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 /* 226bc16138Sjj * Copyright 2008 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> 800209230bSgjelinek #include <sys/vm_usage.h> 817c478bd9Sstevel@tonic-gate #include <fs/fs_subr.h> 82cee1d74bSjfrank #include <sys/ddi.h> 83cee1d74bSjfrank #include <sys/modctl.h> 847c478bd9Sstevel@tonic-gate 857c478bd9Sstevel@tonic-gate static int nopageage = 0; 867c478bd9Sstevel@tonic-gate 877c478bd9Sstevel@tonic-gate static pgcnt_t max_page_get; /* max page_get request size in pages */ 887c478bd9Sstevel@tonic-gate pgcnt_t total_pages = 0; /* total number of pages (used by /proc) */ 897c478bd9Sstevel@tonic-gate 907c478bd9Sstevel@tonic-gate /* 917c478bd9Sstevel@tonic-gate * freemem_lock protects all freemem variables: 927c478bd9Sstevel@tonic-gate * availrmem. Also this lock protects the globals which track the 937c478bd9Sstevel@tonic-gate * availrmem changes for accurate kernel footprint calculation. 947c478bd9Sstevel@tonic-gate * See below for an explanation of these 957c478bd9Sstevel@tonic-gate * globals. 967c478bd9Sstevel@tonic-gate */ 977c478bd9Sstevel@tonic-gate kmutex_t freemem_lock; 987c478bd9Sstevel@tonic-gate pgcnt_t availrmem; 997c478bd9Sstevel@tonic-gate pgcnt_t availrmem_initial; 1007c478bd9Sstevel@tonic-gate 1017c478bd9Sstevel@tonic-gate /* 1027c478bd9Sstevel@tonic-gate * These globals track availrmem changes to get a more accurate 1037c478bd9Sstevel@tonic-gate * estimate of tke kernel size. Historically pp_kernel is used for 1047c478bd9Sstevel@tonic-gate * kernel size and is based on availrmem. But availrmem is adjusted for 1057c478bd9Sstevel@tonic-gate * locked pages in the system not just for kernel locked pages. 1067c478bd9Sstevel@tonic-gate * These new counters will track the pages locked through segvn and 1077c478bd9Sstevel@tonic-gate * by explicit user locking. 1087c478bd9Sstevel@tonic-gate * 109da6c28aaSamw * pages_locked : How many pages are locked because of user specified 1107c478bd9Sstevel@tonic-gate * locking through mlock or plock. 1117c478bd9Sstevel@tonic-gate * 1127c478bd9Sstevel@tonic-gate * pages_useclaim,pages_claimed : These two variables track the 113da6c28aaSamw * claim adjustments because of the protection changes on a segvn segment. 1147c478bd9Sstevel@tonic-gate * 1157c478bd9Sstevel@tonic-gate * All these globals are protected by the same lock which protects availrmem. 1167c478bd9Sstevel@tonic-gate */ 117a98e9dbfSaguzovsk pgcnt_t pages_locked = 0; 118a98e9dbfSaguzovsk pgcnt_t pages_useclaim = 0; 119a98e9dbfSaguzovsk pgcnt_t pages_claimed = 0; 1207c478bd9Sstevel@tonic-gate 1217c478bd9Sstevel@tonic-gate 1227c478bd9Sstevel@tonic-gate /* 1237c478bd9Sstevel@tonic-gate * new_freemem_lock protects freemem, freemem_wait & freemem_cv. 1247c478bd9Sstevel@tonic-gate */ 1257c478bd9Sstevel@tonic-gate static kmutex_t new_freemem_lock; 1267c478bd9Sstevel@tonic-gate static uint_t freemem_wait; /* someone waiting for freemem */ 1277c478bd9Sstevel@tonic-gate static kcondvar_t freemem_cv; 1287c478bd9Sstevel@tonic-gate 1297c478bd9Sstevel@tonic-gate /* 1307c478bd9Sstevel@tonic-gate * The logical page free list is maintained as two lists, the 'free' 1317c478bd9Sstevel@tonic-gate * and the 'cache' lists. 1327c478bd9Sstevel@tonic-gate * The free list contains those pages that should be reused first. 1337c478bd9Sstevel@tonic-gate * 1347c478bd9Sstevel@tonic-gate * The implementation of the lists is machine dependent. 1357c478bd9Sstevel@tonic-gate * page_get_freelist(), page_get_cachelist(), 1367c478bd9Sstevel@tonic-gate * page_list_sub(), and page_list_add() 1377c478bd9Sstevel@tonic-gate * form the interface to the machine dependent implementation. 1387c478bd9Sstevel@tonic-gate * 1397c478bd9Sstevel@tonic-gate * Pages with p_free set are on the cache list. 1407c478bd9Sstevel@tonic-gate * Pages with p_free and p_age set are on the free list, 1417c478bd9Sstevel@tonic-gate * 1427c478bd9Sstevel@tonic-gate * A page may be locked while on either list. 1437c478bd9Sstevel@tonic-gate */ 1447c478bd9Sstevel@tonic-gate 1457c478bd9Sstevel@tonic-gate /* 1467c478bd9Sstevel@tonic-gate * free list accounting stuff. 1477c478bd9Sstevel@tonic-gate * 1487c478bd9Sstevel@tonic-gate * 1497c478bd9Sstevel@tonic-gate * Spread out the value for the number of pages on the 1507c478bd9Sstevel@tonic-gate * page free and page cache lists. If there is just one 1517c478bd9Sstevel@tonic-gate * value, then it must be under just one lock. 1527c478bd9Sstevel@tonic-gate * The lock contention and cache traffic are a real bother. 1537c478bd9Sstevel@tonic-gate * 1547c478bd9Sstevel@tonic-gate * When we acquire and then drop a single pcf lock 1557c478bd9Sstevel@tonic-gate * we can start in the middle of the array of pcf structures. 1567c478bd9Sstevel@tonic-gate * If we acquire more than one pcf lock at a time, we need to 1577c478bd9Sstevel@tonic-gate * start at the front to avoid deadlocking. 1587c478bd9Sstevel@tonic-gate * 1597c478bd9Sstevel@tonic-gate * pcf_count holds the number of pages in each pool. 1607c478bd9Sstevel@tonic-gate * 1617c478bd9Sstevel@tonic-gate * pcf_block is set when page_create_get_something() has asked the 1627c478bd9Sstevel@tonic-gate * PSM page freelist and page cachelist routines without specifying 1637c478bd9Sstevel@tonic-gate * a color and nothing came back. This is used to block anything 1647c478bd9Sstevel@tonic-gate * else from moving pages from one list to the other while the 1657c478bd9Sstevel@tonic-gate * lists are searched again. If a page is freeed while pcf_block is 1667c478bd9Sstevel@tonic-gate * set, then pcf_reserve is incremented. pcgs_unblock() takes care 1677c478bd9Sstevel@tonic-gate * of clearning pcf_block, doing the wakeups, etc. 1687c478bd9Sstevel@tonic-gate */ 1697c478bd9Sstevel@tonic-gate 170*06fb6a36Sdv #define MAX_PCF_FANOUT NCPU 171*06fb6a36Sdv static uint_t pcf_fanout = 1; /* Will get changed at boot time */ 172*06fb6a36Sdv static uint_t pcf_fanout_mask = 0; 1737c478bd9Sstevel@tonic-gate 1747c478bd9Sstevel@tonic-gate struct pcf { 1757c478bd9Sstevel@tonic-gate kmutex_t pcf_lock; /* protects the structure */ 176f2b37d75Sfr uint_t pcf_count; /* page count */ 1777c478bd9Sstevel@tonic-gate uint_t pcf_wait; /* number of waiters */ 1787c478bd9Sstevel@tonic-gate uint_t pcf_block; /* pcgs flag to page_free() */ 1797c478bd9Sstevel@tonic-gate uint_t pcf_reserve; /* pages freed after pcf_block set */ 180*06fb6a36Sdv uint_t pcf_fill[10]; /* to line up on the caches */ 1817c478bd9Sstevel@tonic-gate }; 1827c478bd9Sstevel@tonic-gate 183*06fb6a36Sdv /* 184*06fb6a36Sdv * PCF_INDEX hash needs to be dynamic (every so often the hash changes where 185*06fb6a36Sdv * it will hash the cpu to). This is done to prevent a drain condition 186*06fb6a36Sdv * from happening. This drain condition will occur when pcf_count decrement 187*06fb6a36Sdv * occurs on cpu A and the increment of pcf_count always occurs on cpu B. An 188*06fb6a36Sdv * example of this shows up with device interrupts. The dma buffer is allocated 189*06fb6a36Sdv * by the cpu requesting the IO thus the pcf_count is decremented based on that. 190*06fb6a36Sdv * When the memory is returned by the interrupt thread, the pcf_count will be 191*06fb6a36Sdv * incremented based on the cpu servicing the interrupt. 192*06fb6a36Sdv */ 193*06fb6a36Sdv static struct pcf pcf[MAX_PCF_FANOUT]; 194*06fb6a36Sdv #define PCF_INDEX() ((int)(((long)CPU->cpu_seqid) + \ 195*06fb6a36Sdv (randtick() >> 24)) & (pcf_fanout_mask)) 196*06fb6a36Sdv 197*06fb6a36Sdv static int pcf_decrement_bucket(pgcnt_t); 198*06fb6a36Sdv static int pcf_decrement_multiple(pgcnt_t *, pgcnt_t, int); 1997c478bd9Sstevel@tonic-gate 2007c478bd9Sstevel@tonic-gate kmutex_t pcgs_lock; /* serializes page_create_get_ */ 2017c478bd9Sstevel@tonic-gate kmutex_t pcgs_cagelock; /* serializes NOSLEEP cage allocs */ 2027c478bd9Sstevel@tonic-gate kmutex_t pcgs_wait_lock; /* used for delay in pcgs */ 2037c478bd9Sstevel@tonic-gate static kcondvar_t pcgs_cv; /* cv for delay in pcgs */ 2047c478bd9Sstevel@tonic-gate 2057c478bd9Sstevel@tonic-gate #ifdef VM_STATS 2067c478bd9Sstevel@tonic-gate 2077c478bd9Sstevel@tonic-gate /* 2087c478bd9Sstevel@tonic-gate * No locks, but so what, they are only statistics. 2097c478bd9Sstevel@tonic-gate */ 2107c478bd9Sstevel@tonic-gate 2117c478bd9Sstevel@tonic-gate static struct page_tcnt { 2127c478bd9Sstevel@tonic-gate int pc_free_cache; /* free's into cache list */ 2137c478bd9Sstevel@tonic-gate int pc_free_dontneed; /* free's with dontneed */ 2147c478bd9Sstevel@tonic-gate int pc_free_pageout; /* free's from pageout */ 2157c478bd9Sstevel@tonic-gate int pc_free_free; /* free's into free list */ 2167c478bd9Sstevel@tonic-gate int pc_free_pages; /* free's into large page free list */ 2177c478bd9Sstevel@tonic-gate int pc_destroy_pages; /* large page destroy's */ 2187c478bd9Sstevel@tonic-gate int pc_get_cache; /* get's from cache list */ 2197c478bd9Sstevel@tonic-gate int pc_get_free; /* get's from free list */ 2207c478bd9Sstevel@tonic-gate int pc_reclaim; /* reclaim's */ 2217c478bd9Sstevel@tonic-gate int pc_abortfree; /* abort's of free pages */ 2227c478bd9Sstevel@tonic-gate int pc_find_hit; /* find's that find page */ 2237c478bd9Sstevel@tonic-gate int pc_find_miss; /* find's that don't find page */ 2247c478bd9Sstevel@tonic-gate int pc_destroy_free; /* # of free pages destroyed */ 2257c478bd9Sstevel@tonic-gate #define PC_HASH_CNT (4*PAGE_HASHAVELEN) 2267c478bd9Sstevel@tonic-gate int pc_find_hashlen[PC_HASH_CNT+1]; 2277c478bd9Sstevel@tonic-gate int pc_addclaim_pages; 2287c478bd9Sstevel@tonic-gate int pc_subclaim_pages; 2297c478bd9Sstevel@tonic-gate int pc_free_replacement_page[2]; 2307c478bd9Sstevel@tonic-gate int pc_try_demote_pages[6]; 2317c478bd9Sstevel@tonic-gate int pc_demote_pages[2]; 2327c478bd9Sstevel@tonic-gate } pagecnt; 2337c478bd9Sstevel@tonic-gate 2347c478bd9Sstevel@tonic-gate uint_t hashin_count; 2357c478bd9Sstevel@tonic-gate uint_t hashin_not_held; 2367c478bd9Sstevel@tonic-gate uint_t hashin_already; 2377c478bd9Sstevel@tonic-gate 2387c478bd9Sstevel@tonic-gate uint_t hashout_count; 2397c478bd9Sstevel@tonic-gate uint_t hashout_not_held; 2407c478bd9Sstevel@tonic-gate 2417c478bd9Sstevel@tonic-gate uint_t page_create_count; 2427c478bd9Sstevel@tonic-gate uint_t page_create_not_enough; 2437c478bd9Sstevel@tonic-gate uint_t page_create_not_enough_again; 2447c478bd9Sstevel@tonic-gate uint_t page_create_zero; 2457c478bd9Sstevel@tonic-gate uint_t page_create_hashout; 2467c478bd9Sstevel@tonic-gate uint_t page_create_page_lock_failed; 2477c478bd9Sstevel@tonic-gate uint_t page_create_trylock_failed; 2487c478bd9Sstevel@tonic-gate uint_t page_create_found_one; 2497c478bd9Sstevel@tonic-gate uint_t page_create_hashin_failed; 2507c478bd9Sstevel@tonic-gate uint_t page_create_dropped_phm; 2517c478bd9Sstevel@tonic-gate 2527c478bd9Sstevel@tonic-gate uint_t page_create_new; 2537c478bd9Sstevel@tonic-gate uint_t page_create_exists; 2547c478bd9Sstevel@tonic-gate uint_t page_create_putbacks; 2557c478bd9Sstevel@tonic-gate uint_t page_create_overshoot; 2567c478bd9Sstevel@tonic-gate 2577c478bd9Sstevel@tonic-gate uint_t page_reclaim_zero; 2587c478bd9Sstevel@tonic-gate uint_t page_reclaim_zero_locked; 2597c478bd9Sstevel@tonic-gate 2607c478bd9Sstevel@tonic-gate uint_t page_rename_exists; 2617c478bd9Sstevel@tonic-gate uint_t page_rename_count; 2627c478bd9Sstevel@tonic-gate 2637c478bd9Sstevel@tonic-gate uint_t page_lookup_cnt[20]; 2647c478bd9Sstevel@tonic-gate uint_t page_lookup_nowait_cnt[10]; 2657c478bd9Sstevel@tonic-gate uint_t page_find_cnt; 2667c478bd9Sstevel@tonic-gate uint_t page_exists_cnt; 2677c478bd9Sstevel@tonic-gate uint_t page_exists_forreal_cnt; 2687c478bd9Sstevel@tonic-gate uint_t page_lookup_dev_cnt; 2697c478bd9Sstevel@tonic-gate uint_t get_cachelist_cnt; 2707c478bd9Sstevel@tonic-gate uint_t page_create_cnt[10]; 27178b03d3aSkchow uint_t alloc_pages[9]; 2727c478bd9Sstevel@tonic-gate uint_t page_exphcontg[19]; 2737c478bd9Sstevel@tonic-gate uint_t page_create_large_cnt[10]; 2747c478bd9Sstevel@tonic-gate 2757c478bd9Sstevel@tonic-gate /* 2767c478bd9Sstevel@tonic-gate * Collects statistics. 2777c478bd9Sstevel@tonic-gate */ 2787c478bd9Sstevel@tonic-gate #define PAGE_HASH_SEARCH(index, pp, vp, off) { \ 2797c478bd9Sstevel@tonic-gate uint_t mylen = 0; \ 2807c478bd9Sstevel@tonic-gate \ 2817c478bd9Sstevel@tonic-gate for ((pp) = page_hash[(index)]; (pp); (pp) = (pp)->p_hash, mylen++) { \ 2827c478bd9Sstevel@tonic-gate if ((pp)->p_vnode == (vp) && (pp)->p_offset == (off)) \ 2837c478bd9Sstevel@tonic-gate break; \ 2847c478bd9Sstevel@tonic-gate } \ 2857c478bd9Sstevel@tonic-gate if ((pp) != NULL) \ 2867c478bd9Sstevel@tonic-gate pagecnt.pc_find_hit++; \ 2877c478bd9Sstevel@tonic-gate else \ 2887c478bd9Sstevel@tonic-gate pagecnt.pc_find_miss++; \ 2897c478bd9Sstevel@tonic-gate if (mylen > PC_HASH_CNT) \ 2907c478bd9Sstevel@tonic-gate mylen = PC_HASH_CNT; \ 2917c478bd9Sstevel@tonic-gate pagecnt.pc_find_hashlen[mylen]++; \ 2927c478bd9Sstevel@tonic-gate } 2937c478bd9Sstevel@tonic-gate 2947c478bd9Sstevel@tonic-gate #else /* VM_STATS */ 2957c478bd9Sstevel@tonic-gate 2967c478bd9Sstevel@tonic-gate /* 2977c478bd9Sstevel@tonic-gate * Don't collect statistics 2987c478bd9Sstevel@tonic-gate */ 2997c478bd9Sstevel@tonic-gate #define PAGE_HASH_SEARCH(index, pp, vp, off) { \ 3007c478bd9Sstevel@tonic-gate for ((pp) = page_hash[(index)]; (pp); (pp) = (pp)->p_hash) { \ 3017c478bd9Sstevel@tonic-gate if ((pp)->p_vnode == (vp) && (pp)->p_offset == (off)) \ 3027c478bd9Sstevel@tonic-gate break; \ 3037c478bd9Sstevel@tonic-gate } \ 3047c478bd9Sstevel@tonic-gate } 3057c478bd9Sstevel@tonic-gate 3067c478bd9Sstevel@tonic-gate #endif /* VM_STATS */ 3077c478bd9Sstevel@tonic-gate 3087c478bd9Sstevel@tonic-gate 3097c478bd9Sstevel@tonic-gate 3107c478bd9Sstevel@tonic-gate #ifdef DEBUG 3117c478bd9Sstevel@tonic-gate #define MEMSEG_SEARCH_STATS 3127c478bd9Sstevel@tonic-gate #endif 3137c478bd9Sstevel@tonic-gate 3147c478bd9Sstevel@tonic-gate #ifdef MEMSEG_SEARCH_STATS 3157c478bd9Sstevel@tonic-gate struct memseg_stats { 3167c478bd9Sstevel@tonic-gate uint_t nsearch; 3177c478bd9Sstevel@tonic-gate uint_t nlastwon; 3187c478bd9Sstevel@tonic-gate uint_t nhashwon; 3197c478bd9Sstevel@tonic-gate uint_t nnotfound; 3207c478bd9Sstevel@tonic-gate } memseg_stats; 3217c478bd9Sstevel@tonic-gate 3227c478bd9Sstevel@tonic-gate #define MEMSEG_STAT_INCR(v) \ 3237c478bd9Sstevel@tonic-gate atomic_add_32(&memseg_stats.v, 1) 3247c478bd9Sstevel@tonic-gate #else 3257c478bd9Sstevel@tonic-gate #define MEMSEG_STAT_INCR(x) 3267c478bd9Sstevel@tonic-gate #endif 3277c478bd9Sstevel@tonic-gate 3287c478bd9Sstevel@tonic-gate struct memseg *memsegs; /* list of memory segments */ 3297c478bd9Sstevel@tonic-gate 3302be2af34Smec /* 3312be2af34Smec * /etc/system tunable to control large page allocation hueristic. 3322be2af34Smec * 3332be2af34Smec * Setting to LPAP_LOCAL will heavily prefer the local lgroup over remote lgroup 3342be2af34Smec * for large page allocation requests. If a large page is not readily 3352be2af34Smec * avaliable on the local freelists we will go through additional effort 3362be2af34Smec * to create a large page, potentially moving smaller pages around to coalesce 3372be2af34Smec * larger pages in the local lgroup. 3382be2af34Smec * Default value of LPAP_DEFAULT will go to remote freelists if large pages 3392be2af34Smec * are not readily available in the local lgroup. 3402be2af34Smec */ 3412be2af34Smec enum lpap { 3422be2af34Smec LPAP_DEFAULT, /* default large page allocation policy */ 3432be2af34Smec LPAP_LOCAL /* local large page allocation policy */ 3442be2af34Smec }; 3452be2af34Smec 3462be2af34Smec enum lpap lpg_alloc_prefer = LPAP_DEFAULT; 3477c478bd9Sstevel@tonic-gate 3487c478bd9Sstevel@tonic-gate static void page_init_mem_config(void); 3497c478bd9Sstevel@tonic-gate static int page_do_hashin(page_t *, vnode_t *, u_offset_t); 3507c478bd9Sstevel@tonic-gate static void page_do_hashout(page_t *); 3518b464eb8Smec static void page_capture_init(); 3528b464eb8Smec int page_capture_take_action(page_t *, uint_t, void *); 3537c478bd9Sstevel@tonic-gate 3547c478bd9Sstevel@tonic-gate static void page_demote_vp_pages(page_t *); 3557c478bd9Sstevel@tonic-gate 356*06fb6a36Sdv 357*06fb6a36Sdv void 358*06fb6a36Sdv pcf_init(void) 359*06fb6a36Sdv 360*06fb6a36Sdv { 361*06fb6a36Sdv int i; 362*06fb6a36Sdv 363*06fb6a36Sdv if (boot_ncpus != -1) { 364*06fb6a36Sdv pcf_fanout = boot_ncpus; 365*06fb6a36Sdv } else { 366*06fb6a36Sdv pcf_fanout = max_ncpus; 367*06fb6a36Sdv } 368*06fb6a36Sdv #ifdef sun4v 369*06fb6a36Sdv /* 370*06fb6a36Sdv * Force at least 4 buckets if possible for sun4v. 371*06fb6a36Sdv */ 372*06fb6a36Sdv pcf_fanout = MAX(pcf_fanout, 4); 373*06fb6a36Sdv #endif /* sun4v */ 374*06fb6a36Sdv 375*06fb6a36Sdv /* 376*06fb6a36Sdv * Round up to the nearest power of 2. 377*06fb6a36Sdv */ 378*06fb6a36Sdv pcf_fanout = MIN(pcf_fanout, MAX_PCF_FANOUT); 379*06fb6a36Sdv if (!ISP2(pcf_fanout)) { 380*06fb6a36Sdv pcf_fanout = 1 << highbit(pcf_fanout); 381*06fb6a36Sdv 382*06fb6a36Sdv if (pcf_fanout > MAX_PCF_FANOUT) { 383*06fb6a36Sdv pcf_fanout = 1 << (highbit(MAX_PCF_FANOUT) - 1); 384*06fb6a36Sdv } 385*06fb6a36Sdv } 386*06fb6a36Sdv pcf_fanout_mask = pcf_fanout - 1; 387*06fb6a36Sdv } 388*06fb6a36Sdv 3897c478bd9Sstevel@tonic-gate /* 3907c478bd9Sstevel@tonic-gate * vm subsystem related initialization 3917c478bd9Sstevel@tonic-gate */ 3927c478bd9Sstevel@tonic-gate void 3937c478bd9Sstevel@tonic-gate vm_init(void) 3947c478bd9Sstevel@tonic-gate { 3957c478bd9Sstevel@tonic-gate boolean_t callb_vm_cpr(void *, int); 3967c478bd9Sstevel@tonic-gate 3977c478bd9Sstevel@tonic-gate (void) callb_add(callb_vm_cpr, 0, CB_CL_CPR_VM, "vm"); 3987c478bd9Sstevel@tonic-gate page_init_mem_config(); 399db874c57Selowe page_retire_init(); 4000209230bSgjelinek vm_usage_init(); 4018b464eb8Smec page_capture_init(); 4027c478bd9Sstevel@tonic-gate } 4037c478bd9Sstevel@tonic-gate 4047c478bd9Sstevel@tonic-gate /* 4057c478bd9Sstevel@tonic-gate * This function is called at startup and when memory is added or deleted. 4067c478bd9Sstevel@tonic-gate */ 4077c478bd9Sstevel@tonic-gate void 4087c478bd9Sstevel@tonic-gate init_pages_pp_maximum() 4097c478bd9Sstevel@tonic-gate { 4107c478bd9Sstevel@tonic-gate static pgcnt_t p_min; 4117c478bd9Sstevel@tonic-gate static pgcnt_t pages_pp_maximum_startup; 4127c478bd9Sstevel@tonic-gate static pgcnt_t avrmem_delta; 4137c478bd9Sstevel@tonic-gate static int init_done; 4147c478bd9Sstevel@tonic-gate static int user_set; /* true if set in /etc/system */ 4157c478bd9Sstevel@tonic-gate 4167c478bd9Sstevel@tonic-gate if (init_done == 0) { 4177c478bd9Sstevel@tonic-gate 4187c478bd9Sstevel@tonic-gate /* If the user specified a value, save it */ 4197c478bd9Sstevel@tonic-gate if (pages_pp_maximum != 0) { 4207c478bd9Sstevel@tonic-gate user_set = 1; 4217c478bd9Sstevel@tonic-gate pages_pp_maximum_startup = pages_pp_maximum; 4227c478bd9Sstevel@tonic-gate } 4237c478bd9Sstevel@tonic-gate 4247c478bd9Sstevel@tonic-gate /* 4257c478bd9Sstevel@tonic-gate * Setting of pages_pp_maximum is based first time 4267c478bd9Sstevel@tonic-gate * on the value of availrmem just after the start-up 4277c478bd9Sstevel@tonic-gate * allocations. To preserve this relationship at run 4287c478bd9Sstevel@tonic-gate * time, use a delta from availrmem_initial. 4297c478bd9Sstevel@tonic-gate */ 4307c478bd9Sstevel@tonic-gate ASSERT(availrmem_initial >= availrmem); 4317c478bd9Sstevel@tonic-gate avrmem_delta = availrmem_initial - availrmem; 4327c478bd9Sstevel@tonic-gate 4337c478bd9Sstevel@tonic-gate /* The allowable floor of pages_pp_maximum */ 4347c478bd9Sstevel@tonic-gate p_min = tune.t_minarmem + 100; 4357c478bd9Sstevel@tonic-gate 4367c478bd9Sstevel@tonic-gate /* Make sure we don't come through here again. */ 4377c478bd9Sstevel@tonic-gate init_done = 1; 4387c478bd9Sstevel@tonic-gate } 4397c478bd9Sstevel@tonic-gate /* 4407c478bd9Sstevel@tonic-gate * Determine pages_pp_maximum, the number of currently available 4417c478bd9Sstevel@tonic-gate * pages (availrmem) that can't be `locked'. If not set by 4427c478bd9Sstevel@tonic-gate * the user, we set it to 4% of the currently available memory 4437c478bd9Sstevel@tonic-gate * plus 4MB. 4447c478bd9Sstevel@tonic-gate * But we also insist that it be greater than tune.t_minarmem; 4457c478bd9Sstevel@tonic-gate * otherwise a process could lock down a lot of memory, get swapped 4467c478bd9Sstevel@tonic-gate * out, and never have enough to get swapped back in. 4477c478bd9Sstevel@tonic-gate */ 4487c478bd9Sstevel@tonic-gate if (user_set) 4497c478bd9Sstevel@tonic-gate pages_pp_maximum = pages_pp_maximum_startup; 4507c478bd9Sstevel@tonic-gate else 4517c478bd9Sstevel@tonic-gate pages_pp_maximum = ((availrmem_initial - avrmem_delta) / 25) 4527c478bd9Sstevel@tonic-gate + btop(4 * 1024 * 1024); 4537c478bd9Sstevel@tonic-gate 4547c478bd9Sstevel@tonic-gate if (pages_pp_maximum <= p_min) { 4557c478bd9Sstevel@tonic-gate pages_pp_maximum = p_min; 4567c478bd9Sstevel@tonic-gate } 4577c478bd9Sstevel@tonic-gate } 4587c478bd9Sstevel@tonic-gate 4597c478bd9Sstevel@tonic-gate void 4607c478bd9Sstevel@tonic-gate set_max_page_get(pgcnt_t target_total_pages) 4617c478bd9Sstevel@tonic-gate { 4627c478bd9Sstevel@tonic-gate max_page_get = target_total_pages / 2; 4637c478bd9Sstevel@tonic-gate } 4647c478bd9Sstevel@tonic-gate 4657c478bd9Sstevel@tonic-gate static pgcnt_t pending_delete; 4667c478bd9Sstevel@tonic-gate 4677c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 4687c478bd9Sstevel@tonic-gate static void 4697c478bd9Sstevel@tonic-gate page_mem_config_post_add( 4707c478bd9Sstevel@tonic-gate void *arg, 4717c478bd9Sstevel@tonic-gate pgcnt_t delta_pages) 4727c478bd9Sstevel@tonic-gate { 4737c478bd9Sstevel@tonic-gate set_max_page_get(total_pages - pending_delete); 4747c478bd9Sstevel@tonic-gate init_pages_pp_maximum(); 4757c478bd9Sstevel@tonic-gate } 4767c478bd9Sstevel@tonic-gate 4777c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 4787c478bd9Sstevel@tonic-gate static int 4797c478bd9Sstevel@tonic-gate page_mem_config_pre_del( 4807c478bd9Sstevel@tonic-gate void *arg, 4817c478bd9Sstevel@tonic-gate pgcnt_t delta_pages) 4827c478bd9Sstevel@tonic-gate { 4837c478bd9Sstevel@tonic-gate pgcnt_t nv; 4847c478bd9Sstevel@tonic-gate 4857c478bd9Sstevel@tonic-gate nv = atomic_add_long_nv(&pending_delete, (spgcnt_t)delta_pages); 4867c478bd9Sstevel@tonic-gate set_max_page_get(total_pages - nv); 4877c478bd9Sstevel@tonic-gate return (0); 4887c478bd9Sstevel@tonic-gate } 4897c478bd9Sstevel@tonic-gate 4907c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 4917c478bd9Sstevel@tonic-gate static void 4927c478bd9Sstevel@tonic-gate page_mem_config_post_del( 4937c478bd9Sstevel@tonic-gate void *arg, 4947c478bd9Sstevel@tonic-gate pgcnt_t delta_pages, 4957c478bd9Sstevel@tonic-gate int cancelled) 4967c478bd9Sstevel@tonic-gate { 4977c478bd9Sstevel@tonic-gate pgcnt_t nv; 4987c478bd9Sstevel@tonic-gate 4997c478bd9Sstevel@tonic-gate nv = atomic_add_long_nv(&pending_delete, -(spgcnt_t)delta_pages); 5007c478bd9Sstevel@tonic-gate set_max_page_get(total_pages - nv); 5017c478bd9Sstevel@tonic-gate if (!cancelled) 5027c478bd9Sstevel@tonic-gate init_pages_pp_maximum(); 5037c478bd9Sstevel@tonic-gate } 5047c478bd9Sstevel@tonic-gate 5057c478bd9Sstevel@tonic-gate static kphysm_setup_vector_t page_mem_config_vec = { 5067c478bd9Sstevel@tonic-gate KPHYSM_SETUP_VECTOR_VERSION, 5077c478bd9Sstevel@tonic-gate page_mem_config_post_add, 5087c478bd9Sstevel@tonic-gate page_mem_config_pre_del, 5097c478bd9Sstevel@tonic-gate page_mem_config_post_del, 5107c478bd9Sstevel@tonic-gate }; 5117c478bd9Sstevel@tonic-gate 5127c478bd9Sstevel@tonic-gate static void 5137c478bd9Sstevel@tonic-gate page_init_mem_config(void) 5147c478bd9Sstevel@tonic-gate { 5157c478bd9Sstevel@tonic-gate int ret; 5167c478bd9Sstevel@tonic-gate 5177c478bd9Sstevel@tonic-gate ret = kphysm_setup_func_register(&page_mem_config_vec, (void *)NULL); 5187c478bd9Sstevel@tonic-gate ASSERT(ret == 0); 5197c478bd9Sstevel@tonic-gate } 5207c478bd9Sstevel@tonic-gate 5217c478bd9Sstevel@tonic-gate /* 5227c478bd9Sstevel@tonic-gate * Evenly spread out the PCF counters for large free pages 5237c478bd9Sstevel@tonic-gate */ 5247c478bd9Sstevel@tonic-gate static void 5257c478bd9Sstevel@tonic-gate page_free_large_ctr(pgcnt_t npages) 5267c478bd9Sstevel@tonic-gate { 5277c478bd9Sstevel@tonic-gate static struct pcf *p = pcf; 5287c478bd9Sstevel@tonic-gate pgcnt_t lump; 5297c478bd9Sstevel@tonic-gate 5307c478bd9Sstevel@tonic-gate freemem += npages; 5317c478bd9Sstevel@tonic-gate 532*06fb6a36Sdv lump = roundup(npages, pcf_fanout) / pcf_fanout; 5337c478bd9Sstevel@tonic-gate 5347c478bd9Sstevel@tonic-gate while (npages > 0) { 5357c478bd9Sstevel@tonic-gate 5367c478bd9Sstevel@tonic-gate ASSERT(!p->pcf_block); 5377c478bd9Sstevel@tonic-gate 5387c478bd9Sstevel@tonic-gate if (lump < npages) { 5397c478bd9Sstevel@tonic-gate p->pcf_count += (uint_t)lump; 5407c478bd9Sstevel@tonic-gate npages -= lump; 5417c478bd9Sstevel@tonic-gate } else { 5427c478bd9Sstevel@tonic-gate p->pcf_count += (uint_t)npages; 5437c478bd9Sstevel@tonic-gate npages = 0; 5447c478bd9Sstevel@tonic-gate } 5457c478bd9Sstevel@tonic-gate 5467c478bd9Sstevel@tonic-gate ASSERT(!p->pcf_wait); 5477c478bd9Sstevel@tonic-gate 548*06fb6a36Sdv if (++p > &pcf[pcf_fanout - 1]) 5497c478bd9Sstevel@tonic-gate p = pcf; 5507c478bd9Sstevel@tonic-gate } 5517c478bd9Sstevel@tonic-gate 5527c478bd9Sstevel@tonic-gate ASSERT(npages == 0); 5537c478bd9Sstevel@tonic-gate } 5547c478bd9Sstevel@tonic-gate 5557c478bd9Sstevel@tonic-gate /* 556da6c28aaSamw * Add a physical chunk of memory to the system free lists during startup. 5577c478bd9Sstevel@tonic-gate * Platform specific startup() allocates the memory for the page structs. 5587c478bd9Sstevel@tonic-gate * 5597c478bd9Sstevel@tonic-gate * num - number of page structures 5607c478bd9Sstevel@tonic-gate * base - page number (pfn) to be associated with the first page. 5617c478bd9Sstevel@tonic-gate * 5627c478bd9Sstevel@tonic-gate * Since we are doing this during startup (ie. single threaded), we will 5637c478bd9Sstevel@tonic-gate * use shortcut routines to avoid any locking overhead while putting all 5647c478bd9Sstevel@tonic-gate * these pages on the freelists. 5657c478bd9Sstevel@tonic-gate * 5667c478bd9Sstevel@tonic-gate * NOTE: Any changes performed to page_free(), must also be performed to 5677c478bd9Sstevel@tonic-gate * add_physmem() since this is how we initialize all page_t's at 5687c478bd9Sstevel@tonic-gate * boot time. 5697c478bd9Sstevel@tonic-gate */ 5707c478bd9Sstevel@tonic-gate void 5717c478bd9Sstevel@tonic-gate add_physmem( 5727c478bd9Sstevel@tonic-gate page_t *pp, 5737c478bd9Sstevel@tonic-gate pgcnt_t num, 5747c478bd9Sstevel@tonic-gate pfn_t pnum) 5757c478bd9Sstevel@tonic-gate { 5767c478bd9Sstevel@tonic-gate page_t *root = NULL; 5777c478bd9Sstevel@tonic-gate uint_t szc = page_num_pagesizes() - 1; 5787c478bd9Sstevel@tonic-gate pgcnt_t large = page_get_pagecnt(szc); 5797c478bd9Sstevel@tonic-gate pgcnt_t cnt = 0; 5807c478bd9Sstevel@tonic-gate 5817c478bd9Sstevel@tonic-gate TRACE_2(TR_FAC_VM, TR_PAGE_INIT, 5826e4dd838Smec "add_physmem:pp %p num %lu", pp, num); 5837c478bd9Sstevel@tonic-gate 5847c478bd9Sstevel@tonic-gate /* 5857c478bd9Sstevel@tonic-gate * Arbitrarily limit the max page_get request 5867c478bd9Sstevel@tonic-gate * to 1/2 of the page structs we have. 5877c478bd9Sstevel@tonic-gate */ 5887c478bd9Sstevel@tonic-gate total_pages += num; 5897c478bd9Sstevel@tonic-gate set_max_page_get(total_pages); 5907c478bd9Sstevel@tonic-gate 591e21bae1bSkchow PLCNT_MODIFY_MAX(pnum, (long)num); 592e21bae1bSkchow 5937c478bd9Sstevel@tonic-gate /* 5947c478bd9Sstevel@tonic-gate * The physical space for the pages array 5957c478bd9Sstevel@tonic-gate * representing ram pages has already been 5967c478bd9Sstevel@tonic-gate * allocated. Here we initialize each lock 5977c478bd9Sstevel@tonic-gate * in the page structure, and put each on 5987c478bd9Sstevel@tonic-gate * the free list 5997c478bd9Sstevel@tonic-gate */ 600affbd3ccSkchow for (; num; pp++, pnum++, num--) { 6017c478bd9Sstevel@tonic-gate 6027c478bd9Sstevel@tonic-gate /* 6037c478bd9Sstevel@tonic-gate * this needs to fill in the page number 6047c478bd9Sstevel@tonic-gate * and do any other arch specific initialization 6057c478bd9Sstevel@tonic-gate */ 6067c478bd9Sstevel@tonic-gate add_physmem_cb(pp, pnum); 6077c478bd9Sstevel@tonic-gate 60807b65a64Saguzovsk pp->p_lckcnt = 0; 60907b65a64Saguzovsk pp->p_cowcnt = 0; 61007b65a64Saguzovsk pp->p_slckcnt = 0; 61107b65a64Saguzovsk 6127c478bd9Sstevel@tonic-gate /* 6137c478bd9Sstevel@tonic-gate * Initialize the page lock as unlocked, since nobody 6147c478bd9Sstevel@tonic-gate * can see or access this page yet. 6157c478bd9Sstevel@tonic-gate */ 6167c478bd9Sstevel@tonic-gate pp->p_selock = 0; 6177c478bd9Sstevel@tonic-gate 6187c478bd9Sstevel@tonic-gate /* 6197c478bd9Sstevel@tonic-gate * Initialize IO lock 6207c478bd9Sstevel@tonic-gate */ 6217c478bd9Sstevel@tonic-gate page_iolock_init(pp); 6227c478bd9Sstevel@tonic-gate 6237c478bd9Sstevel@tonic-gate /* 6247c478bd9Sstevel@tonic-gate * initialize other fields in the page_t 6257c478bd9Sstevel@tonic-gate */ 6267c478bd9Sstevel@tonic-gate PP_SETFREE(pp); 6277c478bd9Sstevel@tonic-gate page_clr_all_props(pp); 6287c478bd9Sstevel@tonic-gate PP_SETAGED(pp); 6297c478bd9Sstevel@tonic-gate pp->p_offset = (u_offset_t)-1; 6307c478bd9Sstevel@tonic-gate pp->p_next = pp; 6317c478bd9Sstevel@tonic-gate pp->p_prev = pp; 6327c478bd9Sstevel@tonic-gate 6337c478bd9Sstevel@tonic-gate /* 6347c478bd9Sstevel@tonic-gate * Simple case: System doesn't support large pages. 6357c478bd9Sstevel@tonic-gate */ 6367c478bd9Sstevel@tonic-gate if (szc == 0) { 6377c478bd9Sstevel@tonic-gate pp->p_szc = 0; 6387c478bd9Sstevel@tonic-gate page_free_at_startup(pp); 6397c478bd9Sstevel@tonic-gate continue; 6407c478bd9Sstevel@tonic-gate } 6417c478bd9Sstevel@tonic-gate 6427c478bd9Sstevel@tonic-gate /* 6437c478bd9Sstevel@tonic-gate * Handle unaligned pages, we collect them up onto 6447c478bd9Sstevel@tonic-gate * the root page until we have a full large page. 6457c478bd9Sstevel@tonic-gate */ 6467c478bd9Sstevel@tonic-gate if (!IS_P2ALIGNED(pnum, large)) { 6477c478bd9Sstevel@tonic-gate 6487c478bd9Sstevel@tonic-gate /* 6497c478bd9Sstevel@tonic-gate * If not in a large page, 6507c478bd9Sstevel@tonic-gate * just free as small page. 6517c478bd9Sstevel@tonic-gate */ 6527c478bd9Sstevel@tonic-gate if (root == NULL) { 6537c478bd9Sstevel@tonic-gate pp->p_szc = 0; 6547c478bd9Sstevel@tonic-gate page_free_at_startup(pp); 6557c478bd9Sstevel@tonic-gate continue; 6567c478bd9Sstevel@tonic-gate } 6577c478bd9Sstevel@tonic-gate 6587c478bd9Sstevel@tonic-gate /* 6597c478bd9Sstevel@tonic-gate * Link a constituent page into the large page. 6607c478bd9Sstevel@tonic-gate */ 6617c478bd9Sstevel@tonic-gate pp->p_szc = szc; 6627c478bd9Sstevel@tonic-gate page_list_concat(&root, &pp); 6637c478bd9Sstevel@tonic-gate 6647c478bd9Sstevel@tonic-gate /* 6657c478bd9Sstevel@tonic-gate * When large page is fully formed, free it. 6667c478bd9Sstevel@tonic-gate */ 6677c478bd9Sstevel@tonic-gate if (++cnt == large) { 6687c478bd9Sstevel@tonic-gate page_free_large_ctr(cnt); 6697c478bd9Sstevel@tonic-gate page_list_add_pages(root, PG_LIST_ISINIT); 6707c478bd9Sstevel@tonic-gate root = NULL; 6717c478bd9Sstevel@tonic-gate cnt = 0; 6727c478bd9Sstevel@tonic-gate } 6737c478bd9Sstevel@tonic-gate continue; 6747c478bd9Sstevel@tonic-gate } 6757c478bd9Sstevel@tonic-gate 6767c478bd9Sstevel@tonic-gate /* 6777c478bd9Sstevel@tonic-gate * At this point we have a page number which 6787c478bd9Sstevel@tonic-gate * is aligned. We assert that we aren't already 6797c478bd9Sstevel@tonic-gate * in a different large page. 6807c478bd9Sstevel@tonic-gate */ 6817c478bd9Sstevel@tonic-gate ASSERT(IS_P2ALIGNED(pnum, large)); 6827c478bd9Sstevel@tonic-gate ASSERT(root == NULL && cnt == 0); 6837c478bd9Sstevel@tonic-gate 6847c478bd9Sstevel@tonic-gate /* 6857c478bd9Sstevel@tonic-gate * If insufficient number of pages left to form 6867c478bd9Sstevel@tonic-gate * a large page, just free the small page. 6877c478bd9Sstevel@tonic-gate */ 6887c478bd9Sstevel@tonic-gate if (num < large) { 6897c478bd9Sstevel@tonic-gate pp->p_szc = 0; 6907c478bd9Sstevel@tonic-gate page_free_at_startup(pp); 6917c478bd9Sstevel@tonic-gate continue; 6927c478bd9Sstevel@tonic-gate } 6937c478bd9Sstevel@tonic-gate 6947c478bd9Sstevel@tonic-gate /* 6957c478bd9Sstevel@tonic-gate * Otherwise start a new large page. 6967c478bd9Sstevel@tonic-gate */ 6977c478bd9Sstevel@tonic-gate pp->p_szc = szc; 6987c478bd9Sstevel@tonic-gate cnt++; 6997c478bd9Sstevel@tonic-gate root = pp; 7007c478bd9Sstevel@tonic-gate } 7017c478bd9Sstevel@tonic-gate ASSERT(root == NULL && cnt == 0); 7027c478bd9Sstevel@tonic-gate } 7037c478bd9Sstevel@tonic-gate 7047c478bd9Sstevel@tonic-gate /* 7057c478bd9Sstevel@tonic-gate * Find a page representing the specified [vp, offset]. 7067c478bd9Sstevel@tonic-gate * If we find the page but it is intransit coming in, 7077c478bd9Sstevel@tonic-gate * it will have an "exclusive" lock and we wait for 7087c478bd9Sstevel@tonic-gate * the i/o to complete. A page found on the free list 7097c478bd9Sstevel@tonic-gate * is always reclaimed and then locked. On success, the page 7107c478bd9Sstevel@tonic-gate * is locked, its data is valid and it isn't on the free 7117c478bd9Sstevel@tonic-gate * list, while a NULL is returned if the page doesn't exist. 7127c478bd9Sstevel@tonic-gate */ 7137c478bd9Sstevel@tonic-gate page_t * 7147c478bd9Sstevel@tonic-gate page_lookup(vnode_t *vp, u_offset_t off, se_t se) 7157c478bd9Sstevel@tonic-gate { 7167c478bd9Sstevel@tonic-gate return (page_lookup_create(vp, off, se, NULL, NULL, 0)); 7177c478bd9Sstevel@tonic-gate } 7187c478bd9Sstevel@tonic-gate 7197c478bd9Sstevel@tonic-gate /* 7207c478bd9Sstevel@tonic-gate * Find a page representing the specified [vp, offset]. 7217c478bd9Sstevel@tonic-gate * We either return the one we found or, if passed in, 7227c478bd9Sstevel@tonic-gate * create one with identity of [vp, offset] of the 723da6c28aaSamw * pre-allocated page. If we find existing page but it is 7247c478bd9Sstevel@tonic-gate * intransit coming in, it will have an "exclusive" lock 7257c478bd9Sstevel@tonic-gate * and we wait for the i/o to complete. A page found on 7267c478bd9Sstevel@tonic-gate * the free list is always reclaimed and then locked. 7277c478bd9Sstevel@tonic-gate * On success, the page is locked, its data is valid and 7287c478bd9Sstevel@tonic-gate * it isn't on the free list, while a NULL is returned 7297c478bd9Sstevel@tonic-gate * if the page doesn't exist and newpp is NULL; 7307c478bd9Sstevel@tonic-gate */ 7317c478bd9Sstevel@tonic-gate page_t * 7327c478bd9Sstevel@tonic-gate page_lookup_create( 7337c478bd9Sstevel@tonic-gate vnode_t *vp, 7347c478bd9Sstevel@tonic-gate u_offset_t off, 7357c478bd9Sstevel@tonic-gate se_t se, 7367c478bd9Sstevel@tonic-gate page_t *newpp, 7377c478bd9Sstevel@tonic-gate spgcnt_t *nrelocp, 7387c478bd9Sstevel@tonic-gate int flags) 7397c478bd9Sstevel@tonic-gate { 7407c478bd9Sstevel@tonic-gate page_t *pp; 7417c478bd9Sstevel@tonic-gate kmutex_t *phm; 7427c478bd9Sstevel@tonic-gate ulong_t index; 7437c478bd9Sstevel@tonic-gate uint_t hash_locked; 7447c478bd9Sstevel@tonic-gate uint_t es; 7457c478bd9Sstevel@tonic-gate 7467c478bd9Sstevel@tonic-gate ASSERT(MUTEX_NOT_HELD(page_vnode_mutex(vp))); 7477c478bd9Sstevel@tonic-gate VM_STAT_ADD(page_lookup_cnt[0]); 7487c478bd9Sstevel@tonic-gate ASSERT(newpp ? PAGE_EXCL(newpp) : 1); 7497c478bd9Sstevel@tonic-gate 7507c478bd9Sstevel@tonic-gate /* 7517c478bd9Sstevel@tonic-gate * Acquire the appropriate page hash lock since 7527c478bd9Sstevel@tonic-gate * we have to search the hash list. Pages that 7537c478bd9Sstevel@tonic-gate * hash to this list can't change identity while 7547c478bd9Sstevel@tonic-gate * this lock is held. 7557c478bd9Sstevel@tonic-gate */ 7567c478bd9Sstevel@tonic-gate hash_locked = 0; 7577c478bd9Sstevel@tonic-gate index = PAGE_HASH_FUNC(vp, off); 7587c478bd9Sstevel@tonic-gate phm = NULL; 7597c478bd9Sstevel@tonic-gate top: 7607c478bd9Sstevel@tonic-gate PAGE_HASH_SEARCH(index, pp, vp, off); 7617c478bd9Sstevel@tonic-gate if (pp != NULL) { 7627c478bd9Sstevel@tonic-gate VM_STAT_ADD(page_lookup_cnt[1]); 7637c478bd9Sstevel@tonic-gate es = (newpp != NULL) ? 1 : 0; 7647c478bd9Sstevel@tonic-gate es |= flags; 7657c478bd9Sstevel@tonic-gate if (!hash_locked) { 7667c478bd9Sstevel@tonic-gate VM_STAT_ADD(page_lookup_cnt[2]); 7677c478bd9Sstevel@tonic-gate if (!page_try_reclaim_lock(pp, se, es)) { 7687c478bd9Sstevel@tonic-gate /* 7697c478bd9Sstevel@tonic-gate * On a miss, acquire the phm. Then 7707c478bd9Sstevel@tonic-gate * next time, page_lock() will be called, 7717c478bd9Sstevel@tonic-gate * causing a wait if the page is busy. 7727c478bd9Sstevel@tonic-gate * just looping with page_trylock() would 7737c478bd9Sstevel@tonic-gate * get pretty boring. 7747c478bd9Sstevel@tonic-gate */ 7757c478bd9Sstevel@tonic-gate VM_STAT_ADD(page_lookup_cnt[3]); 7767c478bd9Sstevel@tonic-gate phm = PAGE_HASH_MUTEX(index); 7777c478bd9Sstevel@tonic-gate mutex_enter(phm); 7787c478bd9Sstevel@tonic-gate hash_locked = 1; 7797c478bd9Sstevel@tonic-gate goto top; 7807c478bd9Sstevel@tonic-gate } 7817c478bd9Sstevel@tonic-gate } else { 7827c478bd9Sstevel@tonic-gate VM_STAT_ADD(page_lookup_cnt[4]); 7837c478bd9Sstevel@tonic-gate if (!page_lock_es(pp, se, phm, P_RECLAIM, es)) { 7847c478bd9Sstevel@tonic-gate VM_STAT_ADD(page_lookup_cnt[5]); 7857c478bd9Sstevel@tonic-gate goto top; 7867c478bd9Sstevel@tonic-gate } 7877c478bd9Sstevel@tonic-gate } 7887c478bd9Sstevel@tonic-gate 7897c478bd9Sstevel@tonic-gate /* 7907c478bd9Sstevel@tonic-gate * Since `pp' is locked it can not change identity now. 7917c478bd9Sstevel@tonic-gate * Reconfirm we locked the correct page. 7927c478bd9Sstevel@tonic-gate * 7937c478bd9Sstevel@tonic-gate * Both the p_vnode and p_offset *must* be cast volatile 7947c478bd9Sstevel@tonic-gate * to force a reload of their values: The PAGE_HASH_SEARCH 7957c478bd9Sstevel@tonic-gate * macro will have stuffed p_vnode and p_offset into 7967c478bd9Sstevel@tonic-gate * registers before calling page_trylock(); another thread, 7977c478bd9Sstevel@tonic-gate * actually holding the hash lock, could have changed the 7987c478bd9Sstevel@tonic-gate * page's identity in memory, but our registers would not 7997c478bd9Sstevel@tonic-gate * be changed, fooling the reconfirmation. If the hash 8007c478bd9Sstevel@tonic-gate * lock was held during the search, the casting would 8017c478bd9Sstevel@tonic-gate * not be needed. 8027c478bd9Sstevel@tonic-gate */ 8037c478bd9Sstevel@tonic-gate VM_STAT_ADD(page_lookup_cnt[6]); 8047c478bd9Sstevel@tonic-gate if (((volatile struct vnode *)(pp->p_vnode) != vp) || 8057c478bd9Sstevel@tonic-gate ((volatile u_offset_t)(pp->p_offset) != off)) { 8067c478bd9Sstevel@tonic-gate VM_STAT_ADD(page_lookup_cnt[7]); 8077c478bd9Sstevel@tonic-gate if (hash_locked) { 8087c478bd9Sstevel@tonic-gate panic("page_lookup_create: lost page %p", 8097c478bd9Sstevel@tonic-gate (void *)pp); 8107c478bd9Sstevel@tonic-gate /*NOTREACHED*/ 8117c478bd9Sstevel@tonic-gate } 8127c478bd9Sstevel@tonic-gate page_unlock(pp); 8137c478bd9Sstevel@tonic-gate phm = PAGE_HASH_MUTEX(index); 8147c478bd9Sstevel@tonic-gate mutex_enter(phm); 8157c478bd9Sstevel@tonic-gate hash_locked = 1; 8167c478bd9Sstevel@tonic-gate goto top; 8177c478bd9Sstevel@tonic-gate } 8187c478bd9Sstevel@tonic-gate 8197c478bd9Sstevel@tonic-gate /* 8207c478bd9Sstevel@tonic-gate * If page_trylock() was called, then pp may still be on 8217c478bd9Sstevel@tonic-gate * the cachelist (can't be on the free list, it would not 8227c478bd9Sstevel@tonic-gate * have been found in the search). If it is on the 8237c478bd9Sstevel@tonic-gate * cachelist it must be pulled now. To pull the page from 8247c478bd9Sstevel@tonic-gate * the cachelist, it must be exclusively locked. 8257c478bd9Sstevel@tonic-gate * 8267c478bd9Sstevel@tonic-gate * The other big difference between page_trylock() and 8277c478bd9Sstevel@tonic-gate * page_lock(), is that page_lock() will pull the 8287c478bd9Sstevel@tonic-gate * page from whatever free list (the cache list in this 8297c478bd9Sstevel@tonic-gate * case) the page is on. If page_trylock() was used 8307c478bd9Sstevel@tonic-gate * above, then we have to do the reclaim ourselves. 8317c478bd9Sstevel@tonic-gate */ 8327c478bd9Sstevel@tonic-gate if ((!hash_locked) && (PP_ISFREE(pp))) { 8337c478bd9Sstevel@tonic-gate ASSERT(PP_ISAGED(pp) == 0); 8347c478bd9Sstevel@tonic-gate VM_STAT_ADD(page_lookup_cnt[8]); 8357c478bd9Sstevel@tonic-gate 8367c478bd9Sstevel@tonic-gate /* 8377c478bd9Sstevel@tonic-gate * page_relcaim will insure that we 8387c478bd9Sstevel@tonic-gate * have this page exclusively 8397c478bd9Sstevel@tonic-gate */ 8407c478bd9Sstevel@tonic-gate 8417c478bd9Sstevel@tonic-gate if (!page_reclaim(pp, NULL)) { 8427c478bd9Sstevel@tonic-gate /* 8437c478bd9Sstevel@tonic-gate * Page_reclaim dropped whatever lock 8447c478bd9Sstevel@tonic-gate * we held. 8457c478bd9Sstevel@tonic-gate */ 8467c478bd9Sstevel@tonic-gate VM_STAT_ADD(page_lookup_cnt[9]); 8477c478bd9Sstevel@tonic-gate phm = PAGE_HASH_MUTEX(index); 8487c478bd9Sstevel@tonic-gate mutex_enter(phm); 8497c478bd9Sstevel@tonic-gate hash_locked = 1; 8507c478bd9Sstevel@tonic-gate goto top; 8517c478bd9Sstevel@tonic-gate } else if (se == SE_SHARED && newpp == NULL) { 8527c478bd9Sstevel@tonic-gate VM_STAT_ADD(page_lookup_cnt[10]); 8537c478bd9Sstevel@tonic-gate page_downgrade(pp); 8547c478bd9Sstevel@tonic-gate } 8557c478bd9Sstevel@tonic-gate } 8567c478bd9Sstevel@tonic-gate 8577c478bd9Sstevel@tonic-gate if (hash_locked) { 8587c478bd9Sstevel@tonic-gate mutex_exit(phm); 8597c478bd9Sstevel@tonic-gate } 8607c478bd9Sstevel@tonic-gate 8617c478bd9Sstevel@tonic-gate if (newpp != NULL && pp->p_szc < newpp->p_szc && 8627c478bd9Sstevel@tonic-gate PAGE_EXCL(pp) && nrelocp != NULL) { 8637c478bd9Sstevel@tonic-gate ASSERT(nrelocp != NULL); 8647c478bd9Sstevel@tonic-gate (void) page_relocate(&pp, &newpp, 1, 1, nrelocp, 8657c478bd9Sstevel@tonic-gate NULL); 8667c478bd9Sstevel@tonic-gate if (*nrelocp > 0) { 8677c478bd9Sstevel@tonic-gate VM_STAT_COND_ADD(*nrelocp == 1, 8687c478bd9Sstevel@tonic-gate page_lookup_cnt[11]); 8697c478bd9Sstevel@tonic-gate VM_STAT_COND_ADD(*nrelocp > 1, 8707c478bd9Sstevel@tonic-gate page_lookup_cnt[12]); 8717c478bd9Sstevel@tonic-gate pp = newpp; 8727c478bd9Sstevel@tonic-gate se = SE_EXCL; 8737c478bd9Sstevel@tonic-gate } else { 8747c478bd9Sstevel@tonic-gate if (se == SE_SHARED) { 8757c478bd9Sstevel@tonic-gate page_downgrade(pp); 8767c478bd9Sstevel@tonic-gate } 8777c478bd9Sstevel@tonic-gate VM_STAT_ADD(page_lookup_cnt[13]); 8787c478bd9Sstevel@tonic-gate } 8797c478bd9Sstevel@tonic-gate } else if (newpp != NULL && nrelocp != NULL) { 8807c478bd9Sstevel@tonic-gate if (PAGE_EXCL(pp) && se == SE_SHARED) { 8817c478bd9Sstevel@tonic-gate page_downgrade(pp); 8827c478bd9Sstevel@tonic-gate } 8837c478bd9Sstevel@tonic-gate VM_STAT_COND_ADD(pp->p_szc < newpp->p_szc, 8847c478bd9Sstevel@tonic-gate page_lookup_cnt[14]); 8857c478bd9Sstevel@tonic-gate VM_STAT_COND_ADD(pp->p_szc == newpp->p_szc, 8867c478bd9Sstevel@tonic-gate page_lookup_cnt[15]); 8877c478bd9Sstevel@tonic-gate VM_STAT_COND_ADD(pp->p_szc > newpp->p_szc, 8887c478bd9Sstevel@tonic-gate page_lookup_cnt[16]); 8897c478bd9Sstevel@tonic-gate } else if (newpp != NULL && PAGE_EXCL(pp)) { 8907c478bd9Sstevel@tonic-gate se = SE_EXCL; 8917c478bd9Sstevel@tonic-gate } 8927c478bd9Sstevel@tonic-gate } else if (!hash_locked) { 8937c478bd9Sstevel@tonic-gate VM_STAT_ADD(page_lookup_cnt[17]); 8947c478bd9Sstevel@tonic-gate phm = PAGE_HASH_MUTEX(index); 8957c478bd9Sstevel@tonic-gate mutex_enter(phm); 8967c478bd9Sstevel@tonic-gate hash_locked = 1; 8977c478bd9Sstevel@tonic-gate goto top; 8987c478bd9Sstevel@tonic-gate } else if (newpp != NULL) { 8997c478bd9Sstevel@tonic-gate /* 9007c478bd9Sstevel@tonic-gate * If we have a preallocated page then 9017c478bd9Sstevel@tonic-gate * insert it now and basically behave like 9027c478bd9Sstevel@tonic-gate * page_create. 9037c478bd9Sstevel@tonic-gate */ 9047c478bd9Sstevel@tonic-gate VM_STAT_ADD(page_lookup_cnt[18]); 9057c478bd9Sstevel@tonic-gate /* 9067c478bd9Sstevel@tonic-gate * Since we hold the page hash mutex and 9077c478bd9Sstevel@tonic-gate * just searched for this page, page_hashin 9087c478bd9Sstevel@tonic-gate * had better not fail. If it does, that 9097c478bd9Sstevel@tonic-gate * means some thread did not follow the 9107c478bd9Sstevel@tonic-gate * page hash mutex rules. Panic now and 9117c478bd9Sstevel@tonic-gate * get it over with. As usual, go down 9127c478bd9Sstevel@tonic-gate * holding all the locks. 9137c478bd9Sstevel@tonic-gate */ 9147c478bd9Sstevel@tonic-gate ASSERT(MUTEX_HELD(phm)); 9157c478bd9Sstevel@tonic-gate if (!page_hashin(newpp, vp, off, phm)) { 9167c478bd9Sstevel@tonic-gate ASSERT(MUTEX_HELD(phm)); 9177c478bd9Sstevel@tonic-gate panic("page_lookup_create: hashin failed %p %p %llx %p", 9187c478bd9Sstevel@tonic-gate (void *)newpp, (void *)vp, off, (void *)phm); 9197c478bd9Sstevel@tonic-gate /*NOTREACHED*/ 9207c478bd9Sstevel@tonic-gate } 9217c478bd9Sstevel@tonic-gate ASSERT(MUTEX_HELD(phm)); 9227c478bd9Sstevel@tonic-gate mutex_exit(phm); 9237c478bd9Sstevel@tonic-gate phm = NULL; 9247c478bd9Sstevel@tonic-gate page_set_props(newpp, P_REF); 9257c478bd9Sstevel@tonic-gate page_io_lock(newpp); 9267c478bd9Sstevel@tonic-gate pp = newpp; 9277c478bd9Sstevel@tonic-gate se = SE_EXCL; 9287c478bd9Sstevel@tonic-gate } else { 9297c478bd9Sstevel@tonic-gate VM_STAT_ADD(page_lookup_cnt[19]); 9307c478bd9Sstevel@tonic-gate mutex_exit(phm); 9317c478bd9Sstevel@tonic-gate } 9327c478bd9Sstevel@tonic-gate 9337c478bd9Sstevel@tonic-gate ASSERT(pp ? PAGE_LOCKED_SE(pp, se) : 1); 9347c478bd9Sstevel@tonic-gate 9357c478bd9Sstevel@tonic-gate ASSERT(pp ? ((PP_ISFREE(pp) == 0) && (PP_ISAGED(pp) == 0)) : 1); 9367c478bd9Sstevel@tonic-gate 9377c478bd9Sstevel@tonic-gate return (pp); 9387c478bd9Sstevel@tonic-gate } 9397c478bd9Sstevel@tonic-gate 9407c478bd9Sstevel@tonic-gate /* 9417c478bd9Sstevel@tonic-gate * Search the hash list for the page representing the 9427c478bd9Sstevel@tonic-gate * specified [vp, offset] and return it locked. Skip 9437c478bd9Sstevel@tonic-gate * free pages and pages that cannot be locked as requested. 9447c478bd9Sstevel@tonic-gate * Used while attempting to kluster pages. 