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 502ff05a9Svsakar * Common Development and Distribution License (the "License"). 602ff05a9Svsakar * 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 /* 22*17965fd8SKrishnendu Sadhukhan - Sun Microsystems * Copyright 2010 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) 1984, 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 /* 407c478bd9Sstevel@tonic-gate * VM - address spaces. 417c478bd9Sstevel@tonic-gate */ 427c478bd9Sstevel@tonic-gate 437c478bd9Sstevel@tonic-gate #include <sys/types.h> 447c478bd9Sstevel@tonic-gate #include <sys/t_lock.h> 457c478bd9Sstevel@tonic-gate #include <sys/param.h> 467c478bd9Sstevel@tonic-gate #include <sys/errno.h> 477c478bd9Sstevel@tonic-gate #include <sys/systm.h> 487c478bd9Sstevel@tonic-gate #include <sys/mman.h> 497c478bd9Sstevel@tonic-gate #include <sys/sysmacros.h> 507c478bd9Sstevel@tonic-gate #include <sys/cpuvar.h> 517c478bd9Sstevel@tonic-gate #include <sys/sysinfo.h> 527c478bd9Sstevel@tonic-gate #include <sys/kmem.h> 537c478bd9Sstevel@tonic-gate #include <sys/vnode.h> 547c478bd9Sstevel@tonic-gate #include <sys/vmsystm.h> 557c478bd9Sstevel@tonic-gate #include <sys/cmn_err.h> 567c478bd9Sstevel@tonic-gate #include <sys/debug.h> 577c478bd9Sstevel@tonic-gate #include <sys/tnf_probe.h> 587c478bd9Sstevel@tonic-gate #include <sys/vtrace.h> 597c478bd9Sstevel@tonic-gate 607c478bd9Sstevel@tonic-gate #include <vm/hat.h> 617c478bd9Sstevel@tonic-gate #include <vm/xhat.h> 627c478bd9Sstevel@tonic-gate #include <vm/as.h> 637c478bd9Sstevel@tonic-gate #include <vm/seg.h> 647c478bd9Sstevel@tonic-gate #include <vm/seg_vn.h> 657c478bd9Sstevel@tonic-gate #include <vm/seg_dev.h> 667c478bd9Sstevel@tonic-gate #include <vm/seg_kmem.h> 677c478bd9Sstevel@tonic-gate #include <vm/seg_map.h> 687c478bd9Sstevel@tonic-gate #include <vm/seg_spt.h> 697c478bd9Sstevel@tonic-gate #include <vm/page.h> 707c478bd9Sstevel@tonic-gate 717c478bd9Sstevel@tonic-gate clock_t deadlk_wait = 1; /* number of ticks to wait before retrying */ 727c478bd9Sstevel@tonic-gate 737c478bd9Sstevel@tonic-gate static struct kmem_cache *as_cache; 747c478bd9Sstevel@tonic-gate 757c478bd9Sstevel@tonic-gate static void as_setwatchprot(struct as *, caddr_t, size_t, uint_t); 767c478bd9Sstevel@tonic-gate static void as_clearwatchprot(struct as *, caddr_t, size_t); 7702ff05a9Svsakar int as_map_locked(struct as *, caddr_t, size_t, int ((*)()), void *); 787c478bd9Sstevel@tonic-gate 797c478bd9Sstevel@tonic-gate 807c478bd9Sstevel@tonic-gate /* 817c478bd9Sstevel@tonic-gate * Verifying the segment lists is very time-consuming; it may not be 827c478bd9Sstevel@tonic-gate * desirable always to define VERIFY_SEGLIST when DEBUG is set. 837c478bd9Sstevel@tonic-gate */ 847c478bd9Sstevel@tonic-gate #ifdef DEBUG 857c478bd9Sstevel@tonic-gate #define VERIFY_SEGLIST 867c478bd9Sstevel@tonic-gate int do_as_verify = 0; 877c478bd9Sstevel@tonic-gate #endif 887c478bd9Sstevel@tonic-gate 897c478bd9Sstevel@tonic-gate /* 907c478bd9Sstevel@tonic-gate * Allocate a new callback data structure entry and fill in the events of 917c478bd9Sstevel@tonic-gate * interest, the address range of interest, and the callback argument. 927c478bd9Sstevel@tonic-gate * Link the entry on the as->a_callbacks list. A callback entry for the 937c478bd9Sstevel@tonic-gate * entire address space may be specified with vaddr = 0 and size = -1. 947c478bd9Sstevel@tonic-gate * 957c478bd9Sstevel@tonic-gate * CALLERS RESPONSIBILITY: If not calling from within the process context for 967c478bd9Sstevel@tonic-gate * the specified as, the caller must guarantee persistence of the specified as 977c478bd9Sstevel@tonic-gate * for the duration of this function (eg. pages being locked within the as 987c478bd9Sstevel@tonic-gate * will guarantee persistence). 997c478bd9Sstevel@tonic-gate */ 1007c478bd9Sstevel@tonic-gate int 1017c478bd9Sstevel@tonic-gate as_add_callback(struct as *as, void (*cb_func)(), void *arg, uint_t events, 1027c478bd9Sstevel@tonic-gate caddr_t vaddr, size_t size, int sleepflag) 1037c478bd9Sstevel@tonic-gate { 1047c478bd9Sstevel@tonic-gate struct as_callback *current_head, *cb; 1057c478bd9Sstevel@tonic-gate caddr_t saddr; 1067c478bd9Sstevel@tonic-gate size_t rsize; 1077c478bd9Sstevel@tonic-gate 1087c478bd9Sstevel@tonic-gate /* callback function and an event are mandatory */ 1097c478bd9Sstevel@tonic-gate if ((cb_func == NULL) || ((events & AS_ALL_EVENT) == 0)) 1107c478bd9Sstevel@tonic-gate return (EINVAL); 1117c478bd9Sstevel@tonic-gate 1127c478bd9Sstevel@tonic-gate /* Adding a callback after as_free has been called is not allowed */ 1137c478bd9Sstevel@tonic-gate if (as == &kas) 1147c478bd9Sstevel@tonic-gate return (ENOMEM); 1157c478bd9Sstevel@tonic-gate 1167c478bd9Sstevel@tonic-gate /* 1177c478bd9Sstevel@tonic-gate * vaddr = 0 and size = -1 is used to indicate that the callback range 1187c478bd9Sstevel@tonic-gate * is the entire address space so no rounding is done in that case. 1197c478bd9Sstevel@tonic-gate */ 1207c478bd9Sstevel@tonic-gate if (size != -1) { 1217c478bd9Sstevel@tonic-gate saddr = (caddr_t)((uintptr_t)vaddr & (uintptr_t)PAGEMASK); 1227c478bd9Sstevel@tonic-gate rsize = (((size_t)(vaddr + size) + PAGEOFFSET) & PAGEMASK) - 123843e1988Sjohnlev (size_t)saddr; 1247c478bd9Sstevel@tonic-gate /* check for wraparound */ 1257c478bd9Sstevel@tonic-gate if (saddr + rsize < saddr) 1267c478bd9Sstevel@tonic-gate return (ENOMEM); 1277c478bd9Sstevel@tonic-gate } else { 1287c478bd9Sstevel@tonic-gate if (vaddr != 0) 1297c478bd9Sstevel@tonic-gate return (EINVAL); 1307c478bd9Sstevel@tonic-gate saddr = vaddr; 1317c478bd9Sstevel@tonic-gate rsize = size; 1327c478bd9Sstevel@tonic-gate } 1337c478bd9Sstevel@tonic-gate 1347c478bd9Sstevel@tonic-gate /* Allocate and initialize a callback entry */ 1357c478bd9Sstevel@tonic-gate cb = kmem_zalloc(sizeof (struct as_callback), sleepflag); 1367c478bd9Sstevel@tonic-gate if (cb == NULL) 1377c478bd9Sstevel@tonic-gate return (EAGAIN); 1387c478bd9Sstevel@tonic-gate 1397c478bd9Sstevel@tonic-gate cb->ascb_func = cb_func; 1407c478bd9Sstevel@tonic-gate cb->ascb_arg = arg; 1417c478bd9Sstevel@tonic-gate cb->ascb_events = events; 1427c478bd9Sstevel@tonic-gate cb->ascb_saddr = saddr; 1437c478bd9Sstevel@tonic-gate cb->ascb_len = rsize; 1447c478bd9Sstevel@tonic-gate 1457c478bd9Sstevel@tonic-gate /* Add the entry to the list */ 1467c478bd9Sstevel@tonic-gate mutex_enter(&as->a_contents); 1477c478bd9Sstevel@tonic-gate current_head = as->a_callbacks; 1487c478bd9Sstevel@tonic-gate as->a_callbacks = cb; 1497c478bd9Sstevel@tonic-gate cb->ascb_next = current_head; 1507c478bd9Sstevel@tonic-gate 1517c478bd9Sstevel@tonic-gate /* 1527c478bd9Sstevel@tonic-gate * The call to this function may lose in a race with 1537c478bd9Sstevel@tonic-gate * a pertinent event - eg. a thread does long term memory locking 1547c478bd9Sstevel@tonic-gate * but before the callback is added another thread executes as_unmap. 1557c478bd9Sstevel@tonic-gate * A broadcast here resolves that. 1567c478bd9Sstevel@tonic-gate */ 1577c478bd9Sstevel@tonic-gate if ((cb->ascb_events & AS_UNMAPWAIT_EVENT) && AS_ISUNMAPWAIT(as)) { 1587c478bd9Sstevel@tonic-gate AS_CLRUNMAPWAIT(as); 1597c478bd9Sstevel@tonic-gate cv_broadcast(&as->a_cv); 1607c478bd9Sstevel@tonic-gate } 1617c478bd9Sstevel@tonic-gate 1627c478bd9Sstevel@tonic-gate mutex_exit(&as->a_contents); 1637c478bd9Sstevel@tonic-gate return (0); 1647c478bd9Sstevel@tonic-gate } 1657c478bd9Sstevel@tonic-gate 1667c478bd9Sstevel@tonic-gate /* 1677c478bd9Sstevel@tonic-gate * Search the callback list for an entry which pertains to arg. 1687c478bd9Sstevel@tonic-gate * 1697c478bd9Sstevel@tonic-gate * This is called from within the client upon completion of the callback. 1707c478bd9Sstevel@tonic-gate * RETURN VALUES: 1717c478bd9Sstevel@tonic-gate * AS_CALLBACK_DELETED (callback entry found and deleted) 1727c478bd9Sstevel@tonic-gate * AS_CALLBACK_NOTFOUND (no callback entry found - this is ok) 1737c478bd9Sstevel@tonic-gate * AS_CALLBACK_DELETE_DEFERRED (callback is in process, delete of this 1747c478bd9Sstevel@tonic-gate * entry will be made in as_do_callbacks) 1757c478bd9Sstevel@tonic-gate * 1767c478bd9Sstevel@tonic-gate * If as_delete_callback encounters a matching entry with AS_CALLBACK_CALLED 1777c478bd9Sstevel@tonic-gate * set, it indicates that as_do_callbacks is processing this entry. The 1787c478bd9Sstevel@tonic-gate * AS_ALL_EVENT events are cleared in the entry, and a broadcast is made 1797c478bd9Sstevel@tonic-gate * to unblock as_do_callbacks, in case it is blocked. 1807c478bd9Sstevel@tonic-gate * 1817c478bd9Sstevel@tonic-gate * CALLERS RESPONSIBILITY: If not calling from within the process context for 1827c478bd9Sstevel@tonic-gate * the specified as, the caller must guarantee persistence of the specified as 1837c478bd9Sstevel@tonic-gate * for the duration of this function (eg. pages being locked within the as 1847c478bd9Sstevel@tonic-gate * will guarantee persistence). 1857c478bd9Sstevel@tonic-gate */ 1867c478bd9Sstevel@tonic-gate uint_t 1877c478bd9Sstevel@tonic-gate as_delete_callback(struct as *as, void *arg) 1887c478bd9Sstevel@tonic-gate { 1897c478bd9Sstevel@tonic-gate struct as_callback **prevcb = &as->a_callbacks; 1907c478bd9Sstevel@tonic-gate struct as_callback *cb; 1917c478bd9Sstevel@tonic-gate uint_t rc = AS_CALLBACK_NOTFOUND; 1927c478bd9Sstevel@tonic-gate 1937c478bd9Sstevel@tonic-gate mutex_enter(&as->a_contents); 1947c478bd9Sstevel@tonic-gate for (cb = as->a_callbacks; cb; prevcb = &cb->ascb_next, cb = *prevcb) { 1957c478bd9Sstevel@tonic-gate if (cb->ascb_arg != arg) 1967c478bd9Sstevel@tonic-gate continue; 1977c478bd9Sstevel@tonic-gate 1987c478bd9Sstevel@tonic-gate /* 1997c478bd9Sstevel@tonic-gate * If the events indicate AS_CALLBACK_CALLED, just clear 2007c478bd9Sstevel@tonic-gate * AS_ALL_EVENT in the events field and wakeup the thread 2017c478bd9Sstevel@tonic-gate * that may be waiting in as_do_callbacks. as_do_callbacks 2027c478bd9Sstevel@tonic-gate * will take care of removing this entry from the list. In 2037c478bd9Sstevel@tonic-gate * that case, return AS_CALLBACK_DELETE_DEFERRED. Otherwise 2047c478bd9Sstevel@tonic-gate * (AS_CALLBACK_CALLED not set), just remove it from the 2057c478bd9Sstevel@tonic-gate * list, return the memory and return AS_CALLBACK_DELETED. 2067c478bd9Sstevel@tonic-gate */ 2077c478bd9Sstevel@tonic-gate if ((cb->ascb_events & AS_CALLBACK_CALLED) != 0) { 2087c478bd9Sstevel@tonic-gate /* leave AS_CALLBACK_CALLED */ 2097c478bd9Sstevel@tonic-gate cb->ascb_events &= ~AS_ALL_EVENT; 2107c478bd9Sstevel@tonic-gate rc = AS_CALLBACK_DELETE_DEFERRED; 2117c478bd9Sstevel@tonic-gate cv_broadcast(&as->a_cv); 2127c478bd9Sstevel@tonic-gate } else { 2137c478bd9Sstevel@tonic-gate *prevcb = cb->ascb_next; 2147c478bd9Sstevel@tonic-gate kmem_free(cb, sizeof (struct as_callback)); 2157c478bd9Sstevel@tonic-gate rc = AS_CALLBACK_DELETED; 2167c478bd9Sstevel@tonic-gate } 2177c478bd9Sstevel@tonic-gate break; 2187c478bd9Sstevel@tonic-gate } 2197c478bd9Sstevel@tonic-gate mutex_exit(&as->a_contents); 2207c478bd9Sstevel@tonic-gate return (rc); 2217c478bd9Sstevel@tonic-gate } 2227c478bd9Sstevel@tonic-gate 2237c478bd9Sstevel@tonic-gate /* 2247c478bd9Sstevel@tonic-gate * Searches the as callback list for a matching entry. 2257c478bd9Sstevel@tonic-gate * Returns a pointer to the first matching callback, or NULL if 2267c478bd9Sstevel@tonic-gate * nothing is found. 2277c478bd9Sstevel@tonic-gate * This function never sleeps so it is ok to call it with more 2287c478bd9Sstevel@tonic-gate * locks held but the (required) a_contents mutex. 2297c478bd9Sstevel@tonic-gate * 2307c478bd9Sstevel@tonic-gate * See also comment on as_do_callbacks below. 2317c478bd9Sstevel@tonic-gate */ 2327c478bd9Sstevel@tonic-gate static struct as_callback * 2337c478bd9Sstevel@tonic-gate as_find_callback(struct as *as, uint_t events, caddr_t event_addr, 2347c478bd9Sstevel@tonic-gate size_t event_len) 2357c478bd9Sstevel@tonic-gate { 2367c478bd9Sstevel@tonic-gate struct as_callback *cb; 2377c478bd9Sstevel@tonic-gate 2387c478bd9Sstevel@tonic-gate ASSERT(MUTEX_HELD(&as->a_contents)); 2397c478bd9Sstevel@tonic-gate for (cb = as->a_callbacks; cb != NULL; cb = cb->ascb_next) { 2407c478bd9Sstevel@tonic-gate /* 2417c478bd9Sstevel@tonic-gate * If the callback has not already been called, then 2427c478bd9Sstevel@tonic-gate * check if events or address range pertains. An event_len 2437c478bd9Sstevel@tonic-gate * of zero means do an unconditional callback. 2447c478bd9Sstevel@tonic-gate */ 2457c478bd9Sstevel@tonic-gate if (((cb->ascb_events & AS_CALLBACK_CALLED) != 0) || 2467c478bd9Sstevel@tonic-gate ((event_len != 0) && (((cb->ascb_events & events) == 0) || 2477c478bd9Sstevel@tonic-gate (event_addr + event_len < cb->ascb_saddr) || 2487c478bd9Sstevel@tonic-gate (event_addr > (cb->ascb_saddr + cb->ascb_len))))) { 2497c478bd9Sstevel@tonic-gate continue; 2507c478bd9Sstevel@tonic-gate } 2517c478bd9Sstevel@tonic-gate break; 2527c478bd9Sstevel@tonic-gate } 2537c478bd9Sstevel@tonic-gate return (cb); 2547c478bd9Sstevel@tonic-gate } 2557c478bd9Sstevel@tonic-gate 2567c478bd9Sstevel@tonic-gate /* 2577c478bd9Sstevel@tonic-gate * Executes a given callback and removes it from the callback list for 2587c478bd9Sstevel@tonic-gate * this address space. 2597c478bd9Sstevel@tonic-gate * This function may sleep so the caller must drop all locks except 2607c478bd9Sstevel@tonic-gate * a_contents before calling this func. 2617c478bd9Sstevel@tonic-gate * 2627c478bd9Sstevel@tonic-gate * See also comments on as_do_callbacks below. 2637c478bd9Sstevel@tonic-gate */ 2647c478bd9Sstevel@tonic-gate static void 2657c478bd9Sstevel@tonic-gate as_execute_callback(struct as *as, struct as_callback *cb, 2667c478bd9Sstevel@tonic-gate uint_t events) 2677c478bd9Sstevel@tonic-gate { 2687c478bd9Sstevel@tonic-gate struct as_callback **prevcb; 2697c478bd9Sstevel@tonic-gate void *cb_arg; 2707c478bd9Sstevel@tonic-gate 2717c478bd9Sstevel@tonic-gate ASSERT(MUTEX_HELD(&as->a_contents) && (cb->ascb_events & events)); 2727c478bd9Sstevel@tonic-gate cb->ascb_events |= AS_CALLBACK_CALLED; 2737c478bd9Sstevel@tonic-gate mutex_exit(&as->a_contents); 2747c478bd9Sstevel@tonic-gate (*cb->ascb_func)(as, cb->ascb_arg, events); 2757c478bd9Sstevel@tonic-gate mutex_enter(&as->a_contents); 2767c478bd9Sstevel@tonic-gate /* 2777c478bd9Sstevel@tonic-gate * the callback function is required to delete the callback 2787c478bd9Sstevel@tonic-gate * when the callback function determines it is OK for 2797c478bd9Sstevel@tonic-gate * this thread to continue. as_delete_callback will clear 2807c478bd9Sstevel@tonic-gate * the AS_ALL_EVENT in the events field when it is deleted. 2817c478bd9Sstevel@tonic-gate * If the callback function called as_delete_callback, 2827c478bd9Sstevel@tonic-gate * events will already be cleared and there will be no blocking. 2837c478bd9Sstevel@tonic-gate */ 2847c478bd9Sstevel@tonic-gate while ((cb->ascb_events & events) != 0) { 2857c478bd9Sstevel@tonic-gate cv_wait(&as->a_cv, &as->a_contents); 2867c478bd9Sstevel@tonic-gate } 2877c478bd9Sstevel@tonic-gate /* 2887c478bd9Sstevel@tonic-gate * This entry needs to be taken off the list. Normally, the 2897c478bd9Sstevel@tonic-gate * callback func itself does that, but unfortunately the list 2907c478bd9Sstevel@tonic-gate * may have changed while the callback was running because the 2917c478bd9Sstevel@tonic-gate * a_contents mutex was dropped and someone else other than the 2927c478bd9Sstevel@tonic-gate * callback func itself could have called as_delete_callback, 2937c478bd9Sstevel@tonic-gate * so we have to search to find this entry again. The entry 2947c478bd9Sstevel@tonic-gate * must have AS_CALLBACK_CALLED, and have the same 'arg'. 2957c478bd9Sstevel@tonic-gate */ 2967c478bd9Sstevel@tonic-gate cb_arg = cb->ascb_arg; 2977c478bd9Sstevel@tonic-gate prevcb = &as->a_callbacks; 2987c478bd9Sstevel@tonic-gate for (cb = as->a_callbacks; cb != NULL; 2997c478bd9Sstevel@tonic-gate prevcb = &cb->ascb_next, cb = *prevcb) { 3007c478bd9Sstevel@tonic-gate if (((cb->ascb_events & AS_CALLBACK_CALLED) == 0) || 3017c478bd9Sstevel@tonic-gate (cb_arg != cb->ascb_arg)) { 3027c478bd9Sstevel@tonic-gate continue; 3037c478bd9Sstevel@tonic-gate } 3047c478bd9Sstevel@tonic-gate *prevcb = cb->ascb_next; 3057c478bd9Sstevel@tonic-gate kmem_free(cb, sizeof (struct as_callback)); 3067c478bd9Sstevel@tonic-gate break; 3077c478bd9Sstevel@tonic-gate } 3087c478bd9Sstevel@tonic-gate } 3097c478bd9Sstevel@tonic-gate 3107c478bd9Sstevel@tonic-gate /* 3117c478bd9Sstevel@tonic-gate * Check the callback list for a matching event and intersection of 3127c478bd9Sstevel@tonic-gate * address range. If there is a match invoke the callback. Skip an entry if: 3137c478bd9Sstevel@tonic-gate * - a callback is already in progress for this entry (AS_CALLBACK_CALLED) 3147c478bd9Sstevel@tonic-gate * - not event of interest 3157c478bd9Sstevel@tonic-gate * - not address range of interest 3167c478bd9Sstevel@tonic-gate * 3177c478bd9Sstevel@tonic-gate * An event_len of zero indicates a request for an unconditional callback 3187c478bd9Sstevel@tonic-gate * (regardless of event), only the AS_CALLBACK_CALLED is checked. The 3197c478bd9Sstevel@tonic-gate * a_contents lock must be dropped before a callback, so only one callback 3207c478bd9Sstevel@tonic-gate * can be done before returning. Return -1 (true) if a callback was 3217c478bd9Sstevel@tonic-gate * executed and removed from the list, else return 0 (false). 