create_putback(npages);
22387c478bd9Sstevel@tonic-gate 		VM_STAT_ADD(page_create_large_cnt[5]);
22397c478bd9Sstevel@tonic-gate 		return (NULL);
22407c478bd9Sstevel@tonic-gate 	}
22417c478bd9Sstevel@tonic-gate 
22427c478bd9Sstevel@tonic-gate 	/*
22437c478bd9Sstevel@tonic-gate 	 * if we got the page with the wrong mtype give it back this is a
22447c478bd9Sstevel@tonic-gate 	 * workaround for CR 6249718. When CR 6249718 is fixed we never get
22457c478bd9Sstevel@tonic-gate 	 * inside "if" and the workaround becomes just a nop
22467c478bd9Sstevel@tonic-gate 	 */
22477c478bd9Sstevel@tonic-gate 	if (kcage_on && (flags & PG_NORELOC) && !PP_ISNORELOC(rootpp)) {
22487c478bd9Sstevel@tonic-gate 		page_list_add_pages(rootpp, 0);
22497c478bd9Sstevel@tonic-gate 		page_create_putback(npages);
22507c478bd9Sstevel@tonic-gate 		VM_STAT_ADD(page_create_large_cnt[6]);
22517c478bd9Sstevel@tonic-gate 		return (NULL);
22527c478bd9Sstevel@tonic-gate 	}
22537c478bd9Sstevel@tonic-gate 
22547c478bd9Sstevel@tonic-gate 	/*
22557c478bd9Sstevel@tonic-gate 	 * If satisfying this request has left us with too little
22567c478bd9Sstevel@tonic-gate 	 * memory, start the wheels turning to get some back.  The
22577c478bd9Sstevel@tonic-gate 	 * first clause of the test prevents waking up the pageout
22587c478bd9Sstevel@tonic-gate 	 * daemon in situations where it would decide that there's
22597c478bd9Sstevel@tonic-gate 	 * nothing to do.
22607c478bd9Sstevel@tonic-gate 	 */
22617c478bd9Sstevel@tonic-gate 	if (nscan < desscan && freemem < minfree) {
22627c478bd9Sstevel@tonic-gate 		TRACE_1(TR_FAC_VM, TR_PAGEOUT_CV_SIGNAL,
22637c478bd9Sstevel@tonic-gate 		    "pageout_cv_signal:freemem %ld", freemem);
22647c478bd9Sstevel@tonic-gate 		cv_signal(&proc_pageout->p_cv);
22657c478bd9Sstevel@tonic-gate 	}
22667c478bd9Sstevel@tonic-gate 
22677c478bd9Sstevel@tonic-gate 	pp = rootpp;
22687c478bd9Sstevel@tonic-gate 	while (npages--) {
22697c478bd9Sstevel@tonic-gate 		ASSERT(PAGE_EXCL(pp));
22707c478bd9Sstevel@tonic-gate 		ASSERT(pp->p_vnode == NULL);
22717c478bd9Sstevel@tonic-gate 		ASSERT(!hat_page_is_mapped(pp));
22727c478bd9Sstevel@tonic-gate 		PP_CLRFREE(pp);
22737c478bd9Sstevel@tonic-gate 		PP_CLRAGED(pp);
22747c478bd9Sstevel@tonic-gate 		if (!page_hashin(pp, vp, off, NULL))
22757c478bd9Sstevel@tonic-gate 			panic("page_create_large: hashin failed: page %p",
22767c478bd9Sstevel@tonic-gate 			    (void *)pp);
22777c478bd9Sstevel@tonic-gate 		page_io_lock(pp);
22787c478bd9Sstevel@tonic-gate 		off += PAGESIZE;
22797c478bd9Sstevel@tonic-gate 		pp = pp->p_next;
22807c478bd9Sstevel@tonic-gate 	}
22817c478bd9Sstevel@tonic-gate 
22827c478bd9Sstevel@tonic-gate 	VM_STAT_ADD(page_create_large_cnt[0]);
22837c478bd9Sstevel@tonic-gate 	return (rootpp);
22847c478bd9Sstevel@tonic-gate }
22857c478bd9Sstevel@tonic-gate 
22867c478bd9Sstevel@tonic-gate page_t *
22877c478bd9Sstevel@tonic-gate page_create_va(vnode_t *vp, u_offset_t off, size_t bytes, uint_t flags,
22887c478bd9Sstevel@tonic-gate     struct seg *seg, caddr_t vaddr)
22897c478bd9Sstevel@tonic-gate {
22907c478bd9Sstevel@tonic-gate 	page_t		*plist = NULL;
22917c478bd9Sstevel@tonic-gate 	pgcnt_t		npages;
22927c478bd9Sstevel@tonic-gate 	pgcnt_t		found_on_free = 0;
22937c478bd9Sstevel@tonic-gate 	pgcnt_t		pages_req;
22947c478bd9Sstevel@tonic-gate 	page_t		*npp = NULL;
22957c478bd9Sstevel@tonic-gate 	uint_t		enough;
22967c478bd9Sstevel@tonic-gate 	uint_t		i;
22977c478bd9Sstevel@tonic-gate 	uint_t		pcf_index;
22987c478bd9Sstevel@tonic-gate 	struct pcf	*p;
22997c478bd9Sstevel@tonic-gate 	struct pcf	*q;
23007c478bd9Sstevel@tonic-gate 	lgrp_t		*lgrp;
23017c478bd9Sstevel@tonic-gate 
23027c478bd9Sstevel@tonic-gate 	TRACE_4(TR_FAC_VM, TR_PAGE_CREATE_START,
23037c478bd9Sstevel@tonic-gate 		"page_create_start:vp %p off %llx bytes %lu flags %x",
23047c478bd9Sstevel@tonic-gate 		vp, off, bytes, flags);
23057c478bd9Sstevel@tonic-gate 
23067c478bd9Sstevel@tonic-gate 	ASSERT(bytes != 0 && vp != NULL);
23077c478bd9Sstevel@tonic-gate 
23087c478bd9Sstevel@tonic-gate 	if ((flags & PG_EXCL) == 0 && (flags & PG_WAIT) == 0) {
23097c478bd9Sstevel@tonic-gate 		panic("page_create: invalid flags");
23107c478bd9Sstevel@tonic-gate 		/*NOTREACHED*/
23117c478bd9Sstevel@tonic-gate 	}
23127c478bd9Sstevel@tonic-gate 	ASSERT((flags & ~(PG_EXCL | PG_WAIT |
23137c478bd9Sstevel@tonic-gate 	    PG_NORELOC | PG_PANIC | PG_PUSHPAGE)) == 0);
23147c478bd9Sstevel@tonic-gate 	    /* but no others */
23157c478bd9Sstevel@tonic-gate 
23167c478bd9Sstevel@tonic-gate 	pages_req = npages = btopr(bytes);
23177c478bd9Sstevel@tonic-gate 	/*
23187c478bd9Sstevel@tonic-gate 	 * Try to see whether request is too large to *ever* be
23197c478bd9Sstevel@tonic-gate 	 * satisfied, in order to prevent deadlock.  We arbitrarily
23207c478bd9Sstevel@tonic-gate 	 * decide to limit maximum size requests to max_page_get.
23217c478bd9Sstevel@tonic-gate 	 */
23227c478bd9Sstevel@tonic-gate 	if (npages >= max_page_get) {
23237c478bd9Sstevel@tonic-gate 		if ((flags & PG_WAIT) == 0) {
23247c478bd9Sstevel@tonic-gate 			TRACE_4(TR_FAC_VM, TR_PAGE_CREATE_TOOBIG,
23257c478bd9Sstevel@tonic-gate 			    "page_create_toobig:vp %p off %llx npages "
23267c478bd9Sstevel@tonic-gate 			    "%lu max_page_get %lu",
23277c478bd9Sstevel@tonic-gate 			    vp, off, npages, max_page_get);
23287c478bd9Sstevel@tonic-gate 			return (NULL);
23297c478bd9Sstevel@tonic-gate 		} else {
23307c478bd9Sstevel@tonic-gate 			cmn_err(CE_WARN,
23317c478bd9Sstevel@tonic-gate 			    "Request for too much kernel memory "
23327c478bd9Sstevel@tonic-gate 			    "(%lu bytes), will hang forever", bytes);
23337c478bd9Sstevel@tonic-gate 			for (;;)
23347c478bd9Sstevel@tonic-gate 				delay(1000000000);
23357c478bd9Sstevel@tonic-gate 		}
23367c478bd9Sstevel@tonic-gate 	}
23377c478bd9Sstevel@tonic-gate 
23387c478bd9Sstevel@tonic-gate 	if (!kcage_on || panicstr) {
23397c478bd9Sstevel@tonic-gate 		/*
23407c478bd9Sstevel@tonic-gate 		 * Cage is OFF, or we are single threaded in
23417c478bd9Sstevel@tonic-gate 		 * panic, so make everything a RELOC request.
23427c478bd9Sstevel@tonic-gate 		 */
23437c478bd9Sstevel@tonic-gate 		flags &= ~PG_NORELOC;
23447c478bd9Sstevel@tonic-gate 	}
23457c478bd9Sstevel@tonic-gate 
23467c478bd9Sstevel@tonic-gate 	if (freemem <= throttlefree + npages)
23477c478bd9Sstevel@tonic-gate 		if (!page_create_throttle(npages, flags))
23487c478bd9Sstevel@tonic-gate 			return (NULL);
23497c478bd9Sstevel@tonic-gate 
23507c478bd9Sstevel@tonic-gate 	/*
23517c478bd9Sstevel@tonic-gate 	 * If cage is on, dampen draw from cage when available
23527c478bd9Sstevel@tonic-gate 	 * cage space is low.
23537c478bd9Sstevel@tonic-gate 	 */
23547c478bd9Sstevel@tonic-gate 	if ((flags & PG_NORELOC) &&
23557c478bd9Sstevel@tonic-gate 		kcage_freemem < kcage_throttlefree + npages) {
23567c478bd9Sstevel@tonic-gate 
23577c478bd9Sstevel@tonic-gate 		/*
23587c478bd9Sstevel@tonic-gate 		 * The cage is on, the caller wants PG_NORELOC
23597c478bd9Sstevel@tonic-gate 		 * pages and available cage memory is very low.
23607c478bd9Sstevel@tonic-gate 		 * Call kcage_create_throttle() to attempt to
23617c478bd9Sstevel@tonic-gate 		 * control demand on the cage.
23627c478bd9Sstevel@tonic-gate 		 */
23637c478bd9Sstevel@tonic-gate 		if (kcage_create_throttle(npages, flags) == KCT_FAILURE)
23647c478bd9Sstevel@tonic-gate 			return (NULL);
23657c478bd9Sstevel@tonic-gate 	}
23667c478bd9Sstevel@tonic-gate 
23677c478bd9Sstevel@tonic-gate 	VM_STAT_ADD(page_create_cnt[0]);
23687c478bd9Sstevel@tonic-gate 
23697c478bd9Sstevel@tonic-gate 	enough = 0;
23707c478bd9Sstevel@tonic-gate 	pcf_index = PCF_INDEX();
23717c478bd9Sstevel@tonic-gate 
23727c478bd9Sstevel@tonic-gate 	p = &pcf[pcf_index];
23737c478bd9Sstevel@tonic-gate 	q = &pcf[PCF_FANOUT];
23747c478bd9Sstevel@tonic-gate 	for (i = 0; i < PCF_FANOUT; i++) {
23757c478bd9Sstevel@tonic-gate 		if (p->pcf_count > npages) {
23767c478bd9Sstevel@tonic-gate 			/*
23777c478bd9Sstevel@tonic-gate 			 * a good one to try.
23787c478bd9Sstevel@tonic-gate 			 */
23797c478bd9Sstevel@tonic-gate 			mutex_enter(&p->pcf_lock);
23807c478bd9Sstevel@tonic-gate 			if (p->pcf_count > npages) {
23817c478bd9Sstevel@tonic-gate 				p->pcf_count -= (uint_t)npages;
23827c478bd9Sstevel@tonic-gate 				/*
23837c478bd9Sstevel@tonic-gate 				 * freemem is not protected by any lock.
23847c478bd9Sstevel@tonic-gate 				 * Thus, we cannot have any assertion
23857c478bd9Sstevel@tonic-gate 				 * containing freemem here.
23867c478bd9Sstevel@tonic-gate 				 */
23877c478bd9Sstevel@tonic-gate 				freemem -= npages;
23887c478bd9Sstevel@tonic-gate 				enough = 1;
23897c478bd9Sstevel@tonic-gate 				mutex_exit(&p->pcf_lock);
23907c478bd9Sstevel@tonic-gate 				break;
23917c478bd9Sstevel@tonic-gate 			}
23927c478bd9Sstevel@tonic-gate 			mutex_exit(&p->pcf_lock);
23937c478bd9Sstevel@tonic-gate 		}
23947c478bd9Sstevel@tonic-gate 		p++;
23957c478bd9Sstevel@tonic-gate 		if (p >= q) {
23967c478bd9Sstevel@tonic-gate 			p = pcf;
23977c478bd9Sstevel@tonic-gate 		}
23987c478bd9Sstevel@tonic-gate 	}
23997c478bd9Sstevel@tonic-gate 
24007c478bd9Sstevel@tonic-gate 	if (!enough) {
24017c478bd9Sstevel@tonic-gate 		/*
24027c478bd9Sstevel@tonic-gate 		 * Have to look harder.  If npages is greater than
24037c478bd9Sstevel@tonic-gate 		 * one, then we might have to coalecse the counters.
24047c478bd9Sstevel@tonic-gate 		 *
24057c478bd9Sstevel@tonic-gate 		 * Go wait.  We come back having accounted
24067c478bd9Sstevel@tonic-gate 		 * for the memory.
24077c478bd9Sstevel@tonic-gate 		 */
24087c478bd9Sstevel@tonic-gate 		VM_STAT_ADD(page_create_cnt[1]);
24097c478bd9Sstevel@tonic-gate 		if (!page_create_wait(npages, flags)) {
24107c478bd9Sstevel@tonic-gate 			VM_STAT_ADD(page_create_cnt[2]);
24117c478bd9Sstevel@tonic-gate 			return (NULL);
24127c478bd9Sstevel@tonic-gate 		}
24137c478bd9Sstevel@tonic-gate 	}
24147c478bd9Sstevel@tonic-gate 
24157c478bd9Sstevel@tonic-gate 	TRACE_2(TR_FAC_VM, TR_PAGE_CREATE_SUCCESS,
24167c478bd9Sstevel@tonic-gate 		"page_create_success:vp %p off %llx", vp, off);
24177c478bd9Sstevel@tonic-gate 
24187c478bd9Sstevel@tonic-gate 	/*
24197c478bd9Sstevel@tonic-gate 	 * If satisfying this request has left us with too little
24207c478bd9Sstevel@tonic-gate 	 * memory, start the wheels turning to get some back.  The
24217c478bd9Sstevel@tonic-gate 	 * first clause of the test prevents waking up the pageout
24227c478bd9Sstevel@tonic-gate 	 * daemon in situations where it would decide that there's
24237c478bd9Sstevel@tonic-gate 	 * nothing to do.
24247c478bd9Sstevel@tonic-gate 	 */
24257c478bd9Sstevel@tonic-gate 	if (nscan < desscan && freemem < minfree) {
24267c478bd9Sstevel@tonic-gate 		TRACE_1(TR_FAC_VM, TR_PAGEOUT_CV_SIGNAL,
24277c478bd9Sstevel@tonic-gate 			"pageout_cv_signal:freemem %ld", freemem);
24287c478bd9Sstevel@tonic-gate 		cv_signal(&proc_pageout->p_cv);
24297c478bd9Sstevel@tonic-gate 	}
24307c478bd9Sstevel@tonic-gate 
24317c478bd9Sstevel@tonic-gate 	/*
24327c478bd9Sstevel@tonic-gate 	 * Loop around collecting the requested number of pages.
24337c478bd9Sstevel@tonic-gate 	 * Most of the time, we have to `create' a new page. With
24347c478bd9Sstevel@tonic-gate 	 * this in mind, pull the page off the free list before
24357c478bd9Sstevel@tonic-gate 	 * getting the hash lock.  This will minimize the hash
24367c478bd9Sstevel@tonic-gate 	 * lock hold time, nesting, and the like.  If it turns
24377c478bd9Sstevel@tonic-gate 	 * out we don't need the page, we put it back at the end.
24387c478bd9Sstevel@tonic-gate 	 */
24397c478bd9Sstevel@tonic-gate 	while (npages--) {
24407c478bd9Sstevel@tonic-gate 		page_t		*pp;
24417c478bd9Sstevel@tonic-gate 		kmutex_t	*phm = NULL;
24427c478bd9Sstevel@tonic-gate 		ulong_t		index;
24437c478bd9Sstevel@tonic-gate 
24447c478bd9Sstevel@tonic-gate 		index = PAGE_HASH_FUNC(vp, off);
24457c478bd9Sstevel@tonic-gate top:
24467c478bd9Sstevel@tonic-gate 		ASSERT(phm == NULL);
24477c478bd9Sstevel@tonic-gate 		ASSERT(index == PAGE_HASH_FUNC(vp, off));
24487c478bd9Sstevel@tonic-gate 		ASSERT(MUTEX_NOT_HELD(page_vnode_mutex(vp)));
24497c478bd9Sstevel@tonic-gate 
24507c478bd9Sstevel@tonic-gate 		if (npp == NULL) {
24517c478bd9Sstevel@tonic-gate 			/*
24527c478bd9Sstevel@tonic-gate 			 * Try to get a page from the freelist (ie,
24537c478bd9Sstevel@tonic-gate 			 * a page with no [vp, off] tag).  If that
24547c478bd9Sstevel@tonic-gate 			 * fails, use the cachelist.
24557c478bd9Sstevel@tonic-gate 			 *
24567c478bd9Sstevel@tonic-gate 			 * During the first attempt at both the free
24577c478bd9Sstevel@tonic-gate 			 * and cache lists we try for the correct color.
24587c478bd9Sstevel@tonic-gate 			 */
24597c478bd9Sstevel@tonic-gate 			/*
24607c478bd9Sstevel@tonic-gate 			 * XXXX-how do we deal with virtual indexed
24617c478bd9Sstevel@tonic-gate 			 * caches and and colors?
24627c478bd9Sstevel@tonic-gate 			 */
24637c478bd9Sstevel@tonic-gate 			VM_STAT_ADD(page_create_cnt[4]);
24647c478bd9Sstevel@tonic-gate 			/*
24657c478bd9Sstevel@tonic-gate 			 * Get lgroup to allocate next page of shared memory
24667c478bd9Sstevel@tonic-gate 			 * from and use it to specify where to allocate
24677c478bd9Sstevel@tonic-gate 			 * the physical memory
24687c478bd9Sstevel@tonic-gate 			 */
24697c478bd9Sstevel@tonic-gate 			lgrp = lgrp_mem_choose(seg, vaddr, PAGESIZE);
24707c478bd9Sstevel@tonic-gate 			npp = page_get_freelist(vp, off, seg, vaddr, PAGESIZE,
24717c478bd9Sstevel@tonic-gate 			    flags | PG_MATCH_COLOR, lgrp);
24727c478bd9Sstevel@tonic-gate 			if (npp == NULL) {
24737c478bd9Sstevel@tonic-gate 				npp = page_get_cachelist(vp, off, seg,
24747c478bd9Sstevel@tonic-gate 				    vaddr, flags | PG_MATCH_COLOR, lgrp);
24757c478bd9Sstevel@tonic-gate 				if (npp == NULL) {
24767c478bd9Sstevel@tonic-gate 					npp = page_create_get_something(vp,
24777c478bd9Sstevel@tonic-gate 					    off, seg, vaddr,
24787c478bd9Sstevel@tonic-gate 					    flags & ~PG_MATCH_COLOR);
24797c478bd9Sstevel@tonic-gate 				}
24807c478bd9Sstevel@tonic-gate 
24817c478bd9Sstevel@tonic-gate 				if (PP_ISAGED(npp) == 0) {
24827c478bd9Sstevel@tonic-gate 					/*
24837c478bd9Sstevel@tonic-gate 					 * Since this page came from the
24847c478bd9Sstevel@tonic-gate 					 * cachelist, we must destroy the
24857c478bd9Sstevel@tonic-gate 					 * old vnode association.
24867c478bd9Sstevel@tonic-gate 					 */
24877c478bd9Sstevel@tonic-gate 					page_hashout(npp, NULL);
24887c478bd9Sstevel@tonic-gate 				}
24897c478bd9Sstevel@tonic-gate 			}
24907c478bd9Sstevel@tonic-gate 		}
24917c478bd9Sstevel@tonic-gate 
24927c478bd9Sstevel@tonic-gate 		/*
24937c478bd9Sstevel@tonic-gate 		 * We own this page!
24947c478bd9Sstevel@tonic-gate 		 */
24957c478bd9Sstevel@tonic-gate 		ASSERT(PAGE_EXCL(npp));
24967c478bd9Sstevel@tonic-gate 		ASSERT(npp->p_vnode == NULL);
24977c478bd9Sstevel@tonic-gate 		ASSERT(!hat_page_is_mapped(npp));
24987c478bd9Sstevel@tonic-gate 		PP_CLRFREE(npp);
24997c478bd9Sstevel@tonic-gate 		PP_CLRAGED(npp);
25007c478bd9Sstevel@tonic-gate 
25017c478bd9Sstevel@tonic-gate 		/*
25027c478bd9Sstevel@tonic-gate 		 * Here we have a page in our hot little mits and are
25037c478bd9Sstevel@tonic-gate 		 * just waiting to stuff it on the appropriate lists.
25047c478bd9Sstevel@tonic-gate 		 * Get the mutex and check to see if it really does
25057c478bd9Sstevel@tonic-gate 		 * not exist.
25067c478bd9Sstevel@tonic-gate 		 */
25077c478bd9Sstevel@tonic-gate 		phm = PAGE_HASH_MUTEX(index);
25087c478bd9Sstevel@tonic-gate 		mutex_enter(phm);
25097c478bd9Sstevel@tonic-gate 		PAGE_HASH_SEARCH(index, pp, vp, off);
25107c478bd9Sstevel@tonic-gate 		if (pp == NULL) {
25117c478bd9Sstevel@tonic-gate 			VM_STAT_ADD(page_create_new);
25127c478bd9Sstevel@tonic-gate 			pp = npp;
25137c478bd9Sstevel@tonic-gate 			npp = NULL;
25147c478bd9Sstevel@tonic-gate 			if (!page_hashin(pp, vp, off, phm)) {
25157c478bd9Sstevel@tonic-gate 				/*
25167c478bd9Sstevel@tonic-gate 				 * Since we hold the page hash mutex and
25177c478bd9Sstevel@tonic-gate 				 * just searched for this page, page_hashin
25187c478bd9Sstevel@tonic-gate 				 * had better not fail.  If it does, that
25197c478bd9Sstevel@tonic-gate 				 * means somethread did not follow the
25207c478bd9Sstevel@tonic-gate 				 * page hash mutex rules.  Panic now and
25217c478bd9Sstevel@tonic-gate 				 * get it over with.  As usual, go down
25227c478bd9Sstevel@tonic-gate 				 * holding all the locks.
25237c478bd9Sstevel@tonic-gate 				 */
25247c478bd9Sstevel@tonic-gate 				ASSERT(MUTEX_HELD(phm));
25257c478bd9Sstevel@tonic-gate 				panic("page_create: "
25267c478bd9Sstevel@tonic-gate 				    "hashin failed %p %p %llx %p",
25277c478bd9Sstevel@tonic-gate 				    (void *)pp, (void *)vp, off, (void *)phm);
25287c478bd9Sstevel@tonic-gate 				/*NOTREACHED*/
25297c478bd9Sstevel@tonic-gate 			}
25307c478bd9Sstevel@tonic-gate 			ASSERT(MUTEX_HELD(phm));
25317c478bd9Sstevel@tonic-gate 			mutex_exit(phm);
25327c478bd9Sstevel@tonic-gate 			phm = NULL;
25337c478bd9Sstevel@tonic-gate 
25347c478bd9Sstevel@tonic-gate 			/*
25357c478bd9Sstevel@tonic-gate 			 * Hat layer locking need not be done to set
25367c478bd9Sstevel@tonic-gate 			 * the following bits since the page is not hashed
25377c478bd9Sstevel@tonic-gate 			 * and was on the free list (i.e., had no mappings).
25387c478bd9Sstevel@tonic-gate 			 *
25397c478bd9Sstevel@tonic-gate 			 * Set the reference bit to protect
25407c478bd9Sstevel@tonic-gate 			 * against immediate pageout
25417c478bd9Sstevel@tonic-gate 			 *
25427c478bd9Sstevel@tonic-gate 			 * XXXmh modify freelist code to set reference
25437c478bd9Sstevel@tonic-gate 			 * bit so we don't have to do it here.
25447c478bd9Sstevel@tonic-gate 			 */
25457c478bd9Sstevel@tonic-gate 			page_set_props(pp, P_REF);
25467c478bd9Sstevel@tonic-gate 			found_on_free++;
25477c478bd9Sstevel@tonic-gate 		} else {
25487c478bd9Sstevel@tonic-gate 			VM_STAT_ADD(page_create_exists);
25497c478bd9Sstevel@tonic-gate 			if (flags & PG_EXCL) {
25507c478bd9Sstevel@tonic-gate 				/*
25517c478bd9Sstevel@tonic-gate 				 * Found an existing page, and the caller
25527c478bd9Sstevel@tonic-gate 				 * wanted all new pages.  Undo all of the work
25537c478bd9Sstevel@tonic-gate 				 * we have done.
25547c478bd9Sstevel@tonic-gate 				 */
25557c478bd9Sstevel@tonic-gate 				mutex_exit(phm);
25567c478bd9Sstevel@tonic-gate 				phm = NULL;
25577c478bd9Sstevel@tonic-gate 				while (plist != NULL) {
25587c478bd9Sstevel@tonic-gate 					pp = plist;
25597c478bd9Sstevel@tonic-gate 					page_sub(&plist, pp);
25607c478bd9Sstevel@tonic-gate 					page_io_unlock(pp);
25617c478bd9Sstevel@tonic-gate 					/* large pages should not end up here */
25627c478bd9Sstevel@tonic-gate 					ASSERT(pp->p_szc == 0);
25637c478bd9Sstevel@tonic-gate 					/*LINTED: constant in conditional ctx*/
25647c478bd9Sstevel@tonic-gate 					VN_DISPOSE(pp, B_INVAL, 0, kcred);
25657c478bd9Sstevel@tonic-gate 				}
25667c478bd9Sstevel@tonic-gate 				VM_STAT_ADD(page_create_found_one);
25677c478bd9Sstevel@tonic-gate 				goto fail;
25687c478bd9Sstevel@tonic-gate 			}
25697c478bd9Sstevel@tonic-gate 			ASSERT(flags & PG_WAIT);
25707c478bd9Sstevel@tonic-gate 			if (!page_lock(pp, SE_EXCL, phm, P_NO_RECLAIM)) {
25717c478bd9Sstevel@tonic-gate 				/*
25727c478bd9Sstevel@tonic-gate 				 * Start all over again if we blocked trying
25737c478bd9Sstevel@tonic-gate 				 * to lock the page.
25747c478bd9Sstevel@tonic-gate 				 */
25757c478bd9Sstevel@tonic-gate 				mutex_exit(phm);
25767c478bd9Sstevel@tonic-gate 				VM_STAT_ADD(page_create_page_lock_failed);
25777c478bd9Sstevel@tonic-gate 				phm = NULL;
25787c478bd9Sstevel@tonic-gate 				goto top;
25797c478bd9Sstevel@tonic-gate 			}
25807c478bd9Sstevel@tonic-gate 			mutex_exit(phm);
25817c478bd9Sstevel@tonic-gate 			phm = NULL;
25827c478bd9Sstevel@tonic-gate 
25837c478bd9Sstevel@tonic-gate 			if (PP_ISFREE(pp)) {
25847c478bd9Sstevel@tonic-gate 				ASSERT(PP_ISAGED(pp) == 0);
25857c478bd9Sstevel@tonic-gate 				VM_STAT_ADD(pagecnt.pc_get_cache);
25867c478bd9Sstevel@tonic-gate 				page_list_sub(pp, PG_CACHE_LIST);
25877c478bd9Sstevel@tonic-gate 				PP_CLRFREE(pp);
25887c478bd9Sstevel@tonic-gate 				found_on_free++;
25897c478bd9Sstevel@tonic-gate 			}
25907c478bd9Sstevel@tonic-gate 		}
25917c478bd9Sstevel@tonic-gate 
25927c478bd9Sstevel@tonic-gate 		/*
25937c478bd9Sstevel@tonic-gate 		 * Got a page!  It is locked.  Acquire the i/o
25947c478bd9Sstevel@tonic-gate 		 * lock since we are going to use the p_next and
25957c478bd9Sstevel@tonic-gate 		 * p_prev fields to link the requested pages together.
25967c478bd9Sstevel@tonic-gate 		 */
25977c478bd9Sstevel@tonic-gate 		page_io_lock(pp);
25987c478bd9Sstevel@tonic-gate 		page_add(&plist, pp);
25997c478bd9Sstevel@tonic-gate 		plist = plist->p_next;
26007c478bd9Sstevel@tonic-gate 		off += PAGESIZE;
26017c478bd9Sstevel@tonic-gate 		vaddr += PAGESIZE;
26027c478bd9Sstevel@tonic-gate 	}
26037c478bd9Sstevel@tonic-gate 
26047c478bd9Sstevel@tonic-gate 	ASSERT((flags & PG_EXCL) ? (found_on_free == pages_req) : 1);
26057c478bd9Sstevel@tonic-gate fail:
26067c478bd9Sstevel@tonic-gate 	if (npp != NULL) {
26077c478bd9Sstevel@tonic-gate 		/*
26087c478bd9Sstevel@tonic-gate 		 * Did not need this page after all.
26097c478bd9Sstevel@tonic-gate 		 * Put it back on the free list.
26107c478bd9Sstevel@tonic-gate 		 */
26117c478bd9Sstevel@tonic-gate 		VM_STAT_ADD(page_create_putbacks);
26127c478bd9Sstevel@tonic-gate 		PP_SETFREE(npp);
26137c478bd9Sstevel@tonic-gate 		PP_SETAGED(npp);
26147c478bd9Sstevel@tonic-gate 		npp->p_offset = (u_offset_t)-1;
26157c478bd9Sstevel@tonic-gate 		page_list_add(npp, PG_FREE_LIST | PG_LIST_TAIL);
26167c478bd9Sstevel@tonic-gate 		page_unlock(npp);
26177c478bd9Sstevel@tonic-gate 
26187c478bd9Sstevel@tonic-gate 	}
26197c478bd9Sstevel@tonic-gate 
26207c478bd9Sstevel@tonic-gate 	ASSERT(pages_req >= found_on_free);
26217c478bd9Sstevel@tonic-gate 
26227c478bd9Sstevel@tonic-gate 	{
26237c478bd9Sstevel@tonic-gate 		uint_t overshoot = (uint_t)(pages_req - found_on_free);
26247c478bd9Sstevel@tonic-gate 
26257c478bd9Sstevel@tonic-gate 		if (overshoot) {
26267c478bd9Sstevel@tonic-gate 			VM_STAT_ADD(page_create_overshoot);
26277c478bd9Sstevel@tonic-gate 			p = &pcf[pcf_index];
26287c478bd9Sstevel@tonic-gate 			mutex_enter(&p->pcf_lock);
26297c478bd9Sstevel@tonic-gate 			if (p->pcf_block) {
26307c478bd9Sstevel@tonic-gate 				p->pcf_reserve += overshoot;
26317c478bd9Sstevel@tonic-gate 			} else {
26327c478bd9Sstevel@tonic-gate 				p->pcf_count += overshoot;
26337c478bd9Sstevel@tonic-gate 				if (p->pcf_wait) {
26347c478bd9Sstevel@tonic-gate 					mutex_enter(&new_freemem_lock);
26357c478bd9Sstevel@tonic-gate 					if (freemem_wait) {
26367c478bd9Sstevel@tonic-gate 						cv_signal(&freemem_cv);
26377c478bd9Sstevel@tonic-gate 						p->pcf_wait--;
26387c478bd9Sstevel@tonic-gate 					} else {
26397c478bd9Sstevel@tonic-gate 						p->pcf_wait = 0;
26407c478bd9Sstevel@tonic-gate 					}
26417c478bd9Sstevel@tonic-gate 					mutex_exit(&new_freemem_lock);
26427c478bd9Sstevel@tonic-gate 				}
26437c478bd9Sstevel@tonic-gate 			}
26447c478bd9Sstevel@tonic-gate 			mutex_exit(&p->pcf_lock);
26457c478bd9Sstevel@tonic-gate 			/* freemem is approximate, so this test OK */
26467c478bd9Sstevel@tonic-gate 			if (!p->pcf_block)
26477c478bd9Sstevel@tonic-gate 				freemem += overshoot;
26487c478bd9Sstevel@tonic-gate 		}
26497c478bd9Sstevel@tonic-gate 	}
26507c478bd9Sstevel@tonic-gate 
26517c478bd9Sstevel@tonic-gate 	return (plist);
26527c478bd9Sstevel@tonic-gate }
26537c478bd9Sstevel@tonic-gate 
26547c478bd9Sstevel@tonic-gate /*
26557c478bd9Sstevel@tonic-gate  * One or more constituent pages of this large page has been marked
26567c478bd9Sstevel@tonic-gate  * toxic. Simply demote the large page to PAGESIZE pages and let
26577c478bd9Sstevel@tonic-gate  * page_free() handle it. This routine should only be called by
26587c478bd9Sstevel@tonic-gate  * large page free routines (page_free_pages() and page_destroy_pages().
26597c478bd9Sstevel@tonic-gate  * All pages are locked SE_EXCL and have already been marked free.
26607c478bd9Sstevel@tonic-gate  */
26617c478bd9Sstevel@tonic-gate static void
26627c478bd9Sstevel@tonic-gate page_free_toxic_pages(page_t *rootpp)
26637c478bd9Sstevel@tonic-gate {
26647c478bd9Sstevel@tonic-gate 	page_t	*tpp;
26657c478bd9Sstevel@tonic-gate 	pgcnt_t	i, pgcnt = page_get_pagecnt(rootpp->p_szc);
26667c478bd9Sstevel@tonic-gate 	uint_t	szc = rootpp->p_szc;
26677c478bd9Sstevel@tonic-gate 
26687c478bd9Sstevel@tonic-gate 	for (i = 0, tpp = rootpp; i < pgcnt; i++, tpp = tpp->p_next) {
26697c478bd9Sstevel@tonic-gate 		ASSERT(tpp->p_szc == szc);
26707c478bd9Sstevel@tonic-gate 		ASSERT((PAGE_EXCL(tpp) &&
26717c478bd9Sstevel@tonic-gate 		    !page_iolock_assert(tpp)) || panicstr);
26727c478bd9Sstevel@tonic-gate 		tpp->p_szc = 0;
26737c478bd9Sstevel@tonic-gate 	}
26747c478bd9Sstevel@tonic-gate 
26757c478bd9Sstevel@tonic-gate 	while (rootpp != NULL) {
26767c478bd9Sstevel@tonic-gate 		tpp = rootpp;
26777c478bd9Sstevel@tonic-gate 		page_sub(&rootpp, tpp);
26787c478bd9Sstevel@tonic-gate 		ASSERT(PP_ISFREE(tpp));
26797c478bd9Sstevel@tonic-gate 		PP_CLRFREE(tpp);
26807c478bd9Sstevel@tonic-gate 		page_free(tpp, 1);
26817c478bd9Sstevel@tonic-gate 	}
26827c478bd9Sstevel@tonic-gate }
26837c478bd9Sstevel@tonic-gate 
26847c478bd9Sstevel@tonic-gate /*
26857c478bd9Sstevel@tonic-gate  * Put page on the "free" list.
26867c478bd9Sstevel@tonic-gate  * The free list is really two lists maintained by
26877c478bd9Sstevel@tonic-gate  * the PSM of whatever machine we happen to be on.
26887c478bd9Sstevel@tonic-gate  */
26897c478bd9Sstevel@tonic-gate void
26907c478bd9Sstevel@tonic-gate page_free(page_t *pp, int dontneed)
26917c478bd9Sstevel@tonic-gate {
26927c478bd9Sstevel@tonic-gate 	struct pcf	*p;
26937c478bd9Sstevel@tonic-gate 	uint_t		pcf_index;
26947c478bd9Sstevel@tonic-gate 
26957c478bd9Sstevel@tonic-gate 	ASSERT((PAGE_EXCL(pp) &&
26967c478bd9Sstevel@tonic-gate 	    !page_iolock_assert(pp)) || panicstr);
26977c478bd9Sstevel@tonic-gate 
26987c478bd9Sstevel@tonic-gate 	if (PP_ISFREE(pp)) {
26997c478bd9Sstevel@tonic-gate 		panic("page_free: page %p is free", (void *)pp);
27007c478bd9Sstevel@tonic-gate 	}
27017c478bd9Sstevel@tonic-gate 
27027c478bd9Sstevel@tonic-gate 	if (pp->p_szc != 0) {
27037c478bd9Sstevel@tonic-gate 		if (pp->p_vnode == NULL || IS_SWAPFSVP(pp->p_vnode) ||
27047c478bd9Sstevel@tonic-gate 		    pp->p_vnode == &kvp) {
27057c478bd9Sstevel@tonic-gate 			panic("page_free: anon or kernel "
27067c478bd9Sstevel@tonic-gate 			    "or no vnode large page %p", (void *)pp);
27077c478bd9Sstevel@tonic-gate 		}
27087c478bd9Sstevel@tonic-gate 		page_demote_vp_pages(pp);
27097c478bd9Sstevel@tonic-gate 		ASSERT(pp->p_szc == 0);
27107c478bd9Sstevel@tonic-gate 	}
27117c478bd9Sstevel@tonic-gate 
27127c478bd9Sstevel@tonic-gate 	/*
27137c478bd9Sstevel@tonic-gate 	 * The page_struct_lock need not be acquired to examine these
27147c478bd9Sstevel@tonic-gate 	 * fields since the page has an "exclusive" lock.