9457c478bd9Sstevel@tonic-gate */ 9467c478bd9Sstevel@tonic-gate page_t * 9477c478bd9Sstevel@tonic-gate page_lookup_nowait(vnode_t *vp, u_offset_t off, se_t se) 9487c478bd9Sstevel@tonic-gate { 9497c478bd9Sstevel@tonic-gate page_t *pp; 9507c478bd9Sstevel@tonic-gate kmutex_t *phm; 9517c478bd9Sstevel@tonic-gate ulong_t index; 9527c478bd9Sstevel@tonic-gate uint_t locked; 9537c478bd9Sstevel@tonic-gate 9547c478bd9Sstevel@tonic-gate ASSERT(MUTEX_NOT_HELD(page_vnode_mutex(vp))); 9557c478bd9Sstevel@tonic-gate VM_STAT_ADD(page_lookup_nowait_cnt[0]); 9567c478bd9Sstevel@tonic-gate 9577c478bd9Sstevel@tonic-gate index = PAGE_HASH_FUNC(vp, off); 9587c478bd9Sstevel@tonic-gate PAGE_HASH_SEARCH(index, pp, vp, off); 9597c478bd9Sstevel@tonic-gate locked = 0; 9607c478bd9Sstevel@tonic-gate if (pp == NULL) { 9617c478bd9Sstevel@tonic-gate top: 9627c478bd9Sstevel@tonic-gate VM_STAT_ADD(page_lookup_nowait_cnt[1]); 9637c478bd9Sstevel@tonic-gate locked = 1; 9647c478bd9Sstevel@tonic-gate phm = PAGE_HASH_MUTEX(index); 9657c478bd9Sstevel@tonic-gate mutex_enter(phm); 9667c478bd9Sstevel@tonic-gate PAGE_HASH_SEARCH(index, pp, vp, off); 9677c478bd9Sstevel@tonic-gate } 9687c478bd9Sstevel@tonic-gate 9697c478bd9Sstevel@tonic-gate if (pp == NULL || PP_ISFREE(pp)) { 9707c478bd9Sstevel@tonic-gate VM_STAT_ADD(page_lookup_nowait_cnt[2]); 9717c478bd9Sstevel@tonic-gate pp = NULL; 9727c478bd9Sstevel@tonic-gate } else { 9737c478bd9Sstevel@tonic-gate if (!page_trylock(pp, se)) { 9747c478bd9Sstevel@tonic-gate VM_STAT_ADD(page_lookup_nowait_cnt[3]); 9757c478bd9Sstevel@tonic-gate pp = NULL; 9767c478bd9Sstevel@tonic-gate } else { 9777c478bd9Sstevel@tonic-gate VM_STAT_ADD(page_lookup_nowait_cnt[4]); 9787c478bd9Sstevel@tonic-gate /* 9797c478bd9Sstevel@tonic-gate * See the comment in page_lookup() 9807c478bd9Sstevel@tonic-gate */ 9817c478bd9Sstevel@tonic-gate if (((volatile struct vnode *)(pp->p_vnode) != vp) || 9827c478bd9Sstevel@tonic-gate ((u_offset_t)(pp->p_offset) != off)) { 9837c478bd9Sstevel@tonic-gate VM_STAT_ADD(page_lookup_nowait_cnt[5]); 9847c478bd9Sstevel@tonic-gate if (locked) { 9857c478bd9Sstevel@tonic-gate panic("page_lookup_nowait %p", 9867c478bd9Sstevel@tonic-gate (void *)pp); 9877c478bd9Sstevel@tonic-gate /*NOTREACHED*/ 9887c478bd9Sstevel@tonic-gate } 9897c478bd9Sstevel@tonic-gate page_unlock(pp); 9907c478bd9Sstevel@tonic-gate goto top; 9917c478bd9Sstevel@tonic-gate } 9927c478bd9Sstevel@tonic-gate if (PP_ISFREE(pp)) { 9937c478bd9Sstevel@tonic-gate VM_STAT_ADD(page_lookup_nowait_cnt[6]); 9947c478bd9Sstevel@tonic-gate page_unlock(pp); 9957c478bd9Sstevel@tonic-gate pp = NULL; 9967c478bd9Sstevel@tonic-gate } 9977c478bd9Sstevel@tonic-gate } 9987c478bd9Sstevel@tonic-gate } 9997c478bd9Sstevel@tonic-gate if (locked) { 10007c478bd9Sstevel@tonic-gate VM_STAT_ADD(page_lookup_nowait_cnt[7]); 10017c478bd9Sstevel@tonic-gate mutex_exit(phm); 10027c478bd9Sstevel@tonic-gate } 10037c478bd9Sstevel@tonic-gate 10047c478bd9Sstevel@tonic-gate ASSERT(pp ? PAGE_LOCKED_SE(pp, se) : 1); 10057c478bd9Sstevel@tonic-gate 10067c478bd9Sstevel@tonic-gate return (pp); 10077c478bd9Sstevel@tonic-gate } 10087c478bd9Sstevel@tonic-gate 10097c478bd9Sstevel@tonic-gate /* 10107c478bd9Sstevel@tonic-gate * Search the hash list for a page with the specified [vp, off] 10117c478bd9Sstevel@tonic-gate * that is known to exist and is already locked. This routine 10127c478bd9Sstevel@tonic-gate * is typically used by segment SOFTUNLOCK routines. 10137c478bd9Sstevel@tonic-gate */ 10147c478bd9Sstevel@tonic-gate page_t * 10157c478bd9Sstevel@tonic-gate page_find(vnode_t *vp, u_offset_t off) 10167c478bd9Sstevel@tonic-gate { 10177c478bd9Sstevel@tonic-gate page_t *pp; 10187c478bd9Sstevel@tonic-gate kmutex_t *phm; 10197c478bd9Sstevel@tonic-gate ulong_t index; 10207c478bd9Sstevel@tonic-gate 10217c478bd9Sstevel@tonic-gate ASSERT(MUTEX_NOT_HELD(page_vnode_mutex(vp))); 10227c478bd9Sstevel@tonic-gate VM_STAT_ADD(page_find_cnt); 10237c478bd9Sstevel@tonic-gate 10247c478bd9Sstevel@tonic-gate index = PAGE_HASH_FUNC(vp, off); 10257c478bd9Sstevel@tonic-gate phm = PAGE_HASH_MUTEX(index); 10267c478bd9Sstevel@tonic-gate 10277c478bd9Sstevel@tonic-gate mutex_enter(phm); 10287c478bd9Sstevel@tonic-gate PAGE_HASH_SEARCH(index, pp, vp, off); 10297c478bd9Sstevel@tonic-gate mutex_exit(phm); 10307c478bd9Sstevel@tonic-gate 10314fc2445aSelowe ASSERT(pp == NULL || PAGE_LOCKED(pp) || panicstr); 10327c478bd9Sstevel@tonic-gate return (pp); 10337c478bd9Sstevel@tonic-gate } 10347c478bd9Sstevel@tonic-gate 10357c478bd9Sstevel@tonic-gate /* 10367c478bd9Sstevel@tonic-gate * Determine whether a page with the specified [vp, off] 10377c478bd9Sstevel@tonic-gate * currently exists in the system. Obviously this should 10387c478bd9Sstevel@tonic-gate * only be considered as a hint since nothing prevents the 10397c478bd9Sstevel@tonic-gate * page from disappearing or appearing immediately after 10407c478bd9Sstevel@tonic-gate * the return from this routine. Subsequently, we don't 10417c478bd9Sstevel@tonic-gate * even bother to lock the list. 10427c478bd9Sstevel@tonic-gate */ 10437c478bd9Sstevel@tonic-gate page_t * 10447c478bd9Sstevel@tonic-gate page_exists(vnode_t *vp, u_offset_t off) 10457c478bd9Sstevel@tonic-gate { 10467c478bd9Sstevel@tonic-gate page_t *pp; 10477c478bd9Sstevel@tonic-gate ulong_t index; 10487c478bd9Sstevel@tonic-gate 10497c478bd9Sstevel@tonic-gate ASSERT(MUTEX_NOT_HELD(page_vnode_mutex(vp))); 10507c478bd9Sstevel@tonic-gate VM_STAT_ADD(page_exists_cnt); 10517c478bd9Sstevel@tonic-gate 10527c478bd9Sstevel@tonic-gate index = PAGE_HASH_FUNC(vp, off); 10537c478bd9Sstevel@tonic-gate PAGE_HASH_SEARCH(index, pp, vp, off); 10547c478bd9Sstevel@tonic-gate 10557c478bd9Sstevel@tonic-gate return (pp); 10567c478bd9Sstevel@tonic-gate } 10577c478bd9Sstevel@tonic-gate 10587c478bd9Sstevel@tonic-gate /* 10597c478bd9Sstevel@tonic-gate * Determine if physically contiguous pages exist for [vp, off] - [vp, off + 10607c478bd9Sstevel@tonic-gate * page_size(szc)) range. if they exist and ppa is not NULL fill ppa array 10617c478bd9Sstevel@tonic-gate * with these pages locked SHARED. If necessary reclaim pages from 10627c478bd9Sstevel@tonic-gate * freelist. Return 1 if contiguous pages exist and 0 otherwise. 10637c478bd9Sstevel@tonic-gate * 10647c478bd9Sstevel@tonic-gate * If we fail to lock pages still return 1 if pages exist and contiguous. 10657c478bd9Sstevel@tonic-gate * But in this case return value is just a hint. ppa array won't be filled. 10667c478bd9Sstevel@tonic-gate * Caller should initialize ppa[0] as NULL to distinguish return value. 10677c478bd9Sstevel@tonic-gate * 10687c478bd9Sstevel@tonic-gate * Returns 0 if pages don't exist or not physically contiguous. 10697c478bd9Sstevel@tonic-gate * 10707c478bd9Sstevel@tonic-gate * This routine doesn't work for anonymous(swapfs) pages. 10717c478bd9Sstevel@tonic-gate */ 10727c478bd9Sstevel@tonic-gate int 10737c478bd9Sstevel@tonic-gate page_exists_physcontig(vnode_t *vp, u_offset_t off, uint_t szc, page_t *ppa[]) 10747c478bd9Sstevel@tonic-gate { 10757c478bd9Sstevel@tonic-gate pgcnt_t pages; 10767c478bd9Sstevel@tonic-gate pfn_t pfn; 10777c478bd9Sstevel@tonic-gate page_t *rootpp; 10787c478bd9Sstevel@tonic-gate pgcnt_t i; 10797c478bd9Sstevel@tonic-gate pgcnt_t j; 10807c478bd9Sstevel@tonic-gate u_offset_t save_off = off; 10817c478bd9Sstevel@tonic-gate ulong_t index; 10827c478bd9Sstevel@tonic-gate kmutex_t *phm; 10837c478bd9Sstevel@tonic-gate page_t *pp; 10847c478bd9Sstevel@tonic-gate uint_t pszc; 10857c478bd9Sstevel@tonic-gate int loopcnt = 0; 10867c478bd9Sstevel@tonic-gate 10877c478bd9Sstevel@tonic-gate ASSERT(szc != 0); 10887c478bd9Sstevel@tonic-gate ASSERT(vp != NULL); 10897c478bd9Sstevel@tonic-gate ASSERT(!IS_SWAPFSVP(vp)); 1090ad23a2dbSjohansen ASSERT(!VN_ISKAS(vp)); 10917c478bd9Sstevel@tonic-gate 10927c478bd9Sstevel@tonic-gate again: 10937c478bd9Sstevel@tonic-gate if (++loopcnt > 3) { 10947c478bd9Sstevel@tonic-gate VM_STAT_ADD(page_exphcontg[0]); 10957c478bd9Sstevel@tonic-gate return (0); 10967c478bd9Sstevel@tonic-gate } 10977c478bd9Sstevel@tonic-gate 10987c478bd9Sstevel@tonic-gate index = PAGE_HASH_FUNC(vp, off); 10997c478bd9Sstevel@tonic-gate phm = PAGE_HASH_MUTEX(index); 11007c478bd9Sstevel@tonic-gate 11017c478bd9Sstevel@tonic-gate mutex_enter(phm); 11027c478bd9Sstevel@tonic-gate PAGE_HASH_SEARCH(index, pp, vp, off); 11037c478bd9Sstevel@tonic-gate mutex_exit(phm); 11047c478bd9Sstevel@tonic-gate 11057c478bd9Sstevel@tonic-gate VM_STAT_ADD(page_exphcontg[1]); 11067c478bd9Sstevel@tonic-gate 11077c478bd9Sstevel@tonic-gate if (pp == NULL) { 11087c478bd9Sstevel@tonic-gate VM_STAT_ADD(page_exphcontg[2]); 11097c478bd9Sstevel@tonic-gate return (0); 11107c478bd9Sstevel@tonic-gate } 11117c478bd9Sstevel@tonic-gate 11127c478bd9Sstevel@tonic-gate pages = page_get_pagecnt(szc); 11137c478bd9Sstevel@tonic-gate rootpp = pp; 11147c478bd9Sstevel@tonic-gate pfn = rootpp->p_pagenum; 11157c478bd9Sstevel@tonic-gate 11167c478bd9Sstevel@tonic-gate if ((pszc = pp->p_szc) >= szc && ppa != NULL) { 11177c478bd9Sstevel@tonic-gate VM_STAT_ADD(page_exphcontg[3]); 11187c478bd9Sstevel@tonic-gate if (!page_trylock(pp, SE_SHARED)) { 11197c478bd9Sstevel@tonic-gate VM_STAT_ADD(page_exphcontg[4]); 11207c478bd9Sstevel@tonic-gate return (1); 11217c478bd9Sstevel@tonic-gate } 11227c478bd9Sstevel@tonic-gate if (pp->p_szc != pszc || pp->p_vnode != vp || 11237c478bd9Sstevel@tonic-gate pp->p_offset != off) { 11247c478bd9Sstevel@tonic-gate VM_STAT_ADD(page_exphcontg[5]); 11257c478bd9Sstevel@tonic-gate page_unlock(pp); 11267c478bd9Sstevel@tonic-gate off = save_off; 11277c478bd9Sstevel@tonic-gate goto again; 11287c478bd9Sstevel@tonic-gate } 11297c478bd9Sstevel@tonic-gate /* 11307c478bd9Sstevel@tonic-gate * szc was non zero and vnode and offset matched after we 11317c478bd9Sstevel@tonic-gate * locked the page it means it can't become free on us. 11327c478bd9Sstevel@tonic-gate */ 11337c478bd9Sstevel@tonic-gate ASSERT(!PP_ISFREE(pp)); 11347c478bd9Sstevel@tonic-gate if (!IS_P2ALIGNED(pfn, pages)) { 11357c478bd9Sstevel@tonic-gate page_unlock(pp); 11367c478bd9Sstevel@tonic-gate return (0); 11377c478bd9Sstevel@tonic-gate } 11387c478bd9Sstevel@tonic-gate ppa[0] = pp; 11397c478bd9Sstevel@tonic-gate pp++; 11407c478bd9Sstevel@tonic-gate off += PAGESIZE; 11417c478bd9Sstevel@tonic-gate pfn++; 11427c478bd9Sstevel@tonic-gate for (i = 1; i < pages; i++, pp++, off += PAGESIZE, pfn++) { 11437c478bd9Sstevel@tonic-gate if (!page_trylock(pp, SE_SHARED)) { 11447c478bd9Sstevel@tonic-gate VM_STAT_ADD(page_exphcontg[6]); 11457c478bd9Sstevel@tonic-gate pp--; 11467c478bd9Sstevel@tonic-gate while (i-- > 0) { 11477c478bd9Sstevel@tonic-gate page_unlock(pp); 11487c478bd9Sstevel@tonic-gate pp--; 11497c478bd9Sstevel@tonic-gate } 11507c478bd9Sstevel@tonic-gate ppa[0] = NULL; 11517c478bd9Sstevel@tonic-gate return (1); 11527c478bd9Sstevel@tonic-gate } 11537c478bd9Sstevel@tonic-gate if (pp->p_szc != pszc) { 11547c478bd9Sstevel@tonic-gate VM_STAT_ADD(page_exphcontg[7]); 11557c478bd9Sstevel@tonic-gate page_unlock(pp); 11567c478bd9Sstevel@tonic-gate pp--; 11577c478bd9Sstevel@tonic-gate while (i-- > 0) { 11587c478bd9Sstevel@tonic-gate page_unlock(pp); 11597c478bd9Sstevel@tonic-gate pp--; 11607c478bd9Sstevel@tonic-gate } 11617c478bd9Sstevel@tonic-gate ppa[0] = NULL; 11627c478bd9Sstevel@tonic-gate off = save_off; 11637c478bd9Sstevel@tonic-gate goto again; 11647c478bd9Sstevel@tonic-gate } 11657c478bd9Sstevel@tonic-gate /* 11667c478bd9Sstevel@tonic-gate * szc the same as for previous already locked pages 11677c478bd9Sstevel@tonic-gate * with right identity. Since this page had correct 11687c478bd9Sstevel@tonic-gate * szc after we locked it can't get freed or destroyed 11697c478bd9Sstevel@tonic-gate * and therefore must have the expected identity. 11707c478bd9Sstevel@tonic-gate */ 11717c478bd9Sstevel@tonic-gate ASSERT(!PP_ISFREE(pp)); 11727c478bd9Sstevel@tonic-gate if (pp->p_vnode != vp || 11737c478bd9Sstevel@tonic-gate pp->p_offset != off) { 11747c478bd9Sstevel@tonic-gate panic("page_exists_physcontig: " 11757c478bd9Sstevel@tonic-gate "large page identity doesn't match"); 11767c478bd9Sstevel@tonic-gate } 11777c478bd9Sstevel@tonic-gate ppa[i] = pp; 11787c478bd9Sstevel@tonic-gate ASSERT(pp->p_pagenum == pfn); 11797c478bd9Sstevel@tonic-gate } 11807c478bd9Sstevel@tonic-gate VM_STAT_ADD(page_exphcontg[8]); 11817c478bd9Sstevel@tonic-gate ppa[pages] = NULL; 11827c478bd9Sstevel@tonic-gate return (1); 11837c478bd9Sstevel@tonic-gate } else if (pszc >= szc) { 11847c478bd9Sstevel@tonic-gate VM_STAT_ADD(page_exphcontg[9]); 11857c478bd9Sstevel@tonic-gate if (!IS_P2ALIGNED(pfn, pages)) { 11867c478bd9Sstevel@tonic-gate return (0); 11877c478bd9Sstevel@tonic-gate } 11887c478bd9Sstevel@tonic-gate return (1); 11897c478bd9Sstevel@tonic-gate } 11907c478bd9Sstevel@tonic-gate 11917c478bd9Sstevel@tonic-gate if (!IS_P2ALIGNED(pfn, pages)) { 11927c478bd9Sstevel@tonic-gate VM_STAT_ADD(page_exphcontg[10]); 11937c478bd9Sstevel@tonic-gate return (0); 11947c478bd9Sstevel@tonic-gate } 11957c478bd9Sstevel@tonic-gate 11967c478bd9Sstevel@tonic-gate if (page_numtomemseg_nolock(pfn) != 11977c478bd9Sstevel@tonic-gate page_numtomemseg_nolock(pfn + pages - 1)) { 11987c478bd9Sstevel@tonic-gate VM_STAT_ADD(page_exphcontg[11]); 11997c478bd9Sstevel@tonic-gate return (0); 12007c478bd9Sstevel@tonic-gate } 12017c478bd9Sstevel@tonic-gate 12027c478bd9Sstevel@tonic-gate /* 12037c478bd9Sstevel@tonic-gate * We loop up 4 times across pages to promote page size. 12047c478bd9Sstevel@tonic-gate * We're extra cautious to promote page size atomically with respect 12057c478bd9Sstevel@tonic-gate * to everybody else. But we can probably optimize into 1 loop if 12067c478bd9Sstevel@tonic-gate * this becomes an issue. 12077c478bd9Sstevel@tonic-gate */ 12087c478bd9Sstevel@tonic-gate 12097c478bd9Sstevel@tonic-gate for (i = 0; i < pages; i++, pp++, off += PAGESIZE, pfn++) { 12107c478bd9Sstevel@tonic-gate ASSERT(pp->p_pagenum == pfn); 12117c478bd9Sstevel@tonic-gate if (!page_trylock(pp, SE_EXCL)) { 12127c478bd9Sstevel@tonic-gate VM_STAT_ADD(page_exphcontg[12]); 12137c478bd9Sstevel@tonic-gate break; 12147c478bd9Sstevel@tonic-gate } 12157c478bd9Sstevel@tonic-gate if (pp->p_vnode != vp || 12167c478bd9Sstevel@tonic-gate pp->p_offset != off) { 12177c478bd9Sstevel@tonic-gate VM_STAT_ADD(page_exphcontg[13]); 12187c478bd9Sstevel@tonic-gate page_unlock(pp); 12197c478bd9Sstevel@tonic-gate break; 12207c478bd9Sstevel@tonic-gate } 12217c478bd9Sstevel@tonic-gate if (pp->p_szc >= szc) { 12227c478bd9Sstevel@tonic-gate ASSERT(i == 0); 12237c478bd9Sstevel@tonic-gate page_unlock(pp); 12247c478bd9Sstevel@tonic-gate off = save_off; 12257c478bd9Sstevel@tonic-gate goto again; 12267c478bd9Sstevel@tonic-gate } 12277c478bd9Sstevel@tonic-gate } 12287c478bd9Sstevel@tonic-gate 12297c478bd9Sstevel@tonic-gate if (i != pages) { 12307c478bd9Sstevel@tonic-gate VM_STAT_ADD(page_exphcontg[14]); 12317c478bd9Sstevel@tonic-gate --pp; 12327c478bd9Sstevel@tonic-gate while (i-- > 0) { 12337c478bd9Sstevel@tonic-gate page_unlock(pp); 12347c478bd9Sstevel@tonic-gate --pp; 12357c478bd9Sstevel@tonic-gate } 12367c478bd9Sstevel@tonic-gate return (0); 12377c478bd9Sstevel@tonic-gate } 12387c478bd9Sstevel@tonic-gate 12397c478bd9Sstevel@tonic-gate pp = rootpp; 12407c478bd9Sstevel@tonic-gate for (i = 0; i < pages; i++, pp++) { 12417c478bd9Sstevel@tonic-gate if (PP_ISFREE(pp)) { 12427c478bd9Sstevel@tonic-gate VM_STAT_ADD(page_exphcontg[15]); 12437c478bd9Sstevel@tonic-gate ASSERT(!PP_ISAGED(pp)); 12447c478bd9Sstevel@tonic-gate ASSERT(pp->p_szc == 0); 12457c478bd9Sstevel@tonic-gate if (!page_reclaim(pp, NULL)) { 12467c478bd9Sstevel@tonic-gate break; 12477c478bd9Sstevel@tonic-gate } 12487c478bd9Sstevel@tonic-gate } else { 12497c478bd9Sstevel@tonic-gate ASSERT(pp->p_szc < szc); 12507c478bd9Sstevel@tonic-gate VM_STAT_ADD(page_exphcontg[16]); 12517c478bd9Sstevel@tonic-gate (void) hat_pageunload(pp, HAT_FORCE_PGUNLOAD); 12527c478bd9Sstevel@tonic-gate } 12537c478bd9Sstevel@tonic-gate } 12547c478bd9Sstevel@tonic-gate if (i < pages) { 12557c478bd9Sstevel@tonic-gate VM_STAT_ADD(page_exphcontg[17]); 12567c478bd9Sstevel@tonic-gate /* 12577c478bd9Sstevel@tonic-gate * page_reclaim failed because we were out of memory. 12587c478bd9Sstevel@tonic-gate * drop the rest of the locks and return because this page 12597c478bd9Sstevel@tonic-gate * must be already reallocated anyway. 12607c478bd9Sstevel@tonic-gate */ 12617c478bd9Sstevel@tonic-gate pp = rootpp; 12627c478bd9Sstevel@tonic-gate for (j = 0; j < pages; j++, pp++) { 12637c478bd9Sstevel@tonic-gate if (j != i) { 12647c478bd9Sstevel@tonic-gate page_unlock(pp); 12657c478bd9Sstevel@tonic-gate } 12667c478bd9Sstevel@tonic-gate } 12677c478bd9Sstevel@tonic-gate return (0); 12687c478bd9Sstevel@tonic-gate } 12697c478bd9Sstevel@tonic-gate 12707c478bd9Sstevel@tonic-gate off = save_off; 12717c478bd9Sstevel@tonic-gate pp = rootpp; 12727c478bd9Sstevel@tonic-gate for (i = 0; i < pages; i++, pp++, off += PAGESIZE) { 12737c478bd9Sstevel@tonic-gate ASSERT(PAGE_EXCL(pp)); 12747c478bd9Sstevel@tonic-gate ASSERT(!PP_ISFREE(pp)); 12757c478bd9Sstevel@tonic-gate ASSERT(!hat_page_is_mapped(pp)); 12767c478bd9Sstevel@tonic-gate ASSERT(pp->p_vnode == vp); 12777c478bd9Sstevel@tonic-gate ASSERT(pp->p_offset == off); 12787c478bd9Sstevel@tonic-gate pp->p_szc = szc; 12797c478bd9Sstevel@tonic-gate } 12807c478bd9Sstevel@tonic-gate pp = rootpp; 12817c478bd9Sstevel@tonic-gate for (i = 0; i < pages; i++, pp++) { 12827c478bd9Sstevel@tonic-gate if (ppa == NULL) { 12837c478bd9Sstevel@tonic-gate page_unlock(pp); 12847c478bd9Sstevel@tonic-gate } else { 12857c478bd9Sstevel@tonic-gate ppa[i] = pp; 12867c478bd9Sstevel@tonic-gate page_downgrade(ppa[i]); 12877c478bd9Sstevel@tonic-gate } 12887c478bd9Sstevel@tonic-gate } 12897c478bd9Sstevel@tonic-gate if (ppa != NULL) { 12907c478bd9Sstevel@tonic-gate ppa[pages] = NULL; 12917c478bd9Sstevel@tonic-gate } 12927c478bd9Sstevel@tonic-gate VM_STAT_ADD(page_exphcontg[18]); 12937c478bd9Sstevel@tonic-gate ASSERT(vp->v_pages != NULL); 12947c478bd9Sstevel@tonic-gate return (1); 12957c478bd9Sstevel@tonic-gate } 12967c478bd9Sstevel@tonic-gate 12977c478bd9Sstevel@tonic-gate /* 12987c478bd9Sstevel@tonic-gate * Determine whether a page with the specified [vp, off] 12997c478bd9Sstevel@tonic-gate * currently exists in the system and if so return its 13007c478bd9Sstevel@tonic-gate * size code. Obviously this should only be considered as 13017c478bd9Sstevel@tonic-gate * a hint since nothing prevents the page from disappearing 13027c478bd9Sstevel@tonic-gate * or appearing immediately after the return from this routine. 13037c478bd9Sstevel@tonic-gate */ 13047c478bd9Sstevel@tonic-gate int 13057c478bd9Sstevel@tonic-gate page_exists_forreal(vnode_t *vp, u_offset_t off, uint_t *szc) 13067c478bd9Sstevel@tonic-gate { 13077c478bd9Sstevel@tonic-gate page_t *pp; 13087c478bd9Sstevel@tonic-gate kmutex_t *phm; 13097c478bd9Sstevel@tonic-gate ulong_t index; 13107c478bd9Sstevel@tonic-gate int rc = 0; 13117c478bd9Sstevel@tonic-gate 13127c478bd9Sstevel@tonic-gate ASSERT(MUTEX_NOT_HELD(page_vnode_mutex(vp))); 13137c478bd9Sstevel@tonic-gate ASSERT(szc != NULL); 13147c478bd9Sstevel@tonic-gate VM_STAT_ADD(page_exists_forreal_cnt); 13157c478bd9Sstevel@tonic-gate 13167c478bd9Sstevel@tonic-gate index = PAGE_HASH_FUNC(vp, off); 13177c478bd9Sstevel@tonic-gate phm = PAGE_HASH_MUTEX(index); 13187c478bd9Sstevel@tonic-gate 13197c478bd9Sstevel@tonic-gate mutex_enter(phm); 13207c478bd9Sstevel@tonic-gate PAGE_HASH_SEARCH(index, pp, vp, off); 13217c478bd9Sstevel@tonic-gate if (pp != NULL) { 13227c478bd9Sstevel@tonic-gate *szc = pp->p_szc; 13237c478bd9Sstevel@tonic-gate rc = 1; 13247c478bd9Sstevel@tonic-gate } 13257c478bd9Sstevel@tonic-gate mutex_exit(phm); 13267c478bd9Sstevel@tonic-gate return (rc); 13277c478bd9Sstevel@tonic-gate } 13287c478bd9Sstevel@tonic-gate 13297c478bd9Sstevel@tonic-gate /* wakeup threads waiting for pages in page_create_get_something() */ 13307c478bd9Sstevel@tonic-gate void 13317c478bd9Sstevel@tonic-gate wakeup_pcgs(void) 13327c478bd9Sstevel@tonic-gate { 13337c478bd9Sstevel@tonic-gate if (!CV_HAS_WAITERS(&pcgs_cv)) 13347c478bd9Sstevel@tonic-gate return; 13357c478bd9Sstevel@tonic-gate cv_broadcast(&pcgs_cv); 13367c478bd9Sstevel@tonic-gate } 13377c478bd9Sstevel@tonic-gate 13387c478bd9Sstevel@tonic-gate /* 13397c478bd9Sstevel@tonic-gate * 'freemem' is used all over the kernel as an indication of how many 13407c478bd9Sstevel@tonic-gate * pages are free (either on the cache list or on the free page list) 13417c478bd9Sstevel@tonic-gate * in the system. In very few places is a really accurate 'freemem' 13427c478bd9Sstevel@tonic-gate * needed. To avoid contention of the lock protecting a the 13437c478bd9Sstevel@tonic-gate * single freemem, it was spread out into NCPU buckets. Set_freemem 13447c478bd9Sstevel@tonic-gate * sets freemem to the total of all NCPU buckets. It is called from 13457c478bd9Sstevel@tonic-gate * clock() on each TICK. 13467c478bd9Sstevel@tonic-gate */ 13477c478bd9Sstevel@tonic-gate void 13487c478bd9Sstevel@tonic-gate set_freemem() 13497c478bd9Sstevel@tonic-gate { 13507c478bd9Sstevel@tonic-gate struct pcf *p; 13517c478bd9Sstevel@tonic-gate ulong_t t; 13527c478bd9Sstevel@tonic-gate uint_t i; 13537c478bd9Sstevel@tonic-gate 13547c478bd9Sstevel@tonic-gate t = 0; 13557c478bd9Sstevel@tonic-gate p = pcf; 1356*06fb6a36Sdv for (i = 0; i < pcf_fanout; i++) { 13577c478bd9Sstevel@tonic-gate t += p->pcf_count; 13587c478bd9Sstevel@tonic-gate p++; 13597c478bd9Sstevel@tonic-gate } 13607c478bd9Sstevel@tonic-gate freemem = t; 13617c478bd9Sstevel@tonic-gate 13627c478bd9Sstevel@tonic-gate /* 13637c478bd9Sstevel@tonic-gate * Don't worry about grabbing mutex. It's not that 13647c478bd9Sstevel@tonic-gate * critical if we miss a tick or two. This is 13657c478bd9Sstevel@tonic-gate * where we wakeup possible delayers in 13667c478bd9Sstevel@tonic-gate * page_create_get_something(). 13677c478bd9Sstevel@tonic-gate */ 13687c478bd9Sstevel@tonic-gate wakeup_pcgs(); 13697c478bd9Sstevel@tonic-gate } 13707c478bd9Sstevel@tonic-gate 13717c478bd9Sstevel@tonic-gate ulong_t 13727c478bd9Sstevel@tonic-gate get_freemem() 13737c478bd9Sstevel@tonic-gate { 13747c478bd9Sstevel@tonic-gate struct pcf *p; 13757c478bd9Sstevel@tonic-gate ulong_t t; 13767c478bd9Sstevel@tonic-gate uint_t i; 13777c478bd9Sstevel@tonic-gate 13787c478bd9Sstevel@tonic-gate t = 0; 13797c478bd9Sstevel@tonic-gate p = pcf; 1380*06fb6a36Sdv for (i = 0; i < pcf_fanout; i++) { 13817c478bd9Sstevel@tonic-gate t += p->pcf_count; 13827c478bd9Sstevel@tonic-gate p++; 13837c478bd9Sstevel@tonic-gate } 13847c478bd9Sstevel@tonic-gate /* 13857c478bd9Sstevel@tonic-gate * We just calculated it, might as well set it. 13867c478bd9Sstevel@tonic-gate */ 13877c478bd9Sstevel@tonic-gate freemem = t; 13887c478bd9Sstevel@tonic-gate return (t); 13897c478bd9Sstevel@tonic-gate } 13907c478bd9Sstevel@tonic-gate 13917c478bd9Sstevel@tonic-gate /* 13927c478bd9Sstevel@tonic-gate * Acquire all of the page cache & free (pcf) locks. 13937c478bd9Sstevel@tonic-gate */ 13947c478bd9Sstevel@tonic-gate void 13957c478bd9Sstevel@tonic-gate pcf_acquire_all() 13967c478bd9Sstevel@tonic-gate { 13977c478bd9Sstevel@tonic-gate struct pcf *p; 13987c478bd9Sstevel@tonic-gate uint_t i; 13997c478bd9Sstevel@tonic-gate 14007c478bd9Sstevel@tonic-gate p = pcf; 1401*06fb6a36Sdv for (i = 0; i < pcf_fanout; i++) { 14027c478bd9Sstevel@tonic-gate mutex_enter(&p->pcf_lock); 14037c478bd9Sstevel@tonic-gate p++; 14047c478bd9Sstevel@tonic-gate } 14057c478bd9Sstevel@tonic-gate } 14067c478bd9Sstevel@tonic-gate 14077c478bd9Sstevel@tonic-gate /* 14087c478bd9Sstevel@tonic-gate * Release all the pcf_locks. 14097c478bd9Sstevel@tonic-gate */ 14107c478bd9Sstevel@tonic-gate void 14117c478bd9Sstevel@tonic-gate pcf_release_all() 14127c478bd9Sstevel@tonic-gate { 14137c478bd9Sstevel@tonic-gate struct pcf *p; 14147c478bd9Sstevel@tonic-gate uint_t i; 14157c478bd9Sstevel@tonic-gate 14167c478bd9Sstevel@tonic-gate p = pcf; 1417*06fb6a36Sdv for (i = 0; i < pcf_fanout; i++) { 14187c478bd9Sstevel@tonic-gate mutex_exit(&p->pcf_lock); 14197c478bd9Sstevel@tonic-gate p++; 14207c478bd9Sstevel@tonic-gate } 14217c478bd9Sstevel@tonic-gate } 14227c478bd9Sstevel@tonic-gate 14237c478bd9Sstevel@tonic-gate /* 14247c478bd9Sstevel@tonic-gate * Inform the VM system that we need some pages freed up. 14257c478bd9Sstevel@tonic-gate * Calls must be symmetric, e.g.: 14267c478bd9Sstevel@tonic-gate * 14277c478bd9Sstevel@tonic-gate * page_needfree(100); 14287c478bd9Sstevel@tonic-gate * wait a bit; 14297c478bd9Sstevel@tonic-gate * page_needfree(-100); 14307c478bd9Sstevel@tonic-gate */ 14317c478bd9Sstevel@tonic-gate void 14327c478bd9Sstevel@tonic-gate page_needfree(spgcnt_t npages) 14337c478bd9Sstevel@tonic-gate { 14347c478bd9Sstevel@tonic-gate mutex_enter(&new_freemem_lock); 14357c478bd9Sstevel@tonic-gate needfree += npages; 14367c478bd9Sstevel@tonic-gate mutex_exit(&new_freemem_lock); 14377c478bd9Sstevel@tonic-gate } 14387c478bd9Sstevel@tonic-gate 14397c478bd9Sstevel@tonic-gate /* 14407c478bd9Sstevel@tonic-gate * Throttle for page_create(): try to prevent freemem from dropping 14417c478bd9Sstevel@tonic-gate * below throttlefree. We can't provide a 100% guarantee because 14427c478bd9Sstevel@tonic-gate * KM_NOSLEEP allocations, page_reclaim(), and various other things 14437c478bd9Sstevel@tonic-gate * nibble away at the freelist. However, we can block all PG_WAIT 14447c478bd9Sstevel@tonic-gate * allocations until memory becomes available. The motivation is 14457c478bd9Sstevel@tonic-gate * that several things can fall apart when there's no free memory: 14467c478bd9Sstevel@tonic-gate * 14477c478bd9Sstevel@tonic-gate * (1) If pageout() needs memory to push a page, the system deadlocks. 14487c478bd9Sstevel@tonic-gate * 14497c478bd9Sstevel@tonic-gate * (2) By (broken) specification, timeout(9F) can neither fail nor 14507c478bd9Sstevel@tonic-gate * block, so it has no choice but to panic the system if it 14517c478bd9Sstevel@tonic-gate * cannot allocate a callout structure. 14527c478bd9Sstevel@tonic-gate * 14537c478bd9Sstevel@tonic-gate * (3) Like timeout(), ddi_set_callback() cannot fail and cannot block; 14547c478bd9Sstevel@tonic-gate * it panics if it cannot allocate a callback structure. 14557c478bd9Sstevel@tonic-gate * 14567c478bd9Sstevel@tonic-gate * (4) Untold numbers of third-party drivers have not yet been hardened 14577c478bd9Sstevel@tonic-gate * against KM_NOSLEEP and/or allocb() failures; they simply assume 14587c478bd9Sstevel@tonic-gate * success and panic the system with a data fault on failure. 14597c478bd9Sstevel@tonic-gate * (The long-term solution to this particular problem is to ship 14607c478bd9Sstevel@tonic-gate * hostile fault-injecting DEBUG kernels with the DDK.) 14617c478bd9Sstevel@tonic-gate * 14627c478bd9Sstevel@tonic-gate * It is theoretically impossible to guarantee success of non-blocking 14637c478bd9Sstevel@tonic-gate * allocations, but in practice, this throttle is very hard to break. 14647c478bd9Sstevel@tonic-gate */ 14657c478bd9Sstevel@tonic-gate static int 14667c478bd9Sstevel@tonic-gate page_create_throttle(pgcnt_t npages, int flags) 14677c478bd9Sstevel@tonic-gate { 14687c478bd9Sstevel@tonic-gate ulong_t fm; 14697c478bd9Sstevel@tonic-gate uint_t i; 14707c478bd9Sstevel@tonic-gate pgcnt_t tf; /* effective value of throttlefree */ 14717c478bd9Sstevel@tonic-gate 14727c478bd9Sstevel@tonic-gate /* 14737c478bd9Sstevel@tonic-gate * Never deny pages when: 14747c478bd9Sstevel@tonic-gate * - it's a thread that cannot block [NOMEMWAIT()] 14757c478bd9Sstevel@tonic-gate * - the allocation cannot block and must not fail 14767c478bd9Sstevel@tonic-gate * - the allocation cannot block and is pageout dispensated 14777c478bd9Sstevel@tonic-gate */ 14787c478bd9Sstevel@tonic-gate if (NOMEMWAIT() || 14797c478bd9Sstevel@tonic-gate ((flags & (PG_WAIT | PG_PANIC)) == PG_PANIC) || 14807c478bd9Sstevel@tonic-gate ((flags & (PG_WAIT | PG_PUSHPAGE)) == PG_PUSHPAGE)) 14817c478bd9Sstevel@tonic-gate return (1); 14827c478bd9Sstevel@tonic-gate 14837c478bd9Sstevel@tonic-gate /* 14847c478bd9Sstevel@tonic-gate * If the allocation can't block, we look favorably upon it 14857c478bd9Sstevel@tonic-gate * unless we're below pageout_reserve. In that case we fail 14867c478bd9Sstevel@tonic-gate * the allocation because we want to make sure there are a few 14877c478bd9Sstevel@tonic-gate * pages available for pageout. 14887c478bd9Sstevel@tonic-gate */ 14897c478bd9Sstevel@tonic-gate if ((flags & PG_WAIT) == 0) 14907c478bd9Sstevel@tonic-gate return (freemem >= npages + pageout_reserve); 14917c478bd9Sstevel@tonic-gate 14927c478bd9Sstevel@tonic-gate /* Calculate the effective throttlefree value */ 14937c478bd9Sstevel@tonic-gate tf = throttlefree - 14947c478bd9Sstevel@tonic-gate ((flags & PG_PUSHPAGE) ? pageout_reserve : 0); 14957c478bd9Sstevel@tonic-gate 14967c478bd9Sstevel@tonic-gate cv_signal(&proc_pageout->p_cv); 14977c478bd9Sstevel@tonic-gate 149878b03d3aSkchow for (;;) { 149978b03d3aSkchow fm = 0; 15007c478bd9Sstevel@tonic-gate pcf_acquire_all(); 15017c478bd9Sstevel@tonic-gate mutex_enter(&new_freemem_lock); 1502*06fb6a36Sdv for (i = 0; i < pcf_fanout; i++) { 15037c478bd9Sstevel@tonic-gate fm += pcf[i].pcf_count; 15047c478bd9Sstevel@tonic-gate pcf[i].pcf_wait++; 15057c478bd9Sstevel@tonic-gate mutex_exit(&pcf[i].pcf_lock); 15067c478bd9Sstevel@tonic-gate } 15077c478bd9Sstevel@tonic-gate freemem = fm; 150878b03d3aSkchow if (freemem >= npages + tf) { 150978b03d3aSkchow mutex_exit(&new_freemem_lock); 151078b03d3aSkchow break; 151178b03d3aSkchow } 15127c478bd9Sstevel@tonic-gate needfree += npages; 15137c478bd9Sstevel@tonic-gate freemem_wait++; 15147c478bd9Sstevel@tonic-gate cv_wait(&freemem_cv, &new_freemem_lock); 15157c478bd9Sstevel@tonic-gate freemem_wait--; 15167c478bd9Sstevel@tonic-gate needfree -= npages; 15177c478bd9Sstevel@tonic-gate mutex_exit(&new_freemem_lock); 15187c478bd9Sstevel@tonic-gate } 15197c478bd9Sstevel@tonic-gate return (1); 15207c478bd9Sstevel@tonic-gate } 15217c478bd9Sstevel@tonic-gate 15227c478bd9Sstevel@tonic-gate /* 1523da6c28aaSamw * page_create_wait() is called to either coalesce pages from the 15247c478bd9Sstevel@tonic-gate * different pcf buckets or to wait because there simply are not 15257c478bd9Sstevel@tonic-gate * enough pages to satisfy the caller's request. 15267c478bd9Sstevel@tonic-gate * 15277c478bd9Sstevel@tonic-gate * Sadly, this is called from platform/vm/vm_machdep.c 15287c478bd9Sstevel@tonic-gate */ 15297c478bd9Sstevel@tonic-gate int 1530*06fb6a36Sdv page_create_wait(pgcnt_t npages, uint_t flags) 15317c478bd9Sstevel@tonic-gate { 15327c478bd9Sstevel@tonic-gate pgcnt_t total; 15337c478bd9Sstevel@tonic-gate uint_t i; 15347c478bd9Sstevel@tonic-gate struct pcf *p; 15357c478bd9Sstevel@tonic-gate 15367c478bd9Sstevel@tonic-gate /* 15377c478bd9Sstevel@tonic-gate * Wait until there are enough free pages to satisfy our 15387c478bd9Sstevel@tonic-gate * entire request. 15397c478bd9Sstevel@tonic-gate * We set needfree += npages before prodding pageout, to make sure 15407c478bd9Sstevel@tonic-gate * it does real work when npages > lotsfree > freemem. 15417c478bd9Sstevel@tonic-gate */ 15427c478bd9Sstevel@tonic-gate VM_STAT_ADD(page_create_not_enough); 15437c478bd9Sstevel@tonic-gate 15447c478bd9Sstevel@tonic-gate ASSERT(!kcage_on ? !(flags & PG_NORELOC) : 1); 15457c478bd9Sstevel@tonic-gate checkagain: 15467c478bd9Sstevel@tonic-gate if ((flags & PG_NORELOC) && 15477c478bd9Sstevel@tonic-gate kcage_freemem < kcage_throttlefree + npages) 15487c478bd9Sstevel@tonic-gate (void) kcage_create_throttle(npages, flags); 15497c478bd9Sstevel@tonic-gate 15507c478bd9Sstevel@tonic-gate if (freemem < npages + throttlefree) 15517c478bd9Sstevel@tonic-gate if (!page_create_throttle(npages, flags)) 15527c478bd9Sstevel@tonic-gate return (0); 15537c478bd9Sstevel@tonic-gate 1554*06fb6a36Sdv if (pcf_decrement_bucket(npages) || 1555*06fb6a36Sdv pcf_decrement_multiple(&total, npages, 0)) 1556*06fb6a36Sdv return (1); 15577c478bd9Sstevel@tonic-gate 15587c478bd9Sstevel@tonic-gate /* 15597c478bd9Sstevel@tonic-gate * All of the pcf locks are held, there are not enough pages 15607c478bd9Sstevel@tonic-gate * to satisfy the request (npages < total). 15617c478bd9Sstevel@tonic-gate * Be sure to acquire the new_freemem_lock before dropping 15627c478bd9Sstevel@tonic-gate * the pcf locks. This prevents dropping wakeups in page_free(). 15637c478bd9Sstevel@tonic-gate * The order is always pcf_lock then new_freemem_lock. 15647c478bd9Sstevel@tonic-gate * 15657c478bd9Sstevel@tonic-gate * Since we hold all the pcf locks, it is a good time to set freemem. 15667c478bd9Sstevel@tonic-gate * 15677c478bd9Sstevel@tonic-gate * If the caller does not want to wait, return now. 15687c478bd9Sstevel@tonic-gate * Else turn the pageout daemon loose to find something 15697c478bd9Sstevel@tonic-gate * and wait till it does. 15707c478bd9Sstevel@tonic-gate * 15717c478bd9Sstevel@tonic-gate */ 15727c478bd9Sstevel@tonic-gate freemem = total; 15737c478bd9Sstevel@tonic-gate 15747c478bd9Sstevel@tonic-gate if ((flags & PG_WAIT) == 0) { 15757c478bd9Sstevel@tonic-gate pcf_release_all(); 15767c478bd9Sstevel@tonic-gate 15777c478bd9Sstevel@tonic-gate TRACE_2(TR_FAC_VM, TR_PAGE_CREATE_NOMEM, 15787c478bd9Sstevel@tonic-gate "page_create_nomem:npages %ld freemem %ld", npages, freemem); 15797c478bd9Sstevel@tonic-gate return (0); 15807c478bd9Sstevel@tonic-gate } 15817c478bd9Sstevel@tonic-gate 15827c478bd9Sstevel@tonic-gate ASSERT(proc_pageout != NULL); 15837c478bd9Sstevel@tonic-gate cv_signal(&proc_pageout->p_cv); 15847c478bd9Sstevel@tonic-gate 15857c478bd9Sstevel@tonic-gate TRACE_2(TR_FAC_VM, TR_PAGE_CREATE_SLEEP_START, 15867c478bd9Sstevel@tonic-gate "page_create_sleep_start: freemem %ld needfree %ld", 15877c478bd9Sstevel@tonic-gate freemem, needfree); 15887c478bd9Sstevel@tonic-gate 15897c478bd9Sstevel@tonic-gate /* 15907c478bd9Sstevel@tonic-gate * We are going to wait. 15917c478bd9Sstevel@tonic-gate * We currently hold all of the pcf_locks, 15927c478bd9Sstevel@tonic-gate * get the new_freemem_lock (it protects freemem_wait), 15937c478bd9Sstevel@tonic-gate * before dropping the pcf_locks. 15947c478bd9Sstevel@tonic-gate */ 15957c478bd9Sstevel@tonic-gate mutex_enter(&new_freemem_lock); 15967c478bd9Sstevel@tonic-gate 15977c478bd9Sstevel@tonic-gate p = pcf; 1598*06fb6a36Sdv for (i = 0; i < pcf_fanout; i++) { 15997c478bd9Sstevel@tonic-gate p->pcf_wait++; 16007c478bd9Sstevel@tonic-gate mutex_exit(&p->pcf_lock); 16017c478bd9Sstevel@tonic-gate p++; 16027c478bd9Sstevel@tonic-gate } 16037c478bd9Sstevel@tonic-gate 16047c478bd9Sstevel@tonic-gate needfree += npages; 16057c478bd9Sstevel@tonic-gate freemem_wait++; 16067c478bd9Sstevel@tonic-gate 16077c478bd9Sstevel@tonic-gate cv_wait(&freemem_cv, &new_freemem_lock); 16087c478bd9Sstevel@tonic-gate 16097c478bd9Sstevel@tonic-gate freemem_wait--; 16107c478bd9Sstevel@tonic-gate needfree -= npages; 16117c478bd9Sstevel@tonic-gate 16127c478bd9Sstevel@tonic-gate mutex_exit(&new_freemem_lock); 16137c478bd9Sstevel@tonic-gate 16147c478bd9Sstevel@tonic-gate TRACE_2(TR_FAC_VM, TR_PAGE_CREATE_SLEEP_END, 16157c478bd9Sstevel@tonic-gate "page_create_sleep_end: freemem %ld needfree %ld", 16167c478bd9Sstevel@tonic-gate freemem, needfree); 16177c478bd9Sstevel@tonic-gate 16187c478bd9Sstevel@tonic-gate VM_STAT_ADD(page_create_not_enough_again); 16197c478bd9Sstevel@tonic-gate goto checkagain; 16207c478bd9Sstevel@tonic-gate } 16217c478bd9Sstevel@tonic-gate /* 16227c478bd9Sstevel@tonic-gate * A routine to do the opposite of page_create_wait(). 16237c478bd9Sstevel@tonic-gate */ 16247c478bd9Sstevel@tonic-gate void 16257c478bd9Sstevel@tonic-gate page_create_putback(spgcnt_t npages) 16267c478bd9Sstevel@tonic-gate { 16277c478bd9Sstevel@tonic-gate struct pcf *p; 16287c478bd9Sstevel@tonic-gate pgcnt_t lump; 16297c478bd9Sstevel@tonic-gate uint_t *which; 16307c478bd9Sstevel@tonic-gate 16317c478bd9Sstevel@tonic-gate /* 16327c478bd9Sstevel@tonic-gate * When a contiguous lump is broken up, we have to 16337c478bd9Sstevel@tonic-gate * deal with lots of pages (min 64) so lets spread 16347c478bd9Sstevel@tonic-gate * the wealth around. 16357c478bd9Sstevel@tonic-gate */ 1636*06fb6a36Sdv lump = roundup(npages, pcf_fanout) / pcf_fanout; 16377c478bd9Sstevel@tonic-gate freemem += npages; 16387c478bd9Sstevel@tonic-gate 1639*06fb6a36Sdv for (p = pcf; (npages > 0) && (p < &pcf[pcf_fanout]); p++) { 16407c478bd9Sstevel@tonic-gate which = &p->pcf_count; 16417c478bd9Sstevel@tonic-gate 16427c478bd9Sstevel@tonic-gate mutex_enter(&p->pcf_lock); 16437c478bd9Sstevel@tonic-gate 16447c478bd9Sstevel@tonic-gate if (p->pcf_block) { 16457c478bd9Sstevel@tonic-gate which = &p->pcf_reserve; 16467c478bd9Sstevel@tonic-gate } 16477c478bd9Sstevel@tonic-gate 16487c478bd9Sstevel@tonic-gate if (lump < npages) { 16497c478bd9Sstevel@tonic-gate *which += (uint_t)lump; 16507c478bd9Sstevel@tonic-gate npages -= lump; 16517c478bd9Sstevel@tonic-gate } else { 16527c478bd9Sstevel@tonic-gate *which += (uint_t)npages; 16537c478bd9Sstevel@tonic-gate npages = 0; 16547c478bd9Sstevel@tonic-gate } 16557c478bd9Sstevel@tonic-gate 16567c478bd9Sstevel@tonic-gate if (p->pcf_wait) { 16577c478bd9Sstevel@tonic-gate mutex_enter(&new_freemem_lock); 16587c478bd9Sstevel@tonic-gate /* 16597c478bd9Sstevel@tonic-gate * Check to see if some other thread 16607c478bd9Sstevel@tonic-gate * is actually waiting. Another bucket 16617c478bd9Sstevel@tonic-gate * may have woken it up by now. If there 16627c478bd9Sstevel@tonic-gate * are no waiters, then set our pcf_wait 16637c478bd9Sstevel@tonic-gate * count to zero to avoid coming in here 16647c478bd9Sstevel@tonic-gate * next time. 16657c478bd9Sstevel@tonic-gate */ 16667c478bd9Sstevel@tonic-gate if (freemem_wait) { 16677c478bd9Sstevel@tonic-gate if (npages > 1) { 16687c478bd9Sstevel@tonic-gate cv_broadcast(&freemem_cv); 16697c478bd9Sstevel@tonic-gate } else { 16707c478bd9Sstevel@tonic-gate cv_signal(&freemem_cv); 16717c478bd9Sstevel@tonic-gate } 16727c478bd9Sstevel@tonic-gate p->pcf_wait--; 16737c478bd9Sstevel@tonic-gate } else { 16747c478bd9Sstevel@tonic-gate p->pcf_wait = 0; 16757c478bd9Sstevel@tonic-gate } 16767c478bd9Sstevel@tonic-gate mutex_exit(&new_freemem_lock); 16777c478bd9Sstevel@tonic-gate } 16787c478bd9Sstevel@tonic-gate mutex_exit(&p->pcf_lock); 16797c478bd9Sstevel@tonic-gate } 16807c478bd9Sstevel@tonic-gate ASSERT(npages == 0); 16817c478bd9Sstevel@tonic-gate } 16827c478bd9Sstevel@tonic-gate 16837c478bd9Sstevel@tonic-gate /* 16847c478bd9Sstevel@tonic-gate * A helper routine for page_create_get_something. 16857c478bd9Sstevel@tonic-gate * The indenting got to deep down there. 16867c478bd9Sstevel@tonic-gate * Unblock the pcf counters. Any pages freed after 16877c478bd9Sstevel@tonic-gate * pcf_block got set are moved to pcf_count and 16887c478bd9Sstevel@tonic-gate * wakeups (cv_broadcast() or cv_signal()) are done as needed. 16897c478bd9Sstevel@tonic-gate */ 16907c478bd9Sstevel@tonic-gate static void 16917c478bd9Sstevel@tonic-gate pcgs_unblock(void) 16927c478bd9Sstevel@tonic-gate { 16937c478bd9Sstevel@tonic-gate int i; 16947c478bd9Sstevel@tonic-gate struct pcf *p; 16957c478bd9Sstevel@tonic-gate 16967c478bd9Sstevel@tonic-gate /* Update freemem while we're here. */ 16977c478bd9Sstevel@tonic-gate freemem = 0; 16987c478bd9Sstevel@tonic-gate p = pcf; 1699*06fb6a36Sdv for (i = 0; i < pcf_fanout; i++) { 17007c478bd9Sstevel@tonic-gate mutex_enter(&p->pcf_lock); 17017c478bd9Sstevel@tonic-gate ASSERT(p->pcf_count == 0); 17027c478bd9Sstevel@tonic-gate p->pcf_count = p->pcf_reserve; 17037c478bd9Sstevel@tonic-gate p->pcf_block = 0; 17047c478bd9Sstevel@tonic-gate freemem += p->pcf_count; 17057c478bd9Sstevel@tonic-gate if (p->pcf_wait) { 17067c478bd9Sstevel@tonic-gate mutex_enter(&new_freemem_lock); 17077c478bd9Sstevel@tonic-gate if (freemem_wait) { 17087c478bd9Sstevel@tonic-gate if (p->pcf_reserve > 1) { 17097c478bd9Sstevel@tonic-gate cv_broadcast(&freemem_cv); 17107c478bd9Sstevel@tonic-gate p->pcf_wait = 0; 17117c478bd9Sstevel@tonic-gate } else { 17127c478bd9Sstevel@tonic-gate cv_signal(&freemem_cv); 17137c478bd9Sstevel@tonic-gate p->pcf_wait--; 17147c478bd9Sstevel@tonic-gate } 17157c478bd9Sstevel@tonic-gate } else { 17167c478bd9Sstevel@tonic-gate p->pcf_wait = 0; 17177c478bd9Sstevel@tonic-gate } 17187c478bd9Sstevel@tonic-gate mutex_exit(&new_freemem_lock); 17197c478bd9Sstevel@tonic-gate } 17207c478bd9Sstevel@tonic-gate p->pcf_reserve = 0; 17217c478bd9Sstevel@tonic-gate mutex_exit(&p->pcf_lock); 17227c478bd9Sstevel@tonic-gate p++; 17237c478bd9Sstevel@tonic-gate } 17247c478bd9Sstevel@tonic-gate } 17257c478bd9Sstevel@tonic-gate 17267c478bd9Sstevel@tonic-gate /* 17277c478bd9Sstevel@tonic-gate * Called from page_create_va() when both the cache and free lists 17287c478bd9Sstevel@tonic-gate * have been checked once. 17297c478bd9Sstevel@tonic-gate * 17307c478bd9Sstevel@tonic-gate * Either returns a page or panics since the accounting was done 17317c478bd9Sstevel@tonic-gate * way before we got here. 17327c478bd9Sstevel@tonic-gate * 17337c478bd9Sstevel@tonic-gate * We don't come here often, so leave the accounting on permanently. 17347c478bd9Sstevel@tonic-gate */ 17357c478bd9Sstevel@tonic-gate 17367c478bd9Sstevel@tonic-gate #define MAX_PCGS 100 17377c478bd9Sstevel@tonic-gate 17387c478bd9Sstevel@tonic-gate #ifdef DEBUG 17397c478bd9Sstevel@tonic-gate #define PCGS_TRIES 100 17407c478bd9Sstevel@tonic-gate #else /* DEBUG */ 17417c478bd9Sstevel@tonic-gate #define PCGS_TRIES 10 17427c478bd9Sstevel@tonic-gate #endif /* DEBUG */ 17437c478bd9Sstevel@tonic-gate 17447c478bd9Sstevel@tonic-gate #ifdef VM_STATS 17457c478bd9Sstevel@tonic-gate uint_t pcgs_counts[PCGS_TRIES]; 17467c478bd9Sstevel@tonic-gate uint_t pcgs_too_many; 17477c478bd9Sstevel@tonic-gate uint_t pcgs_entered; 17487c478bd9Sstevel@tonic-gate uint_t pcgs_entered_noreloc; 17497c478bd9Sstevel@tonic-gate uint_t pcgs_locked; 17507c478bd9Sstevel@tonic-gate uint_t pcgs_cagelocked; 17517c478bd9Sstevel@tonic-gate #endif /* VM_STATS */ 17527c478bd9Sstevel@tonic-gate 17537c478bd9Sstevel@tonic-gate static page_t * 17547c478bd9Sstevel@tonic-gate page_create_get_something(vnode_t *vp, u_offset_t off, struct seg *seg, 17557c478bd9Sstevel@tonic-gate caddr_t vaddr, uint_t flags) 17567c478bd9Sstevel@tonic-gate { 17577c478bd9Sstevel@tonic-gate uint_t count; 17587c478bd9Sstevel@tonic-gate page_t *pp; 17597c478bd9Sstevel@tonic-gate uint_t locked, i; 17607c478bd9Sstevel@tonic-gate struct pcf *p; 17617c478bd9Sstevel@tonic-gate lgrp_t *lgrp; 17627c478bd9Sstevel@tonic-gate int cagelocked = 0; 17637c478bd9Sstevel@tonic-gate 17647c478bd9Sstevel@tonic-gate VM_STAT_ADD(pcgs_entered); 17657c478bd9Sstevel@tonic-gate 17667c478bd9Sstevel@tonic-gate /* 17677c478bd9Sstevel@tonic-gate * Tap any reserve freelists: if we fail now, we'll die 17687c478bd9Sstevel@tonic-gate * since the page(s) we're looking for have already been 17697c478bd9Sstevel@tonic-gate * accounted for. 17707c478bd9Sstevel@tonic-gate */ 17717c478bd9Sstevel@tonic-gate flags |= PG_PANIC; 17727c478bd9Sstevel@tonic-gate 17737c478bd9Sstevel@tonic-gate if ((flags & PG_NORELOC) != 0) { 17747c478bd9Sstevel@tonic-gate VM_STAT_ADD(pcgs_entered_noreloc); 17757c478bd9Sstevel@tonic-gate /* 17767c478bd9Sstevel@tonic-gate * Requests for free pages from critical threads 17777c478bd9Sstevel@tonic-gate * such as pageout still won't throttle here, but 17787c478bd9Sstevel@tonic-gate * we must try again, to give the cageout thread 17797c478bd9Sstevel@tonic-gate * another chance to catch up. Since we already 17807c478bd9Sstevel@tonic-gate * accounted for the pages, we had better get them 17817c478bd9Sstevel@tonic-gate * this time. 17827c478bd9Sstevel@tonic-gate * 17837c478bd9Sstevel@tonic-gate * N.B. All non-critical threads acquire the pcgs_cagelock 17847c478bd9Sstevel@tonic-gate * to serialize access to the freelists. This implements a 17857c478bd9Sstevel@tonic-gate * turnstile-type synchornization to avoid starvation of 17867c478bd9Sstevel@tonic-gate * critical requests for PG_NORELOC memory by non-critical 17877c478bd9Sstevel@tonic-gate * threads: all non-critical threads must acquire a 'ticket' 17887c478bd9Sstevel@tonic-gate * before passing through, which entails making sure 17897c478bd9Sstevel@tonic-gate * kcage_freemem won't fall below minfree prior to grabbing 17907c478bd9Sstevel@tonic-gate * pages from the freelists. 17917c478bd9Sstevel@tonic-gate */ 17927c478bd9Sstevel@tonic-gate if (kcage_create_throttle(1, flags) == KCT_NONCRIT) { 17937c478bd9Sstevel@tonic-gate mutex_enter(&pcgs_cagelock); 17947c478bd9Sstevel@tonic-gate cagelocked = 1; 17957c478bd9Sstevel@tonic-gate VM_STAT_ADD(pcgs_cagelocked); 17967c478bd9Sstevel@tonic-gate } 17977c478bd9Sstevel@tonic-gate } 17987c478bd9Sstevel@tonic-gate 17997c478bd9Sstevel@tonic-gate /* 18007c478bd9Sstevel@tonic-gate * Time to get serious. 18017c478bd9Sstevel@tonic-gate * We failed to get a `correctly colored' page from both the 18027c478bd9Sstevel@tonic-gate * free and cache lists. 18037c478bd9Sstevel@tonic-gate * We escalate in stage. 18047c478bd9Sstevel@tonic-gate * 18057c478bd9Sstevel@tonic-gate * First try both lists without worring about color. 18067c478bd9Sstevel@tonic-gate * 18077c478bd9Sstevel@tonic-gate * Then, grab all page accounting locks (ie. pcf[]) and 18087c478bd9Sstevel@tonic-gate * steal any pages that they have and set the pcf_block flag to 18097c478bd9Sstevel@tonic-gate * stop deletions from the lists. This will help because 18107c478bd9Sstevel@tonic-gate * a page can get added to the free list while we are looking 18117c478bd9Sstevel@tonic-gate * at the cache list, then another page could be added to the cache 18127c478bd9Sstevel@tonic-gate * list allowing the page on the free list to be removed as we 18137c478bd9Sstevel@tonic-gate * move from looking at the cache list to the free list. This 18147c478bd9Sstevel@tonic-gate * could happen over and over. We would never find the page 18157c478bd9Sstevel@tonic-gate * we have accounted for. 18167c478bd9Sstevel@tonic-gate * 18177c478bd9Sstevel@tonic-gate * Noreloc pages are a subset of the global (relocatable) page pool. 18187c478bd9Sstevel@tonic-gate * They are not tracked separately in the pcf bins, so it is 18197c478bd9Sstevel@tonic-gate * impossible to know when doing pcf accounting if the available 18207c478bd9Sstevel@tonic-gate * page(s) are noreloc pages or not. When looking for a noreloc page 18217c478bd9Sstevel@tonic-gate * it is quite easy to end up here even if the global (relocatable) 18227c478bd9Sstevel@tonic-gate * page pool has plenty of free pages but the noreloc pool is empty. 18237c478bd9Sstevel@tonic-gate * 18247c478bd9Sstevel@tonic-gate * When the noreloc pool is empty (or low), additional noreloc pages 18257c478bd9Sstevel@tonic-gate * are created by converting pages from the global page pool. This 18267c478bd9Sstevel@tonic-gate * process will stall during pcf accounting if the pcf bins are 18277c478bd9Sstevel@tonic-gate * already locked. Such is the case when a noreloc allocation is 18287c478bd9Sstevel@tonic-gate * looping here in page_create_get_something waiting for more noreloc 18297c478bd9Sstevel@tonic-gate * pages to appear. 