3227c478bd9Sstevel@tonic-gate * 3237c478bd9Sstevel@tonic-gate * The logically separate parts, i.e. finding a matching callback and 3247c478bd9Sstevel@tonic-gate * executing a given callback have been separated into two functions 3257c478bd9Sstevel@tonic-gate * so that they can be called with different sets of locks held beyond 3267c478bd9Sstevel@tonic-gate * the always-required a_contents. as_find_callback does not sleep so 3277c478bd9Sstevel@tonic-gate * it is ok to call it if more locks than a_contents (i.e. the a_lock 3287c478bd9Sstevel@tonic-gate * rwlock) are held. as_execute_callback on the other hand may sleep 3297c478bd9Sstevel@tonic-gate * so all locks beyond a_contents must be dropped by the caller if one 3307c478bd9Sstevel@tonic-gate * does not want to end comatose. 3317c478bd9Sstevel@tonic-gate */ 3327c478bd9Sstevel@tonic-gate static int 3337c478bd9Sstevel@tonic-gate as_do_callbacks(struct as *as, uint_t events, caddr_t event_addr, 3347c478bd9Sstevel@tonic-gate size_t event_len) 3357c478bd9Sstevel@tonic-gate { 3367c478bd9Sstevel@tonic-gate struct as_callback *cb; 3377c478bd9Sstevel@tonic-gate 3387c478bd9Sstevel@tonic-gate if ((cb = as_find_callback(as, events, event_addr, event_len))) { 3397c478bd9Sstevel@tonic-gate as_execute_callback(as, cb, events); 3407c478bd9Sstevel@tonic-gate return (-1); 3417c478bd9Sstevel@tonic-gate } 3427c478bd9Sstevel@tonic-gate return (0); 3437c478bd9Sstevel@tonic-gate } 3447c478bd9Sstevel@tonic-gate 3457c478bd9Sstevel@tonic-gate /* 3467c478bd9Sstevel@tonic-gate * Search for the segment containing addr. If a segment containing addr 3477c478bd9Sstevel@tonic-gate * exists, that segment is returned. If no such segment exists, and 3487c478bd9Sstevel@tonic-gate * the list spans addresses greater than addr, then the first segment 3497c478bd9Sstevel@tonic-gate * whose base is greater than addr is returned; otherwise, NULL is 3507c478bd9Sstevel@tonic-gate * returned unless tail is true, in which case the last element of the 3517c478bd9Sstevel@tonic-gate * list is returned. 3527c478bd9Sstevel@tonic-gate * 3537c478bd9Sstevel@tonic-gate * a_seglast is used to cache the last found segment for repeated 3547c478bd9Sstevel@tonic-gate * searches to the same addr (which happens frequently). 3557c478bd9Sstevel@tonic-gate */ 3567c478bd9Sstevel@tonic-gate struct seg * 3577c478bd9Sstevel@tonic-gate as_findseg(struct as *as, caddr_t addr, int tail) 3587c478bd9Sstevel@tonic-gate { 3597c478bd9Sstevel@tonic-gate struct seg *seg = as->a_seglast; 3607c478bd9Sstevel@tonic-gate avl_index_t where; 3617c478bd9Sstevel@tonic-gate 3627c478bd9Sstevel@tonic-gate ASSERT(AS_LOCK_HELD(as, &as->a_lock)); 3637c478bd9Sstevel@tonic-gate 3647c478bd9Sstevel@tonic-gate if (seg != NULL && 3657c478bd9Sstevel@tonic-gate seg->s_base <= addr && 3667c478bd9Sstevel@tonic-gate addr < seg->s_base + seg->s_size) 3677c478bd9Sstevel@tonic-gate return (seg); 3687c478bd9Sstevel@tonic-gate 3697c478bd9Sstevel@tonic-gate seg = avl_find(&as->a_segtree, &addr, &where); 3707c478bd9Sstevel@tonic-gate if (seg != NULL) 3717c478bd9Sstevel@tonic-gate return (as->a_seglast = seg); 3727c478bd9Sstevel@tonic-gate 3737c478bd9Sstevel@tonic-gate seg = avl_nearest(&as->a_segtree, where, AVL_AFTER); 3747c478bd9Sstevel@tonic-gate if (seg == NULL && tail) 3757c478bd9Sstevel@tonic-gate seg = avl_last(&as->a_segtree); 3767c478bd9Sstevel@tonic-gate return (as->a_seglast = seg); 3777c478bd9Sstevel@tonic-gate } 3787c478bd9Sstevel@tonic-gate 3797c478bd9Sstevel@tonic-gate #ifdef VERIFY_SEGLIST 3807c478bd9Sstevel@tonic-gate /* 3817c478bd9Sstevel@tonic-gate * verify that the linked list is coherent 3827c478bd9Sstevel@tonic-gate */ 3837c478bd9Sstevel@tonic-gate static void 3847c478bd9Sstevel@tonic-gate as_verify(struct as *as) 3857c478bd9Sstevel@tonic-gate { 3867c478bd9Sstevel@tonic-gate struct seg *seg, *seglast, *p, *n; 3877c478bd9Sstevel@tonic-gate uint_t nsegs = 0; 3887c478bd9Sstevel@tonic-gate 3897c478bd9Sstevel@tonic-gate if (do_as_verify == 0) 3907c478bd9Sstevel@tonic-gate return; 3917c478bd9Sstevel@tonic-gate 3927c478bd9Sstevel@tonic-gate seglast = as->a_seglast; 3937c478bd9Sstevel@tonic-gate 3947c478bd9Sstevel@tonic-gate for (seg = AS_SEGFIRST(as); seg != NULL; seg = AS_SEGNEXT(as, seg)) { 3957c478bd9Sstevel@tonic-gate ASSERT(seg->s_as == as); 3967c478bd9Sstevel@tonic-gate p = AS_SEGPREV(as, seg); 3977c478bd9Sstevel@tonic-gate n = AS_SEGNEXT(as, seg); 3987c478bd9Sstevel@tonic-gate ASSERT(p == NULL || p->s_as == as); 3997c478bd9Sstevel@tonic-gate ASSERT(p == NULL || p->s_base < seg->s_base); 4007c478bd9Sstevel@tonic-gate ASSERT(n == NULL || n->s_base > seg->s_base); 4017c478bd9Sstevel@tonic-gate ASSERT(n != NULL || seg == avl_last(&as->a_segtree)); 4027c478bd9Sstevel@tonic-gate if (seg == seglast) 4037c478bd9Sstevel@tonic-gate seglast = NULL; 4047c478bd9Sstevel@tonic-gate nsegs++; 4057c478bd9Sstevel@tonic-gate } 4067c478bd9Sstevel@tonic-gate ASSERT(seglast == NULL); 4077c478bd9Sstevel@tonic-gate ASSERT(avl_numnodes(&as->a_segtree) == nsegs); 4087c478bd9Sstevel@tonic-gate } 4097c478bd9Sstevel@tonic-gate #endif /* VERIFY_SEGLIST */ 4107c478bd9Sstevel@tonic-gate 4117c478bd9Sstevel@tonic-gate /* 4127c478bd9Sstevel@tonic-gate * Add a new segment to the address space. The avl_find() 4137c478bd9Sstevel@tonic-gate * may be expensive so we attempt to use last segment accessed 4147c478bd9Sstevel@tonic-gate * in as_gap() as an insertion point. 4157c478bd9Sstevel@tonic-gate */ 4167c478bd9Sstevel@tonic-gate int 4177c478bd9Sstevel@tonic-gate as_addseg(struct as *as, struct seg *newseg) 4187c478bd9Sstevel@tonic-gate { 4197c478bd9Sstevel@tonic-gate struct seg *seg; 4207c478bd9Sstevel@tonic-gate caddr_t addr; 4217c478bd9Sstevel@tonic-gate caddr_t eaddr; 4227c478bd9Sstevel@tonic-gate avl_index_t where; 4237c478bd9Sstevel@tonic-gate 4247c478bd9Sstevel@tonic-gate ASSERT(AS_WRITE_HELD(as, &as->a_lock)); 4257c478bd9Sstevel@tonic-gate 4267c478bd9Sstevel@tonic-gate as->a_updatedir = 1; /* inform /proc */ 4277c478bd9Sstevel@tonic-gate gethrestime(&as->a_updatetime); 4287c478bd9Sstevel@tonic-gate 4297c478bd9Sstevel@tonic-gate if (as->a_lastgaphl != NULL) { 4307c478bd9Sstevel@tonic-gate struct seg *hseg = NULL; 4317c478bd9Sstevel@tonic-gate struct seg *lseg = NULL; 4327c478bd9Sstevel@tonic-gate 4337c478bd9Sstevel@tonic-gate if (as->a_lastgaphl->s_base > newseg->s_base) { 4347c478bd9Sstevel@tonic-gate hseg = as->a_lastgaphl; 4357c478bd9Sstevel@tonic-gate lseg = AVL_PREV(&as->a_segtree, hseg); 4367c478bd9Sstevel@tonic-gate } else { 4377c478bd9Sstevel@tonic-gate lseg = as->a_lastgaphl; 4387c478bd9Sstevel@tonic-gate hseg = AVL_NEXT(&as->a_segtree, lseg); 4397c478bd9Sstevel@tonic-gate } 4407c478bd9Sstevel@tonic-gate 4417c478bd9Sstevel@tonic-gate if (hseg && lseg && lseg->s_base < newseg->s_base && 4427c478bd9Sstevel@tonic-gate hseg->s_base > newseg->s_base) { 4437c478bd9Sstevel@tonic-gate avl_insert_here(&as->a_segtree, newseg, lseg, 4447c478bd9Sstevel@tonic-gate AVL_AFTER); 4457c478bd9Sstevel@tonic-gate as->a_lastgaphl = NULL; 4467c478bd9Sstevel@tonic-gate as->a_seglast = newseg; 4477c478bd9Sstevel@tonic-gate return (0); 4487c478bd9Sstevel@tonic-gate } 4497c478bd9Sstevel@tonic-gate as->a_lastgaphl = NULL; 4507c478bd9Sstevel@tonic-gate } 4517c478bd9Sstevel@tonic-gate 4527c478bd9Sstevel@tonic-gate addr = newseg->s_base; 4537c478bd9Sstevel@tonic-gate eaddr = addr + newseg->s_size; 4547c478bd9Sstevel@tonic-gate again: 4557c478bd9Sstevel@tonic-gate 4567c478bd9Sstevel@tonic-gate seg = avl_find(&as->a_segtree, &addr, &where); 4577c478bd9Sstevel@tonic-gate 4587c478bd9Sstevel@tonic-gate if (seg == NULL) 4597c478bd9Sstevel@tonic-gate seg = avl_nearest(&as->a_segtree, where, AVL_AFTER); 4607c478bd9Sstevel@tonic-gate 4617c478bd9Sstevel@tonic-gate if (seg == NULL) 4627c478bd9Sstevel@tonic-gate seg = avl_last(&as->a_segtree); 4637c478bd9Sstevel@tonic-gate 4647c478bd9Sstevel@tonic-gate if (seg != NULL) { 4657c478bd9Sstevel@tonic-gate caddr_t base = seg->s_base; 4667c478bd9Sstevel@tonic-gate 4677c478bd9Sstevel@tonic-gate /* 4687c478bd9Sstevel@tonic-gate * If top of seg is below the requested address, then 4697c478bd9Sstevel@tonic-gate * the insertion point is at the end of the linked list, 4707c478bd9Sstevel@tonic-gate * and seg points to the tail of the list. Otherwise, 4717c478bd9Sstevel@tonic-gate * the insertion point is immediately before seg. 4727c478bd9Sstevel@tonic-gate */ 4737c478bd9Sstevel@tonic-gate if (base + seg->s_size > addr) { 4747c478bd9Sstevel@tonic-gate if (addr >= base || eaddr > base) { 4757c478bd9Sstevel@tonic-gate #ifdef __sparc 4767c478bd9Sstevel@tonic-gate extern struct seg_ops segnf_ops; 4777c478bd9Sstevel@tonic-gate 4787c478bd9Sstevel@tonic-gate /* 4797c478bd9Sstevel@tonic-gate * no-fault segs must disappear if overlaid. 4807c478bd9Sstevel@tonic-gate * XXX need new segment type so 4817c478bd9Sstevel@tonic-gate * we don't have to check s_ops 4827c478bd9Sstevel@tonic-gate */ 4837c478bd9Sstevel@tonic-gate if (seg->s_ops == &segnf_ops) { 4847c478bd9Sstevel@tonic-gate seg_unmap(seg); 4857c478bd9Sstevel@tonic-gate goto again; 4867c478bd9Sstevel@tonic-gate } 4877c478bd9Sstevel@tonic-gate #endif 4887c478bd9Sstevel@tonic-gate return (-1); /* overlapping segment */ 4897c478bd9Sstevel@tonic-gate } 4907c478bd9Sstevel@tonic-gate } 4917c478bd9Sstevel@tonic-gate } 4927c478bd9Sstevel@tonic-gate as->a_seglast = newseg; 4937c478bd9Sstevel@tonic-gate avl_insert(&as->a_segtree, newseg, where); 4947c478bd9Sstevel@tonic-gate 4957c478bd9Sstevel@tonic-gate #ifdef VERIFY_SEGLIST 4967c478bd9Sstevel@tonic-gate as_verify(as); 4977c478bd9Sstevel@tonic-gate #endif 4987c478bd9Sstevel@tonic-gate return (0); 4997c478bd9Sstevel@tonic-gate } 5007c478bd9Sstevel@tonic-gate 5017c478bd9Sstevel@tonic-gate struct seg * 5027c478bd9Sstevel@tonic-gate as_removeseg(struct as *as, struct seg *seg) 5037c478bd9Sstevel@tonic-gate { 5047c478bd9Sstevel@tonic-gate avl_tree_t *t; 5057c478bd9Sstevel@tonic-gate 5067c478bd9Sstevel@tonic-gate ASSERT(AS_WRITE_HELD(as, &as->a_lock)); 5077c478bd9Sstevel@tonic-gate 5087c478bd9Sstevel@tonic-gate as->a_updatedir = 1; /* inform /proc */ 5097c478bd9Sstevel@tonic-gate gethrestime(&as->a_updatetime); 5107c478bd9Sstevel@tonic-gate 5117c478bd9Sstevel@tonic-gate if (seg == NULL) 5127c478bd9Sstevel@tonic-gate return (NULL); 5137c478bd9Sstevel@tonic-gate 5147c478bd9Sstevel@tonic-gate t = &as->a_segtree; 5157c478bd9Sstevel@tonic-gate if (as->a_seglast == seg) 5167c478bd9Sstevel@tonic-gate as->a_seglast = NULL; 5177c478bd9Sstevel@tonic-gate as->a_lastgaphl = NULL; 5187c478bd9Sstevel@tonic-gate 5197c478bd9Sstevel@tonic-gate /* 5207c478bd9Sstevel@tonic-gate * if this segment is at an address higher than 5217c478bd9Sstevel@tonic-gate * a_lastgap, set a_lastgap to the next segment (NULL if last segment) 5227c478bd9Sstevel@tonic-gate */ 5237c478bd9Sstevel@tonic-gate if (as->a_lastgap && 5247c478bd9Sstevel@tonic-gate (seg == as->a_lastgap || seg->s_base > as->a_lastgap->s_base)) 5257c478bd9Sstevel@tonic-gate as->a_lastgap = AVL_NEXT(t, seg); 5267c478bd9Sstevel@tonic-gate 5277c478bd9Sstevel@tonic-gate /* 5287c478bd9Sstevel@tonic-gate * remove the segment from the seg tree 5297c478bd9Sstevel@tonic-gate */ 5307c478bd9Sstevel@tonic-gate avl_remove(t, seg); 5317c478bd9Sstevel@tonic-gate 5327c478bd9Sstevel@tonic-gate #ifdef VERIFY_SEGLIST 5337c478bd9Sstevel@tonic-gate as_verify(as); 5347c478bd9Sstevel@tonic-gate #endif 5357c478bd9Sstevel@tonic-gate return (seg); 5367c478bd9Sstevel@tonic-gate } 5377c478bd9Sstevel@tonic-gate 5387c478bd9Sstevel@tonic-gate /* 5397c478bd9Sstevel@tonic-gate * Find a segment containing addr. 5407c478bd9Sstevel@tonic-gate */ 5417c478bd9Sstevel@tonic-gate struct seg * 5427c478bd9Sstevel@tonic-gate as_segat(struct as *as, caddr_t addr) 5437c478bd9Sstevel@tonic-gate { 5447c478bd9Sstevel@tonic-gate struct seg *seg = as->a_seglast; 5457c478bd9Sstevel@tonic-gate 5467c478bd9Sstevel@tonic-gate ASSERT(AS_LOCK_HELD(as, &as->a_lock)); 5477c478bd9Sstevel@tonic-gate 5487c478bd9Sstevel@tonic-gate if (seg != NULL && seg->s_base <= addr && 5497c478bd9Sstevel@tonic-gate addr < seg->s_base + seg->s_size) 5507c478bd9Sstevel@tonic-gate return (seg); 5517c478bd9Sstevel@tonic-gate 5527c478bd9Sstevel@tonic-gate seg = avl_find(&as->a_segtree, &addr, NULL); 5537c478bd9Sstevel@tonic-gate return (seg); 5547c478bd9Sstevel@tonic-gate } 5557c478bd9Sstevel@tonic-gate 5567c478bd9Sstevel@tonic-gate /* 5577c478bd9Sstevel@tonic-gate * Serialize all searches for holes in an address space to 5587c478bd9Sstevel@tonic-gate * prevent two or more threads from allocating the same virtual 5597c478bd9Sstevel@tonic-gate * address range. The address space must not be "read/write" 5607c478bd9Sstevel@tonic-gate * locked by the caller since we may block. 5617c478bd9Sstevel@tonic-gate */ 5627c478bd9Sstevel@tonic-gate void 5637c478bd9Sstevel@tonic-gate as_rangelock(struct as *as) 5647c478bd9Sstevel@tonic-gate { 5657c478bd9Sstevel@tonic-gate mutex_enter(&as->a_contents); 5667c478bd9Sstevel@tonic-gate while (AS_ISCLAIMGAP(as)) 5677c478bd9Sstevel@tonic-gate cv_wait(&as->a_cv, &as->a_contents); 5687c478bd9Sstevel@tonic-gate AS_SETCLAIMGAP(as); 5697c478bd9Sstevel@tonic-gate mutex_exit(&as->a_contents); 5707c478bd9Sstevel@tonic-gate } 5717c478bd9Sstevel@tonic-gate 5727c478bd9Sstevel@tonic-gate /* 5737c478bd9Sstevel@tonic-gate * Release hold on a_state & AS_CLAIMGAP and signal any other blocked threads. 5747c478bd9Sstevel@tonic-gate */ 5757c478bd9Sstevel@tonic-gate void 5767c478bd9Sstevel@tonic-gate as_rangeunlock(struct as *as) 5777c478bd9Sstevel@tonic-gate { 5787c478bd9Sstevel@tonic-gate mutex_enter(&as->a_contents); 5797c478bd9Sstevel@tonic-gate AS_CLRCLAIMGAP(as); 5807c478bd9Sstevel@tonic-gate cv_signal(&as->a_cv); 5817c478bd9Sstevel@tonic-gate mutex_exit(&as->a_contents); 5827c478bd9Sstevel@tonic-gate } 5837c478bd9Sstevel@tonic-gate 5847c478bd9Sstevel@tonic-gate /* 5857c478bd9Sstevel@tonic-gate * compar segments (or just an address) by segment address range 5867c478bd9Sstevel@tonic-gate */ 5877c478bd9Sstevel@tonic-gate static int 5887c478bd9Sstevel@tonic-gate as_segcompar(const void *x, const void *y) 5897c478bd9Sstevel@tonic-gate { 5907c478bd9Sstevel@tonic-gate struct seg *a = (struct seg *)x; 5917c478bd9Sstevel@tonic-gate struct seg *b = (struct seg *)y; 5927c478bd9Sstevel@tonic-gate 5937c478bd9Sstevel@tonic-gate if (a->s_base < b->s_base) 5947c478bd9Sstevel@tonic-gate return (-1); 5957c478bd9Sstevel@tonic-gate if (a->s_base >= b->s_base + b->s_size) 5967c478bd9Sstevel@tonic-gate return (1); 5977c478bd9Sstevel@tonic-gate return (0); 5987c478bd9Sstevel@tonic-gate } 5997c478bd9Sstevel@tonic-gate 6007c478bd9Sstevel@tonic-gate 6017c478bd9Sstevel@tonic-gate void 6027c478bd9Sstevel@tonic-gate as_avlinit(struct as *as) 6037c478bd9Sstevel@tonic-gate { 6047c478bd9Sstevel@tonic-gate avl_create(&as->a_segtree, as_segcompar, sizeof (struct seg), 6057c478bd9Sstevel@tonic-gate offsetof(struct seg, s_tree)); 6067c478bd9Sstevel@tonic-gate avl_create(&as->a_wpage, wp_compare, sizeof (struct watched_page), 6077c478bd9Sstevel@tonic-gate offsetof(struct watched_page, wp_link)); 6087c478bd9Sstevel@tonic-gate } 6097c478bd9Sstevel@tonic-gate 6107c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 6117c478bd9Sstevel@tonic-gate static int 6127c478bd9Sstevel@tonic-gate as_constructor(void *buf, void *cdrarg, int kmflags) 6137c478bd9Sstevel@tonic-gate { 6147c478bd9Sstevel@tonic-gate struct as *as = buf; 6157c478bd9Sstevel@tonic-gate 6167c478bd9Sstevel@tonic-gate mutex_init(&as->a_contents, NULL, MUTEX_DEFAULT, NULL); 6177c478bd9Sstevel@tonic-gate cv_init(&as->a_cv, NULL, CV_DEFAULT, NULL); 6187c478bd9Sstevel@tonic-gate rw_init(&as->a_lock, NULL, RW_DEFAULT, NULL); 6197c478bd9Sstevel@tonic-gate as_avlinit(as); 6207c478bd9Sstevel@tonic-gate return (0); 6217c478bd9Sstevel@tonic-gate } 6227c478bd9Sstevel@tonic-gate 6237c478bd9Sstevel@tonic-gate /*ARGSUSED1*/ 6247c478bd9Sstevel@tonic-gate static void 6257c478bd9Sstevel@tonic-gate as_destructor(void *buf, void *cdrarg) 6267c478bd9Sstevel@tonic-gate { 6277c478bd9Sstevel@tonic-gate struct as *as = buf; 6287c478bd9Sstevel@tonic-gate 6297c478bd9Sstevel@tonic-gate avl_destroy(&as->a_segtree); 6307c478bd9Sstevel@tonic-gate mutex_destroy(&as->a_contents); 6317c478bd9Sstevel@tonic-gate cv_destroy(&as->a_cv); 6327c478bd9Sstevel@tonic-gate rw_destroy(&as->a_lock); 6337c478bd9Sstevel@tonic-gate } 6347c478bd9Sstevel@tonic-gate 6357c478bd9Sstevel@tonic-gate void 6367c478bd9Sstevel@tonic-gate as_init(void) 6377c478bd9Sstevel@tonic-gate { 6387c478bd9Sstevel@tonic-gate as_cache = kmem_cache_create("as_cache", sizeof (struct as), 0, 639843e1988Sjohnlev as_constructor, as_destructor, NULL, NULL, NULL, 0); 6407c478bd9Sstevel@tonic-gate } 6417c478bd9Sstevel@tonic-gate 6427c478bd9Sstevel@tonic-gate /* 6437c478bd9Sstevel@tonic-gate * Allocate and initialize an address space data structure. 