27157c478bd9Sstevel@tonic-gate 	 */
271607b65a64Saguzovsk 	if (hat_page_is_mapped(pp) || pp->p_lckcnt != 0 || pp->p_cowcnt != 0 ||
271707b65a64Saguzovsk 	    pp->p_slckcnt != 0) {
271807b65a64Saguzovsk 		panic("page_free pp=%p, pfn=%lx, lckcnt=%d, cowcnt=%d "
271907b65a64Saguzovsk 		    "slckcnt = %d", pp, page_pptonum(pp), pp->p_lckcnt,
272007b65a64Saguzovsk 		    pp->p_cowcnt, pp->p_slckcnt);
27217c478bd9Sstevel@tonic-gate 		/*NOTREACHED*/
27227c478bd9Sstevel@tonic-gate 	}
27237c478bd9Sstevel@tonic-gate 
27247c478bd9Sstevel@tonic-gate 	ASSERT(!hat_page_getshare(pp));
27257c478bd9Sstevel@tonic-gate 
27267c478bd9Sstevel@tonic-gate 	PP_SETFREE(pp);
27277c478bd9Sstevel@tonic-gate 	ASSERT(pp->p_vnode == NULL || !IS_VMODSORT(pp->p_vnode) ||
27287c478bd9Sstevel@tonic-gate 	    !hat_ismod(pp));
27297c478bd9Sstevel@tonic-gate 	page_clr_all_props(pp);
27307c478bd9Sstevel@tonic-gate 	ASSERT(!hat_page_getshare(pp));
27317c478bd9Sstevel@tonic-gate 
27327c478bd9Sstevel@tonic-gate 	/*
27337c478bd9Sstevel@tonic-gate 	 * Now we add the page to the head of the free list.
27347c478bd9Sstevel@tonic-gate 	 * But if this page is associated with a paged vnode
27357c478bd9Sstevel@tonic-gate 	 * then we adjust the head forward so that the page is
27367c478bd9Sstevel@tonic-gate 	 * effectively at the end of the list.
27377c478bd9Sstevel@tonic-gate 	 */
27387c478bd9Sstevel@tonic-gate 	if (pp->p_vnode == NULL) {
27397c478bd9Sstevel@tonic-gate 		/*
27407c478bd9Sstevel@tonic-gate 		 * Page has no identity, put it on the free list.
27417c478bd9Sstevel@tonic-gate 		 */
27427c478bd9Sstevel@tonic-gate 		PP_SETAGED(pp);
27437c478bd9Sstevel@tonic-gate 		pp->p_offset = (u_offset_t)-1;
27447c478bd9Sstevel@tonic-gate 		page_list_add(pp, PG_FREE_LIST | PG_LIST_TAIL);
27457c478bd9Sstevel@tonic-gate 		VM_STAT_ADD(pagecnt.pc_free_free);
27467c478bd9Sstevel@tonic-gate 		TRACE_1(TR_FAC_VM, TR_PAGE_FREE_FREE,
27477c478bd9Sstevel@tonic-gate 		    "page_free_free:pp %p", pp);
27487c478bd9Sstevel@tonic-gate 	} else {
27497c478bd9Sstevel@tonic-gate 		PP_CLRAGED(pp);
27507c478bd9Sstevel@tonic-gate 
27517c478bd9Sstevel@tonic-gate 		if (!dontneed || nopageage) {
27527c478bd9Sstevel@tonic-gate 			/* move it to the tail of the list */
27537c478bd9Sstevel@tonic-gate 			page_list_add(pp, PG_CACHE_LIST | PG_LIST_TAIL);
27547c478bd9Sstevel@tonic-gate 
27557c478bd9Sstevel@tonic-gate 			VM_STAT_ADD(pagecnt.pc_free_cache);
27567c478bd9Sstevel@tonic-gate 			TRACE_1(TR_FAC_VM, TR_PAGE_FREE_CACHE_TAIL,
27577c478bd9Sstevel@tonic-gate 			    "page_free_cache_tail:pp %p", pp);
27587c478bd9Sstevel@tonic-gate 		} else {
27597c478bd9Sstevel@tonic-gate 			page_list_add(pp, PG_CACHE_LIST | PG_LIST_HEAD);
27607c478bd9Sstevel@tonic-gate 
27617c478bd9Sstevel@tonic-gate 			VM_STAT_ADD(pagecnt.pc_free_dontneed);
27627c478bd9Sstevel@tonic-gate 			TRACE_1(TR_FAC_VM, TR_PAGE_FREE_CACHE_HEAD,
27637c478bd9Sstevel@tonic-gate 			    "page_free_cache_head:pp %p", pp);
27647c478bd9Sstevel@tonic-gate 		}
27657c478bd9Sstevel@tonic-gate 	}
27667c478bd9Sstevel@tonic-gate 	page_unlock(pp);
27677c478bd9Sstevel@tonic-gate 
27687c478bd9Sstevel@tonic-gate 	/*
27697c478bd9Sstevel@tonic-gate 	 * Now do the `freemem' accounting.
27707c478bd9Sstevel@tonic-gate 	 */
27717c478bd9Sstevel@tonic-gate 	pcf_index = PCF_INDEX();
27727c478bd9Sstevel@tonic-gate 	p = &pcf[pcf_index];
27737c478bd9Sstevel@tonic-gate 
27747c478bd9Sstevel@tonic-gate 	mutex_enter(&p->pcf_lock);
27757c478bd9Sstevel@tonic-gate 	if (p->pcf_block) {
27767c478bd9Sstevel@tonic-gate 		p->pcf_reserve += 1;
27777c478bd9Sstevel@tonic-gate 	} else {
27787c478bd9Sstevel@tonic-gate 		p->pcf_count += 1;
27797c478bd9Sstevel@tonic-gate 		if (p->pcf_wait) {
27807c478bd9Sstevel@tonic-gate 			mutex_enter(&new_freemem_lock);
27817c478bd9Sstevel@tonic-gate 			/*
27827c478bd9Sstevel@tonic-gate 			 * Check to see if some other thread
27837c478bd9Sstevel@tonic-gate 			 * is actually waiting.  Another bucket
27847c478bd9Sstevel@tonic-gate 			 * may have woken it up by now.  If there
27857c478bd9Sstevel@tonic-gate 			 * are no waiters, then set our pcf_wait
27867c478bd9Sstevel@tonic-gate 			 * count to zero to avoid coming in here
27877c478bd9Sstevel@tonic-gate 			 * next time.  Also, since only one page
27887c478bd9Sstevel@tonic-gate 			 * was put on the free list, just wake
27897c478bd9Sstevel@tonic-gate 			 * up one waiter.
27907c478bd9Sstevel@tonic-gate 			 */
27917c478bd9Sstevel@tonic-gate 			if (freemem_wait) {
27927c478bd9Sstevel@tonic-gate 				cv_signal(&freemem_cv);
27937c478bd9Sstevel@tonic-gate 				p->pcf_wait--;
27947c478bd9Sstevel@tonic-gate 			} else {
27957c478bd9Sstevel@tonic-gate 				p->pcf_wait = 0;
27967c478bd9Sstevel@tonic-gate 			}
27977c478bd9Sstevel@tonic-gate 			mutex_exit(&new_freemem_lock);
27987c478bd9Sstevel@tonic-gate 		}
27997c478bd9Sstevel@tonic-gate 	}
28007c478bd9Sstevel@tonic-gate 	mutex_exit(&p->pcf_lock);
28017c478bd9Sstevel@tonic-gate 
28027c478bd9Sstevel@tonic-gate 	/* freemem is approximate, so this test OK */
28037c478bd9Sstevel@tonic-gate 	if (!p->pcf_block)
28047c478bd9Sstevel@tonic-gate 		freemem += 1;
28057c478bd9Sstevel@tonic-gate }
28067c478bd9Sstevel@tonic-gate 
28077c478bd9Sstevel@tonic-gate /*
28087c478bd9Sstevel@tonic-gate  * Put page on the "free" list during intial startup.
28097c478bd9Sstevel@tonic-gate  * This happens during initial single threaded execution.
28107c478bd9Sstevel@tonic-gate  */
28117c478bd9Sstevel@tonic-gate void
28127c478bd9Sstevel@tonic-gate page_free_at_startup(page_t *pp)
28137c478bd9Sstevel@tonic-gate {
28147c478bd9Sstevel@tonic-gate 	struct pcf	*p;
28157c478bd9Sstevel@tonic-gate 	uint_t		pcf_index;
28167c478bd9Sstevel@tonic-gate 
28177c478bd9Sstevel@tonic-gate 	page_list_add(pp, PG_FREE_LIST | PG_LIST_HEAD | PG_LIST_ISINIT);
28187c478bd9Sstevel@tonic-gate 	VM_STAT_ADD(pagecnt.pc_free_free);
28197c478bd9Sstevel@tonic-gate 
28207c478bd9Sstevel@tonic-gate 	/*
28217c478bd9Sstevel@tonic-gate 	 * Now do the `freemem' accounting.
28227c478bd9Sstevel@tonic-gate 	 */
28237c478bd9Sstevel@tonic-gate 	pcf_index = PCF_INDEX();
28247c478bd9Sstevel@tonic-gate 	p = &pcf[pcf_index];
28257c478bd9Sstevel@tonic-gate 
28267c478bd9Sstevel@tonic-gate 	ASSERT(p->pcf_block == 0);
28277c478bd9Sstevel@tonic-gate 	ASSERT(p->pcf_wait == 0);
28287c478bd9Sstevel@tonic-gate 	p->pcf_count += 1;
28297c478bd9Sstevel@tonic-gate 
28307c478bd9Sstevel@tonic-gate 	/* freemem is approximate, so this is OK */
28317c478bd9Sstevel@tonic-gate 	freemem += 1;
28327c478bd9Sstevel@tonic-gate }
28337c478bd9Sstevel@tonic-gate 
28347c478bd9Sstevel@tonic-gate void
28357c478bd9Sstevel@tonic-gate page_free_pages(page_t *pp)
28367c478bd9Sstevel@tonic-gate {
28377c478bd9Sstevel@tonic-gate 	page_t	*tpp, *rootpp = NULL;
28387c478bd9Sstevel@tonic-gate 	pgcnt_t	pgcnt = page_get_pagecnt(pp->p_szc);
28397c478bd9Sstevel@tonic-gate 	pgcnt_t	i;
28407c478bd9Sstevel@tonic-gate 	uint_t	szc = pp->p_szc;
28417c478bd9Sstevel@tonic-gate 
28427c478bd9Sstevel@tonic-gate 	VM_STAT_ADD(pagecnt.pc_free_pages);
28437c478bd9Sstevel@tonic-gate 	TRACE_1(TR_FAC_VM, TR_PAGE_FREE_FREE,
28447c478bd9Sstevel@tonic-gate 	    "page_free_free:pp %p", pp);
28457c478bd9Sstevel@tonic-gate 
28467c478bd9Sstevel@tonic-gate 	ASSERT(pp->p_szc != 0 && pp->p_szc < page_num_pagesizes());
28477c478bd9Sstevel@tonic-gate 	if ((page_pptonum(pp) & (pgcnt - 1)) != 0) {
28487c478bd9Sstevel@tonic-gate 		panic("page_free_pages: not root page %p", (void *)pp);
28497c478bd9Sstevel@tonic-gate 		/*NOTREACHED*/
28507c478bd9Sstevel@tonic-gate 	}
28517c478bd9Sstevel@tonic-gate 
2852affbd3ccSkchow 	for (i = 0, tpp = pp; i < pgcnt; i++, tpp++) {
28537c478bd9Sstevel@tonic-gate 		ASSERT((PAGE_EXCL(tpp) &&
28547c478bd9Sstevel@tonic-gate 		    !page_iolock_assert(tpp)) || panicstr);
28557c478bd9Sstevel@tonic-gate 		if (PP_ISFREE(tpp)) {
28567c478bd9Sstevel@tonic-gate 			panic("page_free_pages: page %p is free", (void *)tpp);
28577c478bd9Sstevel@tonic-gate 			/*NOTREACHED*/
28587c478bd9Sstevel@tonic-gate 		}
28597c478bd9Sstevel@tonic-gate 		if (hat_page_is_mapped(tpp) || tpp->p_lckcnt != 0 ||
286007b65a64Saguzovsk 		    tpp->p_cowcnt != 0 || tpp->p_slckcnt != 0) {
28617c478bd9Sstevel@tonic-gate 			panic("page_free_pages %p", (void *)tpp);
28627c478bd9Sstevel@tonic-gate 			/*NOTREACHED*/
28637c478bd9Sstevel@tonic-gate 		}
28647c478bd9Sstevel@tonic-gate 
28657c478bd9Sstevel@tonic-gate 		ASSERT(!hat_page_getshare(tpp));
28667c478bd9Sstevel@tonic-gate 		ASSERT(tpp->p_vnode == NULL);
28677c478bd9Sstevel@tonic-gate 		ASSERT(tpp->p_szc == szc);
28687c478bd9Sstevel@tonic-gate 
28697c478bd9Sstevel@tonic-gate 		PP_SETFREE(tpp);
28707c478bd9Sstevel@tonic-gate 		page_clr_all_props(tpp);
28717c478bd9Sstevel@tonic-gate 		PP_SETAGED(tpp);
28727c478bd9Sstevel@tonic-gate 		tpp->p_offset = (u_offset_t)-1;
28737c478bd9Sstevel@tonic-gate 		ASSERT(tpp->p_next == tpp);
28747c478bd9Sstevel@tonic-gate 		ASSERT(tpp->p_prev == tpp);
28757c478bd9Sstevel@tonic-gate 		page_list_concat(&rootpp, &tpp);
28767c478bd9Sstevel@tonic-gate 	}
28777c478bd9Sstevel@tonic-gate 	ASSERT(rootpp == pp);
28787c478bd9Sstevel@tonic-gate 
28797c478bd9Sstevel@tonic-gate 	page_list_add_pages(rootpp, 0);
28807c478bd9Sstevel@tonic-gate 	page_create_putback(pgcnt);
28817c478bd9Sstevel@tonic-gate }
28827c478bd9Sstevel@tonic-gate 
28837c478bd9Sstevel@tonic-gate int free_pages = 1;
28847c478bd9Sstevel@tonic-gate 
28857c478bd9Sstevel@tonic-gate /*
28867c478bd9Sstevel@tonic-gate  * This routine attempts to return pages to the cachelist via page_release().
28877c478bd9Sstevel@tonic-gate  * It does not *have* to be successful in all cases, since the pageout scanner
28887c478bd9Sstevel@tonic-gate  * will catch any pages it misses.  It does need to be fast and not introduce
28897c478bd9Sstevel@tonic-gate  * too much overhead.
28907c478bd9Sstevel@tonic-gate  *
28917c478bd9Sstevel@tonic-gate  * If a page isn't found on the unlocked sweep of the page_hash bucket, we
28927c478bd9Sstevel@tonic-gate  * don't lock and retry.  This is ok, since the page scanner will eventually
28937c478bd9Sstevel@tonic-gate  * find any page we miss in free_vp_pages().
28947c478bd9Sstevel@tonic-gate  */
28957c478bd9Sstevel@tonic-gate void
28967c478bd9Sstevel@tonic-gate free_vp_pages(vnode_t *vp, u_offset_t off, size_t len)
28977c478bd9Sstevel@tonic-gate {
28987c478bd9Sstevel@tonic-gate 	page_t *pp;
28997c478bd9Sstevel@tonic-gate 	u_offset_t eoff;
29007c478bd9Sstevel@tonic-gate 	extern int swap_in_range(vnode_t *, u_offset_t, size_t);
29017c478bd9Sstevel@tonic-gate 
29027c478bd9Sstevel@tonic-gate 	eoff = off + len;
29037c478bd9Sstevel@tonic-gate 
29047c478bd9Sstevel@tonic-gate 	if (free_pages == 0)
29057c478bd9Sstevel@tonic-gate 		return;
29067c478bd9Sstevel@tonic-gate 	if (swap_in_range(vp, off, len))
29077c478bd9Sstevel@tonic-gate 		return;
29087c478bd9Sstevel@tonic-gate 
29097c478bd9Sstevel@tonic-gate 	for (; off < eoff; off += PAGESIZE) {
29107c478bd9Sstevel@tonic-gate 
29117c478bd9Sstevel@tonic-gate 		/*
29127c478bd9Sstevel@tonic-gate 		 * find the page using a fast, but inexact search. It'll be OK
29137c478bd9Sstevel@tonic-gate 		 * if a few pages slip through the cracks here.
29147c478bd9Sstevel@tonic-gate 		 */
29157c478bd9Sstevel@tonic-gate 		pp = page_exists(vp, off);
29167c478bd9Sstevel@tonic-gate 
29177c478bd9Sstevel@tonic-gate 		/*
29187c478bd9Sstevel@tonic-gate 		 * If we didn't find the page (it may not exist), the page
29197c478bd9Sstevel@tonic-gate 		 * is free, looks still in use (shared), or we can't lock it,
29207c478bd9Sstevel@tonic-gate 		 * just give up.
29217c478bd9Sstevel@tonic-gate 		 */
29227c478bd9Sstevel@tonic-gate 		if (pp == NULL ||
29237c478bd9Sstevel@tonic-gate 		    PP_ISFREE(pp) ||
29247c478bd9Sstevel@tonic-gate 		    page_share_cnt(pp) > 0 ||
29257c478bd9Sstevel@tonic-gate 		    !page_trylock(pp, SE_EXCL))
29267c478bd9Sstevel@tonic-gate 			continue;
29277c478bd9Sstevel@tonic-gate 
29287c478bd9Sstevel@tonic-gate 		/*
29297c478bd9Sstevel@tonic-gate 		 * Once we have locked pp, verify that it's still the
29307c478bd9Sstevel@tonic-gate 		 * correct page and not already free
29317c478bd9Sstevel@tonic-gate 		 */
29327c478bd9Sstevel@tonic-gate 		ASSERT(PAGE_LOCKED_SE(pp, SE_EXCL));
29337c478bd9Sstevel@tonic-gate 		if (pp->p_vnode != vp || pp->p_offset != off || PP_ISFREE(pp)) {
29347c478bd9Sstevel@tonic-gate 			page_unlock(pp);
29357c478bd9Sstevel@tonic-gate 			continue;
29367c478bd9Sstevel@tonic-gate 		}
29377c478bd9Sstevel@tonic-gate 
29387c478bd9Sstevel@tonic-gate 		/*
29397c478bd9Sstevel@tonic-gate 		 * try to release the page...
29407c478bd9Sstevel@tonic-gate 		 */
29417c478bd9Sstevel@tonic-gate 		(void) page_release(pp, 1);
29427c478bd9Sstevel@tonic-gate 	}
29437c478bd9Sstevel@tonic-gate }
29447c478bd9Sstevel@tonic-gate 
29457c478bd9Sstevel@tonic-gate /*
29467c478bd9Sstevel@tonic-gate  * Reclaim the given page from the free list.
29477c478bd9Sstevel@tonic-gate  * Returns 1 on success or 0 on failure.
29487c478bd9Sstevel@tonic-gate  *
29497c478bd9Sstevel@tonic-gate  * The page is unlocked if it can't be reclaimed (when freemem == 0).
29507c478bd9Sstevel@tonic-gate  * If `lock' is non-null, it will be dropped and re-acquired if
29517c478bd9Sstevel@tonic-gate  * the routine must wait while freemem is 0.
29527c478bd9Sstevel@tonic-gate  *
29537c478bd9Sstevel@tonic-gate  * As it turns out, boot_getpages() does this.  It picks a page,
29547c478bd9Sstevel@tonic-gate  * based on where OBP mapped in some address, gets its pfn, searches
29557c478bd9Sstevel@tonic-gate  * the memsegs, locks the page, then pulls it off the free list!
29567c478bd9Sstevel@tonic-gate  */
29577c478bd9Sstevel@tonic-gate int
29587c478bd9Sstevel@tonic-gate page_reclaim(page_t *pp, kmutex_t *lock)
29597c478bd9Sstevel@tonic-gate {
29607c478bd9Sstevel@tonic-gate 	struct pcf	*p;
29617c478bd9Sstevel@tonic-gate 	uint_t		pcf_index;
29627c478bd9Sstevel@tonic-gate 	struct cpu	*cpup;
29637c478bd9Sstevel@tonic-gate 	uint_t		i;
2964de62167aSmec 	pgcnt_t		npgs, need;
2965de62167aSmec 	pgcnt_t		collected = 0;
29667c478bd9Sstevel@tonic-gate 
29677c478bd9Sstevel@tonic-gate 	ASSERT(lock != NULL ? MUTEX_HELD(lock) : 1);
29687c478bd9Sstevel@tonic-gate 	ASSERT(PAGE_EXCL(pp) && PP_ISFREE(pp));
2969db874c57Selowe 
2970db874c57Selowe 	npgs = page_get_pagecnt(pp->p_szc);
29717c478bd9Sstevel@tonic-gate 
29727c478bd9Sstevel@tonic-gate 	/*
29737c478bd9Sstevel@tonic-gate 	 * If `freemem' is 0, we cannot reclaim this page from the
29747c478bd9Sstevel@tonic-gate 	 * freelist, so release every lock we might hold: the page,
29757c478bd9Sstevel@tonic-gate 	 * and the `lock' before blocking.
29767c478bd9Sstevel@tonic-gate 	 *
29777c478bd9Sstevel@tonic-gate 	 * The only way `freemem' can become 0 while there are pages
29787c478bd9Sstevel@tonic-gate 	 * marked free (have their p->p_free bit set) is when the
29797c478bd9Sstevel@tonic-gate 	 * system is low on memory and doing a page_create().  In
29807c478bd9Sstevel@tonic-gate 	 * order to guarantee that once page_create() starts acquiring
29817c478bd9Sstevel@tonic-gate 	 * pages it will be able to get all that it needs since `freemem'
29827c478bd9Sstevel@tonic-gate 	 * was decreased by the requested amount.  So, we need to release
29837c478bd9Sstevel@tonic-gate 	 * this page, and let page_create() have it.
29847c478bd9Sstevel@tonic-gate 	 *
29857c478bd9Sstevel@tonic-gate 	 * Since `freemem' being zero is not supposed to happen, just
29867c478bd9Sstevel@tonic-gate 	 * use the usual hash stuff as a starting point.  If that bucket
29877c478bd9Sstevel@tonic-gate 	 * is empty, then assume the worst, and start at the beginning
29887c478bd9Sstevel@tonic-gate 	 * of the pcf array.  If we always start at the beginning
29897c478bd9Sstevel@tonic-gate 	 * when acquiring more than one pcf lock, there won't be any
29907c478bd9Sstevel@tonic-gate 	 * deadlock problems.
29917c478bd9Sstevel@tonic-gate 	 */
29927c478bd9Sstevel@tonic-gate 
29937c478bd9Sstevel@tonic-gate 	/* TODO: Do we need to test kcage_freemem if PG_NORELOC(pp)? */
29947c478bd9Sstevel@tonic-gate 
299524e9c58bSelowe 	if (freemem <= throttlefree && !page_create_throttle(npgs, 0)) {
29967c478bd9Sstevel@tonic-gate 		pcf_acquire_all();
29977c478bd9Sstevel@tonic-gate 		goto page_reclaim_nomem;
29987c478bd9Sstevel@tonic-gate 	}
29997c478bd9Sstevel@tonic-gate 
30007c478bd9Sstevel@tonic-gate 	pcf_index = PCF_INDEX();
30017c478bd9Sstevel@tonic-gate 	p = &pcf[pcf_index];
30027c478bd9Sstevel@tonic-gate 	mutex_enter(&p->pcf_lock);
3003db874c57Selowe 	if (p->pcf_count >= npgs) {
3004db874c57Selowe 		collected = npgs;
3005db874c57Selowe 		p->pcf_count -= npgs;
30067c478bd9Sstevel@tonic-gate 	}
30077c478bd9Sstevel@tonic-gate 	mutex_exit(&p->pcf_lock);
3008db874c57Selowe 	need = npgs - collected;
30097c478bd9Sstevel@tonic-gate 
3010db874c57Selowe 	if (need > 0) {
30117c478bd9Sstevel@tonic-gate 		VM_STAT_ADD(page_reclaim_zero);
30127c478bd9Sstevel@tonic-gate 		/*
30137c478bd9Sstevel@tonic-gate 		 * Check again. Its possible that some other thread
30147c478bd9Sstevel@tonic-gate 		 * could have been right behind us, and added one
30157c478bd9Sstevel@tonic-gate 		 * to a list somewhere.  Acquire each of the pcf locks
30167c478bd9Sstevel@tonic-gate 		 * until we find a page.
30177c478bd9Sstevel@tonic-gate 		 */
30187c478bd9Sstevel@tonic-gate 		p = pcf;
30197c478bd9Sstevel@tonic-gate 		for (i = 0; i < PCF_FANOUT; i++) {
30207c478bd9Sstevel@tonic-gate 			mutex_enter(&p->pcf_lock);
3021db874c57Selowe 			if (p->pcf_count) {
3022db874c57Selowe 				if (p->pcf_count >= need) {
3023db874c57Selowe 					p->pcf_count -= need;
3024db874c57Selowe 					collected += need;
3025db874c57Selowe 					need = 0;
3026db874c57Selowe 					break;
3027db874c57Selowe 				} else if (p->pcf_count) {
3028db874c57Selowe 					collected += p->pcf_count;
3029db874c57Selowe 					need -= p->pcf_count;
3030db874c57Selowe 					p->pcf_count = 0;
3031db874c57Selowe 				}
30327c478bd9Sstevel@tonic-gate 			}
30337c478bd9Sstevel@tonic-gate 			p++;
30347c478bd9Sstevel@tonic-gate 		}
30357c478bd9Sstevel@tonic-gate 
3036db874c57Selowe 		if (need > 0) {
30377c478bd9Sstevel@tonic-gate page_reclaim_nomem:
30387c478bd9Sstevel@tonic-gate 			/*
30397c478bd9Sstevel@tonic-gate 			 * We really can't have page `pp'.
30407c478bd9Sstevel@tonic-gate 			 * Time for the no-memory dance with
30417c478bd9Sstevel@tonic-gate 			 * page_free().  This is just like
30427c478bd9Sstevel@tonic-gate 			 * page_create_wait().  Plus the added
30437c478bd9Sstevel@tonic-gate 			 * attraction of releasing whatever mutex
30447c478bd9Sstevel@tonic-gate 			 * we held when we were called with in `lock'.
30457c478bd9Sstevel@tonic-gate 			 * Page_unlock() will wakeup any thread
30467c478bd9Sstevel@tonic-gate 			 * waiting around for this page.
30477c478bd9Sstevel@tonic-gate 			 */
30487c478bd9Sstevel@tonic-gate 			if (lock) {
30497c478bd9Sstevel@tonic-gate 				VM_STAT_ADD(page_reclaim_zero_locked);
30507c478bd9Sstevel@tonic-gate 				mutex_exit(lock);
30517c478bd9Sstevel@tonic-gate 			}
30527c478bd9Sstevel@tonic-gate 			page_unlock(pp);
30537c478bd9Sstevel@tonic-gate 
30547c478bd9Sstevel@tonic-gate 			/*
30557c478bd9Sstevel@tonic-gate 			 * get this before we drop all the pcf locks.
30567c478bd9Sstevel@tonic-gate 			 */
30577c478bd9Sstevel@tonic-gate 			mutex_enter(&new_freemem_lock);
30587c478bd9Sstevel@tonic-gate 
30597c478bd9Sstevel@tonic-gate 			p = pcf;
3060db874c57Selowe 			p->pcf_count += collected;
30617c478bd9Sstevel@tonic-gate 			for (i = 0; i < PCF_FANOUT; i++) {
30627c478bd9Sstevel@tonic-gate 				p->pcf_wait++;
30637c478bd9Sstevel@tonic-gate 				mutex_exit(&p->pcf_lock);
30647c478bd9Sstevel@tonic-gate 				p++;
30657c478bd9Sstevel@tonic-gate 			}
30667c478bd9Sstevel@tonic-gate 
30677c478bd9Sstevel@tonic-gate 			freemem_wait++;
30687c478bd9Sstevel@tonic-gate 			cv_wait(&freemem_cv, &new_freemem_lock);
30697c478bd9Sstevel@tonic-gate 			freemem_wait--;
30707c478bd9Sstevel@tonic-gate 
30717c478bd9Sstevel@tonic-gate 			mutex_exit(&new_freemem_lock);
30727c478bd9Sstevel@tonic-gate 
30737c478bd9Sstevel@tonic-gate 			if (lock) {
30747c478bd9Sstevel@tonic-gate 				mutex_enter(lock);
30757c478bd9Sstevel@tonic-gate 			}
30767c478bd9Sstevel@tonic-gate 			return (0);
30777c478bd9Sstevel@tonic-gate 		}
30787c478bd9Sstevel@tonic-gate 
30797c478bd9Sstevel@tonic-gate 		/*
3080db874c57Selowe 		 * We beat the PCF bins over the head until
3081db874c57Selowe 		 * we got the memory that we wanted.
30827c478bd9Sstevel@tonic-gate 		 * The pcf accounting has been done,
30837c478bd9Sstevel@tonic-gate 		 * though none of the pcf_wait flags have been set,
30847c478bd9Sstevel@tonic-gate 		 * drop the locks and continue on.
30857c478bd9Sstevel@tonic-gate 		 */
3086db874c57Selowe 		ASSERT(collected == npgs);
30877c478bd9Sstevel@tonic-gate 		while (p >= pcf) {
30887c478bd9Sstevel@tonic-gate 			mutex_exit(&p->pcf_lock);
30897c478bd9Sstevel@tonic-gate 			p--;
30907c478bd9Sstevel@tonic-gate 		}
30917c478bd9Sstevel@tonic-gate 	}
30927c478bd9Sstevel@tonic-gate 
30937c478bd9Sstevel@tonic-gate 	/*
30947c478bd9Sstevel@tonic-gate 	 * freemem is not protected by any lock. Thus, we cannot
30957c478bd9Sstevel@tonic-gate 	 * have any assertion containing freemem here.
30967c478bd9Sstevel@tonic-gate 	 */
3097db874c57Selowe 	freemem -= npgs;
30987c478bd9Sstevel@tonic-gate 
30997c478bd9Sstevel@tonic-gate 	VM_STAT_ADD(pagecnt.pc_reclaim);
31007c478bd9Sstevel@tonic-gate 	if (PP_ISAGED(pp)) {
3101db874c57Selowe 		if (npgs > 1) {
3102db874c57Selowe 			page_list_sub_pages(pp, pp->p_szc);
3103db874c57Selowe 		} else {
3104db874c57Selowe 			page_list_sub(pp, PG_FREE_LIST);
3105db874c57Selowe 		}
31067c478bd9Sstevel@tonic-gate 		TRACE_1(TR_FAC_VM, TR_PAGE_UNFREE_FREE,
31077c478bd9Sstevel@tonic-gate 		    "page_reclaim_free:pp %p", pp);
31087c478bd9Sstevel@tonic-gate 	} else {
3109db874c57Selowe 		ASSERT(npgs == 1);
31107c478bd9Sstevel@tonic-gate 		page_list_sub(pp, PG_CACHE_LIST);
31117c478bd9Sstevel@tonic-gate 		TRACE_1(TR_FAC_VM, TR_PAGE_UNFREE_CACHE,
31127c478bd9Sstevel@tonic-gate 		    "page_reclaim_cache:pp %p", pp);
31137c478bd9Sstevel@tonic-gate 	}
31147c478bd9Sstevel@tonic-gate 
31157c478bd9Sstevel@tonic-gate 	/*
31167c478bd9Sstevel@tonic-gate 	 * clear the p_free & p_age bits since this page is no longer
31177c478bd9Sstevel@tonic-gate 	 * on the free list.  Notice that there was a brief time where
31187c478bd9Sstevel@tonic-gate 	 * a page is marked as free, but is not on the list.
31197c478bd9Sstevel@tonic-gate 	 *
31207c478bd9Sstevel@tonic-gate 	 * Set the reference bit to protect against immediate pageout.
31217c478bd9Sstevel@tonic-gate 	 */
312224e9c58bSelowe 	for (i = 0; i < npgs; i++, pp++) {
3123db874c57Selowe 		PP_CLRFREE(pp);
3124db874c57Selowe 		PP_CLRAGED(pp);
3125db874c57Selowe 		page_set_props(pp, P_REF);
3126db874c57Selowe 	}
31277c478bd9Sstevel@tonic-gate 
31287c478bd9Sstevel@tonic-gate 	CPU_STATS_ENTER_K();
31297c478bd9Sstevel@tonic-gate 	cpup = CPU;	/* get cpup now that CPU cannot change */
31307c478bd9Sstevel@tonic-gate 	CPU_STATS_ADDQ(cpup, vm, pgrec, 1);
31317c478bd9Sstevel@tonic-gate 	CPU_STATS_ADDQ(cpup, vm, pgfrec, 1);
31327c478bd9Sstevel@tonic-gate 	CPU_STATS_EXIT_K();
31337c478bd9Sstevel@tonic-gate 
31347c478bd9Sstevel@tonic-gate 	return (1);
31357c478bd9Sstevel@tonic-gate }
31367c478bd9Sstevel@tonic-gate 
31377c478bd9Sstevel@tonic-gate 
31387c478bd9Sstevel@tonic-gate 
31397c478bd9Sstevel@tonic-gate /*
31407c478bd9Sstevel@tonic-gate  * Destroy identity of the page and put it back on
31417c478bd9Sstevel@tonic-gate  * the page free list.  Assumes that the caller has
31427c478bd9Sstevel@tonic-gate  * acquired the "exclusive" lock on the page.
31437c478bd9Sstevel@tonic-gate  */
31447c478bd9Sstevel@tonic-gate void
31457c478bd9Sstevel@tonic-gate page_destroy(page_t *pp, int dontfree)
31467c478bd9Sstevel@tonic-gate {
31477c478bd9Sstevel@tonic-gate 	ASSERT((PAGE_EXCL(pp) &&
31487c478bd9Sstevel@tonic-gate 	    !page_iolock_assert(pp)) || panicstr);
314907b65a64Saguzovsk 	ASSERT(pp->p_slckcnt == 0 || panicstr);
31507c478bd9Sstevel@tonic-gate 
31517c478bd9Sstevel@tonic-gate 	if (pp->p_szc != 0) {
31527c478bd9Sstevel@tonic-gate 		if (pp->p_vnode == NULL || IS_SWAPFSVP(pp->p_vnode) ||
31537c478bd9Sstevel@tonic-gate 		    pp->p_vnode == &kvp) {
31547c478bd9Sstevel@tonic-gate 			panic("page_destroy: anon or kernel or no vnode "
31557c478bd9Sstevel@tonic-gate 			    "large page %p", (void *)pp);
31567c478bd9Sstevel@tonic-gate 		}
31577c478bd9Sstevel@tonic-gate 		page_demote_vp_pages(pp);
31587c478bd9Sstevel@tonic-gate 		ASSERT(pp->p_szc == 0);
31597c478bd9Sstevel@tonic-gate 	}
31607c478bd9Sstevel@tonic-gate 
31617c478bd9Sstevel@tonic-gate 	TRACE_1(TR_FAC_VM, TR_PAGE_DESTROY, "page_destroy:pp %p", pp);
31627c478bd9Sstevel@tonic-gate 
31637c478bd9Sstevel@tonic-gate 	/*
31647c478bd9Sstevel@tonic-gate 	 * Unload translations, if any, then hash out the
31657c478bd9Sstevel@tonic-gate 	 * page to erase its identity.
31667c478bd9Sstevel@tonic-gate 	 */
31677c478bd9Sstevel@tonic-gate 	(void) hat_pageunload(pp, HAT_FORCE_PGUNLOAD);
31687c478bd9Sstevel@tonic-gate 	page_hashout(pp, NULL);
31697c478bd9Sstevel@tonic-gate 
31707c478bd9Sstevel@tonic-gate 	if (!dontfree) {
31717c478bd9Sstevel@tonic-gate 		/*
31727c478bd9Sstevel@tonic-gate 		 * Acquire the "freemem_lock" for availrmem.
31737c478bd9Sstevel@tonic-gate 		 * The page_struct_lock need not be acquired for lckcnt
31747c478bd9Sstevel@tonic-gate 		 * and cowcnt since the page has an "exclusive" lock.
31757c478bd9Sstevel@tonic-gate 		 */
31767c478bd9Sstevel@tonic-gate 		if ((pp->p_lckcnt != 0) || (pp->p_cowcnt != 0)) {
31777c478bd9Sstevel@tonic-gate 			mutex_enter(&freemem_lock);
31787c478bd9Sstevel@tonic-gate 			if (pp->p_lckcnt != 0) {
31797c478bd9Sstevel@tonic-gate 				availrmem++;
31807c478bd9Sstevel@tonic-gate 				pp->p_lckcnt = 0;
31817c478bd9Sstevel@tonic-gate 			}
31827c478bd9Sstevel@tonic-gate 			if (pp->p_cowcnt != 0) {
31837c478bd9Sstevel@tonic-gate 				availrmem += pp->p_cowcnt;
31847c478bd9Sstevel@tonic-gate 				pp->p_cowcnt = 0;
31857c478bd9Sstevel@tonic-gate 			}
31867c478bd9Sstevel@tonic-gate 			mutex_exit(&freemem_lock);
31877c478bd9Sstevel@tonic-gate 		}
31887c478bd9Sstevel@tonic-gate 		/*
31897c478bd9Sstevel@tonic-gate 		 * Put the page on the "free" list.