18307c478bd9Sstevel@tonic-gate * 18317c478bd9Sstevel@tonic-gate * Short of adding a new field to the pcf bins to accurately track 18327c478bd9Sstevel@tonic-gate * the number of free noreloc pages, we instead do not grab the 18337c478bd9Sstevel@tonic-gate * pcgs_lock, do not set the pcf blocks and do not timeout when 18347c478bd9Sstevel@tonic-gate * allocating a noreloc page. This allows noreloc allocations to 18357c478bd9Sstevel@tonic-gate * loop without blocking global page pool allocations. 18367c478bd9Sstevel@tonic-gate * 18377c478bd9Sstevel@tonic-gate * NOTE: the behaviour of page_create_get_something has not changed 18387c478bd9Sstevel@tonic-gate * for the case of global page pool allocations. 18397c478bd9Sstevel@tonic-gate */ 18407c478bd9Sstevel@tonic-gate 18417c478bd9Sstevel@tonic-gate flags &= ~PG_MATCH_COLOR; 18427c478bd9Sstevel@tonic-gate locked = 0; 184307ad560dSkchow #if defined(__i386) || defined(__amd64) 1844843e1988Sjohnlev flags = page_create_update_flags_x86(flags); 18457c478bd9Sstevel@tonic-gate #endif 18467c478bd9Sstevel@tonic-gate 18477c478bd9Sstevel@tonic-gate lgrp = lgrp_mem_choose(seg, vaddr, PAGESIZE); 18487c478bd9Sstevel@tonic-gate 18497c478bd9Sstevel@tonic-gate for (count = 0; kcage_on || count < MAX_PCGS; count++) { 18507c478bd9Sstevel@tonic-gate pp = page_get_freelist(vp, off, seg, vaddr, PAGESIZE, 18517c478bd9Sstevel@tonic-gate flags, lgrp); 18527c478bd9Sstevel@tonic-gate if (pp == NULL) { 18537c478bd9Sstevel@tonic-gate pp = page_get_cachelist(vp, off, seg, vaddr, 18546e4dd838Smec flags, lgrp); 18557c478bd9Sstevel@tonic-gate } 18567c478bd9Sstevel@tonic-gate if (pp == NULL) { 18577c478bd9Sstevel@tonic-gate /* 18587c478bd9Sstevel@tonic-gate * Serialize. Don't fight with other pcgs(). 18597c478bd9Sstevel@tonic-gate */ 18607c478bd9Sstevel@tonic-gate if (!locked && (!kcage_on || !(flags & PG_NORELOC))) { 18617c478bd9Sstevel@tonic-gate mutex_enter(&pcgs_lock); 18627c478bd9Sstevel@tonic-gate VM_STAT_ADD(pcgs_locked); 18637c478bd9Sstevel@tonic-gate locked = 1; 18647c478bd9Sstevel@tonic-gate p = pcf; 1865*06fb6a36Sdv for (i = 0; i < pcf_fanout; i++) { 18667c478bd9Sstevel@tonic-gate mutex_enter(&p->pcf_lock); 18677c478bd9Sstevel@tonic-gate ASSERT(p->pcf_block == 0); 18687c478bd9Sstevel@tonic-gate p->pcf_block = 1; 18697c478bd9Sstevel@tonic-gate p->pcf_reserve = p->pcf_count; 18707c478bd9Sstevel@tonic-gate p->pcf_count = 0; 18717c478bd9Sstevel@tonic-gate mutex_exit(&p->pcf_lock); 18727c478bd9Sstevel@tonic-gate p++; 18737c478bd9Sstevel@tonic-gate } 18747c478bd9Sstevel@tonic-gate freemem = 0; 18757c478bd9Sstevel@tonic-gate } 18767c478bd9Sstevel@tonic-gate 18777c478bd9Sstevel@tonic-gate if (count) { 18787c478bd9Sstevel@tonic-gate /* 18797c478bd9Sstevel@tonic-gate * Since page_free() puts pages on 18807c478bd9Sstevel@tonic-gate * a list then accounts for it, we 18817c478bd9Sstevel@tonic-gate * just have to wait for page_free() 18827c478bd9Sstevel@tonic-gate * to unlock any page it was working 18837c478bd9Sstevel@tonic-gate * with. The page_lock()-page_reclaim() 18847c478bd9Sstevel@tonic-gate * path falls in the same boat. 18857c478bd9Sstevel@tonic-gate * 18867c478bd9Sstevel@tonic-gate * We don't need to check on the 18877c478bd9Sstevel@tonic-gate * PG_WAIT flag, we have already 18887c478bd9Sstevel@tonic-gate * accounted for the page we are 18897c478bd9Sstevel@tonic-gate * looking for in page_create_va(). 18907c478bd9Sstevel@tonic-gate * 18917c478bd9Sstevel@tonic-gate * We just wait a moment to let any 18927c478bd9Sstevel@tonic-gate * locked pages on the lists free up, 18937c478bd9Sstevel@tonic-gate * then continue around and try again. 18947c478bd9Sstevel@tonic-gate * 18957c478bd9Sstevel@tonic-gate * Will be awakened by set_freemem(). 18967c478bd9Sstevel@tonic-gate */ 18977c478bd9Sstevel@tonic-gate mutex_enter(&pcgs_wait_lock); 18987c478bd9Sstevel@tonic-gate cv_wait(&pcgs_cv, &pcgs_wait_lock); 18997c478bd9Sstevel@tonic-gate mutex_exit(&pcgs_wait_lock); 19007c478bd9Sstevel@tonic-gate } 19017c478bd9Sstevel@tonic-gate } else { 19027c478bd9Sstevel@tonic-gate #ifdef VM_STATS 19037c478bd9Sstevel@tonic-gate if (count >= PCGS_TRIES) { 19047c478bd9Sstevel@tonic-gate VM_STAT_ADD(pcgs_too_many); 19057c478bd9Sstevel@tonic-gate } else { 19067c478bd9Sstevel@tonic-gate VM_STAT_ADD(pcgs_counts[count]); 19077c478bd9Sstevel@tonic-gate } 19087c478bd9Sstevel@tonic-gate #endif 19097c478bd9Sstevel@tonic-gate if (locked) { 19107c478bd9Sstevel@tonic-gate pcgs_unblock(); 19117c478bd9Sstevel@tonic-gate mutex_exit(&pcgs_lock); 19127c478bd9Sstevel@tonic-gate } 19137c478bd9Sstevel@tonic-gate if (cagelocked) 19147c478bd9Sstevel@tonic-gate mutex_exit(&pcgs_cagelock); 19157c478bd9Sstevel@tonic-gate return (pp); 19167c478bd9Sstevel@tonic-gate } 19177c478bd9Sstevel@tonic-gate } 19187c478bd9Sstevel@tonic-gate /* 19197c478bd9Sstevel@tonic-gate * we go down holding the pcf locks. 19207c478bd9Sstevel@tonic-gate */ 19217c478bd9Sstevel@tonic-gate panic("no %spage found %d", 19227c478bd9Sstevel@tonic-gate ((flags & PG_NORELOC) ? "non-reloc " : ""), count); 19237c478bd9Sstevel@tonic-gate /*NOTREACHED*/ 19247c478bd9Sstevel@tonic-gate } 19257c478bd9Sstevel@tonic-gate 19267c478bd9Sstevel@tonic-gate /* 19277c478bd9Sstevel@tonic-gate * Create enough pages for "bytes" worth of data starting at 19287c478bd9Sstevel@tonic-gate * "off" in "vp". 19297c478bd9Sstevel@tonic-gate * 19307c478bd9Sstevel@tonic-gate * Where flag must be one of: 19317c478bd9Sstevel@tonic-gate * 19327c478bd9Sstevel@tonic-gate * PG_EXCL: Exclusive create (fail if any page already 19337c478bd9Sstevel@tonic-gate * exists in the page cache) which does not 19347c478bd9Sstevel@tonic-gate * wait for memory to become available. 19357c478bd9Sstevel@tonic-gate * 19367c478bd9Sstevel@tonic-gate * PG_WAIT: Non-exclusive create which can wait for 19377c478bd9Sstevel@tonic-gate * memory to become available. 19387c478bd9Sstevel@tonic-gate * 19397c478bd9Sstevel@tonic-gate * PG_PHYSCONTIG: Allocate physically contiguous pages. 19407c478bd9Sstevel@tonic-gate * (Not Supported) 19417c478bd9Sstevel@tonic-gate * 19427c478bd9Sstevel@tonic-gate * A doubly linked list of pages is returned to the caller. Each page 19437c478bd9Sstevel@tonic-gate * on the list has the "exclusive" (p_selock) lock and "iolock" (p_iolock) 19447c478bd9Sstevel@tonic-gate * lock. 19457c478bd9Sstevel@tonic-gate * 19467c478bd9Sstevel@tonic-gate * Unable to change the parameters to page_create() in a minor release, 19477c478bd9Sstevel@tonic-gate * we renamed page_create() to page_create_va(), changed all known calls 19487c478bd9Sstevel@tonic-gate * from page_create() to page_create_va(), and created this wrapper. 19497c478bd9Sstevel@tonic-gate * 19507c478bd9Sstevel@tonic-gate * Upon a major release, we should break compatibility by deleting this 19517c478bd9Sstevel@tonic-gate * wrapper, and replacing all the strings "page_create_va", with "page_create". 19527c478bd9Sstevel@tonic-gate * 19537c478bd9Sstevel@tonic-gate * NOTE: There is a copy of this interface as page_create_io() in 19547c478bd9Sstevel@tonic-gate * i86/vm/vm_machdep.c. Any bugs fixed here should be applied 19557c478bd9Sstevel@tonic-gate * there. 19567c478bd9Sstevel@tonic-gate */ 19577c478bd9Sstevel@tonic-gate page_t * 19587c478bd9Sstevel@tonic-gate page_create(vnode_t *vp, u_offset_t off, size_t bytes, uint_t flags) 19597c478bd9Sstevel@tonic-gate { 19607c478bd9Sstevel@tonic-gate caddr_t random_vaddr; 19617c478bd9Sstevel@tonic-gate struct seg kseg; 19627c478bd9Sstevel@tonic-gate 19637c478bd9Sstevel@tonic-gate #ifdef DEBUG 19647c478bd9Sstevel@tonic-gate cmn_err(CE_WARN, "Using deprecated interface page_create: caller %p", 19657c478bd9Sstevel@tonic-gate (void *)caller()); 19667c478bd9Sstevel@tonic-gate #endif 19677c478bd9Sstevel@tonic-gate 19687c478bd9Sstevel@tonic-gate random_vaddr = (caddr_t)(((uintptr_t)vp >> 7) ^ 19697c478bd9Sstevel@tonic-gate (uintptr_t)(off >> PAGESHIFT)); 19707c478bd9Sstevel@tonic-gate kseg.s_as = &kas; 19717c478bd9Sstevel@tonic-gate 19727c478bd9Sstevel@tonic-gate return (page_create_va(vp, off, bytes, flags, &kseg, random_vaddr)); 19737c478bd9Sstevel@tonic-gate } 19747c478bd9Sstevel@tonic-gate 19757c478bd9Sstevel@tonic-gate #ifdef DEBUG 19767c478bd9Sstevel@tonic-gate uint32_t pg_alloc_pgs_mtbf = 0; 19777c478bd9Sstevel@tonic-gate #endif 19787c478bd9Sstevel@tonic-gate 19797c478bd9Sstevel@tonic-gate /* 19807c478bd9Sstevel@tonic-gate * Used for large page support. It will attempt to allocate 19817c478bd9Sstevel@tonic-gate * a large page(s) off the freelist. 19827c478bd9Sstevel@tonic-gate * 19837c478bd9Sstevel@tonic-gate * Returns non zero on failure. 19847c478bd9Sstevel@tonic-gate */ 19857c478bd9Sstevel@tonic-gate int 1986e44bd21cSsusans page_alloc_pages(struct vnode *vp, struct seg *seg, caddr_t addr, 19872cb27123Saguzovsk page_t **basepp, page_t *ppa[], uint_t szc, int anypgsz, int pgflags) 19887c478bd9Sstevel@tonic-gate { 19897c478bd9Sstevel@tonic-gate pgcnt_t npgs, curnpgs, totpgs; 19907c478bd9Sstevel@tonic-gate size_t pgsz; 19917c478bd9Sstevel@tonic-gate page_t *pplist = NULL, *pp; 19927c478bd9Sstevel@tonic-gate int err = 0; 19937c478bd9Sstevel@tonic-gate lgrp_t *lgrp; 19947c478bd9Sstevel@tonic-gate 19957c478bd9Sstevel@tonic-gate ASSERT(szc != 0 && szc <= (page_num_pagesizes() - 1)); 19962cb27123Saguzovsk ASSERT(pgflags == 0 || pgflags == PG_LOCAL); 19977c478bd9Sstevel@tonic-gate 19982be2af34Smec /* 19992be2af34Smec * Check if system heavily prefers local large pages over remote 20002be2af34Smec * on systems with multiple lgroups. 20012be2af34Smec */ 20022be2af34Smec if (lpg_alloc_prefer == LPAP_LOCAL && nlgrps > 1) { 20032be2af34Smec pgflags = PG_LOCAL; 20042be2af34Smec } 20052be2af34Smec 20067c478bd9Sstevel@tonic-gate VM_STAT_ADD(alloc_pages[0]); 20077c478bd9Sstevel@tonic-gate 20087c478bd9Sstevel@tonic-gate #ifdef DEBUG 20097c478bd9Sstevel@tonic-gate if (pg_alloc_pgs_mtbf && !(gethrtime() % pg_alloc_pgs_mtbf)) { 20107c478bd9Sstevel@tonic-gate return (ENOMEM); 20117c478bd9Sstevel@tonic-gate } 20127c478bd9Sstevel@tonic-gate #endif 20137c478bd9Sstevel@tonic-gate 20147c478bd9Sstevel@tonic-gate /* 20157c478bd9Sstevel@tonic-gate * One must be NULL but not both. 20167c478bd9Sstevel@tonic-gate * And one must be non NULL but not both. 20177c478bd9Sstevel@tonic-gate */ 20187c478bd9Sstevel@tonic-gate ASSERT(basepp != NULL || ppa != NULL); 20197c478bd9Sstevel@tonic-gate ASSERT(basepp == NULL || ppa == NULL); 20207c478bd9Sstevel@tonic-gate 202178b03d3aSkchow #if defined(__i386) || defined(__amd64) 202278b03d3aSkchow while (page_chk_freelist(szc) == 0) { 202378b03d3aSkchow VM_STAT_ADD(alloc_pages[8]); 202478b03d3aSkchow if (anypgsz == 0 || --szc == 0) 202578b03d3aSkchow return (ENOMEM); 202678b03d3aSkchow } 202778b03d3aSkchow #endif 202878b03d3aSkchow 202978b03d3aSkchow pgsz = page_get_pagesize(szc); 203078b03d3aSkchow totpgs = curnpgs = npgs = pgsz >> PAGESHIFT; 203178b03d3aSkchow 203278b03d3aSkchow ASSERT(((uintptr_t)addr & (pgsz - 1)) == 0); 203378b03d3aSkchow 20347c478bd9Sstevel@tonic-gate (void) page_create_wait(npgs, PG_WAIT); 20357c478bd9Sstevel@tonic-gate 20367c478bd9Sstevel@tonic-gate while (npgs && szc) { 20377c478bd9Sstevel@tonic-gate lgrp = lgrp_mem_choose(seg, addr, pgsz); 20382cb27123Saguzovsk if (pgflags == PG_LOCAL) { 20392cb27123Saguzovsk pp = page_get_freelist(vp, 0, seg, addr, pgsz, 20402cb27123Saguzovsk pgflags, lgrp); 20412cb27123Saguzovsk if (pp == NULL) { 20422cb27123Saguzovsk pp = page_get_freelist(vp, 0, seg, addr, pgsz, 20432cb27123Saguzovsk 0, lgrp); 20442cb27123Saguzovsk } 20452cb27123Saguzovsk } else { 20462cb27123Saguzovsk pp = page_get_freelist(vp, 0, seg, addr, pgsz, 20472cb27123Saguzovsk 0, lgrp); 20482cb27123Saguzovsk } 20497c478bd9Sstevel@tonic-gate if (pp != NULL) { 20507c478bd9Sstevel@tonic-gate VM_STAT_ADD(alloc_pages[1]); 20517c478bd9Sstevel@tonic-gate page_list_concat(&pplist, &pp); 20527c478bd9Sstevel@tonic-gate ASSERT(npgs >= curnpgs); 20537c478bd9Sstevel@tonic-gate npgs -= curnpgs; 20547c478bd9Sstevel@tonic-gate } else if (anypgsz) { 20557c478bd9Sstevel@tonic-gate VM_STAT_ADD(alloc_pages[2]); 20567c478bd9Sstevel@tonic-gate szc--; 20577c478bd9Sstevel@tonic-gate pgsz = page_get_pagesize(szc); 20587c478bd9Sstevel@tonic-gate curnpgs = pgsz >> PAGESHIFT; 20597c478bd9Sstevel@tonic-gate } else { 20607c478bd9Sstevel@tonic-gate VM_STAT_ADD(alloc_pages[3]); 20617c478bd9Sstevel@tonic-gate ASSERT(npgs == totpgs); 20627c478bd9Sstevel@tonic-gate page_create_putback(npgs); 20637c478bd9Sstevel@tonic-gate return (ENOMEM); 20647c478bd9Sstevel@tonic-gate } 20657c478bd9Sstevel@tonic-gate } 20667c478bd9Sstevel@tonic-gate if (szc == 0) { 20677c478bd9Sstevel@tonic-gate VM_STAT_ADD(alloc_pages[4]); 20687c478bd9Sstevel@tonic-gate ASSERT(npgs != 0); 20697c478bd9Sstevel@tonic-gate page_create_putback(npgs); 20707c478bd9Sstevel@tonic-gate err = ENOMEM; 20717c478bd9Sstevel@tonic-gate } else if (basepp != NULL) { 20727c478bd9Sstevel@tonic-gate ASSERT(npgs == 0); 20737c478bd9Sstevel@tonic-gate ASSERT(ppa == NULL); 20747c478bd9Sstevel@tonic-gate *basepp = pplist; 20757c478bd9Sstevel@tonic-gate } 20767c478bd9Sstevel@tonic-gate 20777c478bd9Sstevel@tonic-gate npgs = totpgs - npgs; 20787c478bd9Sstevel@tonic-gate pp = pplist; 20797c478bd9Sstevel@tonic-gate 20807c478bd9Sstevel@tonic-gate /* 20817c478bd9Sstevel@tonic-gate * Clear the free and age bits. Also if we were passed in a ppa then 20827c478bd9Sstevel@tonic-gate * fill it in with all the constituent pages from the large page. But 20837c478bd9Sstevel@tonic-gate * if we failed to allocate all the pages just free what we got. 20847c478bd9Sstevel@tonic-gate */ 20857c478bd9Sstevel@tonic-gate while (npgs != 0) { 20867c478bd9Sstevel@tonic-gate ASSERT(PP_ISFREE(pp)); 20877c478bd9Sstevel@tonic-gate ASSERT(PP_ISAGED(pp)); 20887c478bd9Sstevel@tonic-gate if (ppa != NULL || err != 0) { 20897c478bd9Sstevel@tonic-gate if (err == 0) { 20907c478bd9Sstevel@tonic-gate VM_STAT_ADD(alloc_pages[5]); 20917c478bd9Sstevel@tonic-gate PP_CLRFREE(pp); 20927c478bd9Sstevel@tonic-gate PP_CLRAGED(pp); 20937c478bd9Sstevel@tonic-gate page_sub(&pplist, pp); 20947c478bd9Sstevel@tonic-gate *ppa++ = pp; 20957c478bd9Sstevel@tonic-gate npgs--; 20967c478bd9Sstevel@tonic-gate } else { 20977c478bd9Sstevel@tonic-gate VM_STAT_ADD(alloc_pages[6]); 20987c478bd9Sstevel@tonic-gate ASSERT(pp->p_szc != 0); 20997c478bd9Sstevel@tonic-gate curnpgs = page_get_pagecnt(pp->p_szc); 21007c478bd9Sstevel@tonic-gate page_list_break(&pp, &pplist, curnpgs); 21017c478bd9Sstevel@tonic-gate page_list_add_pages(pp, 0); 21027c478bd9Sstevel@tonic-gate page_create_putback(curnpgs); 21037c478bd9Sstevel@tonic-gate ASSERT(npgs >= curnpgs); 21047c478bd9Sstevel@tonic-gate npgs -= curnpgs; 21057c478bd9Sstevel@tonic-gate } 21067c478bd9Sstevel@tonic-gate pp = pplist; 21077c478bd9Sstevel@tonic-gate } else { 21087c478bd9Sstevel@tonic-gate VM_STAT_ADD(alloc_pages[7]); 21097c478bd9Sstevel@tonic-gate PP_CLRFREE(pp); 21107c478bd9Sstevel@tonic-gate PP_CLRAGED(pp); 21117c478bd9Sstevel@tonic-gate pp = pp->p_next; 21127c478bd9Sstevel@tonic-gate npgs--; 21137c478bd9Sstevel@tonic-gate } 21147c478bd9Sstevel@tonic-gate } 21157c478bd9Sstevel@tonic-gate return (err); 21167c478bd9Sstevel@tonic-gate } 21177c478bd9Sstevel@tonic-gate 21187c478bd9Sstevel@tonic-gate /* 21197c478bd9Sstevel@tonic-gate * Get a single large page off of the freelists, and set it up for use. 21207c478bd9Sstevel@tonic-gate * Number of bytes requested must be a supported page size. 21217c478bd9Sstevel@tonic-gate * 21227c478bd9Sstevel@tonic-gate * Note that this call may fail even if there is sufficient 21237c478bd9Sstevel@tonic-gate * memory available or PG_WAIT is set, so the caller must 21247c478bd9Sstevel@tonic-gate * be willing to fallback on page_create_va(), block and retry, 21257c478bd9Sstevel@tonic-gate * or fail the requester. 21267c478bd9Sstevel@tonic-gate */ 21277c478bd9Sstevel@tonic-gate page_t * 21287c478bd9Sstevel@tonic-gate page_create_va_large(vnode_t *vp, u_offset_t off, size_t bytes, uint_t flags, 21297c478bd9Sstevel@tonic-gate struct seg *seg, caddr_t vaddr, void *arg) 21307c478bd9Sstevel@tonic-gate { 2131*06fb6a36Sdv pgcnt_t npages; 21327c478bd9Sstevel@tonic-gate page_t *pp; 21337c478bd9Sstevel@tonic-gate page_t *rootpp; 21347c478bd9Sstevel@tonic-gate lgrp_t *lgrp; 21357c478bd9Sstevel@tonic-gate lgrp_id_t *lgrpid = (lgrp_id_t *)arg; 21367c478bd9Sstevel@tonic-gate 21377c478bd9Sstevel@tonic-gate ASSERT(vp != NULL); 21387c478bd9Sstevel@tonic-gate 21397c478bd9Sstevel@tonic-gate ASSERT((flags & ~(PG_EXCL | PG_WAIT | 21406e4dd838Smec PG_NORELOC | PG_PANIC | PG_PUSHPAGE)) == 0); 21417c478bd9Sstevel@tonic-gate /* but no others */ 21427c478bd9Sstevel@tonic-gate 21437c478bd9Sstevel@tonic-gate ASSERT((flags & PG_EXCL) == PG_EXCL); 21447c478bd9Sstevel@tonic-gate 21457c478bd9Sstevel@tonic-gate npages = btop(bytes); 21467c478bd9Sstevel@tonic-gate 21477c478bd9Sstevel@tonic-gate if (!kcage_on || panicstr) { 21487c478bd9Sstevel@tonic-gate /* 21497c478bd9Sstevel@tonic-gate * Cage is OFF, or we are single threaded in 21507c478bd9Sstevel@tonic-gate * panic, so make everything a RELOC request. 21517c478bd9Sstevel@tonic-gate */ 21527c478bd9Sstevel@tonic-gate flags &= ~PG_NORELOC; 21537c478bd9Sstevel@tonic-gate } 21547c478bd9Sstevel@tonic-gate 21557c478bd9Sstevel@tonic-gate /* 21567c478bd9Sstevel@tonic-gate * Make sure there's adequate physical memory available. 21577c478bd9Sstevel@tonic-gate * Note: PG_WAIT is ignored here. 21587c478bd9Sstevel@tonic-gate */ 21597c478bd9Sstevel@tonic-gate if (freemem <= throttlefree + npages) { 21607c478bd9Sstevel@tonic-gate VM_STAT_ADD(page_create_large_cnt[1]); 21617c478bd9Sstevel@tonic-gate return (NULL); 21627c478bd9Sstevel@tonic-gate } 21637c478bd9Sstevel@tonic-gate 21647c478bd9Sstevel@tonic-gate /* 21657c478bd9Sstevel@tonic-gate * If cage is on, dampen draw from cage when available 21667c478bd9Sstevel@tonic-gate * cage space is low. 21677c478bd9Sstevel@tonic-gate */ 21687c478bd9Sstevel@tonic-gate if ((flags & (PG_NORELOC | PG_WAIT)) == (PG_NORELOC | PG_WAIT) && 21697c478bd9Sstevel@tonic-gate kcage_freemem < kcage_throttlefree + npages) { 21707c478bd9Sstevel@tonic-gate 21717c478bd9Sstevel@tonic-gate /* 21727c478bd9Sstevel@tonic-gate * The cage is on, the caller wants PG_NORELOC 21737c478bd9Sstevel@tonic-gate * pages and available cage memory is very low. 21747c478bd9Sstevel@tonic-gate * Call kcage_create_throttle() to attempt to 21757c478bd9Sstevel@tonic-gate * control demand on the cage. 21767c478bd9Sstevel@tonic-gate */ 21777c478bd9Sstevel@tonic-gate if (kcage_create_throttle(npages, flags) == KCT_FAILURE) { 21787c478bd9Sstevel@tonic-gate VM_STAT_ADD(page_create_large_cnt[2]); 21797c478bd9Sstevel@tonic-gate return (NULL); 21807c478bd9Sstevel@tonic-gate } 21817c478bd9Sstevel@tonic-gate } 21827c478bd9Sstevel@tonic-gate 2183*06fb6a36Sdv if (!pcf_decrement_bucket(npages) && 2184*06fb6a36Sdv !pcf_decrement_multiple(NULL, npages, 1)) { 2185*06fb6a36Sdv VM_STAT_ADD(page_create_large_cnt[4]); 2186*06fb6a36Sdv return (NULL); 21877c478bd9Sstevel@tonic-gate } 21887c478bd9Sstevel@tonic-gate 21897c478bd9Sstevel@tonic-gate /* 21907c478bd9Sstevel@tonic-gate * This is where this function behaves fundamentally differently 21917c478bd9Sstevel@tonic-gate * than page_create_va(); since we're intending to map the page 21927c478bd9Sstevel@tonic-gate * with a single TTE, we have to get it as a physically contiguous 21937c478bd9Sstevel@tonic-gate * hardware pagesize chunk. If we can't, we fail. 21947c478bd9Sstevel@tonic-gate */ 21957c478bd9Sstevel@tonic-gate if (lgrpid != NULL && *lgrpid >= 0 && *lgrpid <= lgrp_alloc_max && 21966e4dd838Smec LGRP_EXISTS(lgrp_table[*lgrpid])) 21977c478bd9Sstevel@tonic-gate lgrp = lgrp_table[*lgrpid]; 21987c478bd9Sstevel@tonic-gate else 21997c478bd9Sstevel@tonic-gate lgrp = lgrp_mem_choose(seg, vaddr, bytes); 22007c478bd9Sstevel@tonic-gate 22017c478bd9Sstevel@tonic-gate if ((rootpp = page_get_freelist(&kvp, off, seg, vaddr, 22027c478bd9Sstevel@tonic-gate bytes, flags & ~PG_MATCH_COLOR, lgrp)) == NULL) { 22037c478bd9Sstevel@tonic-gate page_create_putback(npages); 22047c478bd9Sstevel@tonic-gate VM_STAT_ADD(page_create_large_cnt[5]); 22057c478bd9Sstevel@tonic-gate return (NULL); 22067c478bd9Sstevel@tonic-gate } 22077c478bd9Sstevel@tonic-gate 22087c478bd9Sstevel@tonic-gate /* 22097c478bd9Sstevel@tonic-gate * if we got the page with the wrong mtype give it back this is a 22107c478bd9Sstevel@tonic-gate * workaround for CR 6249718. When CR 6249718 is fixed we never get 22117c478bd9Sstevel@tonic-gate * inside "if" and the workaround becomes just a nop 22127c478bd9Sstevel@tonic-gate */ 22137c478bd9Sstevel@tonic-gate if (kcage_on && (flags & PG_NORELOC) && !PP_ISNORELOC(rootpp)) { 22147c478bd9Sstevel@tonic-gate page_list_add_pages(rootpp, 0); 22157c478bd9Sstevel@tonic-gate page_create_putback(npages); 22167c478bd9Sstevel@tonic-gate VM_STAT_ADD(page_create_large_cnt[6]); 22177c478bd9Sstevel@tonic-gate return (NULL); 22187c478bd9Sstevel@tonic-gate } 22197c478bd9Sstevel@tonic-gate 22207c478bd9Sstevel@tonic-gate /* 22217c478bd9Sstevel@tonic-gate * If satisfying this request has left us with too little 22227c478bd9Sstevel@tonic-gate * memory, start the wheels turning to get some back. The 22237c478bd9Sstevel@tonic-gate * first clause of the test prevents waking up the pageout 22247c478bd9Sstevel@tonic-gate * daemon in situations where it would decide that there's 22257c478bd9Sstevel@tonic-gate * nothing to do. 22267c478bd9Sstevel@tonic-gate */ 22277c478bd9Sstevel@tonic-gate if (nscan < desscan && freemem < minfree) { 22287c478bd9Sstevel@tonic-gate TRACE_1(TR_FAC_VM, TR_PAGEOUT_CV_SIGNAL, 22297c478bd9Sstevel@tonic-gate "pageout_cv_signal:freemem %ld", freemem); 22307c478bd9Sstevel@tonic-gate cv_signal(&proc_pageout->p_cv); 22317c478bd9Sstevel@tonic-gate } 22327c478bd9Sstevel@tonic-gate 22337c478bd9Sstevel@tonic-gate pp = rootpp; 22347c478bd9Sstevel@tonic-gate while (npages--) { 22357c478bd9Sstevel@tonic-gate ASSERT(PAGE_EXCL(pp)); 22367c478bd9Sstevel@tonic-gate ASSERT(pp->p_vnode == NULL); 22377c478bd9Sstevel@tonic-gate ASSERT(!hat_page_is_mapped(pp)); 22387c478bd9Sstevel@tonic-gate PP_CLRFREE(pp); 22397c478bd9Sstevel@tonic-gate PP_CLRAGED(pp); 22407c478bd9Sstevel@tonic-gate if (!page_hashin(pp, vp, off, NULL)) 22417c478bd9Sstevel@tonic-gate panic("page_create_large: hashin failed: page %p", 22427c478bd9Sstevel@tonic-gate (void *)pp); 22437c478bd9Sstevel@tonic-gate page_io_lock(pp); 22447c478bd9Sstevel@tonic-gate off += PAGESIZE; 22457c478bd9Sstevel@tonic-gate pp = pp->p_next; 22467c478bd9Sstevel@tonic-gate } 22477c478bd9Sstevel@tonic-gate 22487c478bd9Sstevel@tonic-gate VM_STAT_ADD(page_create_large_cnt[0]); 22497c478bd9Sstevel@tonic-gate return (rootpp); 22507c478bd9Sstevel@tonic-gate } 22517c478bd9Sstevel@tonic-gate 22527c478bd9Sstevel@tonic-gate page_t * 22537c478bd9Sstevel@tonic-gate page_create_va(vnode_t *vp, u_offset_t off, size_t bytes, uint_t flags, 22547c478bd9Sstevel@tonic-gate struct seg *seg, caddr_t vaddr) 22557c478bd9Sstevel@tonic-gate { 22567c478bd9Sstevel@tonic-gate page_t *plist = NULL; 22577c478bd9Sstevel@tonic-gate pgcnt_t npages; 22587c478bd9Sstevel@tonic-gate pgcnt_t found_on_free = 0; 22597c478bd9Sstevel@tonic-gate pgcnt_t pages_req; 22607c478bd9Sstevel@tonic-gate page_t *npp = NULL; 22617c478bd9Sstevel@tonic-gate struct pcf *p; 22627c478bd9Sstevel@tonic-gate lgrp_t *lgrp; 22637c478bd9Sstevel@tonic-gate 22647c478bd9Sstevel@tonic-gate TRACE_4(TR_FAC_VM, TR_PAGE_CREATE_START, 22656e4dd838Smec "page_create_start:vp %p off %llx bytes %lu flags %x", 22666e4dd838Smec vp, off, bytes, flags); 22677c478bd9Sstevel@tonic-gate 22687c478bd9Sstevel@tonic-gate ASSERT(bytes != 0 && vp != NULL); 22697c478bd9Sstevel@tonic-gate 22707c478bd9Sstevel@tonic-gate if ((flags & PG_EXCL) == 0 && (flags & PG_WAIT) == 0) { 22717c478bd9Sstevel@tonic-gate panic("page_create: invalid flags"); 22727c478bd9Sstevel@tonic-gate /*NOTREACHED*/ 22737c478bd9Sstevel@tonic-gate } 22747c478bd9Sstevel@tonic-gate ASSERT((flags & ~(PG_EXCL | PG_WAIT | 22757c478bd9Sstevel@tonic-gate PG_NORELOC | PG_PANIC | PG_PUSHPAGE)) == 0); 22767c478bd9Sstevel@tonic-gate /* but no others */ 22777c478bd9Sstevel@tonic-gate 22787c478bd9Sstevel@tonic-gate pages_req = npages = btopr(bytes); 22797c478bd9Sstevel@tonic-gate /* 22807c478bd9Sstevel@tonic-gate * Try to see whether request is too large to *ever* be 22817c478bd9Sstevel@tonic-gate * satisfied, in order to prevent deadlock. We arbitrarily 22827c478bd9Sstevel@tonic-gate * decide to limit maximum size requests to max_page_get. 22837c478bd9Sstevel@tonic-gate */ 22847c478bd9Sstevel@tonic-gate if (npages >= max_page_get) { 22857c478bd9Sstevel@tonic-gate if ((flags & PG_WAIT) == 0) { 22867c478bd9Sstevel@tonic-gate TRACE_4(TR_FAC_VM, TR_PAGE_CREATE_TOOBIG, 22877c478bd9Sstevel@tonic-gate "page_create_toobig:vp %p off %llx npages " 22887c478bd9Sstevel@tonic-gate "%lu max_page_get %lu", 22897c478bd9Sstevel@tonic-gate vp, off, npages, max_page_get); 22907c478bd9Sstevel@tonic-gate return (NULL); 22917c478bd9Sstevel@tonic-gate } else { 22927c478bd9Sstevel@tonic-gate cmn_err(CE_WARN, 22937c478bd9Sstevel@tonic-gate "Request for too much kernel memory " 22947c478bd9Sstevel@tonic-gate "(%lu bytes), will hang forever", bytes); 22957c478bd9Sstevel@tonic-gate for (;;) 22967c478bd9Sstevel@tonic-gate delay(1000000000); 22977c478bd9Sstevel@tonic-gate } 22987c478bd9Sstevel@tonic-gate } 22997c478bd9Sstevel@tonic-gate 23007c478bd9Sstevel@tonic-gate if (!kcage_on || panicstr) { 23017c478bd9Sstevel@tonic-gate /* 23027c478bd9Sstevel@tonic-gate * Cage is OFF, or we are single threaded in 23037c478bd9Sstevel@tonic-gate * panic, so make everything a RELOC request. 23047c478bd9Sstevel@tonic-gate */ 23057c478bd9Sstevel@tonic-gate flags &= ~PG_NORELOC; 23067c478bd9Sstevel@tonic-gate } 23077c478bd9Sstevel@tonic-gate 23087c478bd9Sstevel@tonic-gate if (freemem <= throttlefree + npages) 23097c478bd9Sstevel@tonic-gate if (!page_create_throttle(npages, flags)) 23107c478bd9Sstevel@tonic-gate return (NULL); 23117c478bd9Sstevel@tonic-gate 23127c478bd9Sstevel@tonic-gate /* 23137c478bd9Sstevel@tonic-gate * If cage is on, dampen draw from cage when available 23147c478bd9Sstevel@tonic-gate * cage space is low. 23157c478bd9Sstevel@tonic-gate */ 23167c478bd9Sstevel@tonic-gate if ((flags & PG_NORELOC) && 23176e4dd838Smec kcage_freemem < kcage_throttlefree + npages) { 23187c478bd9Sstevel@tonic-gate 23197c478bd9Sstevel@tonic-gate /* 23207c478bd9Sstevel@tonic-gate * The cage is on, the caller wants PG_NORELOC 23217c478bd9Sstevel@tonic-gate * pages and available cage memory is very low. 23227c478bd9Sstevel@tonic-gate * Call kcage_create_throttle() to attempt to 23237c478bd9Sstevel@tonic-gate * control demand on the cage. 23247c478bd9Sstevel@tonic-gate */ 23257c478bd9Sstevel@tonic-gate if (kcage_create_throttle(npages, flags) == KCT_FAILURE) 23267c478bd9Sstevel@tonic-gate return (NULL); 23277c478bd9Sstevel@tonic-gate } 23287c478bd9Sstevel@tonic-gate 23297c478bd9Sstevel@tonic-gate VM_STAT_ADD(page_create_cnt[0]); 23307c478bd9Sstevel@tonic-gate 2331*06fb6a36Sdv if (!pcf_decrement_bucket(npages)) { 23327c478bd9Sstevel@tonic-gate /* 23337c478bd9Sstevel@tonic-gate * Have to look harder. If npages is greater than 2334da6c28aaSamw * one, then we might have to coalesce the counters. 23357c478bd9Sstevel@tonic-gate * 23367c478bd9Sstevel@tonic-gate * Go wait. We come back having accounted 23377c478bd9Sstevel@tonic-gate * for the memory. 23387c478bd9Sstevel@tonic-gate */ 23397c478bd9Sstevel@tonic-gate VM_STAT_ADD(page_create_cnt[1]); 23407c478bd9Sstevel@tonic-gate if (!page_create_wait(npages, flags)) { 23417c478bd9Sstevel@tonic-gate VM_STAT_ADD(page_create_cnt[2]); 23427c478bd9Sstevel@tonic-gate return (NULL); 23437c478bd9Sstevel@tonic-gate } 23447c478bd9Sstevel@tonic-gate } 23457c478bd9Sstevel@tonic-gate 23467c478bd9Sstevel@tonic-gate TRACE_2(TR_FAC_VM, TR_PAGE_CREATE_SUCCESS, 23476e4dd838Smec "page_create_success:vp %p off %llx", vp, off); 23487c478bd9Sstevel@tonic-gate 23497c478bd9Sstevel@tonic-gate /* 23507c478bd9Sstevel@tonic-gate * If satisfying this request has left us with too little 23517c478bd9Sstevel@tonic-gate * memory, start the wheels turning to get some back. The 23527c478bd9Sstevel@tonic-gate * first clause of the test prevents waking up the pageout 23537c478bd9Sstevel@tonic-gate * daemon in situations where it would decide that there's 23547c478bd9Sstevel@tonic-gate * nothing to do. 23557c478bd9Sstevel@tonic-gate */ 23567c478bd9Sstevel@tonic-gate if (nscan < desscan && freemem < minfree) { 23577c478bd9Sstevel@tonic-gate TRACE_1(TR_FAC_VM, TR_PAGEOUT_CV_SIGNAL, 23586e4dd838Smec "pageout_cv_signal:freemem %ld", freemem); 23597c478bd9Sstevel@tonic-gate cv_signal(&proc_pageout->p_cv); 23607c478bd9Sstevel@tonic-gate } 23617c478bd9Sstevel@tonic-gate 23627c478bd9Sstevel@tonic-gate /* 23637c478bd9Sstevel@tonic-gate * Loop around collecting the requested number of pages. 23647c478bd9Sstevel@tonic-gate * Most of the time, we have to `create' a new page. With 23657c478bd9Sstevel@tonic-gate * this in mind, pull the page off the free list before 23667c478bd9Sstevel@tonic-gate * getting the hash lock. This will minimize the hash 23677c478bd9Sstevel@tonic-gate * lock hold time, nesting, and the like. If it turns 23687c478bd9Sstevel@tonic-gate * out we don't need the page, we put it back at the end. 23697c478bd9Sstevel@tonic-gate */ 23707c478bd9Sstevel@tonic-gate while (npages--) { 23717c478bd9Sstevel@tonic-gate page_t *pp; 23727c478bd9Sstevel@tonic-gate kmutex_t *phm = NULL; 23737c478bd9Sstevel@tonic-gate ulong_t index; 23747c478bd9Sstevel@tonic-gate 23757c478bd9Sstevel@tonic-gate index = PAGE_HASH_FUNC(vp, off); 23767c478bd9Sstevel@tonic-gate top: 23777c478bd9Sstevel@tonic-gate ASSERT(phm == NULL); 23787c478bd9Sstevel@tonic-gate ASSERT(index == PAGE_HASH_FUNC(vp, off)); 23797c478bd9Sstevel@tonic-gate ASSERT(MUTEX_NOT_HELD(page_vnode_mutex(vp))); 23807c478bd9Sstevel@tonic-gate 23817c478bd9Sstevel@tonic-gate if (npp == NULL) { 23827c478bd9Sstevel@tonic-gate /* 23837c478bd9Sstevel@tonic-gate * Try to get a page from the freelist (ie, 23847c478bd9Sstevel@tonic-gate * a page with no [vp, off] tag). If that 23857c478bd9Sstevel@tonic-gate * fails, use the cachelist. 23867c478bd9Sstevel@tonic-gate * 23877c478bd9Sstevel@tonic-gate * During the first attempt at both the free 23887c478bd9Sstevel@tonic-gate * and cache lists we try for the correct color. 23897c478bd9Sstevel@tonic-gate */ 23907c478bd9Sstevel@tonic-gate /* 23917c478bd9Sstevel@tonic-gate * XXXX-how do we deal with virtual indexed 23927c478bd9Sstevel@tonic-gate * caches and and colors? 23937c478bd9Sstevel@tonic-gate */ 23947c478bd9Sstevel@tonic-gate VM_STAT_ADD(page_create_cnt[4]); 23957c478bd9Sstevel@tonic-gate /* 23967c478bd9Sstevel@tonic-gate * Get lgroup to allocate next page of shared memory 23977c478bd9Sstevel@tonic-gate * from and use it to specify where to allocate 23987c478bd9Sstevel@tonic-gate * the physical memory 23997c478bd9Sstevel@tonic-gate */ 24007c478bd9Sstevel@tonic-gate lgrp = lgrp_mem_choose(seg, vaddr, PAGESIZE); 24017c478bd9Sstevel@tonic-gate npp = page_get_freelist(vp, off, seg, vaddr, PAGESIZE, 24027c478bd9Sstevel@tonic-gate flags | PG_MATCH_COLOR, lgrp); 24037c478bd9Sstevel@tonic-gate if (npp == NULL) { 24047c478bd9Sstevel@tonic-gate npp = page_get_cachelist(vp, off, seg, 24057c478bd9Sstevel@tonic-gate vaddr, flags | PG_MATCH_COLOR, lgrp); 24067c478bd9Sstevel@tonic-gate if (npp == NULL) { 24077c478bd9Sstevel@tonic-gate npp = page_create_get_something(vp, 24087c478bd9Sstevel@tonic-gate off, seg, vaddr, 24097c478bd9Sstevel@tonic-gate flags & ~PG_MATCH_COLOR); 24107c478bd9Sstevel@tonic-gate } 24117c478bd9Sstevel@tonic-gate 24127c478bd9Sstevel@tonic-gate if (PP_ISAGED(npp) == 0) { 24137c478bd9Sstevel@tonic-gate /* 24147c478bd9Sstevel@tonic-gate * Since this page came from the 24157c478bd9Sstevel@tonic-gate * cachelist, we must destroy the 24167c478bd9Sstevel@tonic-gate * old vnode association. 24177c478bd9Sstevel@tonic-gate */ 24187c478bd9Sstevel@tonic-gate page_hashout(npp, NULL); 24197c478bd9Sstevel@tonic-gate } 24207c478bd9Sstevel@tonic-gate } 24217c478bd9Sstevel@tonic-gate } 24227c478bd9Sstevel@tonic-gate 24237c478bd9Sstevel@tonic-gate /* 24247c478bd9Sstevel@tonic-gate * We own this page! 24257c478bd9Sstevel@tonic-gate */ 24267c478bd9Sstevel@tonic-gate ASSERT(PAGE_EXCL(npp)); 24277c478bd9Sstevel@tonic-gate ASSERT(npp->p_vnode == NULL); 24287c478bd9Sstevel@tonic-gate ASSERT(!hat_page_is_mapped(npp)); 24297c478bd9Sstevel@tonic-gate PP_CLRFREE(npp); 24307c478bd9Sstevel@tonic-gate PP_CLRAGED(npp); 24317c478bd9Sstevel@tonic-gate 24327c478bd9Sstevel@tonic-gate /* 24337c478bd9Sstevel@tonic-gate * Here we have a page in our hot little mits and are 24347c478bd9Sstevel@tonic-gate * just waiting to stuff it on the appropriate lists. 24357c478bd9Sstevel@tonic-gate * Get the mutex and check to see if it really does 24367c478bd9Sstevel@tonic-gate * not exist. 24377c478bd9Sstevel@tonic-gate */ 24387c478bd9Sstevel@tonic-gate phm = PAGE_HASH_MUTEX(index); 24397c478bd9Sstevel@tonic-gate mutex_enter(phm); 24407c478bd9Sstevel@tonic-gate PAGE_HASH_SEARCH(index, pp, vp, off); 24417c478bd9Sstevel@tonic-gate if (pp == NULL) { 24427c478bd9Sstevel@tonic-gate VM_STAT_ADD(page_create_new); 24437c478bd9Sstevel@tonic-gate pp = npp; 24447c478bd9Sstevel@tonic-gate npp = NULL; 24457c478bd9Sstevel@tonic-gate if (!page_hashin(pp, vp, off, phm)) { 24467c478bd9Sstevel@tonic-gate /* 24477c478bd9Sstevel@tonic-gate * Since we hold the page hash mutex and 24487c478bd9Sstevel@tonic-gate * just searched for this page, page_hashin 24497c478bd9Sstevel@tonic-gate * had better not fail. If it does, that 24507c478bd9Sstevel@tonic-gate * means somethread did not follow the 24517c478bd9Sstevel@tonic-gate * page hash mutex rules. Panic now and 24527c478bd9Sstevel@tonic-gate * get it over with. As usual, go down 24537c478bd9Sstevel@tonic-gate * holding all the locks. 24547c478bd9Sstevel@tonic-gate */ 24557c478bd9Sstevel@tonic-gate ASSERT(MUTEX_HELD(phm)); 24567c478bd9Sstevel@tonic-gate panic("page_create: " 24577c478bd9Sstevel@tonic-gate "hashin failed %p %p %llx %p", 24587c478bd9Sstevel@tonic-gate (void *)pp, (void *)vp, off, (void *)phm); 24597c478bd9Sstevel@tonic-gate /*NOTREACHED*/ 24607c478bd9Sstevel@tonic-gate } 24617c478bd9Sstevel@tonic-gate ASSERT(MUTEX_HELD(phm)); 24627c478bd9Sstevel@tonic-gate mutex_exit(phm); 24637c478bd9Sstevel@tonic-gate phm = NULL; 24647c478bd9Sstevel@tonic-gate 24657c478bd9Sstevel@tonic-gate /* 24667c478bd9Sstevel@tonic-gate * Hat layer locking need not be done to set 24677c478bd9Sstevel@tonic-gate * the following bits since the page is not hashed 24687c478bd9Sstevel@tonic-gate * and was on the free list (i.e., had no mappings). 24697c478bd9Sstevel@tonic-gate * 24707c478bd9Sstevel@tonic-gate * Set the reference bit to protect 24717c478bd9Sstevel@tonic-gate * against immediate pageout 24727c478bd9Sstevel@tonic-gate * 24737c478bd9Sstevel@tonic-gate * XXXmh modify freelist code to set reference 24747c478bd9Sstevel@tonic-gate * bit so we don't have to do it here. 24757c478bd9Sstevel@tonic-gate */ 24767c478bd9Sstevel@tonic-gate page_set_props(pp, P_REF); 24777c478bd9Sstevel@tonic-gate found_on_free++; 24787c478bd9Sstevel@tonic-gate } else { 24797c478bd9Sstevel@tonic-gate VM_STAT_ADD(page_create_exists); 24807c478bd9Sstevel@tonic-gate if (flags & PG_EXCL) { 24817c478bd9Sstevel@tonic-gate /* 24827c478bd9Sstevel@tonic-gate * Found an existing page, and the caller 24837c478bd9Sstevel@tonic-gate * wanted all new pages. Undo all of the work 24847c478bd9Sstevel@tonic-gate * we have done. 24857c478bd9Sstevel@tonic-gate */ 24867c478bd9Sstevel@tonic-gate mutex_exit(phm); 24877c478bd9Sstevel@tonic-gate phm = NULL; 24887c478bd9Sstevel@tonic-gate while (plist != NULL) { 24897c478bd9Sstevel@tonic-gate pp = plist; 24907c478bd9Sstevel@tonic-gate page_sub(&plist, pp); 24917c478bd9Sstevel@tonic-gate page_io_unlock(pp); 24927c478bd9Sstevel@tonic-gate /* large pages should not end up here */ 24937c478bd9Sstevel@tonic-gate ASSERT(pp->p_szc == 0); 24947c478bd9Sstevel@tonic-gate /*LINTED: constant in conditional ctx*/ 24957c478bd9Sstevel@tonic-gate VN_DISPOSE(pp, B_INVAL, 0, kcred); 24967c478bd9Sstevel@tonic-gate } 24977c478bd9Sstevel@tonic-gate VM_STAT_ADD(page_create_found_one); 24987c478bd9Sstevel@tonic-gate goto fail; 24997c478bd9Sstevel@tonic-gate } 25007c478bd9Sstevel@tonic-gate ASSERT(flags & PG_WAIT); 25017c478bd9Sstevel@tonic-gate if (!page_lock(pp, SE_EXCL, phm, P_NO_RECLAIM)) { 25027c478bd9Sstevel@tonic-gate /* 25037c478bd9Sstevel@tonic-gate * Start all over again if we blocked trying 25047c478bd9Sstevel@tonic-gate * to lock the page. 25057c478bd9Sstevel@tonic-gate */ 25067c478bd9Sstevel@tonic-gate mutex_exit(phm); 25077c478bd9Sstevel@tonic-gate VM_STAT_ADD(page_create_page_lock_failed); 25087c478bd9Sstevel@tonic-gate phm = NULL; 25097c478bd9Sstevel@tonic-gate goto top; 25107c478bd9Sstevel@tonic-gate } 25117c478bd9Sstevel@tonic-gate mutex_exit(phm); 25127c478bd9Sstevel@tonic-gate phm = NULL; 25137c478bd9Sstevel@tonic-gate 25147c478bd9Sstevel@tonic-gate if (PP_ISFREE(pp)) { 25157c478bd9Sstevel@tonic-gate ASSERT(PP_ISAGED(pp) == 0); 25167c478bd9Sstevel@tonic-gate VM_STAT_ADD(pagecnt.pc_get_cache); 25177c478bd9Sstevel@tonic-gate page_list_sub(pp, PG_CACHE_LIST); 25187c478bd9Sstevel@tonic-gate PP_CLRFREE(pp); 25197c478bd9Sstevel@tonic-gate found_on_free++; 25207c478bd9Sstevel@tonic-gate } 25217c478bd9Sstevel@tonic-gate } 25227c478bd9Sstevel@tonic-gate 25237c478bd9Sstevel@tonic-gate /* 25247c478bd9Sstevel@tonic-gate * Got a page! It is locked. Acquire the i/o 25257c478bd9Sstevel@tonic-gate * lock since we are going to use the p_next and 25267c478bd9Sstevel@tonic-gate * p_prev fields to link the requested pages together. 25277c478bd9Sstevel@tonic-gate */ 25287c478bd9Sstevel@tonic-gate page_io_lock(pp); 25297c478bd9Sstevel@tonic-gate page_add(&plist, pp); 25307c478bd9Sstevel@tonic-gate plist = plist->p_next; 25317c478bd9Sstevel@tonic-gate off += PAGESIZE; 25327c478bd9Sstevel@tonic-gate vaddr += PAGESIZE; 25337c478bd9Sstevel@tonic-gate } 25347c478bd9Sstevel@tonic-gate 25357c478bd9Sstevel@tonic-gate ASSERT((flags & PG_EXCL) ? (found_on_free == pages_req) : 1); 25367c478bd9Sstevel@tonic-gate fail: 25377c478bd9Sstevel@tonic-gate if (npp != NULL) { 25387c478bd9Sstevel@tonic-gate /* 25397c478bd9Sstevel@tonic-gate * Did not need this page after all. 25407c478bd9Sstevel@tonic-gate * Put it back on the free list. 25417c478bd9Sstevel@tonic-gate */ 25427c478bd9Sstevel@tonic-gate VM_STAT_ADD(page_create_putbacks); 25437c478bd9Sstevel@tonic-gate PP_SETFREE(npp); 25447c478bd9Sstevel@tonic-gate PP_SETAGED(npp); 25457c478bd9Sstevel@tonic-gate npp->p_offset = (u_offset_t)-1; 25467c478bd9Sstevel@tonic-gate page_list_add(npp, PG_FREE_LIST | PG_LIST_TAIL); 25477c478bd9Sstevel@tonic-gate page_unlock(npp); 25487c478bd9Sstevel@tonic-gate 25497c478bd9Sstevel@tonic-gate } 25507c478bd9Sstevel@tonic-gate 25517c478bd9Sstevel@tonic-gate ASSERT(pages_req >= found_on_free); 25527c478bd9Sstevel@tonic-gate 25537c478bd9Sstevel@tonic-gate { 25547c478bd9Sstevel@tonic-gate uint_t overshoot = (uint_t)(pages_req - found_on_free); 25557c478bd9Sstevel@tonic-gate 25567c478bd9Sstevel@tonic-gate if (overshoot) { 25577c478bd9Sstevel@tonic-gate VM_STAT_ADD(page_create_overshoot); 2558*06fb6a36Sdv p = &pcf[PCF_INDEX()]; 25597c478bd9Sstevel@tonic-gate mutex_enter(&p->pcf_lock); 25607c478bd9Sstevel@tonic-gate if (p->pcf_block) { 25617c478bd9Sstevel@tonic-gate p->pcf_reserve += overshoot; 25627c478bd9Sstevel@tonic-gate } else { 25637c478bd9Sstevel@tonic-gate p->pcf_count += overshoot; 25647c478bd9Sstevel@tonic-gate if (p->pcf_wait) { 25657c478bd9Sstevel@tonic-gate mutex_enter(&new_freemem_lock); 25667c478bd9Sstevel@tonic-gate if (freemem_wait) { 25677c478bd9Sstevel@tonic-gate cv_signal(&freemem_cv); 25687c478bd9Sstevel@tonic-gate p->pcf_wait--; 25697c478bd9Sstevel@tonic-gate } else { 25707c478bd9Sstevel@tonic-gate p->pcf_wait = 0; 25717c478bd9Sstevel@tonic-gate } 25727c478bd9Sstevel@tonic-gate mutex_exit(&new_freemem_lock); 25737c478bd9Sstevel@tonic-gate } 25747c478bd9Sstevel@tonic-gate } 25757c478bd9Sstevel@tonic-gate mutex_exit(&p->pcf_lock); 25767c478bd9Sstevel@tonic-gate /* freemem is approximate, so this test OK */ 25777c478bd9Sstevel@tonic-gate if (!p->pcf_block) 25787c478bd9Sstevel@tonic-gate freemem += overshoot; 25797c478bd9Sstevel@tonic-gate } 25807c478bd9Sstevel@tonic-gate } 25817c478bd9Sstevel@tonic-gate 25827c478bd9Sstevel@tonic-gate return (plist); 25837c478bd9Sstevel@tonic-gate } 25847c478bd9Sstevel@tonic-gate 25857c478bd9Sstevel@tonic-gate /* 25867c478bd9Sstevel@tonic-gate * One or more constituent pages of this large page has been marked 25877c478bd9Sstevel@tonic-gate * toxic. Simply demote the large page to PAGESIZE pages and let 25887c478bd9Sstevel@tonic-gate * page_free() handle it. This routine should only be called by 25897c478bd9Sstevel@tonic-gate * large page free routines (page_free_pages() and page_destroy_pages(). 25907c478bd9Sstevel@tonic-gate * All pages are locked SE_EXCL and have already been marked free. 25917c478bd9Sstevel@tonic-gate */ 25927c478bd9Sstevel@tonic-gate static void 25937c478bd9Sstevel@tonic-gate page_free_toxic_pages(page_t *rootpp) 25947c478bd9Sstevel@tonic-gate { 25957c478bd9Sstevel@tonic-gate page_t *tpp; 25967c478bd9Sstevel@tonic-gate pgcnt_t i, pgcnt = page_get_pagecnt(rootpp->p_szc); 25977c478bd9Sstevel@tonic-gate uint_t szc = rootpp->p_szc; 25987c478bd9Sstevel@tonic-gate 25997c478bd9Sstevel@tonic-gate for (i = 0, tpp = rootpp; i < pgcnt; i++, tpp = tpp->p_next) { 26007c478bd9Sstevel@tonic-gate ASSERT(tpp->p_szc == szc); 26017c478bd9Sstevel@tonic-gate ASSERT((PAGE_EXCL(tpp) && 26027c478bd9Sstevel@tonic-gate !page_iolock_assert(tpp)) || panicstr); 26037c478bd9Sstevel@tonic-gate tpp->p_szc = 0; 26047c478bd9Sstevel@tonic-gate } 26057c478bd9Sstevel@tonic-gate 26067c478bd9Sstevel@tonic-gate while (rootpp != NULL) { 26077c478bd9Sstevel@tonic-gate tpp = rootpp; 26087c478bd9Sstevel@tonic-gate page_sub(&rootpp, tpp); 26097c478bd9Sstevel@tonic-gate ASSERT(PP_ISFREE(tpp)); 26107c478bd9Sstevel@tonic-gate PP_CLRFREE(tpp); 26117c478bd9Sstevel@tonic-gate page_free(tpp, 1); 26127c478bd9Sstevel@tonic-gate } 26137c478bd9Sstevel@tonic-gate } 26147c478bd9Sstevel@tonic-gate 26157c478bd9Sstevel@tonic-gate /* 26167c478bd9Sstevel@tonic-gate * Put page on the "free" list. 26177c478bd9Sstevel@tonic-gate * The free list is really two lists maintained by 26187c478bd9Sstevel@tonic-gate * the PSM of whatever machine we happen to be on. 26197c478bd9Sstevel@tonic-gate */ 26207c478bd9Sstevel@tonic-gate void 26217c478bd9Sstevel@tonic-gate page_free(page_t *pp, int dontneed) 26227c478bd9Sstevel@tonic-gate { 26237c478bd9Sstevel@tonic-gate struct pcf *p; 26247c478bd9Sstevel@tonic-gate uint_t pcf_index; 26257c478bd9Sstevel@tonic-gate 26267c478bd9Sstevel@tonic-gate ASSERT((PAGE_EXCL(pp) && 26277c478bd9Sstevel@tonic-gate !page_iolock_assert(pp)) || panicstr); 26287c478bd9Sstevel@tonic-gate 26297c478bd9Sstevel@tonic-gate if (PP_ISFREE(pp)) { 26307c478bd9Sstevel@tonic-gate panic("page_free: page %p is free", (void *)pp); 26317c478bd9Sstevel@tonic-gate } 26327c478bd9Sstevel@tonic-gate 26337c478bd9Sstevel@tonic-gate if (pp->p_szc != 0) { 26347c478bd9Sstevel@tonic-gate if (pp->p_vnode == NULL || IS_SWAPFSVP(pp->p_vnode) || 2635ad23a2dbSjohansen PP_ISKAS(pp)) { 26367c478bd9Sstevel@tonic-gate panic("page_free: anon or kernel " 26377c478bd9Sstevel@tonic-gate "or no vnode large page %p", (void *)pp); 26387c478bd9Sstevel@tonic-gate } 26397c478bd9Sstevel@tonic-gate page_demote_vp_pages(pp); 26407c478bd9Sstevel@tonic-gate ASSERT(pp->p_szc == 0); 26417c478bd9Sstevel@tonic-gate } 26427c478bd9Sstevel@tonic-gate 26437c478bd9Sstevel@tonic-gate /* 26447c478bd9Sstevel@tonic-gate * The page_struct_lock need not be acquired to examine these 26457c478bd9Sstevel@tonic-gate * fields since the page has an "exclusive" lock. 26467c478bd9Sstevel@tonic-gate */ 264707b65a64Saguzovsk if (hat_page_is_mapped(pp) || pp->p_lckcnt != 0 || pp->p_cowcnt != 0 || 264807b65a64Saguzovsk pp->p_slckcnt != 0) { 264907b65a64Saguzovsk panic("page_free pp=%p, pfn=%lx, lckcnt=%d, cowcnt=%d " 265007b65a64Saguzovsk "slckcnt = %d", pp, page_pptonum(pp), pp->p_lckcnt, 265107b65a64Saguzovsk pp->p_cowcnt, pp->p_slckcnt); 26527c478bd9Sstevel@tonic-gate /*NOTREACHED*/ 26537c478bd9Sstevel@tonic-gate } 26547c478bd9Sstevel@tonic-gate 26557c478bd9Sstevel@tonic-gate ASSERT(!hat_page_getshare(pp)); 26567c478bd9Sstevel@tonic-gate 26577c478bd9Sstevel@tonic-gate PP_SETFREE(pp); 26587c478bd9Sstevel@tonic-gate ASSERT(pp->p_vnode == NULL || !IS_VMODSORT(pp->p_vnode) || 26597c478bd9Sstevel@tonic-gate !hat_ismod(pp)); 26607c478bd9Sstevel@tonic-gate page_clr_all_props(pp); 26617c478bd9Sstevel@tonic-gate ASSERT(!hat_page_getshare(pp)); 26627c478bd9Sstevel@tonic-gate 26637c478bd9Sstevel@tonic-gate /* 26647c478bd9Sstevel@tonic-gate * Now we add the page to the head of the free list. 26657c478bd9Sstevel@tonic-gate * But if this page is associated with a paged vnode 26667c478bd9Sstevel@tonic-gate * then we adjust the head forward so that the page is 26677c478bd9Sstevel@tonic-gate * effectively at the end of the list. 26687c478bd9Sstevel@tonic-gate */ 26697c478bd9Sstevel@tonic-gate if (pp->p_vnode == NULL) { 26707c478bd9Sstevel@tonic-gate /* 26717c478bd9Sstevel@tonic-gate * Page has no identity, put it on the free list. 26727c478bd9Sstevel@tonic-gate */ 26737c478bd9Sstevel@tonic-gate PP_SETAGED(pp); 26747c478bd9Sstevel@tonic-gate pp->p_offset = (u_offset_t)-1; 26757c478bd9Sstevel@tonic-gate page_list_add(pp, PG_FREE_LIST | PG_LIST_TAIL); 26767c478bd9Sstevel@tonic-gate VM_STAT_ADD(pagecnt.pc_free_free); 26777c478bd9Sstevel@tonic-gate TRACE_1(TR_FAC_VM, TR_PAGE_FREE_FREE, 26787c478bd9Sstevel@tonic-gate "page_free_free:pp %p", pp); 26797c478bd9Sstevel@tonic-gate } else { 26807c478bd9Sstevel@tonic-gate PP_CLRAGED(pp); 26817c478bd9Sstevel@tonic-gate 26827c478bd9Sstevel@tonic-gate if (!dontneed || nopageage) { 26837c478bd9Sstevel@tonic-gate /* move it to the tail of the list */ 26847c478bd9Sstevel@tonic-gate page_list_add(pp, PG_CACHE_LIST | PG_LIST_TAIL); 26857c478bd9Sstevel@tonic-gate 26867c478bd9Sstevel@tonic-gate VM_STAT_ADD(pagecnt.pc_free_cache); 26877c478bd9Sstevel@tonic-gate TRACE_1(TR_FAC_VM, TR_PAGE_FREE_CACHE_TAIL, 26887c478bd9Sstevel@tonic-gate "page_free_cache_tail:pp %p", pp); 26897c478bd9Sstevel@tonic-gate } else { 26907c478bd9Sstevel@tonic-gate page_list_add(pp, PG_CACHE_LIST | PG_LIST_HEAD); 26917c478bd9Sstevel@tonic-gate 26927c478bd9Sstevel@tonic-gate VM_STAT_ADD(pagecnt.pc_free_dontneed); 26937c478bd9Sstevel@tonic-gate TRACE_1(TR_FAC_VM, TR_PAGE_FREE_CACHE_HEAD, 26947c478bd9Sstevel@tonic-gate "page_free_cache_head:pp %p", pp); 26957c478bd9Sstevel@tonic-gate } 26967c478bd9Sstevel@tonic-gate } 26977c478bd9Sstevel@tonic-gate page_unlock(pp); 26987c478bd9Sstevel@tonic-gate 26997c478bd9Sstevel@tonic-gate /* 27007c478bd9Sstevel@tonic-gate * Now do the `freemem' accounting. 