6447c478bd9Sstevel@tonic-gate * We call hat_alloc to allow any machine dependent 6457c478bd9Sstevel@tonic-gate * information in the hat structure to be initialized. 6467c478bd9Sstevel@tonic-gate */ 6477c478bd9Sstevel@tonic-gate struct as * 6487c478bd9Sstevel@tonic-gate as_alloc(void) 6497c478bd9Sstevel@tonic-gate { 6507c478bd9Sstevel@tonic-gate struct as *as; 6517c478bd9Sstevel@tonic-gate 6527c478bd9Sstevel@tonic-gate as = kmem_cache_alloc(as_cache, KM_SLEEP); 6537c478bd9Sstevel@tonic-gate 6547c478bd9Sstevel@tonic-gate as->a_flags = 0; 6557c478bd9Sstevel@tonic-gate as->a_vbits = 0; 6567c478bd9Sstevel@tonic-gate as->a_hrm = NULL; 6577c478bd9Sstevel@tonic-gate as->a_seglast = NULL; 6587c478bd9Sstevel@tonic-gate as->a_size = 0; 65940688216SSudheer A as->a_resvsize = 0; 6607c478bd9Sstevel@tonic-gate as->a_updatedir = 0; 6617c478bd9Sstevel@tonic-gate gethrestime(&as->a_updatetime); 6627c478bd9Sstevel@tonic-gate as->a_objectdir = NULL; 6637c478bd9Sstevel@tonic-gate as->a_sizedir = 0; 6647c478bd9Sstevel@tonic-gate as->a_userlimit = (caddr_t)USERLIMIT; 6657c478bd9Sstevel@tonic-gate as->a_lastgap = NULL; 6667c478bd9Sstevel@tonic-gate as->a_lastgaphl = NULL; 6677c478bd9Sstevel@tonic-gate as->a_callbacks = NULL; 6687c478bd9Sstevel@tonic-gate 6697c478bd9Sstevel@tonic-gate AS_LOCK_ENTER(as, &as->a_lock, RW_WRITER); 6707c478bd9Sstevel@tonic-gate as->a_hat = hat_alloc(as); /* create hat for default system mmu */ 6717c478bd9Sstevel@tonic-gate AS_LOCK_EXIT(as, &as->a_lock); 6727c478bd9Sstevel@tonic-gate 6737c478bd9Sstevel@tonic-gate as->a_xhat = NULL; 6747c478bd9Sstevel@tonic-gate 6757c478bd9Sstevel@tonic-gate return (as); 6767c478bd9Sstevel@tonic-gate } 6777c478bd9Sstevel@tonic-gate 6787c478bd9Sstevel@tonic-gate /* 6797c478bd9Sstevel@tonic-gate * Free an address space data structure. 6807c478bd9Sstevel@tonic-gate * Need to free the hat first and then 6817c478bd9Sstevel@tonic-gate * all the segments on this as and finally 6827c478bd9Sstevel@tonic-gate * the space for the as struct itself. 6837c478bd9Sstevel@tonic-gate */ 6847c478bd9Sstevel@tonic-gate void 6857c478bd9Sstevel@tonic-gate as_free(struct as *as) 6867c478bd9Sstevel@tonic-gate { 6877c478bd9Sstevel@tonic-gate struct hat *hat = as->a_hat; 6887c478bd9Sstevel@tonic-gate struct seg *seg, *next; 6897c478bd9Sstevel@tonic-gate int called = 0; 6907c478bd9Sstevel@tonic-gate 6917c478bd9Sstevel@tonic-gate top: 6927c478bd9Sstevel@tonic-gate /* 6937c478bd9Sstevel@tonic-gate * Invoke ALL callbacks. as_do_callbacks will do one callback 6947c478bd9Sstevel@tonic-gate * per call, and not return (-1) until the callback has completed. 6957c478bd9Sstevel@tonic-gate * When as_do_callbacks returns zero, all callbacks have completed. 6967c478bd9Sstevel@tonic-gate */ 6977c478bd9Sstevel@tonic-gate mutex_enter(&as->a_contents); 698843e1988Sjohnlev while (as->a_callbacks && as_do_callbacks(as, AS_ALL_EVENT, 0, 0)) 699843e1988Sjohnlev ; 7007c478bd9Sstevel@tonic-gate 7017c478bd9Sstevel@tonic-gate /* This will prevent new XHATs from attaching to as */ 7027c478bd9Sstevel@tonic-gate if (!called) 7037c478bd9Sstevel@tonic-gate AS_SETBUSY(as); 7047c478bd9Sstevel@tonic-gate mutex_exit(&as->a_contents); 7057c478bd9Sstevel@tonic-gate AS_LOCK_ENTER(as, &as->a_lock, RW_WRITER); 7067c478bd9Sstevel@tonic-gate 7077c478bd9Sstevel@tonic-gate if (!called) { 7087c478bd9Sstevel@tonic-gate called = 1; 7097c478bd9Sstevel@tonic-gate hat_free_start(hat); 7107c478bd9Sstevel@tonic-gate if (as->a_xhat != NULL) 7117c478bd9Sstevel@tonic-gate xhat_free_start_all(as); 7127c478bd9Sstevel@tonic-gate } 7137c478bd9Sstevel@tonic-gate for (seg = AS_SEGFIRST(as); seg != NULL; seg = next) { 7147c478bd9Sstevel@tonic-gate int err; 7157c478bd9Sstevel@tonic-gate 7167c478bd9Sstevel@tonic-gate next = AS_SEGNEXT(as, seg); 717a98e9dbfSaguzovsk retry: 7187c478bd9Sstevel@tonic-gate err = SEGOP_UNMAP(seg, seg->s_base, seg->s_size); 7197c478bd9Sstevel@tonic-gate if (err == EAGAIN) { 7207c478bd9Sstevel@tonic-gate mutex_enter(&as->a_contents); 7217c478bd9Sstevel@tonic-gate if (as->a_callbacks) { 7227c478bd9Sstevel@tonic-gate AS_LOCK_EXIT(as, &as->a_lock); 723a98e9dbfSaguzovsk } else if (!AS_ISNOUNMAPWAIT(as)) { 7247c478bd9Sstevel@tonic-gate /* 7257c478bd9Sstevel@tonic-gate * Memory is currently locked. Wait for a 7267c478bd9Sstevel@tonic-gate * cv_signal that it has been unlocked, then 7277c478bd9Sstevel@tonic-gate * try the operation again. 7287c478bd9Sstevel@tonic-gate */ 7297c478bd9Sstevel@tonic-gate if (AS_ISUNMAPWAIT(as) == 0) 7307c478bd9Sstevel@tonic-gate cv_broadcast(&as->a_cv); 7317c478bd9Sstevel@tonic-gate AS_SETUNMAPWAIT(as); 7327c478bd9Sstevel@tonic-gate AS_LOCK_EXIT(as, &as->a_lock); 7337c478bd9Sstevel@tonic-gate while (AS_ISUNMAPWAIT(as)) 7347c478bd9Sstevel@tonic-gate cv_wait(&as->a_cv, &as->a_contents); 735a98e9dbfSaguzovsk } else { 736a98e9dbfSaguzovsk /* 737a98e9dbfSaguzovsk * We may have raced with 738a98e9dbfSaguzovsk * segvn_reclaim()/segspt_reclaim(). In this 739a98e9dbfSaguzovsk * case clean nounmapwait flag and retry since 740a98e9dbfSaguzovsk * softlockcnt in this segment may be already 741a98e9dbfSaguzovsk * 0. We don't drop as writer lock so our 742a98e9dbfSaguzovsk * number of retries without sleeping should 743a98e9dbfSaguzovsk * be very small. See segvn_reclaim() for 744a98e9dbfSaguzovsk * more comments. 745a98e9dbfSaguzovsk */ 746a98e9dbfSaguzovsk AS_CLRNOUNMAPWAIT(as); 747a98e9dbfSaguzovsk mutex_exit(&as->a_contents); 748a98e9dbfSaguzovsk goto retry; 7497c478bd9Sstevel@tonic-gate } 7507c478bd9Sstevel@tonic-gate mutex_exit(&as->a_contents); 7517c478bd9Sstevel@tonic-gate goto top; 7527c478bd9Sstevel@tonic-gate } else { 7537c478bd9Sstevel@tonic-gate /* 7547c478bd9Sstevel@tonic-gate * We do not expect any other error return at this 7557c478bd9Sstevel@tonic-gate * time. This is similar to an ASSERT in seg_unmap() 7567c478bd9Sstevel@tonic-gate */ 7577c478bd9Sstevel@tonic-gate ASSERT(err == 0); 7587c478bd9Sstevel@tonic-gate } 7597c478bd9Sstevel@tonic-gate } 7607c478bd9Sstevel@tonic-gate hat_free_end(hat); 7617c478bd9Sstevel@tonic-gate if (as->a_xhat != NULL) 7627c478bd9Sstevel@tonic-gate xhat_free_end_all(as); 7637c478bd9Sstevel@tonic-gate AS_LOCK_EXIT(as, &as->a_lock); 7647c478bd9Sstevel@tonic-gate 7657c478bd9Sstevel@tonic-gate /* /proc stuff */ 7667c478bd9Sstevel@tonic-gate ASSERT(avl_numnodes(&as->a_wpage) == 0); 7677c478bd9Sstevel@tonic-gate if (as->a_objectdir) { 7687c478bd9Sstevel@tonic-gate kmem_free(as->a_objectdir, as->a_sizedir * sizeof (vnode_t *)); 7697c478bd9Sstevel@tonic-gate as->a_objectdir = NULL; 7707c478bd9Sstevel@tonic-gate as->a_sizedir = 0; 7717c478bd9Sstevel@tonic-gate } 7727c478bd9Sstevel@tonic-gate 7737c478bd9Sstevel@tonic-gate /* 7747c478bd9Sstevel@tonic-gate * Free the struct as back to kmem. Assert it has no segments. 7757c478bd9Sstevel@tonic-gate */ 7767c478bd9Sstevel@tonic-gate ASSERT(avl_numnodes(&as->a_segtree) == 0); 7777c478bd9Sstevel@tonic-gate kmem_cache_free(as_cache, as); 7787c478bd9Sstevel@tonic-gate } 7797c478bd9Sstevel@tonic-gate 7807c478bd9Sstevel@tonic-gate int 781bb5ca623SVamsi Nagineni as_dup(struct as *as, struct proc *forkedproc) 7827c478bd9Sstevel@tonic-gate { 7837c478bd9Sstevel@tonic-gate struct as *newas; 7847c478bd9Sstevel@tonic-gate struct seg *seg, *newseg; 78540688216SSudheer A size_t purgesize = 0; 7867c478bd9Sstevel@tonic-gate int error; 7877c478bd9Sstevel@tonic-gate 7887c478bd9Sstevel@tonic-gate AS_LOCK_ENTER(as, &as->a_lock, RW_WRITER); 7897c478bd9Sstevel@tonic-gate as_clearwatch(as); 7907c478bd9Sstevel@tonic-gate newas = as_alloc(); 7917c478bd9Sstevel@tonic-gate newas->a_userlimit = as->a_userlimit; 792bb5ca623SVamsi Nagineni newas->a_proc = forkedproc; 793250f6acdSvk 7947c478bd9Sstevel@tonic-gate AS_LOCK_ENTER(newas, &newas->a_lock, RW_WRITER); 7957c478bd9Sstevel@tonic-gate 7967c478bd9Sstevel@tonic-gate /* This will prevent new XHATs from attaching */ 7977c478bd9Sstevel@tonic-gate mutex_enter(&as->a_contents); 7987c478bd9Sstevel@tonic-gate AS_SETBUSY(as); 7997c478bd9Sstevel@tonic-gate mutex_exit(&as->a_contents); 8007c478bd9Sstevel@tonic-gate mutex_enter(&newas->a_contents); 8017c478bd9Sstevel@tonic-gate AS_SETBUSY(newas); 8027c478bd9Sstevel@tonic-gate mutex_exit(&newas->a_contents); 8037c478bd9Sstevel@tonic-gate 80405d3dc4bSpaulsan (void) hat_dup(as->a_hat, newas->a_hat, NULL, 0, HAT_DUP_SRD); 8057c478bd9Sstevel@tonic-gate 8067c478bd9Sstevel@tonic-gate for (seg = AS_SEGFIRST(as); seg != NULL; seg = AS_SEGNEXT(as, seg)) { 8077c478bd9Sstevel@tonic-gate 80840688216SSudheer A if (seg->s_flags & S_PURGE) { 80940688216SSudheer A purgesize += seg->s_size; 8107c478bd9Sstevel@tonic-gate continue; 81140688216SSudheer A } 8127c478bd9Sstevel@tonic-gate 8137c478bd9Sstevel@tonic-gate newseg = seg_alloc(newas, seg->s_base, seg->s_size); 8147c478bd9Sstevel@tonic-gate if (newseg == NULL) { 8157c478bd9Sstevel@tonic-gate AS_LOCK_EXIT(newas, &newas->a_lock); 8167c478bd9Sstevel@tonic-gate as_setwatch(as); 8177c478bd9Sstevel@tonic-gate mutex_enter(&as->a_contents); 8187c478bd9Sstevel@tonic-gate AS_CLRBUSY(as); 8197c478bd9Sstevel@tonic-gate mutex_exit(&as->a_contents); 8207c478bd9Sstevel@tonic-gate AS_LOCK_EXIT(as, &as->a_lock); 8217c478bd9Sstevel@tonic-gate as_free(newas); 8227c478bd9Sstevel@tonic-gate return (-1); 8237c478bd9Sstevel@tonic-gate } 8247c478bd9Sstevel@tonic-gate if ((error = SEGOP_DUP(seg, newseg)) != 0) { 8257c478bd9Sstevel@tonic-gate /* 8267c478bd9Sstevel@tonic-gate * We call seg_free() on the new seg 8277c478bd9Sstevel@tonic-gate * because the segment is not set up 8287c478bd9Sstevel@tonic-gate * completely; i.e. it has no ops. 8297c478bd9Sstevel@tonic-gate */ 8307c478bd9Sstevel@tonic-gate as_setwatch(as); 8317c478bd9Sstevel@tonic-gate mutex_enter(&as->a_contents); 8327c478bd9Sstevel@tonic-gate AS_CLRBUSY(as); 8337c478bd9Sstevel@tonic-gate mutex_exit(&as->a_contents); 8347c478bd9Sstevel@tonic-gate AS_LOCK_EXIT(as, &as->a_lock); 8357c478bd9Sstevel@tonic-gate seg_free(newseg); 8367c478bd9Sstevel@tonic-gate AS_LOCK_EXIT(newas, &newas->a_lock); 8377c478bd9Sstevel@tonic-gate as_free(newas); 8387c478bd9Sstevel@tonic-gate return (error); 8397c478bd9Sstevel@tonic-gate } 8407c478bd9Sstevel@tonic-gate newas->a_size += seg->s_size; 8417c478bd9Sstevel@tonic-gate } 84240688216SSudheer A newas->a_resvsize = as->a_resvsize - purgesize; 8437c478bd9Sstevel@tonic-gate 8447c478bd9Sstevel@tonic-gate error = hat_dup(as->a_hat, newas->a_hat, NULL, 0, HAT_DUP_ALL); 8457c478bd9Sstevel@tonic-gate if (as->a_xhat != NULL) 8467c478bd9Sstevel@tonic-gate error |= xhat_dup_all(as, newas, NULL, 0, HAT_DUP_ALL); 8477c478bd9Sstevel@tonic-gate 8487c478bd9Sstevel@tonic-gate mutex_enter(&newas->a_contents); 8497c478bd9Sstevel@tonic-gate AS_CLRBUSY(newas); 8507c478bd9Sstevel@tonic-gate mutex_exit(&newas->a_contents); 8517c478bd9Sstevel@tonic-gate AS_LOCK_EXIT(newas, &newas->a_lock); 8527c478bd9Sstevel@tonic-gate 8537c478bd9Sstevel@tonic-gate as_setwatch(as); 8547c478bd9Sstevel@tonic-gate mutex_enter(&as->a_contents); 8557c478bd9Sstevel@tonic-gate AS_CLRBUSY(as); 8567c478bd9Sstevel@tonic-gate mutex_exit(&as->a_contents); 8577c478bd9Sstevel@tonic-gate AS_LOCK_EXIT(as, &as->a_lock); 8587c478bd9Sstevel@tonic-gate if (error != 0) { 8597c478bd9Sstevel@tonic-gate as_free(newas); 8607c478bd9Sstevel@tonic-gate return (error); 8617c478bd9Sstevel@tonic-gate } 862bb5ca623SVamsi Nagineni forkedproc->p_as = newas; 8637c478bd9Sstevel@tonic-gate return (0); 8647c478bd9Sstevel@tonic-gate } 8657c478bd9Sstevel@tonic-gate 8667c478bd9Sstevel@tonic-gate /* 8677c478bd9Sstevel@tonic-gate * Handle a ``fault'' at addr for size bytes. 8687c478bd9Sstevel@tonic-gate */ 8697c478bd9Sstevel@tonic-gate faultcode_t 8707c478bd9Sstevel@tonic-gate as_fault(struct hat *hat, struct as *as, caddr_t addr, size_t size, 8717c478bd9Sstevel@tonic-gate enum fault_type type, enum seg_rw rw) 8727c478bd9Sstevel@tonic-gate { 8737c478bd9Sstevel@tonic-gate struct seg *seg; 8747c478bd9Sstevel@tonic-gate caddr_t raddr; /* rounded down addr */ 8757c478bd9Sstevel@tonic-gate size_t rsize; /* rounded up size */ 8767c478bd9Sstevel@tonic-gate size_t ssize; 8777c478bd9Sstevel@tonic-gate faultcode_t res = 0; 8787c478bd9Sstevel@tonic-gate caddr_t addrsav; 8797c478bd9Sstevel@tonic-gate struct seg *segsav; 8807c478bd9Sstevel@tonic-gate int as_lock_held; 8817c478bd9Sstevel@tonic-gate klwp_t *lwp = ttolwp(curthread); 8827c478bd9Sstevel@tonic-gate int is_xhat = 0; 8837c478bd9Sstevel@tonic-gate int holding_wpage = 0; 8847c478bd9Sstevel@tonic-gate extern struct seg_ops segdev_ops; 8857c478bd9Sstevel@tonic-gate 8867c478bd9Sstevel@tonic-gate 8877c478bd9Sstevel@tonic-gate 8887c478bd9Sstevel@tonic-gate if (as->a_hat != hat) { 8897c478bd9Sstevel@tonic-gate /* This must be an XHAT then */ 8907c478bd9Sstevel@tonic-gate is_xhat = 1; 8917c478bd9Sstevel@tonic-gate 8927c478bd9Sstevel@tonic-gate if ((type != F_INVAL) || (as == &kas)) 8937c478bd9Sstevel@tonic-gate return (FC_NOSUPPORT); 8947c478bd9Sstevel@tonic-gate } 8957c478bd9Sstevel@tonic-gate 8967c478bd9Sstevel@tonic-gate retry: 8977c478bd9Sstevel@tonic-gate if (!is_xhat) { 8987c478bd9Sstevel@tonic-gate /* 8997c478bd9Sstevel@tonic-gate * Indicate that the lwp is not to be stopped while waiting 9007c478bd9Sstevel@tonic-gate * for a pagefault. This is to avoid deadlock while debugging 9017c478bd9Sstevel@tonic-gate * a process via /proc over NFS (in particular). 9027c478bd9Sstevel@tonic-gate */ 9038548bf79Snr if (lwp != NULL) 9047c478bd9Sstevel@tonic-gate lwp->lwp_nostop++; 9057c478bd9Sstevel@tonic-gate 9067c478bd9Sstevel@tonic-gate /* 9077c478bd9Sstevel@tonic-gate * same length must be used when we softlock and softunlock. 9087c478bd9Sstevel@tonic-gate * We don't support softunlocking lengths less than 9097c478bd9Sstevel@tonic-gate * the original length when there is largepage support. 9107c478bd9Sstevel@tonic-gate * See seg_dev.c for more comments. 9117c478bd9Sstevel@tonic-gate */ 9127c478bd9Sstevel@tonic-gate switch (type) { 9137c478bd9Sstevel@tonic-gate 9147c478bd9Sstevel@tonic-gate case F_SOFTLOCK: 9157c478bd9Sstevel@tonic-gate CPU_STATS_ADD_K(vm, softlock, 1); 9167c478bd9Sstevel@tonic-gate break; 9177c478bd9Sstevel@tonic-gate 9187c478bd9Sstevel@tonic-gate case F_SOFTUNLOCK: 9197c478bd9Sstevel@tonic-gate break; 9207c478bd9Sstevel@tonic-gate 9217c478bd9Sstevel@tonic-gate case F_PROT: 9227c478bd9Sstevel@tonic-gate CPU_STATS_ADD_K(vm, prot_fault, 1); 9237c478bd9Sstevel@tonic-gate break; 9247c478bd9Sstevel@tonic-gate 9257c478bd9Sstevel@tonic-gate case F_INVAL: 9267c478bd9Sstevel@tonic-gate CPU_STATS_ENTER_K(); 9277c478bd9Sstevel@tonic-gate CPU_STATS_ADDQ(CPU, vm, as_fault, 1); 9287c478bd9Sstevel@tonic-gate if (as == &kas) 9297c478bd9Sstevel@tonic-gate CPU_STATS_ADDQ(CPU, vm, kernel_asflt, 1); 9307c478bd9Sstevel@tonic-gate CPU_STATS_EXIT_K(); 9317c478bd9Sstevel@tonic-gate break; 9327c478bd9Sstevel@tonic-gate } 9337c478bd9Sstevel@tonic-gate } 9347c478bd9Sstevel@tonic-gate 9357c478bd9Sstevel@tonic-gate /* Kernel probe */ 9367c478bd9Sstevel@tonic-gate TNF_PROBE_3(address_fault, "vm pagefault", /* CSTYLED */, 937843e1988Sjohnlev tnf_opaque, address, addr, 938843e1988Sjohnlev tnf_fault_type, fault_type, type, 939843e1988Sjohnlev tnf_seg_access, access, rw); 9407c478bd9Sstevel@tonic-gate 9417c478bd9Sstevel@tonic-gate raddr = (caddr_t)((uintptr_t)addr & (uintptr_t)PAGEMASK); 9427c478bd9Sstevel@tonic-gate rsize = (((size_t)(addr + size) + PAGEOFFSET) & PAGEMASK) - 943843e1988Sjohnlev (size_t)raddr; 9447c478bd9Sstevel@tonic-gate 9457c478bd9Sstevel@tonic-gate /* 9467c478bd9Sstevel@tonic-gate * XXX -- Don't grab the as lock for segkmap. We should grab it for 9477c478bd9Sstevel@tonic-gate * correctness, but then we could be stuck holding this lock for 9487c478bd9Sstevel@tonic-gate * a LONG time if the fault needs to be resolved on a slow 9497c478bd9Sstevel@tonic-gate * filesystem, and then no-one will be able to exec new commands, 9507c478bd9Sstevel@tonic-gate * as exec'ing requires the write lock on the as. 9517c478bd9Sstevel@tonic-gate */ 9527c478bd9Sstevel@tonic-gate if (as == &kas && segkmap && segkmap->s_base <= raddr && 9537c478bd9Sstevel@tonic-gate raddr + size < segkmap->s_base + segkmap->s_size) { 9547c478bd9Sstevel@tonic-gate /* 9557c478bd9Sstevel@tonic-gate * if (as==&kas), this can't be XHAT: we've already returned 9567c478bd9Sstevel@tonic-gate * FC_NOSUPPORT. 