31907c478bd9Sstevel@tonic-gate 		 */
31917c478bd9Sstevel@tonic-gate 		page_free(pp, 0);
31927c478bd9Sstevel@tonic-gate 	}
31937c478bd9Sstevel@tonic-gate }
31947c478bd9Sstevel@tonic-gate 
31957c478bd9Sstevel@tonic-gate void
31967c478bd9Sstevel@tonic-gate page_destroy_pages(page_t *pp)
31977c478bd9Sstevel@tonic-gate {
31987c478bd9Sstevel@tonic-gate 
31997c478bd9Sstevel@tonic-gate 	page_t	*tpp, *rootpp = NULL;
32007c478bd9Sstevel@tonic-gate 	pgcnt_t	pgcnt = page_get_pagecnt(pp->p_szc);
32017c478bd9Sstevel@tonic-gate 	pgcnt_t	i, pglcks = 0;
32027c478bd9Sstevel@tonic-gate 	uint_t	szc = pp->p_szc;
32037c478bd9Sstevel@tonic-gate 
32047c478bd9Sstevel@tonic-gate 	ASSERT(pp->p_szc != 0 && pp->p_szc < page_num_pagesizes());
32057c478bd9Sstevel@tonic-gate 
32067c478bd9Sstevel@tonic-gate 	VM_STAT_ADD(pagecnt.pc_destroy_pages);
32077c478bd9Sstevel@tonic-gate 
32087c478bd9Sstevel@tonic-gate 	TRACE_1(TR_FAC_VM, TR_PAGE_DESTROY, "page_destroy_pages:pp %p", pp);
32097c478bd9Sstevel@tonic-gate 
32107c478bd9Sstevel@tonic-gate 	if ((page_pptonum(pp) & (pgcnt - 1)) != 0) {
32117c478bd9Sstevel@tonic-gate 		panic("page_destroy_pages: not root page %p", (void *)pp);
32127c478bd9Sstevel@tonic-gate 		/*NOTREACHED*/
32137c478bd9Sstevel@tonic-gate 	}
32147c478bd9Sstevel@tonic-gate 
3215affbd3ccSkchow 	for (i = 0, tpp = pp; i < pgcnt; i++, tpp++) {
32167c478bd9Sstevel@tonic-gate 		ASSERT((PAGE_EXCL(tpp) &&
32177c478bd9Sstevel@tonic-gate 		    !page_iolock_assert(tpp)) || panicstr);
321807b65a64Saguzovsk 		ASSERT(tpp->p_slckcnt == 0 || panicstr);
32197c478bd9Sstevel@tonic-gate 		(void) hat_pageunload(tpp, HAT_FORCE_PGUNLOAD);
32207c478bd9Sstevel@tonic-gate 		page_hashout(tpp, NULL);
32217c478bd9Sstevel@tonic-gate 		ASSERT(tpp->p_offset == (u_offset_t)-1);
32227c478bd9Sstevel@tonic-gate 		if (tpp->p_lckcnt != 0) {
32237c478bd9Sstevel@tonic-gate 			pglcks++;
32247c478bd9Sstevel@tonic-gate 			tpp->p_lckcnt = 0;
32257c478bd9Sstevel@tonic-gate 		} else if (tpp->p_cowcnt != 0) {
32267c478bd9Sstevel@tonic-gate 			pglcks += tpp->p_cowcnt;
32277c478bd9Sstevel@tonic-gate 			tpp->p_cowcnt = 0;
32287c478bd9Sstevel@tonic-gate 		}
32297c478bd9Sstevel@tonic-gate 		ASSERT(!hat_page_getshare(tpp));
32307c478bd9Sstevel@tonic-gate 		ASSERT(tpp->p_vnode == NULL);
32317c478bd9Sstevel@tonic-gate 		ASSERT(tpp->p_szc == szc);
32327c478bd9Sstevel@tonic-gate 
32337c478bd9Sstevel@tonic-gate 		PP_SETFREE(tpp);
32347c478bd9Sstevel@tonic-gate 		page_clr_all_props(tpp);
32357c478bd9Sstevel@tonic-gate 		PP_SETAGED(tpp);
32367c478bd9Sstevel@tonic-gate 		ASSERT(tpp->p_next == tpp);
32377c478bd9Sstevel@tonic-gate 		ASSERT(tpp->p_prev == tpp);
32387c478bd9Sstevel@tonic-gate 		page_list_concat(&rootpp, &tpp);
32397c478bd9Sstevel@tonic-gate 	}
32407c478bd9Sstevel@tonic-gate 
32417c478bd9Sstevel@tonic-gate 	ASSERT(rootpp == pp);
32427c478bd9Sstevel@tonic-gate 	if (pglcks != 0) {
32437c478bd9Sstevel@tonic-gate 		mutex_enter(&freemem_lock);
32447c478bd9Sstevel@tonic-gate 		availrmem += pglcks;
32457c478bd9Sstevel@tonic-gate 		mutex_exit(&freemem_lock);
32467c478bd9Sstevel@tonic-gate 	}
32477c478bd9Sstevel@tonic-gate 
32487c478bd9Sstevel@tonic-gate 	page_list_add_pages(rootpp, 0);
32497c478bd9Sstevel@tonic-gate 	page_create_putback(pgcnt);
32507c478bd9Sstevel@tonic-gate }
32517c478bd9Sstevel@tonic-gate 
32527c478bd9Sstevel@tonic-gate /*
32537c478bd9Sstevel@tonic-gate  * Similar to page_destroy(), but destroys pages which are
32547c478bd9Sstevel@tonic-gate  * locked and known to be on the page free list.  Since
32557c478bd9Sstevel@tonic-gate  * the page is known to be free and locked, no one can access
32567c478bd9Sstevel@tonic-gate  * it.
32577c478bd9Sstevel@tonic-gate  *
32587c478bd9Sstevel@tonic-gate  * Also, the number of free pages does not change.
32597c478bd9Sstevel@tonic-gate  */
32607c478bd9Sstevel@tonic-gate void
32617c478bd9Sstevel@tonic-gate page_destroy_free(page_t *pp)
32627c478bd9Sstevel@tonic-gate {
32637c478bd9Sstevel@tonic-gate 	ASSERT(PAGE_EXCL(pp));
32647c478bd9Sstevel@tonic-gate 	ASSERT(PP_ISFREE(pp));
32657c478bd9Sstevel@tonic-gate 	ASSERT(pp->p_vnode);
32667c478bd9Sstevel@tonic-gate 	ASSERT(hat_page_getattr(pp, P_MOD | P_REF | P_RO) == 0);
32677c478bd9Sstevel@tonic-gate 	ASSERT(!hat_page_is_mapped(pp));
32687c478bd9Sstevel@tonic-gate 	ASSERT(PP_ISAGED(pp) == 0);
32697c478bd9Sstevel@tonic-gate 	ASSERT(pp->p_szc == 0);
32707c478bd9Sstevel@tonic-gate 
32717c478bd9Sstevel@tonic-gate 	VM_STAT_ADD(pagecnt.pc_destroy_free);
32727c478bd9Sstevel@tonic-gate 	page_list_sub(pp, PG_CACHE_LIST);
32737c478bd9Sstevel@tonic-gate 
32747c478bd9Sstevel@tonic-gate 	page_hashout(pp, NULL);
32757c478bd9Sstevel@tonic-gate 	ASSERT(pp->p_vnode == NULL);
32767c478bd9Sstevel@tonic-gate 	ASSERT(pp->p_offset == (u_offset_t)-1);
32777c478bd9Sstevel@tonic-gate 	ASSERT(pp->p_hash == NULL);
32787c478bd9Sstevel@tonic-gate 
32797c478bd9Sstevel@tonic-gate 	PP_SETAGED(pp);
32807c478bd9Sstevel@tonic-gate 	page_list_add(pp, PG_FREE_LIST | PG_LIST_TAIL);
32817c478bd9Sstevel@tonic-gate 	page_unlock(pp);
32827c478bd9Sstevel@tonic-gate 
32837c478bd9Sstevel@tonic-gate 	mutex_enter(&new_freemem_lock);
32847c478bd9Sstevel@tonic-gate 	if (freemem_wait) {
32857c478bd9Sstevel@tonic-gate 		cv_signal(&freemem_cv);
32867c478bd9Sstevel@tonic-gate 	}
32877c478bd9Sstevel@tonic-gate 	mutex_exit(&new_freemem_lock);
32887c478bd9Sstevel@tonic-gate }
32897c478bd9Sstevel@tonic-gate 
32907c478bd9Sstevel@tonic-gate /*
32917c478bd9Sstevel@tonic-gate  * Rename the page "opp" to have an identity specified
32927c478bd9Sstevel@tonic-gate  * by [vp, off].  If a page already exists with this name
32937c478bd9Sstevel@tonic-gate  * it is locked and destroyed.  Note that the page's
32947c478bd9Sstevel@tonic-gate  * translations are not unloaded during the rename.
32957c478bd9Sstevel@tonic-gate  *
32967c478bd9Sstevel@tonic-gate  * This routine is used by the anon layer to "steal" the
32977c478bd9Sstevel@tonic-gate  * original page and is not unlike destroying a page and
32987c478bd9Sstevel@tonic-gate  * creating a new page using the same page frame.
32997c478bd9Sstevel@tonic-gate  *
33007c478bd9Sstevel@tonic-gate  * XXX -- Could deadlock if caller 1 tries to rename A to B while
33017c478bd9Sstevel@tonic-gate  * caller 2 tries to rename B to A.
33027c478bd9Sstevel@tonic-gate  */
33037c478bd9Sstevel@tonic-gate void
33047c478bd9Sstevel@tonic-gate page_rename(page_t *opp, vnode_t *vp, u_offset_t off)
33057c478bd9Sstevel@tonic-gate {
33067c478bd9Sstevel@tonic-gate 	page_t		*pp;
33077c478bd9Sstevel@tonic-gate 	int		olckcnt = 0;
33087c478bd9Sstevel@tonic-gate 	int		ocowcnt = 0;
33097c478bd9Sstevel@tonic-gate 	kmutex_t	*phm;
33107c478bd9Sstevel@tonic-gate 	ulong_t		index;
33117c478bd9Sstevel@tonic-gate 
33127c478bd9Sstevel@tonic-gate 	ASSERT(PAGE_EXCL(opp) && !page_iolock_assert(opp));
33137c478bd9Sstevel@tonic-gate 	ASSERT(MUTEX_NOT_HELD(page_vnode_mutex(vp)));
33147c478bd9Sstevel@tonic-gate 	ASSERT(PP_ISFREE(opp) == 0);
33157c478bd9Sstevel@tonic-gate 
33167c478bd9Sstevel@tonic-gate 	VM_STAT_ADD(page_rename_count);
33177c478bd9Sstevel@tonic-gate 
33187c478bd9Sstevel@tonic-gate 	TRACE_3(TR_FAC_VM, TR_PAGE_RENAME,
33197c478bd9Sstevel@tonic-gate 		"page rename:pp %p vp %p off %llx", opp, vp, off);
33207c478bd9Sstevel@tonic-gate 
332137fbc076Saguzovsk 	/*
332237fbc076Saguzovsk 	 * CacheFS may call page_rename for a large NFS page
332337fbc076Saguzovsk 	 * when both CacheFS and NFS mount points are used
332437fbc076Saguzovsk 	 * by applications. Demote this large page before
332537fbc076Saguzovsk 	 * renaming it, to ensure that there are no "partial"
332637fbc076Saguzovsk 	 * large pages left lying around.
332737fbc076Saguzovsk 	 */
332837fbc076Saguzovsk 	if (opp->p_szc != 0) {
332937fbc076Saguzovsk 		vnode_t *ovp = opp->p_vnode;
333037fbc076Saguzovsk 		ASSERT(ovp != NULL);
333137fbc076Saguzovsk 		ASSERT(!IS_SWAPFSVP(ovp));
333237fbc076Saguzovsk 		ASSERT(ovp != &kvp);
333337fbc076Saguzovsk 		page_demote_vp_pages(opp);
333437fbc076Saguzovsk 		ASSERT(opp->p_szc == 0);
333537fbc076Saguzovsk 	}
333637fbc076Saguzovsk 
33377c478bd9Sstevel@tonic-gate 	page_hashout(opp, NULL);
33387c478bd9Sstevel@tonic-gate 	PP_CLRAGED(opp);
33397c478bd9Sstevel@tonic-gate 
33407c478bd9Sstevel@tonic-gate 	/*
33417c478bd9Sstevel@tonic-gate 	 * Acquire the appropriate page hash lock, since
33427c478bd9Sstevel@tonic-gate 	 * we're going to rename the page.
33437c478bd9Sstevel@tonic-gate 	 */
33447c478bd9Sstevel@tonic-gate 	index = PAGE_HASH_FUNC(vp, off);
33457c478bd9Sstevel@tonic-gate 	phm = PAGE_HASH_MUTEX(index);
33467c478bd9Sstevel@tonic-gate 	mutex_enter(phm);
33477c478bd9Sstevel@tonic-gate top:
33487c478bd9Sstevel@tonic-gate 	/*
33497c478bd9Sstevel@tonic-gate 	 * Look for an existing page with this name and destroy it if found.
33507c478bd9Sstevel@tonic-gate 	 * By holding the page hash lock all the way to the page_hashin()
33517c478bd9Sstevel@tonic-gate 	 * call, we are assured that no page can be created with this
33527c478bd9Sstevel@tonic-gate 	 * identity.  In the case when the phm lock is dropped to undo any
33537c478bd9Sstevel@tonic-gate 	 * hat layer mappings, the existing page is held with an "exclusive"
33547c478bd9Sstevel@tonic-gate 	 * lock, again preventing another page from being created with
33557c478bd9Sstevel@tonic-gate 	 * this identity.
33567c478bd9Sstevel@tonic-gate 	 */
33577c478bd9Sstevel@tonic-gate 	PAGE_HASH_SEARCH(index, pp, vp, off);
33587c478bd9Sstevel@tonic-gate 	if (pp != NULL) {
33597c478bd9Sstevel@tonic-gate 		VM_STAT_ADD(page_rename_exists);
33607c478bd9Sstevel@tonic-gate 
33617c478bd9Sstevel@tonic-gate 		/*
33627c478bd9Sstevel@tonic-gate 		 * As it turns out, this is one of only two places where
33637c478bd9Sstevel@tonic-gate 		 * page_lock() needs to hold the passed in lock in the
33647c478bd9Sstevel@tonic-gate 		 * successful case.  In all of the others, the lock could
33657c478bd9Sstevel@tonic-gate 		 * be dropped as soon as the attempt is made to lock
33667c478bd9Sstevel@tonic-gate 		 * the page.  It is tempting to add yet another arguement,
33677c478bd9Sstevel@tonic-gate 		 * PL_KEEP or PL_DROP, to let page_lock know what to do.
33687c478bd9Sstevel@tonic-gate 		 */
33697c478bd9Sstevel@tonic-gate 		if (!page_lock(pp, SE_EXCL, phm, P_RECLAIM)) {
33707c478bd9Sstevel@tonic-gate 			/*
33717c478bd9Sstevel@tonic-gate 			 * Went to sleep because the page could not
33727c478bd9Sstevel@tonic-gate 			 * be locked.  We were woken up when the page
33737c478bd9Sstevel@tonic-gate 			 * was unlocked, or when the page was destroyed.
33747c478bd9Sstevel@tonic-gate 			 * In either case, `phm' was dropped while we
33757c478bd9Sstevel@tonic-gate 			 * slept.  Hence we should not just roar through
33767c478bd9Sstevel@tonic-gate 			 * this loop.
33777c478bd9Sstevel@tonic-gate 			 */
33787c478bd9Sstevel@tonic-gate 			goto top;
33797c478bd9Sstevel@tonic-gate 		}
33807c478bd9Sstevel@tonic-gate 
338137fbc076Saguzovsk 		/*
338237fbc076Saguzovsk 		 * If an existing page is a large page, then demote
338337fbc076Saguzovsk 		 * it to ensure that no "partial" large pages are
338437fbc076Saguzovsk 		 * "created" after page_rename. An existing page
338537fbc076Saguzovsk 		 * can be a CacheFS page, and can't belong to swapfs.
338637fbc076Saguzovsk 		 */
33877c478bd9Sstevel@tonic-gate 		if (hat_page_is_mapped(pp)) {
33887c478bd9Sstevel@tonic-gate 			/*
33897c478bd9Sstevel@tonic-gate 			 * Unload translations.  Since we hold the
33907c478bd9Sstevel@tonic-gate 			 * exclusive lock on this page, the page
33917c478bd9Sstevel@tonic-gate 			 * can not be changed while we drop phm.
33927c478bd9Sstevel@tonic-gate 			 * This is also not a lock protocol violation,
33937c478bd9Sstevel@tonic-gate 			 * but rather the proper way to do things.
33947c478bd9Sstevel@tonic-gate 			 */
33957c478bd9Sstevel@tonic-gate 			mutex_exit(phm);
33967c478bd9Sstevel@tonic-gate 			(void) hat_pageunload(pp, HAT_FORCE_PGUNLOAD);
339737fbc076Saguzovsk 			if (pp->p_szc != 0) {
339837fbc076Saguzovsk 				ASSERT(!IS_SWAPFSVP(vp));
339937fbc076Saguzovsk 				ASSERT(vp != &kvp);
340037fbc076Saguzovsk 				page_demote_vp_pages(pp);
340137fbc076Saguzovsk 				ASSERT(pp->p_szc == 0);
340237fbc076Saguzovsk 			}
340337fbc076Saguzovsk 			mutex_enter(phm);
340437fbc076Saguzovsk 		} else if (pp->p_szc != 0) {
340537fbc076Saguzovsk 			ASSERT(!IS_SWAPFSVP(vp));
340637fbc076Saguzovsk 			ASSERT(vp != &kvp);
340737fbc076Saguzovsk 			mutex_exit(phm);
340837fbc076Saguzovsk 			page_demote_vp_pages(pp);
340937fbc076Saguzovsk 			ASSERT(pp->p_szc == 0);
34107c478bd9Sstevel@tonic-gate 			mutex_enter(phm);
34117c478bd9Sstevel@tonic-gate 		}
34127c478bd9Sstevel@tonic-gate 		page_hashout(pp, phm);
34137c478bd9Sstevel@tonic-gate 	}
34147c478bd9Sstevel@tonic-gate 	/*
34157c478bd9Sstevel@tonic-gate 	 * Hash in the page with the new identity.
34167c478bd9Sstevel@tonic-gate 	 */
34177c478bd9Sstevel@tonic-gate 	if (!page_hashin(opp, vp, off, phm)) {
34187c478bd9Sstevel@tonic-gate 		/*
34197c478bd9Sstevel@tonic-gate 		 * We were holding phm while we searched for [vp, off]
34207c478bd9Sstevel@tonic-gate 		 * and only dropped phm if we found and locked a page.
34217c478bd9Sstevel@tonic-gate 		 * If we can't create this page now, then some thing
34227c478bd9Sstevel@tonic-gate 		 * is really broken.
34237c478bd9Sstevel@tonic-gate 		 */
34247c478bd9Sstevel@tonic-gate 		panic("page_rename: Can't hash in page: %p", (void *)pp);
34257c478bd9Sstevel@tonic-gate 		/*NOTREACHED*/
34267c478bd9Sstevel@tonic-gate 	}
34277c478bd9Sstevel@tonic-gate 
34287c478bd9Sstevel@tonic-gate 	ASSERT(MUTEX_HELD(phm));
34297c478bd9Sstevel@tonic-gate 	mutex_exit(phm);
34307c478bd9Sstevel@tonic-gate 
34317c478bd9Sstevel@tonic-gate 	/*
34327c478bd9Sstevel@tonic-gate 	 * Now that we have dropped phm, lets get around to finishing up
34337c478bd9Sstevel@tonic-gate 	 * with pp.
34347c478bd9Sstevel@tonic-gate 	 */
34357c478bd9Sstevel@tonic-gate 	if (pp != NULL) {
34367c478bd9Sstevel@tonic-gate 		ASSERT(!hat_page_is_mapped(pp));
34377c478bd9Sstevel@tonic-gate 		/* for now large pages should not end up here */
34387c478bd9Sstevel@tonic-gate 		ASSERT(pp->p_szc == 0);
34397c478bd9Sstevel@tonic-gate 		/*
34407c478bd9Sstevel@tonic-gate 		 * Save the locks for transfer to the new page and then
34417c478bd9Sstevel@tonic-gate 		 * clear them so page_free doesn't think they're important.
34427c478bd9Sstevel@tonic-gate 		 * The page_struct_lock need not be acquired for lckcnt and
34437c478bd9Sstevel@tonic-gate 		 * cowcnt since the page has an "exclusive" lock.
34447c478bd9Sstevel@tonic-gate 		 */
34457c478bd9Sstevel@tonic-gate 		olckcnt = pp->p_lckcnt;
34467c478bd9Sstevel@tonic-gate 		ocowcnt = pp->p_cowcnt;
34477c478bd9Sstevel@tonic-gate 		pp->p_lckcnt = pp->p_cowcnt = 0;
34487c478bd9Sstevel@tonic-gate 
34497c478bd9Sstevel@tonic-gate 		/*
34507c478bd9Sstevel@tonic-gate 		 * Put the page on the "free" list after we drop
34517c478bd9Sstevel@tonic-gate 		 * the lock.  The less work under the lock the better.
34527c478bd9Sstevel@tonic-gate 		 */
34537c478bd9Sstevel@tonic-gate 		/*LINTED: constant in conditional context*/
34547c478bd9Sstevel@tonic-gate 		VN_DISPOSE(pp, B_FREE, 0, kcred);
34557c478bd9Sstevel@tonic-gate 	}
34567c478bd9Sstevel@tonic-gate 
34577c478bd9Sstevel@tonic-gate 	/*
34587c478bd9Sstevel@tonic-gate 	 * Transfer the lock count from the old page (if any).
34597c478bd9Sstevel@tonic-gate 	 * The page_struct_lock need not be acquired for lckcnt and
34607c478bd9Sstevel@tonic-gate 	 * cowcnt since the page has an "exclusive" lock.
34617c478bd9Sstevel@tonic-gate 	 */
34627c478bd9Sstevel@tonic-gate 	opp->p_lckcnt += olckcnt;
34637c478bd9Sstevel@tonic-gate 	opp->p_cowcnt += ocowcnt;
34647c478bd9Sstevel@tonic-gate }
34657c478bd9Sstevel@tonic-gate 
34667c478bd9Sstevel@tonic-gate /*
34677c478bd9Sstevel@tonic-gate  * low level routine to add page `pp' to the hash and vp chains for [vp, offset]
34687c478bd9Sstevel@tonic-gate  *
34697c478bd9Sstevel@tonic-gate  * Pages are normally inserted at the start of a vnode's v_pages list.
34707c478bd9Sstevel@tonic-gate  * If the vnode is VMODSORT and the page is modified, it goes at the end.
34717c478bd9Sstevel@tonic-gate  * This can happen when a modified page is relocated for DR.
34727c478bd9Sstevel@tonic-gate  *
34737c478bd9Sstevel@tonic-gate  * Returns 1 on success and 0 on failure.
34747c478bd9Sstevel@tonic-gate  */
34757c478bd9Sstevel@tonic-gate static int
34767c478bd9Sstevel@tonic-gate page_do_hashin(page_t *pp, vnode_t *vp, u_offset_t offset)
34777c478bd9Sstevel@tonic-gate {
34787c478bd9Sstevel@tonic-gate 	page_t		**listp;
34797c478bd9Sstevel@tonic-gate 	page_t		*tp;
34807c478bd9Sstevel@tonic-gate 	ulong_t		index;
34817c478bd9Sstevel@tonic-gate 
34827c478bd9Sstevel@tonic-gate 	ASSERT(PAGE_EXCL(pp));
34837c478bd9Sstevel@tonic-gate 	ASSERT(vp != NULL);
34847c478bd9Sstevel@tonic-gate 	ASSERT(MUTEX_HELD(page_vnode_mutex(vp)));
34857c478bd9Sstevel@tonic-gate 
34867c478bd9Sstevel@tonic-gate 	/*
34877c478bd9Sstevel@tonic-gate 	 * Be sure to set these up before the page is inserted on the hash
34887c478bd9Sstevel@tonic-gate 	 * list.  As soon as the page is placed on the list some other
34897c478bd9Sstevel@tonic-gate 	 * thread might get confused and wonder how this page could
34907c478bd9Sstevel@tonic-gate 	 * possibly hash to this list.
34917c478bd9Sstevel@tonic-gate 	 */
34927c478bd9Sstevel@tonic-gate 	pp->p_vnode = vp;
34937c478bd9Sstevel@tonic-gate 	pp->p_offset = offset;
34947c478bd9Sstevel@tonic-gate 
34957c478bd9Sstevel@tonic-gate 	/*
34967c478bd9Sstevel@tonic-gate 	 * record if this page is on a swap vnode
34977c478bd9Sstevel@tonic-gate 	 */
34987c478bd9Sstevel@tonic-gate 	if ((vp->v_flag & VISSWAP) != 0)
34997c478bd9Sstevel@tonic-gate 		PP_SETSWAP(pp);
35007c478bd9Sstevel@tonic-gate 
35017c478bd9Sstevel@tonic-gate 	index = PAGE_HASH_FUNC(vp, offset);
35027c478bd9Sstevel@tonic-gate 	ASSERT(MUTEX_HELD(PAGE_HASH_MUTEX(index)));
35037c478bd9Sstevel@tonic-gate 	listp = &page_hash[index];
35047c478bd9Sstevel@tonic-gate 
35057c478bd9Sstevel@tonic-gate 	/*
35067c478bd9Sstevel@tonic-gate 	 * If this page is already hashed in, fail this attempt to add it.
35077c478bd9Sstevel@tonic-gate 	 */
35087c478bd9Sstevel@tonic-gate 	for (tp = *listp; tp != NULL; tp = tp->p_hash) {
35097c478bd9Sstevel@tonic-gate 		if (tp->p_vnode == vp && tp->p_offset == offset) {
35107c478bd9Sstevel@tonic-gate 			pp->p_vnode = NULL;
35117c478bd9Sstevel@tonic-gate 			pp->p_offset = (u_offset_t)(-1);
35127c478bd9Sstevel@tonic-gate 			return (0);
35137c478bd9Sstevel@tonic-gate 		}
35147c478bd9Sstevel@tonic-gate 	}
35157c478bd9Sstevel@tonic-gate 	pp->p_hash = *listp;
35167c478bd9Sstevel@tonic-gate 	*listp = pp;
35177c478bd9Sstevel@tonic-gate 
35187c478bd9Sstevel@tonic-gate 	/*
35197c478bd9Sstevel@tonic-gate 	 * Add the page to the vnode's list of pages
35207c478bd9Sstevel@tonic-gate 	 */
35217c478bd9Sstevel@tonic-gate 	if (vp->v_pages != NULL && IS_VMODSORT(vp) && hat_ismod(pp))
35227c478bd9Sstevel@tonic-gate 		listp = &vp->v_pages->p_vpprev->p_vpnext;
35237c478bd9Sstevel@tonic-gate 	else
35247c478bd9Sstevel@tonic-gate 		listp = &vp->v_pages;
35257c478bd9Sstevel@tonic-gate 
35267c478bd9Sstevel@tonic-gate 	page_vpadd(listp, pp);
35277c478bd9Sstevel@tonic-gate 
35287c478bd9Sstevel@tonic-gate 	return (1);
35297c478bd9Sstevel@tonic-gate }
35307c478bd9Sstevel@tonic-gate 
35317c478bd9Sstevel@tonic-gate /*
35327c478bd9Sstevel@tonic-gate  * Add page `pp' to both the hash and vp chains for [vp, offset].
35337c478bd9Sstevel@tonic-gate  *
35347c478bd9Sstevel@tonic-gate  * Returns 1 on success and 0 on failure.
35357c478bd9Sstevel@tonic-gate  * If hold is passed in, it is not dropped.
35367c478bd9Sstevel@tonic-gate  */
35377c478bd9Sstevel@tonic-gate int
35387c478bd9Sstevel@tonic-gate page_hashin(page_t *pp, vnode_t *vp, u_offset_t offset, kmutex_t *hold)
35397c478bd9Sstevel@tonic-gate {
35407c478bd9Sstevel@tonic-gate 	kmutex_t	*phm = NULL;
35417c478bd9Sstevel@tonic-gate 	kmutex_t	*vphm;
35427c478bd9Sstevel@tonic-gate 	int		rc;
35437c478bd9Sstevel@tonic-gate 
35447c478bd9Sstevel@tonic-gate 	ASSERT(MUTEX_NOT_HELD(page_vnode_mutex(vp)));
35457c478bd9Sstevel@tonic-gate 
35467c478bd9Sstevel@tonic-gate 	TRACE_3(TR_FAC_VM, TR_PAGE_HASHIN,
35477c478bd9Sstevel@tonic-gate 		"page_hashin:pp %p vp %p offset %llx",
35487c478bd9Sstevel@tonic-gate 		pp, vp, offset);
35497c478bd9Sstevel@tonic-gate 
35507c478bd9Sstevel@tonic-gate 	VM_STAT_ADD(hashin_count);
35517c478bd9Sstevel@tonic-gate 
35527c478bd9Sstevel@tonic-gate 	if (hold != NULL)
35537c478bd9Sstevel@tonic-gate 		phm = hold;
35547c478bd9Sstevel@tonic-gate 	else {
35557c478bd9Sstevel@tonic-gate 		VM_STAT_ADD(hashin_not_held);
35567c478bd9Sstevel@tonic-gate 		phm = PAGE_HASH_MUTEX(PAGE_HASH_FUNC(vp, offset));
35577c478bd9Sstevel@tonic-gate 		mutex_enter(phm);
35587c478bd9Sstevel@tonic-gate 	}
35597c478bd9Sstevel@tonic-gate 
35607c478bd9Sstevel@tonic-gate 	vphm = page_vnode_mutex(vp);
35617c478bd9Sstevel@tonic-gate 	mutex_enter(vphm);
35627c478bd9Sstevel@tonic-gate 	rc = page_do_hashin(pp, vp, offset);
35637c478bd9Sstevel@tonic-gate 	mutex_exit(vphm);
35647c478bd9Sstevel@tonic-gate 	if (hold == NULL)
35657c478bd9Sstevel@tonic-gate 		mutex_exit(phm);
35667c478bd9Sstevel@tonic-gate 	if (rc == 0)
35677c478bd9Sstevel@tonic-gate 		VM_STAT_ADD(hashin_already);
35687c478bd9Sstevel@tonic-gate 	return (rc);
35697c478bd9Sstevel@tonic-gate }
35707c478bd9Sstevel@tonic-gate 
35717c478bd9Sstevel@tonic-gate /*
35727c478bd9Sstevel@tonic-gate  * Remove page ``pp'' from the hash and vp chains and remove vp association.
35737c478bd9Sstevel@tonic-gate  * All mutexes must be held
35747c478bd9Sstevel@tonic-gate  */
35757c478bd9Sstevel@tonic-gate static void
35767c478bd9Sstevel@tonic-gate page_do_hashout(page_t *pp)
35777c478bd9Sstevel@tonic-gate {
35787c478bd9Sstevel@tonic-gate 	page_t	**hpp;
35797c478bd9Sstevel@tonic-gate 	page_t	*hp;
35807c478bd9Sstevel@tonic-gate 	vnode_t	*vp = pp->p_vnode;
35817c478bd9Sstevel@tonic-gate 
35827c478bd9Sstevel@tonic-gate 	ASSERT(vp != NULL);
35837c478bd9Sstevel@tonic-gate 	ASSERT(MUTEX_HELD(page_vnode_mutex(vp)));
35847c478bd9Sstevel@tonic-gate 
35857c478bd9Sstevel@tonic-gate 	/*
35867c478bd9Sstevel@tonic-gate 	 * First, take pp off of its hash chain.
35877c478bd9Sstevel@tonic-gate 	 */
35887c478bd9Sstevel@tonic-gate 	hpp = &page_hash[PAGE_HASH_FUNC(vp, pp->p_offset)];
35897c478bd9Sstevel@tonic-gate 
35907c478bd9Sstevel@tonic-gate 	for (;;) {
35917c478bd9Sstevel@tonic-gate 		hp = *hpp;
35927c478bd9Sstevel@tonic-gate 		if (hp == pp)
35937c478bd9Sstevel@tonic-gate 			break;
35947c478bd9Sstevel@tonic-gate 		if (hp == NULL) {
35957c478bd9Sstevel@tonic-gate 			panic("page_do_hashout");
35967c478bd9Sstevel@tonic-gate 			/*NOTREACHED*/
35977c478bd9Sstevel@tonic-gate 		}
35987c478bd9Sstevel@tonic-gate 		hpp = &hp->p_hash;
35997c478bd9Sstevel@tonic-gate 	}
36007c478bd9Sstevel@tonic-gate 	*hpp = pp->p_hash;
36017c478bd9Sstevel@tonic-gate 
36027c478bd9Sstevel@tonic-gate 	/*
36037c478bd9Sstevel@tonic-gate 	 * Now remove it from its associated vnode.
36047c478bd9Sstevel@tonic-gate 	 */
36057c478bd9Sstevel@tonic-gate 	if (vp->v_pages)
36067c478bd9Sstevel@tonic-gate 		page_vpsub(&vp->v_pages, pp);
36077c478bd9Sstevel@tonic-gate 
36087c478bd9Sstevel@tonic-gate 	pp->p_hash = NULL;
36097c478bd9Sstevel@tonic-gate 	page_clr_all_props(pp);
36107c478bd9Sstevel@tonic-gate 	PP_CLRSWAP(pp);
36117c478bd9Sstevel@tonic-gate 	pp->p_vnode = NULL;
36127c478bd9Sstevel@tonic-gate 	pp->p_offset = (u_offset_t)-1;
36137c478bd9Sstevel@tonic-gate }
36147c478bd9Sstevel@tonic-gate 
36157c478bd9Sstevel@tonic-gate /*
36167c478bd9Sstevel@tonic-gate  * Remove page ``pp'' from the hash and vp chains and remove vp association.
36177c478bd9Sstevel@tonic-gate  *
36187c478bd9Sstevel@tonic-gate  * When `phm' is non-NULL it contains the address of the mutex protecting the
36197c478bd9Sstevel@tonic-gate  * hash list pp is on.  It is not dropped.
36207c478bd9Sstevel@tonic-gate  */
36217c478bd9Sstevel@tonic-gate void
36227c478bd9Sstevel@tonic-gate page_hashout(page_t *pp, kmutex_t *phm)
36237c478bd9Sstevel@tonic-gate {
36247c478bd9Sstevel@tonic-gate 	vnode_t		*vp;
36257c478bd9Sstevel@tonic-gate 	ulong_t		index;
36267c478bd9Sstevel@tonic-gate 	kmutex_t	*nphm;
36277c478bd9Sstevel@tonic-gate 	kmutex_t	*vphm;
36287c478bd9Sstevel@tonic-gate 	kmutex_t	*sep;
36297c478bd9Sstevel@tonic-gate 
36307c478bd9Sstevel@tonic-gate 	ASSERT(phm != NULL ? MUTEX_HELD(phm) : 1);
36317c478bd9Sstevel@tonic-gate 	ASSERT(pp->p_vnode != NULL);
36327c478bd9Sstevel@tonic-gate 	ASSERT((PAGE_EXCL(pp) && !page_iolock_assert(pp)) || panicstr);
36337c478bd9Sstevel@tonic-gate 	ASSERT(MUTEX_NOT_HELD(page_vnode_mutex(pp->p_vnode)));
36347c478bd9Sstevel@tonic-gate 
36357c478bd9Sstevel@tonic-gate 	vp = pp->p_vnode;
36367c478bd9Sstevel@tonic-gate 
36377c478bd9Sstevel@tonic-gate 	TRACE_2(TR_FAC_VM, TR_PAGE_HASHOUT,
36387c478bd9Sstevel@tonic-gate 		"page_hashout:pp %p vp %p", pp, vp);
36397c478bd9Sstevel@tonic-gate 
36407c478bd9Sstevel@tonic-gate 	/* Kernel probe */
36417c478bd9Sstevel@tonic-gate 	TNF_PROBE_2(page_unmap, "vm pagefault", /* CSTYLED */,
36427c478bd9Sstevel@tonic-gate 	    tnf_opaque, vnode, vp,
36437c478bd9Sstevel@tonic-gate 	    tnf_offset, offset, pp->p_offset);
36447c478bd9Sstevel@tonic-gate 
36457c478bd9Sstevel@tonic-gate 	/*
36467c478bd9Sstevel@tonic-gate 	 *
36477c478bd9Sstevel@tonic-gate 	 */
36487c478bd9Sstevel@tonic-gate 	VM_STAT_ADD(hashout_count);
36497c478bd9Sstevel@tonic-gate 	index = PAGE_HASH_FUNC(vp, pp->p_offset);
36507c478bd9Sstevel@tonic-gate 	if (phm == NULL) {
36517c478bd9Sstevel@tonic-gate 		VM_STAT_ADD(hashout_not_held);
36527c478bd9Sstevel@tonic-gate 		nphm = PAGE_HASH_MUTEX(index);
36537c478bd9Sstevel@tonic-gate 		mutex_enter(nphm);
36547c478bd9Sstevel@tonic-gate 	}
36557c478bd9Sstevel@tonic-gate 	ASSERT(phm ? phm == PAGE_HASH_MUTEX(index) : 1);
36567c478bd9Sstevel@tonic-gate 
36577c478bd9Sstevel@tonic-gate 
36587c478bd9Sstevel@tonic-gate 	/*
36597c478bd9Sstevel@tonic-gate 	 * grab page vnode mutex and remove it...
36607c478bd9Sstevel@tonic-gate 	 */
36617c478bd9Sstevel@tonic-gate 	vphm = page_vnode_mutex(vp);
36627c478bd9Sstevel@tonic-gate 	mutex_enter(vphm);
36637c478bd9Sstevel@tonic-gate 
36647c478bd9Sstevel@tonic-gate 	page_do_hashout(pp);
36657c478bd9Sstevel@tonic-gate 
36667c478bd9Sstevel@tonic-gate 	mutex_exit(vphm);
36677c478bd9Sstevel@tonic-gate 	if (phm == NULL)
36687c478bd9Sstevel@tonic-gate 		mutex_exit(nphm);
36697c478bd9Sstevel@tonic-gate 
36707c478bd9Sstevel@tonic-gate 	/*
36717c478bd9Sstevel@tonic-gate 	 * Wake up processes waiting for this page.  The page's
36727c478bd9Sstevel@tonic-gate 	 * identity has been changed, and is probably not the
36737c478bd9Sstevel@tonic-gate 	 * desired page any longer.
36747c478bd9Sstevel@tonic-gate 	 */
36757c478bd9Sstevel@tonic-gate 	sep = page_se_mutex(pp);
36767c478bd9Sstevel@tonic-gate 	mutex_enter(sep);
367742787a71Sstans 	pp->p_selock &= ~SE_EWANTED;
36787c478bd9Sstevel@tonic-gate 	if (CV_HAS_WAITERS(&pp->p_cv))
36797c478bd9Sstevel@tonic-gate 		cv_broadcast(&pp->p_cv);
36807c478bd9Sstevel@tonic-gate 	mutex_exit(sep);
36817c478bd9Sstevel@tonic-gate }
36827c478bd9Sstevel@tonic-gate 
36837c478bd9Sstevel@tonic-gate /*
36847c478bd9Sstevel@tonic-gate  * Add the page to the front of a linked list of pages
36857c478bd9Sstevel@tonic-gate  * using the p_next & p_prev pointers for the list.