27017c478bd9Sstevel@tonic-gate */ 27027c478bd9Sstevel@tonic-gate pcf_index = PCF_INDEX(); 27037c478bd9Sstevel@tonic-gate p = &pcf[pcf_index]; 27047c478bd9Sstevel@tonic-gate 27057c478bd9Sstevel@tonic-gate mutex_enter(&p->pcf_lock); 27067c478bd9Sstevel@tonic-gate if (p->pcf_block) { 27077c478bd9Sstevel@tonic-gate p->pcf_reserve += 1; 27087c478bd9Sstevel@tonic-gate } else { 27097c478bd9Sstevel@tonic-gate p->pcf_count += 1; 27107c478bd9Sstevel@tonic-gate if (p->pcf_wait) { 27117c478bd9Sstevel@tonic-gate mutex_enter(&new_freemem_lock); 27127c478bd9Sstevel@tonic-gate /* 27137c478bd9Sstevel@tonic-gate * Check to see if some other thread 27147c478bd9Sstevel@tonic-gate * is actually waiting. Another bucket 27157c478bd9Sstevel@tonic-gate * may have woken it up by now. If there 27167c478bd9Sstevel@tonic-gate * are no waiters, then set our pcf_wait 27177c478bd9Sstevel@tonic-gate * count to zero to avoid coming in here 27187c478bd9Sstevel@tonic-gate * next time. Also, since only one page 27197c478bd9Sstevel@tonic-gate * was put on the free list, just wake 27207c478bd9Sstevel@tonic-gate * up one waiter. 27217c478bd9Sstevel@tonic-gate */ 27227c478bd9Sstevel@tonic-gate if (freemem_wait) { 27237c478bd9Sstevel@tonic-gate cv_signal(&freemem_cv); 27247c478bd9Sstevel@tonic-gate p->pcf_wait--; 27257c478bd9Sstevel@tonic-gate } else { 27267c478bd9Sstevel@tonic-gate p->pcf_wait = 0; 27277c478bd9Sstevel@tonic-gate } 27287c478bd9Sstevel@tonic-gate mutex_exit(&new_freemem_lock); 27297c478bd9Sstevel@tonic-gate } 27307c478bd9Sstevel@tonic-gate } 27317c478bd9Sstevel@tonic-gate mutex_exit(&p->pcf_lock); 27327c478bd9Sstevel@tonic-gate 27337c478bd9Sstevel@tonic-gate /* freemem is approximate, so this test OK */ 27347c478bd9Sstevel@tonic-gate if (!p->pcf_block) 27357c478bd9Sstevel@tonic-gate freemem += 1; 27367c478bd9Sstevel@tonic-gate } 27377c478bd9Sstevel@tonic-gate 27387c478bd9Sstevel@tonic-gate /* 27397c478bd9Sstevel@tonic-gate * Put page on the "free" list during intial startup. 27407c478bd9Sstevel@tonic-gate * This happens during initial single threaded execution. 27417c478bd9Sstevel@tonic-gate */ 27427c478bd9Sstevel@tonic-gate void 27437c478bd9Sstevel@tonic-gate page_free_at_startup(page_t *pp) 27447c478bd9Sstevel@tonic-gate { 27457c478bd9Sstevel@tonic-gate struct pcf *p; 27467c478bd9Sstevel@tonic-gate uint_t pcf_index; 27477c478bd9Sstevel@tonic-gate 27487c478bd9Sstevel@tonic-gate page_list_add(pp, PG_FREE_LIST | PG_LIST_HEAD | PG_LIST_ISINIT); 27497c478bd9Sstevel@tonic-gate VM_STAT_ADD(pagecnt.pc_free_free); 27507c478bd9Sstevel@tonic-gate 27517c478bd9Sstevel@tonic-gate /* 27527c478bd9Sstevel@tonic-gate * Now do the `freemem' accounting. 27537c478bd9Sstevel@tonic-gate */ 27547c478bd9Sstevel@tonic-gate pcf_index = PCF_INDEX(); 27557c478bd9Sstevel@tonic-gate p = &pcf[pcf_index]; 27567c478bd9Sstevel@tonic-gate 27577c478bd9Sstevel@tonic-gate ASSERT(p->pcf_block == 0); 27587c478bd9Sstevel@tonic-gate ASSERT(p->pcf_wait == 0); 27597c478bd9Sstevel@tonic-gate p->pcf_count += 1; 27607c478bd9Sstevel@tonic-gate 27617c478bd9Sstevel@tonic-gate /* freemem is approximate, so this is OK */ 27627c478bd9Sstevel@tonic-gate freemem += 1; 27637c478bd9Sstevel@tonic-gate } 27647c478bd9Sstevel@tonic-gate 27657c478bd9Sstevel@tonic-gate void 27667c478bd9Sstevel@tonic-gate page_free_pages(page_t *pp) 27677c478bd9Sstevel@tonic-gate { 27687c478bd9Sstevel@tonic-gate page_t *tpp, *rootpp = NULL; 27697c478bd9Sstevel@tonic-gate pgcnt_t pgcnt = page_get_pagecnt(pp->p_szc); 27707c478bd9Sstevel@tonic-gate pgcnt_t i; 27717c478bd9Sstevel@tonic-gate uint_t szc = pp->p_szc; 27727c478bd9Sstevel@tonic-gate 27737c478bd9Sstevel@tonic-gate VM_STAT_ADD(pagecnt.pc_free_pages); 27747c478bd9Sstevel@tonic-gate TRACE_1(TR_FAC_VM, TR_PAGE_FREE_FREE, 27757c478bd9Sstevel@tonic-gate "page_free_free:pp %p", pp); 27767c478bd9Sstevel@tonic-gate 27777c478bd9Sstevel@tonic-gate ASSERT(pp->p_szc != 0 && pp->p_szc < page_num_pagesizes()); 27787c478bd9Sstevel@tonic-gate if ((page_pptonum(pp) & (pgcnt - 1)) != 0) { 27797c478bd9Sstevel@tonic-gate panic("page_free_pages: not root page %p", (void *)pp); 27807c478bd9Sstevel@tonic-gate /*NOTREACHED*/ 27817c478bd9Sstevel@tonic-gate } 27827c478bd9Sstevel@tonic-gate 2783affbd3ccSkchow for (i = 0, tpp = pp; i < pgcnt; i++, tpp++) { 27847c478bd9Sstevel@tonic-gate ASSERT((PAGE_EXCL(tpp) && 27857c478bd9Sstevel@tonic-gate !page_iolock_assert(tpp)) || panicstr); 27867c478bd9Sstevel@tonic-gate if (PP_ISFREE(tpp)) { 27877c478bd9Sstevel@tonic-gate panic("page_free_pages: page %p is free", (void *)tpp); 27887c478bd9Sstevel@tonic-gate /*NOTREACHED*/ 27897c478bd9Sstevel@tonic-gate } 27907c478bd9Sstevel@tonic-gate if (hat_page_is_mapped(tpp) || tpp->p_lckcnt != 0 || 279107b65a64Saguzovsk tpp->p_cowcnt != 0 || tpp->p_slckcnt != 0) { 27927c478bd9Sstevel@tonic-gate panic("page_free_pages %p", (void *)tpp); 27937c478bd9Sstevel@tonic-gate /*NOTREACHED*/ 27947c478bd9Sstevel@tonic-gate } 27957c478bd9Sstevel@tonic-gate 27967c478bd9Sstevel@tonic-gate ASSERT(!hat_page_getshare(tpp)); 27977c478bd9Sstevel@tonic-gate ASSERT(tpp->p_vnode == NULL); 27987c478bd9Sstevel@tonic-gate ASSERT(tpp->p_szc == szc); 27997c478bd9Sstevel@tonic-gate 28007c478bd9Sstevel@tonic-gate PP_SETFREE(tpp); 28017c478bd9Sstevel@tonic-gate page_clr_all_props(tpp); 28027c478bd9Sstevel@tonic-gate PP_SETAGED(tpp); 28037c478bd9Sstevel@tonic-gate tpp->p_offset = (u_offset_t)-1; 28047c478bd9Sstevel@tonic-gate ASSERT(tpp->p_next == tpp); 28057c478bd9Sstevel@tonic-gate ASSERT(tpp->p_prev == tpp); 28067c478bd9Sstevel@tonic-gate page_list_concat(&rootpp, &tpp); 28077c478bd9Sstevel@tonic-gate } 28087c478bd9Sstevel@tonic-gate ASSERT(rootpp == pp); 28097c478bd9Sstevel@tonic-gate 28107c478bd9Sstevel@tonic-gate page_list_add_pages(rootpp, 0); 28117c478bd9Sstevel@tonic-gate page_create_putback(pgcnt); 28127c478bd9Sstevel@tonic-gate } 28137c478bd9Sstevel@tonic-gate 28147c478bd9Sstevel@tonic-gate int free_pages = 1; 28157c478bd9Sstevel@tonic-gate 28167c478bd9Sstevel@tonic-gate /* 28177c478bd9Sstevel@tonic-gate * This routine attempts to return pages to the cachelist via page_release(). 28187c478bd9Sstevel@tonic-gate * It does not *have* to be successful in all cases, since the pageout scanner 28197c478bd9Sstevel@tonic-gate * will catch any pages it misses. It does need to be fast and not introduce 28207c478bd9Sstevel@tonic-gate * too much overhead. 28217c478bd9Sstevel@tonic-gate * 28227c478bd9Sstevel@tonic-gate * If a page isn't found on the unlocked sweep of the page_hash bucket, we 28237c478bd9Sstevel@tonic-gate * don't lock and retry. This is ok, since the page scanner will eventually 28247c478bd9Sstevel@tonic-gate * find any page we miss in free_vp_pages(). 28257c478bd9Sstevel@tonic-gate */ 28267c478bd9Sstevel@tonic-gate void 28277c478bd9Sstevel@tonic-gate free_vp_pages(vnode_t *vp, u_offset_t off, size_t len) 28287c478bd9Sstevel@tonic-gate { 28297c478bd9Sstevel@tonic-gate page_t *pp; 28307c478bd9Sstevel@tonic-gate u_offset_t eoff; 28317c478bd9Sstevel@tonic-gate extern int swap_in_range(vnode_t *, u_offset_t, size_t); 28327c478bd9Sstevel@tonic-gate 28337c478bd9Sstevel@tonic-gate eoff = off + len; 28347c478bd9Sstevel@tonic-gate 28357c478bd9Sstevel@tonic-gate if (free_pages == 0) 28367c478bd9Sstevel@tonic-gate return; 28377c478bd9Sstevel@tonic-gate if (swap_in_range(vp, off, len)) 28387c478bd9Sstevel@tonic-gate return; 28397c478bd9Sstevel@tonic-gate 28407c478bd9Sstevel@tonic-gate for (; off < eoff; off += PAGESIZE) { 28417c478bd9Sstevel@tonic-gate 28427c478bd9Sstevel@tonic-gate /* 28437c478bd9Sstevel@tonic-gate * find the page using a fast, but inexact search. It'll be OK 28447c478bd9Sstevel@tonic-gate * if a few pages slip through the cracks here. 28457c478bd9Sstevel@tonic-gate */ 28467c478bd9Sstevel@tonic-gate pp = page_exists(vp, off); 28477c478bd9Sstevel@tonic-gate 28487c478bd9Sstevel@tonic-gate /* 28497c478bd9Sstevel@tonic-gate * If we didn't find the page (it may not exist), the page 28507c478bd9Sstevel@tonic-gate * is free, looks still in use (shared), or we can't lock it, 28517c478bd9Sstevel@tonic-gate * just give up. 28527c478bd9Sstevel@tonic-gate */ 28537c478bd9Sstevel@tonic-gate if (pp == NULL || 28547c478bd9Sstevel@tonic-gate PP_ISFREE(pp) || 28557c478bd9Sstevel@tonic-gate page_share_cnt(pp) > 0 || 28567c478bd9Sstevel@tonic-gate !page_trylock(pp, SE_EXCL)) 28577c478bd9Sstevel@tonic-gate continue; 28587c478bd9Sstevel@tonic-gate 28597c478bd9Sstevel@tonic-gate /* 28607c478bd9Sstevel@tonic-gate * Once we have locked pp, verify that it's still the 28617c478bd9Sstevel@tonic-gate * correct page and not already free 28627c478bd9Sstevel@tonic-gate */ 28637c478bd9Sstevel@tonic-gate ASSERT(PAGE_LOCKED_SE(pp, SE_EXCL)); 28647c478bd9Sstevel@tonic-gate if (pp->p_vnode != vp || pp->p_offset != off || PP_ISFREE(pp)) { 28657c478bd9Sstevel@tonic-gate page_unlock(pp); 28667c478bd9Sstevel@tonic-gate continue; 28677c478bd9Sstevel@tonic-gate } 28687c478bd9Sstevel@tonic-gate 28697c478bd9Sstevel@tonic-gate /* 28707c478bd9Sstevel@tonic-gate * try to release the page... 28717c478bd9Sstevel@tonic-gate */ 28727c478bd9Sstevel@tonic-gate (void) page_release(pp, 1); 28737c478bd9Sstevel@tonic-gate } 28747c478bd9Sstevel@tonic-gate } 28757c478bd9Sstevel@tonic-gate 28767c478bd9Sstevel@tonic-gate /* 28777c478bd9Sstevel@tonic-gate * Reclaim the given page from the free list. 28786e4dd838Smec * If pp is part of a large pages, only the given constituent page is reclaimed 28796e4dd838Smec * and the large page it belonged to will be demoted. This can only happen 28806e4dd838Smec * if the page is not on the cachelist. 28816e4dd838Smec * 28827c478bd9Sstevel@tonic-gate * Returns 1 on success or 0 on failure. 28837c478bd9Sstevel@tonic-gate * 28847c478bd9Sstevel@tonic-gate * The page is unlocked if it can't be reclaimed (when freemem == 0). 28857c478bd9Sstevel@tonic-gate * If `lock' is non-null, it will be dropped and re-acquired if 28867c478bd9Sstevel@tonic-gate * the routine must wait while freemem is 0. 28877c478bd9Sstevel@tonic-gate * 28887c478bd9Sstevel@tonic-gate * As it turns out, boot_getpages() does this. It picks a page, 28897c478bd9Sstevel@tonic-gate * based on where OBP mapped in some address, gets its pfn, searches 28907c478bd9Sstevel@tonic-gate * the memsegs, locks the page, then pulls it off the free list! 28917c478bd9Sstevel@tonic-gate */ 28927c478bd9Sstevel@tonic-gate int 28937c478bd9Sstevel@tonic-gate page_reclaim(page_t *pp, kmutex_t *lock) 28947c478bd9Sstevel@tonic-gate { 28957c478bd9Sstevel@tonic-gate struct pcf *p; 28967c478bd9Sstevel@tonic-gate struct cpu *cpup; 28976e4dd838Smec int enough; 28987c478bd9Sstevel@tonic-gate uint_t i; 28997c478bd9Sstevel@tonic-gate 29007c478bd9Sstevel@tonic-gate ASSERT(lock != NULL ? MUTEX_HELD(lock) : 1); 29017c478bd9Sstevel@tonic-gate ASSERT(PAGE_EXCL(pp) && PP_ISFREE(pp)); 2902db874c57Selowe 29037c478bd9Sstevel@tonic-gate /* 29047c478bd9Sstevel@tonic-gate * If `freemem' is 0, we cannot reclaim this page from the 29057c478bd9Sstevel@tonic-gate * freelist, so release every lock we might hold: the page, 29067c478bd9Sstevel@tonic-gate * and the `lock' before blocking. 29077c478bd9Sstevel@tonic-gate * 29087c478bd9Sstevel@tonic-gate * The only way `freemem' can become 0 while there are pages 29097c478bd9Sstevel@tonic-gate * marked free (have their p->p_free bit set) is when the 29107c478bd9Sstevel@tonic-gate * system is low on memory and doing a page_create(). In 29117c478bd9Sstevel@tonic-gate * order to guarantee that once page_create() starts acquiring 29127c478bd9Sstevel@tonic-gate * pages it will be able to get all that it needs since `freemem' 29137c478bd9Sstevel@tonic-gate * was decreased by the requested amount. So, we need to release 29147c478bd9Sstevel@tonic-gate * this page, and let page_create() have it. 29157c478bd9Sstevel@tonic-gate * 29167c478bd9Sstevel@tonic-gate * Since `freemem' being zero is not supposed to happen, just 29177c478bd9Sstevel@tonic-gate * use the usual hash stuff as a starting point. If that bucket 29187c478bd9Sstevel@tonic-gate * is empty, then assume the worst, and start at the beginning 29197c478bd9Sstevel@tonic-gate * of the pcf array. If we always start at the beginning 29207c478bd9Sstevel@tonic-gate * when acquiring more than one pcf lock, there won't be any 29217c478bd9Sstevel@tonic-gate * deadlock problems. 29227c478bd9Sstevel@tonic-gate */ 29237c478bd9Sstevel@tonic-gate 29247c478bd9Sstevel@tonic-gate /* TODO: Do we need to test kcage_freemem if PG_NORELOC(pp)? */ 29257c478bd9Sstevel@tonic-gate 29266e4dd838Smec if (freemem <= throttlefree && !page_create_throttle(1l, 0)) { 29277c478bd9Sstevel@tonic-gate pcf_acquire_all(); 29287c478bd9Sstevel@tonic-gate goto page_reclaim_nomem; 29297c478bd9Sstevel@tonic-gate } 29307c478bd9Sstevel@tonic-gate 2931*06fb6a36Sdv enough = pcf_decrement_bucket(1); 29327c478bd9Sstevel@tonic-gate 29336e4dd838Smec if (!enough) { 29347c478bd9Sstevel@tonic-gate VM_STAT_ADD(page_reclaim_zero); 29357c478bd9Sstevel@tonic-gate /* 29367c478bd9Sstevel@tonic-gate * Check again. Its possible that some other thread 29377c478bd9Sstevel@tonic-gate * could have been right behind us, and added one 29387c478bd9Sstevel@tonic-gate * to a list somewhere. Acquire each of the pcf locks 29397c478bd9Sstevel@tonic-gate * until we find a page. 29407c478bd9Sstevel@tonic-gate */ 29417c478bd9Sstevel@tonic-gate p = pcf; 2942*06fb6a36Sdv for (i = 0; i < pcf_fanout; i++) { 29437c478bd9Sstevel@tonic-gate mutex_enter(&p->pcf_lock); 29446e4dd838Smec if (p->pcf_count >= 1) { 29456e4dd838Smec p->pcf_count -= 1; 29466e4dd838Smec enough = 1; 29476e4dd838Smec break; 29487c478bd9Sstevel@tonic-gate } 29497c478bd9Sstevel@tonic-gate p++; 29507c478bd9Sstevel@tonic-gate } 29517c478bd9Sstevel@tonic-gate 29526e4dd838Smec if (!enough) { 29537c478bd9Sstevel@tonic-gate page_reclaim_nomem: 29547c478bd9Sstevel@tonic-gate /* 29557c478bd9Sstevel@tonic-gate * We really can't have page `pp'. 29567c478bd9Sstevel@tonic-gate * Time for the no-memory dance with 29577c478bd9Sstevel@tonic-gate * page_free(). This is just like 29587c478bd9Sstevel@tonic-gate * page_create_wait(). Plus the added 29597c478bd9Sstevel@tonic-gate * attraction of releasing whatever mutex 29607c478bd9Sstevel@tonic-gate * we held when we were called with in `lock'. 29617c478bd9Sstevel@tonic-gate * Page_unlock() will wakeup any thread 29627c478bd9Sstevel@tonic-gate * waiting around for this page. 29637c478bd9Sstevel@tonic-gate */ 29647c478bd9Sstevel@tonic-gate if (lock) { 29657c478bd9Sstevel@tonic-gate VM_STAT_ADD(page_reclaim_zero_locked); 29667c478bd9Sstevel@tonic-gate mutex_exit(lock); 29677c478bd9Sstevel@tonic-gate } 29687c478bd9Sstevel@tonic-gate page_unlock(pp); 29697c478bd9Sstevel@tonic-gate 29707c478bd9Sstevel@tonic-gate /* 29717c478bd9Sstevel@tonic-gate * get this before we drop all the pcf locks. 29727c478bd9Sstevel@tonic-gate */ 29737c478bd9Sstevel@tonic-gate mutex_enter(&new_freemem_lock); 29747c478bd9Sstevel@tonic-gate 29757c478bd9Sstevel@tonic-gate p = pcf; 2976*06fb6a36Sdv for (i = 0; i < pcf_fanout; i++) { 29777c478bd9Sstevel@tonic-gate p->pcf_wait++; 29787c478bd9Sstevel@tonic-gate mutex_exit(&p->pcf_lock); 29797c478bd9Sstevel@tonic-gate p++; 29807c478bd9Sstevel@tonic-gate } 29817c478bd9Sstevel@tonic-gate 29827c478bd9Sstevel@tonic-gate freemem_wait++; 29837c478bd9Sstevel@tonic-gate cv_wait(&freemem_cv, &new_freemem_lock); 29847c478bd9Sstevel@tonic-gate freemem_wait--; 29857c478bd9Sstevel@tonic-gate 29867c478bd9Sstevel@tonic-gate mutex_exit(&new_freemem_lock); 29877c478bd9Sstevel@tonic-gate 29887c478bd9Sstevel@tonic-gate if (lock) { 29897c478bd9Sstevel@tonic-gate mutex_enter(lock); 29907c478bd9Sstevel@tonic-gate } 29917c478bd9Sstevel@tonic-gate return (0); 29927c478bd9Sstevel@tonic-gate } 29937c478bd9Sstevel@tonic-gate 29947c478bd9Sstevel@tonic-gate /* 29957c478bd9Sstevel@tonic-gate * The pcf accounting has been done, 29967c478bd9Sstevel@tonic-gate * though none of the pcf_wait flags have been set, 29977c478bd9Sstevel@tonic-gate * drop the locks and continue on. 29987c478bd9Sstevel@tonic-gate */ 29997c478bd9Sstevel@tonic-gate while (p >= pcf) { 30007c478bd9Sstevel@tonic-gate mutex_exit(&p->pcf_lock); 30017c478bd9Sstevel@tonic-gate p--; 30027c478bd9Sstevel@tonic-gate } 30037c478bd9Sstevel@tonic-gate } 30047c478bd9Sstevel@tonic-gate 30057c478bd9Sstevel@tonic-gate /* 30067c478bd9Sstevel@tonic-gate * freemem is not protected by any lock. Thus, we cannot 30077c478bd9Sstevel@tonic-gate * have any assertion containing freemem here. 30087c478bd9Sstevel@tonic-gate */ 30096e4dd838Smec freemem -= 1; 30107c478bd9Sstevel@tonic-gate 30117c478bd9Sstevel@tonic-gate VM_STAT_ADD(pagecnt.pc_reclaim); 30126e4dd838Smec 30136e4dd838Smec /* 30146e4dd838Smec * page_list_sub will handle the case where pp is a large page. 30156e4dd838Smec * It's possible that the page was promoted while on the freelist 30166e4dd838Smec */ 30177c478bd9Sstevel@tonic-gate if (PP_ISAGED(pp)) { 30186e4dd838Smec page_list_sub(pp, PG_FREE_LIST); 30197c478bd9Sstevel@tonic-gate TRACE_1(TR_FAC_VM, TR_PAGE_UNFREE_FREE, 30207c478bd9Sstevel@tonic-gate "page_reclaim_free:pp %p", pp); 30217c478bd9Sstevel@tonic-gate } else { 30227c478bd9Sstevel@tonic-gate page_list_sub(pp, PG_CACHE_LIST); 30237c478bd9Sstevel@tonic-gate TRACE_1(TR_FAC_VM, TR_PAGE_UNFREE_CACHE, 30247c478bd9Sstevel@tonic-gate "page_reclaim_cache:pp %p", pp); 30257c478bd9Sstevel@tonic-gate } 30267c478bd9Sstevel@tonic-gate 30277c478bd9Sstevel@tonic-gate /* 30287c478bd9Sstevel@tonic-gate * clear the p_free & p_age bits since this page is no longer 30297c478bd9Sstevel@tonic-gate * on the free list. Notice that there was a brief time where 30307c478bd9Sstevel@tonic-gate * a page is marked as free, but is not on the list. 30317c478bd9Sstevel@tonic-gate * 30327c478bd9Sstevel@tonic-gate * Set the reference bit to protect against immediate pageout. 30337c478bd9Sstevel@tonic-gate */ 30346e4dd838Smec PP_CLRFREE(pp); 30356e4dd838Smec PP_CLRAGED(pp); 30366e4dd838Smec page_set_props(pp, P_REF); 30377c478bd9Sstevel@tonic-gate 30387c478bd9Sstevel@tonic-gate CPU_STATS_ENTER_K(); 30397c478bd9Sstevel@tonic-gate cpup = CPU; /* get cpup now that CPU cannot change */ 30407c478bd9Sstevel@tonic-gate CPU_STATS_ADDQ(cpup, vm, pgrec, 1); 30417c478bd9Sstevel@tonic-gate CPU_STATS_ADDQ(cpup, vm, pgfrec, 1); 30427c478bd9Sstevel@tonic-gate CPU_STATS_EXIT_K(); 30436e4dd838Smec ASSERT(pp->p_szc == 0); 30447c478bd9Sstevel@tonic-gate 30457c478bd9Sstevel@tonic-gate return (1); 30467c478bd9Sstevel@tonic-gate } 30477c478bd9Sstevel@tonic-gate 30487c478bd9Sstevel@tonic-gate /* 30497c478bd9Sstevel@tonic-gate * Destroy identity of the page and put it back on 30507c478bd9Sstevel@tonic-gate * the page free list. Assumes that the caller has 30517c478bd9Sstevel@tonic-gate * acquired the "exclusive" lock on the page. 30527c478bd9Sstevel@tonic-gate */ 30537c478bd9Sstevel@tonic-gate void 30547c478bd9Sstevel@tonic-gate page_destroy(page_t *pp, int dontfree) 30557c478bd9Sstevel@tonic-gate { 30567c478bd9Sstevel@tonic-gate ASSERT((PAGE_EXCL(pp) && 30577c478bd9Sstevel@tonic-gate !page_iolock_assert(pp)) || panicstr); 305807b65a64Saguzovsk ASSERT(pp->p_slckcnt == 0 || panicstr); 30597c478bd9Sstevel@tonic-gate 30607c478bd9Sstevel@tonic-gate if (pp->p_szc != 0) { 30617c478bd9Sstevel@tonic-gate if (pp->p_vnode == NULL || IS_SWAPFSVP(pp->p_vnode) || 3062ad23a2dbSjohansen PP_ISKAS(pp)) { 30637c478bd9Sstevel@tonic-gate panic("page_destroy: anon or kernel or no vnode " 30647c478bd9Sstevel@tonic-gate "large page %p", (void *)pp); 30657c478bd9Sstevel@tonic-gate } 30667c478bd9Sstevel@tonic-gate page_demote_vp_pages(pp); 30677c478bd9Sstevel@tonic-gate ASSERT(pp->p_szc == 0); 30687c478bd9Sstevel@tonic-gate } 30697c478bd9Sstevel@tonic-gate 30707c478bd9Sstevel@tonic-gate TRACE_1(TR_FAC_VM, TR_PAGE_DESTROY, "page_destroy:pp %p", pp); 30717c478bd9Sstevel@tonic-gate 30727c478bd9Sstevel@tonic-gate /* 30737c478bd9Sstevel@tonic-gate * Unload translations, if any, then hash out the 30747c478bd9Sstevel@tonic-gate * page to erase its identity. 30757c478bd9Sstevel@tonic-gate */ 30767c478bd9Sstevel@tonic-gate (void) hat_pageunload(pp, HAT_FORCE_PGUNLOAD); 30777c478bd9Sstevel@tonic-gate page_hashout(pp, NULL); 30787c478bd9Sstevel@tonic-gate 30797c478bd9Sstevel@tonic-gate if (!dontfree) { 30807c478bd9Sstevel@tonic-gate /* 30817c478bd9Sstevel@tonic-gate * Acquire the "freemem_lock" for availrmem. 30827c478bd9Sstevel@tonic-gate * The page_struct_lock need not be acquired for lckcnt 30837c478bd9Sstevel@tonic-gate * and cowcnt since the page has an "exclusive" lock. 30847c478bd9Sstevel@tonic-gate */ 30857c478bd9Sstevel@tonic-gate if ((pp->p_lckcnt != 0) || (pp->p_cowcnt != 0)) { 30867c478bd9Sstevel@tonic-gate mutex_enter(&freemem_lock); 30877c478bd9Sstevel@tonic-gate if (pp->p_lckcnt != 0) { 30887c478bd9Sstevel@tonic-gate availrmem++; 30897c478bd9Sstevel@tonic-gate pp->p_lckcnt = 0; 30907c478bd9Sstevel@tonic-gate } 30917c478bd9Sstevel@tonic-gate if (pp->p_cowcnt != 0) { 30927c478bd9Sstevel@tonic-gate availrmem += pp->p_cowcnt; 30937c478bd9Sstevel@tonic-gate pp->p_cowcnt = 0; 30947c478bd9Sstevel@tonic-gate } 30957c478bd9Sstevel@tonic-gate mutex_exit(&freemem_lock); 30967c478bd9Sstevel@tonic-gate } 30977c478bd9Sstevel@tonic-gate /* 30987c478bd9Sstevel@tonic-gate * Put the page on the "free" list. 30997c478bd9Sstevel@tonic-gate */ 31007c478bd9Sstevel@tonic-gate page_free(pp, 0); 31017c478bd9Sstevel@tonic-gate } 31027c478bd9Sstevel@tonic-gate } 31037c478bd9Sstevel@tonic-gate 31047c478bd9Sstevel@tonic-gate void 31057c478bd9Sstevel@tonic-gate page_destroy_pages(page_t *pp) 31067c478bd9Sstevel@tonic-gate { 31077c478bd9Sstevel@tonic-gate 31087c478bd9Sstevel@tonic-gate page_t *tpp, *rootpp = NULL; 31097c478bd9Sstevel@tonic-gate pgcnt_t pgcnt = page_get_pagecnt(pp->p_szc); 31107c478bd9Sstevel@tonic-gate pgcnt_t i, pglcks = 0; 31117c478bd9Sstevel@tonic-gate uint_t szc = pp->p_szc; 31127c478bd9Sstevel@tonic-gate 31137c478bd9Sstevel@tonic-gate ASSERT(pp->p_szc != 0 && pp->p_szc < page_num_pagesizes()); 31147c478bd9Sstevel@tonic-gate 31157c478bd9Sstevel@tonic-gate VM_STAT_ADD(pagecnt.pc_destroy_pages); 31167c478bd9Sstevel@tonic-gate 31177c478bd9Sstevel@tonic-gate TRACE_1(TR_FAC_VM, TR_PAGE_DESTROY, "page_destroy_pages:pp %p", pp); 31187c478bd9Sstevel@tonic-gate 31197c478bd9Sstevel@tonic-gate if ((page_pptonum(pp) & (pgcnt - 1)) != 0) { 31207c478bd9Sstevel@tonic-gate panic("page_destroy_pages: not root page %p", (void *)pp); 31217c478bd9Sstevel@tonic-gate /*NOTREACHED*/ 31227c478bd9Sstevel@tonic-gate } 31237c478bd9Sstevel@tonic-gate 3124affbd3ccSkchow for (i = 0, tpp = pp; i < pgcnt; i++, tpp++) { 31257c478bd9Sstevel@tonic-gate ASSERT((PAGE_EXCL(tpp) && 31267c478bd9Sstevel@tonic-gate !page_iolock_assert(tpp)) || panicstr); 312707b65a64Saguzovsk ASSERT(tpp->p_slckcnt == 0 || panicstr); 31287c478bd9Sstevel@tonic-gate (void) hat_pageunload(tpp, HAT_FORCE_PGUNLOAD); 31297c478bd9Sstevel@tonic-gate page_hashout(tpp, NULL); 31307c478bd9Sstevel@tonic-gate ASSERT(tpp->p_offset == (u_offset_t)-1); 31317c478bd9Sstevel@tonic-gate if (tpp->p_lckcnt != 0) { 31327c478bd9Sstevel@tonic-gate pglcks++; 31337c478bd9Sstevel@tonic-gate tpp->p_lckcnt = 0; 31347c478bd9Sstevel@tonic-gate } else if (tpp->p_cowcnt != 0) { 31357c478bd9Sstevel@tonic-gate pglcks += tpp->p_cowcnt; 31367c478bd9Sstevel@tonic-gate tpp->p_cowcnt = 0; 31377c478bd9Sstevel@tonic-gate } 31387c478bd9Sstevel@tonic-gate ASSERT(!hat_page_getshare(tpp)); 31397c478bd9Sstevel@tonic-gate ASSERT(tpp->p_vnode == NULL); 31407c478bd9Sstevel@tonic-gate ASSERT(tpp->p_szc == szc); 31417c478bd9Sstevel@tonic-gate 31427c478bd9Sstevel@tonic-gate PP_SETFREE(tpp); 31437c478bd9Sstevel@tonic-gate page_clr_all_props(tpp); 31447c478bd9Sstevel@tonic-gate PP_SETAGED(tpp); 31457c478bd9Sstevel@tonic-gate ASSERT(tpp->p_next == tpp); 31467c478bd9Sstevel@tonic-gate ASSERT(tpp->p_prev == tpp); 31477c478bd9Sstevel@tonic-gate page_list_concat(&rootpp, &tpp); 31487c478bd9Sstevel@tonic-gate } 31497c478bd9Sstevel@tonic-gate 31507c478bd9Sstevel@tonic-gate ASSERT(rootpp == pp); 31517c478bd9Sstevel@tonic-gate if (pglcks != 0) { 31527c478bd9Sstevel@tonic-gate mutex_enter(&freemem_lock); 31537c478bd9Sstevel@tonic-gate availrmem += pglcks; 31547c478bd9Sstevel@tonic-gate mutex_exit(&freemem_lock); 31557c478bd9Sstevel@tonic-gate } 31567c478bd9Sstevel@tonic-gate 31577c478bd9Sstevel@tonic-gate page_list_add_pages(rootpp, 0); 31587c478bd9Sstevel@tonic-gate page_create_putback(pgcnt); 31597c478bd9Sstevel@tonic-gate } 31607c478bd9Sstevel@tonic-gate 31617c478bd9Sstevel@tonic-gate /* 31627c478bd9Sstevel@tonic-gate * Similar to page_destroy(), but destroys pages which are 31637c478bd9Sstevel@tonic-gate * locked and known to be on the page free list. Since 31647c478bd9Sstevel@tonic-gate * the page is known to be free and locked, no one can access 31657c478bd9Sstevel@tonic-gate * it. 31667c478bd9Sstevel@tonic-gate * 31677c478bd9Sstevel@tonic-gate * Also, the number of free pages does not change. 31687c478bd9Sstevel@tonic-gate */ 31697c478bd9Sstevel@tonic-gate void 31707c478bd9Sstevel@tonic-gate page_destroy_free(page_t *pp) 31717c478bd9Sstevel@tonic-gate { 31727c478bd9Sstevel@tonic-gate ASSERT(PAGE_EXCL(pp)); 31737c478bd9Sstevel@tonic-gate ASSERT(PP_ISFREE(pp)); 31747c478bd9Sstevel@tonic-gate ASSERT(pp->p_vnode); 31757c478bd9Sstevel@tonic-gate ASSERT(hat_page_getattr(pp, P_MOD | P_REF | P_RO) == 0); 31767c478bd9Sstevel@tonic-gate ASSERT(!hat_page_is_mapped(pp)); 31777c478bd9Sstevel@tonic-gate ASSERT(PP_ISAGED(pp) == 0); 31787c478bd9Sstevel@tonic-gate ASSERT(pp->p_szc == 0); 31797c478bd9Sstevel@tonic-gate 31807c478bd9Sstevel@tonic-gate VM_STAT_ADD(pagecnt.pc_destroy_free); 31817c478bd9Sstevel@tonic-gate page_list_sub(pp, PG_CACHE_LIST); 31827c478bd9Sstevel@tonic-gate 31837c478bd9Sstevel@tonic-gate page_hashout(pp, NULL); 31847c478bd9Sstevel@tonic-gate ASSERT(pp->p_vnode == NULL); 31857c478bd9Sstevel@tonic-gate ASSERT(pp->p_offset == (u_offset_t)-1); 31867c478bd9Sstevel@tonic-gate ASSERT(pp->p_hash == NULL); 31877c478bd9Sstevel@tonic-gate 31887c478bd9Sstevel@tonic-gate PP_SETAGED(pp); 31897c478bd9Sstevel@tonic-gate page_list_add(pp, PG_FREE_LIST | PG_LIST_TAIL); 31907c478bd9Sstevel@tonic-gate page_unlock(pp); 31917c478bd9Sstevel@tonic-gate 31927c478bd9Sstevel@tonic-gate mutex_enter(&new_freemem_lock); 31937c478bd9Sstevel@tonic-gate if (freemem_wait) { 31947c478bd9Sstevel@tonic-gate cv_signal(&freemem_cv); 31957c478bd9Sstevel@tonic-gate } 31967c478bd9Sstevel@tonic-gate mutex_exit(&new_freemem_lock); 31977c478bd9Sstevel@tonic-gate } 31987c478bd9Sstevel@tonic-gate 31997c478bd9Sstevel@tonic-gate /* 32007c478bd9Sstevel@tonic-gate * Rename the page "opp" to have an identity specified 32017c478bd9Sstevel@tonic-gate * by [vp, off]. If a page already exists with this name 32027c478bd9Sstevel@tonic-gate * it is locked and destroyed. Note that the page's 32037c478bd9Sstevel@tonic-gate * translations are not unloaded during the rename. 32047c478bd9Sstevel@tonic-gate * 32057c478bd9Sstevel@tonic-gate * This routine is used by the anon layer to "steal" the 32067c478bd9Sstevel@tonic-gate * original page and is not unlike destroying a page and 32077c478bd9Sstevel@tonic-gate * creating a new page using the same page frame. 32087c478bd9Sstevel@tonic-gate * 32097c478bd9Sstevel@tonic-gate * XXX -- Could deadlock if caller 1 tries to rename A to B while 32107c478bd9Sstevel@tonic-gate * caller 2 tries to rename B to A. 32117c478bd9Sstevel@tonic-gate */ 32127c478bd9Sstevel@tonic-gate void 32137c478bd9Sstevel@tonic-gate page_rename(page_t *opp, vnode_t *vp, u_offset_t off) 32147c478bd9Sstevel@tonic-gate { 32157c478bd9Sstevel@tonic-gate page_t *pp; 32167c478bd9Sstevel@tonic-gate int olckcnt = 0; 32177c478bd9Sstevel@tonic-gate int ocowcnt = 0; 32187c478bd9Sstevel@tonic-gate kmutex_t *phm; 32197c478bd9Sstevel@tonic-gate ulong_t index; 32207c478bd9Sstevel@tonic-gate 32217c478bd9Sstevel@tonic-gate ASSERT(PAGE_EXCL(opp) && !page_iolock_assert(opp)); 32227c478bd9Sstevel@tonic-gate ASSERT(MUTEX_NOT_HELD(page_vnode_mutex(vp))); 32237c478bd9Sstevel@tonic-gate ASSERT(PP_ISFREE(opp) == 0); 32247c478bd9Sstevel@tonic-gate 32257c478bd9Sstevel@tonic-gate VM_STAT_ADD(page_rename_count); 32267c478bd9Sstevel@tonic-gate 32277c478bd9Sstevel@tonic-gate TRACE_3(TR_FAC_VM, TR_PAGE_RENAME, 32286e4dd838Smec "page rename:pp %p vp %p off %llx", opp, vp, off); 32297c478bd9Sstevel@tonic-gate 323037fbc076Saguzovsk /* 323137fbc076Saguzovsk * CacheFS may call page_rename for a large NFS page 323237fbc076Saguzovsk * when both CacheFS and NFS mount points are used 323337fbc076Saguzovsk * by applications. Demote this large page before 323437fbc076Saguzovsk * renaming it, to ensure that there are no "partial" 323537fbc076Saguzovsk * large pages left lying around. 323637fbc076Saguzovsk */ 323737fbc076Saguzovsk if (opp->p_szc != 0) { 323837fbc076Saguzovsk vnode_t *ovp = opp->p_vnode; 323937fbc076Saguzovsk ASSERT(ovp != NULL); 324037fbc076Saguzovsk ASSERT(!IS_SWAPFSVP(ovp)); 3241ad23a2dbSjohansen ASSERT(!VN_ISKAS(ovp)); 324237fbc076Saguzovsk page_demote_vp_pages(opp); 324337fbc076Saguzovsk ASSERT(opp->p_szc == 0); 324437fbc076Saguzovsk } 324537fbc076Saguzovsk 32467c478bd9Sstevel@tonic-gate page_hashout(opp, NULL); 32477c478bd9Sstevel@tonic-gate PP_CLRAGED(opp); 32487c478bd9Sstevel@tonic-gate 32497c478bd9Sstevel@tonic-gate /* 32507c478bd9Sstevel@tonic-gate * Acquire the appropriate page hash lock, since 32517c478bd9Sstevel@tonic-gate * we're going to rename the page. 32527c478bd9Sstevel@tonic-gate */ 32537c478bd9Sstevel@tonic-gate index = PAGE_HASH_FUNC(vp, off); 32547c478bd9Sstevel@tonic-gate phm = PAGE_HASH_MUTEX(index); 32557c478bd9Sstevel@tonic-gate mutex_enter(phm); 32567c478bd9Sstevel@tonic-gate top: 32577c478bd9Sstevel@tonic-gate /* 32587c478bd9Sstevel@tonic-gate * Look for an existing page with this name and destroy it if found. 32597c478bd9Sstevel@tonic-gate * By holding the page hash lock all the way to the page_hashin() 32607c478bd9Sstevel@tonic-gate * call, we are assured that no page can be created with this 32617c478bd9Sstevel@tonic-gate * identity. In the case when the phm lock is dropped to undo any 32627c478bd9Sstevel@tonic-gate * hat layer mappings, the existing page is held with an "exclusive" 32637c478bd9Sstevel@tonic-gate * lock, again preventing another page from being created with 32647c478bd9Sstevel@tonic-gate * this identity. 32657c478bd9Sstevel@tonic-gate */ 32667c478bd9Sstevel@tonic-gate PAGE_HASH_SEARCH(index, pp, vp, off); 32677c478bd9Sstevel@tonic-gate if (pp != NULL) { 32687c478bd9Sstevel@tonic-gate VM_STAT_ADD(page_rename_exists); 32697c478bd9Sstevel@tonic-gate 32707c478bd9Sstevel@tonic-gate /* 32717c478bd9Sstevel@tonic-gate * As it turns out, this is one of only two places where 32727c478bd9Sstevel@tonic-gate * page_lock() needs to hold the passed in lock in the 32737c478bd9Sstevel@tonic-gate * successful case. In all of the others, the lock could 32747c478bd9Sstevel@tonic-gate * be dropped as soon as the attempt is made to lock 32757c478bd9Sstevel@tonic-gate * the page. It is tempting to add yet another arguement, 32767c478bd9Sstevel@tonic-gate * PL_KEEP or PL_DROP, to let page_lock know what to do. 32777c478bd9Sstevel@tonic-gate */ 32787c478bd9Sstevel@tonic-gate if (!page_lock(pp, SE_EXCL, phm, P_RECLAIM)) { 32797c478bd9Sstevel@tonic-gate /* 32807c478bd9Sstevel@tonic-gate * Went to sleep because the page could not 32817c478bd9Sstevel@tonic-gate * be locked. We were woken up when the page 32827c478bd9Sstevel@tonic-gate * was unlocked, or when the page was destroyed. 32837c478bd9Sstevel@tonic-gate * In either case, `phm' was dropped while we 32847c478bd9Sstevel@tonic-gate * slept. Hence we should not just roar through 32857c478bd9Sstevel@tonic-gate * this loop. 32867c478bd9Sstevel@tonic-gate */ 32877c478bd9Sstevel@tonic-gate goto top; 32887c478bd9Sstevel@tonic-gate } 32897c478bd9Sstevel@tonic-gate 329037fbc076Saguzovsk /* 329137fbc076Saguzovsk * If an existing page is a large page, then demote 329237fbc076Saguzovsk * it to ensure that no "partial" large pages are 329337fbc076Saguzovsk * "created" after page_rename. An existing page 329437fbc076Saguzovsk * can be a CacheFS page, and can't belong to swapfs. 329537fbc076Saguzovsk */ 32967c478bd9Sstevel@tonic-gate if (hat_page_is_mapped(pp)) { 32977c478bd9Sstevel@tonic-gate /* 32987c478bd9Sstevel@tonic-gate * Unload translations. Since we hold the 32997c478bd9Sstevel@tonic-gate * exclusive lock on this page, the page 33007c478bd9Sstevel@tonic-gate * can not be changed while we drop phm. 33017c478bd9Sstevel@tonic-gate * This is also not a lock protocol violation, 33027c478bd9Sstevel@tonic-gate * but rather the proper way to do things. 33037c478bd9Sstevel@tonic-gate */ 33047c478bd9Sstevel@tonic-gate mutex_exit(phm); 33057c478bd9Sstevel@tonic-gate (void) hat_pageunload(pp, HAT_FORCE_PGUNLOAD); 330637fbc076Saguzovsk if (pp->p_szc != 0) { 330737fbc076Saguzovsk ASSERT(!IS_SWAPFSVP(vp)); 3308ad23a2dbSjohansen ASSERT(!VN_ISKAS(vp)); 330937fbc076Saguzovsk page_demote_vp_pages(pp); 331037fbc076Saguzovsk ASSERT(pp->p_szc == 0); 331137fbc076Saguzovsk } 331237fbc076Saguzovsk mutex_enter(phm); 331337fbc076Saguzovsk } else if (pp->p_szc != 0) { 331437fbc076Saguzovsk ASSERT(!IS_SWAPFSVP(vp)); 3315ad23a2dbSjohansen ASSERT(!VN_ISKAS(vp)); 331637fbc076Saguzovsk mutex_exit(phm); 331737fbc076Saguzovsk page_demote_vp_pages(pp); 331837fbc076Saguzovsk ASSERT(pp->p_szc == 0); 33197c478bd9Sstevel@tonic-gate mutex_enter(phm); 33207c478bd9Sstevel@tonic-gate } 33217c478bd9Sstevel@tonic-gate page_hashout(pp, phm); 33227c478bd9Sstevel@tonic-gate } 33237c478bd9Sstevel@tonic-gate /* 33247c478bd9Sstevel@tonic-gate * Hash in the page with the new identity. 33257c478bd9Sstevel@tonic-gate */ 33267c478bd9Sstevel@tonic-gate if (!page_hashin(opp, vp, off, phm)) { 33277c478bd9Sstevel@tonic-gate /* 33287c478bd9Sstevel@tonic-gate * We were holding phm while we searched for [vp, off] 33297c478bd9Sstevel@tonic-gate * and only dropped phm if we found and locked a page. 33307c478bd9Sstevel@tonic-gate * If we can't create this page now, then some thing 33317c478bd9Sstevel@tonic-gate * is really broken. 33327c478bd9Sstevel@tonic-gate */ 33337c478bd9Sstevel@tonic-gate panic("page_rename: Can't hash in page: %p", (void *)pp); 33347c478bd9Sstevel@tonic-gate /*NOTREACHED*/ 33357c478bd9Sstevel@tonic-gate } 33367c478bd9Sstevel@tonic-gate 33377c478bd9Sstevel@tonic-gate ASSERT(MUTEX_HELD(phm)); 33387c478bd9Sstevel@tonic-gate mutex_exit(phm); 33397c478bd9Sstevel@tonic-gate 33407c478bd9Sstevel@tonic-gate /* 33417c478bd9Sstevel@tonic-gate * Now that we have dropped phm, lets get around to finishing up 33427c478bd9Sstevel@tonic-gate * with pp. 33437c478bd9Sstevel@tonic-gate */ 33447c478bd9Sstevel@tonic-gate if (pp != NULL) { 33457c478bd9Sstevel@tonic-gate ASSERT(!hat_page_is_mapped(pp)); 33467c478bd9Sstevel@tonic-gate /* for now large pages should not end up here */ 33477c478bd9Sstevel@tonic-gate ASSERT(pp->p_szc == 0); 33487c478bd9Sstevel@tonic-gate /* 33497c478bd9Sstevel@tonic-gate * Save the locks for transfer to the new page and then 33507c478bd9Sstevel@tonic-gate * clear them so page_free doesn't think they're important. 33517c478bd9Sstevel@tonic-gate * The page_struct_lock need not be acquired for lckcnt and 33527c478bd9Sstevel@tonic-gate * cowcnt since the page has an "exclusive" lock. 33537c478bd9Sstevel@tonic-gate */ 33547c478bd9Sstevel@tonic-gate olckcnt = pp->p_lckcnt; 33557c478bd9Sstevel@tonic-gate ocowcnt = pp->p_cowcnt; 33567c478bd9Sstevel@tonic-gate pp->p_lckcnt = pp->p_cowcnt = 0; 33577c478bd9Sstevel@tonic-gate 33587c478bd9Sstevel@tonic-gate /* 33597c478bd9Sstevel@tonic-gate * Put the page on the "free" list after we drop 33607c478bd9Sstevel@tonic-gate * the lock. The less work under the lock the better. 33617c478bd9Sstevel@tonic-gate */ 33627c478bd9Sstevel@tonic-gate /*LINTED: constant in conditional context*/ 33637c478bd9Sstevel@tonic-gate VN_DISPOSE(pp, B_FREE, 0, kcred); 33647c478bd9Sstevel@tonic-gate } 33657c478bd9Sstevel@tonic-gate 33667c478bd9Sstevel@tonic-gate /* 33677c478bd9Sstevel@tonic-gate * Transfer the lock count from the old page (if any). 33687c478bd9Sstevel@tonic-gate * The page_struct_lock need not be acquired for lckcnt and 33697c478bd9Sstevel@tonic-gate * cowcnt since the page has an "exclusive" lock. 33707c478bd9Sstevel@tonic-gate */ 33717c478bd9Sstevel@tonic-gate opp->p_lckcnt += olckcnt; 33727c478bd9Sstevel@tonic-gate opp->p_cowcnt += ocowcnt; 33737c478bd9Sstevel@tonic-gate } 33747c478bd9Sstevel@tonic-gate 33757c478bd9Sstevel@tonic-gate /* 33767c478bd9Sstevel@tonic-gate * low level routine to add page `pp' to the hash and vp chains for [vp, offset] 33777c478bd9Sstevel@tonic-gate * 33787c478bd9Sstevel@tonic-gate * Pages are normally inserted at the start of a vnode's v_pages list. 33797c478bd9Sstevel@tonic-gate * If the vnode is VMODSORT and the page is modified, it goes at the end. 33807c478bd9Sstevel@tonic-gate * This can happen when a modified page is relocated for DR. 33817c478bd9Sstevel@tonic-gate * 33827c478bd9Sstevel@tonic-gate * Returns 1 on success and 0 on failure. 33837c478bd9Sstevel@tonic-gate */ 33847c478bd9Sstevel@tonic-gate static int 33857c478bd9Sstevel@tonic-gate page_do_hashin(page_t *pp, vnode_t *vp, u_offset_t offset) 33867c478bd9Sstevel@tonic-gate { 33877c478bd9Sstevel@tonic-gate page_t **listp; 33887c478bd9Sstevel@tonic-gate page_t *tp; 33897c478bd9Sstevel@tonic-gate ulong_t index; 33907c478bd9Sstevel@tonic-gate 33917c478bd9Sstevel@tonic-gate ASSERT(PAGE_EXCL(pp)); 33927c478bd9Sstevel@tonic-gate ASSERT(vp != NULL); 33937c478bd9Sstevel@tonic-gate ASSERT(MUTEX_HELD(page_vnode_mutex(vp))); 33947c478bd9Sstevel@tonic-gate 33957c478bd9Sstevel@tonic-gate /* 33967c478bd9Sstevel@tonic-gate * Be sure to set these up before the page is inserted on the hash 33977c478bd9Sstevel@tonic-gate * list. As soon as the page is placed on the list some other 33987c478bd9Sstevel@tonic-gate * thread might get confused and wonder how this page could 33997c478bd9Sstevel@tonic-gate * possibly hash to this list. 34007c478bd9Sstevel@tonic-gate */ 34017c478bd9Sstevel@tonic-gate pp->p_vnode = vp; 34027c478bd9Sstevel@tonic-gate pp->p_offset = offset; 34037c478bd9Sstevel@tonic-gate 34047c478bd9Sstevel@tonic-gate /* 34057c478bd9Sstevel@tonic-gate * record if this page is on a swap vnode 34067c478bd9Sstevel@tonic-gate */ 34077c478bd9Sstevel@tonic-gate if ((vp->v_flag & VISSWAP) != 0) 34087c478bd9Sstevel@tonic-gate PP_SETSWAP(pp); 34097c478bd9Sstevel@tonic-gate 34107c478bd9Sstevel@tonic-gate index = PAGE_HASH_FUNC(vp, offset); 34117c478bd9Sstevel@tonic-gate ASSERT(MUTEX_HELD(PAGE_HASH_MUTEX(index))); 34127c478bd9Sstevel@tonic-gate listp = &page_hash[index]; 34137c478bd9Sstevel@tonic-gate 34147c478bd9Sstevel@tonic-gate /* 34157c478bd9Sstevel@tonic-gate * If this page is already hashed in, fail this attempt to add it. 34167c478bd9Sstevel@tonic-gate */ 34177c478bd9Sstevel@tonic-gate for (tp = *listp; tp != NULL; tp = tp->p_hash) { 34187c478bd9Sstevel@tonic-gate if (tp->p_vnode == vp && tp->p_offset == offset) { 34197c478bd9Sstevel@tonic-gate pp->p_vnode = NULL; 34207c478bd9Sstevel@tonic-gate pp->p_offset = (u_offset_t)(-1); 34217c478bd9Sstevel@tonic-gate return (0); 34227c478bd9Sstevel@tonic-gate } 34237c478bd9Sstevel@tonic-gate } 34247c478bd9Sstevel@tonic-gate pp->p_hash = *listp; 34257c478bd9Sstevel@tonic-gate *listp = pp; 34267c478bd9Sstevel@tonic-gate 34277c478bd9Sstevel@tonic-gate /* 34287c478bd9Sstevel@tonic-gate * Add the page to the vnode's list of pages 34297c478bd9Sstevel@tonic-gate */ 34307c478bd9Sstevel@tonic-gate if (vp->v_pages != NULL && IS_VMODSORT(vp) && hat_ismod(pp)) 34317c478bd9Sstevel@tonic-gate listp = &vp->v_pages->p_vpprev->p_vpnext; 34327c478bd9Sstevel@tonic-gate else 34337c478bd9Sstevel@tonic-gate listp = &vp->v_pages; 34347c478bd9Sstevel@tonic-gate 34357c478bd9Sstevel@tonic-gate page_vpadd(listp, pp); 34367c478bd9Sstevel@tonic-gate 34377c478bd9Sstevel@tonic-gate return (1); 34387c478bd9Sstevel@tonic-gate } 34397c478bd9Sstevel@tonic-gate 34407c478bd9Sstevel@tonic-gate /* 34417c478bd9Sstevel@tonic-gate * Add page `pp' to both the hash and vp chains for [vp, offset]. 34427c478bd9Sstevel@tonic-gate * 34437c478bd9Sstevel@tonic-gate * Returns 1 on success and 0 on failure. 34447c478bd9Sstevel@tonic-gate * If hold is passed in, it is not dropped. 34457c478bd9Sstevel@tonic-gate */ 34467c478bd9Sstevel@tonic-gate int 34477c478bd9Sstevel@tonic-gate page_hashin(page_t *pp, vnode_t *vp, u_offset_t offset, kmutex_t *hold) 34487c478bd9Sstevel@tonic-gate { 34497c478bd9Sstevel@tonic-gate kmutex_t *phm = NULL; 34507c478bd9Sstevel@tonic-gate kmutex_t *vphm; 34517c478bd9Sstevel@tonic-gate int rc; 34527c478bd9Sstevel@tonic-gate 34537c478bd9Sstevel@tonic-gate ASSERT(MUTEX_NOT_HELD(page_vnode_mutex(vp))); 34547c478bd9Sstevel@tonic-gate 34557c478bd9Sstevel@tonic-gate TRACE_3(TR_FAC_VM, TR_PAGE_HASHIN, 34566e4dd838Smec "page_hashin:pp %p vp %p offset %llx", 34576e4dd838Smec pp, vp, offset); 34587c478bd9Sstevel@tonic-gate 34597c478bd9Sstevel@tonic-gate VM_STAT_ADD(hashin_count); 34607c478bd9Sstevel@tonic-gate 34617c478bd9Sstevel@tonic-gate if (hold != NULL) 34627c478bd9Sstevel@tonic-gate phm = hold; 34637c478bd9Sstevel@tonic-gate else { 34647c478bd9Sstevel@tonic-gate VM_STAT_ADD(hashin_not_held); 34657c478bd9Sstevel@tonic-gate phm = PAGE_HASH_MUTEX(PAGE_HASH_FUNC(vp, offset)); 34667c478bd9Sstevel@tonic-gate mutex_enter(phm); 34677c478bd9Sstevel@tonic-gate } 34687c478bd9Sstevel@tonic-gate 34697c478bd9Sstevel@tonic-gate vphm = page_vnode_mutex(vp); 34707c478bd9Sstevel@tonic-gate mutex_enter(vphm); 34717c478bd9Sstevel@tonic-gate rc = page_do_hashin(pp, vp, offset); 34727c478bd9Sstevel@tonic-gate mutex_exit(vphm); 34737c478bd9Sstevel@tonic-gate if (hold == NULL) 34747c478bd9Sstevel@tonic-gate mutex_exit(phm); 34757c478bd9Sstevel@tonic-gate if (rc == 0) 34767c478bd9Sstevel@tonic-gate VM_STAT_ADD(hashin_already); 34777c478bd9Sstevel@tonic-gate return (rc); 34787c478bd9Sstevel@tonic-gate } 34797c478bd9Sstevel@tonic-gate 34807c478bd9Sstevel@tonic-gate /* 34817c478bd9Sstevel@tonic-gate * Remove page ``pp'' from the hash and vp chains and remove vp association. 34827c478bd9Sstevel@tonic-gate * All mutexes must be held 34837c478bd9Sstevel@tonic-gate */ 34847c478bd9Sstevel@tonic-gate static void 34857c478bd9Sstevel@tonic-gate page_do_hashout(page_t *pp) 34867c478bd9Sstevel@tonic-gate { 34877c478bd9Sstevel@tonic-gate page_t **hpp; 34887c478bd9Sstevel@tonic-gate page_t *hp; 34897c478bd9Sstevel@tonic-gate vnode_t *vp = pp->p_vnode; 34907c478bd9Sstevel@tonic-gate 34917c478bd9Sstevel@tonic-gate ASSERT(vp != NULL); 34927c478bd9Sstevel@tonic-gate ASSERT(MUTEX_HELD(page_vnode_mutex(vp))); 34937c478bd9Sstevel@tonic-gate 34947c478bd9Sstevel@tonic-gate /* 34957c478bd9Sstevel@tonic-gate * First, take pp off of its hash chain. 34967c478bd9Sstevel@tonic-gate */ 34977c478bd9Sstevel@tonic-gate hpp = &page_hash[PAGE_HASH_FUNC(vp, pp->p_offset)]; 34987c478bd9Sstevel@tonic-gate 34997c478bd9Sstevel@tonic-gate for (;;) { 35007c478bd9Sstevel@tonic-gate hp = *hpp; 35017c478bd9Sstevel@tonic-gate if (hp == pp) 35027c478bd9Sstevel@tonic-gate break; 35037c478bd9Sstevel@tonic-gate if (hp == NULL) { 35047c478bd9Sstevel@tonic-gate panic("page_do_hashout"); 35057c478bd9Sstevel@tonic-gate /*NOTREACHED*/ 35067c478bd9Sstevel@tonic-gate } 35077c478bd9Sstevel@tonic-gate hpp = &hp->p_hash; 35087c478bd9Sstevel@tonic-gate } 35097c478bd9Sstevel@tonic-gate *hpp = pp->p_hash; 35107c478bd9Sstevel@tonic-gate 35117c478bd9Sstevel@tonic-gate /* 35127c478bd9Sstevel@tonic-gate * Now remove it from its associated vnode. 35137c478bd9Sstevel@tonic-gate */ 35147c478bd9Sstevel@tonic-gate if (vp->v_pages) 35157c478bd9Sstevel@tonic-gate page_vpsub(&vp->v_pages, pp); 35167c478bd9Sstevel@tonic-gate 35177c478bd9Sstevel@tonic-gate pp->p_hash = NULL; 35187c478bd9Sstevel@tonic-gate page_clr_all_props(pp); 35197c478bd9Sstevel@tonic-gate PP_CLRSWAP(pp); 35207c478bd9Sstevel@tonic-gate pp->p_vnode = NULL; 35217c478bd9Sstevel@tonic-gate pp->p_offset = (u_offset_t)-1; 35227c478bd9Sstevel@tonic-gate } 35237c478bd9Sstevel@tonic-gate 35247c478bd9Sstevel@tonic-gate /* 35257c478bd9Sstevel@tonic-gate * Remove page ``pp'' from the hash and vp chains and remove vp association. 35267c478bd9Sstevel@tonic-gate * 35277c478bd9Sstevel@tonic-gate * When `phm' is non-NULL it contains the address of the mutex protecting the 35287c478bd9Sstevel@tonic-gate * hash list pp is on. It is not dropped. 35297c478bd9Sstevel@tonic-gate */ 35307c478bd9Sstevel@tonic-gate void 35317c478bd9Sstevel@tonic-gate page_hashout(page_t *pp, kmutex_t *phm) 35327c478bd9Sstevel@tonic-gate { 35337c478bd9Sstevel@tonic-gate vnode_t *vp; 35347c478bd9Sstevel@tonic-gate ulong_t index; 35357c478bd9Sstevel@tonic-gate kmutex_t *nphm; 35367c478bd9Sstevel@tonic-gate kmutex_t *vphm; 35377c478bd9Sstevel@tonic-gate kmutex_t *sep; 35387c478bd9Sstevel@tonic-gate 35397c478bd9Sstevel@tonic-gate ASSERT(phm != NULL ? MUTEX_HELD(phm) : 1); 35407c478bd9Sstevel@tonic-gate ASSERT(pp->p_vnode != NULL); 35417c478bd9Sstevel@tonic-gate ASSERT((PAGE_EXCL(pp) && !page_iolock_assert(pp)) || panicstr); 35427c478bd9Sstevel@tonic-gate ASSERT(MUTEX_NOT_HELD(page_vnode_mutex(pp->p_vnode))); 35437c478bd9Sstevel@tonic-gate 35447c478bd9Sstevel@tonic-gate vp = pp->p_vnode; 35457c478bd9Sstevel@tonic-gate 35467c478bd9Sstevel@tonic-gate TRACE_2(TR_FAC_VM, TR_PAGE_HASHOUT, 35476e4dd838Smec "page_hashout:pp %p vp %p", pp, vp); 35487c478bd9Sstevel@tonic-gate 35497c478bd9Sstevel@tonic-gate /* Kernel probe */ 35507c478bd9Sstevel@tonic-gate TNF_PROBE_2(page_unmap, "vm pagefault", /* CSTYLED */, 35517c478bd9Sstevel@tonic-gate tnf_opaque, vnode, vp, 35527c478bd9Sstevel@tonic-gate tnf_offset, offset, pp->p_offset); 35537c478bd9Sstevel@tonic-gate 35547c478bd9Sstevel@tonic-gate /* 35557c478bd9Sstevel@tonic-gate * 35567c478bd9Sstevel@tonic-gate */ 35577c478bd9Sstevel@tonic-gate VM_STAT_ADD(hashout_count); 35587c478bd9Sstevel@tonic-gate index = PAGE_HASH_FUNC(vp, pp->p_offset); 35597c478bd9Sstevel@tonic-gate if (phm == NULL) { 35607c478bd9Sstevel@tonic-gate VM_STAT_ADD(hashout_not_held); 35617c478bd9Sstevel@tonic-gate nphm = PAGE_HASH_MUTEX(index); 35627c478bd9Sstevel@tonic-gate mutex_enter(nphm); 35637c478bd9Sstevel@tonic-gate } 35647c478bd9Sstevel@tonic-gate ASSERT(phm ? phm == PAGE_HASH_MUTEX(index) : 1); 35657c478bd9Sstevel@tonic-gate 35667c478bd9Sstevel@tonic-gate 35677c478bd9Sstevel@tonic-gate /* 35687c478bd9Sstevel@tonic-gate * grab page vnode mutex and remove it... 35697c478bd9Sstevel@tonic-gate */ 35707c478bd9Sstevel@tonic-gate vphm = page_vnode_mutex(vp); 35717c478bd9Sstevel@tonic-gate mutex_enter(vphm); 35727c478bd9Sstevel@tonic-gate 35737c478bd9Sstevel@tonic-gate page_do_hashout(pp); 35747c478bd9Sstevel@tonic-gate 35757c478bd9Sstevel@tonic-gate mutex_exit(vphm); 35767c478bd9Sstevel@tonic-gate if (phm == NULL) 35777c478bd9Sstevel@tonic-gate mutex_exit(nphm); 35787c478bd9Sstevel@tonic-gate 35797c478bd9Sstevel@tonic-gate /* 35807c478bd9Sstevel@tonic-gate * Wake up processes waiting for this page. The page's 35817c478bd9Sstevel@tonic-gate * identity has been changed, and is probably not the 35827c478bd9Sstevel@tonic-gate * desired page any longer. 