9577c478bd9Sstevel@tonic-gate */ 9587c478bd9Sstevel@tonic-gate seg = segkmap; 9597c478bd9Sstevel@tonic-gate as_lock_held = 0; 9607c478bd9Sstevel@tonic-gate } else { 9617c478bd9Sstevel@tonic-gate AS_LOCK_ENTER(as, &as->a_lock, RW_READER); 9627c478bd9Sstevel@tonic-gate if (is_xhat && avl_numnodes(&as->a_wpage) != 0) { 9637c478bd9Sstevel@tonic-gate /* 9647c478bd9Sstevel@tonic-gate * Grab and hold the writers' lock on the as 9657c478bd9Sstevel@tonic-gate * if the fault is to a watched page. 9667c478bd9Sstevel@tonic-gate * This will keep CPUs from "peeking" at the 9677c478bd9Sstevel@tonic-gate * address range while we're temporarily boosting 9687c478bd9Sstevel@tonic-gate * the permissions for the XHAT device to 9697c478bd9Sstevel@tonic-gate * resolve the fault in the segment layer. 9707c478bd9Sstevel@tonic-gate * 9717c478bd9Sstevel@tonic-gate * We could check whether faulted address 9727c478bd9Sstevel@tonic-gate * is within a watched page and only then grab 9737c478bd9Sstevel@tonic-gate * the writer lock, but this is simpler. 9747c478bd9Sstevel@tonic-gate */ 9757c478bd9Sstevel@tonic-gate AS_LOCK_EXIT(as, &as->a_lock); 9767c478bd9Sstevel@tonic-gate AS_LOCK_ENTER(as, &as->a_lock, RW_WRITER); 9777c478bd9Sstevel@tonic-gate } 9787c478bd9Sstevel@tonic-gate 9797c478bd9Sstevel@tonic-gate seg = as_segat(as, raddr); 9807c478bd9Sstevel@tonic-gate if (seg == NULL) { 9817c478bd9Sstevel@tonic-gate AS_LOCK_EXIT(as, &as->a_lock); 9828548bf79Snr if ((lwp != NULL) && (!is_xhat)) 9837c478bd9Sstevel@tonic-gate lwp->lwp_nostop--; 9847c478bd9Sstevel@tonic-gate return (FC_NOMAP); 9857c478bd9Sstevel@tonic-gate } 9867c478bd9Sstevel@tonic-gate 9877c478bd9Sstevel@tonic-gate as_lock_held = 1; 9887c478bd9Sstevel@tonic-gate } 9897c478bd9Sstevel@tonic-gate 9907c478bd9Sstevel@tonic-gate addrsav = raddr; 9917c478bd9Sstevel@tonic-gate segsav = seg; 9927c478bd9Sstevel@tonic-gate 9937c478bd9Sstevel@tonic-gate for (; rsize != 0; rsize -= ssize, raddr += ssize) { 9947c478bd9Sstevel@tonic-gate if (raddr >= seg->s_base + seg->s_size) { 9957c478bd9Sstevel@tonic-gate seg = AS_SEGNEXT(as, seg); 9967c478bd9Sstevel@tonic-gate if (seg == NULL || raddr != seg->s_base) { 9977c478bd9Sstevel@tonic-gate res = FC_NOMAP; 9987c478bd9Sstevel@tonic-gate break; 9997c478bd9Sstevel@tonic-gate } 10007c478bd9Sstevel@tonic-gate } 10017c478bd9Sstevel@tonic-gate if (raddr + rsize > seg->s_base + seg->s_size) 10027c478bd9Sstevel@tonic-gate ssize = seg->s_base + seg->s_size - raddr; 10037c478bd9Sstevel@tonic-gate else 10047c478bd9Sstevel@tonic-gate ssize = rsize; 10057c478bd9Sstevel@tonic-gate 10067c478bd9Sstevel@tonic-gate if (!is_xhat || (seg->s_ops != &segdev_ops)) { 10077c478bd9Sstevel@tonic-gate 10087c478bd9Sstevel@tonic-gate if (is_xhat && avl_numnodes(&as->a_wpage) != 0 && 10097c478bd9Sstevel@tonic-gate pr_is_watchpage_as(raddr, rw, as)) { 10107c478bd9Sstevel@tonic-gate /* 10117c478bd9Sstevel@tonic-gate * Handle watch pages. If we're faulting on a 10127c478bd9Sstevel@tonic-gate * watched page from an X-hat, we have to 10137c478bd9Sstevel@tonic-gate * restore the original permissions while we 10147c478bd9Sstevel@tonic-gate * handle the fault. 10157c478bd9Sstevel@tonic-gate */ 10167c478bd9Sstevel@tonic-gate as_clearwatch(as); 10177c478bd9Sstevel@tonic-gate holding_wpage = 1; 10187c478bd9Sstevel@tonic-gate } 10197c478bd9Sstevel@tonic-gate 10207c478bd9Sstevel@tonic-gate res = SEGOP_FAULT(hat, seg, raddr, ssize, type, rw); 10217c478bd9Sstevel@tonic-gate 10227c478bd9Sstevel@tonic-gate /* Restore watchpoints */ 10237c478bd9Sstevel@tonic-gate if (holding_wpage) { 10247c478bd9Sstevel@tonic-gate as_setwatch(as); 10257c478bd9Sstevel@tonic-gate holding_wpage = 0; 10267c478bd9Sstevel@tonic-gate } 10277c478bd9Sstevel@tonic-gate 10287c478bd9Sstevel@tonic-gate if (res != 0) 10297c478bd9Sstevel@tonic-gate break; 10307c478bd9Sstevel@tonic-gate } else { 10317c478bd9Sstevel@tonic-gate /* XHAT does not support seg_dev */ 10327c478bd9Sstevel@tonic-gate res = FC_NOSUPPORT; 10337c478bd9Sstevel@tonic-gate break; 10347c478bd9Sstevel@tonic-gate } 10357c478bd9Sstevel@tonic-gate } 10367c478bd9Sstevel@tonic-gate 10377c478bd9Sstevel@tonic-gate /* 10387c478bd9Sstevel@tonic-gate * If we were SOFTLOCKing and encountered a failure, 10397c478bd9Sstevel@tonic-gate * we must SOFTUNLOCK the range we already did. (Maybe we 10407c478bd9Sstevel@tonic-gate * should just panic if we are SOFTLOCKing or even SOFTUNLOCKing 10417c478bd9Sstevel@tonic-gate * right here...) 10427c478bd9Sstevel@tonic-gate */ 10437c478bd9Sstevel@tonic-gate if (res != 0 && type == F_SOFTLOCK) { 10447c478bd9Sstevel@tonic-gate for (seg = segsav; addrsav < raddr; addrsav += ssize) { 10457c478bd9Sstevel@tonic-gate if (addrsav >= seg->s_base + seg->s_size) 10467c478bd9Sstevel@tonic-gate seg = AS_SEGNEXT(as, seg); 10477c478bd9Sstevel@tonic-gate ASSERT(seg != NULL); 10487c478bd9Sstevel@tonic-gate /* 10497c478bd9Sstevel@tonic-gate * Now call the fault routine again to perform the 10507c478bd9Sstevel@tonic-gate * unlock using S_OTHER instead of the rw variable 10517c478bd9Sstevel@tonic-gate * since we never got a chance to touch the pages. 10527c478bd9Sstevel@tonic-gate */ 10537c478bd9Sstevel@tonic-gate if (raddr > seg->s_base + seg->s_size) 10547c478bd9Sstevel@tonic-gate ssize = seg->s_base + seg->s_size - addrsav; 10557c478bd9Sstevel@tonic-gate else 10567c478bd9Sstevel@tonic-gate ssize = raddr - addrsav; 10577c478bd9Sstevel@tonic-gate (void) SEGOP_FAULT(hat, seg, addrsav, ssize, 10587c478bd9Sstevel@tonic-gate F_SOFTUNLOCK, S_OTHER); 10597c478bd9Sstevel@tonic-gate } 10607c478bd9Sstevel@tonic-gate } 10617c478bd9Sstevel@tonic-gate if (as_lock_held) 10627c478bd9Sstevel@tonic-gate AS_LOCK_EXIT(as, &as->a_lock); 10638548bf79Snr if ((lwp != NULL) && (!is_xhat)) 10647c478bd9Sstevel@tonic-gate lwp->lwp_nostop--; 10658548bf79Snr 10667c478bd9Sstevel@tonic-gate /* 10677c478bd9Sstevel@tonic-gate * If the lower levels returned EDEADLK for a fault, 10687c478bd9Sstevel@tonic-gate * It means that we should retry the fault. Let's wait 10697c478bd9Sstevel@tonic-gate * a bit also to let the deadlock causing condition clear. 10707c478bd9Sstevel@tonic-gate * This is part of a gross hack to work around a design flaw 10717c478bd9Sstevel@tonic-gate * in the ufs/sds logging code and should go away when the 10727c478bd9Sstevel@tonic-gate * logging code is re-designed to fix the problem. See bug 10737c478bd9Sstevel@tonic-gate * 4125102 for details of the problem. 10747c478bd9Sstevel@tonic-gate */ 10757c478bd9Sstevel@tonic-gate if (FC_ERRNO(res) == EDEADLK) { 10767c478bd9Sstevel@tonic-gate delay(deadlk_wait); 10777c478bd9Sstevel@tonic-gate res = 0; 10787c478bd9Sstevel@tonic-gate goto retry; 10797c478bd9Sstevel@tonic-gate } 10807c478bd9Sstevel@tonic-gate return (res); 10817c478bd9Sstevel@tonic-gate } 10827c478bd9Sstevel@tonic-gate 10837c478bd9Sstevel@tonic-gate 10847c478bd9Sstevel@tonic-gate 10857c478bd9Sstevel@tonic-gate /* 10867c478bd9Sstevel@tonic-gate * Asynchronous ``fault'' at addr for size bytes. 10877c478bd9Sstevel@tonic-gate */ 10887c478bd9Sstevel@tonic-gate faultcode_t 10897c478bd9Sstevel@tonic-gate as_faulta(struct as *as, caddr_t addr, size_t size) 10907c478bd9Sstevel@tonic-gate { 10917c478bd9Sstevel@tonic-gate struct seg *seg; 10927c478bd9Sstevel@tonic-gate caddr_t raddr; /* rounded down addr */ 10937c478bd9Sstevel@tonic-gate size_t rsize; /* rounded up size */ 10947c478bd9Sstevel@tonic-gate faultcode_t res = 0; 10957c478bd9Sstevel@tonic-gate klwp_t *lwp = ttolwp(curthread); 10967c478bd9Sstevel@tonic-gate 10977c478bd9Sstevel@tonic-gate retry: 10987c478bd9Sstevel@tonic-gate /* 10997c478bd9Sstevel@tonic-gate * Indicate that the lwp is not to be stopped while waiting 11007c478bd9Sstevel@tonic-gate * for a pagefault. This is to avoid deadlock while debugging 11017c478bd9Sstevel@tonic-gate * a process via /proc over NFS (in particular). 11027c478bd9Sstevel@tonic-gate */ 11038548bf79Snr if (lwp != NULL) 11047c478bd9Sstevel@tonic-gate lwp->lwp_nostop++; 11057c478bd9Sstevel@tonic-gate 11067c478bd9Sstevel@tonic-gate raddr = (caddr_t)((uintptr_t)addr & (uintptr_t)PAGEMASK); 11077c478bd9Sstevel@tonic-gate rsize = (((size_t)(addr + size) + PAGEOFFSET) & PAGEMASK) - 1108843e1988Sjohnlev (size_t)raddr; 11097c478bd9Sstevel@tonic-gate 11107c478bd9Sstevel@tonic-gate AS_LOCK_ENTER(as, &as->a_lock, RW_READER); 11117c478bd9Sstevel@tonic-gate seg = as_segat(as, raddr); 11127c478bd9Sstevel@tonic-gate if (seg == NULL) { 11137c478bd9Sstevel@tonic-gate AS_LOCK_EXIT(as, &as->a_lock); 11148548bf79Snr if (lwp != NULL) 11157c478bd9Sstevel@tonic-gate lwp->lwp_nostop--; 11167c478bd9Sstevel@tonic-gate return (FC_NOMAP); 11177c478bd9Sstevel@tonic-gate } 11187c478bd9Sstevel@tonic-gate 11197c478bd9Sstevel@tonic-gate for (; rsize != 0; rsize -= PAGESIZE, raddr += PAGESIZE) { 11207c478bd9Sstevel@tonic-gate if (raddr >= seg->s_base + seg->s_size) { 11217c478bd9Sstevel@tonic-gate seg = AS_SEGNEXT(as, seg); 11227c478bd9Sstevel@tonic-gate if (seg == NULL || raddr != seg->s_base) { 11237c478bd9Sstevel@tonic-gate res = FC_NOMAP; 11247c478bd9Sstevel@tonic-gate break; 11257c478bd9Sstevel@tonic-gate } 11267c478bd9Sstevel@tonic-gate } 11277c478bd9Sstevel@tonic-gate res = SEGOP_FAULTA(seg, raddr); 11287c478bd9Sstevel@tonic-gate if (res != 0) 11297c478bd9Sstevel@tonic-gate break; 11307c478bd9Sstevel@tonic-gate } 11317c478bd9Sstevel@tonic-gate AS_LOCK_EXIT(as, &as->a_lock); 11328548bf79Snr if (lwp != NULL) 11337c478bd9Sstevel@tonic-gate lwp->lwp_nostop--; 11347c478bd9Sstevel@tonic-gate /* 11357c478bd9Sstevel@tonic-gate * If the lower levels returned EDEADLK for a fault, 11367c478bd9Sstevel@tonic-gate * It means that we should retry the fault. Let's wait 11377c478bd9Sstevel@tonic-gate * a bit also to let the deadlock causing condition clear. 11387c478bd9Sstevel@tonic-gate * This is part of a gross hack to work around a design flaw 11397c478bd9Sstevel@tonic-gate * in the ufs/sds logging code and should go away when the 11407c478bd9Sstevel@tonic-gate * logging code is re-designed to fix the problem. See bug 11417c478bd9Sstevel@tonic-gate * 4125102 for details of the problem. 11427c478bd9Sstevel@tonic-gate */ 11437c478bd9Sstevel@tonic-gate if (FC_ERRNO(res) == EDEADLK) { 11447c478bd9Sstevel@tonic-gate delay(deadlk_wait); 11457c478bd9Sstevel@tonic-gate res = 0; 11467c478bd9Sstevel@tonic-gate goto retry; 11477c478bd9Sstevel@tonic-gate } 11487c478bd9Sstevel@tonic-gate return (res); 11497c478bd9Sstevel@tonic-gate } 11507c478bd9Sstevel@tonic-gate 11517c478bd9Sstevel@tonic-gate /* 11527c478bd9Sstevel@tonic-gate * Set the virtual mapping for the interval from [addr : addr + size) 11537c478bd9Sstevel@tonic-gate * in address space `as' to have the specified protection. 11547c478bd9Sstevel@tonic-gate * It is ok for the range to cross over several segments, 11557c478bd9Sstevel@tonic-gate * as long as they are contiguous. 11567c478bd9Sstevel@tonic-gate */ 11577c478bd9Sstevel@tonic-gate int 11587c478bd9Sstevel@tonic-gate as_setprot(struct as *as, caddr_t addr, size_t size, uint_t prot) 11597c478bd9Sstevel@tonic-gate { 11607c478bd9Sstevel@tonic-gate struct seg *seg; 11617c478bd9Sstevel@tonic-gate struct as_callback *cb; 11627c478bd9Sstevel@tonic-gate size_t ssize; 11637c478bd9Sstevel@tonic-gate caddr_t raddr; /* rounded down addr */ 11647c478bd9Sstevel@tonic-gate size_t rsize; /* rounded up size */ 11657c478bd9Sstevel@tonic-gate int error = 0, writer = 0; 11667c478bd9Sstevel@tonic-gate caddr_t saveraddr; 11677c478bd9Sstevel@tonic-gate size_t saversize; 11687c478bd9Sstevel@tonic-gate 11697c478bd9Sstevel@tonic-gate setprot_top: 11707c478bd9Sstevel@tonic-gate raddr = (caddr_t)((uintptr_t)addr & (uintptr_t)PAGEMASK); 11717c478bd9Sstevel@tonic-gate rsize = (((size_t)(addr + size) + PAGEOFFSET) & PAGEMASK) - 1172843e1988Sjohnlev (size_t)raddr; 11737c478bd9Sstevel@tonic-gate 11747c478bd9Sstevel@tonic-gate if (raddr + rsize < raddr) /* check for wraparound */ 11757c478bd9Sstevel@tonic-gate return (ENOMEM); 11767c478bd9Sstevel@tonic-gate 11777c478bd9Sstevel@tonic-gate saveraddr = raddr; 11787c478bd9Sstevel@tonic-gate saversize = rsize; 11797c478bd9Sstevel@tonic-gate 11807c478bd9Sstevel@tonic-gate /* 11817c478bd9Sstevel@tonic-gate * Normally we only lock the as as a reader. But 11827c478bd9Sstevel@tonic-gate * if due to setprot the segment driver needs to split 1183da6c28aaSamw * a segment it will return IE_RETRY. Therefore we re-acquire 11847c478bd9Sstevel@tonic-gate * the as lock as a writer so the segment driver can change 11857c478bd9Sstevel@tonic-gate * the seg list. Also the segment driver will return IE_RETRY 11867c478bd9Sstevel@tonic-gate * after it has changed the segment list so we therefore keep 11877c478bd9Sstevel@tonic-gate * locking as a writer. Since these opeartions should be rare 11887c478bd9Sstevel@tonic-gate * want to only lock as a writer when necessary. 11897c478bd9Sstevel@tonic-gate */ 11907c478bd9Sstevel@tonic-gate if (writer || avl_numnodes(&as->a_wpage) != 0) { 11917c478bd9Sstevel@tonic-gate AS_LOCK_ENTER(as, &as->a_lock, RW_WRITER); 11927c478bd9Sstevel@tonic-gate } else { 11937c478bd9Sstevel@tonic-gate AS_LOCK_ENTER(as, &as->a_lock, RW_READER); 11947c478bd9Sstevel@tonic-gate } 11957c478bd9Sstevel@tonic-gate 11967c478bd9Sstevel@tonic-gate as_clearwatchprot(as, raddr, rsize); 11977c478bd9Sstevel@tonic-gate seg = as_segat(as, raddr); 11987c478bd9Sstevel@tonic-gate if (seg == NULL) { 11997c478bd9Sstevel@tonic-gate as_setwatch(as); 12007c478bd9Sstevel@tonic-gate AS_LOCK_EXIT(as, &as->a_lock); 12017c478bd9Sstevel@tonic-gate return (ENOMEM); 12027c478bd9Sstevel@tonic-gate } 12037c478bd9Sstevel@tonic-gate 12047c478bd9Sstevel@tonic-gate for (; rsize != 0; rsize -= ssize, raddr += ssize) { 12057c478bd9Sstevel@tonic-gate if (raddr >= seg->s_base + seg->s_size) { 12067c478bd9Sstevel@tonic-gate seg = AS_SEGNEXT(as, seg); 12077c478bd9Sstevel@tonic-gate if (seg == NULL || raddr != seg->s_base) { 12087c478bd9Sstevel@tonic-gate error = ENOMEM; 12097c478bd9Sstevel@tonic-gate break; 12107c478bd9Sstevel@tonic-gate } 12117c478bd9Sstevel@tonic-gate } 12127c478bd9Sstevel@tonic-gate if ((raddr + rsize) > (seg->s_base + seg->s_size)) 12137c478bd9Sstevel@tonic-gate ssize = seg->s_base + seg->s_size - raddr; 12147c478bd9Sstevel@tonic-gate else 12157c478bd9Sstevel@tonic-gate ssize = rsize; 1216a98e9dbfSaguzovsk retry: 12177c478bd9Sstevel@tonic-gate error = SEGOP_SETPROT(seg, raddr, ssize, prot); 12187c478bd9Sstevel@tonic-gate 12197c478bd9Sstevel@tonic-gate if (error == IE_NOMEM) { 12207c478bd9Sstevel@tonic-gate error = EAGAIN; 12217c478bd9Sstevel@tonic-gate break; 12227c478bd9Sstevel@tonic-gate } 12237c478bd9Sstevel@tonic-gate 12247c478bd9Sstevel@tonic-gate if (error == IE_RETRY) { 12257c478bd9Sstevel@tonic-gate AS_LOCK_EXIT(as, &as->a_lock); 12267c478bd9Sstevel@tonic-gate writer = 1; 12277c478bd9Sstevel@tonic-gate goto setprot_top; 12287c478bd9Sstevel@tonic-gate } 12297c478bd9Sstevel@tonic-gate 12307c478bd9Sstevel@tonic-gate if (error == EAGAIN) { 12317c478bd9Sstevel@tonic-gate /* 12327c478bd9Sstevel@tonic-gate * Make sure we have a_lock as writer. 12337c478bd9Sstevel@tonic-gate */ 12347c478bd9Sstevel@tonic-gate if (writer == 0) { 12357c478bd9Sstevel@tonic-gate AS_LOCK_EXIT(as, &as->a_lock); 12367c478bd9Sstevel@tonic-gate writer = 1; 12377c478bd9Sstevel@tonic-gate goto setprot_top; 12387c478bd9Sstevel@tonic-gate } 12397c478bd9Sstevel@tonic-gate 12407c478bd9Sstevel@tonic-gate /* 12417c478bd9Sstevel@tonic-gate * Memory is currently locked. It must be unlocked 12427c478bd9Sstevel@tonic-gate * before this operation can succeed through a retry. 12437c478bd9Sstevel@tonic-gate * The possible reasons for locked memory and 12447c478bd9Sstevel@tonic-gate * corresponding strategies for unlocking are: 12457c478bd9Sstevel@tonic-gate * (1) Normal I/O 12467c478bd9Sstevel@tonic-gate * wait for a signal that the I/O operation 12477c478bd9Sstevel@tonic-gate * has completed and the memory is unlocked. 