36867c478bd9Sstevel@tonic-gate  * The caller is responsible for protecting the list pointers.
36877c478bd9Sstevel@tonic-gate  */
36887c478bd9Sstevel@tonic-gate void
36897c478bd9Sstevel@tonic-gate page_add(page_t **ppp, page_t *pp)
36907c478bd9Sstevel@tonic-gate {
36917c478bd9Sstevel@tonic-gate 	ASSERT(PAGE_EXCL(pp) || (PAGE_SHARED(pp) && page_iolock_assert(pp)));
36927c478bd9Sstevel@tonic-gate 
36937c478bd9Sstevel@tonic-gate 	page_add_common(ppp, pp);
36947c478bd9Sstevel@tonic-gate }
36957c478bd9Sstevel@tonic-gate 
36967c478bd9Sstevel@tonic-gate 
36977c478bd9Sstevel@tonic-gate 
36987c478bd9Sstevel@tonic-gate /*
36997c478bd9Sstevel@tonic-gate  *  Common code for page_add() and mach_page_add()
37007c478bd9Sstevel@tonic-gate  */
37017c478bd9Sstevel@tonic-gate void
37027c478bd9Sstevel@tonic-gate page_add_common(page_t **ppp, page_t *pp)
37037c478bd9Sstevel@tonic-gate {
37047c478bd9Sstevel@tonic-gate 	if (*ppp == NULL) {
37057c478bd9Sstevel@tonic-gate 		pp->p_next = pp->p_prev = pp;
37067c478bd9Sstevel@tonic-gate 	} else {
37077c478bd9Sstevel@tonic-gate 		pp->p_next = *ppp;
37087c478bd9Sstevel@tonic-gate 		pp->p_prev = (*ppp)->p_prev;
37097c478bd9Sstevel@tonic-gate 		(*ppp)->p_prev = pp;
37107c478bd9Sstevel@tonic-gate 		pp->p_prev->p_next = pp;
37117c478bd9Sstevel@tonic-gate 	}
37127c478bd9Sstevel@tonic-gate 	*ppp = pp;
37137c478bd9Sstevel@tonic-gate }
37147c478bd9Sstevel@tonic-gate 
37157c478bd9Sstevel@tonic-gate 
37167c478bd9Sstevel@tonic-gate /*
37177c478bd9Sstevel@tonic-gate  * Remove this page from a linked list of pages
37187c478bd9Sstevel@tonic-gate  * using the p_next & p_prev pointers for the list.
37197c478bd9Sstevel@tonic-gate  *
37207c478bd9Sstevel@tonic-gate  * The caller is responsible for protecting the list pointers.
37217c478bd9Sstevel@tonic-gate  */
37227c478bd9Sstevel@tonic-gate void
37237c478bd9Sstevel@tonic-gate page_sub(page_t **ppp, page_t *pp)
37247c478bd9Sstevel@tonic-gate {
37257c478bd9Sstevel@tonic-gate 	ASSERT((PP_ISFREE(pp)) ? 1 :
37267c478bd9Sstevel@tonic-gate 	    (PAGE_EXCL(pp)) || (PAGE_SHARED(pp) && page_iolock_assert(pp)));
37277c478bd9Sstevel@tonic-gate 
37287c478bd9Sstevel@tonic-gate 	if (*ppp == NULL || pp == NULL) {
37297c478bd9Sstevel@tonic-gate 		panic("page_sub: bad arg(s): pp %p, *ppp %p",
37307c478bd9Sstevel@tonic-gate 		    (void *)pp, (void *)(*ppp));
37317c478bd9Sstevel@tonic-gate 		/*NOTREACHED*/
37327c478bd9Sstevel@tonic-gate 	}
37337c478bd9Sstevel@tonic-gate 
37347c478bd9Sstevel@tonic-gate 	page_sub_common(ppp, pp);
37357c478bd9Sstevel@tonic-gate }
37367c478bd9Sstevel@tonic-gate 
37377c478bd9Sstevel@tonic-gate 
37387c478bd9Sstevel@tonic-gate /*
37397c478bd9Sstevel@tonic-gate  *  Common code for page_sub() and mach_page_sub()
37407c478bd9Sstevel@tonic-gate  */
37417c478bd9Sstevel@tonic-gate void
37427c478bd9Sstevel@tonic-gate page_sub_common(page_t **ppp, page_t *pp)
37437c478bd9Sstevel@tonic-gate {
37447c478bd9Sstevel@tonic-gate 	if (*ppp == pp)
37457c478bd9Sstevel@tonic-gate 		*ppp = pp->p_next;		/* go to next page */
37467c478bd9Sstevel@tonic-gate 
37477c478bd9Sstevel@tonic-gate 	if (*ppp == pp)
37487c478bd9Sstevel@tonic-gate 		*ppp = NULL;			/* page list is gone */
37497c478bd9Sstevel@tonic-gate 	else {
37507c478bd9Sstevel@tonic-gate 		pp->p_prev->p_next = pp->p_next;
37517c478bd9Sstevel@tonic-gate 		pp->p_next->p_prev = pp->p_prev;
37527c478bd9Sstevel@tonic-gate 	}
37537c478bd9Sstevel@tonic-gate 	pp->p_prev = pp->p_next = pp;		/* make pp a list of one */
37547c478bd9Sstevel@tonic-gate }
37557c478bd9Sstevel@tonic-gate 
37567c478bd9Sstevel@tonic-gate 
37577c478bd9Sstevel@tonic-gate /*
37587c478bd9Sstevel@tonic-gate  * Break page list cppp into two lists with npages in the first list.
37597c478bd9Sstevel@tonic-gate  * The tail is returned in nppp.
37607c478bd9Sstevel@tonic-gate  */
37617c478bd9Sstevel@tonic-gate void
37627c478bd9Sstevel@tonic-gate page_list_break(page_t **oppp, page_t **nppp, pgcnt_t npages)
37637c478bd9Sstevel@tonic-gate {
37647c478bd9Sstevel@tonic-gate 	page_t *s1pp = *oppp;
37657c478bd9Sstevel@tonic-gate 	page_t *s2pp;
37667c478bd9Sstevel@tonic-gate 	page_t *e1pp, *e2pp;
37677c478bd9Sstevel@tonic-gate 	long n = 0;
37687c478bd9Sstevel@tonic-gate 
37697c478bd9Sstevel@tonic-gate 	if (s1pp == NULL) {
37707c478bd9Sstevel@tonic-gate 		*nppp = NULL;
37717c478bd9Sstevel@tonic-gate 		return;
37727c478bd9Sstevel@tonic-gate 	}
37737c478bd9Sstevel@tonic-gate 	if (npages == 0) {
37747c478bd9Sstevel@tonic-gate 		*nppp = s1pp;
37757c478bd9Sstevel@tonic-gate 		*oppp = NULL;
37767c478bd9Sstevel@tonic-gate 		return;
37777c478bd9Sstevel@tonic-gate 	}
37787c478bd9Sstevel@tonic-gate 	for (n = 0, s2pp = *oppp; n < npages; n++) {
37797c478bd9Sstevel@tonic-gate 		s2pp = s2pp->p_next;
37807c478bd9Sstevel@tonic-gate 	}
37817c478bd9Sstevel@tonic-gate 	/* Fix head and tail of new lists */
37827c478bd9Sstevel@tonic-gate 	e1pp = s2pp->p_prev;
37837c478bd9Sstevel@tonic-gate 	e2pp = s1pp->p_prev;
37847c478bd9Sstevel@tonic-gate 	s1pp->p_prev = e1pp;
37857c478bd9Sstevel@tonic-gate 	e1pp->p_next = s1pp;
37867c478bd9Sstevel@tonic-gate 	s2pp->p_prev = e2pp;
37877c478bd9Sstevel@tonic-gate 	e2pp->p_next = s2pp;
37887c478bd9Sstevel@tonic-gate 
37897c478bd9Sstevel@tonic-gate 	/* second list empty */
37907c478bd9Sstevel@tonic-gate 	if (s2pp == s1pp) {
37917c478bd9Sstevel@tonic-gate 		*oppp = s1pp;
37927c478bd9Sstevel@tonic-gate 		*nppp = NULL;
37937c478bd9Sstevel@tonic-gate 	} else {
37947c478bd9Sstevel@tonic-gate 		*oppp = s1pp;
37957c478bd9Sstevel@tonic-gate 		*nppp = s2pp;
37967c478bd9Sstevel@tonic-gate 	}
37977c478bd9Sstevel@tonic-gate }
37987c478bd9Sstevel@tonic-gate 
37997c478bd9Sstevel@tonic-gate /*
38007c478bd9Sstevel@tonic-gate  * Concatenate page list nppp onto the end of list ppp.
38017c478bd9Sstevel@tonic-gate  */
38027c478bd9Sstevel@tonic-gate void
38037c478bd9Sstevel@tonic-gate page_list_concat(page_t **ppp, page_t **nppp)
38047c478bd9Sstevel@tonic-gate {
38057c478bd9Sstevel@tonic-gate 	page_t *s1pp, *s2pp, *e1pp, *e2pp;
38067c478bd9Sstevel@tonic-gate 
38077c478bd9Sstevel@tonic-gate 	if (*nppp == NULL) {
38087c478bd9Sstevel@tonic-gate 		return;
38097c478bd9Sstevel@tonic-gate 	}
38107c478bd9Sstevel@tonic-gate 	if (*ppp == NULL) {
38117c478bd9Sstevel@tonic-gate 		*ppp = *nppp;
38127c478bd9Sstevel@tonic-gate 		return;
38137c478bd9Sstevel@tonic-gate 	}
38147c478bd9Sstevel@tonic-gate 	s1pp = *ppp;
38157c478bd9Sstevel@tonic-gate 	e1pp =  s1pp->p_prev;
38167c478bd9Sstevel@tonic-gate 	s2pp = *nppp;
38177c478bd9Sstevel@tonic-gate 	e2pp = s2pp->p_prev;
38187c478bd9Sstevel@tonic-gate 	s1pp->p_prev = e2pp;
38197c478bd9Sstevel@tonic-gate 	e2pp->p_next = s1pp;
38207c478bd9Sstevel@tonic-gate 	e1pp->p_next = s2pp;
38217c478bd9Sstevel@tonic-gate 	s2pp->p_prev = e1pp;
38227c478bd9Sstevel@tonic-gate }
38237c478bd9Sstevel@tonic-gate 
38247c478bd9Sstevel@tonic-gate /*
38257c478bd9Sstevel@tonic-gate  * return the next page in the page list
38267c478bd9Sstevel@tonic-gate  */
38277c478bd9Sstevel@tonic-gate page_t *
38287c478bd9Sstevel@tonic-gate page_list_next(page_t *pp)
38297c478bd9Sstevel@tonic-gate {
38307c478bd9Sstevel@tonic-gate 	return (pp->p_next);
38317c478bd9Sstevel@tonic-gate }
38327c478bd9Sstevel@tonic-gate 
38337c478bd9Sstevel@tonic-gate 
38347c478bd9Sstevel@tonic-gate /*
38357c478bd9Sstevel@tonic-gate  * Add the page to the front of the linked list of pages
38367c478bd9Sstevel@tonic-gate  * using p_vpnext/p_vpprev pointers for the list.
38377c478bd9Sstevel@tonic-gate  *
38387c478bd9Sstevel@tonic-gate  * The caller is responsible for protecting the lists.
38397c478bd9Sstevel@tonic-gate  */
38407c478bd9Sstevel@tonic-gate void
38417c478bd9Sstevel@tonic-gate page_vpadd(page_t **ppp, page_t *pp)
38427c478bd9Sstevel@tonic-gate {
38437c478bd9Sstevel@tonic-gate 	if (*ppp == NULL) {
38447c478bd9Sstevel@tonic-gate 		pp->p_vpnext = pp->p_vpprev = pp;
38457c478bd9Sstevel@tonic-gate 	} else {
38467c478bd9Sstevel@tonic-gate 		pp->p_vpnext = *ppp;
38477c478bd9Sstevel@tonic-gate 		pp->p_vpprev = (*ppp)->p_vpprev;
38487c478bd9Sstevel@tonic-gate 		(*ppp)->p_vpprev = pp;
38497c478bd9Sstevel@tonic-gate 		pp->p_vpprev->p_vpnext = pp;
38507c478bd9Sstevel@tonic-gate 	}
38517c478bd9Sstevel@tonic-gate 	*ppp = pp;
38527c478bd9Sstevel@tonic-gate }
38537c478bd9Sstevel@tonic-gate 
38547c478bd9Sstevel@tonic-gate /*
38557c478bd9Sstevel@tonic-gate  * Remove this page from the linked list of pages
38567c478bd9Sstevel@tonic-gate  * using p_vpnext/p_vpprev pointers for the list.
38577c478bd9Sstevel@tonic-gate  *
38587c478bd9Sstevel@tonic-gate  * The caller is responsible for protecting the lists.
38597c478bd9Sstevel@tonic-gate  */
38607c478bd9Sstevel@tonic-gate void
38617c478bd9Sstevel@tonic-gate page_vpsub(page_t **ppp, page_t *pp)
38627c478bd9Sstevel@tonic-gate {
38637c478bd9Sstevel@tonic-gate 	if (*ppp == NULL || pp == NULL) {
38647c478bd9Sstevel@tonic-gate 		panic("page_vpsub: bad arg(s): pp %p, *ppp %p",
38657c478bd9Sstevel@tonic-gate 		    (void *)pp, (void *)(*ppp));
38667c478bd9Sstevel@tonic-gate 		/*NOTREACHED*/
38677c478bd9Sstevel@tonic-gate 	}
38687c478bd9Sstevel@tonic-gate 
38697c478bd9Sstevel@tonic-gate 	if (*ppp == pp)
38707c478bd9Sstevel@tonic-gate 		*ppp = pp->p_vpnext;		/* go to next page */
38717c478bd9Sstevel@tonic-gate 
38727c478bd9Sstevel@tonic-gate 	if (*ppp == pp)
38737c478bd9Sstevel@tonic-gate 		*ppp = NULL;			/* page list is gone */
38747c478bd9Sstevel@tonic-gate 	else {
38757c478bd9Sstevel@tonic-gate 		pp->p_vpprev->p_vpnext = pp->p_vpnext;
38767c478bd9Sstevel@tonic-gate 		pp->p_vpnext->p_vpprev = pp->p_vpprev;
38777c478bd9Sstevel@tonic-gate 	}
38787c478bd9Sstevel@tonic-gate 	pp->p_vpprev = pp->p_vpnext = pp;	/* make pp a list of one */
38797c478bd9Sstevel@tonic-gate }
38807c478bd9Sstevel@tonic-gate 
38817c478bd9Sstevel@tonic-gate /*
38827c478bd9Sstevel@tonic-gate  * Lock a physical page into memory "long term".  Used to support "lock
38837c478bd9Sstevel@tonic-gate  * in memory" functions.  Accepts the page to be locked, and a cow variable
38847c478bd9Sstevel@tonic-gate  * to indicate whether a the lock will travel to the new page during
38857c478bd9Sstevel@tonic-gate  * a potential copy-on-write.
38867c478bd9Sstevel@tonic-gate  */
38877c478bd9Sstevel@tonic-gate int
38887c478bd9Sstevel@tonic-gate page_pp_lock(
38897c478bd9Sstevel@tonic-gate 	page_t *pp,			/* page to be locked */
38907c478bd9Sstevel@tonic-gate 	int cow,			/* cow lock */
38917c478bd9Sstevel@tonic-gate 	int kernel)			/* must succeed -- ignore checking */
38927c478bd9Sstevel@tonic-gate {
38937c478bd9Sstevel@tonic-gate 	int r = 0;			/* result -- assume failure */
38947c478bd9Sstevel@tonic-gate 
38957c478bd9Sstevel@tonic-gate 	ASSERT(PAGE_LOCKED(pp));
38967c478bd9Sstevel@tonic-gate 
38977c478bd9Sstevel@tonic-gate 	page_struct_lock(pp);
38987c478bd9Sstevel@tonic-gate 	/*
38997c478bd9Sstevel@tonic-gate 	 * Acquire the "freemem_lock" for availrmem.
39007c478bd9Sstevel@tonic-gate 	 */
39017c478bd9Sstevel@tonic-gate 	if (cow) {
39027c478bd9Sstevel@tonic-gate 		mutex_enter(&freemem_lock);
39037c478bd9Sstevel@tonic-gate 		if ((availrmem > pages_pp_maximum) &&
39047c478bd9Sstevel@tonic-gate 		    (pp->p_cowcnt < (ushort_t)PAGE_LOCK_MAXIMUM)) {
39057c478bd9Sstevel@tonic-gate 			availrmem--;
39067c478bd9Sstevel@tonic-gate 			pages_locked++;
39077c478bd9Sstevel@tonic-gate 			mutex_exit(&freemem_lock);
39087c478bd9Sstevel@tonic-gate 			r = 1;
39097c478bd9Sstevel@tonic-gate 			if (++pp->p_cowcnt == (ushort_t)PAGE_LOCK_MAXIMUM) {
39107c478bd9Sstevel@tonic-gate 				cmn_err(CE_WARN,
39117c478bd9Sstevel@tonic-gate 				    "COW lock limit reached on pfn 0x%lx",
39127c478bd9Sstevel@tonic-gate 				    page_pptonum(pp));
39137c478bd9Sstevel@tonic-gate 			}
39147c478bd9Sstevel@tonic-gate 		} else
39157c478bd9Sstevel@tonic-gate 			mutex_exit(&freemem_lock);
39167c478bd9Sstevel@tonic-gate 	} else {
39177c478bd9Sstevel@tonic-gate 		if (pp->p_lckcnt) {
39187c478bd9Sstevel@tonic-gate 			if (pp->p_lckcnt < (ushort_t)PAGE_LOCK_MAXIMUM) {
39197c478bd9Sstevel@tonic-gate 				r = 1;
39207c478bd9Sstevel@tonic-gate 				if (++pp->p_lckcnt ==
39217c478bd9Sstevel@tonic-gate 				    (ushort_t)PAGE_LOCK_MAXIMUM) {
39227c478bd9Sstevel@tonic-gate 					cmn_err(CE_WARN, "Page lock limit "
39237c478bd9Sstevel@tonic-gate 					    "reached on pfn 0x%lx",
39247c478bd9Sstevel@tonic-gate 					    page_pptonum(pp));
39257c478bd9Sstevel@tonic-gate 				}
39267c478bd9Sstevel@tonic-gate 			}
39277c478bd9Sstevel@tonic-gate 		} else {
39287c478bd9Sstevel@tonic-gate 			if (kernel) {
39297c478bd9Sstevel@tonic-gate 				/* availrmem accounting done by caller */
39307c478bd9Sstevel@tonic-gate 				++pp->p_lckcnt;
39317c478bd9Sstevel@tonic-gate 				r = 1;
39327c478bd9Sstevel@tonic-gate 			} else {
39337c478bd9Sstevel@tonic-gate 				mutex_enter(&freemem_lock);
39347c478bd9Sstevel@tonic-gate 				if (availrmem > pages_pp_maximum) {
39357c478bd9Sstevel@tonic-gate 					availrmem--;
39367c478bd9Sstevel@tonic-gate 					pages_locked++;
39377c478bd9Sstevel@tonic-gate 					++pp->p_lckcnt;
39387c478bd9Sstevel@tonic-gate 					r = 1;
39397c478bd9Sstevel@tonic-gate 				}
39407c478bd9Sstevel@tonic-gate 				mutex_exit(&freemem_lock);
39417c478bd9Sstevel@tonic-gate 			}
39427c478bd9Sstevel@tonic-gate 		}
39437c478bd9Sstevel@tonic-gate 	}
39447c478bd9Sstevel@tonic-gate 	page_struct_unlock(pp);
39457c478bd9Sstevel@tonic-gate 	return (r);
39467c478bd9Sstevel@tonic-gate }
39477c478bd9Sstevel@tonic-gate 
39487c478bd9Sstevel@tonic-gate /*
39497c478bd9Sstevel@tonic-gate  * Decommit a lock on a physical page frame.  Account for cow locks if
39507c478bd9Sstevel@tonic-gate  * appropriate.
39517c478bd9Sstevel@tonic-gate  */
39527c478bd9Sstevel@tonic-gate void
39537c478bd9Sstevel@tonic-gate page_pp_unlock(
39547c478bd9Sstevel@tonic-gate 	page_t *pp,			/* page to be unlocked */
39557c478bd9Sstevel@tonic-gate 	int cow,			/* expect cow lock */
39567c478bd9Sstevel@tonic-gate 	int kernel)			/* this was a kernel lock */
39577c478bd9Sstevel@tonic-gate {
39587c478bd9Sstevel@tonic-gate 	ASSERT(PAGE_LOCKED(pp));
39597c478bd9Sstevel@tonic-gate 
39607c478bd9Sstevel@tonic-gate 	page_struct_lock(pp);
39617c478bd9Sstevel@tonic-gate 	/*
39627c478bd9Sstevel@tonic-gate 	 * Acquire the "freemem_lock" for availrmem.
39637c478bd9Sstevel@tonic-gate 	 * If cowcnt or lcknt is already 0 do nothing; i.e., we
39647c478bd9Sstevel@tonic-gate 	 * could be called to unlock even if nothing is locked. This could
39657c478bd9Sstevel@tonic-gate 	 * happen if locked file pages were truncated (removing the lock)
39667c478bd9Sstevel@tonic-gate 	 * and the file was grown again and new pages faulted in; the new
39677c478bd9Sstevel@tonic-gate 	 * pages are unlocked but the segment still thinks they're locked.
39687c478bd9Sstevel@tonic-gate 	 */
39697c478bd9Sstevel@tonic-gate 	if (cow) {
39707c478bd9Sstevel@tonic-gate 		if (pp->p_cowcnt) {
39717c478bd9Sstevel@tonic-gate 			mutex_enter(&freemem_lock);
39727c478bd9Sstevel@tonic-gate 			pp->p_cowcnt--;
39737c478bd9Sstevel@tonic-gate 			availrmem++;
39747c478bd9Sstevel@tonic-gate 			pages_locked--;
39757c478bd9Sstevel@tonic-gate 			mutex_exit(&freemem_lock);
39767c478bd9Sstevel@tonic-gate 		}
39777c478bd9Sstevel@tonic-gate 	} else {
39787c478bd9Sstevel@tonic-gate 		if (pp->p_lckcnt && --pp->p_lckcnt == 0) {
39797c478bd9Sstevel@tonic-gate 			if (!kernel) {
39807c478bd9Sstevel@tonic-gate 				mutex_enter(&freemem_lock);
39817c478bd9Sstevel@tonic-gate 				availrmem++;
39827c478bd9Sstevel@tonic-gate 				pages_locked--;
39837c478bd9Sstevel@tonic-gate 				mutex_exit(&freemem_lock);
39847c478bd9Sstevel@tonic-gate 			}
39857c478bd9Sstevel@tonic-gate 		}
39867c478bd9Sstevel@tonic-gate 	}
39877c478bd9Sstevel@tonic-gate 	page_struct_unlock(pp);
39887c478bd9Sstevel@tonic-gate }
39897c478bd9Sstevel@tonic-gate 
39907c478bd9Sstevel@tonic-gate /*
39917c478bd9Sstevel@tonic-gate  * This routine reserves availrmem for npages;
39927c478bd9Sstevel@tonic-gate  * 	flags: KM_NOSLEEP or KM_SLEEP
39937c478bd9Sstevel@tonic-gate  * 	returns 1 on success or 0 on failure
39947c478bd9Sstevel@tonic-gate  */
39957c478bd9Sstevel@tonic-gate int
39967c478bd9Sstevel@tonic-gate page_resv(pgcnt_t npages, uint_t flags)
39977c478bd9Sstevel@tonic-gate {
39987c478bd9Sstevel@tonic-gate 	mutex_enter(&freemem_lock);
39997c478bd9Sstevel@tonic-gate 	while (availrmem < tune.t_minarmem + npages) {
40007c478bd9Sstevel@tonic-gate 		if (flags & KM_NOSLEEP) {
40017c478bd9Sstevel@tonic-gate 			mutex_exit(&freemem_lock);
40027c478bd9Sstevel@tonic-gate 			return (0);
40037c478bd9Sstevel@tonic-gate 		}
40047c478bd9Sstevel@tonic-gate 		mutex_exit(&freemem_lock);
40057c478bd9Sstevel@tonic-gate 		page_needfree(npages);
40067c478bd9Sstevel@tonic-gate 		kmem_reap();
40077c478bd9Sstevel@tonic-gate 		delay(hz >> 2);
40087c478bd9Sstevel@tonic-gate 		page_needfree(-(spgcnt_t)npages);
40097c478bd9Sstevel@tonic-gate 		mutex_enter(&freemem_lock);
40107c478bd9Sstevel@tonic-gate 	}
40117c478bd9Sstevel@tonic-gate 	availrmem -= npages;
40127c478bd9Sstevel@tonic-gate 	mutex_exit(&freemem_lock);
40137c478bd9Sstevel@tonic-gate 	return (1);
40147c478bd9Sstevel@tonic-gate }
40157c478bd9Sstevel@tonic-gate 
40167c478bd9Sstevel@tonic-gate /*
40177c478bd9Sstevel@tonic-gate  * This routine unreserves availrmem for npages;
40187c478bd9Sstevel@tonic-gate  */
40197c478bd9Sstevel@tonic-gate void
40207c478bd9Sstevel@tonic-gate page_unresv(pgcnt_t npages)
40217c478bd9Sstevel@tonic-gate {
40227c478bd9Sstevel@tonic-gate 	mutex_enter(&freemem_lock);
40237c478bd9Sstevel@tonic-gate 	availrmem += npages;
40247c478bd9Sstevel@tonic-gate 	mutex_exit(&freemem_lock);
40257c478bd9Sstevel@tonic-gate }
40267c478bd9Sstevel@tonic-gate 
40277c478bd9Sstevel@tonic-gate /*
40287c478bd9Sstevel@tonic-gate  * See Statement at the beginning of segvn_lockop() regarding
40297c478bd9Sstevel@tonic-gate  * the way we handle cowcnts and lckcnts.
40307c478bd9Sstevel@tonic-gate  *
40317c478bd9Sstevel@tonic-gate  * Transfer cowcnt on 'opp' to cowcnt on 'npp' if the vpage
40327c478bd9Sstevel@tonic-gate  * that breaks COW has PROT_WRITE.
40337c478bd9Sstevel@tonic-gate  *
40347c478bd9Sstevel@tonic-gate  * Note that, we may also break COW in case we are softlocking
40357c478bd9Sstevel@tonic-gate  * on read access during physio;
40367c478bd9Sstevel@tonic-gate  * in this softlock case, the vpage may not have PROT_WRITE.
40377c478bd9Sstevel@tonic-gate  * So, we need to transfer lckcnt on 'opp' to lckcnt on 'npp'
40387c478bd9Sstevel@tonic-gate  * if the vpage doesn't have PROT_WRITE.
40397c478bd9Sstevel@tonic-gate  *
40407c478bd9Sstevel@tonic-gate  * This routine is never called if we are stealing a page
40417c478bd9Sstevel@tonic-gate  * in anon_private.
40427c478bd9Sstevel@tonic-gate  *
40437c478bd9Sstevel@tonic-gate  * The caller subtracted from availrmem for read only mapping.
40447c478bd9Sstevel@tonic-gate  * if lckcnt is 1 increment availrmem.
40457c478bd9Sstevel@tonic-gate  */
40467c478bd9Sstevel@tonic-gate void
40477c478bd9Sstevel@tonic-gate page_pp_useclaim(
40487c478bd9Sstevel@tonic-gate 	page_t *opp,		/* original page frame losing lock */
40497c478bd9Sstevel@tonic-gate 	page_t *npp,		/* new page frame gaining lock */
40507c478bd9Sstevel@tonic-gate 	uint_t	write_perm) 	/* set if vpage has PROT_WRITE */
40517c478bd9Sstevel@tonic-gate {
40527c478bd9Sstevel@tonic-gate 	int payback = 0;
40537c478bd9Sstevel@tonic-gate 
40547c478bd9Sstevel@tonic-gate 	ASSERT(PAGE_LOCKED(opp));
40557c478bd9Sstevel@tonic-gate 	ASSERT(PAGE_LOCKED(npp));
40567c478bd9Sstevel@tonic-gate 
40577c478bd9Sstevel@tonic-gate 	page_struct_lock(opp);
40587c478bd9Sstevel@tonic-gate 
40597c478bd9Sstevel@tonic-gate 	ASSERT(npp->p_cowcnt == 0);
40607c478bd9Sstevel@tonic-gate 	ASSERT(npp->p_lckcnt == 0);
40617c478bd9Sstevel@tonic-gate 
40627c478bd9Sstevel@tonic-gate 	/* Don't use claim if nothing is locked (see page_pp_unlock above) */
40637c478bd9Sstevel@tonic-gate 	if ((write_perm && opp->p_cowcnt != 0) ||
40647c478bd9Sstevel@tonic-gate 	    (!write_perm && opp->p_lckcnt != 0)) {
40657c478bd9Sstevel@tonic-gate 
40667c478bd9Sstevel@tonic-gate 		if (write_perm) {
40677c478bd9Sstevel@tonic-gate 			npp->p_cowcnt++;
40687c478bd9Sstevel@tonic-gate 			ASSERT(opp->p_cowcnt != 0);
40697c478bd9Sstevel@tonic-gate 			opp->p_cowcnt--;
40707c478bd9Sstevel@tonic-gate 		} else {
40717c478bd9Sstevel@tonic-gate 
40727c478bd9Sstevel@tonic-gate 			ASSERT(opp->p_lckcnt != 0);
40737c478bd9Sstevel@tonic-gate 
40747c478bd9Sstevel@tonic-gate 			/*
40757c478bd9Sstevel@tonic-gate 			 * We didn't need availrmem decremented if p_lckcnt on
40767c478bd9Sstevel@tonic-gate 			 * original page is 1. Here, we are unlocking
40777c478bd9Sstevel@tonic-gate 			 * read-only copy belonging to original page and
40787c478bd9Sstevel@tonic-gate 			 * are locking a copy belonging to new page.
40797c478bd9Sstevel@tonic-gate 			 */
40807c478bd9Sstevel@tonic-gate 			if (opp->p_lckcnt == 1)
40817c478bd9Sstevel@tonic-gate 				payback = 1;
40827c478bd9Sstevel@tonic-gate 
40837c478bd9Sstevel@tonic-gate 			npp->p_lckcnt++;
40847c478bd9Sstevel@tonic-gate 			opp->p_lckcnt--;
40857c478bd9Sstevel@tonic-gate 		}
40867c478bd9Sstevel@tonic-gate 	}
40877c478bd9Sstevel@tonic-gate 	if (payback) {
40887c478bd9Sstevel@tonic-gate 		mutex_enter(&freemem_lock);
40897c478bd9Sstevel@tonic-gate 		availrmem++;
40907c478bd9Sstevel@tonic-gate 		pages_useclaim--;
40917c478bd9Sstevel@tonic-gate 		mutex_exit(&freemem_lock);
40927c478bd9Sstevel@tonic-gate 	}
40937c478bd9Sstevel@tonic-gate 	page_struct_unlock(opp);
40947c478bd9Sstevel@tonic-gate }
40957c478bd9Sstevel@tonic-gate 
40967c478bd9Sstevel@tonic-gate /*
40977c478bd9Sstevel@tonic-gate  * Simple claim adjust functions -- used to support changes in
40987c478bd9Sstevel@tonic-gate  * claims due to changes in access permissions.  Used by segvn_setprot().
40997c478bd9Sstevel@tonic-gate  */
41007c478bd9Sstevel@tonic-gate int
41017c478bd9Sstevel@tonic-gate page_addclaim(page_t *pp)
41027c478bd9Sstevel@tonic-gate {
41037c478bd9Sstevel@tonic-gate 	int r = 0;			/* result */
41047c478bd9Sstevel@tonic-gate 
41057c478bd9Sstevel@tonic-gate 	ASSERT(PAGE_LOCKED(pp));
41067c478bd9Sstevel@tonic-gate 
41077c478bd9Sstevel@tonic-gate 	page_struct_lock(pp);
41087c478bd9Sstevel@tonic-gate 	ASSERT(pp->p_lckcnt != 0);
41097c478bd9Sstevel@tonic-gate 
41107c478bd9Sstevel@tonic-gate 	if (pp->p_lckcnt == 1) {
41117c478bd9Sstevel@tonic-gate 		if (pp->p_cowcnt < (ushort_t)PAGE_LOCK_MAXIMUM) {
41127c478bd9Sstevel@tonic-gate 			--pp->p_lckcnt;
41137c478bd9Sstevel@tonic-gate 			r = 1;
41147c478bd9Sstevel@tonic-gate 			if (++pp->p_cowcnt == (ushort_t)PAGE_LOCK_MAXIMUM) {
41157c478bd9Sstevel@tonic-gate 				cmn_err(CE_WARN,
41167c478bd9Sstevel@tonic-gate 				    "COW lock limit reached on pfn 0x%lx",
41177c478bd9Sstevel@tonic-gate 				    page_pptonum(pp));
41187c478bd9Sstevel@tonic-gate 			}
41197c478bd9Sstevel@tonic-gate 		}
41207c478bd9Sstevel@tonic-gate 	} else {
41217c478bd9Sstevel@tonic-gate 		mutex_enter(&freemem_lock);
41227c478bd9Sstevel@tonic-gate 		if ((availrmem > pages_pp_maximum) &&
41237c478bd9Sstevel@tonic-gate 		    (pp->p_cowcnt < (ushort_t)PAGE_LOCK_MAXIMUM)) {
41247c478bd9Sstevel@tonic-gate 			--availrmem;
41257c478bd9Sstevel@tonic-gate 			++pages_claimed;
41267c478bd9Sstevel@tonic-gate 			mutex_exit(&freemem_lock);
41277c478bd9Sstevel@tonic-gate 			--pp->p_lckcnt;
41287c478bd9Sstevel@tonic-gate 			r = 1;
41297c478bd9Sstevel@tonic-gate 			if (++pp->p_cowcnt == (ushort_t)PAGE_LOCK_MAXIMUM) {
41307c478bd9Sstevel@tonic-gate 				cmn_err(CE_WARN,
41317c478bd9Sstevel@tonic-gate 				    "COW lock limit reached on pfn 0x%lx",
41327c478bd9Sstevel@tonic-gate 				    page_pptonum(pp));
41337c478bd9Sstevel@tonic-gate 			}
41347c478bd9Sstevel@tonic-gate 		} else
41357c478bd9Sstevel@tonic-gate 			mutex_exit(&freemem_lock);
41367c478bd9Sstevel@tonic-gate 	}
41377c478bd9Sstevel@tonic-gate 	page_struct_unlock(pp);
41387c478bd9Sstevel@tonic-gate 	return (r);
41397c478bd9Sstevel@tonic-gate }
41407c478bd9Sstevel@tonic-gate 
41417c478bd9Sstevel@tonic-gate int
41427c478bd9Sstevel@tonic-gate page_subclaim(page_t *pp)
41437c478bd9Sstevel@tonic-gate {
41447c478bd9Sstevel@tonic-gate 	int r = 0;
41457c478bd9Sstevel@tonic-gate 
41467c478bd9Sstevel@tonic-gate 	ASSERT(PAGE_LOCKED(pp));
41477c478bd9Sstevel@tonic-gate 
41487c478bd9Sstevel@tonic-gate 	page_struct_lock(pp);
41497c478bd9Sstevel@tonic-gate 	ASSERT(pp->p_cowcnt != 0);
41507c478bd9Sstevel@tonic-gate 
41517c478bd9Sstevel@tonic-gate 	if (pp->p_lckcnt) {
41527c478bd9Sstevel@tonic-gate 		if (pp->p_lckcnt < (ushort_t)PAGE_LOCK_MAXIMUM) {
41537c478bd9Sstevel@tonic-gate 			r = 1;
41547c478bd9Sstevel@tonic-gate 			/*
41557c478bd9Sstevel@tonic-gate 			 * for availrmem
41567c478bd9Sstevel@tonic-gate 			 */
41577c478bd9Sstevel@tonic-gate 			mutex_enter(&freemem_lock);
41587c478bd9Sstevel@tonic-gate 			availrmem++;
41597c478bd9Sstevel@tonic-gate 			pages_claimed--;
41607c478bd9Sstevel@tonic-gate 			mutex_exit(&freemem_lock);
41617c478bd9Sstevel@tonic-gate 
41627c478bd9Sstevel@tonic-gate 			pp->p_cowcnt--;
41637c478bd9Sstevel@tonic-gate 
41647c478bd9Sstevel@tonic-gate 			if (++pp->p_lckcnt == (ushort_t)PAGE_LOCK_MAXIMUM) {
41657c478bd9Sstevel@tonic-gate 				cmn_err(CE_WARN,
41667c478bd9Sstevel@tonic-gate 				    "Page lock limit reached on pfn 0x%lx",
41677c478bd9Sstevel@tonic-gate 				    page_pptonum(pp));
41687c478bd9Sstevel@tonic-gate 			}
41697c478bd9Sstevel@tonic-gate 		}
41707c478bd9Sstevel@tonic-gate 	} else {
41717c478bd9Sstevel@tonic-gate 		r = 1;
41727c478bd9Sstevel@tonic-gate 		pp->p_cowcnt--;
41737c478bd9Sstevel@tonic-gate 		pp->p_lckcnt++;
41747c478bd9Sstevel@tonic-gate 	}
41757c478bd9Sstevel@tonic-gate 	page_struct_unlock(pp);
41767c478bd9Sstevel@tonic-gate 	return (r);
41777c478bd9Sstevel@tonic-gate }
41787c478bd9Sstevel@tonic-gate 
41797c478bd9Sstevel@tonic-gate int
41807c478bd9Sstevel@tonic-gate page_addclaim_pages(page_t  **ppa)
41817c478bd9Sstevel@tonic-gate {
41827c478bd9Sstevel@tonic-gate 
41837c478bd9Sstevel@tonic-gate 	pgcnt_t	lckpgs = 0, pg_idx;
41847c478bd9Sstevel@tonic-gate 
41857c478bd9Sstevel@tonic-gate 	VM_STAT_ADD(pagecnt.pc_addclaim_pages);
41867c478bd9Sstevel@tonic-gate 
41877c478bd9Sstevel@tonic-gate 	mutex_enter(&page_llock);
41887c478bd9Sstevel@tonic-gate 	for (pg_idx = 0; ppa[pg_idx] != NULL; pg_idx++) {
41897c478bd9Sstevel@tonic-gate 
41907c478bd9Sstevel@tonic-gate 		ASSERT(PAGE_LOCKED(ppa[pg_idx]));
41917c478bd9Sstevel@tonic-gate 		ASSERT(ppa[pg_idx]->p_lckcnt != 0);
41927c478bd9Sstevel@tonic-gate 		if (ppa[pg_idx]->p_cowcnt == (ushort_t)PAGE_LOCK_MAXIMUM) {
41937c478bd9Sstevel@tonic-gate 			mutex_exit(&page_llock);
41947c478bd9Sstevel@tonic-gate 			return (0);
41957c478bd9Sstevel@tonic-gate 		}
41967c478bd9Sstevel@tonic-gate 		if (ppa[pg_idx]->p_lckcnt > 1)
41977c478bd9Sstevel@tonic-gate 			lckpgs++;
41987c478bd9Sstevel@tonic-gate 	}
41997c478bd9Sstevel@tonic-gate 
42007c478bd9Sstevel@tonic-gate 	if (lckpgs != 0) {
42017c478bd9Sstevel@tonic-gate 		mutex_enter(&freemem_lock);
42027c478bd9Sstevel@tonic-gate 		if (availrmem >= pages_pp_maximum + lckpgs) {
42037c478bd9Sstevel@tonic-gate 			availrmem -= lckpgs;
42047c478bd9Sstevel@tonic-gate 			pages_claimed += lckpgs;
42057c478bd9Sstevel@tonic-gate 		} else {
42067c478bd9Sstevel@tonic-gate 			mutex_exit(&freemem_lock);
42077c478bd9Sstevel@tonic-gate 			mutex_exit(&page_llock);
42087c478bd9Sstevel@tonic-gate 			return (0);
42097c478bd9Sstevel@tonic-gate 		}
42107c478bd9Sstevel@tonic-gate 		mutex_exit(&freemem_lock);
42117c478bd9Sstevel@tonic-gate 	}
42127c478bd9Sstevel@tonic-gate 
42137c478bd9Sstevel@tonic-gate 	for (pg_idx = 0; ppa[pg_idx] != NULL; pg_idx++) {
42147c478bd9Sstevel@tonic-gate 		ppa[pg_idx]->p_lckcnt--;
42157c478bd9Sstevel@tonic-gate 		ppa[pg_idx]->p_cowcnt++;
42167c478bd9Sstevel@tonic-gate 	}
42177c478bd9Sstevel@tonic-gate 	mutex_exit(&page_llock);
42187c478bd9Sstevel@tonic-gate 	return (1);
42197c478bd9Sstevel@tonic-gate }
42207c478bd9Sstevel@tonic-gate 
42217c478bd9Sstevel@tonic-gate int
42227c478bd9Sstevel@tonic-gate page_subclaim_pages(page_t  **ppa)
42237c478bd9Sstevel@tonic-gate {
42247c478bd9Sstevel@tonic-gate 	pgcnt_t	ulckpgs = 0, pg_idx;
42257c478bd9Sstevel@tonic-gate 
42267c478bd9Sstevel@tonic-gate 	VM_STAT_ADD(pagecnt.pc_subclaim_pages);
42277c478bd9Sstevel@tonic-gate 
42287c478bd9Sstevel@tonic-gate 	mutex_enter(&page_llock);
42297c478bd9Sstevel@tonic-gate 	for (pg_idx = 0; ppa[pg_idx] != NULL; pg_idx++) {
42307c478bd9Sstevel@tonic-gate 
42317c478bd9Sstevel@tonic-gate 		ASSERT(PAGE_LOCKED(ppa[pg_idx]));
42327c478bd9Sstevel@tonic-gate 		ASSERT(ppa[pg_idx]->p_cowcnt != 0);
42337c478bd9Sstevel@tonic-gate 		if (ppa[pg_idx]->p_lckcnt == (ushort_t)PAGE_LOCK_MAXIMUM) {
42347c478bd9Sstevel@tonic-gate 			mutex_exit(&page_llock);
42357c478bd9Sstevel@tonic-gate 			return (0);
42367c478bd9Sstevel@tonic-gate 		}
42377c478bd9Sstevel@tonic-gate 		if (ppa[pg_idx]->p_lckcnt != 0)
42387c478bd9Sstevel@tonic-gate 			ulckpgs++;
42397c478bd9Sstevel@tonic-gate 	}
42407c478bd9Sstevel@tonic-gate 
42417c478bd9Sstevel@tonic-gate 	if (ulckpgs != 0) {
42427c478bd9Sstevel@tonic-gate 		mutex_enter(&freemem_lock);
42437c478bd9Sstevel@tonic-gate 		availrmem += ulckpgs;
42447c478bd9Sstevel@tonic-gate 		pages_claimed -= ulckpgs;
42457c478bd9Sstevel@tonic-gate 		mutex_exit(&freemem_lock);
42467c478bd9Sstevel@tonic-gate 	}
42477c478bd9Sstevel@tonic-gate 
42487c478bd9Sstevel@tonic-gate 	for (pg_idx = 0; ppa[pg_idx] != NULL; pg_idx++) {
42497c478bd9Sstevel@tonic-gate 		ppa[pg_idx]->p_cowcnt--;
42507c478bd9Sstevel@tonic-gate 		ppa[pg_idx]->p_lckcnt++;
42517c478bd9Sstevel@tonic-gate 
42527c478bd9Sstevel@tonic-gate 	}
42537c478bd9Sstevel@tonic-gate 	mutex_exit(&page_llock);
42547c478bd9Sstevel@tonic-gate 	return (1);
42557c478bd9Sstevel@tonic-gate }
42567c478bd9Sstevel@tonic-gate 
42577c478bd9Sstevel@tonic-gate page_t *
42587c478bd9Sstevel@tonic-gate page_numtopp(pfn_t pfnum, se_t se)
42597c478bd9Sstevel@tonic-gate {
42607c478bd9Sstevel@tonic-gate 	page_t *pp;
42617c478bd9Sstevel@tonic-gate 
42627c478bd9Sstevel@tonic-gate retry:
42637c478bd9Sstevel@tonic-gate 	pp = page_numtopp_nolock(pfnum);
42647c478bd9Sstevel@tonic-gate 	if (pp == NULL) {
42657c478bd9Sstevel@tonic-gate 		return ((page_t *)NULL);
42667c478bd9Sstevel@tonic-gate 	}
42677c478bd9Sstevel@tonic-gate 
42687c478bd9Sstevel@tonic-gate 	/*
42697c478bd9Sstevel@tonic-gate 	 * Acquire the appropriate lock on the page.