35837c478bd9Sstevel@tonic-gate */ 35847c478bd9Sstevel@tonic-gate sep = page_se_mutex(pp); 35857c478bd9Sstevel@tonic-gate mutex_enter(sep); 358642787a71Sstans pp->p_selock &= ~SE_EWANTED; 35877c478bd9Sstevel@tonic-gate if (CV_HAS_WAITERS(&pp->p_cv)) 35887c478bd9Sstevel@tonic-gate cv_broadcast(&pp->p_cv); 35897c478bd9Sstevel@tonic-gate mutex_exit(sep); 35907c478bd9Sstevel@tonic-gate } 35917c478bd9Sstevel@tonic-gate 35927c478bd9Sstevel@tonic-gate /* 35937c478bd9Sstevel@tonic-gate * Add the page to the front of a linked list of pages 35947c478bd9Sstevel@tonic-gate * using the p_next & p_prev pointers for the list. 35957c478bd9Sstevel@tonic-gate * The caller is responsible for protecting the list pointers. 35967c478bd9Sstevel@tonic-gate */ 35977c478bd9Sstevel@tonic-gate void 35987c478bd9Sstevel@tonic-gate page_add(page_t **ppp, page_t *pp) 35997c478bd9Sstevel@tonic-gate { 36007c478bd9Sstevel@tonic-gate ASSERT(PAGE_EXCL(pp) || (PAGE_SHARED(pp) && page_iolock_assert(pp))); 36017c478bd9Sstevel@tonic-gate 36027c478bd9Sstevel@tonic-gate page_add_common(ppp, pp); 36037c478bd9Sstevel@tonic-gate } 36047c478bd9Sstevel@tonic-gate 36057c478bd9Sstevel@tonic-gate 36067c478bd9Sstevel@tonic-gate 36077c478bd9Sstevel@tonic-gate /* 36087c478bd9Sstevel@tonic-gate * Common code for page_add() and mach_page_add() 36097c478bd9Sstevel@tonic-gate */ 36107c478bd9Sstevel@tonic-gate void 36117c478bd9Sstevel@tonic-gate page_add_common(page_t **ppp, page_t *pp) 36127c478bd9Sstevel@tonic-gate { 36137c478bd9Sstevel@tonic-gate if (*ppp == NULL) { 36147c478bd9Sstevel@tonic-gate pp->p_next = pp->p_prev = pp; 36157c478bd9Sstevel@tonic-gate } else { 36167c478bd9Sstevel@tonic-gate pp->p_next = *ppp; 36177c478bd9Sstevel@tonic-gate pp->p_prev = (*ppp)->p_prev; 36187c478bd9Sstevel@tonic-gate (*ppp)->p_prev = pp; 36197c478bd9Sstevel@tonic-gate pp->p_prev->p_next = pp; 36207c478bd9Sstevel@tonic-gate } 36217c478bd9Sstevel@tonic-gate *ppp = pp; 36227c478bd9Sstevel@tonic-gate } 36237c478bd9Sstevel@tonic-gate 36247c478bd9Sstevel@tonic-gate 36257c478bd9Sstevel@tonic-gate /* 36267c478bd9Sstevel@tonic-gate * Remove this page from a linked list of pages 36277c478bd9Sstevel@tonic-gate * using the p_next & p_prev pointers for the list. 36287c478bd9Sstevel@tonic-gate * 36297c478bd9Sstevel@tonic-gate * The caller is responsible for protecting the list pointers. 36307c478bd9Sstevel@tonic-gate */ 36317c478bd9Sstevel@tonic-gate void 36327c478bd9Sstevel@tonic-gate page_sub(page_t **ppp, page_t *pp) 36337c478bd9Sstevel@tonic-gate { 36347c478bd9Sstevel@tonic-gate ASSERT((PP_ISFREE(pp)) ? 1 : 36357c478bd9Sstevel@tonic-gate (PAGE_EXCL(pp)) || (PAGE_SHARED(pp) && page_iolock_assert(pp))); 36367c478bd9Sstevel@tonic-gate 36377c478bd9Sstevel@tonic-gate if (*ppp == NULL || pp == NULL) { 36387c478bd9Sstevel@tonic-gate panic("page_sub: bad arg(s): pp %p, *ppp %p", 36397c478bd9Sstevel@tonic-gate (void *)pp, (void *)(*ppp)); 36407c478bd9Sstevel@tonic-gate /*NOTREACHED*/ 36417c478bd9Sstevel@tonic-gate } 36427c478bd9Sstevel@tonic-gate 36437c478bd9Sstevel@tonic-gate page_sub_common(ppp, pp); 36447c478bd9Sstevel@tonic-gate } 36457c478bd9Sstevel@tonic-gate 36467c478bd9Sstevel@tonic-gate 36477c478bd9Sstevel@tonic-gate /* 36487c478bd9Sstevel@tonic-gate * Common code for page_sub() and mach_page_sub() 36497c478bd9Sstevel@tonic-gate */ 36507c478bd9Sstevel@tonic-gate void 36517c478bd9Sstevel@tonic-gate page_sub_common(page_t **ppp, page_t *pp) 36527c478bd9Sstevel@tonic-gate { 36537c478bd9Sstevel@tonic-gate if (*ppp == pp) 36547c478bd9Sstevel@tonic-gate *ppp = pp->p_next; /* go to next page */ 36557c478bd9Sstevel@tonic-gate 36567c478bd9Sstevel@tonic-gate if (*ppp == pp) 36577c478bd9Sstevel@tonic-gate *ppp = NULL; /* page list is gone */ 36587c478bd9Sstevel@tonic-gate else { 36597c478bd9Sstevel@tonic-gate pp->p_prev->p_next = pp->p_next; 36607c478bd9Sstevel@tonic-gate pp->p_next->p_prev = pp->p_prev; 36617c478bd9Sstevel@tonic-gate } 36627c478bd9Sstevel@tonic-gate pp->p_prev = pp->p_next = pp; /* make pp a list of one */ 36637c478bd9Sstevel@tonic-gate } 36647c478bd9Sstevel@tonic-gate 36657c478bd9Sstevel@tonic-gate 36667c478bd9Sstevel@tonic-gate /* 36677c478bd9Sstevel@tonic-gate * Break page list cppp into two lists with npages in the first list. 36687c478bd9Sstevel@tonic-gate * The tail is returned in nppp. 36697c478bd9Sstevel@tonic-gate */ 36707c478bd9Sstevel@tonic-gate void 36717c478bd9Sstevel@tonic-gate page_list_break(page_t **oppp, page_t **nppp, pgcnt_t npages) 36727c478bd9Sstevel@tonic-gate { 36737c478bd9Sstevel@tonic-gate page_t *s1pp = *oppp; 36747c478bd9Sstevel@tonic-gate page_t *s2pp; 36757c478bd9Sstevel@tonic-gate page_t *e1pp, *e2pp; 36767c478bd9Sstevel@tonic-gate long n = 0; 36777c478bd9Sstevel@tonic-gate 36787c478bd9Sstevel@tonic-gate if (s1pp == NULL) { 36797c478bd9Sstevel@tonic-gate *nppp = NULL; 36807c478bd9Sstevel@tonic-gate return; 36817c478bd9Sstevel@tonic-gate } 36827c478bd9Sstevel@tonic-gate if (npages == 0) { 36837c478bd9Sstevel@tonic-gate *nppp = s1pp; 36847c478bd9Sstevel@tonic-gate *oppp = NULL; 36857c478bd9Sstevel@tonic-gate return; 36867c478bd9Sstevel@tonic-gate } 36877c478bd9Sstevel@tonic-gate for (n = 0, s2pp = *oppp; n < npages; n++) { 36887c478bd9Sstevel@tonic-gate s2pp = s2pp->p_next; 36897c478bd9Sstevel@tonic-gate } 36907c478bd9Sstevel@tonic-gate /* Fix head and tail of new lists */ 36917c478bd9Sstevel@tonic-gate e1pp = s2pp->p_prev; 36927c478bd9Sstevel@tonic-gate e2pp = s1pp->p_prev; 36937c478bd9Sstevel@tonic-gate s1pp->p_prev = e1pp; 36947c478bd9Sstevel@tonic-gate e1pp->p_next = s1pp; 36957c478bd9Sstevel@tonic-gate s2pp->p_prev = e2pp; 36967c478bd9Sstevel@tonic-gate e2pp->p_next = s2pp; 36977c478bd9Sstevel@tonic-gate 36987c478bd9Sstevel@tonic-gate /* second list empty */ 36997c478bd9Sstevel@tonic-gate if (s2pp == s1pp) { 37007c478bd9Sstevel@tonic-gate *oppp = s1pp; 37017c478bd9Sstevel@tonic-gate *nppp = NULL; 37027c478bd9Sstevel@tonic-gate } else { 37037c478bd9Sstevel@tonic-gate *oppp = s1pp; 37047c478bd9Sstevel@tonic-gate *nppp = s2pp; 37057c478bd9Sstevel@tonic-gate } 37067c478bd9Sstevel@tonic-gate } 37077c478bd9Sstevel@tonic-gate 37087c478bd9Sstevel@tonic-gate /* 37097c478bd9Sstevel@tonic-gate * Concatenate page list nppp onto the end of list ppp. 37107c478bd9Sstevel@tonic-gate */ 37117c478bd9Sstevel@tonic-gate void 37127c478bd9Sstevel@tonic-gate page_list_concat(page_t **ppp, page_t **nppp) 37137c478bd9Sstevel@tonic-gate { 37147c478bd9Sstevel@tonic-gate page_t *s1pp, *s2pp, *e1pp, *e2pp; 37157c478bd9Sstevel@tonic-gate 37167c478bd9Sstevel@tonic-gate if (*nppp == NULL) { 37177c478bd9Sstevel@tonic-gate return; 37187c478bd9Sstevel@tonic-gate } 37197c478bd9Sstevel@tonic-gate if (*ppp == NULL) { 37207c478bd9Sstevel@tonic-gate *ppp = *nppp; 37217c478bd9Sstevel@tonic-gate return; 37227c478bd9Sstevel@tonic-gate } 37237c478bd9Sstevel@tonic-gate s1pp = *ppp; 37247c478bd9Sstevel@tonic-gate e1pp = s1pp->p_prev; 37257c478bd9Sstevel@tonic-gate s2pp = *nppp; 37267c478bd9Sstevel@tonic-gate e2pp = s2pp->p_prev; 37277c478bd9Sstevel@tonic-gate s1pp->p_prev = e2pp; 37287c478bd9Sstevel@tonic-gate e2pp->p_next = s1pp; 37297c478bd9Sstevel@tonic-gate e1pp->p_next = s2pp; 37307c478bd9Sstevel@tonic-gate s2pp->p_prev = e1pp; 37317c478bd9Sstevel@tonic-gate } 37327c478bd9Sstevel@tonic-gate 37337c478bd9Sstevel@tonic-gate /* 37347c478bd9Sstevel@tonic-gate * return the next page in the page list 37357c478bd9Sstevel@tonic-gate */ 37367c478bd9Sstevel@tonic-gate page_t * 37377c478bd9Sstevel@tonic-gate page_list_next(page_t *pp) 37387c478bd9Sstevel@tonic-gate { 37397c478bd9Sstevel@tonic-gate return (pp->p_next); 37407c478bd9Sstevel@tonic-gate } 37417c478bd9Sstevel@tonic-gate 37427c478bd9Sstevel@tonic-gate 37437c478bd9Sstevel@tonic-gate /* 37447c478bd9Sstevel@tonic-gate * Add the page to the front of the linked list of pages 37457c478bd9Sstevel@tonic-gate * using p_vpnext/p_vpprev pointers for the list. 37467c478bd9Sstevel@tonic-gate * 37477c478bd9Sstevel@tonic-gate * The caller is responsible for protecting the lists. 37487c478bd9Sstevel@tonic-gate */ 37497c478bd9Sstevel@tonic-gate void 37507c478bd9Sstevel@tonic-gate page_vpadd(page_t **ppp, page_t *pp) 37517c478bd9Sstevel@tonic-gate { 37527c478bd9Sstevel@tonic-gate if (*ppp == NULL) { 37537c478bd9Sstevel@tonic-gate pp->p_vpnext = pp->p_vpprev = pp; 37547c478bd9Sstevel@tonic-gate } else { 37557c478bd9Sstevel@tonic-gate pp->p_vpnext = *ppp; 37567c478bd9Sstevel@tonic-gate pp->p_vpprev = (*ppp)->p_vpprev; 37577c478bd9Sstevel@tonic-gate (*ppp)->p_vpprev = pp; 37587c478bd9Sstevel@tonic-gate pp->p_vpprev->p_vpnext = pp; 37597c478bd9Sstevel@tonic-gate } 37607c478bd9Sstevel@tonic-gate *ppp = pp; 37617c478bd9Sstevel@tonic-gate } 37627c478bd9Sstevel@tonic-gate 37637c478bd9Sstevel@tonic-gate /* 37647c478bd9Sstevel@tonic-gate * Remove this page from the linked list of pages 37657c478bd9Sstevel@tonic-gate * using p_vpnext/p_vpprev pointers for the list. 37667c478bd9Sstevel@tonic-gate * 37677c478bd9Sstevel@tonic-gate * The caller is responsible for protecting the lists. 37687c478bd9Sstevel@tonic-gate */ 37697c478bd9Sstevel@tonic-gate void 37707c478bd9Sstevel@tonic-gate page_vpsub(page_t **ppp, page_t *pp) 37717c478bd9Sstevel@tonic-gate { 37727c478bd9Sstevel@tonic-gate if (*ppp == NULL || pp == NULL) { 37737c478bd9Sstevel@tonic-gate panic("page_vpsub: bad arg(s): pp %p, *ppp %p", 37747c478bd9Sstevel@tonic-gate (void *)pp, (void *)(*ppp)); 37757c478bd9Sstevel@tonic-gate /*NOTREACHED*/ 37767c478bd9Sstevel@tonic-gate } 37777c478bd9Sstevel@tonic-gate 37787c478bd9Sstevel@tonic-gate if (*ppp == pp) 37797c478bd9Sstevel@tonic-gate *ppp = pp->p_vpnext; /* go to next page */ 37807c478bd9Sstevel@tonic-gate 37817c478bd9Sstevel@tonic-gate if (*ppp == pp) 37827c478bd9Sstevel@tonic-gate *ppp = NULL; /* page list is gone */ 37837c478bd9Sstevel@tonic-gate else { 37847c478bd9Sstevel@tonic-gate pp->p_vpprev->p_vpnext = pp->p_vpnext; 37857c478bd9Sstevel@tonic-gate pp->p_vpnext->p_vpprev = pp->p_vpprev; 37867c478bd9Sstevel@tonic-gate } 37877c478bd9Sstevel@tonic-gate pp->p_vpprev = pp->p_vpnext = pp; /* make pp a list of one */ 37887c478bd9Sstevel@tonic-gate } 37897c478bd9Sstevel@tonic-gate 37907c478bd9Sstevel@tonic-gate /* 37917c478bd9Sstevel@tonic-gate * Lock a physical page into memory "long term". Used to support "lock 37927c478bd9Sstevel@tonic-gate * in memory" functions. Accepts the page to be locked, and a cow variable 37937c478bd9Sstevel@tonic-gate * to indicate whether a the lock will travel to the new page during 37947c478bd9Sstevel@tonic-gate * a potential copy-on-write. 37957c478bd9Sstevel@tonic-gate */ 37967c478bd9Sstevel@tonic-gate int 37977c478bd9Sstevel@tonic-gate page_pp_lock( 37987c478bd9Sstevel@tonic-gate page_t *pp, /* page to be locked */ 37997c478bd9Sstevel@tonic-gate int cow, /* cow lock */ 38007c478bd9Sstevel@tonic-gate int kernel) /* must succeed -- ignore checking */ 38017c478bd9Sstevel@tonic-gate { 38027c478bd9Sstevel@tonic-gate int r = 0; /* result -- assume failure */ 38037c478bd9Sstevel@tonic-gate 38047c478bd9Sstevel@tonic-gate ASSERT(PAGE_LOCKED(pp)); 38057c478bd9Sstevel@tonic-gate 38067c478bd9Sstevel@tonic-gate page_struct_lock(pp); 38077c478bd9Sstevel@tonic-gate /* 38087c478bd9Sstevel@tonic-gate * Acquire the "freemem_lock" for availrmem. 38097c478bd9Sstevel@tonic-gate */ 38107c478bd9Sstevel@tonic-gate if (cow) { 38117c478bd9Sstevel@tonic-gate mutex_enter(&freemem_lock); 38127c478bd9Sstevel@tonic-gate if ((availrmem > pages_pp_maximum) && 38137c478bd9Sstevel@tonic-gate (pp->p_cowcnt < (ushort_t)PAGE_LOCK_MAXIMUM)) { 38147c478bd9Sstevel@tonic-gate availrmem--; 38157c478bd9Sstevel@tonic-gate pages_locked++; 38167c478bd9Sstevel@tonic-gate mutex_exit(&freemem_lock); 38177c478bd9Sstevel@tonic-gate r = 1; 38187c478bd9Sstevel@tonic-gate if (++pp->p_cowcnt == (ushort_t)PAGE_LOCK_MAXIMUM) { 38197c478bd9Sstevel@tonic-gate cmn_err(CE_WARN, 38207c478bd9Sstevel@tonic-gate "COW lock limit reached on pfn 0x%lx", 38217c478bd9Sstevel@tonic-gate page_pptonum(pp)); 38227c478bd9Sstevel@tonic-gate } 38237c478bd9Sstevel@tonic-gate } else 38247c478bd9Sstevel@tonic-gate mutex_exit(&freemem_lock); 38257c478bd9Sstevel@tonic-gate } else { 38267c478bd9Sstevel@tonic-gate if (pp->p_lckcnt) { 38277c478bd9Sstevel@tonic-gate if (pp->p_lckcnt < (ushort_t)PAGE_LOCK_MAXIMUM) { 38287c478bd9Sstevel@tonic-gate r = 1; 38297c478bd9Sstevel@tonic-gate if (++pp->p_lckcnt == 38307c478bd9Sstevel@tonic-gate (ushort_t)PAGE_LOCK_MAXIMUM) { 38317c478bd9Sstevel@tonic-gate cmn_err(CE_WARN, "Page lock limit " 38327c478bd9Sstevel@tonic-gate "reached on pfn 0x%lx", 38337c478bd9Sstevel@tonic-gate page_pptonum(pp)); 38347c478bd9Sstevel@tonic-gate } 38357c478bd9Sstevel@tonic-gate } 38367c478bd9Sstevel@tonic-gate } else { 38377c478bd9Sstevel@tonic-gate if (kernel) { 38387c478bd9Sstevel@tonic-gate /* availrmem accounting done by caller */ 38397c478bd9Sstevel@tonic-gate ++pp->p_lckcnt; 38407c478bd9Sstevel@tonic-gate r = 1; 38417c478bd9Sstevel@tonic-gate } else { 38427c478bd9Sstevel@tonic-gate mutex_enter(&freemem_lock); 38437c478bd9Sstevel@tonic-gate if (availrmem > pages_pp_maximum) { 38447c478bd9Sstevel@tonic-gate availrmem--; 38457c478bd9Sstevel@tonic-gate pages_locked++; 38467c478bd9Sstevel@tonic-gate ++pp->p_lckcnt; 38477c478bd9Sstevel@tonic-gate r = 1; 38487c478bd9Sstevel@tonic-gate } 38497c478bd9Sstevel@tonic-gate mutex_exit(&freemem_lock); 38507c478bd9Sstevel@tonic-gate } 38517c478bd9Sstevel@tonic-gate } 38527c478bd9Sstevel@tonic-gate } 38537c478bd9Sstevel@tonic-gate page_struct_unlock(pp); 38547c478bd9Sstevel@tonic-gate return (r); 38557c478bd9Sstevel@tonic-gate } 38567c478bd9Sstevel@tonic-gate 38577c478bd9Sstevel@tonic-gate /* 38587c478bd9Sstevel@tonic-gate * Decommit a lock on a physical page frame. Account for cow locks if 38597c478bd9Sstevel@tonic-gate * appropriate. 38607c478bd9Sstevel@tonic-gate */ 38617c478bd9Sstevel@tonic-gate void 38627c478bd9Sstevel@tonic-gate page_pp_unlock( 38637c478bd9Sstevel@tonic-gate page_t *pp, /* page to be unlocked */ 38647c478bd9Sstevel@tonic-gate int cow, /* expect cow lock */ 38657c478bd9Sstevel@tonic-gate int kernel) /* this was a kernel lock */ 38667c478bd9Sstevel@tonic-gate { 38677c478bd9Sstevel@tonic-gate ASSERT(PAGE_LOCKED(pp)); 38687c478bd9Sstevel@tonic-gate 38697c478bd9Sstevel@tonic-gate page_struct_lock(pp); 38707c478bd9Sstevel@tonic-gate /* 38717c478bd9Sstevel@tonic-gate * Acquire the "freemem_lock" for availrmem. 38727c478bd9Sstevel@tonic-gate * If cowcnt or lcknt is already 0 do nothing; i.e., we 38737c478bd9Sstevel@tonic-gate * could be called to unlock even if nothing is locked. This could 38747c478bd9Sstevel@tonic-gate * happen if locked file pages were truncated (removing the lock) 38757c478bd9Sstevel@tonic-gate * and the file was grown again and new pages faulted in; the new 38767c478bd9Sstevel@tonic-gate * pages are unlocked but the segment still thinks they're locked. 38777c478bd9Sstevel@tonic-gate */ 38787c478bd9Sstevel@tonic-gate if (cow) { 38797c478bd9Sstevel@tonic-gate if (pp->p_cowcnt) { 38807c478bd9Sstevel@tonic-gate mutex_enter(&freemem_lock); 38817c478bd9Sstevel@tonic-gate pp->p_cowcnt--; 38827c478bd9Sstevel@tonic-gate availrmem++; 38837c478bd9Sstevel@tonic-gate pages_locked--; 38847c478bd9Sstevel@tonic-gate mutex_exit(&freemem_lock); 38857c478bd9Sstevel@tonic-gate } 38867c478bd9Sstevel@tonic-gate } else { 38877c478bd9Sstevel@tonic-gate if (pp->p_lckcnt && --pp->p_lckcnt == 0) { 38887c478bd9Sstevel@tonic-gate if (!kernel) { 38897c478bd9Sstevel@tonic-gate mutex_enter(&freemem_lock); 38907c478bd9Sstevel@tonic-gate availrmem++; 38917c478bd9Sstevel@tonic-gate pages_locked--; 38927c478bd9Sstevel@tonic-gate mutex_exit(&freemem_lock); 38937c478bd9Sstevel@tonic-gate } 38947c478bd9Sstevel@tonic-gate } 38957c478bd9Sstevel@tonic-gate } 38967c478bd9Sstevel@tonic-gate page_struct_unlock(pp); 38977c478bd9Sstevel@tonic-gate } 38987c478bd9Sstevel@tonic-gate 38997c478bd9Sstevel@tonic-gate /* 39007c478bd9Sstevel@tonic-gate * This routine reserves availrmem for npages; 39017c478bd9Sstevel@tonic-gate * flags: KM_NOSLEEP or KM_SLEEP 39027c478bd9Sstevel@tonic-gate * returns 1 on success or 0 on failure 39037c478bd9Sstevel@tonic-gate */ 39047c478bd9Sstevel@tonic-gate int 39057c478bd9Sstevel@tonic-gate page_resv(pgcnt_t npages, uint_t flags) 39067c478bd9Sstevel@tonic-gate { 39077c478bd9Sstevel@tonic-gate mutex_enter(&freemem_lock); 39087c478bd9Sstevel@tonic-gate while (availrmem < tune.t_minarmem + npages) { 39097c478bd9Sstevel@tonic-gate if (flags & KM_NOSLEEP) { 39107c478bd9Sstevel@tonic-gate mutex_exit(&freemem_lock); 39117c478bd9Sstevel@tonic-gate return (0); 39127c478bd9Sstevel@tonic-gate } 39137c478bd9Sstevel@tonic-gate mutex_exit(&freemem_lock); 39147c478bd9Sstevel@tonic-gate page_needfree(npages); 39157c478bd9Sstevel@tonic-gate kmem_reap(); 39167c478bd9Sstevel@tonic-gate delay(hz >> 2); 39177c478bd9Sstevel@tonic-gate page_needfree(-(spgcnt_t)npages); 39187c478bd9Sstevel@tonic-gate mutex_enter(&freemem_lock); 39197c478bd9Sstevel@tonic-gate } 39207c478bd9Sstevel@tonic-gate availrmem -= npages; 39217c478bd9Sstevel@tonic-gate mutex_exit(&freemem_lock); 39227c478bd9Sstevel@tonic-gate return (1); 39237c478bd9Sstevel@tonic-gate } 39247c478bd9Sstevel@tonic-gate 39257c478bd9Sstevel@tonic-gate /* 39267c478bd9Sstevel@tonic-gate * This routine unreserves availrmem for npages; 39277c478bd9Sstevel@tonic-gate */ 39287c478bd9Sstevel@tonic-gate void 39297c478bd9Sstevel@tonic-gate page_unresv(pgcnt_t npages) 39307c478bd9Sstevel@tonic-gate { 39317c478bd9Sstevel@tonic-gate mutex_enter(&freemem_lock); 39327c478bd9Sstevel@tonic-gate availrmem += npages; 39337c478bd9Sstevel@tonic-gate mutex_exit(&freemem_lock); 39347c478bd9Sstevel@tonic-gate } 39357c478bd9Sstevel@tonic-gate 39367c478bd9Sstevel@tonic-gate /* 39377c478bd9Sstevel@tonic-gate * See Statement at the beginning of segvn_lockop() regarding 39387c478bd9Sstevel@tonic-gate * the way we handle cowcnts and lckcnts. 39397c478bd9Sstevel@tonic-gate * 39407c478bd9Sstevel@tonic-gate * Transfer cowcnt on 'opp' to cowcnt on 'npp' if the vpage 39417c478bd9Sstevel@tonic-gate * that breaks COW has PROT_WRITE. 39427c478bd9Sstevel@tonic-gate * 39437c478bd9Sstevel@tonic-gate * Note that, we may also break COW in case we are softlocking 39447c478bd9Sstevel@tonic-gate * on read access during physio; 39457c478bd9Sstevel@tonic-gate * in this softlock case, the vpage may not have PROT_WRITE. 39467c478bd9Sstevel@tonic-gate * So, we need to transfer lckcnt on 'opp' to lckcnt on 'npp' 39477c478bd9Sstevel@tonic-gate * if the vpage doesn't have PROT_WRITE. 39487c478bd9Sstevel@tonic-gate * 39497c478bd9Sstevel@tonic-gate * This routine is never called if we are stealing a page 39507c478bd9Sstevel@tonic-gate * in anon_private. 39517c478bd9Sstevel@tonic-gate * 39527c478bd9Sstevel@tonic-gate * The caller subtracted from availrmem for read only mapping. 39537c478bd9Sstevel@tonic-gate * if lckcnt is 1 increment availrmem. 39547c478bd9Sstevel@tonic-gate */ 39557c478bd9Sstevel@tonic-gate void 39567c478bd9Sstevel@tonic-gate page_pp_useclaim( 39577c478bd9Sstevel@tonic-gate page_t *opp, /* original page frame losing lock */ 39587c478bd9Sstevel@tonic-gate page_t *npp, /* new page frame gaining lock */ 39597c478bd9Sstevel@tonic-gate uint_t write_perm) /* set if vpage has PROT_WRITE */ 39607c478bd9Sstevel@tonic-gate { 39617c478bd9Sstevel@tonic-gate int payback = 0; 39627c478bd9Sstevel@tonic-gate 39637c478bd9Sstevel@tonic-gate ASSERT(PAGE_LOCKED(opp)); 39647c478bd9Sstevel@tonic-gate ASSERT(PAGE_LOCKED(npp)); 39657c478bd9Sstevel@tonic-gate 39667c478bd9Sstevel@tonic-gate page_struct_lock(opp); 39677c478bd9Sstevel@tonic-gate 39687c478bd9Sstevel@tonic-gate ASSERT(npp->p_cowcnt == 0); 39697c478bd9Sstevel@tonic-gate ASSERT(npp->p_lckcnt == 0); 39707c478bd9Sstevel@tonic-gate 39717c478bd9Sstevel@tonic-gate /* Don't use claim if nothing is locked (see page_pp_unlock above) */ 39727c478bd9Sstevel@tonic-gate if ((write_perm && opp->p_cowcnt != 0) || 39737c478bd9Sstevel@tonic-gate (!write_perm && opp->p_lckcnt != 0)) { 39747c478bd9Sstevel@tonic-gate 39757c478bd9Sstevel@tonic-gate if (write_perm) { 39767c478bd9Sstevel@tonic-gate npp->p_cowcnt++; 39777c478bd9Sstevel@tonic-gate ASSERT(opp->p_cowcnt != 0); 39787c478bd9Sstevel@tonic-gate opp->p_cowcnt--; 39797c478bd9Sstevel@tonic-gate } else { 39807c478bd9Sstevel@tonic-gate 39817c478bd9Sstevel@tonic-gate ASSERT(opp->p_lckcnt != 0); 39827c478bd9Sstevel@tonic-gate 39837c478bd9Sstevel@tonic-gate /* 39847c478bd9Sstevel@tonic-gate * We didn't need availrmem decremented if p_lckcnt on 39857c478bd9Sstevel@tonic-gate * original page is 1. Here, we are unlocking 39867c478bd9Sstevel@tonic-gate * read-only copy belonging to original page and 39877c478bd9Sstevel@tonic-gate * are locking a copy belonging to new page. 39887c478bd9Sstevel@tonic-gate */ 39897c478bd9Sstevel@tonic-gate if (opp->p_lckcnt == 1) 39907c478bd9Sstevel@tonic-gate payback = 1; 39917c478bd9Sstevel@tonic-gate 39927c478bd9Sstevel@tonic-gate npp->p_lckcnt++; 39937c478bd9Sstevel@tonic-gate opp->p_lckcnt--; 39947c478bd9Sstevel@tonic-gate } 39957c478bd9Sstevel@tonic-gate } 39967c478bd9Sstevel@tonic-gate if (payback) { 39977c478bd9Sstevel@tonic-gate mutex_enter(&freemem_lock); 39987c478bd9Sstevel@tonic-gate availrmem++; 39997c478bd9Sstevel@tonic-gate pages_useclaim--; 40007c478bd9Sstevel@tonic-gate mutex_exit(&freemem_lock); 40017c478bd9Sstevel@tonic-gate } 40027c478bd9Sstevel@tonic-gate page_struct_unlock(opp); 40037c478bd9Sstevel@tonic-gate } 40047c478bd9Sstevel@tonic-gate 40057c478bd9Sstevel@tonic-gate /* 40067c478bd9Sstevel@tonic-gate * Simple claim adjust functions -- used to support changes in 40077c478bd9Sstevel@tonic-gate * claims due to changes in access permissions. Used by segvn_setprot(). 40087c478bd9Sstevel@tonic-gate */ 40097c478bd9Sstevel@tonic-gate int 40107c478bd9Sstevel@tonic-gate page_addclaim(page_t *pp) 40117c478bd9Sstevel@tonic-gate { 40127c478bd9Sstevel@tonic-gate int r = 0; /* result */ 40137c478bd9Sstevel@tonic-gate 40147c478bd9Sstevel@tonic-gate ASSERT(PAGE_LOCKED(pp)); 40157c478bd9Sstevel@tonic-gate 40167c478bd9Sstevel@tonic-gate page_struct_lock(pp); 40177c478bd9Sstevel@tonic-gate ASSERT(pp->p_lckcnt != 0); 40187c478bd9Sstevel@tonic-gate 40197c478bd9Sstevel@tonic-gate if (pp->p_lckcnt == 1) { 40207c478bd9Sstevel@tonic-gate if (pp->p_cowcnt < (ushort_t)PAGE_LOCK_MAXIMUM) { 40217c478bd9Sstevel@tonic-gate --pp->p_lckcnt; 40227c478bd9Sstevel@tonic-gate r = 1; 40237c478bd9Sstevel@tonic-gate if (++pp->p_cowcnt == (ushort_t)PAGE_LOCK_MAXIMUM) { 40247c478bd9Sstevel@tonic-gate cmn_err(CE_WARN, 40257c478bd9Sstevel@tonic-gate "COW lock limit reached on pfn 0x%lx", 40267c478bd9Sstevel@tonic-gate page_pptonum(pp)); 40277c478bd9Sstevel@tonic-gate } 40287c478bd9Sstevel@tonic-gate } 40297c478bd9Sstevel@tonic-gate } else { 40307c478bd9Sstevel@tonic-gate mutex_enter(&freemem_lock); 40317c478bd9Sstevel@tonic-gate if ((availrmem > pages_pp_maximum) && 40327c478bd9Sstevel@tonic-gate (pp->p_cowcnt < (ushort_t)PAGE_LOCK_MAXIMUM)) { 40337c478bd9Sstevel@tonic-gate --availrmem; 40347c478bd9Sstevel@tonic-gate ++pages_claimed; 40357c478bd9Sstevel@tonic-gate mutex_exit(&freemem_lock); 40367c478bd9Sstevel@tonic-gate --pp->p_lckcnt; 40377c478bd9Sstevel@tonic-gate r = 1; 40387c478bd9Sstevel@tonic-gate if (++pp->p_cowcnt == (ushort_t)PAGE_LOCK_MAXIMUM) { 40397c478bd9Sstevel@tonic-gate cmn_err(CE_WARN, 40407c478bd9Sstevel@tonic-gate "COW lock limit reached on pfn 0x%lx", 40417c478bd9Sstevel@tonic-gate page_pptonum(pp)); 40427c478bd9Sstevel@tonic-gate } 40437c478bd9Sstevel@tonic-gate } else 40447c478bd9Sstevel@tonic-gate mutex_exit(&freemem_lock); 40457c478bd9Sstevel@tonic-gate } 40467c478bd9Sstevel@tonic-gate page_struct_unlock(pp); 40477c478bd9Sstevel@tonic-gate return (r); 40487c478bd9Sstevel@tonic-gate } 40497c478bd9Sstevel@tonic-gate 40507c478bd9Sstevel@tonic-gate int 40517c478bd9Sstevel@tonic-gate page_subclaim(page_t *pp) 40527c478bd9Sstevel@tonic-gate { 40537c478bd9Sstevel@tonic-gate int r = 0; 40547c478bd9Sstevel@tonic-gate 40557c478bd9Sstevel@tonic-gate ASSERT(PAGE_LOCKED(pp)); 40567c478bd9Sstevel@tonic-gate 40577c478bd9Sstevel@tonic-gate page_struct_lock(pp); 40587c478bd9Sstevel@tonic-gate ASSERT(pp->p_cowcnt != 0); 40597c478bd9Sstevel@tonic-gate 40607c478bd9Sstevel@tonic-gate if (pp->p_lckcnt) { 40617c478bd9Sstevel@tonic-gate if (pp->p_lckcnt < (ushort_t)PAGE_LOCK_MAXIMUM) { 40627c478bd9Sstevel@tonic-gate r = 1; 40637c478bd9Sstevel@tonic-gate /* 40647c478bd9Sstevel@tonic-gate * for availrmem 40657c478bd9Sstevel@tonic-gate */ 40667c478bd9Sstevel@tonic-gate mutex_enter(&freemem_lock); 40677c478bd9Sstevel@tonic-gate availrmem++; 40687c478bd9Sstevel@tonic-gate pages_claimed--; 40697c478bd9Sstevel@tonic-gate mutex_exit(&freemem_lock); 40707c478bd9Sstevel@tonic-gate 40717c478bd9Sstevel@tonic-gate pp->p_cowcnt--; 40727c478bd9Sstevel@tonic-gate 40737c478bd9Sstevel@tonic-gate if (++pp->p_lckcnt == (ushort_t)PAGE_LOCK_MAXIMUM) { 40747c478bd9Sstevel@tonic-gate cmn_err(CE_WARN, 40757c478bd9Sstevel@tonic-gate "Page lock limit reached on pfn 0x%lx", 40767c478bd9Sstevel@tonic-gate page_pptonum(pp)); 40777c478bd9Sstevel@tonic-gate } 40787c478bd9Sstevel@tonic-gate } 40797c478bd9Sstevel@tonic-gate } else { 40807c478bd9Sstevel@tonic-gate r = 1; 40817c478bd9Sstevel@tonic-gate pp->p_cowcnt--; 40827c478bd9Sstevel@tonic-gate pp->p_lckcnt++; 40837c478bd9Sstevel@tonic-gate } 40847c478bd9Sstevel@tonic-gate page_struct_unlock(pp); 40857c478bd9Sstevel@tonic-gate return (r); 40867c478bd9Sstevel@tonic-gate } 40877c478bd9Sstevel@tonic-gate 40887c478bd9Sstevel@tonic-gate int 40897c478bd9Sstevel@tonic-gate page_addclaim_pages(page_t **ppa) 40907c478bd9Sstevel@tonic-gate { 40917c478bd9Sstevel@tonic-gate 40927c478bd9Sstevel@tonic-gate pgcnt_t lckpgs = 0, pg_idx; 40937c478bd9Sstevel@tonic-gate 40947c478bd9Sstevel@tonic-gate VM_STAT_ADD(pagecnt.pc_addclaim_pages); 40957c478bd9Sstevel@tonic-gate 40967c478bd9Sstevel@tonic-gate mutex_enter(&page_llock); 40977c478bd9Sstevel@tonic-gate for (pg_idx = 0; ppa[pg_idx] != NULL; pg_idx++) { 40987c478bd9Sstevel@tonic-gate 40997c478bd9Sstevel@tonic-gate ASSERT(PAGE_LOCKED(ppa[pg_idx])); 41007c478bd9Sstevel@tonic-gate ASSERT(ppa[pg_idx]->p_lckcnt != 0); 41017c478bd9Sstevel@tonic-gate if (ppa[pg_idx]->p_cowcnt == (ushort_t)PAGE_LOCK_MAXIMUM) { 41027c478bd9Sstevel@tonic-gate mutex_exit(&page_llock); 41037c478bd9Sstevel@tonic-gate return (0); 41047c478bd9Sstevel@tonic-gate } 41057c478bd9Sstevel@tonic-gate if (ppa[pg_idx]->p_lckcnt > 1) 41067c478bd9Sstevel@tonic-gate lckpgs++; 41077c478bd9Sstevel@tonic-gate } 41087c478bd9Sstevel@tonic-gate 41097c478bd9Sstevel@tonic-gate if (lckpgs != 0) { 41107c478bd9Sstevel@tonic-gate mutex_enter(&freemem_lock); 41117c478bd9Sstevel@tonic-gate if (availrmem >= pages_pp_maximum + lckpgs) { 41127c478bd9Sstevel@tonic-gate availrmem -= lckpgs; 41137c478bd9Sstevel@tonic-gate pages_claimed += lckpgs; 41147c478bd9Sstevel@tonic-gate } else { 41157c478bd9Sstevel@tonic-gate mutex_exit(&freemem_lock); 41167c478bd9Sstevel@tonic-gate mutex_exit(&page_llock); 41177c478bd9Sstevel@tonic-gate return (0); 41187c478bd9Sstevel@tonic-gate } 41197c478bd9Sstevel@tonic-gate mutex_exit(&freemem_lock); 41207c478bd9Sstevel@tonic-gate } 41217c478bd9Sstevel@tonic-gate 41227c478bd9Sstevel@tonic-gate for (pg_idx = 0; ppa[pg_idx] != NULL; pg_idx++) { 41237c478bd9Sstevel@tonic-gate ppa[pg_idx]->p_lckcnt--; 41247c478bd9Sstevel@tonic-gate ppa[pg_idx]->p_cowcnt++; 41257c478bd9Sstevel@tonic-gate } 41267c478bd9Sstevel@tonic-gate mutex_exit(&page_llock); 41277c478bd9Sstevel@tonic-gate return (1); 41287c478bd9Sstevel@tonic-gate } 41297c478bd9Sstevel@tonic-gate 41307c478bd9Sstevel@tonic-gate int 41317c478bd9Sstevel@tonic-gate page_subclaim_pages(page_t **ppa) 41327c478bd9Sstevel@tonic-gate { 41337c478bd9Sstevel@tonic-gate pgcnt_t ulckpgs = 0, pg_idx; 41347c478bd9Sstevel@tonic-gate 41357c478bd9Sstevel@tonic-gate VM_STAT_ADD(pagecnt.pc_subclaim_pages); 41367c478bd9Sstevel@tonic-gate 41377c478bd9Sstevel@tonic-gate mutex_enter(&page_llock); 41387c478bd9Sstevel@tonic-gate for (pg_idx = 0; ppa[pg_idx] != NULL; pg_idx++) { 41397c478bd9Sstevel@tonic-gate 41407c478bd9Sstevel@tonic-gate ASSERT(PAGE_LOCKED(ppa[pg_idx])); 41417c478bd9Sstevel@tonic-gate ASSERT(ppa[pg_idx]->p_cowcnt != 0); 41427c478bd9Sstevel@tonic-gate if (ppa[pg_idx]->p_lckcnt == (ushort_t)PAGE_LOCK_MAXIMUM) { 41437c478bd9Sstevel@tonic-gate mutex_exit(&page_llock); 41447c478bd9Sstevel@tonic-gate return (0); 41457c478bd9Sstevel@tonic-gate } 41467c478bd9Sstevel@tonic-gate if (ppa[pg_idx]->p_lckcnt != 0) 41477c478bd9Sstevel@tonic-gate ulckpgs++; 41487c478bd9Sstevel@tonic-gate } 41497c478bd9Sstevel@tonic-gate 41507c478bd9Sstevel@tonic-gate if (ulckpgs != 0) { 41517c478bd9Sstevel@tonic-gate mutex_enter(&freemem_lock); 41527c478bd9Sstevel@tonic-gate availrmem += ulckpgs; 41537c478bd9Sstevel@tonic-gate pages_claimed -= ulckpgs; 41547c478bd9Sstevel@tonic-gate mutex_exit(&freemem_lock); 41557c478bd9Sstevel@tonic-gate } 41567c478bd9Sstevel@tonic-gate 41577c478bd9Sstevel@tonic-gate for (pg_idx = 0; ppa[pg_idx] != NULL; pg_idx++) { 41587c478bd9Sstevel@tonic-gate ppa[pg_idx]->p_cowcnt--; 41597c478bd9Sstevel@tonic-gate ppa[pg_idx]->p_lckcnt++; 41607c478bd9Sstevel@tonic-gate 41617c478bd9Sstevel@tonic-gate } 41627c478bd9Sstevel@tonic-gate mutex_exit(&page_llock); 41637c478bd9Sstevel@tonic-gate return (1); 41647c478bd9Sstevel@tonic-gate } 41657c478bd9Sstevel@tonic-gate 41667c478bd9Sstevel@tonic-gate page_t * 41677c478bd9Sstevel@tonic-gate page_numtopp(pfn_t pfnum, se_t se) 41687c478bd9Sstevel@tonic-gate { 41697c478bd9Sstevel@tonic-gate page_t *pp; 41707c478bd9Sstevel@tonic-gate 41717c478bd9Sstevel@tonic-gate retry: 41727c478bd9Sstevel@tonic-gate pp = page_numtopp_nolock(pfnum); 41737c478bd9Sstevel@tonic-gate if (pp == NULL) { 41747c478bd9Sstevel@tonic-gate return ((page_t *)NULL); 41757c478bd9Sstevel@tonic-gate } 41767c478bd9Sstevel@tonic-gate 41777c478bd9Sstevel@tonic-gate /* 41787c478bd9Sstevel@tonic-gate * Acquire the appropriate lock on the page. 41797c478bd9Sstevel@tonic-gate */ 41807c478bd9Sstevel@tonic-gate while (!page_lock(pp, se, (kmutex_t *)NULL, P_RECLAIM)) { 41817c478bd9Sstevel@tonic-gate if (page_pptonum(pp) != pfnum) 41827c478bd9Sstevel@tonic-gate goto retry; 41837c478bd9Sstevel@tonic-gate continue; 41847c478bd9Sstevel@tonic-gate } 41857c478bd9Sstevel@tonic-gate 41867c478bd9Sstevel@tonic-gate if (page_pptonum(pp) != pfnum) { 41877c478bd9Sstevel@tonic-gate page_unlock(pp); 41887c478bd9Sstevel@tonic-gate goto retry; 41897c478bd9Sstevel@tonic-gate } 41907c478bd9Sstevel@tonic-gate 41917c478bd9Sstevel@tonic-gate return (pp); 41927c478bd9Sstevel@tonic-gate } 41937c478bd9Sstevel@tonic-gate 41947c478bd9Sstevel@tonic-gate page_t * 41957c478bd9Sstevel@tonic-gate page_numtopp_noreclaim(pfn_t pfnum, se_t se) 41967c478bd9Sstevel@tonic-gate { 41977c478bd9Sstevel@tonic-gate page_t *pp; 41987c478bd9Sstevel@tonic-gate 41997c478bd9Sstevel@tonic-gate retry: 42007c478bd9Sstevel@tonic-gate pp = page_numtopp_nolock(pfnum); 42017c478bd9Sstevel@tonic-gate if (pp == NULL) { 42027c478bd9Sstevel@tonic-gate return ((page_t *)NULL); 42037c478bd9Sstevel@tonic-gate } 42047c478bd9Sstevel@tonic-gate 42057c478bd9Sstevel@tonic-gate /* 42067c478bd9Sstevel@tonic-gate * Acquire the appropriate lock on the page. 42077c478bd9Sstevel@tonic-gate */ 42087c478bd9Sstevel@tonic-gate while (!page_lock(pp, se, (kmutex_t *)NULL, P_NO_RECLAIM)) { 42097c478bd9Sstevel@tonic-gate if (page_pptonum(pp) != pfnum) 42107c478bd9Sstevel@tonic-gate goto retry; 42117c478bd9Sstevel@tonic-gate continue; 42127c478bd9Sstevel@tonic-gate } 42137c478bd9Sstevel@tonic-gate 42147c478bd9Sstevel@tonic-gate if (page_pptonum(pp) != pfnum) { 42157c478bd9Sstevel@tonic-gate page_unlock(pp); 42167c478bd9Sstevel@tonic-gate goto retry; 42177c478bd9Sstevel@tonic-gate } 42187c478bd9Sstevel@tonic-gate 42197c478bd9Sstevel@tonic-gate return (pp); 42207c478bd9Sstevel@tonic-gate } 42217c478bd9Sstevel@tonic-gate 42227c478bd9Sstevel@tonic-gate /* 42237c478bd9Sstevel@tonic-gate * This routine is like page_numtopp, but will only return page structs 42247c478bd9Sstevel@tonic-gate * for pages which are ok for loading into hardware using the page struct. 42257c478bd9Sstevel@tonic-gate */ 42267c478bd9Sstevel@tonic-gate page_t * 42277c478bd9Sstevel@tonic-gate page_numtopp_nowait(pfn_t pfnum, se_t se) 42287c478bd9Sstevel@tonic-gate { 42297c478bd9Sstevel@tonic-gate page_t *pp; 42307c478bd9Sstevel@tonic-gate 42317c478bd9Sstevel@tonic-gate retry: 42327c478bd9Sstevel@tonic-gate pp = page_numtopp_nolock(pfnum); 42337c478bd9Sstevel@tonic-gate if (pp == NULL) { 42347c478bd9Sstevel@tonic-gate return ((page_t *)NULL); 42357c478bd9Sstevel@tonic-gate } 42367c478bd9Sstevel@tonic-gate 42377c478bd9Sstevel@tonic-gate /* 42387c478bd9Sstevel@tonic-gate * Try to acquire the appropriate lock on the page. 42397c478bd9Sstevel@tonic-gate */ 42407c478bd9Sstevel@tonic-gate if (PP_ISFREE(pp)) 42417c478bd9Sstevel@tonic-gate pp = NULL; 42427c478bd9Sstevel@tonic-gate else { 42437c478bd9Sstevel@tonic-gate if (!page_trylock(pp, se)) 42447c478bd9Sstevel@tonic-gate pp = NULL; 42457c478bd9Sstevel@tonic-gate else { 42467c478bd9Sstevel@tonic-gate if (page_pptonum(pp) != pfnum) { 42477c478bd9Sstevel@tonic-gate page_unlock(pp); 42487c478bd9Sstevel@tonic-gate goto retry; 42497c478bd9Sstevel@tonic-gate } 42507c478bd9Sstevel@tonic-gate if (PP_ISFREE(pp)) { 42517c478bd9Sstevel@tonic-gate page_unlock(pp); 42527c478bd9Sstevel@tonic-gate pp = NULL; 42537c478bd9Sstevel@tonic-gate } 42547c478bd9Sstevel@tonic-gate } 42557c478bd9Sstevel@tonic-gate } 42567c478bd9Sstevel@tonic-gate return (pp); 42577c478bd9Sstevel@tonic-gate } 42587c478bd9Sstevel@tonic-gate 42597c478bd9Sstevel@tonic-gate /* 42607c478bd9Sstevel@tonic-gate * Returns a count of dirty pages that are in the process 42617c478bd9Sstevel@tonic-gate * of being written out. If 'cleanit' is set, try to push the page. 42627c478bd9Sstevel@tonic-gate */ 42637c478bd9Sstevel@tonic-gate pgcnt_t 42647c478bd9Sstevel@tonic-gate page_busy(int cleanit) 42657c478bd9Sstevel@tonic-gate { 42667c478bd9Sstevel@tonic-gate page_t *page0 = page_first(); 42677c478bd9Sstevel@tonic-gate page_t *pp = page0; 42687c478bd9Sstevel@tonic-gate pgcnt_t nppbusy = 0; 42697c478bd9Sstevel@tonic-gate u_offset_t off; 42707c478bd9Sstevel@tonic-gate 42717c478bd9Sstevel@tonic-gate do { 42727c478bd9Sstevel@tonic-gate vnode_t *vp = pp->p_vnode; 42737c478bd9Sstevel@tonic-gate 42747c478bd9Sstevel@tonic-gate /* 42757c478bd9Sstevel@tonic-gate * A page is a candidate for syncing if it is: 42767c478bd9Sstevel@tonic-gate * 42777c478bd9Sstevel@tonic-gate * (a) On neither the freelist nor the cachelist 42787c478bd9Sstevel@tonic-gate * (b) Hashed onto a vnode 42797c478bd9Sstevel@tonic-gate * (c) Not a kernel page 42807c478bd9Sstevel@tonic-gate * (d) Dirty 42817c478bd9Sstevel@tonic-gate * (e) Not part of a swapfile 42827c478bd9Sstevel@tonic-gate * (f) a page which belongs to a real vnode; eg has a non-null 42837c478bd9Sstevel@tonic-gate * v_vfsp pointer. 42847c478bd9Sstevel@tonic-gate * (g) Backed by a filesystem which doesn't have a 42857c478bd9Sstevel@tonic-gate * stubbed-out sync operation 42867c478bd9Sstevel@tonic-gate */ 4287ad23a2dbSjohansen if (!PP_ISFREE(pp) && vp != NULL && !VN_ISKAS(vp) && 42887c478bd9Sstevel@tonic-gate hat_ismod(pp) && !IS_SWAPVP(vp) && vp->v_vfsp != NULL && 42897c478bd9Sstevel@tonic-gate vfs_can_sync(vp->v_vfsp)) { 42907c478bd9Sstevel@tonic-gate nppbusy++; 42917c478bd9Sstevel@tonic-gate vfs_syncprogress(); 42927c478bd9Sstevel@tonic-gate 42937c478bd9Sstevel@tonic-gate if (!cleanit) 42947c478bd9Sstevel@tonic-gate continue; 42957c478bd9Sstevel@tonic-gate if (!page_trylock(pp, SE_EXCL)) 42967c478bd9Sstevel@tonic-gate continue; 42977c478bd9Sstevel@tonic-gate 42987c478bd9Sstevel@tonic-gate if (PP_ISFREE(pp) || vp == NULL || IS_SWAPVP(vp) || 42997c478bd9Sstevel@tonic-gate pp->p_lckcnt != 0 || pp->p_cowcnt != 0 || 43007c478bd9Sstevel@tonic-gate !(hat_pagesync(pp, 43017c478bd9Sstevel@tonic-gate HAT_SYNC_DONTZERO | HAT_SYNC_STOPON_MOD) & P_MOD)) { 43027c478bd9Sstevel@tonic-gate page_unlock(pp); 43037c478bd9Sstevel@tonic-gate continue; 43047c478bd9Sstevel@tonic-gate } 43057c478bd9Sstevel@tonic-gate off = pp->p_offset; 43067c478bd9Sstevel@tonic-gate VN_HOLD(vp); 43077c478bd9Sstevel@tonic-gate page_unlock(pp); 43087c478bd9Sstevel@tonic-gate (void) VOP_PUTPAGE(vp, off, PAGESIZE, 4309da6c28aaSamw B_ASYNC | B_FREE, kcred, NULL); 43107c478bd9Sstevel@tonic-gate VN_RELE(vp); 43117c478bd9Sstevel@tonic-gate } 43127c478bd9Sstevel@tonic-gate } while ((pp = page_next(pp)) != page0); 43137c478bd9Sstevel@tonic-gate 43147c478bd9Sstevel@tonic-gate return (nppbusy); 43157c478bd9Sstevel@tonic-gate } 43167c478bd9Sstevel@tonic-gate 43177c478bd9Sstevel@tonic-gate void page_invalidate_pages(void); 43187c478bd9Sstevel@tonic-gate 43197c478bd9Sstevel@tonic-gate /* 43207c478bd9Sstevel@tonic-gate * callback handler to vm sub-system 43217c478bd9Sstevel@tonic-gate * 43227c478bd9Sstevel@tonic-gate * callers make sure no recursive entries to this func. 43237c478bd9Sstevel@tonic-gate */ 43247c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 43257c478bd9Sstevel@tonic-gate boolean_t 43267c478bd9Sstevel@tonic-gate callb_vm_cpr(void *arg, int code) 43277c478bd9Sstevel@tonic-gate { 43287c478bd9Sstevel@tonic-gate if (code == CB_CODE_CPR_CHKPT) 43297c478bd9Sstevel@tonic-gate page_invalidate_pages(); 43307c478bd9Sstevel@tonic-gate return (B_TRUE); 43317c478bd9Sstevel@tonic-gate } 43327c478bd9Sstevel@tonic-gate 43337c478bd9Sstevel@tonic-gate /* 43347c478bd9Sstevel@tonic-gate * Invalidate all pages of the system. 43357c478bd9Sstevel@tonic-gate * It shouldn't be called until all user page activities are all stopped. 43367c478bd9Sstevel@tonic-gate */ 43377c478bd9Sstevel@tonic-gate void 43387c478bd9Sstevel@tonic-gate page_invalidate_pages() 43397c478bd9Sstevel@tonic-gate { 43407c478bd9Sstevel@tonic-gate page_t *pp; 43417c478bd9Sstevel@tonic-gate page_t *page0; 43427c478bd9Sstevel@tonic-gate pgcnt_t nbusypages; 43437c478bd9Sstevel@tonic-gate int retry = 0; 43447c478bd9Sstevel@tonic-gate const int MAXRETRIES = 4; 43457c478bd9Sstevel@tonic-gate #if defined(__sparc) 43467c478bd9Sstevel@tonic-gate extern struct vnode prom_ppages; 43477c478bd9Sstevel@tonic-gate #endif /* __sparc */ 43487c478bd9Sstevel@tonic-gate 43497c478bd9Sstevel@tonic-gate top: 43507c478bd9Sstevel@tonic-gate /* 43518b464eb8Smec * Flush dirty pages and destroy the clean ones. 43527c478bd9Sstevel@tonic-gate */ 43537c478bd9Sstevel@tonic-gate nbusypages = 0; 43547c478bd9Sstevel@tonic-gate 43557c478bd9Sstevel@tonic-gate pp = page0 = page_first(); 43567c478bd9Sstevel@tonic-gate do { 43577c478bd9Sstevel@tonic-gate struct vnode *vp; 43587c478bd9Sstevel@tonic-gate u_offset_t offset; 43597c478bd9Sstevel@tonic-gate int mod; 43607c478bd9Sstevel@tonic-gate 43617c478bd9Sstevel@tonic-gate /* 43627c478bd9Sstevel@tonic-gate * skip the page if it has no vnode or the page associated 43637c478bd9Sstevel@tonic-gate * with the kernel vnode or prom allocated kernel mem. 43647c478bd9Sstevel@tonic-gate */ 43657c478bd9Sstevel@tonic-gate #if defined(__sparc) 4366ad23a2dbSjohansen if ((vp = pp->p_vnode) == NULL || VN_ISKAS(vp) || 43677c478bd9Sstevel@tonic-gate vp == &prom_ppages) 43687c478bd9Sstevel@tonic-gate #else /* x86 doesn't have prom or prom_ppage */ 4369ad23a2dbSjohansen if ((vp = pp->p_vnode) == NULL || VN_ISKAS(vp)) 43707c478bd9Sstevel@tonic-gate #endif /* __sparc */ 43717c478bd9Sstevel@tonic-gate continue; 43727c478bd9Sstevel@tonic-gate 43737c478bd9Sstevel@tonic-gate /* 43747c478bd9Sstevel@tonic-gate * skip the page which is already free invalidated. 43757c478bd9Sstevel@tonic-gate */ 43767c478bd9Sstevel@tonic-gate if (PP_ISFREE(pp) && PP_ISAGED(pp)) 43777c478bd9Sstevel@tonic-gate continue; 43787c478bd9Sstevel@tonic-gate 43797c478bd9Sstevel@tonic-gate /* 43807c478bd9Sstevel@tonic-gate * skip pages that are already locked or can't be "exclusively" 43817c478bd9Sstevel@tonic-gate * locked or are already free. After we lock the page, check 43827c478bd9Sstevel@tonic-gate * the free and age bits again to be sure it's not destroied 43837c478bd9Sstevel@tonic-gate * yet. 43847c478bd9Sstevel@tonic-gate * To achieve max. parallelization, we use page_trylock instead 43857c478bd9Sstevel@tonic-gate * of page_lock so that we don't get block on individual pages 43867c478bd9Sstevel@tonic-gate * while we have thousands of other pages to process. 43877c478bd9Sstevel@tonic-gate */ 43887c478bd9Sstevel@tonic-gate if (!page_trylock(pp, SE_EXCL)) { 43897c478bd9Sstevel@tonic-gate nbusypages++; 43907c478bd9Sstevel@tonic-gate continue; 43917c478bd9Sstevel@tonic-gate } else if (PP_ISFREE(pp)) { 43927c478bd9Sstevel@tonic-gate if (!PP_ISAGED(pp)) { 43937c478bd9Sstevel@tonic-gate page_destroy_free(pp); 43947c478bd9Sstevel@tonic-gate } else { 43957c478bd9Sstevel@tonic-gate page_unlock(pp); 43967c478bd9Sstevel@tonic-gate } 43977c478bd9Sstevel@tonic-gate continue; 43987c478bd9Sstevel@tonic-gate } 43997c478bd9Sstevel@tonic-gate /* 44007c478bd9Sstevel@tonic-gate * Is this page involved in some I/O? shared? 44017c478bd9Sstevel@tonic-gate * 44027c478bd9Sstevel@tonic-gate * The page_struct_lock need not be acquired to 44037c478bd9Sstevel@tonic-gate * examine these fields since the page has an 44047c478bd9Sstevel@tonic-gate * "exclusive" lock. 44057c478bd9Sstevel@tonic-gate */ 44067c478bd9Sstevel@tonic-gate if (pp->p_lckcnt != 0 || pp->p_cowcnt != 0) { 44077c478bd9Sstevel@tonic-gate page_unlock(pp); 44087c478bd9Sstevel@tonic-gate continue; 44097c478bd9Sstevel@tonic-gate } 44107c478bd9Sstevel@tonic-gate 44117c478bd9Sstevel@tonic-gate if (vp->v_type == VCHR) { 44127c478bd9Sstevel@tonic-gate panic("vp->v_type == VCHR"); 44137c478bd9Sstevel@tonic-gate /*NOTREACHED*/ 44147c478bd9Sstevel@tonic-gate } 44157c478bd9Sstevel@tonic-gate 44167c478bd9Sstevel@tonic-gate if (!page_try_demote_pages(pp)) { 44177c478bd9Sstevel@tonic-gate page_unlock(pp); 44187c478bd9Sstevel@tonic-gate continue; 44197c478bd9Sstevel@tonic-gate } 44207c478bd9Sstevel@tonic-gate 44217c478bd9Sstevel@tonic-gate /* 44227c478bd9Sstevel@tonic-gate * Check the modified bit. Leave the bits alone in hardware 44237c478bd9Sstevel@tonic-gate * (they will be modified if we do the putpage). 44247c478bd9Sstevel@tonic-gate */ 44257c478bd9Sstevel@tonic-gate mod = (hat_pagesync(pp, HAT_SYNC_DONTZERO | HAT_SYNC_STOPON_MOD) 44266e4dd838Smec & P_MOD); 44277c478bd9Sstevel@tonic-gate if (mod) { 44287c478bd9Sstevel@tonic-gate offset = pp->p_offset; 44297c478bd9Sstevel@tonic-gate /* 44307c478bd9Sstevel@tonic-gate * Hold the vnode before releasing the page lock 44317c478bd9Sstevel@tonic-gate * to prevent it from being freed and re-used by 44327c478bd9Sstevel@tonic-gate * some other thread. 44337c478bd9Sstevel@tonic-gate */ 44347c478bd9Sstevel@tonic-gate VN_HOLD(vp); 44357c478bd9Sstevel@tonic-gate page_unlock(pp); 44367c478bd9Sstevel@tonic-gate /* 44377c478bd9Sstevel@tonic-gate * No error return is checked here. Callers such as 44387c478bd9Sstevel@tonic-gate * cpr deals with the dirty pages at the dump time 44397c478bd9Sstevel@tonic-gate * if this putpage fails. 44407c478bd9Sstevel@tonic-gate */ 44417c478bd9Sstevel@tonic-gate (void) VOP_PUTPAGE(vp, offset, PAGESIZE, B_INVAL, 4442da6c28aaSamw kcred, NULL); 44437c478bd9Sstevel@tonic-gate VN_RELE(vp); 44447c478bd9Sstevel@tonic-gate } else { 44457c478bd9Sstevel@tonic-gate page_destroy(pp, 0); 44467c478bd9Sstevel@tonic-gate } 44477c478bd9Sstevel@tonic-gate } while ((pp = page_next(pp)) != page0); 44487c478bd9Sstevel@tonic-gate if (nbusypages && retry++ < MAXRETRIES) { 44497c478bd9Sstevel@tonic-gate delay(1); 44507c478bd9Sstevel@tonic-gate goto top; 44517c478bd9Sstevel@tonic-gate } 44527c478bd9Sstevel@tonic-gate } 44537c478bd9Sstevel@tonic-gate 44547c478bd9Sstevel@tonic-gate /* 44557c478bd9Sstevel@tonic-gate * Replace the page "old" with the page "new" on the page hash and vnode lists 44567c478bd9Sstevel@tonic-gate * 4457da6c28aaSamw * the replacement must be done in place, ie the equivalent sequence: 44587c478bd9Sstevel@tonic-gate * 44597c478bd9Sstevel@tonic-gate * vp = old->p_vnode; 44607c478bd9Sstevel@tonic-gate * off = old->p_offset; 44617c478bd9Sstevel@tonic-gate * page_do_hashout(old) 44627c478bd9Sstevel@tonic-gate * page_do_hashin(new, vp, off) 44637c478bd9Sstevel@tonic-gate * 44647c478bd9Sstevel@tonic-gate * doesn't work, since 44657c478bd9Sstevel@tonic-gate * 1) if old is the only page on the vnode, the v_pages list has a window 44667c478bd9Sstevel@tonic-gate * where it looks empty. This will break file system assumptions. 