12487c478bd9Sstevel@tonic-gate * (2) Asynchronous I/O 12497c478bd9Sstevel@tonic-gate * The aio subsystem does not unlock pages when 12507c478bd9Sstevel@tonic-gate * the I/O is completed. Those pages are unlocked 12517c478bd9Sstevel@tonic-gate * when the application calls aiowait/aioerror. 12527c478bd9Sstevel@tonic-gate * So, to prevent blocking forever, cv_broadcast() 12537c478bd9Sstevel@tonic-gate * is done to wake up aio_cleanup_thread. 12547c478bd9Sstevel@tonic-gate * Subsequently, segvn_reclaim will be called, and 12557c478bd9Sstevel@tonic-gate * that will do AS_CLRUNMAPWAIT() and wake us up. 12567c478bd9Sstevel@tonic-gate * (3) Long term page locking: 12577c478bd9Sstevel@tonic-gate * Drivers intending to have pages locked for a 12587c478bd9Sstevel@tonic-gate * period considerably longer than for normal I/O 12597c478bd9Sstevel@tonic-gate * (essentially forever) may have registered for a 12607c478bd9Sstevel@tonic-gate * callback so they may unlock these pages on 12617c478bd9Sstevel@tonic-gate * request. This is needed to allow this operation 12627c478bd9Sstevel@tonic-gate * to succeed. Each entry on the callback list is 12637c478bd9Sstevel@tonic-gate * examined. If the event or address range pertains 12647c478bd9Sstevel@tonic-gate * the callback is invoked (unless it already is in 12657c478bd9Sstevel@tonic-gate * progress). The a_contents lock must be dropped 12667c478bd9Sstevel@tonic-gate * before the callback, so only one callback can 12677c478bd9Sstevel@tonic-gate * be done at a time. Go to the top and do more 12687c478bd9Sstevel@tonic-gate * until zero is returned. If zero is returned, 12697c478bd9Sstevel@tonic-gate * either there were no callbacks for this event 12707c478bd9Sstevel@tonic-gate * or they were already in progress. 12717c478bd9Sstevel@tonic-gate */ 12727c478bd9Sstevel@tonic-gate mutex_enter(&as->a_contents); 12737c478bd9Sstevel@tonic-gate if (as->a_callbacks && 1274843e1988Sjohnlev (cb = as_find_callback(as, AS_SETPROT_EVENT, 1275843e1988Sjohnlev seg->s_base, seg->s_size))) { 12767c478bd9Sstevel@tonic-gate AS_LOCK_EXIT(as, &as->a_lock); 12777c478bd9Sstevel@tonic-gate as_execute_callback(as, cb, AS_SETPROT_EVENT); 1278a98e9dbfSaguzovsk } else if (!AS_ISNOUNMAPWAIT(as)) { 12797c478bd9Sstevel@tonic-gate if (AS_ISUNMAPWAIT(as) == 0) 12807c478bd9Sstevel@tonic-gate cv_broadcast(&as->a_cv); 12817c478bd9Sstevel@tonic-gate AS_SETUNMAPWAIT(as); 12827c478bd9Sstevel@tonic-gate AS_LOCK_EXIT(as, &as->a_lock); 12837c478bd9Sstevel@tonic-gate while (AS_ISUNMAPWAIT(as)) 12847c478bd9Sstevel@tonic-gate cv_wait(&as->a_cv, &as->a_contents); 1285a98e9dbfSaguzovsk } else { 1286a98e9dbfSaguzovsk /* 1287a98e9dbfSaguzovsk * We may have raced with 1288a98e9dbfSaguzovsk * segvn_reclaim()/segspt_reclaim(). In this 1289a98e9dbfSaguzovsk * case clean nounmapwait flag and retry since 1290a98e9dbfSaguzovsk * softlockcnt in this segment may be already 1291a98e9dbfSaguzovsk * 0. We don't drop as writer lock so our 1292a98e9dbfSaguzovsk * number of retries without sleeping should 1293a98e9dbfSaguzovsk * be very small. See segvn_reclaim() for 1294a98e9dbfSaguzovsk * more comments. 1295a98e9dbfSaguzovsk */ 1296a98e9dbfSaguzovsk AS_CLRNOUNMAPWAIT(as); 1297a98e9dbfSaguzovsk mutex_exit(&as->a_contents); 1298a98e9dbfSaguzovsk goto retry; 12997c478bd9Sstevel@tonic-gate } 13007c478bd9Sstevel@tonic-gate mutex_exit(&as->a_contents); 13017c478bd9Sstevel@tonic-gate goto setprot_top; 13027c478bd9Sstevel@tonic-gate } else if (error != 0) 13037c478bd9Sstevel@tonic-gate break; 13047c478bd9Sstevel@tonic-gate } 13057c478bd9Sstevel@tonic-gate if (error != 0) { 13067c478bd9Sstevel@tonic-gate as_setwatch(as); 13077c478bd9Sstevel@tonic-gate } else { 13087c478bd9Sstevel@tonic-gate as_setwatchprot(as, saveraddr, saversize, prot); 13097c478bd9Sstevel@tonic-gate } 13107c478bd9Sstevel@tonic-gate AS_LOCK_EXIT(as, &as->a_lock); 13117c478bd9Sstevel@tonic-gate return (error); 13127c478bd9Sstevel@tonic-gate } 13137c478bd9Sstevel@tonic-gate 13147c478bd9Sstevel@tonic-gate /* 13157c478bd9Sstevel@tonic-gate * Check to make sure that the interval [addr, addr + size) 13167c478bd9Sstevel@tonic-gate * in address space `as' has at least the specified protection. 13177c478bd9Sstevel@tonic-gate * It is ok for the range to cross over several segments, as long 13187c478bd9Sstevel@tonic-gate * as they are contiguous. 13197c478bd9Sstevel@tonic-gate */ 13207c478bd9Sstevel@tonic-gate int 13217c478bd9Sstevel@tonic-gate as_checkprot(struct as *as, caddr_t addr, size_t size, uint_t prot) 13227c478bd9Sstevel@tonic-gate { 13237c478bd9Sstevel@tonic-gate struct seg *seg; 13247c478bd9Sstevel@tonic-gate size_t ssize; 13257c478bd9Sstevel@tonic-gate caddr_t raddr; /* rounded down addr */ 13267c478bd9Sstevel@tonic-gate size_t rsize; /* rounded up size */ 13277c478bd9Sstevel@tonic-gate int error = 0; 13287c478bd9Sstevel@tonic-gate 13297c478bd9Sstevel@tonic-gate raddr = (caddr_t)((uintptr_t)addr & (uintptr_t)PAGEMASK); 13307c478bd9Sstevel@tonic-gate rsize = (((size_t)(addr + size) + PAGEOFFSET) & PAGEMASK) - 1331843e1988Sjohnlev (size_t)raddr; 13327c478bd9Sstevel@tonic-gate 13337c478bd9Sstevel@tonic-gate if (raddr + rsize < raddr) /* check for wraparound */ 13347c478bd9Sstevel@tonic-gate return (ENOMEM); 13357c478bd9Sstevel@tonic-gate 13367c478bd9Sstevel@tonic-gate /* 13377c478bd9Sstevel@tonic-gate * This is ugly as sin... 13387c478bd9Sstevel@tonic-gate * Normally, we only acquire the address space readers lock. 13397c478bd9Sstevel@tonic-gate * However, if the address space has watchpoints present, 13407c478bd9Sstevel@tonic-gate * we must acquire the writer lock on the address space for 13417c478bd9Sstevel@tonic-gate * the benefit of as_clearwatchprot() and as_setwatchprot(). 13427c478bd9Sstevel@tonic-gate */ 13437c478bd9Sstevel@tonic-gate if (avl_numnodes(&as->a_wpage) != 0) 13447c478bd9Sstevel@tonic-gate AS_LOCK_ENTER(as, &as->a_lock, RW_WRITER); 13457c478bd9Sstevel@tonic-gate else 13467c478bd9Sstevel@tonic-gate AS_LOCK_ENTER(as, &as->a_lock, RW_READER); 13477c478bd9Sstevel@tonic-gate as_clearwatchprot(as, raddr, rsize); 13487c478bd9Sstevel@tonic-gate seg = as_segat(as, raddr); 13497c478bd9Sstevel@tonic-gate if (seg == NULL) { 13507c478bd9Sstevel@tonic-gate as_setwatch(as); 13517c478bd9Sstevel@tonic-gate AS_LOCK_EXIT(as, &as->a_lock); 13527c478bd9Sstevel@tonic-gate return (ENOMEM); 13537c478bd9Sstevel@tonic-gate } 13547c478bd9Sstevel@tonic-gate 13557c478bd9Sstevel@tonic-gate for (; rsize != 0; rsize -= ssize, raddr += ssize) { 13567c478bd9Sstevel@tonic-gate if (raddr >= seg->s_base + seg->s_size) { 13577c478bd9Sstevel@tonic-gate seg = AS_SEGNEXT(as, seg); 13587c478bd9Sstevel@tonic-gate if (seg == NULL || raddr != seg->s_base) { 13597c478bd9Sstevel@tonic-gate error = ENOMEM; 13607c478bd9Sstevel@tonic-gate break; 13617c478bd9Sstevel@tonic-gate } 13627c478bd9Sstevel@tonic-gate } 13637c478bd9Sstevel@tonic-gate if ((raddr + rsize) > (seg->s_base + seg->s_size)) 13647c478bd9Sstevel@tonic-gate ssize = seg->s_base + seg->s_size - raddr; 13657c478bd9Sstevel@tonic-gate else 13667c478bd9Sstevel@tonic-gate ssize = rsize; 13677c478bd9Sstevel@tonic-gate 13687c478bd9Sstevel@tonic-gate error = SEGOP_CHECKPROT(seg, raddr, ssize, prot); 13697c478bd9Sstevel@tonic-gate if (error != 0) 13707c478bd9Sstevel@tonic-gate break; 13717c478bd9Sstevel@tonic-gate } 13727c478bd9Sstevel@tonic-gate as_setwatch(as); 13737c478bd9Sstevel@tonic-gate AS_LOCK_EXIT(as, &as->a_lock); 13747c478bd9Sstevel@tonic-gate return (error); 13757c478bd9Sstevel@tonic-gate } 13767c478bd9Sstevel@tonic-gate 13777c478bd9Sstevel@tonic-gate int 13787c478bd9Sstevel@tonic-gate as_unmap(struct as *as, caddr_t addr, size_t size) 13797c478bd9Sstevel@tonic-gate { 13807c478bd9Sstevel@tonic-gate struct seg *seg, *seg_next; 13817c478bd9Sstevel@tonic-gate struct as_callback *cb; 13827c478bd9Sstevel@tonic-gate caddr_t raddr, eaddr; 138340688216SSudheer A size_t ssize, rsize = 0; 13847c478bd9Sstevel@tonic-gate int err; 13857c478bd9Sstevel@tonic-gate 13867c478bd9Sstevel@tonic-gate top: 13877c478bd9Sstevel@tonic-gate raddr = (caddr_t)((uintptr_t)addr & (uintptr_t)PAGEMASK); 13887c478bd9Sstevel@tonic-gate eaddr = (caddr_t)(((uintptr_t)(addr + size) + PAGEOFFSET) & 13897c478bd9Sstevel@tonic-gate (uintptr_t)PAGEMASK); 13907c478bd9Sstevel@tonic-gate 13917c478bd9Sstevel@tonic-gate AS_LOCK_ENTER(as, &as->a_lock, RW_WRITER); 13927c478bd9Sstevel@tonic-gate 13937c478bd9Sstevel@tonic-gate as->a_updatedir = 1; /* inform /proc */ 13947c478bd9Sstevel@tonic-gate gethrestime(&as->a_updatetime); 13957c478bd9Sstevel@tonic-gate 13967c478bd9Sstevel@tonic-gate /* 13977c478bd9Sstevel@tonic-gate * Use as_findseg to find the first segment in the range, then 13987c478bd9Sstevel@tonic-gate * step through the segments in order, following s_next. 13997c478bd9Sstevel@tonic-gate */ 14007c478bd9Sstevel@tonic-gate as_clearwatchprot(as, raddr, eaddr - raddr); 14017c478bd9Sstevel@tonic-gate 14027c478bd9Sstevel@tonic-gate for (seg = as_findseg(as, raddr, 0); seg != NULL; seg = seg_next) { 14037c478bd9Sstevel@tonic-gate if (eaddr <= seg->s_base) 14047c478bd9Sstevel@tonic-gate break; /* eaddr was in a gap; all done */ 14057c478bd9Sstevel@tonic-gate 14067c478bd9Sstevel@tonic-gate /* this is implied by the test above */ 14077c478bd9Sstevel@tonic-gate ASSERT(raddr < eaddr); 14087c478bd9Sstevel@tonic-gate 14097c478bd9Sstevel@tonic-gate if (raddr < seg->s_base) 14107c478bd9Sstevel@tonic-gate raddr = seg->s_base; /* raddr was in a gap */ 14117c478bd9Sstevel@tonic-gate 14127c478bd9Sstevel@tonic-gate if (eaddr > (seg->s_base + seg->s_size)) 14137c478bd9Sstevel@tonic-gate ssize = seg->s_base + seg->s_size - raddr; 14147c478bd9Sstevel@tonic-gate else 14157c478bd9Sstevel@tonic-gate ssize = eaddr - raddr; 14167c478bd9Sstevel@tonic-gate 14177c478bd9Sstevel@tonic-gate /* 14187c478bd9Sstevel@tonic-gate * Save next segment pointer since seg can be 14197c478bd9Sstevel@tonic-gate * destroyed during the segment unmap operation. 14207c478bd9Sstevel@tonic-gate */ 14217c478bd9Sstevel@tonic-gate seg_next = AS_SEGNEXT(as, seg); 14227c478bd9Sstevel@tonic-gate 142340688216SSudheer A /* 142440688216SSudheer A * We didn't count /dev/null mappings, so ignore them here. 142540688216SSudheer A * We'll handle MAP_NORESERVE cases in segvn_unmap(). (Again, 142640688216SSudheer A * we have to do this check here while we have seg.) 142740688216SSudheer A */ 1428*17965fd8SKrishnendu Sadhukhan - Sun Microsystems rsize = 0; 142940688216SSudheer A if (!SEG_IS_DEVNULL_MAPPING(seg) && 143040688216SSudheer A !SEG_IS_PARTIAL_RESV(seg)) 143140688216SSudheer A rsize = ssize; 143240688216SSudheer A 1433a98e9dbfSaguzovsk retry: 14347c478bd9Sstevel@tonic-gate err = SEGOP_UNMAP(seg, raddr, ssize); 14357c478bd9Sstevel@tonic-gate if (err == EAGAIN) { 14367c478bd9Sstevel@tonic-gate /* 14377c478bd9Sstevel@tonic-gate * Memory is currently locked. It must be unlocked 14387c478bd9Sstevel@tonic-gate * before this operation can succeed through a retry. 14397c478bd9Sstevel@tonic-gate * The possible reasons for locked memory and 14407c478bd9Sstevel@tonic-gate * corresponding strategies for unlocking are: 14417c478bd9Sstevel@tonic-gate * (1) Normal I/O 14427c478bd9Sstevel@tonic-gate * wait for a signal that the I/O operation 14437c478bd9Sstevel@tonic-gate * has completed and the memory is unlocked. 14447c478bd9Sstevel@tonic-gate * (2) Asynchronous I/O 14457c478bd9Sstevel@tonic-gate * The aio subsystem does not unlock pages when 14467c478bd9Sstevel@tonic-gate * the I/O is completed. Those pages are unlocked 14477c478bd9Sstevel@tonic-gate * when the application calls aiowait/aioerror. 14487c478bd9Sstevel@tonic-gate * So, to prevent blocking forever, cv_broadcast() 14497c478bd9Sstevel@tonic-gate * is done to wake up aio_cleanup_thread. 14507c478bd9Sstevel@tonic-gate * Subsequently, segvn_reclaim will be called, and 14517c478bd9Sstevel@tonic-gate * that will do AS_CLRUNMAPWAIT() and wake us up. 14527c478bd9Sstevel@tonic-gate * (3) Long term page locking: 14537c478bd9Sstevel@tonic-gate * Drivers intending to have pages locked for a 14547c478bd9Sstevel@tonic-gate * period considerably longer than for normal I/O 14557c478bd9Sstevel@tonic-gate * (essentially forever) may have registered for a 14567c478bd9Sstevel@tonic-gate * callback so they may unlock these pages on 14577c478bd9Sstevel@tonic-gate * request. This is needed to allow this operation 14587c478bd9Sstevel@tonic-gate * to succeed. Each entry on the callback list is 14597c478bd9Sstevel@tonic-gate * examined. If the event or address range pertains 14607c478bd9Sstevel@tonic-gate * the callback is invoked (unless it already is in 14617c478bd9Sstevel@tonic-gate * progress). The a_contents lock must be dropped 14627c478bd9Sstevel@tonic-gate * before the callback, so only one callback can 14637c478bd9Sstevel@tonic-gate * be done at a time. Go to the top and do more 14647c478bd9Sstevel@tonic-gate * until zero is returned. If zero is returned, 14657c478bd9Sstevel@tonic-gate * either there were no callbacks for this event 14667c478bd9Sstevel@tonic-gate * or they were already in progress. 14677c478bd9Sstevel@tonic-gate */ 14687c478bd9Sstevel@tonic-gate mutex_enter(&as->a_contents); 14697c478bd9Sstevel@tonic-gate if (as->a_callbacks && 1470843e1988Sjohnlev (cb = as_find_callback(as, AS_UNMAP_EVENT, 1471843e1988Sjohnlev seg->s_base, seg->s_size))) { 14727c478bd9Sstevel@tonic-gate AS_LOCK_EXIT(as, &as->a_lock); 14737c478bd9Sstevel@tonic-gate as_execute_callback(as, cb, AS_UNMAP_EVENT); 1474a98e9dbfSaguzovsk } else if (!AS_ISNOUNMAPWAIT(as)) { 14757c478bd9Sstevel@tonic-gate if (AS_ISUNMAPWAIT(as) == 0) 14767c478bd9Sstevel@tonic-gate cv_broadcast(&as->a_cv); 14777c478bd9Sstevel@tonic-gate AS_SETUNMAPWAIT(as); 14787c478bd9Sstevel@tonic-gate AS_LOCK_EXIT(as, &as->a_lock); 14797c478bd9Sstevel@tonic-gate while (AS_ISUNMAPWAIT(as)) 14807c478bd9Sstevel@tonic-gate cv_wait(&as->a_cv, &as->a_contents); 1481a98e9dbfSaguzovsk } else { 1482a98e9dbfSaguzovsk /* 1483a98e9dbfSaguzovsk * We may have raced with 1484a98e9dbfSaguzovsk * segvn_reclaim()/segspt_reclaim(). In this 1485a98e9dbfSaguzovsk * case clean nounmapwait flag and retry since 1486a98e9dbfSaguzovsk * softlockcnt in this segment may be already 1487a98e9dbfSaguzovsk * 0. We don't drop as writer lock so our 1488a98e9dbfSaguzovsk * number of retries without sleeping should 1489a98e9dbfSaguzovsk * be very small. See segvn_reclaim() for 1490a98e9dbfSaguzovsk * more comments. 