42707c478bd9Sstevel@tonic-gate 	 */
42717c478bd9Sstevel@tonic-gate 	while (!page_lock(pp, se, (kmutex_t *)NULL, P_RECLAIM)) {
42727c478bd9Sstevel@tonic-gate 		if (page_pptonum(pp) != pfnum)
42737c478bd9Sstevel@tonic-gate 			goto retry;
42747c478bd9Sstevel@tonic-gate 		continue;
42757c478bd9Sstevel@tonic-gate 	}
42767c478bd9Sstevel@tonic-gate 
42777c478bd9Sstevel@tonic-gate 	if (page_pptonum(pp) != pfnum) {
42787c478bd9Sstevel@tonic-gate 		page_unlock(pp);
42797c478bd9Sstevel@tonic-gate 		goto retry;
42807c478bd9Sstevel@tonic-gate 	}
42817c478bd9Sstevel@tonic-gate 
42827c478bd9Sstevel@tonic-gate 	return (pp);
42837c478bd9Sstevel@tonic-gate }
42847c478bd9Sstevel@tonic-gate 
42857c478bd9Sstevel@tonic-gate page_t *
42867c478bd9Sstevel@tonic-gate page_numtopp_noreclaim(pfn_t pfnum, se_t se)
42877c478bd9Sstevel@tonic-gate {
42887c478bd9Sstevel@tonic-gate 	page_t *pp;
42897c478bd9Sstevel@tonic-gate 
42907c478bd9Sstevel@tonic-gate retry:
42917c478bd9Sstevel@tonic-gate 	pp = page_numtopp_nolock(pfnum);
42927c478bd9Sstevel@tonic-gate 	if (pp == NULL) {
42937c478bd9Sstevel@tonic-gate 		return ((page_t *)NULL);
42947c478bd9Sstevel@tonic-gate 	}
42957c478bd9Sstevel@tonic-gate 
42967c478bd9Sstevel@tonic-gate 	/*
42977c478bd9Sstevel@tonic-gate 	 * Acquire the appropriate lock on the page.
42987c478bd9Sstevel@tonic-gate 	 */
42997c478bd9Sstevel@tonic-gate 	while (!page_lock(pp, se, (kmutex_t *)NULL, P_NO_RECLAIM)) {
43007c478bd9Sstevel@tonic-gate 		if (page_pptonum(pp) != pfnum)
43017c478bd9Sstevel@tonic-gate 			goto retry;
43027c478bd9Sstevel@tonic-gate 		continue;
43037c478bd9Sstevel@tonic-gate 	}
43047c478bd9Sstevel@tonic-gate 
43057c478bd9Sstevel@tonic-gate 	if (page_pptonum(pp) != pfnum) {
43067c478bd9Sstevel@tonic-gate 		page_unlock(pp);
43077c478bd9Sstevel@tonic-gate 		goto retry;
43087c478bd9Sstevel@tonic-gate 	}
43097c478bd9Sstevel@tonic-gate 
43107c478bd9Sstevel@tonic-gate 	return (pp);
43117c478bd9Sstevel@tonic-gate }
43127c478bd9Sstevel@tonic-gate 
43137c478bd9Sstevel@tonic-gate /*
43147c478bd9Sstevel@tonic-gate  * This routine is like page_numtopp, but will only return page structs
43157c478bd9Sstevel@tonic-gate  * for pages which are ok for loading into hardware using the page struct.
43167c478bd9Sstevel@tonic-gate  */
43177c478bd9Sstevel@tonic-gate page_t *
43187c478bd9Sstevel@tonic-gate page_numtopp_nowait(pfn_t pfnum, se_t se)
43197c478bd9Sstevel@tonic-gate {
43207c478bd9Sstevel@tonic-gate 	page_t *pp;
43217c478bd9Sstevel@tonic-gate 
43227c478bd9Sstevel@tonic-gate retry:
43237c478bd9Sstevel@tonic-gate 	pp = page_numtopp_nolock(pfnum);
43247c478bd9Sstevel@tonic-gate 	if (pp == NULL) {
43257c478bd9Sstevel@tonic-gate 		return ((page_t *)NULL);
43267c478bd9Sstevel@tonic-gate 	}
43277c478bd9Sstevel@tonic-gate 
43287c478bd9Sstevel@tonic-gate 	/*
43297c478bd9Sstevel@tonic-gate 	 * Try to acquire the appropriate lock on the page.
43307c478bd9Sstevel@tonic-gate 	 */
43317c478bd9Sstevel@tonic-gate 	if (PP_ISFREE(pp))
43327c478bd9Sstevel@tonic-gate 		pp = NULL;
43337c478bd9Sstevel@tonic-gate 	else {
43347c478bd9Sstevel@tonic-gate 		if (!page_trylock(pp, se))
43357c478bd9Sstevel@tonic-gate 			pp = NULL;
43367c478bd9Sstevel@tonic-gate 		else {
43377c478bd9Sstevel@tonic-gate 			if (page_pptonum(pp) != pfnum) {
43387c478bd9Sstevel@tonic-gate 				page_unlock(pp);
43397c478bd9Sstevel@tonic-gate 				goto retry;
43407c478bd9Sstevel@tonic-gate 			}
43417c478bd9Sstevel@tonic-gate 			if (PP_ISFREE(pp)) {
43427c478bd9Sstevel@tonic-gate 				page_unlock(pp);
43437c478bd9Sstevel@tonic-gate 				pp = NULL;
43447c478bd9Sstevel@tonic-gate 			}
43457c478bd9Sstevel@tonic-gate 		}
43467c478bd9Sstevel@tonic-gate 	}
43477c478bd9Sstevel@tonic-gate 	return (pp);
43487c478bd9Sstevel@tonic-gate }
43497c478bd9Sstevel@tonic-gate 
43507c478bd9Sstevel@tonic-gate /*
43517c478bd9Sstevel@tonic-gate  * Returns a count of dirty pages that are in the process
43527c478bd9Sstevel@tonic-gate  * of being written out.  If 'cleanit' is set, try to push the page.
43537c478bd9Sstevel@tonic-gate  */
43547c478bd9Sstevel@tonic-gate pgcnt_t
43557c478bd9Sstevel@tonic-gate page_busy(int cleanit)
43567c478bd9Sstevel@tonic-gate {
43577c478bd9Sstevel@tonic-gate 	page_t *page0 = page_first();
43587c478bd9Sstevel@tonic-gate 	page_t *pp = page0;
43597c478bd9Sstevel@tonic-gate 	pgcnt_t nppbusy = 0;
43607c478bd9Sstevel@tonic-gate 	u_offset_t off;
43617c478bd9Sstevel@tonic-gate 
43627c478bd9Sstevel@tonic-gate 	do {
43637c478bd9Sstevel@tonic-gate 		vnode_t *vp = pp->p_vnode;
43647c478bd9Sstevel@tonic-gate 
43657c478bd9Sstevel@tonic-gate 		/*
43667c478bd9Sstevel@tonic-gate 		 * A page is a candidate for syncing if it is:
43677c478bd9Sstevel@tonic-gate 		 *
43687c478bd9Sstevel@tonic-gate 		 * (a)	On neither the freelist nor the cachelist
43697c478bd9Sstevel@tonic-gate 		 * (b)	Hashed onto a vnode
43707c478bd9Sstevel@tonic-gate 		 * (c)	Not a kernel page
43717c478bd9Sstevel@tonic-gate 		 * (d)	Dirty
43727c478bd9Sstevel@tonic-gate 		 * (e)	Not part of a swapfile
43737c478bd9Sstevel@tonic-gate 		 * (f)	a page which belongs to a real vnode; eg has a non-null
43747c478bd9Sstevel@tonic-gate 		 *	v_vfsp pointer.
43757c478bd9Sstevel@tonic-gate 		 * (g)	Backed by a filesystem which doesn't have a
43767c478bd9Sstevel@tonic-gate 		 *	stubbed-out sync operation
43777c478bd9Sstevel@tonic-gate 		 */
43787c478bd9Sstevel@tonic-gate 		if (!PP_ISFREE(pp) && vp != NULL && vp != &kvp &&
43797c478bd9Sstevel@tonic-gate 		    hat_ismod(pp) && !IS_SWAPVP(vp) && vp->v_vfsp != NULL &&
43807c478bd9Sstevel@tonic-gate 		    vfs_can_sync(vp->v_vfsp)) {
43817c478bd9Sstevel@tonic-gate 			nppbusy++;
43827c478bd9Sstevel@tonic-gate 			vfs_syncprogress();
43837c478bd9Sstevel@tonic-gate 
43847c478bd9Sstevel@tonic-gate 			if (!cleanit)
43857c478bd9Sstevel@tonic-gate 				continue;
43867c478bd9Sstevel@tonic-gate 			if (!page_trylock(pp, SE_EXCL))
43877c478bd9Sstevel@tonic-gate 				continue;
43887c478bd9Sstevel@tonic-gate 
43897c478bd9Sstevel@tonic-gate 			if (PP_ISFREE(pp) || vp == NULL || IS_SWAPVP(vp) ||
43907c478bd9Sstevel@tonic-gate 			    pp->p_lckcnt != 0 || pp->p_cowcnt != 0 ||
43917c478bd9Sstevel@tonic-gate 			    !(hat_pagesync(pp,
43927c478bd9Sstevel@tonic-gate 			    HAT_SYNC_DONTZERO | HAT_SYNC_STOPON_MOD) & P_MOD)) {
43937c478bd9Sstevel@tonic-gate 				page_unlock(pp);
43947c478bd9Sstevel@tonic-gate 				continue;
43957c478bd9Sstevel@tonic-gate 			}
43967c478bd9Sstevel@tonic-gate 			off = pp->p_offset;
43977c478bd9Sstevel@tonic-gate 			VN_HOLD(vp);
43987c478bd9Sstevel@tonic-gate 			page_unlock(pp);
43997c478bd9Sstevel@tonic-gate 			(void) VOP_PUTPAGE(vp, off, PAGESIZE,
44007c478bd9Sstevel@tonic-gate 			    B_ASYNC | B_FREE, kcred);
44017c478bd9Sstevel@tonic-gate 			VN_RELE(vp);
44027c478bd9Sstevel@tonic-gate 		}
44037c478bd9Sstevel@tonic-gate 	} while ((pp = page_next(pp)) != page0);
44047c478bd9Sstevel@tonic-gate 
44057c478bd9Sstevel@tonic-gate 	return (nppbusy);
44067c478bd9Sstevel@tonic-gate }
44077c478bd9Sstevel@tonic-gate 
44087c478bd9Sstevel@tonic-gate void page_invalidate_pages(void);
44097c478bd9Sstevel@tonic-gate 
44107c478bd9Sstevel@tonic-gate /*
44117c478bd9Sstevel@tonic-gate  * callback handler to vm sub-system
44127c478bd9Sstevel@tonic-gate  *
44137c478bd9Sstevel@tonic-gate  * callers make sure no recursive entries to this func.
44147c478bd9Sstevel@tonic-gate  */
44157c478bd9Sstevel@tonic-gate /*ARGSUSED*/
44167c478bd9Sstevel@tonic-gate boolean_t
44177c478bd9Sstevel@tonic-gate callb_vm_cpr(void *arg, int code)
44187c478bd9Sstevel@tonic-gate {
44197c478bd9Sstevel@tonic-gate 	if (code == CB_CODE_CPR_CHKPT)
44207c478bd9Sstevel@tonic-gate 		page_invalidate_pages();
44217c478bd9Sstevel@tonic-gate 	return (B_TRUE);
44227c478bd9Sstevel@tonic-gate }
44237c478bd9Sstevel@tonic-gate 
44247c478bd9Sstevel@tonic-gate /*
44257c478bd9Sstevel@tonic-gate  * Invalidate all pages of the system.
44267c478bd9Sstevel@tonic-gate  * It shouldn't be called until all user page activities are all stopped.
44277c478bd9Sstevel@tonic-gate  */
44287c478bd9Sstevel@tonic-gate void
44297c478bd9Sstevel@tonic-gate page_invalidate_pages()
44307c478bd9Sstevel@tonic-gate {
44317c478bd9Sstevel@tonic-gate 	page_t *pp;
44327c478bd9Sstevel@tonic-gate 	page_t *page0;
44337c478bd9Sstevel@tonic-gate 	pgcnt_t nbusypages;
44347c478bd9Sstevel@tonic-gate 	int retry = 0;
44357c478bd9Sstevel@tonic-gate 	const int MAXRETRIES = 4;
44367c478bd9Sstevel@tonic-gate #if defined(__sparc)
44377c478bd9Sstevel@tonic-gate 	extern struct vnode prom_ppages;
44387c478bd9Sstevel@tonic-gate #endif /* __sparc */
44397c478bd9Sstevel@tonic-gate 
44407c478bd9Sstevel@tonic-gate top:
44417c478bd9Sstevel@tonic-gate 	/*
44427c478bd9Sstevel@tonic-gate 	 * Flush dirty pages and destory the clean ones.
44437c478bd9Sstevel@tonic-gate 	 */
44447c478bd9Sstevel@tonic-gate 	nbusypages = 0;
44457c478bd9Sstevel@tonic-gate 
44467c478bd9Sstevel@tonic-gate 	pp = page0 = page_first();
44477c478bd9Sstevel@tonic-gate 	do {
44487c478bd9Sstevel@tonic-gate 		struct vnode	*vp;
44497c478bd9Sstevel@tonic-gate 		u_offset_t	offset;
44507c478bd9Sstevel@tonic-gate 		int		mod;
44517c478bd9Sstevel@tonic-gate 
44527c478bd9Sstevel@tonic-gate 		/*
44537c478bd9Sstevel@tonic-gate 		 * skip the page if it has no vnode or the page associated
44547c478bd9Sstevel@tonic-gate 		 * with the kernel vnode or prom allocated kernel mem.
44557c478bd9Sstevel@tonic-gate 		 */
44567c478bd9Sstevel@tonic-gate #if defined(__sparc)
44577c478bd9Sstevel@tonic-gate 		if ((vp = pp->p_vnode) == NULL || vp == &kvp ||
44587c478bd9Sstevel@tonic-gate 		    vp == &prom_ppages)
44597c478bd9Sstevel@tonic-gate #else /* x86 doesn't have prom or prom_ppage */
44607c478bd9Sstevel@tonic-gate 		if ((vp = pp->p_vnode) == NULL || vp == &kvp)
44617c478bd9Sstevel@tonic-gate #endif /* __sparc */
44627c478bd9Sstevel@tonic-gate 			continue;
44637c478bd9Sstevel@tonic-gate 
44647c478bd9Sstevel@tonic-gate 		/*
44657c478bd9Sstevel@tonic-gate 		 * skip the page which is already free invalidated.
44667c478bd9Sstevel@tonic-gate 		 */
44677c478bd9Sstevel@tonic-gate 		if (PP_ISFREE(pp) && PP_ISAGED(pp))
44687c478bd9Sstevel@tonic-gate 			continue;
44697c478bd9Sstevel@tonic-gate 
44707c478bd9Sstevel@tonic-gate 		/*
44717c478bd9Sstevel@tonic-gate 		 * skip pages that are already locked or can't be "exclusively"
44727c478bd9Sstevel@tonic-gate 		 * locked or are already free.  After we lock the page, check
44737c478bd9Sstevel@tonic-gate 		 * the free and age bits again to be sure it's not destroied
44747c478bd9Sstevel@tonic-gate 		 * yet.
44757c478bd9Sstevel@tonic-gate 		 * To achieve max. parallelization, we use page_trylock instead
44767c478bd9Sstevel@tonic-gate 		 * of page_lock so that we don't get block on individual pages
44777c478bd9Sstevel@tonic-gate 		 * while we have thousands of other pages to process.
44787c478bd9Sstevel@tonic-gate 		 */
44797c478bd9Sstevel@tonic-gate 		if (!page_trylock(pp, SE_EXCL)) {
44807c478bd9Sstevel@tonic-gate 			nbusypages++;
44817c478bd9Sstevel@tonic-gate 			continue;
44827c478bd9Sstevel@tonic-gate 		} else if (PP_ISFREE(pp)) {
44837c478bd9Sstevel@tonic-gate 			if (!PP_ISAGED(pp)) {
44847c478bd9Sstevel@tonic-gate 				page_destroy_free(pp);
44857c478bd9Sstevel@tonic-gate 			} else {
44867c478bd9Sstevel@tonic-gate 				page_unlock(pp);
44877c478bd9Sstevel@tonic-gate 			}
44887c478bd9Sstevel@tonic-gate 			continue;
44897c478bd9Sstevel@tonic-gate 		}
44907c478bd9Sstevel@tonic-gate 		/*
44917c478bd9Sstevel@tonic-gate 		 * Is this page involved in some I/O? shared?
44927c478bd9Sstevel@tonic-gate 		 *
44937c478bd9Sstevel@tonic-gate 		 * The page_struct_lock need not be acquired to
44947c478bd9Sstevel@tonic-gate 		 * examine these fields since the page has an
44957c478bd9Sstevel@tonic-gate 		 * "exclusive" lock.
44967c478bd9Sstevel@tonic-gate 		 */
44977c478bd9Sstevel@tonic-gate 		if (pp->p_lckcnt != 0 || pp->p_cowcnt != 0) {
44987c478bd9Sstevel@tonic-gate 			page_unlock(pp);
44997c478bd9Sstevel@tonic-gate 			continue;
45007c478bd9Sstevel@tonic-gate 		}
45017c478bd9Sstevel@tonic-gate 
45027c478bd9Sstevel@tonic-gate 		if (vp->v_type == VCHR) {
45037c478bd9Sstevel@tonic-gate 			panic("vp->v_type == VCHR");
45047c478bd9Sstevel@tonic-gate 			/*NOTREACHED*/
45057c478bd9Sstevel@tonic-gate 		}
45067c478bd9Sstevel@tonic-gate 
45077c478bd9Sstevel@tonic-gate 		if (!page_try_demote_pages(pp)) {
45087c478bd9Sstevel@tonic-gate 			page_unlock(pp);
45097c478bd9Sstevel@tonic-gate 			continue;
45107c478bd9Sstevel@tonic-gate 		}
45117c478bd9Sstevel@tonic-gate 
45127c478bd9Sstevel@tonic-gate 		/*
45137c478bd9Sstevel@tonic-gate 		 * Check the modified bit. Leave the bits alone in hardware
45147c478bd9Sstevel@tonic-gate 		 * (they will be modified if we do the putpage).
45157c478bd9Sstevel@tonic-gate 		 */
45167c478bd9Sstevel@tonic-gate 		mod = (hat_pagesync(pp, HAT_SYNC_DONTZERO | HAT_SYNC_STOPON_MOD)
45177c478bd9Sstevel@tonic-gate 			& P_MOD);
45187c478bd9Sstevel@tonic-gate 		if (mod) {
45197c478bd9Sstevel@tonic-gate 			offset = pp->p_offset;
45207c478bd9Sstevel@tonic-gate 			/*
45217c478bd9Sstevel@tonic-gate 			 * Hold the vnode before releasing the page lock
45227c478bd9Sstevel@tonic-gate 			 * to prevent it from being freed and re-used by
45237c478bd9Sstevel@tonic-gate 			 * some other thread.
45247c478bd9Sstevel@tonic-gate 			 */
45257c478bd9Sstevel@tonic-gate 			VN_HOLD(vp);
45267c478bd9Sstevel@tonic-gate 			page_unlock(pp);
45277c478bd9Sstevel@tonic-gate 			/*
45287c478bd9Sstevel@tonic-gate 			 * No error return is checked here. Callers such as
45297c478bd9Sstevel@tonic-gate 			 * cpr deals with the dirty pages at the dump time
45307c478bd9Sstevel@tonic-gate 			 * if this putpage fails.
45317c478bd9Sstevel@tonic-gate 			 */
45327c478bd9Sstevel@tonic-gate 			(void) VOP_PUTPAGE(vp, offset, PAGESIZE, B_INVAL,
45337c478bd9Sstevel@tonic-gate 			    kcred);
45347c478bd9Sstevel@tonic-gate 			VN_RELE(vp);
45357c478bd9Sstevel@tonic-gate 		} else {
45367c478bd9Sstevel@tonic-gate 			page_destroy(pp, 0);
45377c478bd9Sstevel@tonic-gate 		}
45387c478bd9Sstevel@tonic-gate 	} while ((pp = page_next(pp)) != page0);
45397c478bd9Sstevel@tonic-gate 	if (nbusypages && retry++ < MAXRETRIES) {
45407c478bd9Sstevel@tonic-gate 		delay(1);
45417c478bd9Sstevel@tonic-gate 		goto top;
45427c478bd9Sstevel@tonic-gate 	}
45437c478bd9Sstevel@tonic-gate }
45447c478bd9Sstevel@tonic-gate 
45457c478bd9Sstevel@tonic-gate /*
45467c478bd9Sstevel@tonic-gate  * Replace the page "old" with the page "new" on the page hash and vnode lists
45477c478bd9Sstevel@tonic-gate  *
45487c478bd9Sstevel@tonic-gate  * the replacemnt must be done in place, ie the equivalent sequence:
45497c478bd9Sstevel@tonic-gate  *
45507c478bd9Sstevel@tonic-gate  *	vp = old->p_vnode;
45517c478bd9Sstevel@tonic-gate  *	off = old->p_offset;
45527c478bd9Sstevel@tonic-gate  *	page_do_hashout(old)
45537c478bd9Sstevel@tonic-gate  *	page_do_hashin(new, vp, off)
45547c478bd9Sstevel@tonic-gate  *
45557c478bd9Sstevel@tonic-gate  * doesn't work, since
45567c478bd9Sstevel@tonic-gate  *  1) if old is the only page on the vnode, the v_pages list has a window
45577c478bd9Sstevel@tonic-gate  *     where it looks empty. This will break file system assumptions.
45587c478bd9Sstevel@tonic-gate  * and
45597c478bd9Sstevel@tonic-gate  *  2) pvn_vplist_dirty() can't deal with pages moving on the v_pages list.
45607c478bd9Sstevel@tonic-gate  */
45617c478bd9Sstevel@tonic-gate static void
45627c478bd9Sstevel@tonic-gate page_do_relocate_hash(page_t *new, page_t *old)
45637c478bd9Sstevel@tonic-gate {
45647c478bd9Sstevel@tonic-gate 	page_t	**hash_list;
45657c478bd9Sstevel@tonic-gate 	vnode_t	*vp = old->p_vnode;
45667c478bd9Sstevel@tonic-gate 	kmutex_t *sep;
45677c478bd9Sstevel@tonic-gate 
45687c478bd9Sstevel@tonic-gate 	ASSERT(PAGE_EXCL(old));
45697c478bd9Sstevel@tonic-gate 	ASSERT(PAGE_EXCL(new));
45707c478bd9Sstevel@tonic-gate 	ASSERT(vp != NULL);
45717c478bd9Sstevel@tonic-gate 	ASSERT(MUTEX_HELD(page_vnode_mutex(vp)));
45727c478bd9Sstevel@tonic-gate 	ASSERT(MUTEX_HELD(PAGE_HASH_MUTEX(PAGE_HASH_FUNC(vp, old->p_offset))));
45737c478bd9Sstevel@tonic-gate 
45747c478bd9Sstevel@tonic-gate 	/*
45757c478bd9Sstevel@tonic-gate 	 * First find old page on the page hash list
45767c478bd9Sstevel@tonic-gate 	 */
45777c478bd9Sstevel@tonic-gate 	hash_list = &page_hash[PAGE_HASH_FUNC(vp, old->p_offset)];
45787c478bd9Sstevel@tonic-gate 
45797c478bd9Sstevel@tonic-gate 	for (;;) {
45807c478bd9Sstevel@tonic-gate 		if (*hash_list == old)
45817c478bd9Sstevel@tonic-gate 			break;
45827c478bd9Sstevel@tonic-gate 		if (*hash_list == NULL) {
45837c478bd9Sstevel@tonic-gate 			panic("page_do_hashout");
45847c478bd9Sstevel@tonic-gate 			/*NOTREACHED*/
45857c478bd9Sstevel@tonic-gate 		}
45867c478bd9Sstevel@tonic-gate 		hash_list = &(*hash_list)->p_hash;
45877c478bd9Sstevel@tonic-gate 	}
45887c478bd9Sstevel@tonic-gate 
45897c478bd9Sstevel@tonic-gate 	/*
45907c478bd9Sstevel@tonic-gate 	 * update new and replace old with new on the page hash list
45917c478bd9Sstevel@tonic-gate 	 */
45927c478bd9Sstevel@tonic-gate 	new->p_vnode = old->p_vnode;
45937c478bd9Sstevel@tonic-gate 	new->p_offset = old->p_offset;
45947c478bd9Sstevel@tonic-gate 	new->p_hash = old->p_hash;
45957c478bd9Sstevel@tonic-gate 	*hash_list = new;
45967c478bd9Sstevel@tonic-gate 
45977c478bd9Sstevel@tonic-gate 	if ((new->p_vnode->v_flag & VISSWAP) != 0)
45987c478bd9Sstevel@tonic-gate 		PP_SETSWAP(new);
45997c478bd9Sstevel@tonic-gate 
46007c478bd9Sstevel@tonic-gate 	/*
46017c478bd9Sstevel@tonic-gate 	 * replace old with new on the vnode's page list
46027c478bd9Sstevel@tonic-gate 	 */
46037c478bd9Sstevel@tonic-gate 	if (old->p_vpnext == old) {
46047c478bd9Sstevel@tonic-gate 		new->p_vpnext = new;
46057c478bd9Sstevel@tonic-gate 		new->p_vpprev = new;
46067c478bd9Sstevel@tonic-gate 	} else {
46077c478bd9Sstevel@tonic-gate 		new->p_vpnext = old->p_vpnext;
46087c478bd9Sstevel@tonic-gate 		new->p_vpprev = old->p_vpprev;
46097c478bd9Sstevel@tonic-gate 		new->p_vpnext->p_vpprev = new;
46107c478bd9Sstevel@tonic-gate 		new->p_vpprev->p_vpnext = new;
46117c478bd9Sstevel@tonic-gate 	}
46127c478bd9Sstevel@tonic-gate 	if (vp->v_pages == old)
46137c478bd9Sstevel@tonic-gate 		vp->v_pages = new;
46147c478bd9Sstevel@tonic-gate 
46157c478bd9Sstevel@tonic-gate 	/*
46167c478bd9Sstevel@tonic-gate 	 * clear out the old page
46177c478bd9Sstevel@tonic-gate 	 */
46187c478bd9Sstevel@tonic-gate 	old->p_hash = NULL;
46197c478bd9Sstevel@tonic-gate 	old->p_vpnext = NULL;
46207c478bd9Sstevel@tonic-gate 	old->p_vpprev = NULL;
46217c478bd9Sstevel@tonic-gate 	old->p_vnode = NULL;
46227c478bd9Sstevel@tonic-gate 	PP_CLRSWAP(old);
46237c478bd9Sstevel@tonic-gate 	old->p_offset = (u_offset_t)-1;
46247c478bd9Sstevel@tonic-gate 	page_clr_all_props(old);
46257c478bd9Sstevel@tonic-gate 
46267c478bd9Sstevel@tonic-gate 	/*
46277c478bd9Sstevel@tonic-gate 	 * Wake up processes waiting for this page.  The page's
46287c478bd9Sstevel@tonic-gate 	 * identity has been changed, and is probably not the
46297c478bd9Sstevel@tonic-gate 	 * desired page any longer.
46307c478bd9Sstevel@tonic-gate 	 */
46317c478bd9Sstevel@tonic-gate 	sep = page_se_mutex(old);
46327c478bd9Sstevel@tonic-gate 	mutex_enter(sep);
463342787a71Sstans 	old->p_selock &= ~SE_EWANTED;
46347c478bd9Sstevel@tonic-gate 	if (CV_HAS_WAITERS(&old->p_cv))
46357c478bd9Sstevel@tonic-gate 		cv_broadcast(&old->p_cv);
46367c478bd9Sstevel@tonic-gate 	mutex_exit(sep);
46377c478bd9Sstevel@tonic-gate }
46387c478bd9Sstevel@tonic-gate 
46397c478bd9Sstevel@tonic-gate /*
46407c478bd9Sstevel@tonic-gate  * This function moves the identity of page "pp_old" to page "pp_new".
46417c478bd9Sstevel@tonic-gate  * Both pages must be locked on entry.  "pp_new" is free, has no identity,
46427c478bd9Sstevel@tonic-gate  * and need not be hashed out from anywhere.
46437c478bd9Sstevel@tonic-gate  */
46447c478bd9Sstevel@tonic-gate void
46457c478bd9Sstevel@tonic-gate page_relocate_hash(page_t *pp_new, page_t *pp_old)
46467c478bd9Sstevel@tonic-gate {
46477c478bd9Sstevel@tonic-gate 	vnode_t *vp = pp_old->p_vnode;
46487c478bd9Sstevel@tonic-gate 	u_offset_t off = pp_old->p_offset;
46497c478bd9Sstevel@tonic-gate 	kmutex_t *phm, *vphm;
46507c478bd9Sstevel@tonic-gate 
46517c478bd9Sstevel@tonic-gate 	/*
46527c478bd9Sstevel@tonic-gate 	 * Rehash two pages
46537c478bd9Sstevel@tonic-gate 	 */
46547c478bd9Sstevel@tonic-gate 	ASSERT(PAGE_EXCL(pp_old));
46557c478bd9Sstevel@tonic-gate 	ASSERT(PAGE_EXCL(pp_new));
46567c478bd9Sstevel@tonic-gate 	ASSERT(vp != NULL);
46577c478bd9Sstevel@tonic-gate 	ASSERT(pp_new->p_vnode == NULL);
46587c478bd9Sstevel@tonic-gate 
46597c478bd9Sstevel@tonic-gate 	/*
46607c478bd9Sstevel@tonic-gate 	 * hashout then hashin while holding the mutexes
46617c478bd9Sstevel@tonic-gate 	 */
46627c478bd9Sstevel@tonic-gate 	phm = PAGE_HASH_MUTEX(PAGE_HASH_FUNC(vp, off));
46637c478bd9Sstevel@tonic-gate 	mutex_enter(phm);
46647c478bd9Sstevel@tonic-gate 	vphm = page_vnode_mutex(vp);
46657c478bd9Sstevel@tonic-gate 	mutex_enter(vphm);
46667c478bd9Sstevel@tonic-gate 
46677c478bd9Sstevel@tonic-gate 	page_do_relocate_hash(pp_new, pp_old);
46687c478bd9Sstevel@tonic-gate 
46697c478bd9Sstevel@tonic-gate 	mutex_exit(vphm);
46707c478bd9Sstevel@tonic-gate 	mutex_exit(phm);
46717c478bd9Sstevel@tonic-gate 
46727c478bd9Sstevel@tonic-gate 	/*
46737c478bd9Sstevel@tonic-gate 	 * The page_struct_lock need not be acquired for lckcnt and
46747c478bd9Sstevel@tonic-gate 	 * cowcnt since the page has an "exclusive" lock.