44677c478bd9Sstevel@tonic-gate * and 44687c478bd9Sstevel@tonic-gate * 2) pvn_vplist_dirty() can't deal with pages moving on the v_pages list. 44697c478bd9Sstevel@tonic-gate */ 44707c478bd9Sstevel@tonic-gate static void 44717c478bd9Sstevel@tonic-gate page_do_relocate_hash(page_t *new, page_t *old) 44727c478bd9Sstevel@tonic-gate { 44737c478bd9Sstevel@tonic-gate page_t **hash_list; 44747c478bd9Sstevel@tonic-gate vnode_t *vp = old->p_vnode; 44757c478bd9Sstevel@tonic-gate kmutex_t *sep; 44767c478bd9Sstevel@tonic-gate 44777c478bd9Sstevel@tonic-gate ASSERT(PAGE_EXCL(old)); 44787c478bd9Sstevel@tonic-gate ASSERT(PAGE_EXCL(new)); 44797c478bd9Sstevel@tonic-gate ASSERT(vp != NULL); 44807c478bd9Sstevel@tonic-gate ASSERT(MUTEX_HELD(page_vnode_mutex(vp))); 44817c478bd9Sstevel@tonic-gate ASSERT(MUTEX_HELD(PAGE_HASH_MUTEX(PAGE_HASH_FUNC(vp, old->p_offset)))); 44827c478bd9Sstevel@tonic-gate 44837c478bd9Sstevel@tonic-gate /* 44847c478bd9Sstevel@tonic-gate * First find old page on the page hash list 44857c478bd9Sstevel@tonic-gate */ 44867c478bd9Sstevel@tonic-gate hash_list = &page_hash[PAGE_HASH_FUNC(vp, old->p_offset)]; 44877c478bd9Sstevel@tonic-gate 44887c478bd9Sstevel@tonic-gate for (;;) { 44897c478bd9Sstevel@tonic-gate if (*hash_list == old) 44907c478bd9Sstevel@tonic-gate break; 44917c478bd9Sstevel@tonic-gate if (*hash_list == NULL) { 44927c478bd9Sstevel@tonic-gate panic("page_do_hashout"); 44937c478bd9Sstevel@tonic-gate /*NOTREACHED*/ 44947c478bd9Sstevel@tonic-gate } 44957c478bd9Sstevel@tonic-gate hash_list = &(*hash_list)->p_hash; 44967c478bd9Sstevel@tonic-gate } 44977c478bd9Sstevel@tonic-gate 44987c478bd9Sstevel@tonic-gate /* 44997c478bd9Sstevel@tonic-gate * update new and replace old with new on the page hash list 45007c478bd9Sstevel@tonic-gate */ 45017c478bd9Sstevel@tonic-gate new->p_vnode = old->p_vnode; 45027c478bd9Sstevel@tonic-gate new->p_offset = old->p_offset; 45037c478bd9Sstevel@tonic-gate new->p_hash = old->p_hash; 45047c478bd9Sstevel@tonic-gate *hash_list = new; 45057c478bd9Sstevel@tonic-gate 45067c478bd9Sstevel@tonic-gate if ((new->p_vnode->v_flag & VISSWAP) != 0) 45077c478bd9Sstevel@tonic-gate PP_SETSWAP(new); 45087c478bd9Sstevel@tonic-gate 45097c478bd9Sstevel@tonic-gate /* 45107c478bd9Sstevel@tonic-gate * replace old with new on the vnode's page list 45117c478bd9Sstevel@tonic-gate */ 45127c478bd9Sstevel@tonic-gate if (old->p_vpnext == old) { 45137c478bd9Sstevel@tonic-gate new->p_vpnext = new; 45147c478bd9Sstevel@tonic-gate new->p_vpprev = new; 45157c478bd9Sstevel@tonic-gate } else { 45167c478bd9Sstevel@tonic-gate new->p_vpnext = old->p_vpnext; 45177c478bd9Sstevel@tonic-gate new->p_vpprev = old->p_vpprev; 45187c478bd9Sstevel@tonic-gate new->p_vpnext->p_vpprev = new; 45197c478bd9Sstevel@tonic-gate new->p_vpprev->p_vpnext = new; 45207c478bd9Sstevel@tonic-gate } 45217c478bd9Sstevel@tonic-gate if (vp->v_pages == old) 45227c478bd9Sstevel@tonic-gate vp->v_pages = new; 45237c478bd9Sstevel@tonic-gate 45247c478bd9Sstevel@tonic-gate /* 45257c478bd9Sstevel@tonic-gate * clear out the old page 45267c478bd9Sstevel@tonic-gate */ 45277c478bd9Sstevel@tonic-gate old->p_hash = NULL; 45287c478bd9Sstevel@tonic-gate old->p_vpnext = NULL; 45297c478bd9Sstevel@tonic-gate old->p_vpprev = NULL; 45307c478bd9Sstevel@tonic-gate old->p_vnode = NULL; 45317c478bd9Sstevel@tonic-gate PP_CLRSWAP(old); 45327c478bd9Sstevel@tonic-gate old->p_offset = (u_offset_t)-1; 45337c478bd9Sstevel@tonic-gate page_clr_all_props(old); 45347c478bd9Sstevel@tonic-gate 45357c478bd9Sstevel@tonic-gate /* 45367c478bd9Sstevel@tonic-gate * Wake up processes waiting for this page. The page's 45377c478bd9Sstevel@tonic-gate * identity has been changed, and is probably not the 45387c478bd9Sstevel@tonic-gate * desired page any longer. 45397c478bd9Sstevel@tonic-gate */ 45407c478bd9Sstevel@tonic-gate sep = page_se_mutex(old); 45417c478bd9Sstevel@tonic-gate mutex_enter(sep); 454242787a71Sstans old->p_selock &= ~SE_EWANTED; 45437c478bd9Sstevel@tonic-gate if (CV_HAS_WAITERS(&old->p_cv)) 45447c478bd9Sstevel@tonic-gate cv_broadcast(&old->p_cv); 45457c478bd9Sstevel@tonic-gate mutex_exit(sep); 45467c478bd9Sstevel@tonic-gate } 45477c478bd9Sstevel@tonic-gate 45487c478bd9Sstevel@tonic-gate /* 45497c478bd9Sstevel@tonic-gate * This function moves the identity of page "pp_old" to page "pp_new". 45507c478bd9Sstevel@tonic-gate * Both pages must be locked on entry. "pp_new" is free, has no identity, 45517c478bd9Sstevel@tonic-gate * and need not be hashed out from anywhere. 45527c478bd9Sstevel@tonic-gate */ 45537c478bd9Sstevel@tonic-gate void 45547c478bd9Sstevel@tonic-gate page_relocate_hash(page_t *pp_new, page_t *pp_old) 45557c478bd9Sstevel@tonic-gate { 45567c478bd9Sstevel@tonic-gate vnode_t *vp = pp_old->p_vnode; 45577c478bd9Sstevel@tonic-gate u_offset_t off = pp_old->p_offset; 45587c478bd9Sstevel@tonic-gate kmutex_t *phm, *vphm; 45597c478bd9Sstevel@tonic-gate 45607c478bd9Sstevel@tonic-gate /* 45617c478bd9Sstevel@tonic-gate * Rehash two pages 45627c478bd9Sstevel@tonic-gate */ 45637c478bd9Sstevel@tonic-gate ASSERT(PAGE_EXCL(pp_old)); 45647c478bd9Sstevel@tonic-gate ASSERT(PAGE_EXCL(pp_new)); 45657c478bd9Sstevel@tonic-gate ASSERT(vp != NULL); 45667c478bd9Sstevel@tonic-gate ASSERT(pp_new->p_vnode == NULL); 45677c478bd9Sstevel@tonic-gate 45687c478bd9Sstevel@tonic-gate /* 45697c478bd9Sstevel@tonic-gate * hashout then hashin while holding the mutexes 45707c478bd9Sstevel@tonic-gate */ 45717c478bd9Sstevel@tonic-gate phm = PAGE_HASH_MUTEX(PAGE_HASH_FUNC(vp, off)); 45727c478bd9Sstevel@tonic-gate mutex_enter(phm); 45737c478bd9Sstevel@tonic-gate vphm = page_vnode_mutex(vp); 45747c478bd9Sstevel@tonic-gate mutex_enter(vphm); 45757c478bd9Sstevel@tonic-gate 45767c478bd9Sstevel@tonic-gate page_do_relocate_hash(pp_new, pp_old); 45777c478bd9Sstevel@tonic-gate 45787c478bd9Sstevel@tonic-gate mutex_exit(vphm); 45797c478bd9Sstevel@tonic-gate mutex_exit(phm); 45807c478bd9Sstevel@tonic-gate 45817c478bd9Sstevel@tonic-gate /* 45827c478bd9Sstevel@tonic-gate * The page_struct_lock need not be acquired for lckcnt and 45837c478bd9Sstevel@tonic-gate * cowcnt since the page has an "exclusive" lock. 45847c478bd9Sstevel@tonic-gate */ 45857c478bd9Sstevel@tonic-gate ASSERT(pp_new->p_lckcnt == 0); 45867c478bd9Sstevel@tonic-gate ASSERT(pp_new->p_cowcnt == 0); 45877c478bd9Sstevel@tonic-gate pp_new->p_lckcnt = pp_old->p_lckcnt; 45887c478bd9Sstevel@tonic-gate pp_new->p_cowcnt = pp_old->p_cowcnt; 45897c478bd9Sstevel@tonic-gate pp_old->p_lckcnt = pp_old->p_cowcnt = 0; 45907c478bd9Sstevel@tonic-gate 45917c478bd9Sstevel@tonic-gate /* The following comment preserved from page_flip(). */ 45927c478bd9Sstevel@tonic-gate /* XXX - Do we need to protect fsdata? */ 45937c478bd9Sstevel@tonic-gate pp_new->p_fsdata = pp_old->p_fsdata; 45947c478bd9Sstevel@tonic-gate } 45957c478bd9Sstevel@tonic-gate 45967c478bd9Sstevel@tonic-gate /* 45977c478bd9Sstevel@tonic-gate * Helper routine used to lock all remaining members of a 45987c478bd9Sstevel@tonic-gate * large page. The caller is responsible for passing in a locked 45997c478bd9Sstevel@tonic-gate * pp. If pp is a large page, then it succeeds in locking all the 46007c478bd9Sstevel@tonic-gate * remaining constituent pages or it returns with only the 46017c478bd9Sstevel@tonic-gate * original page locked. 46027c478bd9Sstevel@tonic-gate * 46037c478bd9Sstevel@tonic-gate * Returns 1 on success, 0 on failure. 46047c478bd9Sstevel@tonic-gate * 4605da6c28aaSamw * If success is returned this routine guarantees p_szc for all constituent 46067c478bd9Sstevel@tonic-gate * pages of a large page pp belongs to can't change. To achieve this we 46077c478bd9Sstevel@tonic-gate * recheck szc of pp after locking all constituent pages and retry if szc 46087c478bd9Sstevel@tonic-gate * changed (it could only decrease). Since hat_page_demote() needs an EXCL 46097c478bd9Sstevel@tonic-gate * lock on one of constituent pages it can't be running after all constituent 46107c478bd9Sstevel@tonic-gate * pages are locked. hat_page_demote() with a lock on a constituent page 46117c478bd9Sstevel@tonic-gate * outside of this large page (i.e. pp belonged to a larger large page) is 46127c478bd9Sstevel@tonic-gate * already done with all constituent pages of pp since the root's p_szc is 4613da6c28aaSamw * changed last. Therefore no need to synchronize with hat_page_demote() that 46147c478bd9Sstevel@tonic-gate * locked a constituent page outside of pp's current large page. 46157c478bd9Sstevel@tonic-gate */ 46167c478bd9Sstevel@tonic-gate #ifdef DEBUG 46177c478bd9Sstevel@tonic-gate uint32_t gpg_trylock_mtbf = 0; 46187c478bd9Sstevel@tonic-gate #endif 46197c478bd9Sstevel@tonic-gate 46207c478bd9Sstevel@tonic-gate int 46217c478bd9Sstevel@tonic-gate group_page_trylock(page_t *pp, se_t se) 46227c478bd9Sstevel@tonic-gate { 46237c478bd9Sstevel@tonic-gate page_t *tpp; 46247c478bd9Sstevel@tonic-gate pgcnt_t npgs, i, j; 46257c478bd9Sstevel@tonic-gate uint_t pszc = pp->p_szc; 46267c478bd9Sstevel@tonic-gate 46277c478bd9Sstevel@tonic-gate #ifdef DEBUG 46287c478bd9Sstevel@tonic-gate if (gpg_trylock_mtbf && !(gethrtime() % gpg_trylock_mtbf)) { 46297c478bd9Sstevel@tonic-gate return (0); 46307c478bd9Sstevel@tonic-gate } 46317c478bd9Sstevel@tonic-gate #endif 46327c478bd9Sstevel@tonic-gate 46337c478bd9Sstevel@tonic-gate if (pp != PP_GROUPLEADER(pp, pszc)) { 46347c478bd9Sstevel@tonic-gate return (0); 46357c478bd9Sstevel@tonic-gate } 46367c478bd9Sstevel@tonic-gate 46377c478bd9Sstevel@tonic-gate retry: 46387c478bd9Sstevel@tonic-gate ASSERT(PAGE_LOCKED_SE(pp, se)); 46397c478bd9Sstevel@tonic-gate ASSERT(!PP_ISFREE(pp)); 46407c478bd9Sstevel@tonic-gate if (pszc == 0) { 46417c478bd9Sstevel@tonic-gate return (1); 46427c478bd9Sstevel@tonic-gate } 46437c478bd9Sstevel@tonic-gate npgs = page_get_pagecnt(pszc); 46447c478bd9Sstevel@tonic-gate tpp = pp + 1; 46457c478bd9Sstevel@tonic-gate for (i = 1; i < npgs; i++, tpp++) { 46467c478bd9Sstevel@tonic-gate if (!page_trylock(tpp, se)) { 46477c478bd9Sstevel@tonic-gate tpp = pp + 1; 46487c478bd9Sstevel@tonic-gate for (j = 1; j < i; j++, tpp++) { 46497c478bd9Sstevel@tonic-gate page_unlock(tpp); 46507c478bd9Sstevel@tonic-gate } 46517c478bd9Sstevel@tonic-gate return (0); 46527c478bd9Sstevel@tonic-gate } 46537c478bd9Sstevel@tonic-gate } 46547c478bd9Sstevel@tonic-gate if (pp->p_szc != pszc) { 46557c478bd9Sstevel@tonic-gate ASSERT(pp->p_szc < pszc); 4656ad23a2dbSjohansen ASSERT(pp->p_vnode != NULL && !PP_ISKAS(pp) && 46577c478bd9Sstevel@tonic-gate !IS_SWAPFSVP(pp->p_vnode)); 46587c478bd9Sstevel@tonic-gate tpp = pp + 1; 46597c478bd9Sstevel@tonic-gate for (i = 1; i < npgs; i++, tpp++) { 46607c478bd9Sstevel@tonic-gate page_unlock(tpp); 46617c478bd9Sstevel@tonic-gate } 46627c478bd9Sstevel@tonic-gate pszc = pp->p_szc; 46637c478bd9Sstevel@tonic-gate goto retry; 46647c478bd9Sstevel@tonic-gate } 46657c478bd9Sstevel@tonic-gate return (1); 46667c478bd9Sstevel@tonic-gate } 46677c478bd9Sstevel@tonic-gate 46687c478bd9Sstevel@tonic-gate void 46697c478bd9Sstevel@tonic-gate group_page_unlock(page_t *pp) 46707c478bd9Sstevel@tonic-gate { 46717c478bd9Sstevel@tonic-gate page_t *tpp; 46727c478bd9Sstevel@tonic-gate pgcnt_t npgs, i; 46737c478bd9Sstevel@tonic-gate 46747c478bd9Sstevel@tonic-gate ASSERT(PAGE_LOCKED(pp)); 46757c478bd9Sstevel@tonic-gate ASSERT(!PP_ISFREE(pp)); 46767c478bd9Sstevel@tonic-gate ASSERT(pp == PP_PAGEROOT(pp)); 46777c478bd9Sstevel@tonic-gate npgs = page_get_pagecnt(pp->p_szc); 46787c478bd9Sstevel@tonic-gate for (i = 1, tpp = pp + 1; i < npgs; i++, tpp++) { 46797c478bd9Sstevel@tonic-gate page_unlock(tpp); 46807c478bd9Sstevel@tonic-gate } 46817c478bd9Sstevel@tonic-gate } 46827c478bd9Sstevel@tonic-gate 46837c478bd9Sstevel@tonic-gate /* 46847c478bd9Sstevel@tonic-gate * returns 46857c478bd9Sstevel@tonic-gate * 0 : on success and *nrelocp is number of relocated PAGESIZE pages 46867c478bd9Sstevel@tonic-gate * ERANGE : this is not a base page 46877c478bd9Sstevel@tonic-gate * EBUSY : failure to get locks on the page/pages 46887c478bd9Sstevel@tonic-gate * ENOMEM : failure to obtain replacement pages 46897c478bd9Sstevel@tonic-gate * EAGAIN : OBP has not yet completed its boot-time handoff to the kernel 46908b464eb8Smec * EIO : An error occurred while trying to copy the page data 46917c478bd9Sstevel@tonic-gate * 46927c478bd9Sstevel@tonic-gate * Return with all constituent members of target and replacement 46937c478bd9Sstevel@tonic-gate * SE_EXCL locked. It is the callers responsibility to drop the 46947c478bd9Sstevel@tonic-gate * locks. 46957c478bd9Sstevel@tonic-gate */ 46967c478bd9Sstevel@tonic-gate int 46977c478bd9Sstevel@tonic-gate do_page_relocate( 46987c478bd9Sstevel@tonic-gate page_t **target, 46997c478bd9Sstevel@tonic-gate page_t **replacement, 47007c478bd9Sstevel@tonic-gate int grouplock, 47017c478bd9Sstevel@tonic-gate spgcnt_t *nrelocp, 47027c478bd9Sstevel@tonic-gate lgrp_t *lgrp) 47037c478bd9Sstevel@tonic-gate { 47047c478bd9Sstevel@tonic-gate page_t *first_repl; 47057c478bd9Sstevel@tonic-gate page_t *repl; 47067c478bd9Sstevel@tonic-gate page_t *targ; 47077c478bd9Sstevel@tonic-gate page_t *pl = NULL; 47087c478bd9Sstevel@tonic-gate uint_t ppattr; 47097c478bd9Sstevel@tonic-gate pfn_t pfn, repl_pfn; 47107c478bd9Sstevel@tonic-gate uint_t szc; 47117c478bd9Sstevel@tonic-gate spgcnt_t npgs, i; 47127c478bd9Sstevel@tonic-gate int repl_contig = 0; 47137c478bd9Sstevel@tonic-gate uint_t flags = 0; 47147c478bd9Sstevel@tonic-gate spgcnt_t dofree = 0; 47157c478bd9Sstevel@tonic-gate 47167c478bd9Sstevel@tonic-gate *nrelocp = 0; 47177c478bd9Sstevel@tonic-gate 47187c478bd9Sstevel@tonic-gate #if defined(__sparc) 47197c478bd9Sstevel@tonic-gate /* 47207c478bd9Sstevel@tonic-gate * We need to wait till OBP has completed 47217c478bd9Sstevel@tonic-gate * its boot-time handoff of its resources to the kernel 47227c478bd9Sstevel@tonic-gate * before we allow page relocation 47237c478bd9Sstevel@tonic-gate */ 47247c478bd9Sstevel@tonic-gate if (page_relocate_ready == 0) { 47257c478bd9Sstevel@tonic-gate return (EAGAIN); 47267c478bd9Sstevel@tonic-gate } 47277c478bd9Sstevel@tonic-gate #endif 47287c478bd9Sstevel@tonic-gate 47297c478bd9Sstevel@tonic-gate /* 47307c478bd9Sstevel@tonic-gate * If this is not a base page, 47317c478bd9Sstevel@tonic-gate * just return with 0x0 pages relocated. 47327c478bd9Sstevel@tonic-gate */ 47337c478bd9Sstevel@tonic-gate targ = *target; 47347c478bd9Sstevel@tonic-gate ASSERT(PAGE_EXCL(targ)); 47357c478bd9Sstevel@tonic-gate ASSERT(!PP_ISFREE(targ)); 47367c478bd9Sstevel@tonic-gate szc = targ->p_szc; 47377c478bd9Sstevel@tonic-gate ASSERT(szc < mmu_page_sizes); 47387c478bd9Sstevel@tonic-gate VM_STAT_ADD(vmm_vmstats.ppr_reloc[szc]); 47397c478bd9Sstevel@tonic-gate pfn = targ->p_pagenum; 47407c478bd9Sstevel@tonic-gate if (pfn != PFN_BASE(pfn, szc)) { 47417c478bd9Sstevel@tonic-gate VM_STAT_ADD(vmm_vmstats.ppr_relocnoroot[szc]); 47427c478bd9Sstevel@tonic-gate return (ERANGE); 47437c478bd9Sstevel@tonic-gate } 47447c478bd9Sstevel@tonic-gate 47457c478bd9Sstevel@tonic-gate if ((repl = *replacement) != NULL && repl->p_szc >= szc) { 47467c478bd9Sstevel@tonic-gate repl_pfn = repl->p_pagenum; 47477c478bd9Sstevel@tonic-gate if (repl_pfn != PFN_BASE(repl_pfn, szc)) { 47487c478bd9Sstevel@tonic-gate VM_STAT_ADD(vmm_vmstats.ppr_reloc_replnoroot[szc]); 47497c478bd9Sstevel@tonic-gate return (ERANGE); 47507c478bd9Sstevel@tonic-gate } 47517c478bd9Sstevel@tonic-gate repl_contig = 1; 47527c478bd9Sstevel@tonic-gate } 47537c478bd9Sstevel@tonic-gate 47547c478bd9Sstevel@tonic-gate /* 47557c478bd9Sstevel@tonic-gate * We must lock all members of this large page or we cannot 47567c478bd9Sstevel@tonic-gate * relocate any part of it. 47577c478bd9Sstevel@tonic-gate */ 47587c478bd9Sstevel@tonic-gate if (grouplock != 0 && !group_page_trylock(targ, SE_EXCL)) { 47597c478bd9Sstevel@tonic-gate VM_STAT_ADD(vmm_vmstats.ppr_relocnolock[targ->p_szc]); 47607c478bd9Sstevel@tonic-gate return (EBUSY); 47617c478bd9Sstevel@tonic-gate } 47627c478bd9Sstevel@tonic-gate 47637c478bd9Sstevel@tonic-gate /* 47647c478bd9Sstevel@tonic-gate * reread szc it could have been decreased before 47657c478bd9Sstevel@tonic-gate * group_page_trylock() was done. 47667c478bd9Sstevel@tonic-gate */ 47677c478bd9Sstevel@tonic-gate szc = targ->p_szc; 47687c478bd9Sstevel@tonic-gate ASSERT(szc < mmu_page_sizes); 47697c478bd9Sstevel@tonic-gate VM_STAT_ADD(vmm_vmstats.ppr_reloc[szc]); 47707c478bd9Sstevel@tonic-gate ASSERT(pfn == PFN_BASE(pfn, szc)); 47717c478bd9Sstevel@tonic-gate 47727c478bd9Sstevel@tonic-gate npgs = page_get_pagecnt(targ->p_szc); 47737c478bd9Sstevel@tonic-gate 47747c478bd9Sstevel@tonic-gate if (repl == NULL) { 47757c478bd9Sstevel@tonic-gate dofree = npgs; /* Size of target page in MMU pages */ 47767c478bd9Sstevel@tonic-gate if (!page_create_wait(dofree, 0)) { 47777c478bd9Sstevel@tonic-gate if (grouplock != 0) { 47787c478bd9Sstevel@tonic-gate group_page_unlock(targ); 47797c478bd9Sstevel@tonic-gate } 47807c478bd9Sstevel@tonic-gate VM_STAT_ADD(vmm_vmstats.ppr_relocnomem[szc]); 47817c478bd9Sstevel@tonic-gate return (ENOMEM); 47827c478bd9Sstevel@tonic-gate } 47837c478bd9Sstevel@tonic-gate 47847c478bd9Sstevel@tonic-gate /* 47857c478bd9Sstevel@tonic-gate * seg kmem pages require that the target and replacement 47867c478bd9Sstevel@tonic-gate * page be the same pagesize. 47877c478bd9Sstevel@tonic-gate */ 4788ad23a2dbSjohansen flags = (VN_ISKAS(targ->p_vnode)) ? PGR_SAMESZC : 0; 47897c478bd9Sstevel@tonic-gate repl = page_get_replacement_page(targ, lgrp, flags); 47907c478bd9Sstevel@tonic-gate if (repl == NULL) { 47917c478bd9Sstevel@tonic-gate if (grouplock != 0) { 47927c478bd9Sstevel@tonic-gate group_page_unlock(targ); 47937c478bd9Sstevel@tonic-gate } 47947c478bd9Sstevel@tonic-gate page_create_putback(dofree); 47957c478bd9Sstevel@tonic-gate VM_STAT_ADD(vmm_vmstats.ppr_relocnomem[szc]); 47967c478bd9Sstevel@tonic-gate return (ENOMEM); 47977c478bd9Sstevel@tonic-gate } 47987c478bd9Sstevel@tonic-gate } 47997c478bd9Sstevel@tonic-gate #ifdef DEBUG 48007c478bd9Sstevel@tonic-gate else { 48017c478bd9Sstevel@tonic-gate ASSERT(PAGE_LOCKED(repl)); 48027c478bd9Sstevel@tonic-gate } 48037c478bd9Sstevel@tonic-gate #endif /* DEBUG */ 48047c478bd9Sstevel@tonic-gate 48057c478bd9Sstevel@tonic-gate #if defined(__sparc) 48067c478bd9Sstevel@tonic-gate /* 48077c478bd9Sstevel@tonic-gate * Let hat_page_relocate() complete the relocation if it's kernel page 48087c478bd9Sstevel@tonic-gate */ 4809ad23a2dbSjohansen if (VN_ISKAS(targ->p_vnode)) { 48107c478bd9Sstevel@tonic-gate *replacement = repl; 48117c478bd9Sstevel@tonic-gate if (hat_page_relocate(target, replacement, nrelocp) != 0) { 48127c478bd9Sstevel@tonic-gate if (grouplock != 0) { 48137c478bd9Sstevel@tonic-gate group_page_unlock(targ); 48147c478bd9Sstevel@tonic-gate } 48157c478bd9Sstevel@tonic-gate if (dofree) { 48167c478bd9Sstevel@tonic-gate *replacement = NULL; 48177c478bd9Sstevel@tonic-gate page_free_replacement_page(repl); 48187c478bd9Sstevel@tonic-gate page_create_putback(dofree); 48197c478bd9Sstevel@tonic-gate } 48207c478bd9Sstevel@tonic-gate VM_STAT_ADD(vmm_vmstats.ppr_krelocfail[szc]); 48217c478bd9Sstevel@tonic-gate return (EAGAIN); 48227c478bd9Sstevel@tonic-gate } 48237c478bd9Sstevel@tonic-gate VM_STAT_ADD(vmm_vmstats.ppr_relocok[szc]); 48247c478bd9Sstevel@tonic-gate return (0); 48257c478bd9Sstevel@tonic-gate } 48267c478bd9Sstevel@tonic-gate #else 48277c478bd9Sstevel@tonic-gate #if defined(lint) 48287c478bd9Sstevel@tonic-gate dofree = dofree; 48297c478bd9Sstevel@tonic-gate #endif 48307c478bd9Sstevel@tonic-gate #endif 48317c478bd9Sstevel@tonic-gate 48327c478bd9Sstevel@tonic-gate first_repl = repl; 48337c478bd9Sstevel@tonic-gate 48347c478bd9Sstevel@tonic-gate for (i = 0; i < npgs; i++) { 48357c478bd9Sstevel@tonic-gate ASSERT(PAGE_EXCL(targ)); 483607b65a64Saguzovsk ASSERT(targ->p_slckcnt == 0); 483707b65a64Saguzovsk ASSERT(repl->p_slckcnt == 0); 48387c478bd9Sstevel@tonic-gate 48397c478bd9Sstevel@tonic-gate (void) hat_pageunload(targ, HAT_FORCE_PGUNLOAD); 48407c478bd9Sstevel@tonic-gate 48417c478bd9Sstevel@tonic-gate ASSERT(hat_page_getshare(targ) == 0); 48427c478bd9Sstevel@tonic-gate ASSERT(!PP_ISFREE(targ)); 48437c478bd9Sstevel@tonic-gate ASSERT(targ->p_pagenum == (pfn + i)); 48447c478bd9Sstevel@tonic-gate ASSERT(repl_contig == 0 || 48457c478bd9Sstevel@tonic-gate repl->p_pagenum == (repl_pfn + i)); 48467c478bd9Sstevel@tonic-gate 48477c478bd9Sstevel@tonic-gate /* 48487c478bd9Sstevel@tonic-gate * Copy the page contents and attributes then 48497c478bd9Sstevel@tonic-gate * relocate the page in the page hash. 48507c478bd9Sstevel@tonic-gate */ 48518b464eb8Smec if (ppcopy(targ, repl) == 0) { 48528b464eb8Smec targ = *target; 48538b464eb8Smec repl = first_repl; 48548b464eb8Smec VM_STAT_ADD(vmm_vmstats.ppr_copyfail); 48558b464eb8Smec if (grouplock != 0) { 48568b464eb8Smec group_page_unlock(targ); 48578b464eb8Smec } 48588b464eb8Smec if (dofree) { 48598b464eb8Smec *replacement = NULL; 48608b464eb8Smec page_free_replacement_page(repl); 48618b464eb8Smec page_create_putback(dofree); 48628b464eb8Smec } 48638b464eb8Smec return (EIO); 48648b464eb8Smec } 48658b464eb8Smec 48668b464eb8Smec targ++; 48678b464eb8Smec if (repl_contig != 0) { 48688b464eb8Smec repl++; 48698b464eb8Smec } else { 48708b464eb8Smec repl = repl->p_next; 48718b464eb8Smec } 48728b464eb8Smec } 48738b464eb8Smec 48748b464eb8Smec repl = first_repl; 48758b464eb8Smec targ = *target; 48768b464eb8Smec 48778b464eb8Smec for (i = 0; i < npgs; i++) { 48787c478bd9Sstevel@tonic-gate ppattr = hat_page_getattr(targ, (P_MOD | P_REF | P_RO)); 48797c478bd9Sstevel@tonic-gate page_clr_all_props(repl); 48807c478bd9Sstevel@tonic-gate page_set_props(repl, ppattr); 48817c478bd9Sstevel@tonic-gate page_relocate_hash(repl, targ); 48827c478bd9Sstevel@tonic-gate 48837c478bd9Sstevel@tonic-gate ASSERT(hat_page_getshare(targ) == 0); 48847c478bd9Sstevel@tonic-gate ASSERT(hat_page_getshare(repl) == 0); 48857c478bd9Sstevel@tonic-gate /* 48867c478bd9Sstevel@tonic-gate * Now clear the props on targ, after the 48877c478bd9Sstevel@tonic-gate * page_relocate_hash(), they no longer 48887c478bd9Sstevel@tonic-gate * have any meaning. 48897c478bd9Sstevel@tonic-gate */ 48907c478bd9Sstevel@tonic-gate page_clr_all_props(targ); 48917c478bd9Sstevel@tonic-gate ASSERT(targ->p_next == targ); 48927c478bd9Sstevel@tonic-gate ASSERT(targ->p_prev == targ); 48937c478bd9Sstevel@tonic-gate page_list_concat(&pl, &targ); 48947c478bd9Sstevel@tonic-gate 48957c478bd9Sstevel@tonic-gate targ++; 48967c478bd9Sstevel@tonic-gate if (repl_contig != 0) { 48977c478bd9Sstevel@tonic-gate repl++; 48987c478bd9Sstevel@tonic-gate } else { 48997c478bd9Sstevel@tonic-gate repl = repl->p_next; 49007c478bd9Sstevel@tonic-gate } 49017c478bd9Sstevel@tonic-gate } 49027c478bd9Sstevel@tonic-gate /* assert that we have come full circle with repl */ 49037c478bd9Sstevel@tonic-gate ASSERT(repl_contig == 1 || first_repl == repl); 49047c478bd9Sstevel@tonic-gate 49057c478bd9Sstevel@tonic-gate *target = pl; 49067c478bd9Sstevel@tonic-gate if (*replacement == NULL) { 49077c478bd9Sstevel@tonic-gate ASSERT(first_repl == repl); 49087c478bd9Sstevel@tonic-gate *replacement = repl; 49097c478bd9Sstevel@tonic-gate } 49107c478bd9Sstevel@tonic-gate VM_STAT_ADD(vmm_vmstats.ppr_relocok[szc]); 49117c478bd9Sstevel@tonic-gate *nrelocp = npgs; 49127c478bd9Sstevel@tonic-gate return (0); 49137c478bd9Sstevel@tonic-gate } 49147c478bd9Sstevel@tonic-gate /* 49157c478bd9Sstevel@tonic-gate * On success returns 0 and *nrelocp the number of PAGESIZE pages relocated. 49167c478bd9Sstevel@tonic-gate */ 49177c478bd9Sstevel@tonic-gate int 49187c478bd9Sstevel@tonic-gate page_relocate( 49197c478bd9Sstevel@tonic-gate page_t **target, 49207c478bd9Sstevel@tonic-gate page_t **replacement, 49217c478bd9Sstevel@tonic-gate int grouplock, 49227c478bd9Sstevel@tonic-gate int freetarget, 49237c478bd9Sstevel@tonic-gate spgcnt_t *nrelocp, 49247c478bd9Sstevel@tonic-gate lgrp_t *lgrp) 49257c478bd9Sstevel@tonic-gate { 49267c478bd9Sstevel@tonic-gate spgcnt_t ret; 49277c478bd9Sstevel@tonic-gate 49287c478bd9Sstevel@tonic-gate /* do_page_relocate returns 0 on success or errno value */ 49297c478bd9Sstevel@tonic-gate ret = do_page_relocate(target, replacement, grouplock, nrelocp, lgrp); 49307c478bd9Sstevel@tonic-gate 49317c478bd9Sstevel@tonic-gate if (ret != 0 || freetarget == 0) { 49327c478bd9Sstevel@tonic-gate return (ret); 49337c478bd9Sstevel@tonic-gate } 49347c478bd9Sstevel@tonic-gate if (*nrelocp == 1) { 49357c478bd9Sstevel@tonic-gate ASSERT(*target != NULL); 49367c478bd9Sstevel@tonic-gate page_free(*target, 1); 49377c478bd9Sstevel@tonic-gate } else { 49387c478bd9Sstevel@tonic-gate page_t *tpp = *target; 49397c478bd9Sstevel@tonic-gate uint_t szc = tpp->p_szc; 49407c478bd9Sstevel@tonic-gate pgcnt_t npgs = page_get_pagecnt(szc); 49417c478bd9Sstevel@tonic-gate ASSERT(npgs > 1); 49427c478bd9Sstevel@tonic-gate ASSERT(szc != 0); 49437c478bd9Sstevel@tonic-gate do { 49447c478bd9Sstevel@tonic-gate ASSERT(PAGE_EXCL(tpp)); 49457c478bd9Sstevel@tonic-gate ASSERT(!hat_page_is_mapped(tpp)); 49467c478bd9Sstevel@tonic-gate ASSERT(tpp->p_szc == szc); 49477c478bd9Sstevel@tonic-gate PP_SETFREE(tpp); 49487c478bd9Sstevel@tonic-gate PP_SETAGED(tpp); 49497c478bd9Sstevel@tonic-gate npgs--; 49507c478bd9Sstevel@tonic-gate } while ((tpp = tpp->p_next) != *target); 49517c478bd9Sstevel@tonic-gate ASSERT(npgs == 0); 49527c478bd9Sstevel@tonic-gate page_list_add_pages(*target, 0); 49537c478bd9Sstevel@tonic-gate npgs = page_get_pagecnt(szc); 49547c478bd9Sstevel@tonic-gate page_create_putback(npgs); 49557c478bd9Sstevel@tonic-gate } 49567c478bd9Sstevel@tonic-gate return (ret); 49577c478bd9Sstevel@tonic-gate } 49587c478bd9Sstevel@tonic-gate 49597c478bd9Sstevel@tonic-gate /* 49607c478bd9Sstevel@tonic-gate * it is up to the caller to deal with pcf accounting. 49617c478bd9Sstevel@tonic-gate */ 49627c478bd9Sstevel@tonic-gate void 49637c478bd9Sstevel@tonic-gate page_free_replacement_page(page_t *pplist) 49647c478bd9Sstevel@tonic-gate { 49657c478bd9Sstevel@tonic-gate page_t *pp; 49667c478bd9Sstevel@tonic-gate 49677c478bd9Sstevel@tonic-gate while (pplist != NULL) { 49687c478bd9Sstevel@tonic-gate /* 49697c478bd9Sstevel@tonic-gate * pp_targ is a linked list. 49707c478bd9Sstevel@tonic-gate */ 49717c478bd9Sstevel@tonic-gate pp = pplist; 49727c478bd9Sstevel@tonic-gate if (pp->p_szc == 0) { 49737c478bd9Sstevel@tonic-gate page_sub(&pplist, pp); 49747c478bd9Sstevel@tonic-gate page_clr_all_props(pp); 49757c478bd9Sstevel@tonic-gate PP_SETFREE(pp); 49767c478bd9Sstevel@tonic-gate PP_SETAGED(pp); 49777c478bd9Sstevel@tonic-gate page_list_add(pp, PG_FREE_LIST | PG_LIST_TAIL); 49787c478bd9Sstevel@tonic-gate page_unlock(pp); 49797c478bd9Sstevel@tonic-gate VM_STAT_ADD(pagecnt.pc_free_replacement_page[0]); 49807c478bd9Sstevel@tonic-gate } else { 49817c478bd9Sstevel@tonic-gate spgcnt_t curnpgs = page_get_pagecnt(pp->p_szc); 49827c478bd9Sstevel@tonic-gate page_t *tpp; 49837c478bd9Sstevel@tonic-gate page_list_break(&pp, &pplist, curnpgs); 49847c478bd9Sstevel@tonic-gate tpp = pp; 49857c478bd9Sstevel@tonic-gate do { 49867c478bd9Sstevel@tonic-gate ASSERT(PAGE_EXCL(tpp)); 49877c478bd9Sstevel@tonic-gate ASSERT(!hat_page_is_mapped(tpp)); 49887c478bd9Sstevel@tonic-gate page_clr_all_props(pp); 49897c478bd9Sstevel@tonic-gate PP_SETFREE(tpp); 49907c478bd9Sstevel@tonic-gate PP_SETAGED(tpp); 49917c478bd9Sstevel@tonic-gate } while ((tpp = tpp->p_next) != pp); 49927c478bd9Sstevel@tonic-gate page_list_add_pages(pp, 0); 49937c478bd9Sstevel@tonic-gate VM_STAT_ADD(pagecnt.pc_free_replacement_page[1]); 49947c478bd9Sstevel@tonic-gate } 49957c478bd9Sstevel@tonic-gate } 49967c478bd9Sstevel@tonic-gate } 49977c478bd9Sstevel@tonic-gate 49987c478bd9Sstevel@tonic-gate /* 49997c478bd9Sstevel@tonic-gate * Relocate target to non-relocatable replacement page. 50007c478bd9Sstevel@tonic-gate */ 50017c478bd9Sstevel@tonic-gate int 50027c478bd9Sstevel@tonic-gate page_relocate_cage(page_t **target, page_t **replacement) 50037c478bd9Sstevel@tonic-gate { 50047c478bd9Sstevel@tonic-gate page_t *tpp, *rpp; 50057c478bd9Sstevel@tonic-gate spgcnt_t pgcnt, npgs; 50067c478bd9Sstevel@tonic-gate int result; 50077c478bd9Sstevel@tonic-gate 50087c478bd9Sstevel@tonic-gate tpp = *target; 50097c478bd9Sstevel@tonic-gate 50107c478bd9Sstevel@tonic-gate ASSERT(PAGE_EXCL(tpp)); 50117c478bd9Sstevel@tonic-gate ASSERT(tpp->p_szc == 0); 50127c478bd9Sstevel@tonic-gate 50137c478bd9Sstevel@tonic-gate pgcnt = btop(page_get_pagesize(tpp->p_szc)); 50147c478bd9Sstevel@tonic-gate 50157c478bd9Sstevel@tonic-gate do { 50167c478bd9Sstevel@tonic-gate (void) page_create_wait(pgcnt, PG_WAIT | PG_NORELOC); 50177c478bd9Sstevel@tonic-gate rpp = page_get_replacement_page(tpp, NULL, PGR_NORELOC); 50187c478bd9Sstevel@tonic-gate if (rpp == NULL) { 50197c478bd9Sstevel@tonic-gate page_create_putback(pgcnt); 50207c478bd9Sstevel@tonic-gate kcage_cageout_wakeup(); 50217c478bd9Sstevel@tonic-gate } 50227c478bd9Sstevel@tonic-gate } while (rpp == NULL); 50237c478bd9Sstevel@tonic-gate 50247c478bd9Sstevel@tonic-gate ASSERT(PP_ISNORELOC(rpp)); 50257c478bd9Sstevel@tonic-gate 50267c478bd9Sstevel@tonic-gate result = page_relocate(&tpp, &rpp, 0, 1, &npgs, NULL); 50277c478bd9Sstevel@tonic-gate 50287c478bd9Sstevel@tonic-gate if (result == 0) { 50297c478bd9Sstevel@tonic-gate *replacement = rpp; 50307c478bd9Sstevel@tonic-gate if (pgcnt != npgs) 50317c478bd9Sstevel@tonic-gate panic("page_relocate_cage: partial relocation"); 50327c478bd9Sstevel@tonic-gate } 50337c478bd9Sstevel@tonic-gate 50347c478bd9Sstevel@tonic-gate return (result); 50357c478bd9Sstevel@tonic-gate } 50367c478bd9Sstevel@tonic-gate 50377c478bd9Sstevel@tonic-gate /* 50387c478bd9Sstevel@tonic-gate * Release the page lock on a page, place on cachelist 50397c478bd9Sstevel@tonic-gate * tail if no longer mapped. Caller can let us know if 50407c478bd9Sstevel@tonic-gate * the page is known to be clean. 50417c478bd9Sstevel@tonic-gate */ 50427c478bd9Sstevel@tonic-gate int 50437c478bd9Sstevel@tonic-gate page_release(page_t *pp, int checkmod) 50447c478bd9Sstevel@tonic-gate { 50457c478bd9Sstevel@tonic-gate int status; 50467c478bd9Sstevel@tonic-gate 50477c478bd9Sstevel@tonic-gate ASSERT(PAGE_LOCKED(pp) && !PP_ISFREE(pp) && 50486e4dd838Smec (pp->p_vnode != NULL)); 50497c478bd9Sstevel@tonic-gate 50507c478bd9Sstevel@tonic-gate if (!hat_page_is_mapped(pp) && !IS_SWAPVP(pp->p_vnode) && 50517c478bd9Sstevel@tonic-gate ((PAGE_SHARED(pp) && page_tryupgrade(pp)) || PAGE_EXCL(pp)) && 50527c478bd9Sstevel@tonic-gate pp->p_lckcnt == 0 && pp->p_cowcnt == 0 && 50537c478bd9Sstevel@tonic-gate !hat_page_is_mapped(pp)) { 50547c478bd9Sstevel@tonic-gate 50557c478bd9Sstevel@tonic-gate /* 50567c478bd9Sstevel@tonic-gate * If page is modified, unlock it 50577c478bd9Sstevel@tonic-gate * 50587c478bd9Sstevel@tonic-gate * (p_nrm & P_MOD) bit has the latest stuff because: 50597c478bd9Sstevel@tonic-gate * (1) We found that this page doesn't have any mappings 50607c478bd9Sstevel@tonic-gate * _after_ holding SE_EXCL and 50617c478bd9Sstevel@tonic-gate * (2) We didn't drop SE_EXCL lock after the check in (1) 50627c478bd9Sstevel@tonic-gate */ 50637c478bd9Sstevel@tonic-gate if (checkmod && hat_ismod(pp)) { 50647c478bd9Sstevel@tonic-gate page_unlock(pp); 50657c478bd9Sstevel@tonic-gate status = PGREL_MOD; 50667c478bd9Sstevel@tonic-gate } else { 50677c478bd9Sstevel@tonic-gate /*LINTED: constant in conditional context*/ 50687c478bd9Sstevel@tonic-gate VN_DISPOSE(pp, B_FREE, 0, kcred); 50697c478bd9Sstevel@tonic-gate status = PGREL_CLEAN; 50707c478bd9Sstevel@tonic-gate } 50717c478bd9Sstevel@tonic-gate } else { 50727c478bd9Sstevel@tonic-gate page_unlock(pp); 50737c478bd9Sstevel@tonic-gate status = PGREL_NOTREL; 50747c478bd9Sstevel@tonic-gate } 50757c478bd9Sstevel@tonic-gate return (status); 50767c478bd9Sstevel@tonic-gate } 50777c478bd9Sstevel@tonic-gate 5078db874c57Selowe /* 5079db874c57Selowe * Given a constituent page, try to demote the large page on the freelist. 5080db874c57Selowe * 5081db874c57Selowe * Returns nonzero if the page could be demoted successfully. Returns with 5082db874c57Selowe * the constituent page still locked. 5083db874c57Selowe */ 5084db874c57Selowe int 5085db874c57Selowe page_try_demote_free_pages(page_t *pp) 5086db874c57Selowe { 5087db874c57Selowe page_t *rootpp = pp; 5088db874c57Selowe pfn_t pfn = page_pptonum(pp); 5089db874c57Selowe spgcnt_t npgs; 5090db874c57Selowe uint_t szc = pp->p_szc; 5091db874c57Selowe 5092db874c57Selowe ASSERT(PP_ISFREE(pp)); 5093db874c57Selowe ASSERT(PAGE_EXCL(pp)); 5094db874c57Selowe 5095db874c57Selowe /* 5096db874c57Selowe * Adjust rootpp and lock it, if `pp' is not the base 5097db874c57Selowe * constituent page. 5098db874c57Selowe */ 5099db874c57Selowe npgs = page_get_pagecnt(pp->p_szc); 5100db874c57Selowe if (npgs == 1) { 5101db874c57Selowe return (0); 5102db874c57Selowe } 5103db874c57Selowe 5104db874c57Selowe if (!IS_P2ALIGNED(pfn, npgs)) { 5105db874c57Selowe pfn = P2ALIGN(pfn, npgs); 5106db874c57Selowe rootpp = page_numtopp_nolock(pfn); 5107db874c57Selowe } 5108db874c57Selowe 5109db874c57Selowe if (pp != rootpp && !page_trylock(rootpp, SE_EXCL)) { 5110db874c57Selowe return (0); 5111db874c57Selowe } 5112db874c57Selowe 5113db874c57Selowe if (rootpp->p_szc != szc) { 5114db874c57Selowe if (pp != rootpp) 5115db874c57Selowe page_unlock(rootpp); 5116db874c57Selowe return (0); 5117db874c57Selowe } 5118db874c57Selowe 5119db874c57Selowe page_demote_free_pages(rootpp); 5120db874c57Selowe 5121db874c57Selowe if (pp != rootpp) 5122db874c57Selowe page_unlock(rootpp); 5123db874c57Selowe 5124db874c57Selowe ASSERT(PP_ISFREE(pp)); 5125db874c57Selowe ASSERT(PAGE_EXCL(pp)); 5126db874c57Selowe return (1); 5127db874c57Selowe } 5128db874c57Selowe 5129db874c57Selowe /* 5130db874c57Selowe * Given a constituent page, try to demote the large page. 5131db874c57Selowe * 5132db874c57Selowe * Returns nonzero if the page could be demoted successfully. Returns with 5133db874c57Selowe * the constituent page still locked. 5134db874c57Selowe */ 51357c478bd9Sstevel@tonic-gate int 51367c478bd9Sstevel@tonic-gate page_try_demote_pages(page_t *pp) 51377c478bd9Sstevel@tonic-gate { 51387c478bd9Sstevel@tonic-gate page_t *tpp, *rootpp = pp; 51397c478bd9Sstevel@tonic-gate pfn_t pfn = page_pptonum(pp); 51407c478bd9Sstevel@tonic-gate spgcnt_t i, npgs; 51417c478bd9Sstevel@tonic-gate uint_t szc = pp->p_szc; 51427c478bd9Sstevel@tonic-gate vnode_t *vp = pp->p_vnode; 51437c478bd9Sstevel@tonic-gate 5144db874c57Selowe ASSERT(PAGE_EXCL(pp)); 51457c478bd9Sstevel@tonic-gate 51467c478bd9Sstevel@tonic-gate VM_STAT_ADD(pagecnt.pc_try_demote_pages[0]); 51477c478bd9Sstevel@tonic-gate 5148db874c57Selowe if (pp->p_szc == 0) { 51497c478bd9Sstevel@tonic-gate VM_STAT_ADD(pagecnt.pc_try_demote_pages[1]); 51507c478bd9Sstevel@tonic-gate return (1); 51517c478bd9Sstevel@tonic-gate } 51527c478bd9Sstevel@tonic-gate 5153ad23a2dbSjohansen if (vp != NULL && !IS_SWAPFSVP(vp) && !VN_ISKAS(vp)) { 51547c478bd9Sstevel@tonic-gate VM_STAT_ADD(pagecnt.pc_try_demote_pages[2]); 5155db874c57Selowe page_demote_vp_pages(pp); 51567c478bd9Sstevel@tonic-gate ASSERT(pp->p_szc == 0); 51577c478bd9Sstevel@tonic-gate return (1); 51587c478bd9Sstevel@tonic-gate } 51597c478bd9Sstevel@tonic-gate 51607c478bd9Sstevel@tonic-gate /* 5161db874c57Selowe * Adjust rootpp if passed in is not the base 51627c478bd9Sstevel@tonic-gate * constituent page. 51637c478bd9Sstevel@tonic-gate */ 5164db874c57Selowe npgs = page_get_pagecnt(pp->p_szc); 51657c478bd9Sstevel@tonic-gate ASSERT(npgs > 1); 51667c478bd9Sstevel@tonic-gate if (!IS_P2ALIGNED(pfn, npgs)) { 51677c478bd9Sstevel@tonic-gate pfn = P2ALIGN(pfn, npgs); 51687c478bd9Sstevel@tonic-gate rootpp = page_numtopp_nolock(pfn); 51697c478bd9Sstevel@tonic-gate VM_STAT_ADD(pagecnt.pc_try_demote_pages[3]); 51707c478bd9Sstevel@tonic-gate ASSERT(rootpp->p_vnode != NULL); 51717c478bd9Sstevel@tonic-gate ASSERT(rootpp->p_szc == szc); 51727c478bd9Sstevel@tonic-gate } 51737c478bd9Sstevel@tonic-gate 51747c478bd9Sstevel@tonic-gate /* 51757c478bd9Sstevel@tonic-gate * We can't demote kernel pages since we can't hat_unload() 51767c478bd9Sstevel@tonic-gate * the mappings. 51777c478bd9Sstevel@tonic-gate */ 5178ad23a2dbSjohansen if (VN_ISKAS(rootpp->p_vnode)) 51797c478bd9Sstevel@tonic-gate return (0); 51807c478bd9Sstevel@tonic-gate 51817c478bd9Sstevel@tonic-gate /* 51827c478bd9Sstevel@tonic-gate * Attempt to lock all constituent pages except the page passed 51837c478bd9Sstevel@tonic-gate * in since it's already locked. 51847c478bd9Sstevel@tonic-gate */ 5185affbd3ccSkchow for (tpp = rootpp, i = 0; i < npgs; i++, tpp++) { 51867c478bd9Sstevel@tonic-gate ASSERT(!PP_ISFREE(tpp)); 51877c478bd9Sstevel@tonic-gate ASSERT(tpp->p_vnode != NULL); 51887c478bd9Sstevel@tonic-gate 51897c478bd9Sstevel@tonic-gate if (tpp != pp && !page_trylock(tpp, SE_EXCL)) 51907c478bd9Sstevel@tonic-gate break; 51917c478bd9Sstevel@tonic-gate ASSERT(tpp->p_szc == rootpp->p_szc); 51927c478bd9Sstevel@tonic-gate ASSERT(page_pptonum(tpp) == page_pptonum(rootpp) + i); 51937c478bd9Sstevel@tonic-gate } 51947c478bd9Sstevel@tonic-gate 51957c478bd9Sstevel@tonic-gate /* 5196db874c57Selowe * If we failed to lock them all then unlock what we have 5197db874c57Selowe * locked so far and bail. 51987c478bd9Sstevel@tonic-gate */ 51997c478bd9Sstevel@tonic-gate if (i < npgs) { 52007c478bd9Sstevel@tonic-gate tpp = rootpp; 52017c478bd9Sstevel@tonic-gate while (i-- > 0) { 52027c478bd9Sstevel@tonic-gate if (tpp != pp) 52037c478bd9Sstevel@tonic-gate page_unlock(tpp); 5204affbd3ccSkchow tpp++; 52057c478bd9Sstevel@tonic-gate } 52067c478bd9Sstevel@tonic-gate VM_STAT_ADD(pagecnt.pc_try_demote_pages[4]); 52077c478bd9Sstevel@tonic-gate return (0); 52087c478bd9Sstevel@tonic-gate } 52097c478bd9Sstevel@tonic-gate 5210affbd3ccSkchow for (tpp = rootpp, i = 0; i < npgs; i++, tpp++) { 52117c478bd9Sstevel@tonic-gate ASSERT(PAGE_EXCL(tpp)); 521207b65a64Saguzovsk ASSERT(tpp->p_slckcnt == 0); 5213db874c57Selowe (void) hat_pageunload(tpp, HAT_FORCE_PGUNLOAD); 52147c478bd9Sstevel@tonic-gate tpp->p_szc = 0; 52157c478bd9Sstevel@tonic-gate } 52167c478bd9Sstevel@tonic-gate 52177c478bd9Sstevel@tonic-gate /* 52187c478bd9Sstevel@tonic-gate * Unlock all pages except the page passed in. 52197c478bd9Sstevel@tonic-gate */ 5220affbd3ccSkchow for (tpp = rootpp, i = 0; i < npgs; i++, tpp++) { 52217c478bd9Sstevel@tonic-gate ASSERT(!hat_page_is_mapped(tpp)); 52227c478bd9Sstevel@tonic-gate if (tpp != pp) 52237c478bd9Sstevel@tonic-gate page_unlock(tpp); 52247c478bd9Sstevel@tonic-gate } 5225db874c57Selowe 52267c478bd9Sstevel@tonic-gate VM_STAT_ADD(pagecnt.pc_try_demote_pages[5]); 52277c478bd9Sstevel@tonic-gate return (1); 52287c478bd9Sstevel@tonic-gate } 52297c478bd9Sstevel@tonic-gate 52307c478bd9Sstevel@tonic-gate /* 52317c478bd9Sstevel@tonic-gate * Called by page_free() and page_destroy() to demote the page size code 52327c478bd9Sstevel@tonic-gate * (p_szc) to 0 (since we can't just put a single PAGESIZE page with non zero 52337c478bd9Sstevel@tonic-gate * p_szc on free list, neither can we just clear p_szc of a single page_t 52347c478bd9Sstevel@tonic-gate * within a large page since it will break other code that relies on p_szc 52357c478bd9Sstevel@tonic-gate * being the same for all page_t's of a large page). Anonymous pages should 52367c478bd9Sstevel@tonic-gate * never end up here because anon_map_getpages() cannot deal with p_szc 52377c478bd9Sstevel@tonic-gate * changes after a single constituent page is locked. While anonymous or 52387c478bd9Sstevel@tonic-gate * kernel large pages are demoted or freed the entire large page at a time 52397c478bd9Sstevel@tonic-gate * with all constituent pages locked EXCL for the file system pages we 52407c478bd9Sstevel@tonic-gate * have to be able to demote a large page (i.e. decrease all constituent pages 52417c478bd9Sstevel@tonic-gate * p_szc) with only just an EXCL lock on one of constituent pages. The reason 52427c478bd9Sstevel@tonic-gate * we can easily deal with anonymous page demotion the entire large page at a 52437c478bd9Sstevel@tonic-gate * time is that those operation originate at address space level and concern 52447c478bd9Sstevel@tonic-gate * the entire large page region with actual demotion only done when pages are 52457c478bd9Sstevel@tonic-gate * not shared with any other processes (therefore we can always get EXCL lock 52467c478bd9Sstevel@tonic-gate * on all anonymous constituent pages after clearing segment page 52477c478bd9Sstevel@tonic-gate * cache). However file system pages can be truncated or invalidated at a 52487c478bd9Sstevel@tonic-gate * PAGESIZE level from the file system side and end up in page_free() or 52497c478bd9Sstevel@tonic-gate * page_destroy() (we also allow only part of the large page to be SOFTLOCKed 5250da6c28aaSamw * and therefore pageout should be able to demote a large page by EXCL locking 52517c478bd9Sstevel@tonic-gate * any constituent page that is not under SOFTLOCK). In those cases we cannot 52527c478bd9Sstevel@tonic-gate * rely on being able to lock EXCL all constituent pages. 52537c478bd9Sstevel@tonic-gate * 52547c478bd9Sstevel@tonic-gate * To prevent szc changes on file system pages one has to lock all constituent 52557c478bd9Sstevel@tonic-gate * pages at least SHARED (or call page_szc_lock()). The only subsystem that 52567c478bd9Sstevel@tonic-gate * doesn't rely on locking all constituent pages (or using page_szc_lock()) to 52577c478bd9Sstevel@tonic-gate * prevent szc changes is hat layer that uses its own page level mlist 52587c478bd9Sstevel@tonic-gate * locks. hat assumes that szc doesn't change after mlist lock for a page is 52597c478bd9Sstevel@tonic-gate * taken. Therefore we need to change szc under hat level locks if we only 52607c478bd9Sstevel@tonic-gate * have an EXCL lock on a single constituent page and hat still references any 52617c478bd9Sstevel@tonic-gate * of constituent pages. (Note we can't "ignore" hat layer by simply 52627c478bd9Sstevel@tonic-gate * hat_pageunload() all constituent pages without having EXCL locks on all of 52637c478bd9Sstevel@tonic-gate * constituent pages). We use hat_page_demote() call to safely demote szc of 52647c478bd9Sstevel@tonic-gate * all constituent pages under hat locks when we only have an EXCL lock on one 52657c478bd9Sstevel@tonic-gate * of constituent pages. 52667c478bd9Sstevel@tonic-gate * 52677c478bd9Sstevel@tonic-gate * This routine calls page_szc_lock() before calling hat_page_demote() to 52687c478bd9Sstevel@tonic-gate * allow segvn in one special case not to lock all constituent pages SHARED 5269da6c28aaSamw * before calling hat_memload_array() that relies on p_szc not changing even 52707c478bd9Sstevel@tonic-gate * before hat level mlist lock is taken. In that case segvn uses 5271da6c28aaSamw * page_szc_lock() to prevent hat_page_demote() changing p_szc values. 52727c478bd9Sstevel@tonic-gate * 52737c478bd9Sstevel@tonic-gate * Anonymous or kernel page demotion still has to lock all pages exclusively 52747c478bd9Sstevel@tonic-gate * and do hat_pageunload() on all constituent pages before demoting the page 52757c478bd9Sstevel@tonic-gate * therefore there's no need for anonymous or kernel page demotion to use 52767c478bd9Sstevel@tonic-gate * hat_page_demote() mechanism. 52777c478bd9Sstevel@tonic-gate * 52787c478bd9Sstevel@tonic-gate * hat_page_demote() removes all large mappings that map pp and then decreases 52797c478bd9Sstevel@tonic-gate * p_szc starting from the last constituent page of the large page. By working 52807c478bd9Sstevel@tonic-gate * from the tail of a large page in pfn decreasing order allows one looking at 52817c478bd9Sstevel@tonic-gate * the root page to know that hat_page_demote() is done for root's szc area. 52827c478bd9Sstevel@tonic-gate * e.g. if a root page has szc 1 one knows it only has to lock all constituent 52837c478bd9Sstevel@tonic-gate * pages within szc 1 area to prevent szc changes because hat_page_demote() 52847c478bd9Sstevel@tonic-gate * that started on this page when it had szc > 1 is done for this szc 1 area. 52857c478bd9Sstevel@tonic-gate * 5286da6c28aaSamw * We are guaranteed that all constituent pages of pp's large page belong to 52877c478bd9Sstevel@tonic-gate * the same vnode with the consecutive offsets increasing in the direction of 52887c478bd9Sstevel@tonic-gate * the pfn i.e. the identity of constituent pages can't change until their 52897c478bd9Sstevel@tonic-gate * p_szc is decreased. Therefore it's safe for hat_page_demote() to remove 52907c478bd9Sstevel@tonic-gate * large mappings to pp even though we don't lock any constituent page except 52917c478bd9Sstevel@tonic-gate * pp (i.e. we won't unload e.g. kernel locked page). 