1491a98e9dbfSaguzovsk */ 1492a98e9dbfSaguzovsk AS_CLRNOUNMAPWAIT(as); 1493a98e9dbfSaguzovsk mutex_exit(&as->a_contents); 1494a98e9dbfSaguzovsk goto retry; 14957c478bd9Sstevel@tonic-gate } 14967c478bd9Sstevel@tonic-gate mutex_exit(&as->a_contents); 14977c478bd9Sstevel@tonic-gate goto top; 14987c478bd9Sstevel@tonic-gate } else if (err == IE_RETRY) { 14997c478bd9Sstevel@tonic-gate AS_LOCK_EXIT(as, &as->a_lock); 15007c478bd9Sstevel@tonic-gate goto top; 15017c478bd9Sstevel@tonic-gate } else if (err) { 15027c478bd9Sstevel@tonic-gate as_setwatch(as); 15037c478bd9Sstevel@tonic-gate AS_LOCK_EXIT(as, &as->a_lock); 15047c478bd9Sstevel@tonic-gate return (-1); 15057c478bd9Sstevel@tonic-gate } 15067c478bd9Sstevel@tonic-gate 15077c478bd9Sstevel@tonic-gate as->a_size -= ssize; 1508*17965fd8SKrishnendu Sadhukhan - Sun Microsystems if (rsize) 1509*17965fd8SKrishnendu Sadhukhan - Sun Microsystems as->a_resvsize -= rsize; 15107c478bd9Sstevel@tonic-gate raddr += ssize; 15117c478bd9Sstevel@tonic-gate } 15127c478bd9Sstevel@tonic-gate AS_LOCK_EXIT(as, &as->a_lock); 15137c478bd9Sstevel@tonic-gate return (0); 15147c478bd9Sstevel@tonic-gate } 15157c478bd9Sstevel@tonic-gate 15167c478bd9Sstevel@tonic-gate static int 151707b65a64Saguzovsk as_map_segvn_segs(struct as *as, caddr_t addr, size_t size, uint_t szcvec, 15187c478bd9Sstevel@tonic-gate int (*crfp)(), struct segvn_crargs *vn_a, int *segcreated) 15197c478bd9Sstevel@tonic-gate { 15207c478bd9Sstevel@tonic-gate uint_t szc; 15217c478bd9Sstevel@tonic-gate uint_t nszc; 15227c478bd9Sstevel@tonic-gate int error; 15237c478bd9Sstevel@tonic-gate caddr_t a; 15247c478bd9Sstevel@tonic-gate caddr_t eaddr; 15257c478bd9Sstevel@tonic-gate size_t segsize; 15267c478bd9Sstevel@tonic-gate struct seg *seg; 15277c478bd9Sstevel@tonic-gate size_t pgsz; 152807b65a64Saguzovsk int do_off = (vn_a->vp != NULL || vn_a->amp != NULL); 152907b65a64Saguzovsk uint_t save_szcvec; 15307c478bd9Sstevel@tonic-gate 15317c478bd9Sstevel@tonic-gate ASSERT(AS_WRITE_HELD(as, &as->a_lock)); 15327c478bd9Sstevel@tonic-gate ASSERT(IS_P2ALIGNED(addr, PAGESIZE)); 15337c478bd9Sstevel@tonic-gate ASSERT(IS_P2ALIGNED(size, PAGESIZE)); 153407b65a64Saguzovsk ASSERT(vn_a->vp == NULL || vn_a->amp == NULL); 153507b65a64Saguzovsk if (!do_off) { 153607b65a64Saguzovsk vn_a->offset = 0; 153707b65a64Saguzovsk } 15387c478bd9Sstevel@tonic-gate 15397c478bd9Sstevel@tonic-gate if (szcvec <= 1) { 15407c478bd9Sstevel@tonic-gate seg = seg_alloc(as, addr, size); 15417c478bd9Sstevel@tonic-gate if (seg == NULL) { 15427c478bd9Sstevel@tonic-gate return (ENOMEM); 15437c478bd9Sstevel@tonic-gate } 15447c478bd9Sstevel@tonic-gate vn_a->szc = 0; 15457c478bd9Sstevel@tonic-gate error = (*crfp)(seg, vn_a); 15467c478bd9Sstevel@tonic-gate if (error != 0) { 15477c478bd9Sstevel@tonic-gate seg_free(seg); 15483a30c6acSsusans } else { 15493a30c6acSsusans as->a_size += size; 1550*17965fd8SKrishnendu Sadhukhan - Sun Microsystems as->a_resvsize += size; 15517c478bd9Sstevel@tonic-gate } 15527c478bd9Sstevel@tonic-gate return (error); 15537c478bd9Sstevel@tonic-gate } 15547c478bd9Sstevel@tonic-gate 15557c478bd9Sstevel@tonic-gate eaddr = addr + size; 15567c478bd9Sstevel@tonic-gate save_szcvec = szcvec; 15577c478bd9Sstevel@tonic-gate szcvec >>= 1; 15587c478bd9Sstevel@tonic-gate szc = 0; 15597c478bd9Sstevel@tonic-gate nszc = 0; 15607c478bd9Sstevel@tonic-gate while (szcvec) { 15617c478bd9Sstevel@tonic-gate if ((szcvec & 0x1) == 0) { 15627c478bd9Sstevel@tonic-gate nszc++; 15637c478bd9Sstevel@tonic-gate szcvec >>= 1; 15647c478bd9Sstevel@tonic-gate continue; 15657c478bd9Sstevel@tonic-gate } 15667c478bd9Sstevel@tonic-gate nszc++; 15677c478bd9Sstevel@tonic-gate pgsz = page_get_pagesize(nszc); 15687c478bd9Sstevel@tonic-gate a = (caddr_t)P2ROUNDUP((uintptr_t)addr, pgsz); 15697c478bd9Sstevel@tonic-gate if (a != addr) { 15707c478bd9Sstevel@tonic-gate ASSERT(a < eaddr); 15717c478bd9Sstevel@tonic-gate segsize = a - addr; 15727c478bd9Sstevel@tonic-gate seg = seg_alloc(as, addr, segsize); 15737c478bd9Sstevel@tonic-gate if (seg == NULL) { 15747c478bd9Sstevel@tonic-gate return (ENOMEM); 15757c478bd9Sstevel@tonic-gate } 15767c478bd9Sstevel@tonic-gate vn_a->szc = szc; 15777c478bd9Sstevel@tonic-gate error = (*crfp)(seg, vn_a); 15787c478bd9Sstevel@tonic-gate if (error != 0) { 15797c478bd9Sstevel@tonic-gate seg_free(seg); 15807c478bd9Sstevel@tonic-gate return (error); 15817c478bd9Sstevel@tonic-gate } 15823a30c6acSsusans as->a_size += segsize; 1583*17965fd8SKrishnendu Sadhukhan - Sun Microsystems as->a_resvsize += segsize; 15847c478bd9Sstevel@tonic-gate *segcreated = 1; 158507b65a64Saguzovsk if (do_off) { 158607b65a64Saguzovsk vn_a->offset += segsize; 158707b65a64Saguzovsk } 15887c478bd9Sstevel@tonic-gate addr = a; 15897c478bd9Sstevel@tonic-gate } 15907c478bd9Sstevel@tonic-gate szc = nszc; 15917c478bd9Sstevel@tonic-gate szcvec >>= 1; 15927c478bd9Sstevel@tonic-gate } 15937c478bd9Sstevel@tonic-gate 15947c478bd9Sstevel@tonic-gate ASSERT(addr < eaddr); 15957c478bd9Sstevel@tonic-gate szcvec = save_szcvec | 1; /* add 8K pages */ 15967c478bd9Sstevel@tonic-gate while (szcvec) { 15977c478bd9Sstevel@tonic-gate a = (caddr_t)P2ALIGN((uintptr_t)eaddr, pgsz); 15987c478bd9Sstevel@tonic-gate ASSERT(a >= addr); 15997c478bd9Sstevel@tonic-gate if (a != addr) { 16007c478bd9Sstevel@tonic-gate segsize = a - addr; 16017c478bd9Sstevel@tonic-gate seg = seg_alloc(as, addr, segsize); 16027c478bd9Sstevel@tonic-gate if (seg == NULL) { 16037c478bd9Sstevel@tonic-gate return (ENOMEM); 16047c478bd9Sstevel@tonic-gate } 16057c478bd9Sstevel@tonic-gate vn_a->szc = szc; 16067c478bd9Sstevel@tonic-gate error = (*crfp)(seg, vn_a); 16077c478bd9Sstevel@tonic-gate if (error != 0) { 16087c478bd9Sstevel@tonic-gate seg_free(seg); 16097c478bd9Sstevel@tonic-gate return (error); 16107c478bd9Sstevel@tonic-gate } 16113a30c6acSsusans as->a_size += segsize; 1612*17965fd8SKrishnendu Sadhukhan - Sun Microsystems as->a_resvsize += segsize; 16137c478bd9Sstevel@tonic-gate *segcreated = 1; 161407b65a64Saguzovsk if (do_off) { 161507b65a64Saguzovsk vn_a->offset += segsize; 161607b65a64Saguzovsk } 16177c478bd9Sstevel@tonic-gate addr = a; 16187c478bd9Sstevel@tonic-gate } 16197c478bd9Sstevel@tonic-gate szcvec &= ~(1 << szc); 16207c478bd9Sstevel@tonic-gate if (szcvec) { 16217c478bd9Sstevel@tonic-gate szc = highbit(szcvec) - 1; 16227c478bd9Sstevel@tonic-gate pgsz = page_get_pagesize(szc); 16237c478bd9Sstevel@tonic-gate } 16247c478bd9Sstevel@tonic-gate } 16257c478bd9Sstevel@tonic-gate ASSERT(addr == eaddr); 16267c478bd9Sstevel@tonic-gate 162707b65a64Saguzovsk return (0); 162807b65a64Saguzovsk } 162907b65a64Saguzovsk 163007b65a64Saguzovsk static int 163107b65a64Saguzovsk as_map_vnsegs(struct as *as, caddr_t addr, size_t size, 163207b65a64Saguzovsk int (*crfp)(), struct segvn_crargs *vn_a, int *segcreated) 163307b65a64Saguzovsk { 1634ec25b48fSsusans uint_t mapflags = vn_a->flags & (MAP_TEXT | MAP_INITDATA); 1635ec25b48fSsusans int type = (vn_a->type == MAP_SHARED) ? MAPPGSZC_SHM : MAPPGSZC_PRIVM; 1636ec25b48fSsusans uint_t szcvec = map_pgszcvec(addr, size, (uintptr_t)addr, mapflags, 1637ec25b48fSsusans type, 0); 163807b65a64Saguzovsk int error; 163907b65a64Saguzovsk struct seg *seg; 164007b65a64Saguzovsk struct vattr va; 164107b65a64Saguzovsk u_offset_t eoff; 164207b65a64Saguzovsk size_t save_size = 0; 16432cb27123Saguzovsk extern size_t textrepl_size_thresh; 164407b65a64Saguzovsk 164507b65a64Saguzovsk ASSERT(AS_WRITE_HELD(as, &as->a_lock)); 164607b65a64Saguzovsk ASSERT(IS_P2ALIGNED(addr, PAGESIZE)); 164707b65a64Saguzovsk ASSERT(IS_P2ALIGNED(size, PAGESIZE)); 164807b65a64Saguzovsk ASSERT(vn_a->vp != NULL); 164907b65a64Saguzovsk ASSERT(vn_a->amp == NULL); 165007b65a64Saguzovsk 165107b65a64Saguzovsk again: 165207b65a64Saguzovsk if (szcvec <= 1) { 165307b65a64Saguzovsk seg = seg_alloc(as, addr, size); 165407b65a64Saguzovsk if (seg == NULL) { 165507b65a64Saguzovsk return (ENOMEM); 165607b65a64Saguzovsk } 165707b65a64Saguzovsk vn_a->szc = 0; 165807b65a64Saguzovsk error = (*crfp)(seg, vn_a); 165907b65a64Saguzovsk if (error != 0) { 166007b65a64Saguzovsk seg_free(seg); 16613a30c6acSsusans } else { 16623a30c6acSsusans as->a_size += size; 1663*17965fd8SKrishnendu Sadhukhan - Sun Microsystems as->a_resvsize += size; 166407b65a64Saguzovsk } 166507b65a64Saguzovsk return (error); 166607b65a64Saguzovsk } 166707b65a64Saguzovsk 166807b65a64Saguzovsk va.va_mask = AT_SIZE; 1669da6c28aaSamw if (VOP_GETATTR(vn_a->vp, &va, ATTR_HINT, vn_a->cred, NULL) != 0) { 167007b65a64Saguzovsk szcvec = 0; 167107b65a64Saguzovsk goto again; 167207b65a64Saguzovsk } 167307b65a64Saguzovsk eoff = vn_a->offset & PAGEMASK; 167407b65a64Saguzovsk if (eoff >= va.va_size) { 167507b65a64Saguzovsk szcvec = 0; 167607b65a64Saguzovsk goto again; 167707b65a64Saguzovsk } 167807b65a64Saguzovsk eoff += size; 167907b65a64Saguzovsk if (btopr(va.va_size) < btopr(eoff)) { 168007b65a64Saguzovsk save_size = size; 168107b65a64Saguzovsk size = va.va_size - (vn_a->offset & PAGEMASK); 168207b65a64Saguzovsk size = P2ROUNDUP_TYPED(size, PAGESIZE, size_t); 1683ec25b48fSsusans szcvec = map_pgszcvec(addr, size, (uintptr_t)addr, mapflags, 1684ec25b48fSsusans type, 0); 168507b65a64Saguzovsk if (szcvec <= 1) { 168607b65a64Saguzovsk size = save_size; 168707b65a64Saguzovsk goto again; 168807b65a64Saguzovsk } 168907b65a64Saguzovsk } 169007b65a64Saguzovsk 16912cb27123Saguzovsk if (size > textrepl_size_thresh) { 16922cb27123Saguzovsk vn_a->flags |= _MAP_TEXTREPL; 16932cb27123Saguzovsk } 169407b65a64Saguzovsk error = as_map_segvn_segs(as, addr, size, szcvec, crfp, vn_a, 169507b65a64Saguzovsk segcreated); 169607b65a64Saguzovsk if (error != 0) { 169707b65a64Saguzovsk return (error); 169807b65a64Saguzovsk } 16997c478bd9Sstevel@tonic-gate if (save_size) { 170007b65a64Saguzovsk addr += size; 17017c478bd9Sstevel@tonic-gate size = save_size - size; 170207b65a64Saguzovsk szcvec = 0; 17037c478bd9Sstevel@tonic-gate goto again; 17047c478bd9Sstevel@tonic-gate } 17057c478bd9Sstevel@tonic-gate return (0); 17067c478bd9Sstevel@tonic-gate } 17077c478bd9Sstevel@tonic-gate 1708ec25b48fSsusans /* 1709ec25b48fSsusans * as_map_ansegs: shared or private anonymous memory. Note that the flags 1710ec25b48fSsusans * passed to map_pgszvec cannot be MAP_INITDATA, for anon. 1711ec25b48fSsusans */ 171207b65a64Saguzovsk static int 1713ec25b48fSsusans as_map_ansegs(struct as *as, caddr_t addr, size_t size, 171407b65a64Saguzovsk int (*crfp)(), struct segvn_crargs *vn_a, int *segcreated) 171507b65a64Saguzovsk { 1716ec25b48fSsusans uint_t szcvec; 1717ec25b48fSsusans uchar_t type; 1718ec25b48fSsusans 1719ec25b48fSsusans ASSERT(vn_a->type == MAP_SHARED || vn_a->type == MAP_PRIVATE); 1720ec25b48fSsusans if (vn_a->type == MAP_SHARED) { 1721ec25b48fSsusans type = MAPPGSZC_SHM; 1722ec25b48fSsusans } else if (vn_a->type == MAP_PRIVATE) { 1723ec25b48fSsusans if (vn_a->szc == AS_MAP_HEAP) { 1724ec25b48fSsusans type = MAPPGSZC_HEAP; 1725ec25b48fSsusans } else if (vn_a->szc == AS_MAP_STACK) { 1726ec25b48fSsusans type = MAPPGSZC_STACK; 1727ec25b48fSsusans } else { 1728ec25b48fSsusans type = MAPPGSZC_PRIVM; 1729ec25b48fSsusans } 1730ec25b48fSsusans } 1731ec25b48fSsusans szcvec = map_pgszcvec(addr, size, vn_a->amp == NULL ? 1732ec25b48fSsusans (uintptr_t)addr : (uintptr_t)P2ROUNDUP(vn_a->offset, PAGESIZE), 1733ec25b48fSsusans (vn_a->flags & MAP_TEXT), type, 0); 173407b65a64Saguzovsk ASSERT(AS_WRITE_HELD(as, &as->a_lock)); 173507b65a64Saguzovsk ASSERT(IS_P2ALIGNED(addr, PAGESIZE)); 173607b65a64Saguzovsk ASSERT(IS_P2ALIGNED(size, PAGESIZE)); 173707b65a64Saguzovsk ASSERT(vn_a->vp == NULL); 173807b65a64Saguzovsk 173907b65a64Saguzovsk return (as_map_segvn_segs(as, addr, size, szcvec, 174007b65a64Saguzovsk crfp, vn_a, segcreated)); 174107b65a64Saguzovsk } 174207b65a64Saguzovsk 17437c478bd9Sstevel@tonic-gate int 17447c478bd9Sstevel@tonic-gate as_map(struct as *as, caddr_t addr, size_t size, int (*crfp)(), void *argsp) 174502ff05a9Svsakar { 174602ff05a9Svsakar AS_LOCK_ENTER(as, &as->a_lock, RW_WRITER); 174702ff05a9Svsakar return (as_map_locked(as, addr, size, crfp, argsp)); 174802ff05a9Svsakar } 174902ff05a9Svsakar 175002ff05a9Svsakar int 175102ff05a9Svsakar as_map_locked(struct as *as, caddr_t addr, size_t size, int (*crfp)(), 175202ff05a9Svsakar void *argsp) 17537c478bd9Sstevel@tonic-gate { 17547c478bd9Sstevel@tonic-gate struct seg *seg = NULL; 17557c478bd9Sstevel@tonic-gate caddr_t raddr; /* rounded down addr */ 17567c478bd9Sstevel@tonic-gate size_t rsize; /* rounded up size */ 17577c478bd9Sstevel@tonic-gate int error; 1758ec25b48fSsusans int unmap = 0; 17597c478bd9Sstevel@tonic-gate struct proc *p = curproc; 1760232cfe63Ssusans struct segvn_crargs crargs; 17617c478bd9Sstevel@tonic-gate 17627c478bd9Sstevel@tonic-gate raddr = (caddr_t)((uintptr_t)addr & (uintptr_t)PAGEMASK); 17637c478bd9Sstevel@tonic-gate rsize = (((size_t)(addr + size) + PAGEOFFSET) & PAGEMASK) - 1764843e1988Sjohnlev (size_t)raddr; 17657c478bd9Sstevel@tonic-gate 17667c478bd9Sstevel@tonic-gate /* 17677c478bd9Sstevel@tonic-gate * check for wrap around 17687c478bd9Sstevel@tonic-gate */ 17697c478bd9Sstevel@tonic-gate if ((raddr + rsize < raddr) || (as->a_size > (ULONG_MAX - size))) { 17707c478bd9Sstevel@tonic-gate AS_LOCK_EXIT(as, &as->a_lock); 17717c478bd9Sstevel@tonic-gate return (ENOMEM); 17727c478bd9Sstevel@tonic-gate } 17737c478bd9Sstevel@tonic-gate 17747c478bd9Sstevel@tonic-gate as->a_updatedir = 1; /* inform /proc */ 17757c478bd9Sstevel@tonic-gate gethrestime(&as->a_updatetime); 17767c478bd9Sstevel@tonic-gate 17777c478bd9Sstevel@tonic-gate if (as != &kas && as->a_size + rsize > (size_t)p->p_vmem_ctl) { 17787c478bd9Sstevel@tonic-gate AS_LOCK_EXIT(as, &as->a_lock); 17797c478bd9Sstevel@tonic-gate 17807c478bd9Sstevel@tonic-gate (void) rctl_action(rctlproc_legacy[RLIMIT_VMEM], p->p_rctls, p, 17817c478bd9Sstevel@tonic-gate RCA_UNSAFE_ALL); 17827c478bd9Sstevel@tonic-gate 17837c478bd9Sstevel@tonic-gate return (ENOMEM); 17847c478bd9Sstevel@tonic-gate } 17857c478bd9Sstevel@tonic-gate 1786ec25b48fSsusans if (AS_MAP_CHECK_VNODE_LPOOB(crfp, argsp)) { 1787232cfe63Ssusans crargs = *(struct segvn_crargs *)argsp; 1788232cfe63Ssusans error = as_map_vnsegs(as, raddr, rsize, crfp, &crargs, &unmap); 1789ec25b48fSsusans if (error != 0) { 1790ec25b48fSsusans AS_LOCK_EXIT(as, &as->a_lock); 1791ec25b48fSsusans if (unmap) { 1792ec25b48fSsusans (void) as_unmap(as, addr, size); 1793ec25b48fSsusans } 1794ec25b48fSsusans return (error); 179507b65a64Saguzovsk } 1796ec25b48fSsusans } else if (AS_MAP_CHECK_ANON_LPOOB(crfp, argsp)) { 1797232cfe63Ssusans crargs = *(struct segvn_crargs *)argsp; 1798232cfe63Ssusans error = as_map_ansegs(as, raddr, rsize, crfp, &crargs, &unmap); 17997c478bd9Sstevel@tonic-gate if (error != 0) { 18007c478bd9Sstevel@tonic-gate AS_LOCK_EXIT(as, &as->a_lock); 18017c478bd9Sstevel@tonic-gate if (unmap) { 18027c478bd9Sstevel@tonic-gate (void) as_unmap(as, addr, size); 18037c478bd9Sstevel@tonic-gate } 18047c478bd9Sstevel@tonic-gate return (error); 18057c478bd9Sstevel@tonic-gate } 18067c478bd9Sstevel@tonic-gate } else { 18077c478bd9Sstevel@tonic-gate seg = seg_alloc(as, addr, size); 18087c478bd9Sstevel@tonic-gate if (seg == NULL) { 18097c478bd9Sstevel@tonic-gate AS_LOCK_EXIT(as, &as->a_lock); 18107c478bd9Sstevel@tonic-gate return (ENOMEM); 18117c478bd9Sstevel@tonic-gate } 18127c478bd9Sstevel@tonic-gate 18137c478bd9Sstevel@tonic-gate error = (*crfp)(seg, argsp); 18147c478bd9Sstevel@tonic-gate if (error != 0) { 18157c478bd9Sstevel@tonic-gate seg_free(seg); 18167c478bd9Sstevel@tonic-gate AS_LOCK_EXIT(as, &as->a_lock); 18177c478bd9Sstevel@tonic-gate return (error); 18187c478bd9Sstevel@tonic-gate } 18193a30c6acSsusans /* 18203a30c6acSsusans * Add size now so as_unmap will work if as_ctl fails. 18213a30c6acSsusans */ 18223a30c6acSsusans as->a_size += rsize; 1823*17965fd8SKrishnendu Sadhukhan - Sun Microsystems as->a_resvsize += rsize; 18247c478bd9Sstevel@tonic-gate } 18257c478bd9Sstevel@tonic-gate 18267c478bd9Sstevel@tonic-gate as_setwatch(as); 18277c478bd9Sstevel@tonic-gate 18287c478bd9Sstevel@tonic-gate /* 18297c478bd9Sstevel@tonic-gate * If the address space is locked, 18307c478bd9Sstevel@tonic-gate * establish memory locks for the new segment. 18317c478bd9Sstevel@tonic-gate */ 18327c478bd9Sstevel@tonic-gate mutex_enter(&as->a_contents); 18337c478bd9Sstevel@tonic-gate if (AS_ISPGLCK(as)) { 18347c478bd9Sstevel@tonic-gate mutex_exit(&as->a_contents); 18357c478bd9Sstevel@tonic-gate AS_LOCK_EXIT(as, &as->a_lock); 18367c478bd9Sstevel@tonic-gate error = as_ctl(as, addr, size, MC_LOCK, 0, 0, NULL, 0); 18377c478bd9Sstevel@tonic-gate if (error != 0) 18387c478bd9Sstevel@tonic-gate (void) as_unmap(as, addr, size); 18397c478bd9Sstevel@tonic-gate } else { 18407c478bd9Sstevel@tonic-gate mutex_exit(&as->a_contents); 18417c478bd9Sstevel@tonic-gate AS_LOCK_EXIT(as, &as->a_lock); 18427c478bd9Sstevel@tonic-gate } 18437c478bd9Sstevel@tonic-gate return (error); 18447c478bd9Sstevel@tonic-gate } 18457c478bd9Sstevel@tonic-gate 18467c478bd9Sstevel@tonic-gate 18477c478bd9Sstevel@tonic-gate /* 18487c478bd9Sstevel@tonic-gate * Delete all segments in the address space marked with S_PURGE. 18497c478bd9Sstevel@tonic-gate * This is currently used for Sparc V9 nofault ASI segments (seg_nf.c). 18507c478bd9Sstevel@tonic-gate * These segments are deleted as a first step before calls to as_gap(), so 18517c478bd9Sstevel@tonic-gate * that they don't affect mmap() or shmat(). 18527c478bd9Sstevel@tonic-gate */ 18537c478bd9Sstevel@tonic-gate void 18547c478bd9Sstevel@tonic-gate as_purge(struct as *as) 18557c478bd9Sstevel@tonic-gate { 18567c478bd9Sstevel@tonic-gate struct seg *seg; 18577c478bd9Sstevel@tonic-gate struct seg *next_seg; 18587c478bd9Sstevel@tonic-gate 18597c478bd9Sstevel@tonic-gate /* 18607c478bd9Sstevel@tonic-gate * the setting of NEEDSPURGE is protect by as_rangelock(), so 18617c478bd9Sstevel@tonic-gate * no need to grab a_contents mutex for this check 18627c478bd9Sstevel@tonic-gate */ 18637c478bd9Sstevel@tonic-gate if ((as->a_flags & AS_NEEDSPURGE) == 0) 18647c478bd9Sstevel@tonic-gate return; 18657c478bd9Sstevel@tonic-gate 18667c478bd9Sstevel@tonic-gate AS_LOCK_ENTER(as, &as->a_lock, RW_WRITER); 18677c478bd9Sstevel@tonic-gate next_seg = NULL; 18687c478bd9Sstevel@tonic-gate seg = AS_SEGFIRST(as); 18697c478bd9Sstevel@tonic-gate while (seg != NULL) { 18707c478bd9Sstevel@tonic-gate next_seg = AS_SEGNEXT(as, seg); 18717c478bd9Sstevel@tonic-gate if (seg->s_flags & S_PURGE) 18727c478bd9Sstevel@tonic-gate SEGOP_UNMAP(seg, seg->s_base, seg->s_size); 18737c478bd9Sstevel@tonic-gate seg = next_seg; 18747c478bd9Sstevel@tonic-gate } 18757c478bd9Sstevel@tonic-gate AS_LOCK_EXIT(as, &as->a_lock); 18767c478bd9Sstevel@tonic-gate 18777c478bd9Sstevel@tonic-gate mutex_enter(&as->a_contents); 18787c478bd9Sstevel@tonic-gate as->a_flags &= ~AS_NEEDSPURGE; 18797c478bd9Sstevel@tonic-gate mutex_exit(&as->a_contents); 18807c478bd9Sstevel@tonic-gate } 18817c478bd9Sstevel@tonic-gate 18827c478bd9Sstevel@tonic-gate /* 188346ab9534Smec * Find a hole within [*basep, *basep + *lenp), which contains a mappable 188446ab9534Smec * range of addresses at least "minlen" long, where the base of the range is 188546ab9534Smec * at "off" phase from an "align" boundary and there is space for a 188646ab9534Smec * "redzone"-sized redzone on eithe rside of the range. Thus, 188746ab9534Smec * if align was 4M and off was 16k, the user wants a hole which will start 188846ab9534Smec * 16k into a 4M page. 18897c478bd9Sstevel@tonic-gate * 18907c478bd9Sstevel@tonic-gate * If flags specifies AH_HI, the hole will have the highest possible address 18917c478bd9Sstevel@tonic-gate * in the range. We use the as->a_lastgap field to figure out where to 18927c478bd9Sstevel@tonic-gate * start looking for a gap. 18937c478bd9Sstevel@tonic-gate * 18947c478bd9Sstevel@tonic-gate * Otherwise, the gap will have the lowest possible address. 