46757c478bd9Sstevel@tonic-gate 	 */
46767c478bd9Sstevel@tonic-gate 	ASSERT(pp_new->p_lckcnt == 0);
46777c478bd9Sstevel@tonic-gate 	ASSERT(pp_new->p_cowcnt == 0);
46787c478bd9Sstevel@tonic-gate 	pp_new->p_lckcnt = pp_old->p_lckcnt;
46797c478bd9Sstevel@tonic-gate 	pp_new->p_cowcnt = pp_old->p_cowcnt;
46807c478bd9Sstevel@tonic-gate 	pp_old->p_lckcnt = pp_old->p_cowcnt = 0;
46817c478bd9Sstevel@tonic-gate 
46827c478bd9Sstevel@tonic-gate 	/* The following comment preserved from page_flip(). */
46837c478bd9Sstevel@tonic-gate 	/* XXX - Do we need to protect fsdata? */
46847c478bd9Sstevel@tonic-gate 	pp_new->p_fsdata = pp_old->p_fsdata;
46857c478bd9Sstevel@tonic-gate }
46867c478bd9Sstevel@tonic-gate 
46877c478bd9Sstevel@tonic-gate /*
46887c478bd9Sstevel@tonic-gate  * Helper routine used to lock all remaining members of a
46897c478bd9Sstevel@tonic-gate  * large page. The caller is responsible for passing in a locked
46907c478bd9Sstevel@tonic-gate  * pp. If pp is a large page, then it succeeds in locking all the
46917c478bd9Sstevel@tonic-gate  * remaining constituent pages or it returns with only the
46927c478bd9Sstevel@tonic-gate  * original page locked.
46937c478bd9Sstevel@tonic-gate  *
46947c478bd9Sstevel@tonic-gate  * Returns 1 on success, 0 on failure.
46957c478bd9Sstevel@tonic-gate  *
46967c478bd9Sstevel@tonic-gate  * If success is returned this routine gurantees p_szc for all constituent
46977c478bd9Sstevel@tonic-gate  * pages of a large page pp belongs to can't change. To achieve this we
46987c478bd9Sstevel@tonic-gate  * recheck szc of pp after locking all constituent pages and retry if szc
46997c478bd9Sstevel@tonic-gate  * changed (it could only decrease). Since hat_page_demote() needs an EXCL
47007c478bd9Sstevel@tonic-gate  * lock on one of constituent pages it can't be running after all constituent
47017c478bd9Sstevel@tonic-gate  * pages are locked.  hat_page_demote() with a lock on a constituent page
47027c478bd9Sstevel@tonic-gate  * outside of this large page (i.e. pp belonged to a larger large page) is
47037c478bd9Sstevel@tonic-gate  * already done with all constituent pages of pp since the root's p_szc is
47047c478bd9Sstevel@tonic-gate  * changed last. Thefore no need to synchronize with hat_page_demote() that
47057c478bd9Sstevel@tonic-gate  * locked a constituent page outside of pp's current large page.
47067c478bd9Sstevel@tonic-gate  */
47077c478bd9Sstevel@tonic-gate #ifdef DEBUG
47087c478bd9Sstevel@tonic-gate uint32_t gpg_trylock_mtbf = 0;
47097c478bd9Sstevel@tonic-gate #endif
47107c478bd9Sstevel@tonic-gate 
47117c478bd9Sstevel@tonic-gate int
47127c478bd9Sstevel@tonic-gate group_page_trylock(page_t *pp, se_t se)
47137c478bd9Sstevel@tonic-gate {
47147c478bd9Sstevel@tonic-gate 	page_t  *tpp;
47157c478bd9Sstevel@tonic-gate 	pgcnt_t	npgs, i, j;
47167c478bd9Sstevel@tonic-gate 	uint_t pszc = pp->p_szc;
47177c478bd9Sstevel@tonic-gate 
47187c478bd9Sstevel@tonic-gate #ifdef DEBUG
47197c478bd9Sstevel@tonic-gate 	if (gpg_trylock_mtbf && !(gethrtime() % gpg_trylock_mtbf)) {
47207c478bd9Sstevel@tonic-gate 		return (0);
47217c478bd9Sstevel@tonic-gate 	}
47227c478bd9Sstevel@tonic-gate #endif
47237c478bd9Sstevel@tonic-gate 
47247c478bd9Sstevel@tonic-gate 	if (pp != PP_GROUPLEADER(pp, pszc)) {
47257c478bd9Sstevel@tonic-gate 		return (0);
47267c478bd9Sstevel@tonic-gate 	}
47277c478bd9Sstevel@tonic-gate 
47287c478bd9Sstevel@tonic-gate retry:
47297c478bd9Sstevel@tonic-gate 	ASSERT(PAGE_LOCKED_SE(pp, se));
47307c478bd9Sstevel@tonic-gate 	ASSERT(!PP_ISFREE(pp));
47317c478bd9Sstevel@tonic-gate 	if (pszc == 0) {
47327c478bd9Sstevel@tonic-gate 		return (1);
47337c478bd9Sstevel@tonic-gate 	}
47347c478bd9Sstevel@tonic-gate 	npgs = page_get_pagecnt(pszc);
47357c478bd9Sstevel@tonic-gate 	tpp = pp + 1;
47367c478bd9Sstevel@tonic-gate 	for (i = 1; i < npgs; i++, tpp++) {
47377c478bd9Sstevel@tonic-gate 		if (!page_trylock(tpp, se)) {
47387c478bd9Sstevel@tonic-gate 			tpp = pp + 1;
47397c478bd9Sstevel@tonic-gate 			for (j = 1; j < i; j++, tpp++) {
47407c478bd9Sstevel@tonic-gate 				page_unlock(tpp);
47417c478bd9Sstevel@tonic-gate 			}
47427c478bd9Sstevel@tonic-gate 			return (0);
47437c478bd9Sstevel@tonic-gate 		}
47447c478bd9Sstevel@tonic-gate 	}
47457c478bd9Sstevel@tonic-gate 	if (pp->p_szc != pszc) {
47467c478bd9Sstevel@tonic-gate 		ASSERT(pp->p_szc < pszc);
47477c478bd9Sstevel@tonic-gate 		ASSERT(pp->p_vnode != NULL && pp->p_vnode != &kvp &&
47487c478bd9Sstevel@tonic-gate 		    !IS_SWAPFSVP(pp->p_vnode));
47497c478bd9Sstevel@tonic-gate 		tpp = pp + 1;
47507c478bd9Sstevel@tonic-gate 		for (i = 1; i < npgs; i++, tpp++) {
47517c478bd9Sstevel@tonic-gate 			page_unlock(tpp);
47527c478bd9Sstevel@tonic-gate 		}
47537c478bd9Sstevel@tonic-gate 		pszc = pp->p_szc;
47547c478bd9Sstevel@tonic-gate 		goto retry;
47557c478bd9Sstevel@tonic-gate 	}
47567c478bd9Sstevel@tonic-gate 	return (1);
47577c478bd9Sstevel@tonic-gate }
47587c478bd9Sstevel@tonic-gate 
47597c478bd9Sstevel@tonic-gate void
47607c478bd9Sstevel@tonic-gate group_page_unlock(page_t *pp)
47617c478bd9Sstevel@tonic-gate {
47627c478bd9Sstevel@tonic-gate 	page_t *tpp;
47637c478bd9Sstevel@tonic-gate 	pgcnt_t	npgs, i;
47647c478bd9Sstevel@tonic-gate 
47657c478bd9Sstevel@tonic-gate 	ASSERT(PAGE_LOCKED(pp));
47667c478bd9Sstevel@tonic-gate 	ASSERT(!PP_ISFREE(pp));
47677c478bd9Sstevel@tonic-gate 	ASSERT(pp == PP_PAGEROOT(pp));
47687c478bd9Sstevel@tonic-gate 	npgs = page_get_pagecnt(pp->p_szc);
47697c478bd9Sstevel@tonic-gate 	for (i = 1, tpp = pp + 1; i < npgs; i++, tpp++) {
47707c478bd9Sstevel@tonic-gate 		page_unlock(tpp);
47717c478bd9Sstevel@tonic-gate 	}
47727c478bd9Sstevel@tonic-gate }
47737c478bd9Sstevel@tonic-gate 
47747c478bd9Sstevel@tonic-gate /*
47757c478bd9Sstevel@tonic-gate  * returns
47767c478bd9Sstevel@tonic-gate  * 0 		: on success and *nrelocp is number of relocated PAGESIZE pages
47777c478bd9Sstevel@tonic-gate  * ERANGE	: this is not a base page
47787c478bd9Sstevel@tonic-gate  * EBUSY	: failure to get locks on the page/pages
47797c478bd9Sstevel@tonic-gate  * ENOMEM	: failure to obtain replacement pages
47807c478bd9Sstevel@tonic-gate  * EAGAIN	: OBP has not yet completed its boot-time handoff to the kernel
47817c478bd9Sstevel@tonic-gate  *
47827c478bd9Sstevel@tonic-gate  * Return with all constituent members of target and replacement
47837c478bd9Sstevel@tonic-gate  * SE_EXCL locked. It is the callers responsibility to drop the
47847c478bd9Sstevel@tonic-gate  * locks.
47857c478bd9Sstevel@tonic-gate  */
47867c478bd9Sstevel@tonic-gate int
47877c478bd9Sstevel@tonic-gate do_page_relocate(
47887c478bd9Sstevel@tonic-gate 	page_t **target,
47897c478bd9Sstevel@tonic-gate 	page_t **replacement,
47907c478bd9Sstevel@tonic-gate 	int grouplock,
47917c478bd9Sstevel@tonic-gate 	spgcnt_t *nrelocp,
47927c478bd9Sstevel@tonic-gate 	lgrp_t *lgrp)
47937c478bd9Sstevel@tonic-gate {
47947c478bd9Sstevel@tonic-gate #ifdef DEBUG
47957c478bd9Sstevel@tonic-gate 	page_t *first_repl;
47967c478bd9Sstevel@tonic-gate #endif /* DEBUG */
47977c478bd9Sstevel@tonic-gate 	page_t *repl;
47987c478bd9Sstevel@tonic-gate 	page_t *targ;
47997c478bd9Sstevel@tonic-gate 	page_t *pl = NULL;
48007c478bd9Sstevel@tonic-gate 	uint_t ppattr;
48017c478bd9Sstevel@tonic-gate 	pfn_t   pfn, repl_pfn;
48027c478bd9Sstevel@tonic-gate 	uint_t	szc;
48037c478bd9Sstevel@tonic-gate 	spgcnt_t npgs, i;
48047c478bd9Sstevel@tonic-gate 	int repl_contig = 0;
48057c478bd9Sstevel@tonic-gate 	uint_t flags = 0;
48067c478bd9Sstevel@tonic-gate 	spgcnt_t dofree = 0;
48077c478bd9Sstevel@tonic-gate 
48087c478bd9Sstevel@tonic-gate 	*nrelocp = 0;
48097c478bd9Sstevel@tonic-gate 
48107c478bd9Sstevel@tonic-gate #if defined(__sparc)
48117c478bd9Sstevel@tonic-gate 	/*
48127c478bd9Sstevel@tonic-gate 	 * We need to wait till OBP has completed
48137c478bd9Sstevel@tonic-gate 	 * its boot-time handoff of its resources to the kernel
48147c478bd9Sstevel@tonic-gate 	 * before we allow page relocation
48157c478bd9Sstevel@tonic-gate 	 */
48167c478bd9Sstevel@tonic-gate 	if (page_relocate_ready == 0) {
48177c478bd9Sstevel@tonic-gate 		return (EAGAIN);
48187c478bd9Sstevel@tonic-gate 	}
48197c478bd9Sstevel@tonic-gate #endif
48207c478bd9Sstevel@tonic-gate 
48217c478bd9Sstevel@tonic-gate 	/*
48227c478bd9Sstevel@tonic-gate 	 * If this is not a base page,
48237c478bd9Sstevel@tonic-gate 	 * just return with 0x0 pages relocated.
48247c478bd9Sstevel@tonic-gate 	 */
48257c478bd9Sstevel@tonic-gate 	targ = *target;
48267c478bd9Sstevel@tonic-gate 	ASSERT(PAGE_EXCL(targ));
48277c478bd9Sstevel@tonic-gate 	ASSERT(!PP_ISFREE(targ));
48287c478bd9Sstevel@tonic-gate 	szc = targ->p_szc;
48297c478bd9Sstevel@tonic-gate 	ASSERT(szc < mmu_page_sizes);
48307c478bd9Sstevel@tonic-gate 	VM_STAT_ADD(vmm_vmstats.ppr_reloc[szc]);
48317c478bd9Sstevel@tonic-gate 	pfn = targ->p_pagenum;
48327c478bd9Sstevel@tonic-gate 	if (pfn != PFN_BASE(pfn, szc)) {
48337c478bd9Sstevel@tonic-gate 		VM_STAT_ADD(vmm_vmstats.ppr_relocnoroot[szc]);
48347c478bd9Sstevel@tonic-gate 		return (ERANGE);
48357c478bd9Sstevel@tonic-gate 	}
48367c478bd9Sstevel@tonic-gate 
48377c478bd9Sstevel@tonic-gate 	if ((repl = *replacement) != NULL && repl->p_szc >= szc) {
48387c478bd9Sstevel@tonic-gate 		repl_pfn = repl->p_pagenum;
48397c478bd9Sstevel@tonic-gate 		if (repl_pfn != PFN_BASE(repl_pfn, szc)) {
48407c478bd9Sstevel@tonic-gate 			VM_STAT_ADD(vmm_vmstats.ppr_reloc_replnoroot[szc]);
48417c478bd9Sstevel@tonic-gate 			return (ERANGE);
48427c478bd9Sstevel@tonic-gate 		}
48437c478bd9Sstevel@tonic-gate 		repl_contig = 1;
48447c478bd9Sstevel@tonic-gate 	}
48457c478bd9Sstevel@tonic-gate 
48467c478bd9Sstevel@tonic-gate 	/*
48477c478bd9Sstevel@tonic-gate 	 * We must lock all members of this large page or we cannot
48487c478bd9Sstevel@tonic-gate 	 * relocate any part of it.
48497c478bd9Sstevel@tonic-gate 	 */
48507c478bd9Sstevel@tonic-gate 	if (grouplock != 0 && !group_page_trylock(targ, SE_EXCL)) {
48517c478bd9Sstevel@tonic-gate 		VM_STAT_ADD(vmm_vmstats.ppr_relocnolock[targ->p_szc]);
48527c478bd9Sstevel@tonic-gate 		return (EBUSY);
48537c478bd9Sstevel@tonic-gate 	}
48547c478bd9Sstevel@tonic-gate 
48557c478bd9Sstevel@tonic-gate 	/*
48567c478bd9Sstevel@tonic-gate 	 * reread szc it could have been decreased before
48577c478bd9Sstevel@tonic-gate 	 * group_page_trylock() was done.
48587c478bd9Sstevel@tonic-gate 	 */
48597c478bd9Sstevel@tonic-gate 	szc = targ->p_szc;
48607c478bd9Sstevel@tonic-gate 	ASSERT(szc < mmu_page_sizes);
48617c478bd9Sstevel@tonic-gate 	VM_STAT_ADD(vmm_vmstats.ppr_reloc[szc]);
48627c478bd9Sstevel@tonic-gate 	ASSERT(pfn == PFN_BASE(pfn, szc));
48637c478bd9Sstevel@tonic-gate 
48647c478bd9Sstevel@tonic-gate 	npgs = page_get_pagecnt(targ->p_szc);
48657c478bd9Sstevel@tonic-gate 
48667c478bd9Sstevel@tonic-gate 	if (repl == NULL) {
48677c478bd9Sstevel@tonic-gate 		dofree = npgs;		/* Size of target page in MMU pages */
48687c478bd9Sstevel@tonic-gate 		if (!page_create_wait(dofree, 0)) {
48697c478bd9Sstevel@tonic-gate 			if (grouplock != 0) {
48707c478bd9Sstevel@tonic-gate 				group_page_unlock(targ);
48717c478bd9Sstevel@tonic-gate 			}
48727c478bd9Sstevel@tonic-gate 			VM_STAT_ADD(vmm_vmstats.ppr_relocnomem[szc]);
48737c478bd9Sstevel@tonic-gate 			return (ENOMEM);
48747c478bd9Sstevel@tonic-gate 		}
48757c478bd9Sstevel@tonic-gate 
48767c478bd9Sstevel@tonic-gate 		/*
48777c478bd9Sstevel@tonic-gate 		 * seg kmem pages require that the target and replacement
48787c478bd9Sstevel@tonic-gate 		 * page be the same pagesize.
48797c478bd9Sstevel@tonic-gate 		 */
48807c478bd9Sstevel@tonic-gate 		flags = (targ->p_vnode == &kvp) ? PGR_SAMESZC : 0;
48817c478bd9Sstevel@tonic-gate 		repl = page_get_replacement_page(targ, lgrp, flags);
48827c478bd9Sstevel@tonic-gate 		if (repl == NULL) {
48837c478bd9Sstevel@tonic-gate 			if (grouplock != 0) {
48847c478bd9Sstevel@tonic-gate 				group_page_unlock(targ);
48857c478bd9Sstevel@tonic-gate 			}
48867c478bd9Sstevel@tonic-gate 			page_create_putback(dofree);
48877c478bd9Sstevel@tonic-gate 			VM_STAT_ADD(vmm_vmstats.ppr_relocnomem[szc]);
48887c478bd9Sstevel@tonic-gate 			return (ENOMEM);
48897c478bd9Sstevel@tonic-gate 		}
48907c478bd9Sstevel@tonic-gate 	}
48917c478bd9Sstevel@tonic-gate #ifdef DEBUG
48927c478bd9Sstevel@tonic-gate 	else {
48937c478bd9Sstevel@tonic-gate 		ASSERT(PAGE_LOCKED(repl));
48947c478bd9Sstevel@tonic-gate 	}
48957c478bd9Sstevel@tonic-gate #endif /* DEBUG */
48967c478bd9Sstevel@tonic-gate 
48977c478bd9Sstevel@tonic-gate #if defined(__sparc)
48987c478bd9Sstevel@tonic-gate 	/*
48997c478bd9Sstevel@tonic-gate 	 * Let hat_page_relocate() complete the relocation if it's kernel page
49007c478bd9Sstevel@tonic-gate 	 */
49017c478bd9Sstevel@tonic-gate 	if (targ->p_vnode == &kvp) {
49027c478bd9Sstevel@tonic-gate 		*replacement = repl;
49037c478bd9Sstevel@tonic-gate 		if (hat_page_relocate(target, replacement, nrelocp) != 0) {
49047c478bd9Sstevel@tonic-gate 			if (grouplock != 0) {
49057c478bd9Sstevel@tonic-gate 				group_page_unlock(targ);
49067c478bd9Sstevel@tonic-gate 			}
49077c478bd9Sstevel@tonic-gate 			if (dofree) {
49087c478bd9Sstevel@tonic-gate 				*replacement = NULL;
49097c478bd9Sstevel@tonic-gate 				page_free_replacement_page(repl);
49107c478bd9Sstevel@tonic-gate 				page_create_putback(dofree);
49117c478bd9Sstevel@tonic-gate 			}
49127c478bd9Sstevel@tonic-gate 			VM_STAT_ADD(vmm_vmstats.ppr_krelocfail[szc]);
49137c478bd9Sstevel@tonic-gate 			return (EAGAIN);
49147c478bd9Sstevel@tonic-gate 		}
49157c478bd9Sstevel@tonic-gate 		VM_STAT_ADD(vmm_vmstats.ppr_relocok[szc]);
49167c478bd9Sstevel@tonic-gate 		return (0);
49177c478bd9Sstevel@tonic-gate 	}
49187c478bd9Sstevel@tonic-gate #else
49197c478bd9Sstevel@tonic-gate #if defined(lint)
49207c478bd9Sstevel@tonic-gate 	dofree = dofree;
49217c478bd9Sstevel@tonic-gate #endif
49227c478bd9Sstevel@tonic-gate #endif
49237c478bd9Sstevel@tonic-gate 
49247c478bd9Sstevel@tonic-gate #ifdef DEBUG
49257c478bd9Sstevel@tonic-gate 	first_repl = repl;
49267c478bd9Sstevel@tonic-gate #endif /* DEBUG */
49277c478bd9Sstevel@tonic-gate 
49287c478bd9Sstevel@tonic-gate 	for (i = 0; i < npgs; i++) {
49297c478bd9Sstevel@tonic-gate 		ASSERT(PAGE_EXCL(targ));
493007b65a64Saguzovsk 		ASSERT(targ->p_slckcnt == 0);
493107b65a64Saguzovsk 		ASSERT(repl->p_slckcnt == 0);
49327c478bd9Sstevel@tonic-gate 
49337c478bd9Sstevel@tonic-gate 		(void) hat_pageunload(targ, HAT_FORCE_PGUNLOAD);
49347c478bd9Sstevel@tonic-gate 
49357c478bd9Sstevel@tonic-gate 		ASSERT(hat_page_getshare(targ) == 0);
49367c478bd9Sstevel@tonic-gate 		ASSERT(!PP_ISFREE(targ));
49377c478bd9Sstevel@tonic-gate 		ASSERT(targ->p_pagenum == (pfn + i));
49387c478bd9Sstevel@tonic-gate 		ASSERT(repl_contig == 0 ||
49397c478bd9Sstevel@tonic-gate 		    repl->p_pagenum == (repl_pfn + i));
49407c478bd9Sstevel@tonic-gate 
49417c478bd9Sstevel@tonic-gate 		/*
49427c478bd9Sstevel@tonic-gate 		 * Copy the page contents and attributes then
49437c478bd9Sstevel@tonic-gate 		 * relocate the page in the page hash.
49447c478bd9Sstevel@tonic-gate 		 */
49457c478bd9Sstevel@tonic-gate 		ppcopy(targ, repl);
49467c478bd9Sstevel@tonic-gate 		ppattr = hat_page_getattr(targ, (P_MOD | P_REF | P_RO));
49477c478bd9Sstevel@tonic-gate 		page_clr_all_props(repl);
49487c478bd9Sstevel@tonic-gate 		page_set_props(repl, ppattr);
49497c478bd9Sstevel@tonic-gate 		page_relocate_hash(repl, targ);
49507c478bd9Sstevel@tonic-gate 
49517c478bd9Sstevel@tonic-gate 		ASSERT(hat_page_getshare(targ) == 0);
49527c478bd9Sstevel@tonic-gate 		ASSERT(hat_page_getshare(repl) == 0);
49537c478bd9Sstevel@tonic-gate 		/*
49547c478bd9Sstevel@tonic-gate 		 * Now clear the props on targ, after the
49557c478bd9Sstevel@tonic-gate 		 * page_relocate_hash(), they no longer
49567c478bd9Sstevel@tonic-gate 		 * have any meaning.
49577c478bd9Sstevel@tonic-gate 		 */
49587c478bd9Sstevel@tonic-gate 		page_clr_all_props(targ);
49597c478bd9Sstevel@tonic-gate 		ASSERT(targ->p_next == targ);
49607c478bd9Sstevel@tonic-gate 		ASSERT(targ->p_prev == targ);
49617c478bd9Sstevel@tonic-gate 		page_list_concat(&pl, &targ);
49627c478bd9Sstevel@tonic-gate 
49637c478bd9Sstevel@tonic-gate 		targ++;
49647c478bd9Sstevel@tonic-gate 		if (repl_contig != 0) {
49657c478bd9Sstevel@tonic-gate 			repl++;
49667c478bd9Sstevel@tonic-gate 		} else {
49677c478bd9Sstevel@tonic-gate 			repl = repl->p_next;
49687c478bd9Sstevel@tonic-gate 		}
49697c478bd9Sstevel@tonic-gate 	}
49707c478bd9Sstevel@tonic-gate 	/* assert that we have come full circle with repl */
49717c478bd9Sstevel@tonic-gate 	ASSERT(repl_contig == 1 || first_repl == repl);
49727c478bd9Sstevel@tonic-gate 
49737c478bd9Sstevel@tonic-gate 	*target = pl;
49747c478bd9Sstevel@tonic-gate 	if (*replacement == NULL) {
49757c478bd9Sstevel@tonic-gate 		ASSERT(first_repl == repl);
49767c478bd9Sstevel@tonic-gate 		*replacement = repl;
49777c478bd9Sstevel@tonic-gate 	}
49787c478bd9Sstevel@tonic-gate 	VM_STAT_ADD(vmm_vmstats.ppr_relocok[szc]);
49797c478bd9Sstevel@tonic-gate 	*nrelocp = npgs;
49807c478bd9Sstevel@tonic-gate 	return (0);
49817c478bd9Sstevel@tonic-gate }
49827c478bd9Sstevel@tonic-gate /*
49837c478bd9Sstevel@tonic-gate  * On success returns 0 and *nrelocp the number of PAGESIZE pages relocated.
49847c478bd9Sstevel@tonic-gate  */
49857c478bd9Sstevel@tonic-gate int
49867c478bd9Sstevel@tonic-gate page_relocate(
49877c478bd9Sstevel@tonic-gate 	page_t **target,
49887c478bd9Sstevel@tonic-gate 	page_t **replacement,
49897c478bd9Sstevel@tonic-gate 	int grouplock,
49907c478bd9Sstevel@tonic-gate 	int freetarget,
49917c478bd9Sstevel@tonic-gate 	spgcnt_t *nrelocp,
49927c478bd9Sstevel@tonic-gate 	lgrp_t *lgrp)
49937c478bd9Sstevel@tonic-gate {
49947c478bd9Sstevel@tonic-gate 	spgcnt_t ret;
49957c478bd9Sstevel@tonic-gate 
49967c478bd9Sstevel@tonic-gate 	/* do_page_relocate returns 0 on success or errno value */
49977c478bd9Sstevel@tonic-gate 	ret = do_page_relocate(target, replacement, grouplock, nrelocp, lgrp);
49987c478bd9Sstevel@tonic-gate 
49997c478bd9Sstevel@tonic-gate 	if (ret != 0 || freetarget == 0) {
50007c478bd9Sstevel@tonic-gate 		return (ret);
50017c478bd9Sstevel@tonic-gate 	}
50027c478bd9Sstevel@tonic-gate 	if (*nrelocp == 1) {
50037c478bd9Sstevel@tonic-gate 		ASSERT(*target != NULL);
50047c478bd9Sstevel@tonic-gate 		page_free(*target, 1);
50057c478bd9Sstevel@tonic-gate 	} else {
50067c478bd9Sstevel@tonic-gate 		page_t *tpp = *target;
50077c478bd9Sstevel@tonic-gate 		uint_t szc = tpp->p_szc;
50087c478bd9Sstevel@tonic-gate 		pgcnt_t npgs = page_get_pagecnt(szc);
50097c478bd9Sstevel@tonic-gate 		ASSERT(npgs > 1);
50107c478bd9Sstevel@tonic-gate 		ASSERT(szc != 0);
50117c478bd9Sstevel@tonic-gate 		do {
50127c478bd9Sstevel@tonic-gate 			ASSERT(PAGE_EXCL(tpp));
50137c478bd9Sstevel@tonic-gate 			ASSERT(!hat_page_is_mapped(tpp));
50147c478bd9Sstevel@tonic-gate 			ASSERT(tpp->p_szc == szc);
50157c478bd9Sstevel@tonic-gate 			PP_SETFREE(tpp);
50167c478bd9Sstevel@tonic-gate 			PP_SETAGED(tpp);
50177c478bd9Sstevel@tonic-gate 			npgs--;
50187c478bd9Sstevel@tonic-gate 		} while ((tpp = tpp->p_next) != *target);
50197c478bd9Sstevel@tonic-gate 		ASSERT(npgs == 0);
50207c478bd9Sstevel@tonic-gate 		page_list_add_pages(*target, 0);
50217c478bd9Sstevel@tonic-gate 		npgs = page_get_pagecnt(szc);
50227c478bd9Sstevel@tonic-gate 		page_create_putback(npgs);
50237c478bd9Sstevel@tonic-gate 	}
50247c478bd9Sstevel@tonic-gate 	return (ret);
50257c478bd9Sstevel@tonic-gate }
50267c478bd9Sstevel@tonic-gate 
50277c478bd9Sstevel@tonic-gate /*
50287c478bd9Sstevel@tonic-gate  * it is up to the caller to deal with pcf accounting.
50297c478bd9Sstevel@tonic-gate  */
50307c478bd9Sstevel@tonic-gate void
50317c478bd9Sstevel@tonic-gate page_free_replacement_page(page_t *pplist)
50327c478bd9Sstevel@tonic-gate {
50337c478bd9Sstevel@tonic-gate 	page_t *pp;
50347c478bd9Sstevel@tonic-gate 
50357c478bd9Sstevel@tonic-gate 	while (pplist != NULL) {
50367c478bd9Sstevel@tonic-gate 		/*
50377c478bd9Sstevel@tonic-gate 		 * pp_targ is a linked list.
50387c478bd9Sstevel@tonic-gate 		 */
50397c478bd9Sstevel@tonic-gate 		pp = pplist;
50407c478bd9Sstevel@tonic-gate 		if (pp->p_szc == 0) {
50417c478bd9Sstevel@tonic-gate 			page_sub(&pplist, pp);
50427c478bd9Sstevel@tonic-gate 			page_clr_all_props(pp);
50437c478bd9Sstevel@tonic-gate 			PP_SETFREE(pp);
50447c478bd9Sstevel@tonic-gate 			PP_SETAGED(pp);
50457c478bd9Sstevel@tonic-gate 			page_list_add(pp, PG_FREE_LIST | PG_LIST_TAIL);
50467c478bd9Sstevel@tonic-gate 			page_unlock(pp);
50477c478bd9Sstevel@tonic-gate 			VM_STAT_ADD(pagecnt.pc_free_replacement_page[0]);
50487c478bd9Sstevel@tonic-gate 		} else {
50497c478bd9Sstevel@tonic-gate 			spgcnt_t curnpgs = page_get_pagecnt(pp->p_szc);
50507c478bd9Sstevel@tonic-gate 			page_t *tpp;
50517c478bd9Sstevel@tonic-gate 			page_list_break(&pp, &pplist, curnpgs);
50527c478bd9Sstevel@tonic-gate 			tpp = pp;
50537c478bd9Sstevel@tonic-gate 			do {
50547c478bd9Sstevel@tonic-gate 				ASSERT(PAGE_EXCL(tpp));
50557c478bd9Sstevel@tonic-gate 				ASSERT(!hat_page_is_mapped(tpp));
50567c478bd9Sstevel@tonic-gate 				page_clr_all_props(pp);
50577c478bd9Sstevel@tonic-gate 				PP_SETFREE(tpp);
50587c478bd9Sstevel@tonic-gate 				PP_SETAGED(tpp);
50597c478bd9Sstevel@tonic-gate 			} while ((tpp = tpp->p_next) != pp);
50607c478bd9Sstevel@tonic-gate 			page_list_add_pages(pp, 0);
50617c478bd9Sstevel@tonic-gate 			VM_STAT_ADD(pagecnt.pc_free_replacement_page[1]);
50627c478bd9Sstevel@tonic-gate 		}
50637c478bd9Sstevel@tonic-gate 	}
50647c478bd9Sstevel@tonic-gate }
50657c478bd9Sstevel@tonic-gate 
50667c478bd9Sstevel@tonic-gate /*
50677c478bd9Sstevel@tonic-gate  * Relocate target to non-relocatable replacement page.
50687c478bd9Sstevel@tonic-gate  */
50697c478bd9Sstevel@tonic-gate int
50707c478bd9Sstevel@tonic-gate page_relocate_cage(page_t **target, page_t **replacement)
50717c478bd9Sstevel@tonic-gate {
50727c478bd9Sstevel@tonic-gate 	page_t *tpp, *rpp;
50737c478bd9Sstevel@tonic-gate 	spgcnt_t pgcnt, npgs;
50747c478bd9Sstevel@tonic-gate 	int result;
50757c478bd9Sstevel@tonic-gate 
50767c478bd9Sstevel@tonic-gate 	tpp = *target;
50777c478bd9Sstevel@tonic-gate 
50787c478bd9Sstevel@tonic-gate 	ASSERT(PAGE_EXCL(tpp));
50797c478bd9Sstevel@tonic-gate 	ASSERT(tpp->p_szc == 0);
50807c478bd9Sstevel@tonic-gate 
50817c478bd9Sstevel@tonic-gate 	pgcnt = btop(page_get_pagesize(tpp->p_szc));
50827c478bd9Sstevel@tonic-gate 
50837c478bd9Sstevel@tonic-gate 	do {
50847c478bd9Sstevel@tonic-gate 		(void) page_create_wait(pgcnt, PG_WAIT | PG_NORELOC);
50857c478bd9Sstevel@tonic-gate 		rpp = page_get_replacement_page(tpp, NULL, PGR_NORELOC);
50867c478bd9Sstevel@tonic-gate 		if (rpp == NULL) {
50877c478bd9Sstevel@tonic-gate 			page_create_putback(pgcnt);
50887c478bd9Sstevel@tonic-gate 			kcage_cageout_wakeup();
50897c478bd9Sstevel@tonic-gate 		}
50907c478bd9Sstevel@tonic-gate 	} while (rpp == NULL);
50917c478bd9Sstevel@tonic-gate 
50927c478bd9Sstevel@tonic-gate 	ASSERT(PP_ISNORELOC(rpp));
50937c478bd9Sstevel@tonic-gate 
50947c478bd9Sstevel@tonic-gate 	result = page_relocate(&tpp, &rpp, 0, 1, &npgs, NULL);
50957c478bd9Sstevel@tonic-gate 
50967c478bd9Sstevel@tonic-gate 	if (result == 0) {
50977c478bd9Sstevel@tonic-gate 		*replacement = rpp;
50987c478bd9Sstevel@tonic-gate 		if (pgcnt != npgs)
50997c478bd9Sstevel@tonic-gate 			panic("page_relocate_cage: partial relocation");
51007c478bd9Sstevel@tonic-gate 	}
51017c478bd9Sstevel@tonic-gate 
51027c478bd9Sstevel@tonic-gate 	return (result);
51037c478bd9Sstevel@tonic-gate }
51047c478bd9Sstevel@tonic-gate 
51057c478bd9Sstevel@tonic-gate /*
51067c478bd9Sstevel@tonic-gate  * Release the page lock on a page, place on cachelist
51077c478bd9Sstevel@tonic-gate  * tail if no longer mapped. Caller can let us know if
51087c478bd9Sstevel@tonic-gate  * the page is known to be clean.
51097c478bd9Sstevel@tonic-gate  */
51107c478bd9Sstevel@tonic-gate int
51117c478bd9Sstevel@tonic-gate page_release(page_t *pp, int checkmod)
51127c478bd9Sstevel@tonic-gate {
51137c478bd9Sstevel@tonic-gate 	int status;
51147c478bd9Sstevel@tonic-gate 
51157c478bd9Sstevel@tonic-gate 	ASSERT(PAGE_LOCKED(pp) && !PP_ISFREE(pp) &&
51167c478bd9Sstevel@tonic-gate 		(pp->p_vnode != NULL));
51177c478bd9Sstevel@tonic-gate 
51187c478bd9Sstevel@tonic-gate 	if (!hat_page_is_mapped(pp) && !IS_SWAPVP(pp->p_vnode) &&
51197c478bd9Sstevel@tonic-gate 	    ((PAGE_SHARED(pp) && page_tryupgrade(pp)) || PAGE_EXCL(pp)) &&
51207c478bd9Sstevel@tonic-gate 	    pp->p_lckcnt == 0 && pp->p_cowcnt == 0 &&
51217c478bd9Sstevel@tonic-gate 	    !hat_page_is_mapped(pp)) {
51227c478bd9Sstevel@tonic-gate 
51237c478bd9Sstevel@tonic-gate 		/*
51247c478bd9Sstevel@tonic-gate 		 * If page is modified, unlock it
51257c478bd9Sstevel@tonic-gate 		 *
51267c478bd9Sstevel@tonic-gate 		 * (p_nrm & P_MOD) bit has the latest stuff because:
51277c478bd9Sstevel@tonic-gate 		 * (1) We found that this page doesn't have any mappings
51287c478bd9Sstevel@tonic-gate 		 *	_after_ holding SE_EXCL and
51297c478bd9Sstevel@tonic-gate 		 * (2) We didn't drop SE_EXCL lock after the check in (1)
51307c478bd9Sstevel@tonic-gate 		 */
51317c478bd9Sstevel@tonic-gate 		if (checkmod && hat_ismod(pp)) {
51327c478bd9Sstevel@tonic-gate 			page_unlock(pp);
51337c478bd9Sstevel@tonic-gate 			status = PGREL_MOD;
51347c478bd9Sstevel@tonic-gate 		} else {
51357c478bd9Sstevel@tonic-gate 			/*LINTED: constant in conditional context*/
51367c478bd9Sstevel@tonic-gate 			VN_DISPOSE(pp, B_FREE, 0, kcred);
51377c478bd9Sstevel@tonic-gate 			status = PGREL_CLEAN;
51387c478bd9Sstevel@tonic-gate 		}
51397c478bd9Sstevel@tonic-gate 	} else {
51407c478bd9Sstevel@tonic-gate 		page_unlock(pp);
51417c478bd9Sstevel@tonic-gate 		status = PGREL_NOTREL;
51427c478bd9Sstevel@tonic-gate 	}
51437c478bd9Sstevel@tonic-gate 	return (status);
51447c478bd9Sstevel@tonic-gate }
51457c478bd9Sstevel@tonic-gate 
5146db874c57Selowe /*
5147db874c57Selowe  * Given a constituent page, try to demote the large page on the freelist.
5148db874c57Selowe  *
5149db874c57Selowe  * Returns nonzero if the page could be demoted successfully. Returns with
5150db874c57Selowe  * the constituent page still locked.
5151db874c57Selowe  */
5152db874c57Selowe int
5153db874c57Selowe page_try_demote_free_pages(page_t *pp)
5154db874c57Selowe {
5155db874c57Selowe 	page_t *rootpp = pp;
5156db874c57Selowe 	pfn_t	pfn = page_pptonum(pp);
5157db874c57Selowe 	spgcnt_t npgs;
5158db874c57Selowe 	uint_t	szc = pp->p_szc;
5159db874c57Selowe 
5160db874c57Selowe 	ASSERT(PP_ISFREE(pp));
5161db874c57Selowe 	ASSERT(PAGE_EXCL(pp));
5162db874c57Selowe 
5163db874c57Selowe 	/*
5164db874c57Selowe 	 * Adjust rootpp and lock it, if `pp' is not the base
5165db874c57Selowe 	 * constituent page.