52927c478bd9Sstevel@tonic-gate */ 52937c478bd9Sstevel@tonic-gate static void 52947c478bd9Sstevel@tonic-gate page_demote_vp_pages(page_t *pp) 52957c478bd9Sstevel@tonic-gate { 52967c478bd9Sstevel@tonic-gate kmutex_t *mtx; 52977c478bd9Sstevel@tonic-gate 52987c478bd9Sstevel@tonic-gate ASSERT(PAGE_EXCL(pp)); 52997c478bd9Sstevel@tonic-gate ASSERT(!PP_ISFREE(pp)); 53007c478bd9Sstevel@tonic-gate ASSERT(pp->p_vnode != NULL); 53017c478bd9Sstevel@tonic-gate ASSERT(!IS_SWAPFSVP(pp->p_vnode)); 5302ad23a2dbSjohansen ASSERT(!PP_ISKAS(pp)); 53037c478bd9Sstevel@tonic-gate 53047c478bd9Sstevel@tonic-gate VM_STAT_ADD(pagecnt.pc_demote_pages[0]); 53057c478bd9Sstevel@tonic-gate 53067c478bd9Sstevel@tonic-gate mtx = page_szc_lock(pp); 53077c478bd9Sstevel@tonic-gate if (mtx != NULL) { 53087c478bd9Sstevel@tonic-gate hat_page_demote(pp); 53097c478bd9Sstevel@tonic-gate mutex_exit(mtx); 53107c478bd9Sstevel@tonic-gate } 53117c478bd9Sstevel@tonic-gate ASSERT(pp->p_szc == 0); 53127c478bd9Sstevel@tonic-gate } 53137c478bd9Sstevel@tonic-gate 53147c478bd9Sstevel@tonic-gate /* 53157c478bd9Sstevel@tonic-gate * Mark any existing pages for migration in the given range 53167c478bd9Sstevel@tonic-gate */ 53177c478bd9Sstevel@tonic-gate void 53187c478bd9Sstevel@tonic-gate page_mark_migrate(struct seg *seg, caddr_t addr, size_t len, 53197c478bd9Sstevel@tonic-gate struct anon_map *amp, ulong_t anon_index, vnode_t *vp, 53207c478bd9Sstevel@tonic-gate u_offset_t vnoff, int rflag) 53217c478bd9Sstevel@tonic-gate { 53227c478bd9Sstevel@tonic-gate struct anon *ap; 53237c478bd9Sstevel@tonic-gate vnode_t *curvp; 53247c478bd9Sstevel@tonic-gate lgrp_t *from; 53257c478bd9Sstevel@tonic-gate pgcnt_t i; 53267c478bd9Sstevel@tonic-gate pgcnt_t nlocked; 53277c478bd9Sstevel@tonic-gate u_offset_t off; 53287c478bd9Sstevel@tonic-gate pfn_t pfn; 53297c478bd9Sstevel@tonic-gate size_t pgsz; 53307c478bd9Sstevel@tonic-gate size_t segpgsz; 53317c478bd9Sstevel@tonic-gate pgcnt_t pages; 53327c478bd9Sstevel@tonic-gate uint_t pszc; 53337c478bd9Sstevel@tonic-gate page_t **ppa; 53347c478bd9Sstevel@tonic-gate pgcnt_t ppa_nentries; 53357c478bd9Sstevel@tonic-gate page_t *pp; 53367c478bd9Sstevel@tonic-gate caddr_t va; 53377c478bd9Sstevel@tonic-gate ulong_t an_idx; 53387c478bd9Sstevel@tonic-gate anon_sync_obj_t cookie; 53397c478bd9Sstevel@tonic-gate 53407c478bd9Sstevel@tonic-gate ASSERT(seg->s_as && AS_LOCK_HELD(seg->s_as, &seg->s_as->a_lock)); 53417c478bd9Sstevel@tonic-gate 53427c478bd9Sstevel@tonic-gate /* 53437c478bd9Sstevel@tonic-gate * Don't do anything if don't need to do lgroup optimizations 53447c478bd9Sstevel@tonic-gate * on this system 53457c478bd9Sstevel@tonic-gate */ 53467c478bd9Sstevel@tonic-gate if (!lgrp_optimizations()) 53477c478bd9Sstevel@tonic-gate return; 53487c478bd9Sstevel@tonic-gate 53497c478bd9Sstevel@tonic-gate /* 53507c478bd9Sstevel@tonic-gate * Align address and length to (potentially large) page boundary 53517c478bd9Sstevel@tonic-gate */ 53527c478bd9Sstevel@tonic-gate segpgsz = page_get_pagesize(seg->s_szc); 53537c478bd9Sstevel@tonic-gate addr = (caddr_t)P2ALIGN((uintptr_t)addr, segpgsz); 53547c478bd9Sstevel@tonic-gate if (rflag) 53557c478bd9Sstevel@tonic-gate len = P2ROUNDUP(len, segpgsz); 53567c478bd9Sstevel@tonic-gate 53577c478bd9Sstevel@tonic-gate /* 5358da6c28aaSamw * Allocate page array to accommodate largest page size 53597c478bd9Sstevel@tonic-gate */ 53607c478bd9Sstevel@tonic-gate pgsz = page_get_pagesize(page_num_pagesizes() - 1); 53617c478bd9Sstevel@tonic-gate ppa_nentries = btop(pgsz); 53627c478bd9Sstevel@tonic-gate ppa = kmem_zalloc(ppa_nentries * sizeof (page_t *), KM_SLEEP); 53637c478bd9Sstevel@tonic-gate 53647c478bd9Sstevel@tonic-gate /* 53657c478bd9Sstevel@tonic-gate * Do one (large) page at a time 53667c478bd9Sstevel@tonic-gate */ 53677c478bd9Sstevel@tonic-gate va = addr; 53687c478bd9Sstevel@tonic-gate while (va < addr + len) { 53697c478bd9Sstevel@tonic-gate /* 53707c478bd9Sstevel@tonic-gate * Lookup (root) page for vnode and offset corresponding to 53717c478bd9Sstevel@tonic-gate * this virtual address 53727c478bd9Sstevel@tonic-gate * Try anonmap first since there may be copy-on-write 53737c478bd9Sstevel@tonic-gate * pages, but initialize vnode pointer and offset using 53747c478bd9Sstevel@tonic-gate * vnode arguments just in case there isn't an amp. 53757c478bd9Sstevel@tonic-gate */ 53767c478bd9Sstevel@tonic-gate curvp = vp; 53777c478bd9Sstevel@tonic-gate off = vnoff + va - seg->s_base; 53787c478bd9Sstevel@tonic-gate if (amp) { 53797c478bd9Sstevel@tonic-gate ANON_LOCK_ENTER(&->a_rwlock, RW_READER); 53807c478bd9Sstevel@tonic-gate an_idx = anon_index + seg_page(seg, va); 53817c478bd9Sstevel@tonic-gate anon_array_enter(amp, an_idx, &cookie); 53827c478bd9Sstevel@tonic-gate ap = anon_get_ptr(amp->ahp, an_idx); 53837c478bd9Sstevel@tonic-gate if (ap) 53847c478bd9Sstevel@tonic-gate swap_xlate(ap, &curvp, &off); 53857c478bd9Sstevel@tonic-gate anon_array_exit(&cookie); 53867c478bd9Sstevel@tonic-gate ANON_LOCK_EXIT(&->a_rwlock); 53877c478bd9Sstevel@tonic-gate } 53887c478bd9Sstevel@tonic-gate 53897c478bd9Sstevel@tonic-gate pp = NULL; 53907c478bd9Sstevel@tonic-gate if (curvp) 53917c478bd9Sstevel@tonic-gate pp = page_lookup(curvp, off, SE_SHARED); 53927c478bd9Sstevel@tonic-gate 53937c478bd9Sstevel@tonic-gate /* 53947c478bd9Sstevel@tonic-gate * If there isn't a page at this virtual address, 53957c478bd9Sstevel@tonic-gate * skip to next page 53967c478bd9Sstevel@tonic-gate */ 53977c478bd9Sstevel@tonic-gate if (pp == NULL) { 53987c478bd9Sstevel@tonic-gate va += PAGESIZE; 53997c478bd9Sstevel@tonic-gate continue; 54007c478bd9Sstevel@tonic-gate } 54017c478bd9Sstevel@tonic-gate 54027c478bd9Sstevel@tonic-gate /* 54037c478bd9Sstevel@tonic-gate * Figure out which lgroup this page is in for kstats 54047c478bd9Sstevel@tonic-gate */ 54057c478bd9Sstevel@tonic-gate pfn = page_pptonum(pp); 54067c478bd9Sstevel@tonic-gate from = lgrp_pfn_to_lgrp(pfn); 54077c478bd9Sstevel@tonic-gate 54087c478bd9Sstevel@tonic-gate /* 54097c478bd9Sstevel@tonic-gate * Get page size, and round up and skip to next page boundary 54107c478bd9Sstevel@tonic-gate * if unaligned address 54117c478bd9Sstevel@tonic-gate */ 54127c478bd9Sstevel@tonic-gate pszc = pp->p_szc; 54137c478bd9Sstevel@tonic-gate pgsz = page_get_pagesize(pszc); 54147c478bd9Sstevel@tonic-gate pages = btop(pgsz); 54157c478bd9Sstevel@tonic-gate if (!IS_P2ALIGNED(va, pgsz) || 54167c478bd9Sstevel@tonic-gate !IS_P2ALIGNED(pfn, pages) || 54177c478bd9Sstevel@tonic-gate pgsz > segpgsz) { 54187c478bd9Sstevel@tonic-gate pgsz = MIN(pgsz, segpgsz); 54197c478bd9Sstevel@tonic-gate page_unlock(pp); 54207c478bd9Sstevel@tonic-gate i = btop(P2END((uintptr_t)va, pgsz) - 54217c478bd9Sstevel@tonic-gate (uintptr_t)va); 54227c478bd9Sstevel@tonic-gate va = (caddr_t)P2END((uintptr_t)va, pgsz); 54237c478bd9Sstevel@tonic-gate lgrp_stat_add(from->lgrp_id, LGRP_PMM_FAIL_PGS, i); 54247c478bd9Sstevel@tonic-gate continue; 54257c478bd9Sstevel@tonic-gate } 54267c478bd9Sstevel@tonic-gate 54277c478bd9Sstevel@tonic-gate /* 54287c478bd9Sstevel@tonic-gate * Upgrade to exclusive lock on page 54297c478bd9Sstevel@tonic-gate */ 54307c478bd9Sstevel@tonic-gate if (!page_tryupgrade(pp)) { 54317c478bd9Sstevel@tonic-gate page_unlock(pp); 54327c478bd9Sstevel@tonic-gate va += pgsz; 54337c478bd9Sstevel@tonic-gate lgrp_stat_add(from->lgrp_id, LGRP_PMM_FAIL_PGS, 54347c478bd9Sstevel@tonic-gate btop(pgsz)); 54357c478bd9Sstevel@tonic-gate continue; 54367c478bd9Sstevel@tonic-gate } 54377c478bd9Sstevel@tonic-gate 54387c478bd9Sstevel@tonic-gate /* 54397c478bd9Sstevel@tonic-gate * Remember pages locked exclusively and how many 54407c478bd9Sstevel@tonic-gate */ 54417c478bd9Sstevel@tonic-gate ppa[0] = pp; 54427c478bd9Sstevel@tonic-gate nlocked = 1; 54437c478bd9Sstevel@tonic-gate 54447c478bd9Sstevel@tonic-gate /* 54457c478bd9Sstevel@tonic-gate * Lock constituent pages if this is large page 54467c478bd9Sstevel@tonic-gate */ 54477c478bd9Sstevel@tonic-gate if (pages > 1) { 54487c478bd9Sstevel@tonic-gate /* 54497c478bd9Sstevel@tonic-gate * Lock all constituents except root page, since it 54507c478bd9Sstevel@tonic-gate * should be locked already. 54517c478bd9Sstevel@tonic-gate */ 54527c478bd9Sstevel@tonic-gate for (i = 1; i < pages; i++) { 5453affbd3ccSkchow pp++; 54547c478bd9Sstevel@tonic-gate if (!page_trylock(pp, SE_EXCL)) { 54557c478bd9Sstevel@tonic-gate break; 54567c478bd9Sstevel@tonic-gate } 54577c478bd9Sstevel@tonic-gate if (PP_ISFREE(pp) || 54587c478bd9Sstevel@tonic-gate pp->p_szc != pszc) { 54597c478bd9Sstevel@tonic-gate /* 54607c478bd9Sstevel@tonic-gate * hat_page_demote() raced in with us. 54617c478bd9Sstevel@tonic-gate */ 54627c478bd9Sstevel@tonic-gate ASSERT(!IS_SWAPFSVP(curvp)); 54637c478bd9Sstevel@tonic-gate page_unlock(pp); 54647c478bd9Sstevel@tonic-gate break; 54657c478bd9Sstevel@tonic-gate } 54667c478bd9Sstevel@tonic-gate ppa[nlocked] = pp; 54677c478bd9Sstevel@tonic-gate nlocked++; 54687c478bd9Sstevel@tonic-gate } 54697c478bd9Sstevel@tonic-gate } 54707c478bd9Sstevel@tonic-gate 54717c478bd9Sstevel@tonic-gate /* 54727c478bd9Sstevel@tonic-gate * If all constituent pages couldn't be locked, 54737c478bd9Sstevel@tonic-gate * unlock pages locked so far and skip to next page. 54747c478bd9Sstevel@tonic-gate */ 54757c478bd9Sstevel@tonic-gate if (nlocked != pages) { 54767c478bd9Sstevel@tonic-gate for (i = 0; i < nlocked; i++) 54777c478bd9Sstevel@tonic-gate page_unlock(ppa[i]); 54787c478bd9Sstevel@tonic-gate va += pgsz; 54797c478bd9Sstevel@tonic-gate lgrp_stat_add(from->lgrp_id, LGRP_PMM_FAIL_PGS, 54807c478bd9Sstevel@tonic-gate btop(pgsz)); 54817c478bd9Sstevel@tonic-gate continue; 54827c478bd9Sstevel@tonic-gate } 54837c478bd9Sstevel@tonic-gate 54847c478bd9Sstevel@tonic-gate /* 54857c478bd9Sstevel@tonic-gate * hat_page_demote() can no longer happen 54867c478bd9Sstevel@tonic-gate * since last cons page had the right p_szc after 54877c478bd9Sstevel@tonic-gate * all cons pages were locked. all cons pages 54887c478bd9Sstevel@tonic-gate * should now have the same p_szc. 54897c478bd9Sstevel@tonic-gate */ 54907c478bd9Sstevel@tonic-gate 54917c478bd9Sstevel@tonic-gate /* 54927c478bd9Sstevel@tonic-gate * All constituent pages locked successfully, so mark 54937c478bd9Sstevel@tonic-gate * large page for migration and unload the mappings of 54947c478bd9Sstevel@tonic-gate * constituent pages, so a fault will occur on any part of the 54957c478bd9Sstevel@tonic-gate * large page 54967c478bd9Sstevel@tonic-gate */ 54977c478bd9Sstevel@tonic-gate PP_SETMIGRATE(ppa[0]); 54987c478bd9Sstevel@tonic-gate for (i = 0; i < nlocked; i++) { 54997c478bd9Sstevel@tonic-gate pp = ppa[i]; 55007c478bd9Sstevel@tonic-gate (void) hat_pageunload(pp, HAT_FORCE_PGUNLOAD); 55017c478bd9Sstevel@tonic-gate ASSERT(hat_page_getshare(pp) == 0); 55027c478bd9Sstevel@tonic-gate page_unlock(pp); 55037c478bd9Sstevel@tonic-gate } 55047c478bd9Sstevel@tonic-gate lgrp_stat_add(from->lgrp_id, LGRP_PMM_PGS, nlocked); 55057c478bd9Sstevel@tonic-gate 55067c478bd9Sstevel@tonic-gate va += pgsz; 55077c478bd9Sstevel@tonic-gate } 55087c478bd9Sstevel@tonic-gate kmem_free(ppa, ppa_nentries * sizeof (page_t *)); 55097c478bd9Sstevel@tonic-gate } 55107c478bd9Sstevel@tonic-gate 55117c478bd9Sstevel@tonic-gate /* 55127c478bd9Sstevel@tonic-gate * Migrate any pages that have been marked for migration in the given range 55137c478bd9Sstevel@tonic-gate */ 55147c478bd9Sstevel@tonic-gate void 55157c478bd9Sstevel@tonic-gate page_migrate( 55167c478bd9Sstevel@tonic-gate struct seg *seg, 55177c478bd9Sstevel@tonic-gate caddr_t addr, 55187c478bd9Sstevel@tonic-gate page_t **ppa, 55197c478bd9Sstevel@tonic-gate pgcnt_t npages) 55207c478bd9Sstevel@tonic-gate { 55217c478bd9Sstevel@tonic-gate lgrp_t *from; 55227c478bd9Sstevel@tonic-gate lgrp_t *to; 55237c478bd9Sstevel@tonic-gate page_t *newpp; 55247c478bd9Sstevel@tonic-gate page_t *pp; 55257c478bd9Sstevel@tonic-gate pfn_t pfn; 55267c478bd9Sstevel@tonic-gate size_t pgsz; 55277c478bd9Sstevel@tonic-gate spgcnt_t page_cnt; 55287c478bd9Sstevel@tonic-gate spgcnt_t i; 55297c478bd9Sstevel@tonic-gate uint_t pszc; 55307c478bd9Sstevel@tonic-gate 55317c478bd9Sstevel@tonic-gate ASSERT(seg->s_as && AS_LOCK_HELD(seg->s_as, &seg->s_as->a_lock)); 55327c478bd9Sstevel@tonic-gate 55337c478bd9Sstevel@tonic-gate while (npages > 0) { 55347c478bd9Sstevel@tonic-gate pp = *ppa; 55357c478bd9Sstevel@tonic-gate pszc = pp->p_szc; 55367c478bd9Sstevel@tonic-gate pgsz = page_get_pagesize(pszc); 55377c478bd9Sstevel@tonic-gate page_cnt = btop(pgsz); 55387c478bd9Sstevel@tonic-gate 55397c478bd9Sstevel@tonic-gate /* 55407c478bd9Sstevel@tonic-gate * Check to see whether this page is marked for migration 55417c478bd9Sstevel@tonic-gate * 55427c478bd9Sstevel@tonic-gate * Assume that root page of large page is marked for 55437c478bd9Sstevel@tonic-gate * migration and none of the other constituent pages 55447c478bd9Sstevel@tonic-gate * are marked. This really simplifies clearing the 55457c478bd9Sstevel@tonic-gate * migrate bit by not having to clear it from each 55467c478bd9Sstevel@tonic-gate * constituent page. 55477c478bd9Sstevel@tonic-gate * 55487c478bd9Sstevel@tonic-gate * note we don't want to relocate an entire large page if 55497c478bd9Sstevel@tonic-gate * someone is only using one subpage. 55507c478bd9Sstevel@tonic-gate */ 55517c478bd9Sstevel@tonic-gate if (npages < page_cnt) 55527c478bd9Sstevel@tonic-gate break; 55537c478bd9Sstevel@tonic-gate 55547c478bd9Sstevel@tonic-gate /* 55557c478bd9Sstevel@tonic-gate * Is it marked for migration? 55567c478bd9Sstevel@tonic-gate */ 55577c478bd9Sstevel@tonic-gate if (!PP_ISMIGRATE(pp)) 55587c478bd9Sstevel@tonic-gate goto next; 55597c478bd9Sstevel@tonic-gate 55607c478bd9Sstevel@tonic-gate /* 55617c478bd9Sstevel@tonic-gate * Determine lgroups that page is being migrated between 55627c478bd9Sstevel@tonic-gate */ 55637c478bd9Sstevel@tonic-gate pfn = page_pptonum(pp); 55647c478bd9Sstevel@tonic-gate if (!IS_P2ALIGNED(pfn, page_cnt)) { 55657c478bd9Sstevel@tonic-gate break; 55667c478bd9Sstevel@tonic-gate } 55677c478bd9Sstevel@tonic-gate from = lgrp_pfn_to_lgrp(pfn); 55687c478bd9Sstevel@tonic-gate to = lgrp_mem_choose(seg, addr, pgsz); 55697c478bd9Sstevel@tonic-gate 55707c478bd9Sstevel@tonic-gate /* 55717c478bd9Sstevel@tonic-gate * Need to get exclusive lock's to migrate 55727c478bd9Sstevel@tonic-gate */ 55737c478bd9Sstevel@tonic-gate for (i = 0; i < page_cnt; i++) { 55747c478bd9Sstevel@tonic-gate ASSERT(PAGE_LOCKED(ppa[i])); 55757c478bd9Sstevel@tonic-gate if (page_pptonum(ppa[i]) != pfn + i || 55767c478bd9Sstevel@tonic-gate ppa[i]->p_szc != pszc) { 55777c478bd9Sstevel@tonic-gate break; 55787c478bd9Sstevel@tonic-gate } 55797c478bd9Sstevel@tonic-gate if (!page_tryupgrade(ppa[i])) { 55807c478bd9Sstevel@tonic-gate lgrp_stat_add(from->lgrp_id, 55817c478bd9Sstevel@tonic-gate LGRP_PM_FAIL_LOCK_PGS, 55827c478bd9Sstevel@tonic-gate page_cnt); 55837c478bd9Sstevel@tonic-gate break; 55847c478bd9Sstevel@tonic-gate } 55856bc16138Sjj 55866bc16138Sjj /* 55876bc16138Sjj * Check to see whether we are trying to migrate 55886bc16138Sjj * page to lgroup where it is allocated already. 55896bc16138Sjj * If so, clear the migrate bit and skip to next 55906bc16138Sjj * page. 55916bc16138Sjj */ 55926bc16138Sjj if (i == 0 && to == from) { 55936bc16138Sjj PP_CLRMIGRATE(ppa[0]); 55946bc16138Sjj page_downgrade(ppa[0]); 55956bc16138Sjj goto next; 55966bc16138Sjj } 55977c478bd9Sstevel@tonic-gate } 55986bc16138Sjj 55996bc16138Sjj /* 56006bc16138Sjj * If all constituent pages couldn't be locked, 56016bc16138Sjj * unlock pages locked so far and skip to next page. 56026bc16138Sjj */ 56037c478bd9Sstevel@tonic-gate if (i != page_cnt) { 56047c478bd9Sstevel@tonic-gate while (--i != -1) { 56057c478bd9Sstevel@tonic-gate page_downgrade(ppa[i]); 56067c478bd9Sstevel@tonic-gate } 56077c478bd9Sstevel@tonic-gate goto next; 56087c478bd9Sstevel@tonic-gate } 56097c478bd9Sstevel@tonic-gate 56107c478bd9Sstevel@tonic-gate (void) page_create_wait(page_cnt, PG_WAIT); 56117c478bd9Sstevel@tonic-gate newpp = page_get_replacement_page(pp, to, PGR_SAMESZC); 56127c478bd9Sstevel@tonic-gate if (newpp == NULL) { 56137c478bd9Sstevel@tonic-gate page_create_putback(page_cnt); 56147c478bd9Sstevel@tonic-gate for (i = 0; i < page_cnt; i++) { 56157c478bd9Sstevel@tonic-gate page_downgrade(ppa[i]); 56167c478bd9Sstevel@tonic-gate } 56177c478bd9Sstevel@tonic-gate lgrp_stat_add(to->lgrp_id, LGRP_PM_FAIL_ALLOC_PGS, 56187c478bd9Sstevel@tonic-gate page_cnt); 56197c478bd9Sstevel@tonic-gate goto next; 56207c478bd9Sstevel@tonic-gate } 56217c478bd9Sstevel@tonic-gate ASSERT(newpp->p_szc == pszc); 56227c478bd9Sstevel@tonic-gate /* 56237c478bd9Sstevel@tonic-gate * Clear migrate bit and relocate page 56247c478bd9Sstevel@tonic-gate */ 56257c478bd9Sstevel@tonic-gate PP_CLRMIGRATE(pp); 56267c478bd9Sstevel@tonic-gate if (page_relocate(&pp, &newpp, 0, 1, &page_cnt, to)) { 56277c478bd9Sstevel@tonic-gate panic("page_migrate: page_relocate failed"); 56287c478bd9Sstevel@tonic-gate } 56297c478bd9Sstevel@tonic-gate ASSERT(page_cnt * PAGESIZE == pgsz); 56307c478bd9Sstevel@tonic-gate 56317c478bd9Sstevel@tonic-gate /* 56327c478bd9Sstevel@tonic-gate * Keep stats for number of pages migrated from and to 56337c478bd9Sstevel@tonic-gate * each lgroup 56347c478bd9Sstevel@tonic-gate */ 56357c478bd9Sstevel@tonic-gate lgrp_stat_add(from->lgrp_id, LGRP_PM_SRC_PGS, page_cnt); 56367c478bd9Sstevel@tonic-gate lgrp_stat_add(to->lgrp_id, LGRP_PM_DEST_PGS, page_cnt); 56377c478bd9Sstevel@tonic-gate /* 56387c478bd9Sstevel@tonic-gate * update the page_t array we were passed in and 56397c478bd9Sstevel@tonic-gate * unlink constituent pages of a large page. 56407c478bd9Sstevel@tonic-gate */ 56417c478bd9Sstevel@tonic-gate for (i = 0; i < page_cnt; ++i, ++pp) { 56427c478bd9Sstevel@tonic-gate ASSERT(PAGE_EXCL(newpp)); 56437c478bd9Sstevel@tonic-gate ASSERT(newpp->p_szc == pszc); 56447c478bd9Sstevel@tonic-gate ppa[i] = newpp; 56457c478bd9Sstevel@tonic-gate pp = newpp; 56467c478bd9Sstevel@tonic-gate page_sub(&newpp, pp); 56477c478bd9Sstevel@tonic-gate page_downgrade(pp); 56487c478bd9Sstevel@tonic-gate } 56497c478bd9Sstevel@tonic-gate ASSERT(newpp == NULL); 56507c478bd9Sstevel@tonic-gate next: 56517c478bd9Sstevel@tonic-gate addr += pgsz; 56527c478bd9Sstevel@tonic-gate ppa += page_cnt; 56537c478bd9Sstevel@tonic-gate npages -= page_cnt; 56547c478bd9Sstevel@tonic-gate } 56557c478bd9Sstevel@tonic-gate } 56567c478bd9Sstevel@tonic-gate 56577c478bd9Sstevel@tonic-gate ulong_t mem_waiters = 0; 56587c478bd9Sstevel@tonic-gate ulong_t max_count = 20; 56597c478bd9Sstevel@tonic-gate #define MAX_DELAY 0x1ff 56607c478bd9Sstevel@tonic-gate 56617c478bd9Sstevel@tonic-gate /* 56627c478bd9Sstevel@tonic-gate * Check if enough memory is available to proceed. 56637c478bd9Sstevel@tonic-gate * Depending on system configuration and how much memory is 56647c478bd9Sstevel@tonic-gate * reserved for swap we need to check against two variables. 56657c478bd9Sstevel@tonic-gate * e.g. on systems with little physical swap availrmem can be 56667c478bd9Sstevel@tonic-gate * more reliable indicator of how much memory is available. 56677c478bd9Sstevel@tonic-gate * On systems with large phys swap freemem can be better indicator. 56687c478bd9Sstevel@tonic-gate * If freemem drops below threshold level don't return an error 56697c478bd9Sstevel@tonic-gate * immediately but wake up pageout to free memory and block. 56707c478bd9Sstevel@tonic-gate * This is done number of times. If pageout is not able to free 56717c478bd9Sstevel@tonic-gate * memory within certain time return an error. 56727c478bd9Sstevel@tonic-gate * The same applies for availrmem but kmem_reap is used to 56737c478bd9Sstevel@tonic-gate * free memory. 56747c478bd9Sstevel@tonic-gate */ 56757c478bd9Sstevel@tonic-gate int 56767c478bd9Sstevel@tonic-gate page_mem_avail(pgcnt_t npages) 56777c478bd9Sstevel@tonic-gate { 56787c478bd9Sstevel@tonic-gate ulong_t count; 56797c478bd9Sstevel@tonic-gate 56807c478bd9Sstevel@tonic-gate #if defined(__i386) 56817c478bd9Sstevel@tonic-gate if (freemem > desfree + npages && 56827c478bd9Sstevel@tonic-gate availrmem > swapfs_reserve + npages && 56837c478bd9Sstevel@tonic-gate btop(vmem_size(heap_arena, VMEM_FREE)) > tune.t_minarmem + 56847c478bd9Sstevel@tonic-gate npages) 56857c478bd9Sstevel@tonic-gate return (1); 56867c478bd9Sstevel@tonic-gate #else 56877c478bd9Sstevel@tonic-gate if (freemem > desfree + npages && 56887c478bd9Sstevel@tonic-gate availrmem > swapfs_reserve + npages) 56897c478bd9Sstevel@tonic-gate return (1); 56907c478bd9Sstevel@tonic-gate #endif 56917c478bd9Sstevel@tonic-gate 56927c478bd9Sstevel@tonic-gate count = max_count; 56937c478bd9Sstevel@tonic-gate atomic_add_long(&mem_waiters, 1); 56947c478bd9Sstevel@tonic-gate 56957c478bd9Sstevel@tonic-gate while (freemem < desfree + npages && --count) { 56967c478bd9Sstevel@tonic-gate cv_signal(&proc_pageout->p_cv); 56977c478bd9Sstevel@tonic-gate if (delay_sig(hz + (mem_waiters & MAX_DELAY))) { 56987c478bd9Sstevel@tonic-gate atomic_add_long(&mem_waiters, -1); 56997c478bd9Sstevel@tonic-gate return (0); 57007c478bd9Sstevel@tonic-gate } 57017c478bd9Sstevel@tonic-gate } 57027c478bd9Sstevel@tonic-gate if (count == 0) { 57037c478bd9Sstevel@tonic-gate atomic_add_long(&mem_waiters, -1); 57047c478bd9Sstevel@tonic-gate return (0); 57057c478bd9Sstevel@tonic-gate } 57067c478bd9Sstevel@tonic-gate 57077c478bd9Sstevel@tonic-gate count = max_count; 57087c478bd9Sstevel@tonic-gate while (availrmem < swapfs_reserve + npages && --count) { 57097c478bd9Sstevel@tonic-gate kmem_reap(); 57107c478bd9Sstevel@tonic-gate if (delay_sig(hz + (mem_waiters & MAX_DELAY))) { 57117c478bd9Sstevel@tonic-gate atomic_add_long(&mem_waiters, -1); 57127c478bd9Sstevel@tonic-gate return (0); 57137c478bd9Sstevel@tonic-gate } 57147c478bd9Sstevel@tonic-gate } 57157c478bd9Sstevel@tonic-gate atomic_add_long(&mem_waiters, -1); 57167c478bd9Sstevel@tonic-gate if (count == 0) 57177c478bd9Sstevel@tonic-gate return (0); 57187c478bd9Sstevel@tonic-gate 57197c478bd9Sstevel@tonic-gate #if defined(__i386) 57207c478bd9Sstevel@tonic-gate if (btop(vmem_size(heap_arena, VMEM_FREE)) < 57217c478bd9Sstevel@tonic-gate tune.t_minarmem + npages) 57227c478bd9Sstevel@tonic-gate return (0); 57237c478bd9Sstevel@tonic-gate #endif 57247c478bd9Sstevel@tonic-gate return (1); 57257c478bd9Sstevel@tonic-gate } 57267c478bd9Sstevel@tonic-gate 57273cff2f43Sstans #define MAX_CNT 60 /* max num of iterations */ 57283cff2f43Sstans /* 57293cff2f43Sstans * Reclaim/reserve availrmem for npages. 57303cff2f43Sstans * If there is not enough memory start reaping seg, kmem caches. 57313cff2f43Sstans * Start pageout scanner (via page_needfree()). 57323cff2f43Sstans * Exit after ~ MAX_CNT s regardless of how much memory has been released. 57333cff2f43Sstans * Note: There is no guarantee that any availrmem will be freed as 57343cff2f43Sstans * this memory typically is locked (kernel heap) or reserved for swap. 57353cff2f43Sstans * Also due to memory fragmentation kmem allocator may not be able 57363cff2f43Sstans * to free any memory (single user allocated buffer will prevent 57373cff2f43Sstans * freeing slab or a page). 57383cff2f43Sstans */ 57393cff2f43Sstans int 57403cff2f43Sstans page_reclaim_mem(pgcnt_t npages, pgcnt_t epages, int adjust) 57413cff2f43Sstans { 57423cff2f43Sstans int i = 0; 57433cff2f43Sstans int ret = 0; 57443cff2f43Sstans pgcnt_t deficit; 57453cff2f43Sstans pgcnt_t old_availrmem; 57463cff2f43Sstans 57473cff2f43Sstans mutex_enter(&freemem_lock); 57483cff2f43Sstans old_availrmem = availrmem - 1; 57493cff2f43Sstans while ((availrmem < tune.t_minarmem + npages + epages) && 57503cff2f43Sstans (old_availrmem < availrmem) && (i++ < MAX_CNT)) { 57513cff2f43Sstans old_availrmem = availrmem; 57523cff2f43Sstans deficit = tune.t_minarmem + npages + epages - availrmem; 57533cff2f43Sstans mutex_exit(&freemem_lock); 57543cff2f43Sstans page_needfree(deficit); 57553cff2f43Sstans kmem_reap(); 57563cff2f43Sstans delay(hz); 57573cff2f43Sstans page_needfree(-(spgcnt_t)deficit); 57583cff2f43Sstans mutex_enter(&freemem_lock); 57593cff2f43Sstans } 57603cff2f43Sstans 57613cff2f43Sstans if (adjust && (availrmem >= tune.t_minarmem + npages + epages)) { 57623cff2f43Sstans availrmem -= npages; 57633cff2f43Sstans ret = 1; 57643cff2f43Sstans } 57653cff2f43Sstans 57663cff2f43Sstans mutex_exit(&freemem_lock); 57673cff2f43Sstans 57683cff2f43Sstans return (ret); 57693cff2f43Sstans } 57707c478bd9Sstevel@tonic-gate 57717c478bd9Sstevel@tonic-gate /* 57727c478bd9Sstevel@tonic-gate * Search the memory segments to locate the desired page. Within a 57737c478bd9Sstevel@tonic-gate * segment, pages increase linearly with one page structure per 57747c478bd9Sstevel@tonic-gate * physical page frame (size PAGESIZE). The search begins 57757c478bd9Sstevel@tonic-gate * with the segment that was accessed last, to take advantage of locality. 57767c478bd9Sstevel@tonic-gate * If the hint misses, we start from the beginning of the sorted memseg list 57777c478bd9Sstevel@tonic-gate */ 57787c478bd9Sstevel@tonic-gate 57797c478bd9Sstevel@tonic-gate 57807c478bd9Sstevel@tonic-gate /* 57817c478bd9Sstevel@tonic-gate * Some data structures for pfn to pp lookup. 57827c478bd9Sstevel@tonic-gate */ 57837c478bd9Sstevel@tonic-gate ulong_t mhash_per_slot; 57847c478bd9Sstevel@tonic-gate struct memseg *memseg_hash[N_MEM_SLOTS]; 57857c478bd9Sstevel@tonic-gate 57867c478bd9Sstevel@tonic-gate page_t * 57877c478bd9Sstevel@tonic-gate page_numtopp_nolock(pfn_t pfnum) 57887c478bd9Sstevel@tonic-gate { 57897c478bd9Sstevel@tonic-gate struct memseg *seg; 57907c478bd9Sstevel@tonic-gate page_t *pp; 5791affbd3ccSkchow vm_cpu_data_t *vc = CPU->cpu_vm_data; 57927c478bd9Sstevel@tonic-gate 5793affbd3ccSkchow ASSERT(vc != NULL); 57947c478bd9Sstevel@tonic-gate 57957c478bd9Sstevel@tonic-gate MEMSEG_STAT_INCR(nsearch); 57967c478bd9Sstevel@tonic-gate 57977c478bd9Sstevel@tonic-gate /* Try last winner first */ 5798affbd3ccSkchow if (((seg = vc->vc_pnum_memseg) != NULL) && 57996e4dd838Smec (pfnum >= seg->pages_base) && (pfnum < seg->pages_end)) { 58007c478bd9Sstevel@tonic-gate MEMSEG_STAT_INCR(nlastwon); 58017c478bd9Sstevel@tonic-gate pp = seg->pages + (pfnum - seg->pages_base); 58027c478bd9Sstevel@tonic-gate if (pp->p_pagenum == pfnum) 58037c478bd9Sstevel@tonic-gate return ((page_t *)pp); 58047c478bd9Sstevel@tonic-gate } 58057c478bd9Sstevel@tonic-gate 58067c478bd9Sstevel@tonic-gate /* Else Try hash */ 58077c478bd9Sstevel@tonic-gate if (((seg = memseg_hash[MEMSEG_PFN_HASH(pfnum)]) != NULL) && 58086e4dd838Smec (pfnum >= seg->pages_base) && (pfnum < seg->pages_end)) { 58097c478bd9Sstevel@tonic-gate MEMSEG_STAT_INCR(nhashwon); 5810affbd3ccSkchow vc->vc_pnum_memseg = seg; 58117c478bd9Sstevel@tonic-gate pp = seg->pages + (pfnum - seg->pages_base); 58127c478bd9Sstevel@tonic-gate if (pp->p_pagenum == pfnum) 58137c478bd9Sstevel@tonic-gate return ((page_t *)pp); 58147c478bd9Sstevel@tonic-gate } 58157c478bd9Sstevel@tonic-gate 58167c478bd9Sstevel@tonic-gate /* Else Brute force */ 58177c478bd9Sstevel@tonic-gate for (seg = memsegs; seg != NULL; seg = seg->next) { 58187c478bd9Sstevel@tonic-gate if (pfnum >= seg->pages_base && pfnum < seg->pages_end) { 5819affbd3ccSkchow vc->vc_pnum_memseg = seg; 58207c478bd9Sstevel@tonic-gate pp = seg->pages + (pfnum - seg->pages_base); 58217c478bd9Sstevel@tonic-gate return ((page_t *)pp); 58227c478bd9Sstevel@tonic-gate } 58237c478bd9Sstevel@tonic-gate } 5824affbd3ccSkchow vc->vc_pnum_memseg = NULL; 58257c478bd9Sstevel@tonic-gate MEMSEG_STAT_INCR(nnotfound); 58267c478bd9Sstevel@tonic-gate return ((page_t *)NULL); 58277c478bd9Sstevel@tonic-gate 58287c478bd9Sstevel@tonic-gate } 58297c478bd9Sstevel@tonic-gate 58307c478bd9Sstevel@tonic-gate struct memseg * 58317c478bd9Sstevel@tonic-gate page_numtomemseg_nolock(pfn_t pfnum) 58327c478bd9Sstevel@tonic-gate { 58337c478bd9Sstevel@tonic-gate struct memseg *seg; 58347c478bd9Sstevel@tonic-gate page_t *pp; 58357c478bd9Sstevel@tonic-gate 58367c478bd9Sstevel@tonic-gate /* Try hash */ 58377c478bd9Sstevel@tonic-gate if (((seg = memseg_hash[MEMSEG_PFN_HASH(pfnum)]) != NULL) && 58386e4dd838Smec (pfnum >= seg->pages_base) && (pfnum < seg->pages_end)) { 58397c478bd9Sstevel@tonic-gate pp = seg->pages + (pfnum - seg->pages_base); 58407c478bd9Sstevel@tonic-gate if (pp->p_pagenum == pfnum) 58417c478bd9Sstevel@tonic-gate return (seg); 58427c478bd9Sstevel@tonic-gate } 58437c478bd9Sstevel@tonic-gate 58447c478bd9Sstevel@tonic-gate /* Else Brute force */ 58457c478bd9Sstevel@tonic-gate for (seg = memsegs; seg != NULL; seg = seg->next) { 58467c478bd9Sstevel@tonic-gate if (pfnum >= seg->pages_base && pfnum < seg->pages_end) { 58477c478bd9Sstevel@tonic-gate return (seg); 58487c478bd9Sstevel@tonic-gate } 58497c478bd9Sstevel@tonic-gate } 58507c478bd9Sstevel@tonic-gate return ((struct memseg *)NULL); 58517c478bd9Sstevel@tonic-gate } 58527c478bd9Sstevel@tonic-gate 58537c478bd9Sstevel@tonic-gate /* 58547c478bd9Sstevel@tonic-gate * Given a page and a count return the page struct that is 58557c478bd9Sstevel@tonic-gate * n structs away from the current one in the global page 58567c478bd9Sstevel@tonic-gate * list. 58577c478bd9Sstevel@tonic-gate * 58587c478bd9Sstevel@tonic-gate * This function wraps to the first page upon 58597c478bd9Sstevel@tonic-gate * reaching the end of the memseg list. 58607c478bd9Sstevel@tonic-gate */ 58617c478bd9Sstevel@tonic-gate page_t * 58627c478bd9Sstevel@tonic-gate page_nextn(page_t *pp, ulong_t n) 58637c478bd9Sstevel@tonic-gate { 58647c478bd9Sstevel@tonic-gate struct memseg *seg; 58657c478bd9Sstevel@tonic-gate page_t *ppn; 5866affbd3ccSkchow vm_cpu_data_t *vc = (vm_cpu_data_t *)CPU->cpu_vm_data; 58677c478bd9Sstevel@tonic-gate 5868affbd3ccSkchow ASSERT(vc != NULL); 5869affbd3ccSkchow 5870affbd3ccSkchow if (((seg = vc->vc_pnext_memseg) == NULL) || 58717c478bd9Sstevel@tonic-gate (seg->pages_base == seg->pages_end) || 58727c478bd9Sstevel@tonic-gate !(pp >= seg->pages && pp < seg->epages)) { 58737c478bd9Sstevel@tonic-gate 58747c478bd9Sstevel@tonic-gate for (seg = memsegs; seg; seg = seg->next) { 58757c478bd9Sstevel@tonic-gate if (pp >= seg->pages && pp < seg->epages) 58767c478bd9Sstevel@tonic-gate break; 58777c478bd9Sstevel@tonic-gate } 58787c478bd9Sstevel@tonic-gate 58797c478bd9Sstevel@tonic-gate if (seg == NULL) { 58807c478bd9Sstevel@tonic-gate /* Memory delete got in, return something valid. */ 58817c478bd9Sstevel@tonic-gate /* TODO: fix me. */ 58827c478bd9Sstevel@tonic-gate seg = memsegs; 58837c478bd9Sstevel@tonic-gate pp = seg->pages; 58847c478bd9Sstevel@tonic-gate } 58857c478bd9Sstevel@tonic-gate } 58867c478bd9Sstevel@tonic-gate 58877c478bd9Sstevel@tonic-gate /* check for wraparound - possible if n is large */ 58887c478bd9Sstevel@tonic-gate while ((ppn = (pp + n)) >= seg->epages || ppn < pp) { 58897c478bd9Sstevel@tonic-gate n -= seg->epages - pp; 58907c478bd9Sstevel@tonic-gate seg = seg->next; 58917c478bd9Sstevel@tonic-gate if (seg == NULL) 58927c478bd9Sstevel@tonic-gate seg = memsegs; 58937c478bd9Sstevel@tonic-gate pp = seg->pages; 58947c478bd9Sstevel@tonic-gate } 5895affbd3ccSkchow vc->vc_pnext_memseg = seg; 58967c478bd9Sstevel@tonic-gate return (ppn); 58977c478bd9Sstevel@tonic-gate } 58987c478bd9Sstevel@tonic-gate 58997c478bd9Sstevel@tonic-gate /* 59007c478bd9Sstevel@tonic-gate * Initialize for a loop using page_next_scan_large(). 59017c478bd9Sstevel@tonic-gate */ 59027c478bd9Sstevel@tonic-gate page_t * 59037c478bd9Sstevel@tonic-gate page_next_scan_init(void **cookie) 59047c478bd9Sstevel@tonic-gate { 59057c478bd9Sstevel@tonic-gate ASSERT(cookie != NULL); 59067c478bd9Sstevel@tonic-gate *cookie = (void *)memsegs; 59077c478bd9Sstevel@tonic-gate return ((page_t *)memsegs->pages); 59087c478bd9Sstevel@tonic-gate } 59097c478bd9Sstevel@tonic-gate 59107c478bd9Sstevel@tonic-gate /* 59117c478bd9Sstevel@tonic-gate * Return the next page in a scan of page_t's, assuming we want 59127c478bd9Sstevel@tonic-gate * to skip over sub-pages within larger page sizes. 59137c478bd9Sstevel@tonic-gate * 59147c478bd9Sstevel@tonic-gate * The cookie is used to keep track of the current memseg. 59157c478bd9Sstevel@tonic-gate */ 59167c478bd9Sstevel@tonic-gate page_t * 59177c478bd9Sstevel@tonic-gate page_next_scan_large( 59187c478bd9Sstevel@tonic-gate page_t *pp, 59197c478bd9Sstevel@tonic-gate ulong_t *n, 59207c478bd9Sstevel@tonic-gate void **cookie) 59217c478bd9Sstevel@tonic-gate { 59227c478bd9Sstevel@tonic-gate struct memseg *seg = (struct memseg *)*cookie; 59237c478bd9Sstevel@tonic-gate page_t *new_pp; 59247c478bd9Sstevel@tonic-gate ulong_t cnt; 59257c478bd9Sstevel@tonic-gate pfn_t pfn; 59267c478bd9Sstevel@tonic-gate 59277c478bd9Sstevel@tonic-gate 59287c478bd9Sstevel@tonic-gate /* 59297c478bd9Sstevel@tonic-gate * get the count of page_t's to skip based on the page size 59307c478bd9Sstevel@tonic-gate */ 59317c478bd9Sstevel@tonic-gate ASSERT(pp != NULL); 59327c478bd9Sstevel@tonic-gate if (pp->p_szc == 0) { 59337c478bd9Sstevel@tonic-gate cnt = 1; 59347c478bd9Sstevel@tonic-gate } else { 59357c478bd9Sstevel@tonic-gate pfn = page_pptonum(pp); 59367c478bd9Sstevel@tonic-gate cnt = page_get_pagecnt(pp->p_szc); 59377c478bd9Sstevel@tonic-gate cnt -= pfn & (cnt - 1); 59387c478bd9Sstevel@tonic-gate } 59397c478bd9Sstevel@tonic-gate *n += cnt; 59407c478bd9Sstevel@tonic-gate new_pp = pp + cnt; 59417c478bd9Sstevel@tonic-gate 59427c478bd9Sstevel@tonic-gate /* 59437c478bd9Sstevel@tonic-gate * Catch if we went past the end of the current memory segment. If so, 59447c478bd9Sstevel@tonic-gate * just move to the next segment with pages. 59457c478bd9Sstevel@tonic-gate */ 59467c478bd9Sstevel@tonic-gate if (new_pp >= seg->epages) { 59477c478bd9Sstevel@tonic-gate do { 59487c478bd9Sstevel@tonic-gate seg = seg->next; 59497c478bd9Sstevel@tonic-gate if (seg == NULL) 59507c478bd9Sstevel@tonic-gate seg = memsegs; 59517c478bd9Sstevel@tonic-gate } while (seg->pages == seg->epages); 59527c478bd9Sstevel@tonic-gate new_pp = seg->pages; 59537c478bd9Sstevel@tonic-gate *cookie = (void *)seg; 59547c478bd9Sstevel@tonic-gate } 59557c478bd9Sstevel@tonic-gate 59567c478bd9Sstevel@tonic-gate return (new_pp); 59577c478bd9Sstevel@tonic-gate } 59587c478bd9Sstevel@tonic-gate 59597c478bd9Sstevel@tonic-gate 59607c478bd9Sstevel@tonic-gate /* 59617c478bd9Sstevel@tonic-gate * Returns next page in list. Note: this function wraps 59627c478bd9Sstevel@tonic-gate * to the first page in the list upon reaching the end 59637c478bd9Sstevel@tonic-gate * of the list. Callers should be aware of this fact. 59647c478bd9Sstevel@tonic-gate */ 59657c478bd9Sstevel@tonic-gate 59667c478bd9Sstevel@tonic-gate /* We should change this be a #define */ 59677c478bd9Sstevel@tonic-gate 59687c478bd9Sstevel@tonic-gate page_t * 59697c478bd9Sstevel@tonic-gate page_next(page_t *pp) 59707c478bd9Sstevel@tonic-gate { 59717c478bd9Sstevel@tonic-gate return (page_nextn(pp, 1)); 59727c478bd9Sstevel@tonic-gate } 59737c478bd9Sstevel@tonic-gate 59747c478bd9Sstevel@tonic-gate page_t * 59757c478bd9Sstevel@tonic-gate page_first() 59767c478bd9Sstevel@tonic-gate { 59777c478bd9Sstevel@tonic-gate return ((page_t *)memsegs->pages); 59787c478bd9Sstevel@tonic-gate } 59797c478bd9Sstevel@tonic-gate 59807c478bd9Sstevel@tonic-gate 59817c478bd9Sstevel@tonic-gate /* 59827c478bd9Sstevel@tonic-gate * This routine is called at boot with the initial memory configuration 59837c478bd9Sstevel@tonic-gate * and when memory is added or removed. 59847c478bd9Sstevel@tonic-gate */ 59857c478bd9Sstevel@tonic-gate void 59867c478bd9Sstevel@tonic-gate build_pfn_hash() 59877c478bd9Sstevel@tonic-gate { 59887c478bd9Sstevel@tonic-gate pfn_t cur; 59897c478bd9Sstevel@tonic-gate pgcnt_t index; 59907c478bd9Sstevel@tonic-gate struct memseg *pseg; 59917c478bd9Sstevel@tonic-gate int i; 59927c478bd9Sstevel@tonic-gate 59937c478bd9Sstevel@tonic-gate /* 59947c478bd9Sstevel@tonic-gate * Clear memseg_hash array. 59957c478bd9Sstevel@tonic-gate * Since memory add/delete is designed to operate concurrently 59967c478bd9Sstevel@tonic-gate * with normal operation, the hash rebuild must be able to run 59977c478bd9Sstevel@tonic-gate * concurrently with page_numtopp_nolock(). To support this 59987c478bd9Sstevel@tonic-gate * functionality, assignments to memseg_hash array members must 59997c478bd9Sstevel@tonic-gate * be done atomically. 60007c478bd9Sstevel@tonic-gate * 60017c478bd9Sstevel@tonic-gate * NOTE: bzero() does not currently guarantee this for kernel 60027c478bd9Sstevel@tonic-gate * threads, and cannot be used here. 60037c478bd9Sstevel@tonic-gate */ 60047c478bd9Sstevel@tonic-gate for (i = 0; i < N_MEM_SLOTS; i++) 60057c478bd9Sstevel@tonic-gate memseg_hash[i] = NULL; 60067c478bd9Sstevel@tonic-gate 60077c478bd9Sstevel@tonic-gate hat_kpm_mseghash_clear(N_MEM_SLOTS); 60087c478bd9Sstevel@tonic-gate 60097c478bd9Sstevel@tonic-gate /* 60107c478bd9Sstevel@tonic-gate * Physmax is the last valid pfn. 60117c478bd9Sstevel@tonic-gate */ 60127c478bd9Sstevel@tonic-gate mhash_per_slot = (physmax + 1) >> MEM_HASH_SHIFT; 60137c478bd9Sstevel@tonic-gate for (pseg = memsegs; pseg != NULL; pseg = pseg->next) { 60147c478bd9Sstevel@tonic-gate index = MEMSEG_PFN_HASH(pseg->pages_base); 60157c478bd9Sstevel@tonic-gate cur = pseg->pages_base; 60167c478bd9Sstevel@tonic-gate do { 60177c478bd9Sstevel@tonic-gate if (index >= N_MEM_SLOTS) 60187c478bd9Sstevel@tonic-gate index = MEMSEG_PFN_HASH(cur); 60197c478bd9Sstevel@tonic-gate 60207c478bd9Sstevel@tonic-gate if (memseg_hash[index] == NULL || 60217c478bd9Sstevel@tonic-gate memseg_hash[index]->pages_base > pseg->pages_base) { 60227c478bd9Sstevel@tonic-gate memseg_hash[index] = pseg; 60237c478bd9Sstevel@tonic-gate hat_kpm_mseghash_update(index, pseg); 60247c478bd9Sstevel@tonic-gate } 60257c478bd9Sstevel@tonic-gate cur += mhash_per_slot; 60267c478bd9Sstevel@tonic-gate index++; 60277c478bd9Sstevel@tonic-gate } while (cur < pseg->pages_end); 60287c478bd9Sstevel@tonic-gate } 60297c478bd9Sstevel@tonic-gate } 60307c478bd9Sstevel@tonic-gate 60317c478bd9Sstevel@tonic-gate /* 60327c478bd9Sstevel@tonic-gate * Return the pagenum for the pp 60337c478bd9Sstevel@tonic-gate */ 60347c478bd9Sstevel@tonic-gate pfn_t 60357c478bd9Sstevel@tonic-gate page_pptonum(page_t *pp) 60367c478bd9Sstevel@tonic-gate { 60377c478bd9Sstevel@tonic-gate return (pp->p_pagenum); 60387c478bd9Sstevel@tonic-gate } 60397c478bd9Sstevel@tonic-gate 60407c478bd9Sstevel@tonic-gate /* 60417c478bd9Sstevel@tonic-gate * interface to the referenced and modified etc bits 60427c478bd9Sstevel@tonic-gate * in the PSM part of the page struct 60437c478bd9Sstevel@tonic-gate * when no locking is desired. 60447c478bd9Sstevel@tonic-gate */ 60457c478bd9Sstevel@tonic-gate void 60467c478bd9Sstevel@tonic-gate page_set_props(page_t *pp, uint_t flags) 60477c478bd9Sstevel@tonic-gate { 60487c478bd9Sstevel@tonic-gate ASSERT((flags & ~(P_MOD | P_REF | P_RO)) == 0); 60497c478bd9Sstevel@tonic-gate pp->p_nrm |= (uchar_t)flags; 60507c478bd9Sstevel@tonic-gate } 60517c478bd9Sstevel@tonic-gate 60527c478bd9Sstevel@tonic-gate void 60537c478bd9Sstevel@tonic-gate page_clr_all_props(page_t *pp) 60547c478bd9Sstevel@tonic-gate { 60557c478bd9Sstevel@tonic-gate pp->p_nrm = 0; 60567c478bd9Sstevel@tonic-gate } 60577c478bd9Sstevel@tonic-gate 60587c478bd9Sstevel@tonic-gate /* 6059db874c57Selowe * Clear p_lckcnt and p_cowcnt, adjusting freemem if required. 60607c478bd9Sstevel@tonic-gate */ 60617c478bd9Sstevel@tonic-gate int 6062db874c57Selowe page_clear_lck_cow(page_t *pp, int adjust) 60637c478bd9Sstevel@tonic-gate { 6064db874c57Selowe int f_amount; 60657c478bd9Sstevel@tonic-gate 6066db874c57Selowe ASSERT(PAGE_EXCL(pp)); 60677c478bd9Sstevel@tonic-gate 60687c478bd9Sstevel@tonic-gate /* 6069db874c57Selowe * The page_struct_lock need not be acquired here since 6070db874c57Selowe * we require the caller hold the page exclusively locked. 60717c478bd9Sstevel@tonic-gate */ 6072db874c57Selowe f_amount = 0; 6073db874c57Selowe if (pp->p_lckcnt) { 6074db874c57Selowe f_amount = 1; 6075db874c57Selowe pp->p_lckcnt = 0; 6076db874c57Selowe } 6077db874c57Selowe if (pp->p_cowcnt) { 6078db874c57Selowe f_amount += pp->p_cowcnt; 6079db874c57Selowe pp->p_cowcnt = 0; 6080db874c57Selowe } 60817c478bd9Sstevel@tonic-gate 6082db874c57Selowe if (adjust && f_amount) { 6083db874c57Selowe mutex_enter(&freemem_lock); 6084db874c57Selowe availrmem += f_amount; 6085db874c57Selowe mutex_exit(&freemem_lock); 60867c478bd9Sstevel@tonic-gate } 60877c478bd9Sstevel@tonic-gate 6088db874c57Selowe return (f_amount); 60897c478bd9Sstevel@tonic-gate } 60907c478bd9Sstevel@tonic-gate 60917c478bd9Sstevel@tonic-gate /* 6092db874c57Selowe * The following functions is called from free_vp_pages() 6093db874c57Selowe * for an inexact estimate of a newly free'd page... 60947c478bd9Sstevel@tonic-gate */ 6095db874c57Selowe ulong_t 6096db874c57Selowe page_share_cnt(page_t *pp) 6097db874c57Selowe { 6098db874c57Selowe return (hat_page_getshare(pp)); 6099db874c57Selowe } 6100db874c57Selowe 61017c478bd9Sstevel@tonic-gate int 61027c478bd9Sstevel@tonic-gate page_isshared(page_t *pp) 61037c478bd9Sstevel@tonic-gate { 610405d3dc4bSpaulsan return (hat_page_checkshare(pp, 1)); 61057c478bd9Sstevel@tonic-gate } 61067c478bd9Sstevel@tonic-gate 61077c478bd9Sstevel@tonic-gate int 61087c478bd9Sstevel@tonic-gate page_isfree(page_t *pp) 61097c478bd9Sstevel@tonic-gate { 61107c478bd9Sstevel@tonic-gate return (PP_ISFREE(pp)); 61117c478bd9Sstevel@tonic-gate } 61127c478bd9Sstevel@tonic-gate 61137c478bd9Sstevel@tonic-gate int 61147c478bd9Sstevel@tonic-gate page_isref(page_t *pp) 61157c478bd9Sstevel@tonic-gate { 61167c478bd9Sstevel@tonic-gate return (hat_page_getattr(pp, P_REF)); 61177c478bd9Sstevel@tonic-gate } 61187c478bd9Sstevel@tonic-gate 61197c478bd9Sstevel@tonic-gate int 61207c478bd9Sstevel@tonic-gate page_ismod(page_t *pp) 61217c478bd9Sstevel@tonic-gate { 61227c478bd9Sstevel@tonic-gate return (hat_page_getattr(pp, P_MOD)); 61237c478bd9Sstevel@tonic-gate } 61248b464eb8Smec 61258b464eb8Smec /* 61268b464eb8Smec * The following code all currently relates to the page capture logic: 61278b464eb8Smec * 61288b464eb8Smec * This logic is used for cases where there is a desire to claim a certain 61298b464eb8Smec * physical page in the system for the caller. As it may not be possible 61308b464eb8Smec * to capture the page immediately, the p_toxic bits are used in the page 61318b464eb8Smec * structure to indicate that someone wants to capture this page. When the 61328b464eb8Smec * page gets unlocked, the toxic flag will be noted and an attempt to capture 61338b464eb8Smec * the page will be made. If it is successful, the original callers callback 61348b464eb8Smec * will be called with the page to do with it what they please. 61358b464eb8Smec * 61368b464eb8Smec * There is also an async thread which wakes up to attempt to capture 61378b464eb8Smec * pages occasionally which have the capture bit set. All of the pages which 61388b464eb8Smec * need to be captured asynchronously have been inserted into the 61398b464eb8Smec * page_capture_hash and thus this thread walks that hash list. Items in the 61408b464eb8Smec * hash have an expiration time so this thread handles that as well by removing 61418b464eb8Smec * the item from the hash if it has expired. 61428b464eb8Smec * 61438b464eb8Smec * Some important things to note are: 61448b464eb8Smec * - if the PR_CAPTURE bit is set on a page, then the page is in the 61458b464eb8Smec * page_capture_hash. The page_capture_hash_head.pchh_mutex is needed 61468b464eb8Smec * to set and clear this bit, and while the lock is held is the only time 61478b464eb8Smec * you can add or remove an entry from the hash. 61488b464eb8Smec * - the PR_CAPTURE bit can only be set and cleared while holding the 61498b464eb8Smec * page_capture_hash_head.pchh_mutex 61508b464eb8Smec * - the t_flag field of the thread struct is used with the T_CAPTURING 61518b464eb8Smec * flag to prevent recursion while dealing with large pages. 61528b464eb8Smec * - pages which need to be retired never expire on the page_capture_hash. 61538b464eb8Smec */ 61548b464eb8Smec 61558b464eb8Smec static void page_capture_thread(void); 61568b464eb8Smec static kthread_t *pc_thread_id; 61578b464eb8Smec kcondvar_t pc_cv; 61588b464eb8Smec static kmutex_t pc_thread_mutex; 61598b464eb8Smec static clock_t pc_thread_shortwait; 61608b464eb8Smec static clock_t pc_thread_longwait; 6161a98e9dbfSaguzovsk static int pc_thread_retry; 61628b464eb8Smec 61638b464eb8Smec struct page_capture_callback pc_cb[PC_NUM_CALLBACKS]; 61648b464eb8Smec 61658b464eb8Smec /* Note that this is a circular linked list */ 61668b464eb8Smec typedef struct page_capture_hash_bucket { 61678b464eb8Smec page_t *pp; 61688b464eb8Smec uint_t szc; 61698b464eb8Smec uint_t flags; 61708b464eb8Smec clock_t expires; /* lbolt at which this request expires. */ 61718b464eb8Smec void *datap; /* Cached data passed in for callback */ 61728b464eb8Smec struct page_capture_hash_bucket *next; 61738b464eb8Smec struct page_capture_hash_bucket *prev; 61748b464eb8Smec } page_capture_hash_bucket_t; 61758b464eb8Smec 61768b464eb8Smec /* 61778b464eb8Smec * Each hash bucket will have it's own mutex and two lists which are: 61788b464eb8Smec * active (0): represents requests which have not been processed by 61798b464eb8Smec * the page_capture async thread yet. 61808b464eb8Smec * walked (1): represents requests which have been processed by the 61818b464eb8Smec * page_capture async thread within it's given walk of this bucket. 61828b464eb8Smec * 61838b464eb8Smec * These are all needed so that we can synchronize all async page_capture 61848b464eb8Smec * events. When the async thread moves to a new bucket, it will append the 61858b464eb8Smec * walked list to the active list and walk each item one at a time, moving it 61868b464eb8Smec * from the active list to the walked list. Thus if there is an async request 61878b464eb8Smec * outstanding for a given page, it will always be in one of the two lists. 61888b464eb8Smec * New requests will always be added to the active list. 61898b464eb8Smec * If we were not able to capture a page before the request expired, we'd free 61908b464eb8Smec * up the request structure which would indicate to page_capture that there is 61918b464eb8Smec * no longer a need for the given page, and clear the PR_CAPTURE flag if 61928b464eb8Smec * possible. 61938b464eb8Smec */ 61948b464eb8Smec typedef struct page_capture_hash_head { 61958b464eb8Smec kmutex_t pchh_mutex; 61968b464eb8Smec uint_t num_pages; 61978b464eb8Smec page_capture_hash_bucket_t lists[2]; /* sentinel nodes */ 61988b464eb8Smec } page_capture_hash_head_t; 61998b464eb8Smec 62008b464eb8Smec #ifdef DEBUG 62018b464eb8Smec #define NUM_PAGE_CAPTURE_BUCKETS 4 62028b464eb8Smec #else 62038b464eb8Smec #define NUM_PAGE_CAPTURE_BUCKETS 64 62048b464eb8Smec #endif 62058b464eb8Smec 62068b464eb8Smec page_capture_hash_head_t page_capture_hash[NUM_PAGE_CAPTURE_BUCKETS]; 62078b464eb8Smec 62088b464eb8Smec /* for now use a very simple hash based upon the size of a page struct */ 62098b464eb8Smec #define PAGE_CAPTURE_HASH(pp) \ 62108b464eb8Smec ((int)(((uintptr_t)pp >> 7) & (NUM_PAGE_CAPTURE_BUCKETS - 1))) 62118b464eb8Smec 62128b464eb8Smec extern pgcnt_t swapfs_minfree; 62138b464eb8Smec 62148b464eb8Smec int page_trycapture(page_t *pp, uint_t szc, uint_t flags, void *datap); 62158b464eb8Smec 62168b464eb8Smec /* 62178b464eb8Smec * a callback function is required for page capture requests. 62188b464eb8Smec */ 62198b464eb8Smec void 62208b464eb8Smec page_capture_register_callback(uint_t index, clock_t duration, 62218b464eb8Smec int (*cb_func)(page_t *, void *, uint_t)) 62228b464eb8Smec { 62238b464eb8Smec ASSERT(pc_cb[index].cb_active == 0); 62248b464eb8Smec ASSERT(cb_func != NULL); 62258b464eb8Smec rw_enter(&pc_cb[index].cb_rwlock, RW_WRITER); 62268b464eb8Smec pc_cb[index].duration = duration; 62278b464eb8Smec pc_cb[index].cb_func = cb_func; 62288b464eb8Smec pc_cb[index].cb_active = 1; 62298b464eb8Smec rw_exit(&pc_cb[index].cb_rwlock); 62308b464eb8Smec } 62318b464eb8Smec 62328b464eb8Smec void 62338b464eb8Smec page_capture_unregister_callback(uint_t index) 62348b464eb8Smec { 62358b464eb8Smec int i, j; 62368b464eb8Smec struct page_capture_hash_bucket *bp1; 62378b464eb8Smec struct page_capture_hash_bucket *bp2; 62388b464eb8Smec struct page_capture_hash_bucket *head = NULL; 62398b464eb8Smec uint_t flags = (1 << index); 62408b464eb8Smec 62418b464eb8Smec rw_enter(&pc_cb[index].cb_rwlock, RW_WRITER); 62428b464eb8Smec ASSERT(pc_cb[index].cb_active == 1); 62438b464eb8Smec pc_cb[index].duration = 0; /* Paranoia */ 62448b464eb8Smec pc_cb[index].cb_func = NULL; /* Paranoia */ 62458b464eb8Smec pc_cb[index].cb_active = 0; 62468b464eb8Smec rw_exit(&pc_cb[index].cb_rwlock); 62478b464eb8Smec 62488b464eb8Smec /* 62498b464eb8Smec * Just move all the entries to a private list which we can walk 62508b464eb8Smec * through without the need to hold any locks. 62518b464eb8Smec * No more requests can get added to the hash lists for this consumer 62528b464eb8Smec * as the cb_active field for the callback has been cleared. 62538b464eb8Smec */ 62548b464eb8Smec for (i = 0; i < NUM_PAGE_CAPTURE_BUCKETS; i++) { 62558b464eb8Smec mutex_enter(&page_capture_hash[i].pchh_mutex); 62568b464eb8Smec for (j = 0; j < 2; j++) { 62578b464eb8Smec bp1 = page_capture_hash[i].lists[j].next; 62588b464eb8Smec /* walk through all but first (sentinel) element */ 62598b464eb8Smec while (bp1 != &page_capture_hash[i].lists[j]) { 62608b464eb8Smec bp2 = bp1; 62618b464eb8Smec if (bp2->flags & flags) { 62628b464eb8Smec bp1 = bp2->next; 62638b464eb8Smec bp1->prev = bp2->prev; 62648b464eb8Smec bp2->prev->next = bp1; 62658b464eb8Smec bp2->next = head; 62668b464eb8Smec head = bp2; 62678b464eb8Smec /* 62688b464eb8Smec * Clear the PR_CAPTURE bit as we 62698b464eb8Smec * hold appropriate locks here. 62708b464eb8Smec */ 62718b464eb8Smec page_clrtoxic(head->pp, PR_CAPTURE); 62728b464eb8Smec page_capture_hash[i].num_pages--; 62738b464eb8Smec continue; 62748b464eb8Smec } 62758b464eb8Smec bp1 = bp1->next; 62768b464eb8Smec } 62778b464eb8Smec } 62788b464eb8Smec mutex_exit(&page_capture_hash[i].pchh_mutex); 62798b464eb8Smec } 62808b464eb8Smec 62818b464eb8Smec while (head != NULL) { 62828b464eb8Smec bp1 = head; 62838b464eb8Smec head = head->next; 62848b464eb8Smec kmem_free(bp1, sizeof (*bp1)); 62858b464eb8Smec } 62868b464eb8Smec } 62878b464eb8Smec 62888b464eb8Smec 62898b464eb8Smec /* 62908b464eb8Smec * Find pp in the active list and move it to the walked list if it 62918b464eb8Smec * exists. 