18957c478bd9Sstevel@tonic-gate * 18967c478bd9Sstevel@tonic-gate * If flags specifies AH_CONTAIN, the hole will contain the address addr. 18977c478bd9Sstevel@tonic-gate * 189846ab9534Smec * If an adequate hole is found, *basep and *lenp are set to reflect the part of 189946ab9534Smec * the hole that is within range, and 0 is returned. On failure, -1 is returned. 19007c478bd9Sstevel@tonic-gate * 19017c478bd9Sstevel@tonic-gate * NOTE: This routine is not correct when base+len overflows caddr_t. 19027c478bd9Sstevel@tonic-gate */ 19037c478bd9Sstevel@tonic-gate int 190446ab9534Smec as_gap_aligned(struct as *as, size_t minlen, caddr_t *basep, size_t *lenp, 190546ab9534Smec uint_t flags, caddr_t addr, size_t align, size_t redzone, size_t off) 19067c478bd9Sstevel@tonic-gate { 19077c478bd9Sstevel@tonic-gate caddr_t lobound = *basep; 19087c478bd9Sstevel@tonic-gate caddr_t hibound = lobound + *lenp; 19097c478bd9Sstevel@tonic-gate struct seg *lseg, *hseg; 19107c478bd9Sstevel@tonic-gate caddr_t lo, hi; 19117c478bd9Sstevel@tonic-gate int forward; 19127c478bd9Sstevel@tonic-gate caddr_t save_base; 19137c478bd9Sstevel@tonic-gate size_t save_len; 19147d87bd0eSmec size_t save_minlen; 19157d87bd0eSmec size_t save_redzone; 19167d87bd0eSmec int fast_path = 1; 19177c478bd9Sstevel@tonic-gate 19187c478bd9Sstevel@tonic-gate save_base = *basep; 19197c478bd9Sstevel@tonic-gate save_len = *lenp; 19207d87bd0eSmec save_minlen = minlen; 19217d87bd0eSmec save_redzone = redzone; 19227d87bd0eSmec 19237d87bd0eSmec /* 19247d87bd0eSmec * For the first pass/fast_path, just add align and redzone into 19257d87bd0eSmec * minlen since if we get an allocation, we can guarantee that it 19267d87bd0eSmec * will fit the alignment and redzone requested. 19277d87bd0eSmec * This increases the chance that hibound will be adjusted to 19287d87bd0eSmec * a_lastgap->s_base which will likely allow us to find an 19297d87bd0eSmec * acceptable hole in the address space quicker. 19307d87bd0eSmec * If we can't find a hole with this fast_path, then we look for 19317d87bd0eSmec * smaller holes in which the alignment and offset may allow 19327d87bd0eSmec * the allocation to fit. 19337d87bd0eSmec */ 19347d87bd0eSmec minlen += align; 19357d87bd0eSmec minlen += 2 * redzone; 19367d87bd0eSmec redzone = 0; 19377d87bd0eSmec 19387c478bd9Sstevel@tonic-gate AS_LOCK_ENTER(as, &as->a_lock, RW_READER); 19397c478bd9Sstevel@tonic-gate if (AS_SEGFIRST(as) == NULL) { 194046ab9534Smec if (valid_va_range_aligned(basep, lenp, minlen, flags & AH_DIR, 194146ab9534Smec align, redzone, off)) { 19427c478bd9Sstevel@tonic-gate AS_LOCK_EXIT(as, &as->a_lock); 19437c478bd9Sstevel@tonic-gate return (0); 19447c478bd9Sstevel@tonic-gate } else { 19457c478bd9Sstevel@tonic-gate AS_LOCK_EXIT(as, &as->a_lock); 19467c478bd9Sstevel@tonic-gate *basep = save_base; 19477c478bd9Sstevel@tonic-gate *lenp = save_len; 19487c478bd9Sstevel@tonic-gate return (-1); 19497c478bd9Sstevel@tonic-gate } 19507c478bd9Sstevel@tonic-gate } 19517c478bd9Sstevel@tonic-gate 19527d87bd0eSmec retry: 19537c478bd9Sstevel@tonic-gate /* 19547c478bd9Sstevel@tonic-gate * Set up to iterate over all the inter-segment holes in the given 19557c478bd9Sstevel@tonic-gate * direction. lseg is NULL for the lowest-addressed hole and hseg is 19567c478bd9Sstevel@tonic-gate * NULL for the highest-addressed hole. If moving backwards, we reset 19577c478bd9Sstevel@tonic-gate * sseg to denote the highest-addressed segment. 19587c478bd9Sstevel@tonic-gate */ 19597c478bd9Sstevel@tonic-gate forward = (flags & AH_DIR) == AH_LO; 19607c478bd9Sstevel@tonic-gate if (forward) { 19617c478bd9Sstevel@tonic-gate hseg = as_findseg(as, lobound, 1); 19627c478bd9Sstevel@tonic-gate lseg = AS_SEGPREV(as, hseg); 19637c478bd9Sstevel@tonic-gate } else { 19647c478bd9Sstevel@tonic-gate 19657c478bd9Sstevel@tonic-gate /* 19667c478bd9Sstevel@tonic-gate * If allocating at least as much as the last allocation, 19677c478bd9Sstevel@tonic-gate * use a_lastgap's base as a better estimate of hibound. 19687c478bd9Sstevel@tonic-gate */ 19697c478bd9Sstevel@tonic-gate if (as->a_lastgap && 19707c478bd9Sstevel@tonic-gate minlen >= as->a_lastgap->s_size && 19717c478bd9Sstevel@tonic-gate hibound >= as->a_lastgap->s_base) 19727c478bd9Sstevel@tonic-gate hibound = as->a_lastgap->s_base; 19737c478bd9Sstevel@tonic-gate 19747c478bd9Sstevel@tonic-gate hseg = as_findseg(as, hibound, 1); 19757c478bd9Sstevel@tonic-gate if (hseg->s_base + hseg->s_size < hibound) { 19767c478bd9Sstevel@tonic-gate lseg = hseg; 19777c478bd9Sstevel@tonic-gate hseg = NULL; 19787c478bd9Sstevel@tonic-gate } else { 19797c478bd9Sstevel@tonic-gate lseg = AS_SEGPREV(as, hseg); 19807c478bd9Sstevel@tonic-gate } 19817c478bd9Sstevel@tonic-gate } 19827c478bd9Sstevel@tonic-gate 19837c478bd9Sstevel@tonic-gate for (;;) { 19847c478bd9Sstevel@tonic-gate /* 19857c478bd9Sstevel@tonic-gate * Set lo and hi to the hole's boundaries. (We should really 19867c478bd9Sstevel@tonic-gate * use MAXADDR in place of hibound in the expression below, 19877c478bd9Sstevel@tonic-gate * but can't express it easily; using hibound in its place is 19887c478bd9Sstevel@tonic-gate * harmless.) 19897c478bd9Sstevel@tonic-gate */ 19907c478bd9Sstevel@tonic-gate lo = (lseg == NULL) ? 0 : lseg->s_base + lseg->s_size; 19917c478bd9Sstevel@tonic-gate hi = (hseg == NULL) ? hibound : hseg->s_base; 19927c478bd9Sstevel@tonic-gate /* 19937c478bd9Sstevel@tonic-gate * If the iteration has moved past the interval from lobound 19947c478bd9Sstevel@tonic-gate * to hibound it's pointless to continue. 19957c478bd9Sstevel@tonic-gate */ 19967c478bd9Sstevel@tonic-gate if ((forward && lo > hibound) || (!forward && hi < lobound)) 19977c478bd9Sstevel@tonic-gate break; 19987c478bd9Sstevel@tonic-gate else if (lo > hibound || hi < lobound) 19997c478bd9Sstevel@tonic-gate goto cont; 20007c478bd9Sstevel@tonic-gate /* 20017c478bd9Sstevel@tonic-gate * Candidate hole lies at least partially within the allowable 20027c478bd9Sstevel@tonic-gate * range. Restrict it to fall completely within that range, 20037c478bd9Sstevel@tonic-gate * i.e., to [max(lo, lobound), min(hi, hibound)]. 20047c478bd9Sstevel@tonic-gate */ 20057c478bd9Sstevel@tonic-gate if (lo < lobound) 20067c478bd9Sstevel@tonic-gate lo = lobound; 20077c478bd9Sstevel@tonic-gate if (hi > hibound) 20087c478bd9Sstevel@tonic-gate hi = hibound; 20097c478bd9Sstevel@tonic-gate /* 20107c478bd9Sstevel@tonic-gate * Verify that the candidate hole is big enough and meets 20117d87bd0eSmec * hardware constraints. If the hole is too small, no need 20127d87bd0eSmec * to do the further checks since they will fail. 20137c478bd9Sstevel@tonic-gate */ 20147c478bd9Sstevel@tonic-gate *basep = lo; 20157c478bd9Sstevel@tonic-gate *lenp = hi - lo; 20167d87bd0eSmec if (*lenp >= minlen && valid_va_range_aligned(basep, lenp, 20177d87bd0eSmec minlen, forward ? AH_LO : AH_HI, align, redzone, off) && 20187c478bd9Sstevel@tonic-gate ((flags & AH_CONTAIN) == 0 || 20197c478bd9Sstevel@tonic-gate (*basep <= addr && *basep + *lenp > addr))) { 20207c478bd9Sstevel@tonic-gate if (!forward) 20217c478bd9Sstevel@tonic-gate as->a_lastgap = hseg; 20227c478bd9Sstevel@tonic-gate if (hseg != NULL) 20237c478bd9Sstevel@tonic-gate as->a_lastgaphl = hseg; 20247c478bd9Sstevel@tonic-gate else 20257c478bd9Sstevel@tonic-gate as->a_lastgaphl = lseg; 20267c478bd9Sstevel@tonic-gate AS_LOCK_EXIT(as, &as->a_lock); 20277c478bd9Sstevel@tonic-gate return (0); 20287c478bd9Sstevel@tonic-gate } 20297c478bd9Sstevel@tonic-gate cont: 20307c478bd9Sstevel@tonic-gate /* 20317c478bd9Sstevel@tonic-gate * Move to the next hole. 20327c478bd9Sstevel@tonic-gate */ 20337c478bd9Sstevel@tonic-gate if (forward) { 20347c478bd9Sstevel@tonic-gate lseg = hseg; 20357c478bd9Sstevel@tonic-gate if (lseg == NULL) 20367c478bd9Sstevel@tonic-gate break; 20377c478bd9Sstevel@tonic-gate hseg = AS_SEGNEXT(as, hseg); 20387c478bd9Sstevel@tonic-gate } else { 20397c478bd9Sstevel@tonic-gate hseg = lseg; 20407c478bd9Sstevel@tonic-gate if (hseg == NULL) 20417c478bd9Sstevel@tonic-gate break; 20427c478bd9Sstevel@tonic-gate lseg = AS_SEGPREV(as, lseg); 20437c478bd9Sstevel@tonic-gate } 20447c478bd9Sstevel@tonic-gate } 20457d87bd0eSmec if (fast_path && (align != 0 || save_redzone != 0)) { 20467d87bd0eSmec fast_path = 0; 20477d87bd0eSmec minlen = save_minlen; 20487d87bd0eSmec redzone = save_redzone; 20497d87bd0eSmec goto retry; 20507d87bd0eSmec } 20517c478bd9Sstevel@tonic-gate *basep = save_base; 20527c478bd9Sstevel@tonic-gate *lenp = save_len; 20537c478bd9Sstevel@tonic-gate AS_LOCK_EXIT(as, &as->a_lock); 20547c478bd9Sstevel@tonic-gate return (-1); 20557c478bd9Sstevel@tonic-gate } 20567c478bd9Sstevel@tonic-gate 205746ab9534Smec /* 205846ab9534Smec * Find a hole of at least size minlen within [*basep, *basep + *lenp). 205946ab9534Smec * 206046ab9534Smec * If flags specifies AH_HI, the hole will have the highest possible address 206146ab9534Smec * in the range. We use the as->a_lastgap field to figure out where to 206246ab9534Smec * start looking for a gap. 206346ab9534Smec * 206446ab9534Smec * Otherwise, the gap will have the lowest possible address. 206546ab9534Smec * 206646ab9534Smec * If flags specifies AH_CONTAIN, the hole will contain the address addr. 206746ab9534Smec * 206846ab9534Smec * If an adequate hole is found, base and len are set to reflect the part of 206946ab9534Smec * the hole that is within range, and 0 is returned, otherwise, 207046ab9534Smec * -1 is returned. 207146ab9534Smec * 207246ab9534Smec * NOTE: This routine is not correct when base+len overflows caddr_t. 207346ab9534Smec */ 207446ab9534Smec int 207546ab9534Smec as_gap(struct as *as, size_t minlen, caddr_t *basep, size_t *lenp, uint_t flags, 207646ab9534Smec caddr_t addr) 207746ab9534Smec { 207846ab9534Smec 207946ab9534Smec return (as_gap_aligned(as, minlen, basep, lenp, flags, addr, 0, 0, 0)); 208046ab9534Smec } 208146ab9534Smec 20827c478bd9Sstevel@tonic-gate /* 20837c478bd9Sstevel@tonic-gate * Return the next range within [base, base + len) that is backed 20847c478bd9Sstevel@tonic-gate * with "real memory". Skip holes and non-seg_vn segments. 20857c478bd9Sstevel@tonic-gate * We're lazy and only return one segment at a time. 20867c478bd9Sstevel@tonic-gate */ 20877c478bd9Sstevel@tonic-gate int 20887c478bd9Sstevel@tonic-gate as_memory(struct as *as, caddr_t *basep, size_t *lenp) 20897c478bd9Sstevel@tonic-gate { 20907c478bd9Sstevel@tonic-gate extern struct seg_ops segspt_shmops; /* needs a header file */ 20917c478bd9Sstevel@tonic-gate struct seg *seg; 20927c478bd9Sstevel@tonic-gate caddr_t addr, eaddr; 20937c478bd9Sstevel@tonic-gate caddr_t segend; 20947c478bd9Sstevel@tonic-gate 20957c478bd9Sstevel@tonic-gate AS_LOCK_ENTER(as, &as->a_lock, RW_READER); 20967c478bd9Sstevel@tonic-gate 20977c478bd9Sstevel@tonic-gate addr = *basep; 20987c478bd9Sstevel@tonic-gate eaddr = addr + *lenp; 20997c478bd9Sstevel@tonic-gate 21007c478bd9Sstevel@tonic-gate seg = as_findseg(as, addr, 0); 21017c478bd9Sstevel@tonic-gate if (seg != NULL) 21027c478bd9Sstevel@tonic-gate addr = MAX(seg->s_base, addr); 21037c478bd9Sstevel@tonic-gate 21047c478bd9Sstevel@tonic-gate for (;;) { 21057c478bd9Sstevel@tonic-gate if (seg == NULL || addr >= eaddr || eaddr <= seg->s_base) { 21067c478bd9Sstevel@tonic-gate AS_LOCK_EXIT(as, &as->a_lock); 21077c478bd9Sstevel@tonic-gate return (EINVAL); 21087c478bd9Sstevel@tonic-gate } 21097c478bd9Sstevel@tonic-gate 21107c478bd9Sstevel@tonic-gate if (seg->s_ops == &segvn_ops) { 21117c478bd9Sstevel@tonic-gate segend = seg->s_base + seg->s_size; 21127c478bd9Sstevel@tonic-gate break; 21137c478bd9Sstevel@tonic-gate } 21147c478bd9Sstevel@tonic-gate 21157c478bd9Sstevel@tonic-gate /* 21167c478bd9Sstevel@tonic-gate * We do ISM by looking into the private data 21177c478bd9Sstevel@tonic-gate * to determine the real size of the segment. 21187c478bd9Sstevel@tonic-gate */ 21197c478bd9Sstevel@tonic-gate if (seg->s_ops == &segspt_shmops) { 21207c478bd9Sstevel@tonic-gate segend = seg->s_base + spt_realsize(seg); 21217c478bd9Sstevel@tonic-gate if (addr < segend) 21227c478bd9Sstevel@tonic-gate break; 21237c478bd9Sstevel@tonic-gate } 21247c478bd9Sstevel@tonic-gate 21257c478bd9Sstevel@tonic-gate seg = AS_SEGNEXT(as, seg); 21267c478bd9Sstevel@tonic-gate 21277c478bd9Sstevel@tonic-gate if (seg != NULL) 21287c478bd9Sstevel@tonic-gate addr = seg->s_base; 21297c478bd9Sstevel@tonic-gate } 21307c478bd9Sstevel@tonic-gate 21317c478bd9Sstevel@tonic-gate *basep = addr; 21327c478bd9Sstevel@tonic-gate 21337c478bd9Sstevel@tonic-gate if (segend > eaddr) 21347c478bd9Sstevel@tonic-gate *lenp = eaddr - addr; 21357c478bd9Sstevel@tonic-gate else 21367c478bd9Sstevel@tonic-gate *lenp = segend - addr; 21377c478bd9Sstevel@tonic-gate 21387c478bd9Sstevel@tonic-gate AS_LOCK_EXIT(as, &as->a_lock); 21397c478bd9Sstevel@tonic-gate return (0); 21407c478bd9Sstevel@tonic-gate } 21417c478bd9Sstevel@tonic-gate 21427c478bd9Sstevel@tonic-gate /* 21437c478bd9Sstevel@tonic-gate * Swap the pages associated with the address space as out to 21447c478bd9Sstevel@tonic-gate * secondary storage, returning the number of bytes actually 21457c478bd9Sstevel@tonic-gate * swapped. 21467c478bd9Sstevel@tonic-gate * 21477c478bd9Sstevel@tonic-gate * The value returned is intended to correlate well with the process's 21487c478bd9Sstevel@tonic-gate * memory requirements. Its usefulness for this purpose depends on 21497c478bd9Sstevel@tonic-gate * how well the segment-level routines do at returning accurate 21507c478bd9Sstevel@tonic-gate * information. 21517c478bd9Sstevel@tonic-gate */ 21527c478bd9Sstevel@tonic-gate size_t 21537c478bd9Sstevel@tonic-gate as_swapout(struct as *as) 21547c478bd9Sstevel@tonic-gate { 21557c478bd9Sstevel@tonic-gate struct seg *seg; 21567c478bd9Sstevel@tonic-gate size_t swpcnt = 0; 21577c478bd9Sstevel@tonic-gate 21587c478bd9Sstevel@tonic-gate /* 21597c478bd9Sstevel@tonic-gate * Kernel-only processes have given up their address 21607c478bd9Sstevel@tonic-gate * spaces. Of course, we shouldn't be attempting to 21617c478bd9Sstevel@tonic-gate * swap out such processes in the first place... 21627c478bd9Sstevel@tonic-gate */ 21637c478bd9Sstevel@tonic-gate if (as == NULL) 21647c478bd9Sstevel@tonic-gate return (0); 21657c478bd9Sstevel@tonic-gate 21667c478bd9Sstevel@tonic-gate AS_LOCK_ENTER(as, &as->a_lock, RW_READER); 21677c478bd9Sstevel@tonic-gate 21687c478bd9Sstevel@tonic-gate /* Prevent XHATs from attaching */ 21697c478bd9Sstevel@tonic-gate mutex_enter(&as->a_contents); 21707c478bd9Sstevel@tonic-gate AS_SETBUSY(as); 21717c478bd9Sstevel@tonic-gate mutex_exit(&as->a_contents); 21727c478bd9Sstevel@tonic-gate 21737c478bd9Sstevel@tonic-gate 21747c478bd9Sstevel@tonic-gate /* 21757c478bd9Sstevel@tonic-gate * Free all mapping resources associated with the address 21767c478bd9Sstevel@tonic-gate * space. The segment-level swapout routines capitalize 21777c478bd9Sstevel@tonic-gate * on this unmapping by scavanging pages that have become 21787c478bd9Sstevel@tonic-gate * unmapped here. 21797c478bd9Sstevel@tonic-gate */ 21807c478bd9Sstevel@tonic-gate hat_swapout(as->a_hat); 21817c478bd9Sstevel@tonic-gate if (as->a_xhat != NULL) 21827c478bd9Sstevel@tonic-gate xhat_swapout_all(as); 21837c478bd9Sstevel@tonic-gate 21847c478bd9Sstevel@tonic-gate mutex_enter(&as->a_contents); 21857c478bd9Sstevel@tonic-gate AS_CLRBUSY(as); 21867c478bd9Sstevel@tonic-gate mutex_exit(&as->a_contents); 21877c478bd9Sstevel@tonic-gate 21887c478bd9Sstevel@tonic-gate /* 21897c478bd9Sstevel@tonic-gate * Call the swapout routines of all segments in the address 21907c478bd9Sstevel@tonic-gate * space to do the actual work, accumulating the amount of 21917c478bd9Sstevel@tonic-gate * space reclaimed. 