5166db874c57Selowe 	 */
5167db874c57Selowe 	npgs = page_get_pagecnt(pp->p_szc);
5168db874c57Selowe 	if (npgs == 1) {
5169db874c57Selowe 		return (0);
5170db874c57Selowe 	}
5171db874c57Selowe 
5172db874c57Selowe 	if (!IS_P2ALIGNED(pfn, npgs)) {
5173db874c57Selowe 		pfn = P2ALIGN(pfn, npgs);
5174db874c57Selowe 		rootpp = page_numtopp_nolock(pfn);
5175db874c57Selowe 	}
5176db874c57Selowe 
5177db874c57Selowe 	if (pp != rootpp && !page_trylock(rootpp, SE_EXCL)) {
5178db874c57Selowe 		return (0);
5179db874c57Selowe 	}
5180db874c57Selowe 
5181db874c57Selowe 	if (rootpp->p_szc != szc) {
5182db874c57Selowe 		if (pp != rootpp)
5183db874c57Selowe 			page_unlock(rootpp);
5184db874c57Selowe 		return (0);
5185db874c57Selowe 	}
5186db874c57Selowe 
5187db874c57Selowe 	page_demote_free_pages(rootpp);
5188db874c57Selowe 
5189db874c57Selowe 	if (pp != rootpp)
5190db874c57Selowe 		page_unlock(rootpp);
5191db874c57Selowe 
5192db874c57Selowe 	ASSERT(PP_ISFREE(pp));
5193db874c57Selowe 	ASSERT(PAGE_EXCL(pp));
5194db874c57Selowe 	return (1);
5195db874c57Selowe }
5196db874c57Selowe 
5197db874c57Selowe /*
5198db874c57Selowe  * Given a constituent page, try to demote the large page.
5199db874c57Selowe  *
5200db874c57Selowe  * Returns nonzero if the page could be demoted successfully. Returns with
5201db874c57Selowe  * the constituent page still locked.
5202db874c57Selowe  */
52037c478bd9Sstevel@tonic-gate int
52047c478bd9Sstevel@tonic-gate page_try_demote_pages(page_t *pp)
52057c478bd9Sstevel@tonic-gate {
52067c478bd9Sstevel@tonic-gate 	page_t *tpp, *rootpp = pp;
52077c478bd9Sstevel@tonic-gate 	pfn_t	pfn = page_pptonum(pp);
52087c478bd9Sstevel@tonic-gate 	spgcnt_t i, npgs;
52097c478bd9Sstevel@tonic-gate 	uint_t	szc = pp->p_szc;
52107c478bd9Sstevel@tonic-gate 	vnode_t *vp = pp->p_vnode;
52117c478bd9Sstevel@tonic-gate 
5212db874c57Selowe 	ASSERT(PAGE_EXCL(pp));
52137c478bd9Sstevel@tonic-gate 
52147c478bd9Sstevel@tonic-gate 	VM_STAT_ADD(pagecnt.pc_try_demote_pages[0]);
52157c478bd9Sstevel@tonic-gate 
5216db874c57Selowe 	if (pp->p_szc == 0) {
52177c478bd9Sstevel@tonic-gate 		VM_STAT_ADD(pagecnt.pc_try_demote_pages[1]);
52187c478bd9Sstevel@tonic-gate 		return (1);
52197c478bd9Sstevel@tonic-gate 	}
52207c478bd9Sstevel@tonic-gate 
52217c478bd9Sstevel@tonic-gate 	if (vp != NULL && !IS_SWAPFSVP(vp) && vp != &kvp) {
52227c478bd9Sstevel@tonic-gate 		VM_STAT_ADD(pagecnt.pc_try_demote_pages[2]);
5223db874c57Selowe 		page_demote_vp_pages(pp);
52247c478bd9Sstevel@tonic-gate 		ASSERT(pp->p_szc == 0);
52257c478bd9Sstevel@tonic-gate 		return (1);
52267c478bd9Sstevel@tonic-gate 	}
52277c478bd9Sstevel@tonic-gate 
52287c478bd9Sstevel@tonic-gate 	/*
5229db874c57Selowe 	 * Adjust rootpp if passed in is not the base
52307c478bd9Sstevel@tonic-gate 	 * constituent page.
52317c478bd9Sstevel@tonic-gate 	 */
5232db874c57Selowe 	npgs = page_get_pagecnt(pp->p_szc);
52337c478bd9Sstevel@tonic-gate 	ASSERT(npgs > 1);
52347c478bd9Sstevel@tonic-gate 	if (!IS_P2ALIGNED(pfn, npgs)) {
52357c478bd9Sstevel@tonic-gate 		pfn = P2ALIGN(pfn, npgs);
52367c478bd9Sstevel@tonic-gate 		rootpp = page_numtopp_nolock(pfn);
52377c478bd9Sstevel@tonic-gate 		VM_STAT_ADD(pagecnt.pc_try_demote_pages[3]);
52387c478bd9Sstevel@tonic-gate 		ASSERT(rootpp->p_vnode != NULL);
52397c478bd9Sstevel@tonic-gate 		ASSERT(rootpp->p_szc == szc);
52407c478bd9Sstevel@tonic-gate 	}
52417c478bd9Sstevel@tonic-gate 
52427c478bd9Sstevel@tonic-gate 	/*
52437c478bd9Sstevel@tonic-gate 	 * We can't demote kernel pages since we can't hat_unload()
52447c478bd9Sstevel@tonic-gate 	 * the mappings.
52457c478bd9Sstevel@tonic-gate 	 */
52467c478bd9Sstevel@tonic-gate 	if (rootpp->p_vnode == &kvp)
52477c478bd9Sstevel@tonic-gate 		return (0);
52487c478bd9Sstevel@tonic-gate 
52497c478bd9Sstevel@tonic-gate 	/*
52507c478bd9Sstevel@tonic-gate 	 * Attempt to lock all constituent pages except the page passed
52517c478bd9Sstevel@tonic-gate 	 * in since it's already locked.
52527c478bd9Sstevel@tonic-gate 	 */
5253affbd3ccSkchow 	for (tpp = rootpp, i = 0; i < npgs; i++, tpp++) {
52547c478bd9Sstevel@tonic-gate 		ASSERT(!PP_ISFREE(tpp));
52557c478bd9Sstevel@tonic-gate 		ASSERT(tpp->p_vnode != NULL);
52567c478bd9Sstevel@tonic-gate 
52577c478bd9Sstevel@tonic-gate 		if (tpp != pp && !page_trylock(tpp, SE_EXCL))
52587c478bd9Sstevel@tonic-gate 			break;
52597c478bd9Sstevel@tonic-gate 		ASSERT(tpp->p_szc == rootpp->p_szc);
52607c478bd9Sstevel@tonic-gate 		ASSERT(page_pptonum(tpp) == page_pptonum(rootpp) + i);
52617c478bd9Sstevel@tonic-gate 	}
52627c478bd9Sstevel@tonic-gate 
52637c478bd9Sstevel@tonic-gate 	/*
5264db874c57Selowe 	 * If we failed to lock them all then unlock what we have
5265db874c57Selowe 	 * locked so far and bail.
52667c478bd9Sstevel@tonic-gate 	 */
52677c478bd9Sstevel@tonic-gate 	if (i < npgs) {
52687c478bd9Sstevel@tonic-gate 		tpp = rootpp;
52697c478bd9Sstevel@tonic-gate 		while (i-- > 0) {
52707c478bd9Sstevel@tonic-gate 			if (tpp != pp)
52717c478bd9Sstevel@tonic-gate 				page_unlock(tpp);
5272affbd3ccSkchow 			tpp++;
52737c478bd9Sstevel@tonic-gate 		}
52747c478bd9Sstevel@tonic-gate 		VM_STAT_ADD(pagecnt.pc_try_demote_pages[4]);
52757c478bd9Sstevel@tonic-gate 		return (0);
52767c478bd9Sstevel@tonic-gate 	}
52777c478bd9Sstevel@tonic-gate 
5278affbd3ccSkchow 	for (tpp = rootpp, i = 0; i < npgs; i++, tpp++) {
52797c478bd9Sstevel@tonic-gate 		ASSERT(PAGE_EXCL(tpp));
528007b65a64Saguzovsk 		ASSERT(tpp->p_slckcnt == 0);
5281db874c57Selowe 		(void) hat_pageunload(tpp, HAT_FORCE_PGUNLOAD);
52827c478bd9Sstevel@tonic-gate 		tpp->p_szc = 0;
52837c478bd9Sstevel@tonic-gate 	}
52847c478bd9Sstevel@tonic-gate 
52857c478bd9Sstevel@tonic-gate 	/*
52867c478bd9Sstevel@tonic-gate 	 * Unlock all pages except the page passed in.
52877c478bd9Sstevel@tonic-gate 	 */
5288affbd3ccSkchow 	for (tpp = rootpp, i = 0; i < npgs; i++, tpp++) {
52897c478bd9Sstevel@tonic-gate 		ASSERT(!hat_page_is_mapped(tpp));
52907c478bd9Sstevel@tonic-gate 		if (tpp != pp)
52917c478bd9Sstevel@tonic-gate 			page_unlock(tpp);
52927c478bd9Sstevel@tonic-gate 	}
5293db874c57Selowe 
52947c478bd9Sstevel@tonic-gate 	VM_STAT_ADD(pagecnt.pc_try_demote_pages[5]);
52957c478bd9Sstevel@tonic-gate 	return (1);
52967c478bd9Sstevel@tonic-gate }
52977c478bd9Sstevel@tonic-gate 
52987c478bd9Sstevel@tonic-gate /*
52997c478bd9Sstevel@tonic-gate  * Called by page_free() and page_destroy() to demote the page size code
53007c478bd9Sstevel@tonic-gate  * (p_szc) to 0 (since we can't just put a single PAGESIZE page with non zero
53017c478bd9Sstevel@tonic-gate  * p_szc on free list, neither can we just clear p_szc of a single page_t
53027c478bd9Sstevel@tonic-gate  * within a large page since it will break other code that relies on p_szc
53037c478bd9Sstevel@tonic-gate  * being the same for all page_t's of a large page). Anonymous pages should
53047c478bd9Sstevel@tonic-gate  * never end up here because anon_map_getpages() cannot deal with p_szc
53057c478bd9Sstevel@tonic-gate  * changes after a single constituent page is locked.  While anonymous or
53067c478bd9Sstevel@tonic-gate  * kernel large pages are demoted or freed the entire large page at a time
53077c478bd9Sstevel@tonic-gate  * with all constituent pages locked EXCL for the file system pages we
53087c478bd9Sstevel@tonic-gate  * have to be able to demote a large page (i.e. decrease all constituent pages
53097c478bd9Sstevel@tonic-gate  * p_szc) with only just an EXCL lock on one of constituent pages. The reason
53107c478bd9Sstevel@tonic-gate  * we can easily deal with anonymous page demotion the entire large page at a
53117c478bd9Sstevel@tonic-gate  * time is that those operation originate at address space level and concern
53127c478bd9Sstevel@tonic-gate  * the entire large page region with actual demotion only done when pages are
53137c478bd9Sstevel@tonic-gate  * not shared with any other processes (therefore we can always get EXCL lock
53147c478bd9Sstevel@tonic-gate  * on all anonymous constituent pages after clearing segment page
53157c478bd9Sstevel@tonic-gate  * cache). However file system pages can be truncated or invalidated at a
53167c478bd9Sstevel@tonic-gate  * PAGESIZE level from the file system side and end up in page_free() or
53177c478bd9Sstevel@tonic-gate  * page_destroy() (we also allow only part of the large page to be SOFTLOCKed
53187c478bd9Sstevel@tonic-gate  * and therfore pageout should be able to demote a large page by EXCL locking
53197c478bd9Sstevel@tonic-gate  * any constituent page that is not under SOFTLOCK). In those cases we cannot
53207c478bd9Sstevel@tonic-gate  * rely on being able to lock EXCL all constituent pages.
53217c478bd9Sstevel@tonic-gate  *
53227c478bd9Sstevel@tonic-gate  * To prevent szc changes on file system pages one has to lock all constituent
53237c478bd9Sstevel@tonic-gate  * pages at least SHARED (or call page_szc_lock()). The only subsystem that
53247c478bd9Sstevel@tonic-gate  * doesn't rely on locking all constituent pages (or using page_szc_lock()) to
53257c478bd9Sstevel@tonic-gate  * prevent szc changes is hat layer that uses its own page level mlist
53267c478bd9Sstevel@tonic-gate  * locks. hat assumes that szc doesn't change after mlist lock for a page is
53277c478bd9Sstevel@tonic-gate  * taken. Therefore we need to change szc under hat level locks if we only
53287c478bd9Sstevel@tonic-gate  * have an EXCL lock on a single constituent page and hat still references any
53297c478bd9Sstevel@tonic-gate  * of constituent pages.  (Note we can't "ignore" hat layer by simply
53307c478bd9Sstevel@tonic-gate  * hat_pageunload() all constituent pages without having EXCL locks on all of
53317c478bd9Sstevel@tonic-gate  * constituent pages). We use hat_page_demote() call to safely demote szc of
53327c478bd9Sstevel@tonic-gate  * all constituent pages under hat locks when we only have an EXCL lock on one
53337c478bd9Sstevel@tonic-gate  * of constituent pages.
53347c478bd9Sstevel@tonic-gate  *
53357c478bd9Sstevel@tonic-gate  * This routine calls page_szc_lock() before calling hat_page_demote() to
53367c478bd9Sstevel@tonic-gate  * allow segvn in one special case not to lock all constituent pages SHARED
53377c478bd9Sstevel@tonic-gate  * before calling hat_memload_array() that relies on p_szc not changeing even
53387c478bd9Sstevel@tonic-gate  * before hat level mlist lock is taken.  In that case segvn uses
53397c478bd9Sstevel@tonic-gate  * page_szc_lock() to prevent hat_page_demote() changeing p_szc values.
53407c478bd9Sstevel@tonic-gate  *
53417c478bd9Sstevel@tonic-gate  * Anonymous or kernel page demotion still has to lock all pages exclusively
53427c478bd9Sstevel@tonic-gate  * and do hat_pageunload() on all constituent pages before demoting the page
53437c478bd9Sstevel@tonic-gate  * therefore there's no need for anonymous or kernel page demotion to use
53447c478bd9Sstevel@tonic-gate  * hat_page_demote() mechanism.
53457c478bd9Sstevel@tonic-gate  *
53467c478bd9Sstevel@tonic-gate  * hat_page_demote() removes all large mappings that map pp and then decreases
53477c478bd9Sstevel@tonic-gate  * p_szc starting from the last constituent page of the large page. By working
53487c478bd9Sstevel@tonic-gate  * from the tail of a large page in pfn decreasing order allows one looking at
53497c478bd9Sstevel@tonic-gate  * the root page to know that hat_page_demote() is done for root's szc area.
53507c478bd9Sstevel@tonic-gate  * e.g. if a root page has szc 1 one knows it only has to lock all constituent
53517c478bd9Sstevel@tonic-gate  * pages within szc 1 area to prevent szc changes because hat_page_demote()
53527c478bd9Sstevel@tonic-gate  * that started on this page when it had szc > 1 is done for this szc 1 area.
53537c478bd9Sstevel@tonic-gate  *
53547c478bd9Sstevel@tonic-gate  * We are guranteed that all constituent pages of pp's large page belong to
53557c478bd9Sstevel@tonic-gate  * the same vnode with the consecutive offsets increasing in the direction of
53567c478bd9Sstevel@tonic-gate  * the pfn i.e. the identity of constituent pages can't change until their
53577c478bd9Sstevel@tonic-gate  * p_szc is decreased. Therefore it's safe for hat_page_demote() to remove
53587c478bd9Sstevel@tonic-gate  * large mappings to pp even though we don't lock any constituent page except
53597c478bd9Sstevel@tonic-gate  * pp (i.e. we won't unload e.g. kernel locked page).
53607c478bd9Sstevel@tonic-gate  */
53617c478bd9Sstevel@tonic-gate static void
53627c478bd9Sstevel@tonic-gate page_demote_vp_pages(page_t *pp)
53637c478bd9Sstevel@tonic-gate {
53647c478bd9Sstevel@tonic-gate 	kmutex_t *mtx;
53657c478bd9Sstevel@tonic-gate 
53667c478bd9Sstevel@tonic-gate 	ASSERT(PAGE_EXCL(pp));
53677c478bd9Sstevel@tonic-gate 	ASSERT(!PP_ISFREE(pp));
53687c478bd9Sstevel@tonic-gate 	ASSERT(pp->p_vnode != NULL);
53697c478bd9Sstevel@tonic-gate 	ASSERT(!IS_SWAPFSVP(pp->p_vnode));
53707c478bd9Sstevel@tonic-gate 	ASSERT(pp->p_vnode != &kvp);
53717c478bd9Sstevel@tonic-gate 
53727c478bd9Sstevel@tonic-gate 	VM_STAT_ADD(pagecnt.pc_demote_pages[0]);
53737c478bd9Sstevel@tonic-gate 
53747c478bd9Sstevel@tonic-gate 	mtx = page_szc_lock(pp);
53757c478bd9Sstevel@tonic-gate 	if (mtx != NULL) {
53767c478bd9Sstevel@tonic-gate 		hat_page_demote(pp);
53777c478bd9Sstevel@tonic-gate 		mutex_exit(mtx);
53787c478bd9Sstevel@tonic-gate 	}
53797c478bd9Sstevel@tonic-gate 	ASSERT(pp->p_szc == 0);
53807c478bd9Sstevel@tonic-gate }
53817c478bd9Sstevel@tonic-gate 
53827c478bd9Sstevel@tonic-gate /*
53837c478bd9Sstevel@tonic-gate  * Mark any existing pages for migration in the given range
53847c478bd9Sstevel@tonic-gate  */
53857c478bd9Sstevel@tonic-gate void
53867c478bd9Sstevel@tonic-gate page_mark_migrate(struct seg *seg, caddr_t addr, size_t len,
53877c478bd9Sstevel@tonic-gate     struct anon_map *amp, ulong_t anon_index, vnode_t *vp,
53887c478bd9Sstevel@tonic-gate     u_offset_t vnoff, int rflag)
53897c478bd9Sstevel@tonic-gate {
53907c478bd9Sstevel@tonic-gate 	struct anon	*ap;
53917c478bd9Sstevel@tonic-gate 	vnode_t		*curvp;
53927c478bd9Sstevel@tonic-gate 	lgrp_t		*from;
53937c478bd9Sstevel@tonic-gate 	pgcnt_t		i;
53947c478bd9Sstevel@tonic-gate 	pgcnt_t		nlocked;
53957c478bd9Sstevel@tonic-gate 	u_offset_t	off;
53967c478bd9Sstevel@tonic-gate 	pfn_t		pfn;
53977c478bd9Sstevel@tonic-gate 	size_t		pgsz;
53987c478bd9Sstevel@tonic-gate 	size_t		segpgsz;
53997c478bd9Sstevel@tonic-gate 	pgcnt_t		pages;
54007c478bd9Sstevel@tonic-gate 	uint_t		pszc;
54017c478bd9Sstevel@tonic-gate 	page_t		**ppa;
54027c478bd9Sstevel@tonic-gate 	pgcnt_t		ppa_nentries;
54037c478bd9Sstevel@tonic-gate 	page_t		*pp;
54047c478bd9Sstevel@tonic-gate 	caddr_t		va;
54057c478bd9Sstevel@tonic-gate 	ulong_t		an_idx;
54067c478bd9Sstevel@tonic-gate 	anon_sync_obj_t	cookie;
54077c478bd9Sstevel@tonic-gate 
54087c478bd9Sstevel@tonic-gate 	ASSERT(seg->s_as && AS_LOCK_HELD(seg->s_as, &seg->s_as->a_lock));
54097c478bd9Sstevel@tonic-gate 
54107c478bd9Sstevel@tonic-gate 	/*
54117c478bd9Sstevel@tonic-gate 	 * Don't do anything if don't need to do lgroup optimizations
54127c478bd9Sstevel@tonic-gate 	 * on this system
54137c478bd9Sstevel@tonic-gate 	 */
54147c478bd9Sstevel@tonic-gate 	if (!lgrp_optimizations())
54157c478bd9Sstevel@tonic-gate 		return;
54167c478bd9Sstevel@tonic-gate 
54177c478bd9Sstevel@tonic-gate 	/*
54187c478bd9Sstevel@tonic-gate 	 * Align address and length to (potentially large) page boundary
54197c478bd9Sstevel@tonic-gate 	 */
54207c478bd9Sstevel@tonic-gate 	segpgsz = page_get_pagesize(seg->s_szc);
54217c478bd9Sstevel@tonic-gate 	addr = (caddr_t)P2ALIGN((uintptr_t)addr, segpgsz);
54227c478bd9Sstevel@tonic-gate 	if (rflag)
54237c478bd9Sstevel@tonic-gate 		len = P2ROUNDUP(len, segpgsz);
54247c478bd9Sstevel@tonic-gate 
54257c478bd9Sstevel@tonic-gate 	/*
54267c478bd9Sstevel@tonic-gate 	 * Allocate page array to accomodate largest page size
54277c478bd9Sstevel@tonic-gate 	 */
54287c478bd9Sstevel@tonic-gate 	pgsz = page_get_pagesize(page_num_pagesizes() - 1);
54297c478bd9Sstevel@tonic-gate 	ppa_nentries = btop(pgsz);
54307c478bd9Sstevel@tonic-gate 	ppa = kmem_zalloc(ppa_nentries * sizeof (page_t *), KM_SLEEP);
54317c478bd9Sstevel@tonic-gate 
54327c478bd9Sstevel@tonic-gate 	/*
54337c478bd9Sstevel@tonic-gate 	 * Do one (large) page at a time
54347c478bd9Sstevel@tonic-gate 	 */
54357c478bd9Sstevel@tonic-gate 	va = addr;
54367c478bd9Sstevel@tonic-gate 	while (va < addr + len) {
54377c478bd9Sstevel@tonic-gate 		/*
54387c478bd9Sstevel@tonic-gate 		 * Lookup (root) page for vnode and offset corresponding to
54397c478bd9Sstevel@tonic-gate 		 * this virtual address
54407c478bd9Sstevel@tonic-gate 		 * Try anonmap first since there may be copy-on-write
54417c478bd9Sstevel@tonic-gate 		 * pages, but initialize vnode pointer and offset using
54427c478bd9Sstevel@tonic-gate 		 * vnode arguments just in case there isn't an amp.
54437c478bd9Sstevel@tonic-gate 		 */
54447c478bd9Sstevel@tonic-gate 		curvp = vp;
54457c478bd9Sstevel@tonic-gate 		off = vnoff + va - seg->s_base;
54467c478bd9Sstevel@tonic-gate 		if (amp) {
54477c478bd9Sstevel@tonic-gate 			ANON_LOCK_ENTER(&amp->a_rwlock, RW_READER);
54487c478bd9Sstevel@tonic-gate 			an_idx = anon_index + seg_page(seg, va);
54497c478bd9Sstevel@tonic-gate 			anon_array_enter(amp, an_idx, &cookie);
54507c478bd9Sstevel@tonic-gate 			ap = anon_get_ptr(amp->ahp, an_idx);
54517c478bd9Sstevel@tonic-gate 			if (ap)
54527c478bd9Sstevel@tonic-gate 				swap_xlate(ap, &curvp, &off);
54537c478bd9Sstevel@tonic-gate 			anon_array_exit(&cookie);
54547c478bd9Sstevel@tonic-gate 			ANON_LOCK_EXIT(&amp->a_rwlock);
54557c478bd9Sstevel@tonic-gate 		}
54567c478bd9Sstevel@tonic-gate 
54577c478bd9Sstevel@tonic-gate 		pp = NULL;
54587c478bd9Sstevel@tonic-gate 		if (curvp)
54597c478bd9Sstevel@tonic-gate 			pp = page_lookup(curvp, off, SE_SHARED);
54607c478bd9Sstevel@tonic-gate 
54617c478bd9Sstevel@tonic-gate 		/*
54627c478bd9Sstevel@tonic-gate 		 * If there isn't a page at this virtual address,
54637c478bd9Sstevel@tonic-gate 		 * skip to next page
54647c478bd9Sstevel@tonic-gate 		 */
54657c478bd9Sstevel@tonic-gate 		if (pp == NULL) {
54667c478bd9Sstevel@tonic-gate 			va += PAGESIZE;
54677c478bd9Sstevel@tonic-gate 			continue;
54687c478bd9Sstevel@tonic-gate 		}
54697c478bd9Sstevel@tonic-gate 
54707c478bd9Sstevel@tonic-gate 		/*
54717c478bd9Sstevel@tonic-gate 		 * Figure out which lgroup this page is in for kstats
54727c478bd9Sstevel@tonic-gate 		 */
54737c478bd9Sstevel@tonic-gate 		pfn = page_pptonum(pp);
54747c478bd9Sstevel@tonic-gate 		from = lgrp_pfn_to_lgrp(pfn);
54757c478bd9Sstevel@tonic-gate 
54767c478bd9Sstevel@tonic-gate 		/*
54777c478bd9Sstevel@tonic-gate 		 * Get page size, and round up and skip to next page boundary
54787c478bd9Sstevel@tonic-gate 		 * if unaligned address
54797c478bd9Sstevel@tonic-gate 		 */
54807c478bd9Sstevel@tonic-gate 		pszc = pp->p_szc;
54817c478bd9Sstevel@tonic-gate 		pgsz = page_get_pagesize(pszc);
54827c478bd9Sstevel@tonic-gate 		pages = btop(pgsz);
54837c478bd9Sstevel@tonic-gate 		if (!IS_P2ALIGNED(va, pgsz) ||
54847c478bd9Sstevel@tonic-gate 		    !IS_P2ALIGNED(pfn, pages) ||
54857c478bd9Sstevel@tonic-gate 		    pgsz > segpgsz) {
54867c478bd9Sstevel@tonic-gate 			pgsz = MIN(pgsz, segpgsz);
54877c478bd9Sstevel@tonic-gate 			page_unlock(pp);
54887c478bd9Sstevel@tonic-gate 			i = btop(P2END((uintptr_t)va, pgsz) -
54897c478bd9Sstevel@tonic-gate 			    (uintptr_t)va);
54907c478bd9Sstevel@tonic-gate 			va = (caddr_t)P2END((uintptr_t)va, pgsz);
54917c478bd9Sstevel@tonic-gate 			lgrp_stat_add(from->lgrp_id, LGRP_PMM_FAIL_PGS, i);
54927c478bd9Sstevel@tonic-gate 			continue;
54937c478bd9Sstevel@tonic-gate 		}
54947c478bd9Sstevel@tonic-gate 
54957c478bd9Sstevel@tonic-gate 		/*
54967c478bd9Sstevel@tonic-gate 		 * Upgrade to exclusive lock on page
54977c478bd9Sstevel@tonic-gate 		 */
54987c478bd9Sstevel@tonic-gate 		if (!page_tryupgrade(pp)) {
54997c478bd9Sstevel@tonic-gate 			page_unlock(pp);
55007c478bd9Sstevel@tonic-gate 			va += pgsz;
55017c478bd9Sstevel@tonic-gate 			lgrp_stat_add(from->lgrp_id, LGRP_PMM_FAIL_PGS,
55027c478bd9Sstevel@tonic-gate 			    btop(pgsz));
55037c478bd9Sstevel@tonic-gate 			continue;
55047c478bd9Sstevel@tonic-gate 		}
55057c478bd9Sstevel@tonic-gate 
55067c478bd9Sstevel@tonic-gate 		/*
55077c478bd9Sstevel@tonic-gate 		 * Remember pages locked exclusively and how many
55087c478bd9Sstevel@tonic-gate 		 */
55097c478bd9Sstevel@tonic-gate 		ppa[0] = pp;
55107c478bd9Sstevel@tonic-gate 		nlocked = 1;
55117c478bd9Sstevel@tonic-gate 
55127c478bd9Sstevel@tonic-gate 		/*
55137c478bd9Sstevel@tonic-gate 		 * Lock constituent pages if this is large page
55147c478bd9Sstevel@tonic-gate 		 */
55157c478bd9Sstevel@tonic-gate 		if (pages > 1) {
55167c478bd9Sstevel@tonic-gate 			/*
55177c478bd9Sstevel@tonic-gate 			 * Lock all constituents except root page, since it
55187c478bd9Sstevel@tonic-gate 			 * should be locked already.
55197c478bd9Sstevel@tonic-gate 			 */
55207c478bd9Sstevel@tonic-gate 			for (i = 1; i < pages; i++) {
5521affbd3ccSkchow 				pp++;
55227c478bd9Sstevel@tonic-gate 				if (!page_trylock(pp, SE_EXCL)) {
55237c478bd9Sstevel@tonic-gate 					break;
55247c478bd9Sstevel@tonic-gate 				}
55257c478bd9Sstevel@tonic-gate 				if (PP_ISFREE(pp) ||
55267c478bd9Sstevel@tonic-gate 				    pp->p_szc != pszc) {
55277c478bd9Sstevel@tonic-gate 					/*
55287c478bd9Sstevel@tonic-gate 					 * hat_page_demote() raced in with us.
55297c478bd9Sstevel@tonic-gate 					 */
55307c478bd9Sstevel@tonic-gate 					ASSERT(!IS_SWAPFSVP(curvp));
55317c478bd9Sstevel@tonic-gate 					page_unlock(pp);
55327c478bd9Sstevel@tonic-gate 					break;
55337c478bd9Sstevel@tonic-gate 				}
55347c478bd9Sstevel@tonic-gate 				ppa[nlocked] = pp;
55357c478bd9Sstevel@tonic-gate 				nlocked++;
55367c478bd9Sstevel@tonic-gate 			}
55377c478bd9Sstevel@tonic-gate 		}
55387c478bd9Sstevel@tonic-gate 
55397c478bd9Sstevel@tonic-gate 		/*
55407c478bd9Sstevel@tonic-gate 		 * If all constituent pages couldn't be locked,
55417c478bd9Sstevel@tonic-gate 		 * unlock pages locked so far and skip to next page.
55427c478bd9Sstevel@tonic-gate 		 */
55437c478bd9Sstevel@tonic-gate 		if (nlocked != pages) {
55447c478bd9Sstevel@tonic-gate 			for (i = 0; i < nlocked; i++)
55457c478bd9Sstevel@tonic-gate 				page_unlock(ppa[i]);
55467c478bd9Sstevel@tonic-gate 			va += pgsz;
55477c478bd9Sstevel@tonic-gate 			lgrp_stat_add(from->lgrp_id, LGRP_PMM_FAIL_PGS,
55487c478bd9Sstevel@tonic-gate 			    btop(pgsz));
55497c478bd9Sstevel@tonic-gate 			continue;
55507c478bd9Sstevel@tonic-gate 		}
55517c478bd9Sstevel@tonic-gate 
55527c478bd9Sstevel@tonic-gate 		/*
55537c478bd9Sstevel@tonic-gate 		 * hat_page_demote() can no longer happen
55547c478bd9Sstevel@tonic-gate 		 * since last cons page had the right p_szc after
55557c478bd9Sstevel@tonic-gate 		 * all cons pages were locked. all cons pages
55567c478bd9Sstevel@tonic-gate 		 * should now have the same p_szc.
55577c478bd9Sstevel@tonic-gate 		 */
55587c478bd9Sstevel@tonic-gate 
55597c478bd9Sstevel@tonic-gate 		/*
55607c478bd9Sstevel@tonic-gate 		 * All constituent pages locked successfully, so mark
55617c478bd9Sstevel@tonic-gate 		 * large page for migration and unload the mappings of
55627c478bd9Sstevel@tonic-gate 		 * constituent pages, so a fault will occur on any part of the
55637c478bd9Sstevel@tonic-gate 		 * large page
55647c478bd9Sstevel@tonic-gate 		 */
55657c478bd9Sstevel@tonic-gate 		PP_SETMIGRATE(ppa[0]);
55667c478bd9Sstevel@tonic-gate 		for (i = 0; i < nlocked; i++) {
55677c478bd9Sstevel@tonic-gate 			pp = ppa[i];
55687c478bd9Sstevel@tonic-gate 			(void) hat_pageunload(pp, HAT_FORCE_PGUNLOAD);
55697c478bd9Sstevel@tonic-gate 			ASSERT(hat_page_getshare(pp) == 0);
55707c478bd9Sstevel@tonic-gate 			page_unlock(pp);
55717c478bd9Sstevel@tonic-gate 		}
55727c478bd9Sstevel@tonic-gate 		lgrp_stat_add(from->lgrp_id, LGRP_PMM_PGS, nlocked);
55737c478bd9Sstevel@tonic-gate 
55747c478bd9Sstevel@tonic-gate 		va += pgsz;
55757c478bd9Sstevel@tonic-gate 	}
55767c478bd9Sstevel@tonic-gate 	kmem_free(ppa, ppa_nentries * sizeof (page_t *));
55777c478bd9Sstevel@tonic-gate }
55787c478bd9Sstevel@tonic-gate 
55797c478bd9Sstevel@tonic-gate /*
55807c478bd9Sstevel@tonic-gate  * Migrate any pages that have been marked for migration in the given range
55817c478bd9Sstevel@tonic-gate  */
55827c478bd9Sstevel@tonic-gate void
55837c478bd9Sstevel@tonic-gate page_migrate(
55847c478bd9Sstevel@tonic-gate 	struct seg	*seg,
55857c478bd9Sstevel@tonic-gate 	caddr_t		addr,
55867c478bd9Sstevel@tonic-gate 	page_t		**ppa,
55877c478bd9Sstevel@tonic-gate 	pgcnt_t		npages)
55887c478bd9Sstevel@tonic-gate {
55897c478bd9Sstevel@tonic-gate 	lgrp_t		*from;
55907c478bd9Sstevel@tonic-gate 	lgrp_t		*to;
55917c478bd9Sstevel@tonic-gate 	page_t		*newpp;
55927c478bd9Sstevel@tonic-gate 	page_t		*pp;
55937c478bd9Sstevel@tonic-gate 	pfn_t		pfn;
55947c478bd9Sstevel@tonic-gate 	size_t		pgsz;
55957c478bd9Sstevel@tonic-gate 	spgcnt_t	page_cnt;
55967c478bd9Sstevel@tonic-gate 	spgcnt_t	i;
55977c478bd9Sstevel@tonic-gate 	uint_t		pszc;
55987c478bd9Sstevel@tonic-gate 
55997c478bd9Sstevel@tonic-gate 	ASSERT(seg->s_as && AS_LOCK_HELD(seg->s_as, &seg->s_as->a_lock));
56007c478bd9Sstevel@tonic-gate 
56017c478bd9Sstevel@tonic-gate 	while (npages > 0) {
56027c478bd9Sstevel@tonic-gate 		pp = *ppa;
56037c478bd9Sstevel@tonic-gate 		pszc = pp->p_szc;
56047c478bd9Sstevel@tonic-gate 		pgsz = page_get_pagesize(pszc);
56057c478bd9Sstevel@tonic-gate 		page_cnt = btop(pgsz);
56067c478bd9Sstevel@tonic-gate 
56077c478bd9Sstevel@tonic-gate 		/*
56087c478bd9Sstevel@tonic-gate 		 * Check to see whether this page is marked for migration
56097c478bd9Sstevel@tonic-gate 		 *
56107c478bd9Sstevel@tonic-gate 		 * Assume that root page of large page is marked for
56117c478bd9Sstevel@tonic-gate 		 * migration and none of the other constituent pages
56127c478bd9Sstevel@tonic-gate 		 * are marked.  This really simplifies clearing the
56137c478bd9Sstevel@tonic-gate 		 * migrate bit by not having to clear it from each
56147c478bd9Sstevel@tonic-gate 		 * constituent page.
56157c478bd9Sstevel@tonic-gate 		 *
56167c478bd9Sstevel@tonic-gate 		 * note we don't want to relocate an entire large page if
56177c478bd9Sstevel@tonic-gate 		 * someone is only using one subpage.
56187c478bd9Sstevel@tonic-gate 		 */
56197c478bd9Sstevel@tonic-gate 		if (npages < page_cnt)
56207c478bd9Sstevel@tonic-gate 			break;
56217c478bd9Sstevel@tonic-gate 
56227c478bd9Sstevel@tonic-gate 		/*
56237c478bd9Sstevel@tonic-gate 		 * Is it marked for migration?