62928b464eb8Smec * Note that most often pp should be at the front of the active list 62938b464eb8Smec * as it is currently used and thus there is no other sort of optimization 62948b464eb8Smec * being done here as this is a linked list data structure. 62958b464eb8Smec * Returns 1 on successful move or 0 if page could not be found. 62968b464eb8Smec */ 62978b464eb8Smec static int 62988b464eb8Smec page_capture_move_to_walked(page_t *pp) 62998b464eb8Smec { 63008b464eb8Smec page_capture_hash_bucket_t *bp; 63018b464eb8Smec int index; 63028b464eb8Smec 63038b464eb8Smec index = PAGE_CAPTURE_HASH(pp); 63048b464eb8Smec 63058b464eb8Smec mutex_enter(&page_capture_hash[index].pchh_mutex); 63068b464eb8Smec bp = page_capture_hash[index].lists[0].next; 63078b464eb8Smec while (bp != &page_capture_hash[index].lists[0]) { 63088b464eb8Smec if (bp->pp == pp) { 63098b464eb8Smec /* Remove from old list */ 63108b464eb8Smec bp->next->prev = bp->prev; 63118b464eb8Smec bp->prev->next = bp->next; 63128b464eb8Smec 63138b464eb8Smec /* Add to new list */ 63148b464eb8Smec bp->next = page_capture_hash[index].lists[1].next; 63158b464eb8Smec bp->prev = &page_capture_hash[index].lists[1]; 63168b464eb8Smec page_capture_hash[index].lists[1].next = bp; 63178b464eb8Smec bp->next->prev = bp; 63188b464eb8Smec mutex_exit(&page_capture_hash[index].pchh_mutex); 63198b464eb8Smec 63208b464eb8Smec return (1); 63218b464eb8Smec } 63228b464eb8Smec bp = bp->next; 63238b464eb8Smec } 63248b464eb8Smec mutex_exit(&page_capture_hash[index].pchh_mutex); 63258b464eb8Smec return (0); 63268b464eb8Smec } 63278b464eb8Smec 63288b464eb8Smec /* 63298b464eb8Smec * Add a new entry to the page capture hash. The only case where a new 63308b464eb8Smec * entry is not added is when the page capture consumer is no longer registered. 63318b464eb8Smec * In this case, we'll silently not add the page to the hash. We know that 63328b464eb8Smec * page retire will always be registered for the case where we are currently 63338b464eb8Smec * unretiring a page and thus there are no conflicts. 63348b464eb8Smec */ 63358b464eb8Smec static void 63368b464eb8Smec page_capture_add_hash(page_t *pp, uint_t szc, uint_t flags, void *datap) 63378b464eb8Smec { 63388b464eb8Smec page_capture_hash_bucket_t *bp1; 63398b464eb8Smec page_capture_hash_bucket_t *bp2; 63408b464eb8Smec int index; 63418b464eb8Smec int cb_index; 63428b464eb8Smec int i; 63438b464eb8Smec #ifdef DEBUG 63448b464eb8Smec page_capture_hash_bucket_t *tp1; 63458b464eb8Smec int l; 63468b464eb8Smec #endif 63478b464eb8Smec 63488b464eb8Smec ASSERT(!(flags & CAPTURE_ASYNC)); 63498b464eb8Smec 63508b464eb8Smec bp1 = kmem_alloc(sizeof (struct page_capture_hash_bucket), KM_SLEEP); 63518b464eb8Smec 63528b464eb8Smec bp1->pp = pp; 63538b464eb8Smec bp1->szc = szc; 63548b464eb8Smec bp1->flags = flags; 63558b464eb8Smec bp1->datap = datap; 63568b464eb8Smec 63578b464eb8Smec for (cb_index = 0; cb_index < PC_NUM_CALLBACKS; cb_index++) { 63588b464eb8Smec if ((flags >> cb_index) & 1) { 63598b464eb8Smec break; 63608b464eb8Smec } 63618b464eb8Smec } 63628b464eb8Smec 63638b464eb8Smec ASSERT(cb_index != PC_NUM_CALLBACKS); 63648b464eb8Smec 63658b464eb8Smec rw_enter(&pc_cb[cb_index].cb_rwlock, RW_READER); 63668b464eb8Smec if (pc_cb[cb_index].cb_active) { 63678b464eb8Smec if (pc_cb[cb_index].duration == -1) { 63688b464eb8Smec bp1->expires = (clock_t)-1; 63698b464eb8Smec } else { 63708b464eb8Smec bp1->expires = lbolt + pc_cb[cb_index].duration; 63718b464eb8Smec } 63728b464eb8Smec } else { 63738b464eb8Smec /* There's no callback registered so don't add to the hash */ 63748b464eb8Smec rw_exit(&pc_cb[cb_index].cb_rwlock); 63758b464eb8Smec kmem_free(bp1, sizeof (*bp1)); 63768b464eb8Smec return; 63778b464eb8Smec } 63788b464eb8Smec 63798b464eb8Smec index = PAGE_CAPTURE_HASH(pp); 63808b464eb8Smec 63818b464eb8Smec /* 63828b464eb8Smec * Only allow capture flag to be modified under this mutex. 63838b464eb8Smec * Prevents multiple entries for same page getting added. 63848b464eb8Smec */ 63858b464eb8Smec mutex_enter(&page_capture_hash[index].pchh_mutex); 63868b464eb8Smec 63878b464eb8Smec /* 63888b464eb8Smec * if not already on the hash, set capture bit and add to the hash 63898b464eb8Smec */ 63908b464eb8Smec if (!(pp->p_toxic & PR_CAPTURE)) { 63918b464eb8Smec #ifdef DEBUG 63928b464eb8Smec /* Check for duplicate entries */ 63938b464eb8Smec for (l = 0; l < 2; l++) { 63948b464eb8Smec tp1 = page_capture_hash[index].lists[l].next; 63958b464eb8Smec while (tp1 != &page_capture_hash[index].lists[l]) { 63968b464eb8Smec if (tp1->pp == pp) { 63978b464eb8Smec panic("page pp 0x%p already on hash " 63988b464eb8Smec "at 0x%p\n", pp, tp1); 63998b464eb8Smec } 64008b464eb8Smec tp1 = tp1->next; 64018b464eb8Smec } 64028b464eb8Smec } 64038b464eb8Smec 64048b464eb8Smec #endif 64058b464eb8Smec page_settoxic(pp, PR_CAPTURE); 64068b464eb8Smec bp1->next = page_capture_hash[index].lists[0].next; 64078b464eb8Smec bp1->prev = &page_capture_hash[index].lists[0]; 64088b464eb8Smec bp1->next->prev = bp1; 64098b464eb8Smec page_capture_hash[index].lists[0].next = bp1; 64108b464eb8Smec page_capture_hash[index].num_pages++; 6411cee1d74bSjfrank if (flags & CAPTURE_RETIRE) { 6412cee1d74bSjfrank page_retire_incr_pend_count(); 6413cee1d74bSjfrank } 64148b464eb8Smec mutex_exit(&page_capture_hash[index].pchh_mutex); 64158b464eb8Smec rw_exit(&pc_cb[cb_index].cb_rwlock); 64168b464eb8Smec cv_signal(&pc_cv); 64178b464eb8Smec return; 64188b464eb8Smec } 64198b464eb8Smec 64208b464eb8Smec /* 64218b464eb8Smec * A page retire request will replace any other request. 64228b464eb8Smec * A second physmem request which is for a different process than 64238b464eb8Smec * the currently registered one will be dropped as there is 64248b464eb8Smec * no way to hold the private data for both calls. 64258b464eb8Smec * In the future, once there are more callers, this will have to 64268b464eb8Smec * be worked out better as there needs to be private storage for 64278b464eb8Smec * at least each type of caller (maybe have datap be an array of 64288b464eb8Smec * *void's so that we can index based upon callers index). 64298b464eb8Smec */ 64308b464eb8Smec 64318b464eb8Smec /* walk hash list to update expire time */ 64328b464eb8Smec for (i = 0; i < 2; i++) { 64338b464eb8Smec bp2 = page_capture_hash[index].lists[i].next; 64348b464eb8Smec while (bp2 != &page_capture_hash[index].lists[i]) { 64358b464eb8Smec if (bp2->pp == pp) { 64368b464eb8Smec if (flags & CAPTURE_RETIRE) { 64378b464eb8Smec if (!(bp2->flags & CAPTURE_RETIRE)) { 6438cee1d74bSjfrank page_retire_incr_pend_count(); 64398b464eb8Smec bp2->flags = flags; 64408b464eb8Smec bp2->expires = bp1->expires; 64418b464eb8Smec bp2->datap = datap; 64428b464eb8Smec } 64438b464eb8Smec } else { 64448b464eb8Smec ASSERT(flags & CAPTURE_PHYSMEM); 64458b464eb8Smec if (!(bp2->flags & CAPTURE_RETIRE) && 64468b464eb8Smec (datap == bp2->datap)) { 64478b464eb8Smec bp2->expires = bp1->expires; 64488b464eb8Smec } 64498b464eb8Smec } 64508b464eb8Smec mutex_exit(&page_capture_hash[index]. 64518b464eb8Smec pchh_mutex); 64528b464eb8Smec rw_exit(&pc_cb[cb_index].cb_rwlock); 64538b464eb8Smec kmem_free(bp1, sizeof (*bp1)); 64548b464eb8Smec return; 64558b464eb8Smec } 64568b464eb8Smec bp2 = bp2->next; 64578b464eb8Smec } 64588b464eb8Smec } 64598b464eb8Smec 64608b464eb8Smec /* 64618b464eb8Smec * the PR_CAPTURE flag is protected by the page_capture_hash mutexes 64628b464eb8Smec * and thus it either has to be set or not set and can't change 64638b464eb8Smec * while holding the mutex above. 64648b464eb8Smec */ 64658b464eb8Smec panic("page_capture_add_hash, PR_CAPTURE flag set on pp %p\n", pp); 64668b464eb8Smec } 64678b464eb8Smec 64688b464eb8Smec /* 64698b464eb8Smec * We have a page in our hands, lets try and make it ours by turning 64708b464eb8Smec * it into a clean page like it had just come off the freelists. 64718b464eb8Smec * 64728b464eb8Smec * Returns 0 on success, with the page still EXCL locked. 64738b464eb8Smec * On failure, the page will be unlocked, and returns EAGAIN 64748b464eb8Smec */ 64758b464eb8Smec static int 64768b464eb8Smec page_capture_clean_page(page_t *pp) 64778b464eb8Smec { 64788b464eb8Smec page_t *newpp; 64798b464eb8Smec int skip_unlock = 0; 64808b464eb8Smec spgcnt_t count; 64818b464eb8Smec page_t *tpp; 64828b464eb8Smec int ret = 0; 64838b464eb8Smec int extra; 64848b464eb8Smec 64858b464eb8Smec ASSERT(PAGE_EXCL(pp)); 64868b464eb8Smec ASSERT(!PP_RETIRED(pp)); 64878b464eb8Smec ASSERT(curthread->t_flag & T_CAPTURING); 64888b464eb8Smec 64898b464eb8Smec if (PP_ISFREE(pp)) { 64906e4dd838Smec if (!page_reclaim(pp, NULL)) { 64918b464eb8Smec skip_unlock = 1; 64928b464eb8Smec ret = EAGAIN; 64938b464eb8Smec goto cleanup; 64948b464eb8Smec } 64956e4dd838Smec ASSERT(pp->p_szc == 0); 64968b464eb8Smec if (pp->p_vnode != NULL) { 64978b464eb8Smec /* 64988b464eb8Smec * Since this page came from the 64998b464eb8Smec * cachelist, we must destroy the 65008b464eb8Smec * old vnode association. 65018b464eb8Smec */ 65028b464eb8Smec page_hashout(pp, NULL); 65038b464eb8Smec } 65048b464eb8Smec goto cleanup; 65058b464eb8Smec } 65068b464eb8Smec 65078b464eb8Smec /* 65088b464eb8Smec * If we know page_relocate will fail, skip it 65098b464eb8Smec * It could still fail due to a UE on another page but we 65108b464eb8Smec * can't do anything about that. 65118b464eb8Smec */ 65128b464eb8Smec if (pp->p_toxic & PR_UE) { 65138b464eb8Smec goto skip_relocate; 65148b464eb8Smec } 65158b464eb8Smec 65168b464eb8Smec /* 65178b464eb8Smec * It's possible that pages can not have a vnode as fsflush comes 65188b464eb8Smec * through and cleans up these pages. It's ugly but that's how it is. 65198b464eb8Smec */ 65208b464eb8Smec if (pp->p_vnode == NULL) { 65218b464eb8Smec goto skip_relocate; 65228b464eb8Smec } 65238b464eb8Smec 65248b464eb8Smec /* 65258b464eb8Smec * Page was not free, so lets try to relocate it. 65268b464eb8Smec * page_relocate only works with root pages, so if this is not a root 65278b464eb8Smec * page, we need to demote it to try and relocate it. 65288b464eb8Smec * Unfortunately this is the best we can do right now. 65298b464eb8Smec */ 65308b464eb8Smec newpp = NULL; 65318b464eb8Smec if ((pp->p_szc > 0) && (pp != PP_PAGEROOT(pp))) { 65328b464eb8Smec if (page_try_demote_pages(pp) == 0) { 65338b464eb8Smec ret = EAGAIN; 65348b464eb8Smec goto cleanup; 65358b464eb8Smec } 65368b464eb8Smec } 65378b464eb8Smec ret = page_relocate(&pp, &newpp, 1, 0, &count, NULL); 65388b464eb8Smec if (ret == 0) { 65398b464eb8Smec page_t *npp; 65408b464eb8Smec /* unlock the new page(s) */ 65418b464eb8Smec while (count-- > 0) { 65428b464eb8Smec ASSERT(newpp != NULL); 65438b464eb8Smec npp = newpp; 65448b464eb8Smec page_sub(&newpp, npp); 65458b464eb8Smec page_unlock(npp); 65468b464eb8Smec } 65478b464eb8Smec ASSERT(newpp == NULL); 65488b464eb8Smec /* 65498b464eb8Smec * Check to see if the page we have is too large. 65508b464eb8Smec * If so, demote it freeing up the extra pages. 65518b464eb8Smec */ 65528b464eb8Smec if (pp->p_szc > 0) { 65538b464eb8Smec /* For now demote extra pages to szc == 0 */ 65548b464eb8Smec extra = page_get_pagecnt(pp->p_szc) - 1; 65558b464eb8Smec while (extra > 0) { 65568b464eb8Smec tpp = pp->p_next; 65578b464eb8Smec page_sub(&pp, tpp); 65588b464eb8Smec tpp->p_szc = 0; 65598b464eb8Smec page_free(tpp, 1); 65608b464eb8Smec extra--; 65618b464eb8Smec } 65628b464eb8Smec /* Make sure to set our page to szc 0 as well */ 65638b464eb8Smec ASSERT(pp->p_next == pp && pp->p_prev == pp); 65648b464eb8Smec pp->p_szc = 0; 65658b464eb8Smec } 65668b464eb8Smec goto cleanup; 65678b464eb8Smec } else if (ret == EIO) { 65688b464eb8Smec ret = EAGAIN; 65698b464eb8Smec goto cleanup; 65708b464eb8Smec } else { 65718b464eb8Smec /* 65728b464eb8Smec * Need to reset return type as we failed to relocate the page 65738b464eb8Smec * but that does not mean that some of the next steps will not 65748b464eb8Smec * work. 65758b464eb8Smec */ 65768b464eb8Smec ret = 0; 65778b464eb8Smec } 65788b464eb8Smec 65798b464eb8Smec skip_relocate: 65808b464eb8Smec 65818b464eb8Smec if (pp->p_szc > 0) { 65828b464eb8Smec if (page_try_demote_pages(pp) == 0) { 65838b464eb8Smec ret = EAGAIN; 65848b464eb8Smec goto cleanup; 65858b464eb8Smec } 65868b464eb8Smec } 65878b464eb8Smec 65888b464eb8Smec ASSERT(pp->p_szc == 0); 65898b464eb8Smec 65908b464eb8Smec if (hat_ismod(pp)) { 65918b464eb8Smec ret = EAGAIN; 65928b464eb8Smec goto cleanup; 65938b464eb8Smec } 6594ad23a2dbSjohansen if (PP_ISKAS(pp)) { 65958b464eb8Smec ret = EAGAIN; 65968b464eb8Smec goto cleanup; 65978b464eb8Smec } 65988b464eb8Smec if (pp->p_lckcnt || pp->p_cowcnt) { 65998b464eb8Smec ret = EAGAIN; 66008b464eb8Smec goto cleanup; 66018b464eb8Smec } 66028b464eb8Smec 66038b464eb8Smec (void) hat_pageunload(pp, HAT_FORCE_PGUNLOAD); 66048b464eb8Smec ASSERT(!hat_page_is_mapped(pp)); 66058b464eb8Smec 66068b464eb8Smec if (hat_ismod(pp)) { 66078b464eb8Smec /* 66088b464eb8Smec * This is a semi-odd case as the page is now modified but not 66098b464eb8Smec * mapped as we just unloaded the mappings above. 66108b464eb8Smec */ 66118b464eb8Smec ret = EAGAIN; 66128b464eb8Smec goto cleanup; 66138b464eb8Smec } 66148b464eb8Smec if (pp->p_vnode != NULL) { 66158b464eb8Smec page_hashout(pp, NULL); 66168b464eb8Smec } 66178b464eb8Smec 66188b464eb8Smec /* 66198b464eb8Smec * At this point, the page should be in a clean state and 66208b464eb8Smec * we can do whatever we want with it. 66218b464eb8Smec */ 66228b464eb8Smec 66238b464eb8Smec cleanup: 66248b464eb8Smec if (ret != 0) { 66258b464eb8Smec if (!skip_unlock) { 66268b464eb8Smec page_unlock(pp); 66278b464eb8Smec } 66288b464eb8Smec } else { 66298b464eb8Smec ASSERT(pp->p_szc == 0); 66308b464eb8Smec ASSERT(PAGE_EXCL(pp)); 66318b464eb8Smec 66328b464eb8Smec pp->p_next = pp; 66338b464eb8Smec pp->p_prev = pp; 66348b464eb8Smec } 66358b464eb8Smec return (ret); 66368b464eb8Smec } 66378b464eb8Smec 66388b464eb8Smec /* 66398b464eb8Smec * Various callers of page_trycapture() can have different restrictions upon 66408b464eb8Smec * what memory they have access to. 66418b464eb8Smec * Returns 0 on success, with the following error codes on failure: 66428b464eb8Smec * EPERM - The requested page is long term locked, and thus repeated 66438b464eb8Smec * requests to capture this page will likely fail. 66448b464eb8Smec * ENOMEM - There was not enough free memory in the system to safely 66458b464eb8Smec * map the requested page. 66468b464eb8Smec * ENOENT - The requested page was inside the kernel cage, and the 66478b464eb8Smec * PHYSMEM_CAGE flag was not set. 66488b464eb8Smec */ 66498b464eb8Smec int 66508b464eb8Smec page_capture_pre_checks(page_t *pp, uint_t flags) 66518b464eb8Smec { 66528b464eb8Smec #if defined(__sparc) 66538b464eb8Smec extern struct vnode prom_ppages; 66548b464eb8Smec #endif /* __sparc */ 66558b464eb8Smec 66568b464eb8Smec ASSERT(pp != NULL); 66578b464eb8Smec 66588b464eb8Smec #if defined(__sparc) 66598b464eb8Smec if (pp->p_vnode == &prom_ppages) { 66608b464eb8Smec return (EPERM); 66618b464eb8Smec } 66628b464eb8Smec 6663a98e9dbfSaguzovsk if (PP_ISNORELOC(pp) && !(flags & CAPTURE_GET_CAGE) && 6664a98e9dbfSaguzovsk (flags & CAPTURE_PHYSMEM)) { 66658b464eb8Smec return (ENOENT); 66668b464eb8Smec } 66678b464eb8Smec 66688b464eb8Smec if (PP_ISNORELOCKERNEL(pp)) { 66698b464eb8Smec return (EPERM); 66708b464eb8Smec } 66718b464eb8Smec #else 6672ad23a2dbSjohansen if (PP_ISKAS(pp)) { 66738b464eb8Smec return (EPERM); 66748b464eb8Smec } 66758b464eb8Smec #endif /* __sparc */ 66768b464eb8Smec 6677a98e9dbfSaguzovsk /* only physmem currently has the restrictions checked below */ 6678a98e9dbfSaguzovsk if (!(flags & CAPTURE_PHYSMEM)) { 6679a98e9dbfSaguzovsk return (0); 6680a98e9dbfSaguzovsk } 6681a98e9dbfSaguzovsk 66828b464eb8Smec if (availrmem < swapfs_minfree) { 66838b464eb8Smec /* 66848b464eb8Smec * We won't try to capture this page as we are 66858b464eb8Smec * running low on memory. 66868b464eb8Smec */ 66878b464eb8Smec return (ENOMEM); 66888b464eb8Smec } 66898b464eb8Smec return (0); 66908b464eb8Smec } 66918b464eb8Smec 66928b464eb8Smec /* 66938b464eb8Smec * Once we have a page in our mits, go ahead and complete the capture 66948b464eb8Smec * operation. 66958b464eb8Smec * Returns 1 on failure where page is no longer needed 66968b464eb8Smec * Returns 0 on success 66978b464eb8Smec * Returns -1 if there was a transient failure. 66988b464eb8Smec * Failure cases must release the SE_EXCL lock on pp (usually via page_free). 66998b464eb8Smec */ 67008b464eb8Smec int 67018b464eb8Smec page_capture_take_action(page_t *pp, uint_t flags, void *datap) 67028b464eb8Smec { 67038b464eb8Smec int cb_index; 67048b464eb8Smec int ret = 0; 67058b464eb8Smec page_capture_hash_bucket_t *bp1; 67068b464eb8Smec page_capture_hash_bucket_t *bp2; 67078b464eb8Smec int index; 67088b464eb8Smec int found = 0; 67098b464eb8Smec int i; 67108b464eb8Smec 67118b464eb8Smec ASSERT(PAGE_EXCL(pp)); 67128b464eb8Smec ASSERT(curthread->t_flag & T_CAPTURING); 67138b464eb8Smec 67148b464eb8Smec for (cb_index = 0; cb_index < PC_NUM_CALLBACKS; cb_index++) { 67158b464eb8Smec if ((flags >> cb_index) & 1) { 67168b464eb8Smec break; 67178b464eb8Smec } 67188b464eb8Smec } 67198b464eb8Smec ASSERT(cb_index < PC_NUM_CALLBACKS); 67208b464eb8Smec 67218b464eb8Smec /* 67228b464eb8Smec * Remove the entry from the page_capture hash, but don't free it yet 67238b464eb8Smec * as we may need to put it back. 67248b464eb8Smec * Since we own the page at this point in time, we should find it 67258b464eb8Smec * in the hash if this is an ASYNC call. If we don't it's likely 67268b464eb8Smec * that the page_capture_async() thread decided that this request 67278b464eb8Smec * had expired, in which case we just continue on. 67288b464eb8Smec */ 67298b464eb8Smec if (flags & CAPTURE_ASYNC) { 67308b464eb8Smec 67318b464eb8Smec index = PAGE_CAPTURE_HASH(pp); 67328b464eb8Smec 67338b464eb8Smec mutex_enter(&page_capture_hash[index].pchh_mutex); 67348b464eb8Smec for (i = 0; i < 2 && !found; i++) { 67358b464eb8Smec bp1 = page_capture_hash[index].lists[i].next; 67368b464eb8Smec while (bp1 != &page_capture_hash[index].lists[i]) { 67378b464eb8Smec if (bp1->pp == pp) { 67388b464eb8Smec bp1->next->prev = bp1->prev; 67398b464eb8Smec bp1->prev->next = bp1->next; 67408b464eb8Smec page_capture_hash[index].num_pages--; 67418b464eb8Smec page_clrtoxic(pp, PR_CAPTURE); 67428b464eb8Smec found = 1; 67438b464eb8Smec break; 67448b464eb8Smec } 67458b464eb8Smec bp1 = bp1->next; 67468b464eb8Smec } 67478b464eb8Smec } 67488b464eb8Smec mutex_exit(&page_capture_hash[index].pchh_mutex); 67498b464eb8Smec } 67508b464eb8Smec 67518b464eb8Smec /* Synchronize with the unregister func. */ 67528b464eb8Smec rw_enter(&pc_cb[cb_index].cb_rwlock, RW_READER); 67538b464eb8Smec if (!pc_cb[cb_index].cb_active) { 67548b464eb8Smec page_free(pp, 1); 67558b464eb8Smec rw_exit(&pc_cb[cb_index].cb_rwlock); 67568b464eb8Smec if (found) { 67578b464eb8Smec kmem_free(bp1, sizeof (*bp1)); 67588b464eb8Smec } 67598b464eb8Smec return (1); 67608b464eb8Smec } 67618b464eb8Smec 67628b464eb8Smec /* 67638b464eb8Smec * We need to remove the entry from the page capture hash and turn off 67648b464eb8Smec * the PR_CAPTURE bit before calling the callback. We'll need to cache 67658b464eb8Smec * the entry here, and then based upon the return value, cleanup 67668b464eb8Smec * appropriately or re-add it to the hash, making sure that someone else 67678b464eb8Smec * hasn't already done so. 67688b464eb8Smec * It should be rare for the callback to fail and thus it's ok for 67698b464eb8Smec * the failure path to be a bit complicated as the success path is 67708b464eb8Smec * cleaner and the locking rules are easier to follow. 67718b464eb8Smec */ 67728b464eb8Smec 67738b464eb8Smec ret = pc_cb[cb_index].cb_func(pp, datap, flags); 67748b464eb8Smec 67758b464eb8Smec rw_exit(&pc_cb[cb_index].cb_rwlock); 67768b464eb8Smec 67778b464eb8Smec /* 67788b464eb8Smec * If this was an ASYNC request, we need to cleanup the hash if the 67798b464eb8Smec * callback was successful or if the request was no longer valid. 67808b464eb8Smec * For non-ASYNC requests, we return failure to map and the caller 67818b464eb8Smec * will take care of adding the request to the hash. 67828b464eb8Smec * Note also that the callback itself is responsible for the page 67838b464eb8Smec * at this point in time in terms of locking ... The most common 67848b464eb8Smec * case for the failure path should just be a page_free. 67858b464eb8Smec */ 67868b464eb8Smec if (ret >= 0) { 67878b464eb8Smec if (found) { 6788cee1d74bSjfrank if (bp1->flags & CAPTURE_RETIRE) { 6789cee1d74bSjfrank page_retire_decr_pend_count(); 6790cee1d74bSjfrank } 67918b464eb8Smec kmem_free(bp1, sizeof (*bp1)); 67928b464eb8Smec } 67938b464eb8Smec return (ret); 67948b464eb8Smec } 67958b464eb8Smec if (!found) { 67968b464eb8Smec return (ret); 67978b464eb8Smec } 67988b464eb8Smec 67998b464eb8Smec ASSERT(flags & CAPTURE_ASYNC); 68008b464eb8Smec 68018b464eb8Smec /* 68028b464eb8Smec * Check for expiration time first as we can just free it up if it's 68038b464eb8Smec * expired. 68048b464eb8Smec */ 68058b464eb8Smec if (lbolt > bp1->expires && bp1->expires != -1) { 68068b464eb8Smec kmem_free(bp1, sizeof (*bp1)); 68078b464eb8Smec return (ret); 68088b464eb8Smec } 68098b464eb8Smec 68108b464eb8Smec /* 68118b464eb8Smec * The callback failed and there used to be an entry in the hash for 68128b464eb8Smec * this page, so we need to add it back to the hash. 68138b464eb8Smec */ 68148b464eb8Smec mutex_enter(&page_capture_hash[index].pchh_mutex); 68158b464eb8Smec if (!(pp->p_toxic & PR_CAPTURE)) { 68168b464eb8Smec /* just add bp1 back to head of walked list */ 68178b464eb8Smec page_settoxic(pp, PR_CAPTURE); 68188b464eb8Smec bp1->next = page_capture_hash[index].lists[1].next; 68198b464eb8Smec bp1->prev = &page_capture_hash[index].lists[1]; 68208b464eb8Smec bp1->next->prev = bp1; 68218b464eb8Smec page_capture_hash[index].lists[1].next = bp1; 68228b464eb8Smec page_capture_hash[index].num_pages++; 68238b464eb8Smec mutex_exit(&page_capture_hash[index].pchh_mutex); 68248b464eb8Smec return (ret); 68258b464eb8Smec } 68268b464eb8Smec 68278b464eb8Smec /* 68288b464eb8Smec * Otherwise there was a new capture request added to list 68298b464eb8Smec * Need to make sure that our original data is represented if 68308b464eb8Smec * appropriate. 68318b464eb8Smec */ 68328b464eb8Smec for (i = 0; i < 2; i++) { 68338b464eb8Smec bp2 = page_capture_hash[index].lists[i].next; 68348b464eb8Smec while (bp2 != &page_capture_hash[index].lists[i]) { 68358b464eb8Smec if (bp2->pp == pp) { 68368b464eb8Smec if (bp1->flags & CAPTURE_RETIRE) { 68378b464eb8Smec if (!(bp2->flags & CAPTURE_RETIRE)) { 68388b464eb8Smec bp2->szc = bp1->szc; 68398b464eb8Smec bp2->flags = bp1->flags; 68408b464eb8Smec bp2->expires = bp1->expires; 68418b464eb8Smec bp2->datap = bp1->datap; 68428b464eb8Smec } 68438b464eb8Smec } else { 68448b464eb8Smec ASSERT(bp1->flags & CAPTURE_PHYSMEM); 68458b464eb8Smec if (!(bp2->flags & CAPTURE_RETIRE)) { 68468b464eb8Smec bp2->szc = bp1->szc; 68478b464eb8Smec bp2->flags = bp1->flags; 68488b464eb8Smec bp2->expires = bp1->expires; 68498b464eb8Smec bp2->datap = bp1->datap; 68508b464eb8Smec } 68518b464eb8Smec } 68528b464eb8Smec mutex_exit(&page_capture_hash[index]. 68538b464eb8Smec pchh_mutex); 68548b464eb8Smec kmem_free(bp1, sizeof (*bp1)); 68558b464eb8Smec return (ret); 68568b464eb8Smec } 68578b464eb8Smec bp2 = bp2->next; 68588b464eb8Smec } 68598b464eb8Smec } 68608b464eb8Smec panic("PR_CAPTURE set but not on hash for pp 0x%p\n", pp); 68618b464eb8Smec /*NOTREACHED*/ 68628b464eb8Smec } 68638b464eb8Smec 68648b464eb8Smec /* 68658b464eb8Smec * Try to capture the given page for the caller specified in the flags 68668b464eb8Smec * parameter. The page will either be captured and handed over to the 68678b464eb8Smec * appropriate callback, or will be queued up in the page capture hash 68688b464eb8Smec * to be captured asynchronously. 68698b464eb8Smec * If the current request is due to an async capture, the page must be 68708b464eb8Smec * exclusively locked before calling this function. 68718b464eb8Smec * Currently szc must be 0 but in the future this should be expandable to 68728b464eb8Smec * other page sizes. 68738b464eb8Smec * Returns 0 on success, with the following error codes on failure: 68748b464eb8Smec * EPERM - The requested page is long term locked, and thus repeated 68758b464eb8Smec * requests to capture this page will likely fail. 68768b464eb8Smec * ENOMEM - There was not enough free memory in the system to safely 68778b464eb8Smec * map the requested page. 68788b464eb8Smec * ENOENT - The requested page was inside the kernel cage, and the 68798b464eb8Smec * CAPTURE_GET_CAGE flag was not set. 68808b464eb8Smec * EAGAIN - The requested page could not be capturead at this point in 68818b464eb8Smec * time but future requests will likely work. 68828b464eb8Smec * EBUSY - The requested page is retired and the CAPTURE_GET_RETIRED flag 68838b464eb8Smec * was not set. 68848b464eb8Smec */ 68858b464eb8Smec int 68868b464eb8Smec page_itrycapture(page_t *pp, uint_t szc, uint_t flags, void *datap) 68878b464eb8Smec { 68888b464eb8Smec int ret; 68898b464eb8Smec int cb_index; 68908b464eb8Smec 68918b464eb8Smec if (flags & CAPTURE_ASYNC) { 68928b464eb8Smec ASSERT(PAGE_EXCL(pp)); 68938b464eb8Smec goto async; 68948b464eb8Smec } 68958b464eb8Smec 68968b464eb8Smec /* Make sure there's enough availrmem ... */ 68978b464eb8Smec ret = page_capture_pre_checks(pp, flags); 68988b464eb8Smec if (ret != 0) { 68998b464eb8Smec return (ret); 69008b464eb8Smec } 69018b464eb8Smec 69028b464eb8Smec if (!page_trylock(pp, SE_EXCL)) { 69038b464eb8Smec for (cb_index = 0; cb_index < PC_NUM_CALLBACKS; cb_index++) { 69048b464eb8Smec if ((flags >> cb_index) & 1) { 69058b464eb8Smec break; 69068b464eb8Smec } 69078b464eb8Smec } 69088b464eb8Smec ASSERT(cb_index < PC_NUM_CALLBACKS); 69098b464eb8Smec ret = EAGAIN; 69108b464eb8Smec /* Special case for retired pages */ 69118b464eb8Smec if (PP_RETIRED(pp)) { 69128b464eb8Smec if (flags & CAPTURE_GET_RETIRED) { 69138b464eb8Smec if (!page_unretire_pp(pp, PR_UNR_TEMP)) { 69148b464eb8Smec /* 69158b464eb8Smec * Need to set capture bit and add to 69168b464eb8Smec * hash so that the page will be 69178b464eb8Smec * retired when freed. 69188b464eb8Smec */ 69198b464eb8Smec page_capture_add_hash(pp, szc, 69208b464eb8Smec CAPTURE_RETIRE, NULL); 69218b464eb8Smec ret = 0; 69228b464eb8Smec goto own_page; 69238b464eb8Smec } 69248b464eb8Smec } else { 69258b464eb8Smec return (EBUSY); 69268b464eb8Smec } 69278b464eb8Smec } 69288b464eb8Smec page_capture_add_hash(pp, szc, flags, datap); 69298b464eb8Smec return (ret); 69308b464eb8Smec } 69318b464eb8Smec 69328b464eb8Smec async: 69338b464eb8Smec ASSERT(PAGE_EXCL(pp)); 69348b464eb8Smec 69358b464eb8Smec /* Need to check for physmem async requests that availrmem is sane */ 69368b464eb8Smec if ((flags & (CAPTURE_ASYNC | CAPTURE_PHYSMEM)) == 69378b464eb8Smec (CAPTURE_ASYNC | CAPTURE_PHYSMEM) && 69388b464eb8Smec (availrmem < swapfs_minfree)) { 69398b464eb8Smec page_unlock(pp); 69408b464eb8Smec return (ENOMEM); 69418b464eb8Smec } 69428b464eb8Smec 69438b464eb8Smec ret = page_capture_clean_page(pp); 69448b464eb8Smec 69458b464eb8Smec if (ret != 0) { 69468b464eb8Smec /* We failed to get the page, so lets add it to the hash */ 69478b464eb8Smec if (!(flags & CAPTURE_ASYNC)) { 69488b464eb8Smec page_capture_add_hash(pp, szc, flags, datap); 69498b464eb8Smec } 69508b464eb8Smec return (ret); 69518b464eb8Smec } 69528b464eb8Smec 69538b464eb8Smec own_page: 69548b464eb8Smec ASSERT(PAGE_EXCL(pp)); 69558b464eb8Smec ASSERT(pp->p_szc == 0); 69568b464eb8Smec 69578b464eb8Smec /* Call the callback */ 69588b464eb8Smec ret = page_capture_take_action(pp, flags, datap); 69598b464eb8Smec 69608b464eb8Smec if (ret == 0) { 69618b464eb8Smec return (0); 69628b464eb8Smec } 69638b464eb8Smec 69648b464eb8Smec /* 69658b464eb8Smec * Note that in the failure cases from page_capture_take_action, the 69668b464eb8Smec * EXCL lock will have already been dropped. 69678b464eb8Smec */ 69688b464eb8Smec if ((ret == -1) && (!(flags & CAPTURE_ASYNC))) { 69698b464eb8Smec page_capture_add_hash(pp, szc, flags, datap); 69708b464eb8Smec } 69718b464eb8Smec return (EAGAIN); 69728b464eb8Smec } 69738b464eb8Smec 69748b464eb8Smec int 69758b464eb8Smec page_trycapture(page_t *pp, uint_t szc, uint_t flags, void *datap) 69768b464eb8Smec { 69778b464eb8Smec int ret; 69788b464eb8Smec 69798b464eb8Smec curthread->t_flag |= T_CAPTURING; 69808b464eb8Smec ret = page_itrycapture(pp, szc, flags, datap); 69818b464eb8Smec curthread->t_flag &= ~T_CAPTURING; /* xor works as we know its set */ 69828b464eb8Smec return (ret); 69838b464eb8Smec } 69848b464eb8Smec 69858b464eb8Smec /* 69868b464eb8Smec * When unlocking a page which has the PR_CAPTURE bit set, this routine 69878b464eb8Smec * gets called to try and capture the page. 69888b464eb8Smec */ 69898b464eb8Smec void 69908b464eb8Smec page_unlock_capture(page_t *pp) 69918b464eb8Smec { 69928b464eb8Smec page_capture_hash_bucket_t *bp; 69938b464eb8Smec int index; 69948b464eb8Smec int i; 69958b464eb8Smec uint_t szc; 69968b464eb8Smec uint_t flags = 0; 69978b464eb8Smec void *datap; 69988b464eb8Smec kmutex_t *mp; 69998b464eb8Smec extern vnode_t retired_pages; 70008b464eb8Smec 70018b464eb8Smec /* 70028b464eb8Smec * We need to protect against a possible deadlock here where we own 70038b464eb8Smec * the vnode page hash mutex and want to acquire it again as there 70048b464eb8Smec * are locations in the code, where we unlock a page while holding 70058b464eb8Smec * the mutex which can lead to the page being captured and eventually 70068b464eb8Smec * end up here. As we may be hashing out the old page and hashing into 70078b464eb8Smec * the retire vnode, we need to make sure we don't own them. 70088b464eb8Smec * Other callbacks who do hash operations also need to make sure that 70098b464eb8Smec * before they hashin to a vnode that they do not currently own the 70108b464eb8Smec * vphm mutex otherwise there will be a panic. 70118b464eb8Smec */ 70128b464eb8Smec if (mutex_owned(page_vnode_mutex(&retired_pages))) { 701322addef7Smec page_unlock_nocapture(pp); 70148b464eb8Smec return; 70158b464eb8Smec } 70168b464eb8Smec if (pp->p_vnode != NULL && mutex_owned(page_vnode_mutex(pp->p_vnode))) { 701722addef7Smec page_unlock_nocapture(pp); 70188b464eb8Smec return; 70198b464eb8Smec } 70208b464eb8Smec 70218b464eb8Smec index = PAGE_CAPTURE_HASH(pp); 70228b464eb8Smec 70238b464eb8Smec mp = &page_capture_hash[index].pchh_mutex; 70248b464eb8Smec mutex_enter(mp); 70258b464eb8Smec for (i = 0; i < 2; i++) { 70268b464eb8Smec bp = page_capture_hash[index].lists[i].next; 70278b464eb8Smec while (bp != &page_capture_hash[index].lists[i]) { 70288b464eb8Smec if (bp->pp == pp) { 70298b464eb8Smec szc = bp->szc; 70308b464eb8Smec flags = bp->flags | CAPTURE_ASYNC; 70318b464eb8Smec datap = bp->datap; 70328b464eb8Smec mutex_exit(mp); 70338b464eb8Smec (void) page_trycapture(pp, szc, flags, datap); 70348b464eb8Smec return; 70358b464eb8Smec } 70368b464eb8Smec bp = bp->next; 70378b464eb8Smec } 70388b464eb8Smec } 70398b464eb8Smec 70408b464eb8Smec /* Failed to find page in hash so clear flags and unlock it. */ 70418b464eb8Smec page_clrtoxic(pp, PR_CAPTURE); 70428b464eb8Smec page_unlock(pp); 70438b464eb8Smec 70448b464eb8Smec mutex_exit(mp); 70458b464eb8Smec } 70468b464eb8Smec 70478b464eb8Smec void 70488b464eb8Smec page_capture_init() 70498b464eb8Smec { 70508b464eb8Smec int i; 70518b464eb8Smec for (i = 0; i < NUM_PAGE_CAPTURE_BUCKETS; i++) { 70528b464eb8Smec page_capture_hash[i].lists[0].next = 70538b464eb8Smec &page_capture_hash[i].lists[0]; 70548b464eb8Smec page_capture_hash[i].lists[0].prev = 70558b464eb8Smec &page_capture_hash[i].lists[0]; 70568b464eb8Smec page_capture_hash[i].lists[1].next = 70578b464eb8Smec &page_capture_hash[i].lists[1]; 70588b464eb8Smec page_capture_hash[i].lists[1].prev = 70598b464eb8Smec &page_capture_hash[i].lists[1]; 70608b464eb8Smec } 70618b464eb8Smec 70628b464eb8Smec pc_thread_shortwait = 23 * hz; 70638b464eb8Smec pc_thread_longwait = 1201 * hz; 7064a98e9dbfSaguzovsk pc_thread_retry = 3; 70658b464eb8Smec mutex_init(&pc_thread_mutex, NULL, MUTEX_DEFAULT, NULL); 70668b464eb8Smec cv_init(&pc_cv, NULL, CV_DEFAULT, NULL); 70678b464eb8Smec pc_thread_id = thread_create(NULL, 0, page_capture_thread, NULL, 0, &p0, 70688b464eb8Smec TS_RUN, minclsyspri); 70698b464eb8Smec } 70708b464eb8Smec 70718b464eb8Smec /* 70728b464eb8Smec * It is necessary to scrub any failing pages prior to reboot in order to 70738b464eb8Smec * prevent a latent error trap from occurring on the next boot. 70748b464eb8Smec */ 70758b464eb8Smec void 70768b464eb8Smec page_retire_mdboot() 70778b464eb8Smec { 70788b464eb8Smec page_t *pp; 70798b464eb8Smec int i, j; 70808b464eb8Smec page_capture_hash_bucket_t *bp; 70818b464eb8Smec 70828b464eb8Smec /* walk lists looking for pages to scrub */ 70838b464eb8Smec for (i = 0; i < NUM_PAGE_CAPTURE_BUCKETS; i++) { 70848b464eb8Smec if (page_capture_hash[i].num_pages == 0) 70858b464eb8Smec continue; 70868b464eb8Smec 70878b464eb8Smec mutex_enter(&page_capture_hash[i].pchh_mutex); 70888b464eb8Smec 70898b464eb8Smec for (j = 0; j < 2; j++) { 70908b464eb8Smec bp = page_capture_hash[i].lists[j].next; 70918b464eb8Smec while (bp != &page_capture_hash[i].lists[j]) { 70928b464eb8Smec pp = bp->pp; 7093ad23a2dbSjohansen if (!PP_ISKAS(pp) && PP_TOXIC(pp)) { 70948b464eb8Smec pp->p_selock = -1; /* pacify ASSERTs */ 70958b464eb8Smec PP_CLRFREE(pp); 70968b464eb8Smec pagescrub(pp, 0, PAGESIZE); 70978b464eb8Smec pp->p_selock = 0; 70988b464eb8Smec } 70998b464eb8Smec bp = bp->next; 71008b464eb8Smec } 71018b464eb8Smec } 71028b464eb8Smec mutex_exit(&page_capture_hash[i].pchh_mutex); 71038b464eb8Smec } 71048b464eb8Smec } 71058b464eb8Smec 71068b464eb8Smec /* 71078b464eb8Smec * Walk the page_capture_hash trying to capture pages and also cleanup old 71088b464eb8Smec * entries which have expired. 71098b464eb8Smec */ 71108b464eb8Smec void 71118b464eb8Smec page_capture_async() 71128b464eb8Smec { 71138b464eb8Smec page_t *pp; 71148b464eb8Smec int i; 71158b464eb8Smec int ret; 71168b464eb8Smec page_capture_hash_bucket_t *bp1, *bp2; 71178b464eb8Smec uint_t szc; 71188b464eb8Smec uint_t flags; 71198b464eb8Smec void *datap; 71208b464eb8Smec 71218b464eb8Smec /* If there are outstanding pages to be captured, get to work */ 71228b464eb8Smec for (i = 0; i < NUM_PAGE_CAPTURE_BUCKETS; i++) { 71238b464eb8Smec if (page_capture_hash[i].num_pages == 0) 71248b464eb8Smec continue; 71258b464eb8Smec /* Append list 1 to list 0 and then walk through list 0 */ 71268b464eb8Smec mutex_enter(&page_capture_hash[i].pchh_mutex); 71278b464eb8Smec bp1 = &page_capture_hash[i].lists[1]; 71288b464eb8Smec bp2 = bp1->next; 71298b464eb8Smec if (bp1 != bp2) { 71308b464eb8Smec bp1->prev->next = page_capture_hash[i].lists[0].next; 71318b464eb8Smec bp2->prev = &page_capture_hash[i].lists[0]; 71328b464eb8Smec page_capture_hash[i].lists[0].next->prev = bp1->prev; 71338b464eb8Smec page_capture_hash[i].lists[0].next = bp2; 71348b464eb8Smec bp1->next = bp1; 71358b464eb8Smec bp1->prev = bp1; 71368b464eb8Smec } 71378b464eb8Smec 71388b464eb8Smec /* list[1] will be empty now */ 71398b464eb8Smec 71408b464eb8Smec bp1 = page_capture_hash[i].lists[0].next; 71418b464eb8Smec while (bp1 != &page_capture_hash[i].lists[0]) { 71428b464eb8Smec /* Check expiration time */ 71438b464eb8Smec if ((lbolt > bp1->expires && bp1->expires != -1) || 71448b464eb8Smec page_deleted(bp1->pp)) { 71458b464eb8Smec page_capture_hash[i].lists[0].next = bp1->next; 71468b464eb8Smec bp1->next->prev = 71478b464eb8Smec &page_capture_hash[i].lists[0]; 71488b464eb8Smec page_capture_hash[i].num_pages--; 71498b464eb8Smec 71508b464eb8Smec /* 71518b464eb8Smec * We can safely remove the PR_CAPTURE bit 71528b464eb8Smec * without holding the EXCL lock on the page 71538b464eb8Smec * as the PR_CAPTURE bit requres that the 71548b464eb8Smec * page_capture_hash[].pchh_mutex be held 71558b464eb8Smec * to modify it. 71568b464eb8Smec */ 71578b464eb8Smec page_clrtoxic(bp1->pp, PR_CAPTURE); 71588b464eb8Smec mutex_exit(&page_capture_hash[i].pchh_mutex); 71598b464eb8Smec kmem_free(bp1, sizeof (*bp1)); 71608b464eb8Smec mutex_enter(&page_capture_hash[i].pchh_mutex); 71618b464eb8Smec bp1 = page_capture_hash[i].lists[0].next; 71628b464eb8Smec continue; 71638b464eb8Smec } 71648b464eb8Smec pp = bp1->pp; 71658b464eb8Smec szc = bp1->szc; 71668b464eb8Smec flags = bp1->flags; 71678b464eb8Smec datap = bp1->datap; 71688b464eb8Smec mutex_exit(&page_capture_hash[i].pchh_mutex); 71698b464eb8Smec if (page_trylock(pp, SE_EXCL)) { 71708b464eb8Smec ret = page_trycapture(pp, szc, 71718b464eb8Smec flags | CAPTURE_ASYNC, datap); 71728b464eb8Smec } else { 71738b464eb8Smec ret = 1; /* move to walked hash */ 71748b464eb8Smec } 71758b464eb8Smec 71768b464eb8Smec if (ret != 0) { 71778b464eb8Smec /* Move to walked hash */ 71788b464eb8Smec (void) page_capture_move_to_walked(pp); 71798b464eb8Smec } 71808b464eb8Smec mutex_enter(&page_capture_hash[i].pchh_mutex); 71818b464eb8Smec bp1 = page_capture_hash[i].lists[0].next; 71828b464eb8Smec } 71838b464eb8Smec 71848b464eb8Smec mutex_exit(&page_capture_hash[i].pchh_mutex); 71858b464eb8Smec } 71868b464eb8Smec } 71878b464eb8Smec 7188cee1d74bSjfrank /* 7189cee1d74bSjfrank * This function is called by the page_capture_thread, and is needed in 7190cee1d74bSjfrank * in order to initiate aio cleanup, so that pages used in aio 7191cee1d74bSjfrank * will be unlocked and subsequently retired by page_capture_thread. 7192cee1d74bSjfrank */ 7193cee1d74bSjfrank static int 7194cee1d74bSjfrank do_aio_cleanup(void) 7195cee1d74bSjfrank { 7196cee1d74bSjfrank proc_t *procp; 7197cee1d74bSjfrank int (*aio_cleanup_dr_delete_memory)(proc_t *); 7198cee1d74bSjfrank int cleaned = 0; 7199cee1d74bSjfrank 7200cee1d74bSjfrank if (modload("sys", "kaio") == -1) { 7201cee1d74bSjfrank cmn_err(CE_WARN, "do_aio_cleanup: cannot load kaio"); 7202cee1d74bSjfrank return (0); 7203cee1d74bSjfrank } 7204cee1d74bSjfrank /* 7205cee1d74bSjfrank * We use the aio_cleanup_dr_delete_memory function to 7206cee1d74bSjfrank * initiate the actual clean up; this function will wake 7207cee1d74bSjfrank * up the per-process aio_cleanup_thread. 7208cee1d74bSjfrank */ 7209cee1d74bSjfrank aio_cleanup_dr_delete_memory = (int (*)(proc_t *)) 7210cee1d74bSjfrank modgetsymvalue("aio_cleanup_dr_delete_memory", 0); 7211cee1d74bSjfrank if (aio_cleanup_dr_delete_memory == NULL) { 7212cee1d74bSjfrank cmn_err(CE_WARN, 7213cee1d74bSjfrank "aio_cleanup_dr_delete_memory not found in kaio"); 7214cee1d74bSjfrank return (0); 7215cee1d74bSjfrank } 7216cee1d74bSjfrank mutex_enter(&pidlock); 7217cee1d74bSjfrank for (procp = practive; (procp != NULL); procp = procp->p_next) { 7218cee1d74bSjfrank mutex_enter(&procp->p_lock); 7219cee1d74bSjfrank if (procp->p_aio != NULL) { 7220cee1d74bSjfrank /* cleanup proc's outstanding kaio */ 7221cee1d74bSjfrank cleaned += (*aio_cleanup_dr_delete_memory)(procp); 7222cee1d74bSjfrank } 7223cee1d74bSjfrank mutex_exit(&procp->p_lock); 7224cee1d74bSjfrank } 7225cee1d74bSjfrank mutex_exit(&pidlock); 7226cee1d74bSjfrank return (cleaned); 7227cee1d74bSjfrank } 7228cee1d74bSjfrank 7229cee1d74bSjfrank /* 7230cee1d74bSjfrank * helper function for page_capture_thread 7231cee1d74bSjfrank */ 7232cee1d74bSjfrank static void 7233cee1d74bSjfrank page_capture_handle_outstanding(void) 7234cee1d74bSjfrank { 7235cee1d74bSjfrank int ntry; 7236cee1d74bSjfrank 7237cee1d74bSjfrank if (!page_retire_pend_count()) { 7238cee1d74bSjfrank /* 7239cee1d74bSjfrank * Do we really want to be this aggressive 7240cee1d74bSjfrank * for things other than page_retire? 7241cee1d74bSjfrank * Maybe have a counter for each callback 7242cee1d74bSjfrank * type to guide how aggressive we should 7243cee1d74bSjfrank * be here. Thus if there's at least one 7244cee1d74bSjfrank * page for page_retire we go ahead and reap 7245cee1d74bSjfrank * like this. 7246cee1d74bSjfrank */ 7247cee1d74bSjfrank kmem_reap(); 7248cee1d74bSjfrank seg_preap(); 7249cee1d74bSjfrank page_capture_async(); 7250cee1d74bSjfrank } else { 7251cee1d74bSjfrank /* 7252cee1d74bSjfrank * There are pages pending retirement, so 7253cee1d74bSjfrank * we reap prior to attempting to capture. 7254cee1d74bSjfrank */ 7255cee1d74bSjfrank kmem_reap(); 7256a98e9dbfSaguzovsk 7257a98e9dbfSaguzovsk /* disable and purge seg_pcache */ 7258a98e9dbfSaguzovsk (void) seg_p_disable(); 7259a98e9dbfSaguzovsk for (ntry = 0; ntry < pc_thread_retry; ntry++) { 7260a98e9dbfSaguzovsk if (!page_retire_pend_count()) 7261a98e9dbfSaguzovsk break; 7262a98e9dbfSaguzovsk if (do_aio_cleanup()) { 7263a98e9dbfSaguzovsk /* 7264a98e9dbfSaguzovsk * allow the apps cleanup threads 7265a98e9dbfSaguzovsk * to run 7266a98e9dbfSaguzovsk */ 7267a98e9dbfSaguzovsk delay(pc_thread_shortwait); 7268cee1d74bSjfrank } 7269cee1d74bSjfrank page_capture_async(); 7270cee1d74bSjfrank } 7271a98e9dbfSaguzovsk /* reenable seg_pcache */ 7272a98e9dbfSaguzovsk seg_p_enable(); 7273cee1d74bSjfrank } 7274cee1d74bSjfrank } 7275cee1d74bSjfrank 72768b464eb8Smec /* 72778b464eb8Smec * The page_capture_thread loops forever, looking to see if there are 72788b464eb8Smec * pages still waiting to be captured. 72798b464eb8Smec */ 72808b464eb8Smec static void 72818b464eb8Smec page_capture_thread(void) 72828b464eb8Smec { 72838b464eb8Smec callb_cpr_t c; 72848b464eb8Smec int outstanding; 72858b464eb8Smec int i; 72868b464eb8Smec 72878b464eb8Smec CALLB_CPR_INIT(&c, &pc_thread_mutex, callb_generic_cpr, "page_capture"); 72888b464eb8Smec 72898b464eb8Smec mutex_enter(&pc_thread_mutex); 72908b464eb8Smec for (;;) { 72918b464eb8Smec outstanding = 0; 72928b464eb8Smec for (i = 0; i < NUM_PAGE_CAPTURE_BUCKETS; i++) 72938b464eb8Smec outstanding += page_capture_hash[i].num_pages; 72948b464eb8Smec if (outstanding) { 7295cee1d74bSjfrank page_capture_handle_outstanding(); 72968b464eb8Smec CALLB_CPR_SAFE_BEGIN(&c); 72978b464eb8Smec (void) cv_timedwait(&pc_cv, &pc_thread_mutex, 72988b464eb8Smec lbolt + pc_thread_shortwait); 72998b464eb8Smec CALLB_CPR_SAFE_END(&c, &pc_thread_mutex); 73008b464eb8Smec } else { 73018b464eb8Smec CALLB_CPR_SAFE_BEGIN(&c); 73028b464eb8Smec (void) cv_timedwait(&pc_cv, &pc_thread_mutex, 73038b464eb8Smec lbolt + pc_thread_longwait); 73048b464eb8Smec CALLB_CPR_SAFE_END(&c, &pc_thread_mutex); 73058b464eb8Smec } 73068b464eb8Smec } 73078b464eb8Smec /*NOTREACHED*/ 73088b464eb8Smec } 7309*06fb6a36Sdv /* 7310*06fb6a36Sdv * Attempt to locate a bucket that has enough pages to satisfy the request. 7311*06fb6a36Sdv * The initial check is done without the lock to avoid unneeded contention. 7312*06fb6a36Sdv * The function returns 1 if enough pages were found, else 0 if it could not 7313*06fb6a36Sdv * find enough pages in a bucket. 7314*06fb6a36Sdv */ 7315*06fb6a36Sdv static int 7316*06fb6a36Sdv pcf_decrement_bucket(pgcnt_t npages) 7317*06fb6a36Sdv { 7318*06fb6a36Sdv struct pcf *p; 7319*06fb6a36Sdv struct pcf *q; 7320*06fb6a36Sdv int i; 7321*06fb6a36Sdv 7322*06fb6a36Sdv p = &pcf[PCF_INDEX()]; 7323*06fb6a36Sdv q = &pcf[pcf_fanout]; 7324*06fb6a36Sdv for (i = 0; i < pcf_fanout; i++) { 7325*06fb6a36Sdv if (p->pcf_count > npages) { 7326*06fb6a36Sdv /* 7327*06fb6a36Sdv * a good one to try. 7328*06fb6a36Sdv */ 7329*06fb6a36Sdv mutex_enter(&p->pcf_lock); 7330*06fb6a36Sdv if (p->pcf_count > npages) { 7331*06fb6a36Sdv p->pcf_count -= (uint_t)npages; 7332*06fb6a36Sdv /* 7333*06fb6a36Sdv * freemem is not protected by any lock. 7334*06fb6a36Sdv * Thus, we cannot have any assertion 7335*06fb6a36Sdv * containing freemem here. 7336*06fb6a36Sdv */ 7337*06fb6a36Sdv freemem -= npages; 7338*06fb6a36Sdv mutex_exit(&p->pcf_lock); 7339*06fb6a36Sdv return (1); 7340*06fb6a36Sdv } 7341*06fb6a36Sdv mutex_exit(&p->pcf_lock); 7342*06fb6a36Sdv } 7343*06fb6a36Sdv p++; 7344*06fb6a36Sdv if (p >= q) { 7345*06fb6a36Sdv p = pcf; 7346*06fb6a36Sdv } 7347*06fb6a36Sdv } 7348*06fb6a36Sdv return (0); 7349*06fb6a36Sdv } 7350*06fb6a36Sdv 7351*06fb6a36Sdv /* 7352*06fb6a36Sdv * Arguments: 7353*06fb6a36Sdv * pcftotal_ret: If the value is not NULL and we have walked all the 7354*06fb6a36Sdv * buckets but did not find enough pages then it will 7355*06fb6a36Sdv * be set to the total number of pages in all the pcf 7356*06fb6a36Sdv * buckets. 7357*06fb6a36Sdv * npages: Is the number of pages we have been requested to 7358*06fb6a36Sdv * find. 7359*06fb6a36Sdv * unlock: If set to 0 we will leave the buckets locked if the 7360*06fb6a36Sdv * requested number of pages are not found. 7361*06fb6a36Sdv * 7362*06fb6a36Sdv * Go and try to satisfy the page request from any number of buckets. 7363*06fb6a36Sdv * This can be a very expensive operation as we have to lock the buckets 7364*06fb6a36Sdv * we are checking (and keep them locked), starting at bucket 0. 7365*06fb6a36Sdv * 7366*06fb6a36Sdv * The function returns 1 if enough pages were found, else 0 if it could not 7367*06fb6a36Sdv * find enough pages in the buckets. 7368*06fb6a36Sdv * 7369*06fb6a36Sdv */ 7370*06fb6a36Sdv static int 7371*06fb6a36Sdv pcf_decrement_multiple(pgcnt_t *pcftotal_ret, pgcnt_t npages, int unlock) 7372*06fb6a36Sdv { 7373*06fb6a36Sdv struct pcf *p; 7374*06fb6a36Sdv pgcnt_t pcftotal; 7375*06fb6a36Sdv int i; 7376*06fb6a36Sdv 7377*06fb6a36Sdv p = pcf; 7378*06fb6a36Sdv /* try to collect pages from several pcf bins */ 7379*06fb6a36Sdv for (pcftotal = 0, i = 0; i < pcf_fanout; i++) { 7380*06fb6a36Sdv mutex_enter(&p->pcf_lock); 7381*06fb6a36Sdv pcftotal += p->pcf_count; 7382*06fb6a36Sdv if (pcftotal >= npages) { 7383*06fb6a36Sdv /* 7384*06fb6a36Sdv * Wow! There are enough pages laying around 7385*06fb6a36Sdv * to satisfy the request. Do the accounting, 7386*06fb6a36Sdv * drop the locks we acquired, and go back. 7387*06fb6a36Sdv * 7388*06fb6a36Sdv * freemem is not protected by any lock. So, 7389*06fb6a36Sdv * we cannot have any assertion containing 7390*06fb6a36Sdv * freemem. 7391*06fb6a36Sdv */ 7392*06fb6a36Sdv freemem -= npages; 7393*06fb6a36Sdv while (p >= pcf) { 7394*06fb6a36Sdv if (p->pcf_count <= npages) { 7395*06fb6a36Sdv npages -= p->pcf_count; 7396*06fb6a36Sdv p->pcf_count = 0; 7397*06fb6a36Sdv } else { 7398*06fb6a36Sdv p->pcf_count -= (uint_t)npages; 7399*06fb6a36Sdv npages = 0; 7400*06fb6a36Sdv } 7401*06fb6a36Sdv mutex_exit(&p->pcf_lock); 7402*06fb6a36Sdv p--; 7403*06fb6a36Sdv } 7404*06fb6a36Sdv ASSERT(npages == 0); 7405*06fb6a36Sdv return (1); 7406*06fb6a36Sdv } 7407*06fb6a36Sdv p++; 7408*06fb6a36Sdv } 7409*06fb6a36Sdv if (unlock) { 7410*06fb6a36Sdv /* failed to collect pages - release the locks */ 7411*06fb6a36Sdv while (--p >= pcf) { 7412*06fb6a36Sdv mutex_exit(&p->pcf_lock); 7413*06fb6a36Sdv } 7414*06fb6a36Sdv } 7415*06fb6a36Sdv if (pcftotal_ret != NULL) 7416*06fb6a36Sdv *pcftotal_ret = pcftotal; 7417*06fb6a36Sdv return (0); 7418*06fb6a36Sdv } 7419