56247c478bd9Sstevel@tonic-gate 		 */
56257c478bd9Sstevel@tonic-gate 		if (!PP_ISMIGRATE(pp))
56267c478bd9Sstevel@tonic-gate 			goto next;
56277c478bd9Sstevel@tonic-gate 
56287c478bd9Sstevel@tonic-gate 		/*
56297c478bd9Sstevel@tonic-gate 		 * Determine lgroups that page is being migrated between
56307c478bd9Sstevel@tonic-gate 		 */
56317c478bd9Sstevel@tonic-gate 		pfn = page_pptonum(pp);
56327c478bd9Sstevel@tonic-gate 		if (!IS_P2ALIGNED(pfn, page_cnt)) {
56337c478bd9Sstevel@tonic-gate 			break;
56347c478bd9Sstevel@tonic-gate 		}
56357c478bd9Sstevel@tonic-gate 		from = lgrp_pfn_to_lgrp(pfn);
56367c478bd9Sstevel@tonic-gate 		to = lgrp_mem_choose(seg, addr, pgsz);
56377c478bd9Sstevel@tonic-gate 
56387c478bd9Sstevel@tonic-gate 		/*
56397c478bd9Sstevel@tonic-gate 		 * Check to see whether we are trying to migrate page to lgroup
56407c478bd9Sstevel@tonic-gate 		 * where it is allocated already
56417c478bd9Sstevel@tonic-gate 		 */
56427c478bd9Sstevel@tonic-gate 		if (to == from) {
56437c478bd9Sstevel@tonic-gate 			PP_CLRMIGRATE(pp);
56447c478bd9Sstevel@tonic-gate 			goto next;
56457c478bd9Sstevel@tonic-gate 		}
56467c478bd9Sstevel@tonic-gate 
56477c478bd9Sstevel@tonic-gate 		/*
56487c478bd9Sstevel@tonic-gate 		 * Need to get exclusive lock's to migrate
56497c478bd9Sstevel@tonic-gate 		 */
56507c478bd9Sstevel@tonic-gate 		for (i = 0; i < page_cnt; i++) {
56517c478bd9Sstevel@tonic-gate 			ASSERT(PAGE_LOCKED(ppa[i]));
56527c478bd9Sstevel@tonic-gate 			if (page_pptonum(ppa[i]) != pfn + i ||
56537c478bd9Sstevel@tonic-gate 			    ppa[i]->p_szc != pszc) {
56547c478bd9Sstevel@tonic-gate 				break;
56557c478bd9Sstevel@tonic-gate 			}
56567c478bd9Sstevel@tonic-gate 			if (!page_tryupgrade(ppa[i])) {
56577c478bd9Sstevel@tonic-gate 				lgrp_stat_add(from->lgrp_id,
56587c478bd9Sstevel@tonic-gate 				    LGRP_PM_FAIL_LOCK_PGS,
56597c478bd9Sstevel@tonic-gate 				    page_cnt);
56607c478bd9Sstevel@tonic-gate 				break;
56617c478bd9Sstevel@tonic-gate 			}
56627c478bd9Sstevel@tonic-gate 		}
56637c478bd9Sstevel@tonic-gate 		if (i != page_cnt) {
56647c478bd9Sstevel@tonic-gate 			while (--i != -1) {
56657c478bd9Sstevel@tonic-gate 				page_downgrade(ppa[i]);
56667c478bd9Sstevel@tonic-gate 			}
56677c478bd9Sstevel@tonic-gate 			goto next;
56687c478bd9Sstevel@tonic-gate 		}
56697c478bd9Sstevel@tonic-gate 
56707c478bd9Sstevel@tonic-gate 		(void) page_create_wait(page_cnt, PG_WAIT);
56717c478bd9Sstevel@tonic-gate 		newpp = page_get_replacement_page(pp, to, PGR_SAMESZC);
56727c478bd9Sstevel@tonic-gate 		if (newpp == NULL) {
56737c478bd9Sstevel@tonic-gate 			page_create_putback(page_cnt);
56747c478bd9Sstevel@tonic-gate 			for (i = 0; i < page_cnt; i++) {
56757c478bd9Sstevel@tonic-gate 				page_downgrade(ppa[i]);
56767c478bd9Sstevel@tonic-gate 			}
56777c478bd9Sstevel@tonic-gate 			lgrp_stat_add(to->lgrp_id, LGRP_PM_FAIL_ALLOC_PGS,
56787c478bd9Sstevel@tonic-gate 			    page_cnt);
56797c478bd9Sstevel@tonic-gate 			goto next;
56807c478bd9Sstevel@tonic-gate 		}
56817c478bd9Sstevel@tonic-gate 		ASSERT(newpp->p_szc == pszc);
56827c478bd9Sstevel@tonic-gate 		/*
56837c478bd9Sstevel@tonic-gate 		 * Clear migrate bit and relocate page
56847c478bd9Sstevel@tonic-gate 		 */
56857c478bd9Sstevel@tonic-gate 		PP_CLRMIGRATE(pp);
56867c478bd9Sstevel@tonic-gate 		if (page_relocate(&pp, &newpp, 0, 1, &page_cnt, to)) {
56877c478bd9Sstevel@tonic-gate 			panic("page_migrate: page_relocate failed");
56887c478bd9Sstevel@tonic-gate 		}
56897c478bd9Sstevel@tonic-gate 		ASSERT(page_cnt * PAGESIZE == pgsz);
56907c478bd9Sstevel@tonic-gate 
56917c478bd9Sstevel@tonic-gate 		/*
56927c478bd9Sstevel@tonic-gate 		 * Keep stats for number of pages migrated from and to
56937c478bd9Sstevel@tonic-gate 		 * each lgroup
56947c478bd9Sstevel@tonic-gate 		 */
56957c478bd9Sstevel@tonic-gate 		lgrp_stat_add(from->lgrp_id, LGRP_PM_SRC_PGS, page_cnt);
56967c478bd9Sstevel@tonic-gate 		lgrp_stat_add(to->lgrp_id, LGRP_PM_DEST_PGS, page_cnt);
56977c478bd9Sstevel@tonic-gate 		/*
56987c478bd9Sstevel@tonic-gate 		 * update the page_t array we were passed in and
56997c478bd9Sstevel@tonic-gate 		 * unlink constituent pages of a large page.
57007c478bd9Sstevel@tonic-gate 		 */
57017c478bd9Sstevel@tonic-gate 		for (i = 0; i < page_cnt; ++i, ++pp) {
57027c478bd9Sstevel@tonic-gate 			ASSERT(PAGE_EXCL(newpp));
57037c478bd9Sstevel@tonic-gate 			ASSERT(newpp->p_szc == pszc);
57047c478bd9Sstevel@tonic-gate 			ppa[i] = newpp;
57057c478bd9Sstevel@tonic-gate 			pp = newpp;
57067c478bd9Sstevel@tonic-gate 			page_sub(&newpp, pp);
57077c478bd9Sstevel@tonic-gate 			page_downgrade(pp);
57087c478bd9Sstevel@tonic-gate 		}
57097c478bd9Sstevel@tonic-gate 		ASSERT(newpp == NULL);
57107c478bd9Sstevel@tonic-gate next:
57117c478bd9Sstevel@tonic-gate 		addr += pgsz;
57127c478bd9Sstevel@tonic-gate 		ppa += page_cnt;
57137c478bd9Sstevel@tonic-gate 		npages -= page_cnt;
57147c478bd9Sstevel@tonic-gate 	}
57157c478bd9Sstevel@tonic-gate }
57167c478bd9Sstevel@tonic-gate 
57177c478bd9Sstevel@tonic-gate ulong_t mem_waiters 	= 0;
57187c478bd9Sstevel@tonic-gate ulong_t	max_count 	= 20;
57197c478bd9Sstevel@tonic-gate #define	MAX_DELAY	0x1ff
57207c478bd9Sstevel@tonic-gate 
57217c478bd9Sstevel@tonic-gate /*
57227c478bd9Sstevel@tonic-gate  * Check if enough memory is available to proceed.
57237c478bd9Sstevel@tonic-gate  * Depending on system configuration and how much memory is
57247c478bd9Sstevel@tonic-gate  * reserved for swap we need to check against two variables.
57257c478bd9Sstevel@tonic-gate  * e.g. on systems with little physical swap availrmem can be
57267c478bd9Sstevel@tonic-gate  * more reliable indicator of how much memory is available.
57277c478bd9Sstevel@tonic-gate  * On systems with large phys swap freemem can be better indicator.
57287c478bd9Sstevel@tonic-gate  * If freemem drops below threshold level don't return an error
57297c478bd9Sstevel@tonic-gate  * immediately but wake up pageout to free memory and block.
57307c478bd9Sstevel@tonic-gate  * This is done number of times. If pageout is not able to free
57317c478bd9Sstevel@tonic-gate  * memory within certain time return an error.
57327c478bd9Sstevel@tonic-gate  * The same applies for availrmem but kmem_reap is used to
57337c478bd9Sstevel@tonic-gate  * free memory.
57347c478bd9Sstevel@tonic-gate  */
57357c478bd9Sstevel@tonic-gate int
57367c478bd9Sstevel@tonic-gate page_mem_avail(pgcnt_t npages)
57377c478bd9Sstevel@tonic-gate {
57387c478bd9Sstevel@tonic-gate 	ulong_t count;
57397c478bd9Sstevel@tonic-gate 
57407c478bd9Sstevel@tonic-gate #if defined(__i386)
57417c478bd9Sstevel@tonic-gate 	if (freemem > desfree + npages &&
57427c478bd9Sstevel@tonic-gate 	    availrmem > swapfs_reserve + npages &&
57437c478bd9Sstevel@tonic-gate 	    btop(vmem_size(heap_arena, VMEM_FREE)) > tune.t_minarmem +
57447c478bd9Sstevel@tonic-gate 	    npages)
57457c478bd9Sstevel@tonic-gate 		return (1);
57467c478bd9Sstevel@tonic-gate #else
57477c478bd9Sstevel@tonic-gate 	if (freemem > desfree + npages &&
57487c478bd9Sstevel@tonic-gate 	    availrmem > swapfs_reserve + npages)
57497c478bd9Sstevel@tonic-gate 		return (1);
57507c478bd9Sstevel@tonic-gate #endif
57517c478bd9Sstevel@tonic-gate 
57527c478bd9Sstevel@tonic-gate 	count = max_count;
57537c478bd9Sstevel@tonic-gate 	atomic_add_long(&mem_waiters, 1);
57547c478bd9Sstevel@tonic-gate 
57557c478bd9Sstevel@tonic-gate 	while (freemem < desfree + npages && --count) {
57567c478bd9Sstevel@tonic-gate 		cv_signal(&proc_pageout->p_cv);
57577c478bd9Sstevel@tonic-gate 		if (delay_sig(hz + (mem_waiters & MAX_DELAY))) {
57587c478bd9Sstevel@tonic-gate 			atomic_add_long(&mem_waiters, -1);
57597c478bd9Sstevel@tonic-gate 			return (0);
57607c478bd9Sstevel@tonic-gate 		}
57617c478bd9Sstevel@tonic-gate 	}
57627c478bd9Sstevel@tonic-gate 	if (count == 0) {
57637c478bd9Sstevel@tonic-gate 		atomic_add_long(&mem_waiters, -1);
57647c478bd9Sstevel@tonic-gate 		return (0);
57657c478bd9Sstevel@tonic-gate 	}
57667c478bd9Sstevel@tonic-gate 
57677c478bd9Sstevel@tonic-gate 	count = max_count;
57687c478bd9Sstevel@tonic-gate 	while (availrmem < swapfs_reserve + npages && --count) {
57697c478bd9Sstevel@tonic-gate 		kmem_reap();
57707c478bd9Sstevel@tonic-gate 		if (delay_sig(hz + (mem_waiters & MAX_DELAY))) {
57717c478bd9Sstevel@tonic-gate 			atomic_add_long(&mem_waiters, -1);
57727c478bd9Sstevel@tonic-gate 			return (0);
57737c478bd9Sstevel@tonic-gate 		}
57747c478bd9Sstevel@tonic-gate 	}
57757c478bd9Sstevel@tonic-gate 	atomic_add_long(&mem_waiters, -1);
57767c478bd9Sstevel@tonic-gate 	if (count == 0)
57777c478bd9Sstevel@tonic-gate 		return (0);
57787c478bd9Sstevel@tonic-gate 
57797c478bd9Sstevel@tonic-gate #if defined(__i386)
57807c478bd9Sstevel@tonic-gate 	if (btop(vmem_size(heap_arena, VMEM_FREE)) <
57817c478bd9Sstevel@tonic-gate 	    tune.t_minarmem + npages)
57827c478bd9Sstevel@tonic-gate 		return (0);
57837c478bd9Sstevel@tonic-gate #endif
57847c478bd9Sstevel@tonic-gate 	return (1);
57857c478bd9Sstevel@tonic-gate }
57867c478bd9Sstevel@tonic-gate 
57873cff2f43Sstans #define	MAX_CNT	60	/* max num of iterations */
57883cff2f43Sstans /*
57893cff2f43Sstans  * Reclaim/reserve availrmem for npages.
57903cff2f43Sstans  * If there is not enough memory start reaping seg, kmem caches.
57913cff2f43Sstans  * Start pageout scanner (via page_needfree()).
57923cff2f43Sstans  * Exit after ~ MAX_CNT s regardless of how much memory has been released.
57933cff2f43Sstans  * Note: There is no guarantee that any availrmem will be freed as
57943cff2f43Sstans  * this memory typically is locked (kernel heap) or reserved for swap.
57953cff2f43Sstans  * Also due to memory fragmentation kmem allocator may not be able
57963cff2f43Sstans  * to free any memory (single user allocated buffer will prevent
57973cff2f43Sstans  * freeing slab or a page).
57983cff2f43Sstans  */
57993cff2f43Sstans int
58003cff2f43Sstans page_reclaim_mem(pgcnt_t npages, pgcnt_t epages, int adjust)
58013cff2f43Sstans {
58023cff2f43Sstans 	int	i = 0;
58033cff2f43Sstans 	int	ret = 0;
58043cff2f43Sstans 	pgcnt_t	deficit;
58053cff2f43Sstans 	pgcnt_t old_availrmem;
58063cff2f43Sstans 
58073cff2f43Sstans 	mutex_enter(&freemem_lock);
58083cff2f43Sstans 	old_availrmem = availrmem - 1;
58093cff2f43Sstans 	while ((availrmem < tune.t_minarmem + npages + epages) &&
58103cff2f43Sstans 	    (old_availrmem < availrmem) && (i++ < MAX_CNT)) {
58113cff2f43Sstans 		old_availrmem = availrmem;
58123cff2f43Sstans 		deficit = tune.t_minarmem + npages + epages - availrmem;
58133cff2f43Sstans 		mutex_exit(&freemem_lock);
58143cff2f43Sstans 		page_needfree(deficit);
58153cff2f43Sstans 		seg_preap();
58163cff2f43Sstans 		kmem_reap();
58173cff2f43Sstans 		delay(hz);
58183cff2f43Sstans 		page_needfree(-(spgcnt_t)deficit);
58193cff2f43Sstans 		mutex_enter(&freemem_lock);
58203cff2f43Sstans 	}
58213cff2f43Sstans 
58223cff2f43Sstans 	if (adjust && (availrmem >= tune.t_minarmem + npages + epages)) {
58233cff2f43Sstans 		availrmem -= npages;
58243cff2f43Sstans 		ret = 1;
58253cff2f43Sstans 	}
58263cff2f43Sstans 
58273cff2f43Sstans 	mutex_exit(&freemem_lock);
58283cff2f43Sstans 
58293cff2f43Sstans 	return (ret);
58303cff2f43Sstans }
58317c478bd9Sstevel@tonic-gate 
58327c478bd9Sstevel@tonic-gate /*
58337c478bd9Sstevel@tonic-gate  * Search the memory segments to locate the desired page.  Within a
58347c478bd9Sstevel@tonic-gate  * segment, pages increase linearly with one page structure per
58357c478bd9Sstevel@tonic-gate  * physical page frame (size PAGESIZE).  The search begins
58367c478bd9Sstevel@tonic-gate  * with the segment that was accessed last, to take advantage of locality.
58377c478bd9Sstevel@tonic-gate  * If the hint misses, we start from the beginning of the sorted memseg list
58387c478bd9Sstevel@tonic-gate  */
58397c478bd9Sstevel@tonic-gate 
58407c478bd9Sstevel@tonic-gate 
58417c478bd9Sstevel@tonic-gate /*
58427c478bd9Sstevel@tonic-gate  * Some data structures for pfn to pp lookup.
58437c478bd9Sstevel@tonic-gate  */
58447c478bd9Sstevel@tonic-gate ulong_t mhash_per_slot;
58457c478bd9Sstevel@tonic-gate struct memseg *memseg_hash[N_MEM_SLOTS];
58467c478bd9Sstevel@tonic-gate 
58477c478bd9Sstevel@tonic-gate page_t *
58487c478bd9Sstevel@tonic-gate page_numtopp_nolock(pfn_t pfnum)
58497c478bd9Sstevel@tonic-gate {
58507c478bd9Sstevel@tonic-gate 	struct memseg *seg;
58517c478bd9Sstevel@tonic-gate 	page_t *pp;
5852affbd3ccSkchow 	vm_cpu_data_t *vc = CPU->cpu_vm_data;
58537c478bd9Sstevel@tonic-gate 
5854affbd3ccSkchow 	ASSERT(vc != NULL);
58557c478bd9Sstevel@tonic-gate 
58567c478bd9Sstevel@tonic-gate 	MEMSEG_STAT_INCR(nsearch);
58577c478bd9Sstevel@tonic-gate 
58587c478bd9Sstevel@tonic-gate 	/* Try last winner first */
5859affbd3ccSkchow 	if (((seg = vc->vc_pnum_memseg) != NULL) &&
58607c478bd9Sstevel@tonic-gate 		(pfnum >= seg->pages_base) && (pfnum < seg->pages_end)) {
58617c478bd9Sstevel@tonic-gate 		MEMSEG_STAT_INCR(nlastwon);
58627c478bd9Sstevel@tonic-gate 		pp = seg->pages + (pfnum - seg->pages_base);
58637c478bd9Sstevel@tonic-gate 		if (pp->p_pagenum == pfnum)
58647c478bd9Sstevel@tonic-gate 			return ((page_t *)pp);
58657c478bd9Sstevel@tonic-gate 	}
58667c478bd9Sstevel@tonic-gate 
58677c478bd9Sstevel@tonic-gate 	/* Else Try hash */
58687c478bd9Sstevel@tonic-gate 	if (((seg = memseg_hash[MEMSEG_PFN_HASH(pfnum)]) != NULL) &&
58697c478bd9Sstevel@tonic-gate 		(pfnum >= seg->pages_base) && (pfnum < seg->pages_end)) {
58707c478bd9Sstevel@tonic-gate 		MEMSEG_STAT_INCR(nhashwon);
5871affbd3ccSkchow 		vc->vc_pnum_memseg = seg;
58727c478bd9Sstevel@tonic-gate 		pp = seg->pages + (pfnum - seg->pages_base);
58737c478bd9Sstevel@tonic-gate 		if (pp->p_pagenum == pfnum)
58747c478bd9Sstevel@tonic-gate 			return ((page_t *)pp);
58757c478bd9Sstevel@tonic-gate 	}
58767c478bd9Sstevel@tonic-gate 
58777c478bd9Sstevel@tonic-gate 	/* Else Brute force */
58787c478bd9Sstevel@tonic-gate 	for (seg = memsegs; seg != NULL; seg = seg->next) {
58797c478bd9Sstevel@tonic-gate 		if (pfnum >= seg->pages_base && pfnum < seg->pages_end) {
5880affbd3ccSkchow 			vc->vc_pnum_memseg = seg;
58817c478bd9Sstevel@tonic-gate 			pp = seg->pages + (pfnum - seg->pages_base);
58827c478bd9Sstevel@tonic-gate 			return ((page_t *)pp);
58837c478bd9Sstevel@tonic-gate 		}
58847c478bd9Sstevel@tonic-gate 	}
5885affbd3ccSkchow 	vc->vc_pnum_memseg = NULL;
58867c478bd9Sstevel@tonic-gate 	MEMSEG_STAT_INCR(nnotfound);
58877c478bd9Sstevel@tonic-gate 	return ((page_t *)NULL);
58887c478bd9Sstevel@tonic-gate 
58897c478bd9Sstevel@tonic-gate }
58907c478bd9Sstevel@tonic-gate 
58917c478bd9Sstevel@tonic-gate struct memseg *
58927c478bd9Sstevel@tonic-gate page_numtomemseg_nolock(pfn_t pfnum)
58937c478bd9Sstevel@tonic-gate {
58947c478bd9Sstevel@tonic-gate 	struct memseg *seg;
58957c478bd9Sstevel@tonic-gate 	page_t *pp;
58967c478bd9Sstevel@tonic-gate 
58977c478bd9Sstevel@tonic-gate 	/* Try hash */
58987c478bd9Sstevel@tonic-gate 	if (((seg = memseg_hash[MEMSEG_PFN_HASH(pfnum)]) != NULL) &&
58997c478bd9Sstevel@tonic-gate 		(pfnum >= seg->pages_base) && (pfnum < seg->pages_end)) {
59007c478bd9Sstevel@tonic-gate 		pp = seg->pages + (pfnum - seg->pages_base);
59017c478bd9Sstevel@tonic-gate 		if (pp->p_pagenum == pfnum)
59027c478bd9Sstevel@tonic-gate 			return (seg);
59037c478bd9Sstevel@tonic-gate 	}
59047c478bd9Sstevel@tonic-gate 
59057c478bd9Sstevel@tonic-gate 	/* Else Brute force */
59067c478bd9Sstevel@tonic-gate 	for (seg = memsegs; seg != NULL; seg = seg->next) {
59077c478bd9Sstevel@tonic-gate 		if (pfnum >= seg->pages_base && pfnum < seg->pages_end) {
59087c478bd9Sstevel@tonic-gate 			return (seg);
59097c478bd9Sstevel@tonic-gate 		}
59107c478bd9Sstevel@tonic-gate 	}
59117c478bd9Sstevel@tonic-gate 	return ((struct memseg *)NULL);
59127c478bd9Sstevel@tonic-gate }
59137c478bd9Sstevel@tonic-gate 
59147c478bd9Sstevel@tonic-gate /*
59157c478bd9Sstevel@tonic-gate  * Given a page and a count return the page struct that is
59167c478bd9Sstevel@tonic-gate  * n structs away from the current one in the global page
59177c478bd9Sstevel@tonic-gate  * list.
59187c478bd9Sstevel@tonic-gate  *
59197c478bd9Sstevel@tonic-gate  * This function wraps to the first page upon
59207c478bd9Sstevel@tonic-gate  * reaching the end of the memseg list.
59217c478bd9Sstevel@tonic-gate  */
59227c478bd9Sstevel@tonic-gate page_t *
59237c478bd9Sstevel@tonic-gate page_nextn(page_t *pp, ulong_t n)
59247c478bd9Sstevel@tonic-gate {
59257c478bd9Sstevel@tonic-gate 	struct memseg *seg;
59267c478bd9Sstevel@tonic-gate 	page_t *ppn;
5927affbd3ccSkchow 	vm_cpu_data_t *vc = (vm_cpu_data_t *)CPU->cpu_vm_data;
59287c478bd9Sstevel@tonic-gate 
5929affbd3ccSkchow 	ASSERT(vc != NULL);
5930affbd3ccSkchow 
5931affbd3ccSkchow 	if (((seg = vc->vc_pnext_memseg) == NULL) ||
59327c478bd9Sstevel@tonic-gate 	    (seg->pages_base == seg->pages_end) ||
59337c478bd9Sstevel@tonic-gate 	    !(pp >= seg->pages && pp < seg->epages)) {
59347c478bd9Sstevel@tonic-gate 
59357c478bd9Sstevel@tonic-gate 		for (seg = memsegs; seg; seg = seg->next) {
59367c478bd9Sstevel@tonic-gate 			if (pp >= seg->pages && pp < seg->epages)
59377c478bd9Sstevel@tonic-gate 				break;
59387c478bd9Sstevel@tonic-gate 		}
59397c478bd9Sstevel@tonic-gate 
59407c478bd9Sstevel@tonic-gate 		if (seg == NULL) {
59417c478bd9Sstevel@tonic-gate 			/* Memory delete got in, return something valid. */
59427c478bd9Sstevel@tonic-gate 			/* TODO: fix me. */
59437c478bd9Sstevel@tonic-gate 			seg = memsegs;
59447c478bd9Sstevel@tonic-gate 			pp = seg->pages;
59457c478bd9Sstevel@tonic-gate 		}
59467c478bd9Sstevel@tonic-gate 	}
59477c478bd9Sstevel@tonic-gate 
59487c478bd9Sstevel@tonic-gate 	/* check for wraparound - possible if n is large */
59497c478bd9Sstevel@tonic-gate 	while ((ppn = (pp + n)) >= seg->epages || ppn < pp) {
59507c478bd9Sstevel@tonic-gate 		n -= seg->epages - pp;
59517c478bd9Sstevel@tonic-gate 		seg = seg->next;
59527c478bd9Sstevel@tonic-gate 		if (seg == NULL)
59537c478bd9Sstevel@tonic-gate 			seg = memsegs;
59547c478bd9Sstevel@tonic-gate 		pp = seg->pages;
59557c478bd9Sstevel@tonic-gate 	}
5956affbd3ccSkchow 	vc->vc_pnext_memseg = seg;
59577c478bd9Sstevel@tonic-gate 	return (ppn);
59587c478bd9Sstevel@tonic-gate }
59597c478bd9Sstevel@tonic-gate 
59607c478bd9Sstevel@tonic-gate /*
59617c478bd9Sstevel@tonic-gate  * Initialize for a loop using page_next_scan_large().
59627c478bd9Sstevel@tonic-gate  */
59637c478bd9Sstevel@tonic-gate page_t *
59647c478bd9Sstevel@tonic-gate page_next_scan_init(void **cookie)
59657c478bd9Sstevel@tonic-gate {
59667c478bd9Sstevel@tonic-gate 	ASSERT(cookie != NULL);
59677c478bd9Sstevel@tonic-gate 	*cookie = (void *)memsegs;
59687c478bd9Sstevel@tonic-gate 	return ((page_t *)memsegs->pages);
59697c478bd9Sstevel@tonic-gate }
59707c478bd9Sstevel@tonic-gate 
59717c478bd9Sstevel@tonic-gate /*
59727c478bd9Sstevel@tonic-gate  * Return the next page in a scan of page_t's, assuming we want
59737c478bd9Sstevel@tonic-gate  * to skip over sub-pages within larger page sizes.
59747c478bd9Sstevel@tonic-gate  *
59757c478bd9Sstevel@tonic-gate  * The cookie is used to keep track of the current memseg.
59767c478bd9Sstevel@tonic-gate  */
59777c478bd9Sstevel@tonic-gate page_t *
59787c478bd9Sstevel@tonic-gate page_next_scan_large(
59797c478bd9Sstevel@tonic-gate 	page_t		*pp,
59807c478bd9Sstevel@tonic-gate 	ulong_t		*n,
59817c478bd9Sstevel@tonic-gate 	void		**cookie)
59827c478bd9Sstevel@tonic-gate {
59837c478bd9Sstevel@tonic-gate 	struct memseg	*seg = (struct memseg *)*cookie;
59847c478bd9Sstevel@tonic-gate 	page_t		*new_pp;
59857c478bd9Sstevel@tonic-gate 	ulong_t		cnt;
59867c478bd9Sstevel@tonic-gate 	pfn_t		pfn;
59877c478bd9Sstevel@tonic-gate 
59887c478bd9Sstevel@tonic-gate 
59897c478bd9Sstevel@tonic-gate 	/*
59907c478bd9Sstevel@tonic-gate 	 * get the count of page_t's to skip based on the page size
59917c478bd9Sstevel@tonic-gate 	 */
59927c478bd9Sstevel@tonic-gate 	ASSERT(pp != NULL);
59937c478bd9Sstevel@tonic-gate 	if (pp->p_szc == 0) {
59947c478bd9Sstevel@tonic-gate 		cnt = 1;
59957c478bd9Sstevel@tonic-gate 	} else {
59967c478bd9Sstevel@tonic-gate 		pfn = page_pptonum(pp);
59977c478bd9Sstevel@tonic-gate 		cnt = page_get_pagecnt(pp->p_szc);
59987c478bd9Sstevel@tonic-gate 		cnt -= pfn & (cnt - 1);
59997c478bd9Sstevel@tonic-gate 	}
60007c478bd9Sstevel@tonic-gate 	*n += cnt;
60017c478bd9Sstevel@tonic-gate 	new_pp = pp + cnt;
60027c478bd9Sstevel@tonic-gate 
60037c478bd9Sstevel@tonic-gate 	/*
60047c478bd9Sstevel@tonic-gate 	 * Catch if we went past the end of the current memory segment. If so,
60057c478bd9Sstevel@tonic-gate 	 * just move to the next segment with pages.
60067c478bd9Sstevel@tonic-gate 	 */
60077c478bd9Sstevel@tonic-gate 	if (new_pp >= seg->epages) {
60087c478bd9Sstevel@tonic-gate 		do {
60097c478bd9Sstevel@tonic-gate 			seg = seg->next;
60107c478bd9Sstevel@tonic-gate 			if (seg == NULL)
60117c478bd9Sstevel@tonic-gate 				seg = memsegs;
60127c478bd9Sstevel@tonic-gate 		} while (seg->pages == seg->epages);
60137c478bd9Sstevel@tonic-gate 		new_pp = seg->pages;
60147c478bd9Sstevel@tonic-gate 		*cookie = (void *)seg;
60157c478bd9Sstevel@tonic-gate 	}
60167c478bd9Sstevel@tonic-gate 
60177c478bd9Sstevel@tonic-gate 	return (new_pp);
60187c478bd9Sstevel@tonic-gate }
60197c478bd9Sstevel@tonic-gate 
60207c478bd9Sstevel@tonic-gate 
60217c478bd9Sstevel@tonic-gate /*
60227c478bd9Sstevel@tonic-gate  * Returns next page in list. Note: this function wraps
60237c478bd9Sstevel@tonic-gate  * to the first page in the list upon reaching the end
60247c478bd9Sstevel@tonic-gate  * of the list. Callers should be aware of this fact.
60257c478bd9Sstevel@tonic-gate  */
60267c478bd9Sstevel@tonic-gate 
60277c478bd9Sstevel@tonic-gate /* We should change this be a #define */
60287c478bd9Sstevel@tonic-gate 
60297c478bd9Sstevel@tonic-gate page_t *
60307c478bd9Sstevel@tonic-gate page_next(page_t *pp)
60317c478bd9Sstevel@tonic-gate {
60327c478bd9Sstevel@tonic-gate 	return (page_nextn(pp, 1));
60337c478bd9Sstevel@tonic-gate }
60347c478bd9Sstevel@tonic-gate 
60357c478bd9Sstevel@tonic-gate page_t *
60367c478bd9Sstevel@tonic-gate page_first()
60377c478bd9Sstevel@tonic-gate {
60387c478bd9Sstevel@tonic-gate 	return ((page_t *)memsegs->pages);
60397c478bd9Sstevel@tonic-gate }
60407c478bd9Sstevel@tonic-gate 
60417c478bd9Sstevel@tonic-gate 
60427c478bd9Sstevel@tonic-gate /*
60437c478bd9Sstevel@tonic-gate  * This routine is called at boot with the initial memory configuration
60447c478bd9Sstevel@tonic-gate  * and when memory is added or removed.
60457c478bd9Sstevel@tonic-gate  */
60467c478bd9Sstevel@tonic-gate void
60477c478bd9Sstevel@tonic-gate build_pfn_hash()
60487c478bd9Sstevel@tonic-gate {
60497c478bd9Sstevel@tonic-gate 	pfn_t cur;
60507c478bd9Sstevel@tonic-gate 	pgcnt_t index;
60517c478bd9Sstevel@tonic-gate 	struct memseg *pseg;
60527c478bd9Sstevel@tonic-gate 	int	i;
60537c478bd9Sstevel@tonic-gate 
60547c478bd9Sstevel@tonic-gate 	/*
60557c478bd9Sstevel@tonic-gate 	 * Clear memseg_hash array.
60567c478bd9Sstevel@tonic-gate 	 * Since memory add/delete is designed to operate concurrently
60577c478bd9Sstevel@tonic-gate 	 * with normal operation, the hash rebuild must be able to run
60587c478bd9Sstevel@tonic-gate 	 * concurrently with page_numtopp_nolock(). To support this
60597c478bd9Sstevel@tonic-gate 	 * functionality, assignments to memseg_hash array members must
60607c478bd9Sstevel@tonic-gate 	 * be done atomically.
60617c478bd9Sstevel@tonic-gate 	 *
60627c478bd9Sstevel@tonic-gate 	 * NOTE: bzero() does not currently guarantee this for kernel
60637c478bd9Sstevel@tonic-gate 	 * threads, and cannot be used here.
60647c478bd9Sstevel@tonic-gate 	 */
60657c478bd9Sstevel@tonic-gate 	for (i = 0; i < N_MEM_SLOTS; i++)
60667c478bd9Sstevel@tonic-gate 		memseg_hash[i] = NULL;
60677c478bd9Sstevel@tonic-gate 
60687c478bd9Sstevel@tonic-gate 	hat_kpm_mseghash_clear(N_MEM_SLOTS);
60697c478bd9Sstevel@tonic-gate 
60707c478bd9Sstevel@tonic-gate 	/*
60717c478bd9Sstevel@tonic-gate 	 * Physmax is the last valid pfn.
60727c478bd9Sstevel@tonic-gate 	 */
60737c478bd9Sstevel@tonic-gate 	mhash_per_slot = (physmax + 1) >> MEM_HASH_SHIFT;
60747c478bd9Sstevel@tonic-gate 	for (pseg = memsegs; pseg != NULL; pseg = pseg->next) {
60757c478bd9Sstevel@tonic-gate 		index = MEMSEG_PFN_HASH(pseg->pages_base);
60767c478bd9Sstevel@tonic-gate 		cur = pseg->pages_base;
60777c478bd9Sstevel@tonic-gate 		do {
60787c478bd9Sstevel@tonic-gate 			if (index >= N_MEM_SLOTS)
60797c478bd9Sstevel@tonic-gate 				index = MEMSEG_PFN_HASH(cur);
60807c478bd9Sstevel@tonic-gate 
60817c478bd9Sstevel@tonic-gate 			if (memseg_hash[index] == NULL ||
60827c478bd9Sstevel@tonic-gate 			    memseg_hash[index]->pages_base > pseg->pages_base) {
60837c478bd9Sstevel@tonic-gate 				memseg_hash[index] = pseg;
60847c478bd9Sstevel@tonic-gate 				hat_kpm_mseghash_update(index, pseg);
60857c478bd9Sstevel@tonic-gate 			}
60867c478bd9Sstevel@tonic-gate 			cur += mhash_per_slot;
60877c478bd9Sstevel@tonic-gate 			index++;
60887c478bd9Sstevel@tonic-gate 		} while (cur < pseg->pages_end);
60897c478bd9Sstevel@tonic-gate 	}
60907c478bd9Sstevel@tonic-gate }
60917c478bd9Sstevel@tonic-gate 
60927c478bd9Sstevel@tonic-gate /*
60937c478bd9Sstevel@tonic-gate  * Return the pagenum for the pp
60947c478bd9Sstevel@tonic-gate  */
60957c478bd9Sstevel@tonic-gate pfn_t
60967c478bd9Sstevel@tonic-gate page_pptonum(page_t *pp)
60977c478bd9Sstevel@tonic-gate {
60987c478bd9Sstevel@tonic-gate 	return (pp->p_pagenum);
60997c478bd9Sstevel@tonic-gate }
61007c478bd9Sstevel@tonic-gate 
61017c478bd9Sstevel@tonic-gate /*
61027c478bd9Sstevel@tonic-gate  * interface to the referenced and modified etc bits
61037c478bd9Sstevel@tonic-gate  * in the PSM part of the page struct
61047c478bd9Sstevel@tonic-gate  * when no locking is desired.
61057c478bd9Sstevel@tonic-gate  */
61067c478bd9Sstevel@tonic-gate void
61077c478bd9Sstevel@tonic-gate page_set_props(page_t *pp, uint_t flags)
61087c478bd9Sstevel@tonic-gate {
61097c478bd9Sstevel@tonic-gate 	ASSERT((flags & ~(P_MOD | P_REF | P_RO)) == 0);
61107c478bd9Sstevel@tonic-gate 	pp->p_nrm |= (uchar_t)flags;
61117c478bd9Sstevel@tonic-gate }
61127c478bd9Sstevel@tonic-gate 
61137c478bd9Sstevel@tonic-gate void
61147c478bd9Sstevel@tonic-gate page_clr_all_props(page_t *pp)
61157c478bd9Sstevel@tonic-gate {
61167c478bd9Sstevel@tonic-gate 	pp->p_nrm = 0;
61177c478bd9Sstevel@tonic-gate }
61187c478bd9Sstevel@tonic-gate 
61197c478bd9Sstevel@tonic-gate /*
6120db874c57Selowe  * Clear p_lckcnt and p_cowcnt, adjusting freemem if required.
61217c478bd9Sstevel@tonic-gate  */
61227c478bd9Sstevel@tonic-gate int
6123db874c57Selowe page_clear_lck_cow(page_t *pp, int adjust)
61247c478bd9Sstevel@tonic-gate {
6125db874c57Selowe 	int	f_amount;
61267c478bd9Sstevel@tonic-gate 
6127db874c57Selowe 	ASSERT(PAGE_EXCL(pp));
61287c478bd9Sstevel@tonic-gate 
61297c478bd9Sstevel@tonic-gate 	/*
6130db874c57Selowe 	 * The page_struct_lock need not be acquired here since
6131db874c57Selowe 	 * we require the caller hold the page exclusively locked.
61327c478bd9Sstevel@tonic-gate 	 */
6133db874c57Selowe 	f_amount = 0;
6134db874c57Selowe 	if (pp->p_lckcnt) {
6135db874c57Selowe 		f_amount = 1;
6136db874c57Selowe 		pp->p_lckcnt = 0;
6137db874c57Selowe 	}
6138db874c57Selowe 	if (pp->p_cowcnt) {
6139db874c57Selowe 		f_amount += pp->p_cowcnt;
6140db874c57Selowe 		pp->p_cowcnt = 0;
6141db874c57Selowe 	}
61427c478bd9Sstevel@tonic-gate 
6143db874c57Selowe 	if (adjust && f_amount) {
6144db874c57Selowe 		mutex_enter(&freemem_lock);
6145db874c57Selowe 		availrmem += f_amount;
6146db874c57Selowe 		mutex_exit(&freemem_lock);
61477c478bd9Sstevel@tonic-gate 	}
61487c478bd9Sstevel@tonic-gate 
6149db874c57Selowe 	return (f_amount);
61507c478bd9Sstevel@tonic-gate }
61517c478bd9Sstevel@tonic-gate 
61527c478bd9Sstevel@tonic-gate /*
6153db874c57Selowe  * The following functions is called from free_vp_pages()
6154db874c57Selowe  * for an inexact estimate of a newly free'd page...
61557c478bd9Sstevel@tonic-gate  */
6156db874c57Selowe ulong_t
6157db874c57Selowe page_share_cnt(page_t *pp)
6158db874c57Selowe {
6159db874c57Selowe 	return (hat_page_getshare(pp));
6160db874c57Selowe }
6161db874c57Selowe 
61627c478bd9Sstevel@tonic-gate int
61637c478bd9Sstevel@tonic-gate page_isshared(page_t *pp)
61647c478bd9Sstevel@tonic-gate {
61657c478bd9Sstevel@tonic-gate 	return (hat_page_getshare(pp) > 1);
61667c478bd9Sstevel@tonic-gate }
61677c478bd9Sstevel@tonic-gate 
61687c478bd9Sstevel@tonic-gate int
61697c478bd9Sstevel@tonic-gate page_isfree(page_t *pp)
61707c478bd9Sstevel@tonic-gate {
61717c478bd9Sstevel@tonic-gate 	return (PP_ISFREE(pp));
61727c478bd9Sstevel@tonic-gate }
61737c478bd9Sstevel@tonic-gate 
61747c478bd9Sstevel@tonic-gate int
61757c478bd9Sstevel@tonic-gate page_isref(page_t *pp)
61767c478bd9Sstevel@tonic-gate {
61777c478bd9Sstevel@tonic-gate 	return (hat_page_getattr(pp, P_REF));
61787c478bd9Sstevel@tonic-gate }
61797c478bd9Sstevel@tonic-gate 
61807c478bd9Sstevel@tonic-gate int
61817c478bd9Sstevel@tonic-gate page_ismod(page_t *pp)
61827c478bd9Sstevel@tonic-gate {
61837c478bd9Sstevel@tonic-gate 	return (hat_page_getattr(pp, P_MOD));
61847c478bd9Sstevel@tonic-gate }
6185