21927c478bd9Sstevel@tonic-gate */ 21937c478bd9Sstevel@tonic-gate for (seg = AS_SEGFIRST(as); seg != NULL; seg = AS_SEGNEXT(as, seg)) { 21947c478bd9Sstevel@tonic-gate struct seg_ops *ov = seg->s_ops; 21957c478bd9Sstevel@tonic-gate 21967c478bd9Sstevel@tonic-gate /* 21977c478bd9Sstevel@tonic-gate * We have to check to see if the seg has 21987c478bd9Sstevel@tonic-gate * an ops vector because the seg may have 21997c478bd9Sstevel@tonic-gate * been in the middle of being set up when 22007c478bd9Sstevel@tonic-gate * the process was picked for swapout. 22017c478bd9Sstevel@tonic-gate */ 22027c478bd9Sstevel@tonic-gate if ((ov != NULL) && (ov->swapout != NULL)) 22037c478bd9Sstevel@tonic-gate swpcnt += SEGOP_SWAPOUT(seg); 22047c478bd9Sstevel@tonic-gate } 22057c478bd9Sstevel@tonic-gate AS_LOCK_EXIT(as, &as->a_lock); 22067c478bd9Sstevel@tonic-gate return (swpcnt); 22077c478bd9Sstevel@tonic-gate } 22087c478bd9Sstevel@tonic-gate 22097c478bd9Sstevel@tonic-gate /* 22107c478bd9Sstevel@tonic-gate * Determine whether data from the mappings in interval [addr, addr + size) 22117c478bd9Sstevel@tonic-gate * are in the primary memory (core) cache. 22127c478bd9Sstevel@tonic-gate */ 22137c478bd9Sstevel@tonic-gate int 22147c478bd9Sstevel@tonic-gate as_incore(struct as *as, caddr_t addr, 22157c478bd9Sstevel@tonic-gate size_t size, char *vec, size_t *sizep) 22167c478bd9Sstevel@tonic-gate { 22177c478bd9Sstevel@tonic-gate struct seg *seg; 22187c478bd9Sstevel@tonic-gate size_t ssize; 22197c478bd9Sstevel@tonic-gate caddr_t raddr; /* rounded down addr */ 22207c478bd9Sstevel@tonic-gate size_t rsize; /* rounded up size */ 22217c478bd9Sstevel@tonic-gate size_t isize; /* iteration size */ 22227c478bd9Sstevel@tonic-gate int error = 0; /* result, assume success */ 22237c478bd9Sstevel@tonic-gate 22247c478bd9Sstevel@tonic-gate *sizep = 0; 22257c478bd9Sstevel@tonic-gate raddr = (caddr_t)((uintptr_t)addr & (uintptr_t)PAGEMASK); 22267c478bd9Sstevel@tonic-gate rsize = ((((size_t)addr + size) + PAGEOFFSET) & PAGEMASK) - 2227843e1988Sjohnlev (size_t)raddr; 22287c478bd9Sstevel@tonic-gate 22297c478bd9Sstevel@tonic-gate if (raddr + rsize < raddr) /* check for wraparound */ 22307c478bd9Sstevel@tonic-gate return (ENOMEM); 22317c478bd9Sstevel@tonic-gate 22327c478bd9Sstevel@tonic-gate AS_LOCK_ENTER(as, &as->a_lock, RW_READER); 22337c478bd9Sstevel@tonic-gate seg = as_segat(as, raddr); 22347c478bd9Sstevel@tonic-gate if (seg == NULL) { 22357c478bd9Sstevel@tonic-gate AS_LOCK_EXIT(as, &as->a_lock); 22367c478bd9Sstevel@tonic-gate return (-1); 22377c478bd9Sstevel@tonic-gate } 22387c478bd9Sstevel@tonic-gate 22397c478bd9Sstevel@tonic-gate for (; rsize != 0; rsize -= ssize, raddr += ssize) { 22407c478bd9Sstevel@tonic-gate if (raddr >= seg->s_base + seg->s_size) { 22417c478bd9Sstevel@tonic-gate seg = AS_SEGNEXT(as, seg); 22427c478bd9Sstevel@tonic-gate if (seg == NULL || raddr != seg->s_base) { 22437c478bd9Sstevel@tonic-gate error = -1; 22447c478bd9Sstevel@tonic-gate break; 22457c478bd9Sstevel@tonic-gate } 22467c478bd9Sstevel@tonic-gate } 22477c478bd9Sstevel@tonic-gate if ((raddr + rsize) > (seg->s_base + seg->s_size)) 22487c478bd9Sstevel@tonic-gate ssize = seg->s_base + seg->s_size - raddr; 22497c478bd9Sstevel@tonic-gate else 22507c478bd9Sstevel@tonic-gate ssize = rsize; 22517c478bd9Sstevel@tonic-gate *sizep += isize = SEGOP_INCORE(seg, raddr, ssize, vec); 22527c478bd9Sstevel@tonic-gate if (isize != ssize) { 22537c478bd9Sstevel@tonic-gate error = -1; 22547c478bd9Sstevel@tonic-gate break; 22557c478bd9Sstevel@tonic-gate } 22567c478bd9Sstevel@tonic-gate vec += btopr(ssize); 22577c478bd9Sstevel@tonic-gate } 22587c478bd9Sstevel@tonic-gate AS_LOCK_EXIT(as, &as->a_lock); 22597c478bd9Sstevel@tonic-gate return (error); 22607c478bd9Sstevel@tonic-gate } 22617c478bd9Sstevel@tonic-gate 22627c478bd9Sstevel@tonic-gate static void 22637c478bd9Sstevel@tonic-gate as_segunlock(struct seg *seg, caddr_t addr, int attr, 22647c478bd9Sstevel@tonic-gate ulong_t *bitmap, size_t position, size_t npages) 22657c478bd9Sstevel@tonic-gate { 22667c478bd9Sstevel@tonic-gate caddr_t range_start; 22677c478bd9Sstevel@tonic-gate size_t pos1 = position; 22687c478bd9Sstevel@tonic-gate size_t pos2; 22697c478bd9Sstevel@tonic-gate size_t size; 22707c478bd9Sstevel@tonic-gate size_t end_pos = npages + position; 22717c478bd9Sstevel@tonic-gate 22727c478bd9Sstevel@tonic-gate while (bt_range(bitmap, &pos1, &pos2, end_pos)) { 22737c478bd9Sstevel@tonic-gate size = ptob((pos2 - pos1)); 22747c478bd9Sstevel@tonic-gate range_start = (caddr_t)((uintptr_t)addr + 2275843e1988Sjohnlev ptob(pos1 - position)); 22767c478bd9Sstevel@tonic-gate 22777c478bd9Sstevel@tonic-gate (void) SEGOP_LOCKOP(seg, range_start, size, attr, MC_UNLOCK, 2278843e1988Sjohnlev (ulong_t *)NULL, (size_t)NULL); 22797c478bd9Sstevel@tonic-gate pos1 = pos2; 22807c478bd9Sstevel@tonic-gate } 22817c478bd9Sstevel@tonic-gate } 22827c478bd9Sstevel@tonic-gate 22837c478bd9Sstevel@tonic-gate static void 22847c478bd9Sstevel@tonic-gate as_unlockerr(struct as *as, int attr, ulong_t *mlock_map, 22857c478bd9Sstevel@tonic-gate caddr_t raddr, size_t rsize) 22867c478bd9Sstevel@tonic-gate { 22877c478bd9Sstevel@tonic-gate struct seg *seg = as_segat(as, raddr); 22887c478bd9Sstevel@tonic-gate size_t ssize; 22897c478bd9Sstevel@tonic-gate 22907c478bd9Sstevel@tonic-gate while (rsize != 0) { 22917c478bd9Sstevel@tonic-gate if (raddr >= seg->s_base + seg->s_size) 22927c478bd9Sstevel@tonic-gate seg = AS_SEGNEXT(as, seg); 22937c478bd9Sstevel@tonic-gate 22947c478bd9Sstevel@tonic-gate if ((raddr + rsize) > (seg->s_base + seg->s_size)) 22957c478bd9Sstevel@tonic-gate ssize = seg->s_base + seg->s_size - raddr; 22967c478bd9Sstevel@tonic-gate else 22977c478bd9Sstevel@tonic-gate ssize = rsize; 22987c478bd9Sstevel@tonic-gate 22997c478bd9Sstevel@tonic-gate as_segunlock(seg, raddr, attr, mlock_map, 0, btopr(ssize)); 23007c478bd9Sstevel@tonic-gate 23017c478bd9Sstevel@tonic-gate rsize -= ssize; 23027c478bd9Sstevel@tonic-gate raddr += ssize; 23037c478bd9Sstevel@tonic-gate } 23047c478bd9Sstevel@tonic-gate } 23057c478bd9Sstevel@tonic-gate 23067c478bd9Sstevel@tonic-gate /* 23077c478bd9Sstevel@tonic-gate * Cache control operations over the interval [addr, addr + size) in 23087c478bd9Sstevel@tonic-gate * address space "as". 23097c478bd9Sstevel@tonic-gate */ 23107c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 23117c478bd9Sstevel@tonic-gate int 23127c478bd9Sstevel@tonic-gate as_ctl(struct as *as, caddr_t addr, size_t size, int func, int attr, 23137c478bd9Sstevel@tonic-gate uintptr_t arg, ulong_t *lock_map, size_t pos) 23147c478bd9Sstevel@tonic-gate { 23157c478bd9Sstevel@tonic-gate struct seg *seg; /* working segment */ 23167c478bd9Sstevel@tonic-gate caddr_t raddr; /* rounded down addr */ 23177c478bd9Sstevel@tonic-gate caddr_t initraddr; /* saved initial rounded down addr */ 23187c478bd9Sstevel@tonic-gate size_t rsize; /* rounded up size */ 23197c478bd9Sstevel@tonic-gate size_t initrsize; /* saved initial rounded up size */ 23207c478bd9Sstevel@tonic-gate size_t ssize; /* size of seg */ 23217c478bd9Sstevel@tonic-gate int error = 0; /* result */ 23227c478bd9Sstevel@tonic-gate size_t mlock_size; /* size of bitmap */ 23237c478bd9Sstevel@tonic-gate ulong_t *mlock_map; /* pointer to bitmap used */ 23247c478bd9Sstevel@tonic-gate /* to represent the locked */ 23257c478bd9Sstevel@tonic-gate /* pages. */ 23267c478bd9Sstevel@tonic-gate retry: 23277c478bd9Sstevel@tonic-gate if (error == IE_RETRY) 23287c478bd9Sstevel@tonic-gate AS_LOCK_ENTER(as, &as->a_lock, RW_WRITER); 23297c478bd9Sstevel@tonic-gate else 23307c478bd9Sstevel@tonic-gate AS_LOCK_ENTER(as, &as->a_lock, RW_READER); 23317c478bd9Sstevel@tonic-gate 23327c478bd9Sstevel@tonic-gate /* 23337c478bd9Sstevel@tonic-gate * If these are address space lock/unlock operations, loop over 23347c478bd9Sstevel@tonic-gate * all segments in the address space, as appropriate. 23357c478bd9Sstevel@tonic-gate */ 23367c478bd9Sstevel@tonic-gate if (func == MC_LOCKAS) { 23377c478bd9Sstevel@tonic-gate size_t npages, idx; 23387c478bd9Sstevel@tonic-gate size_t rlen = 0; /* rounded as length */ 23397c478bd9Sstevel@tonic-gate 23407c478bd9Sstevel@tonic-gate idx = pos; 23417c478bd9Sstevel@tonic-gate 23427c478bd9Sstevel@tonic-gate if (arg & MCL_FUTURE) { 23437c478bd9Sstevel@tonic-gate mutex_enter(&as->a_contents); 23447c478bd9Sstevel@tonic-gate AS_SETPGLCK(as); 23457c478bd9Sstevel@tonic-gate mutex_exit(&as->a_contents); 23467c478bd9Sstevel@tonic-gate } 23477c478bd9Sstevel@tonic-gate if ((arg & MCL_CURRENT) == 0) { 23487c478bd9Sstevel@tonic-gate AS_LOCK_EXIT(as, &as->a_lock); 23497c478bd9Sstevel@tonic-gate return (0); 23507c478bd9Sstevel@tonic-gate } 23517c478bd9Sstevel@tonic-gate 23527c478bd9Sstevel@tonic-gate seg = AS_SEGFIRST(as); 23537c478bd9Sstevel@tonic-gate if (seg == NULL) { 23547c478bd9Sstevel@tonic-gate AS_LOCK_EXIT(as, &as->a_lock); 23557c478bd9Sstevel@tonic-gate return (0); 23567c478bd9Sstevel@tonic-gate } 23577c478bd9Sstevel@tonic-gate 23587c478bd9Sstevel@tonic-gate do { 23597c478bd9Sstevel@tonic-gate raddr = (caddr_t)((uintptr_t)seg->s_base & 23607c478bd9Sstevel@tonic-gate (uintptr_t)PAGEMASK); 23617c478bd9Sstevel@tonic-gate rlen += (((uintptr_t)(seg->s_base + seg->s_size) + 2362843e1988Sjohnlev PAGEOFFSET) & PAGEMASK) - (uintptr_t)raddr; 23637c478bd9Sstevel@tonic-gate } while ((seg = AS_SEGNEXT(as, seg)) != NULL); 23647c478bd9Sstevel@tonic-gate 23657c478bd9Sstevel@tonic-gate mlock_size = BT_BITOUL(btopr(rlen)); 23667c478bd9Sstevel@tonic-gate if ((mlock_map = (ulong_t *)kmem_zalloc(mlock_size * 2367843e1988Sjohnlev sizeof (ulong_t), KM_NOSLEEP)) == NULL) { 23687c478bd9Sstevel@tonic-gate AS_LOCK_EXIT(as, &as->a_lock); 23697c478bd9Sstevel@tonic-gate return (EAGAIN); 23707c478bd9Sstevel@tonic-gate } 23717c478bd9Sstevel@tonic-gate 23727c478bd9Sstevel@tonic-gate for (seg = AS_SEGFIRST(as); seg; seg = AS_SEGNEXT(as, seg)) { 23737c478bd9Sstevel@tonic-gate error = SEGOP_LOCKOP(seg, seg->s_base, 23747c478bd9Sstevel@tonic-gate seg->s_size, attr, MC_LOCK, mlock_map, pos); 23757c478bd9Sstevel@tonic-gate if (error != 0) 23767c478bd9Sstevel@tonic-gate break; 23777c478bd9Sstevel@tonic-gate pos += seg_pages(seg); 23787c478bd9Sstevel@tonic-gate } 23797c478bd9Sstevel@tonic-gate 23807c478bd9Sstevel@tonic-gate if (error) { 23817c478bd9Sstevel@tonic-gate for (seg = AS_SEGFIRST(as); seg != NULL; 2382843e1988Sjohnlev seg = AS_SEGNEXT(as, seg)) { 23837c478bd9Sstevel@tonic-gate 23847c478bd9Sstevel@tonic-gate raddr = (caddr_t)((uintptr_t)seg->s_base & 2385843e1988Sjohnlev (uintptr_t)PAGEMASK); 23867c478bd9Sstevel@tonic-gate npages = seg_pages(seg); 23877c478bd9Sstevel@tonic-gate as_segunlock(seg, raddr, attr, mlock_map, 2388843e1988Sjohnlev idx, npages); 23897c478bd9Sstevel@tonic-gate idx += npages; 23907c478bd9Sstevel@tonic-gate } 23917c478bd9Sstevel@tonic-gate } 23927c478bd9Sstevel@tonic-gate 23937c478bd9Sstevel@tonic-gate kmem_free(mlock_map, mlock_size * sizeof (ulong_t)); 23947c478bd9Sstevel@tonic-gate AS_LOCK_EXIT(as, &as->a_lock); 23957c478bd9Sstevel@tonic-gate goto lockerr; 23967c478bd9Sstevel@tonic-gate } else if (func == MC_UNLOCKAS) { 23977c478bd9Sstevel@tonic-gate mutex_enter(&as->a_contents); 23987c478bd9Sstevel@tonic-gate AS_CLRPGLCK(as); 23997c478bd9Sstevel@tonic-gate mutex_exit(&as->a_contents); 24007c478bd9Sstevel@tonic-gate 24017c478bd9Sstevel@tonic-gate for (seg = AS_SEGFIRST(as); seg; seg = AS_SEGNEXT(as, seg)) { 24027c478bd9Sstevel@tonic-gate error = SEGOP_LOCKOP(seg, seg->s_base, 24037c478bd9Sstevel@tonic-gate seg->s_size, attr, MC_UNLOCK, NULL, 0); 24047c478bd9Sstevel@tonic-gate if (error != 0) 24057c478bd9Sstevel@tonic-gate break; 24067c478bd9Sstevel@tonic-gate } 24077c478bd9Sstevel@tonic-gate 24087c478bd9Sstevel@tonic-gate AS_LOCK_EXIT(as, &as->a_lock); 24097c478bd9Sstevel@tonic-gate goto lockerr; 24107c478bd9Sstevel@tonic-gate } 24117c478bd9Sstevel@tonic-gate 24127c478bd9Sstevel@tonic-gate /* 24137c478bd9Sstevel@tonic-gate * Normalize addresses and sizes. 24147c478bd9Sstevel@tonic-gate */ 24157c478bd9Sstevel@tonic-gate initraddr = raddr = (caddr_t)((uintptr_t)addr & (uintptr_t)PAGEMASK); 24167c478bd9Sstevel@tonic-gate initrsize = rsize = (((size_t)(addr + size) + PAGEOFFSET) & PAGEMASK) - 2417843e1988Sjohnlev (size_t)raddr; 24187c478bd9Sstevel@tonic-gate 24197c478bd9Sstevel@tonic-gate if (raddr + rsize < raddr) { /* check for wraparound */ 24207c478bd9Sstevel@tonic-gate AS_LOCK_EXIT(as, &as->a_lock); 24217c478bd9Sstevel@tonic-gate return (ENOMEM); 24227c478bd9Sstevel@tonic-gate } 24237c478bd9Sstevel@tonic-gate 24247c478bd9Sstevel@tonic-gate /* 24257c478bd9Sstevel@tonic-gate * Get initial segment. 24267c478bd9Sstevel@tonic-gate */ 24277c478bd9Sstevel@tonic-gate if ((seg = as_segat(as, raddr)) == NULL) { 24287c478bd9Sstevel@tonic-gate AS_LOCK_EXIT(as, &as->a_lock); 24297c478bd9Sstevel@tonic-gate return (ENOMEM); 24307c478bd9Sstevel@tonic-gate } 24317c478bd9Sstevel@tonic-gate 24327c478bd9Sstevel@tonic-gate if (func == MC_LOCK) { 24337c478bd9Sstevel@tonic-gate mlock_size = BT_BITOUL(btopr(rsize)); 24347c478bd9Sstevel@tonic-gate if ((mlock_map = (ulong_t *)kmem_zalloc(mlock_size * 2435843e1988Sjohnlev sizeof (ulong_t), KM_NOSLEEP)) == NULL) { 24367c478bd9Sstevel@tonic-gate AS_LOCK_EXIT(as, &as->a_lock); 24377c478bd9Sstevel@tonic-gate return (EAGAIN); 24387c478bd9Sstevel@tonic-gate } 24397c478bd9Sstevel@tonic-gate } 24407c478bd9Sstevel@tonic-gate 24417c478bd9Sstevel@tonic-gate /* 24427c478bd9Sstevel@tonic-gate * Loop over all segments. If a hole in the address range is 24437c478bd9Sstevel@tonic-gate * discovered, then fail. For each segment, perform the appropriate 24447c478bd9Sstevel@tonic-gate * control operation. 24457c478bd9Sstevel@tonic-gate */ 24467c478bd9Sstevel@tonic-gate while (rsize != 0) { 24477c478bd9Sstevel@tonic-gate 24487c478bd9Sstevel@tonic-gate /* 24497c478bd9Sstevel@tonic-gate * Make sure there's no hole, calculate the portion 24507c478bd9Sstevel@tonic-gate * of the next segment to be operated over. 24517c478bd9Sstevel@tonic-gate */ 24527c478bd9Sstevel@tonic-gate if (raddr >= seg->s_base + seg->s_size) { 24537c478bd9Sstevel@tonic-gate seg = AS_SEGNEXT(as, seg); 24547c478bd9Sstevel@tonic-gate if (seg == NULL || raddr != seg->s_base) { 24557c478bd9Sstevel@tonic-gate if (func == MC_LOCK) { 24567c478bd9Sstevel@tonic-gate as_unlockerr(as, attr, mlock_map, 2457843e1988Sjohnlev initraddr, initrsize - rsize); 24587c478bd9Sstevel@tonic-gate kmem_free(mlock_map, 2459843e1988Sjohnlev mlock_size * sizeof (ulong_t)); 24607c478bd9Sstevel@tonic-gate } 24617c478bd9Sstevel@tonic-gate AS_LOCK_EXIT(as, &as->a_lock); 24627c478bd9Sstevel@tonic-gate return (ENOMEM); 24637c478bd9Sstevel@tonic-gate } 24647c478bd9Sstevel@tonic-gate } 24657c478bd9Sstevel@tonic-gate if ((raddr + rsize) > (seg->s_base + seg->s_size)) 24667c478bd9Sstevel@tonic-gate ssize = seg->s_base + seg->s_size - raddr; 24677c478bd9Sstevel@tonic-gate else 24687c478bd9Sstevel@tonic-gate ssize = rsize; 24697c478bd9Sstevel@tonic-gate 24707c478bd9Sstevel@tonic-gate /* 24717c478bd9Sstevel@tonic-gate * Dispatch on specific function. 24727c478bd9Sstevel@tonic-gate */ 24737c478bd9Sstevel@tonic-gate switch (func) { 24747c478bd9Sstevel@tonic-gate 24757c478bd9Sstevel@tonic-gate /* 24767c478bd9Sstevel@tonic-gate * Synchronize cached data from mappings with backing 24777c478bd9Sstevel@tonic-gate * objects. 24787c478bd9Sstevel@tonic-gate */ 24797c478bd9Sstevel@tonic-gate case MC_SYNC: 24807c478bd9Sstevel@tonic-gate if (error = SEGOP_SYNC(seg, raddr, ssize, 24817c478bd9Sstevel@tonic-gate attr, (uint_t)arg)) { 24827c478bd9Sstevel@tonic-gate AS_LOCK_EXIT(as, &as->a_lock); 24837c478bd9Sstevel@tonic-gate return (error); 24847c478bd9Sstevel@tonic-gate } 24857c478bd9Sstevel@tonic-gate break; 24867c478bd9Sstevel@tonic-gate 24877c478bd9Sstevel@tonic-gate /* 24887c478bd9Sstevel@tonic-gate * Lock pages in memory. 24897c478bd9Sstevel@tonic-gate */ 24907c478bd9Sstevel@tonic-gate case MC_LOCK: 24917c478bd9Sstevel@tonic-gate if (error = SEGOP_LOCKOP(seg, raddr, ssize, 2492843e1988Sjohnlev attr, func, mlock_map, pos)) { 24937c478bd9Sstevel@tonic-gate as_unlockerr(as, attr, mlock_map, initraddr, 2494843e1988Sjohnlev initrsize - rsize + ssize); 24957c478bd9Sstevel@tonic-gate kmem_free(mlock_map, mlock_size * 2496843e1988Sjohnlev sizeof (ulong_t)); 24977c478bd9Sstevel@tonic-gate AS_LOCK_EXIT(as, &as->a_lock); 24987c478bd9Sstevel@tonic-gate goto lockerr; 24997c478bd9Sstevel@tonic-gate } 25007c478bd9Sstevel@tonic-gate break; 25017c478bd9Sstevel@tonic-gate 25027c478bd9Sstevel@tonic-gate /* 25037c478bd9Sstevel@tonic-gate * Unlock mapped pages. 25047c478bd9Sstevel@tonic-gate */ 25057c478bd9Sstevel@tonic-gate case MC_UNLOCK: 25067c478bd9Sstevel@tonic-gate (void) SEGOP_LOCKOP(seg, raddr, ssize, attr, func, 2507843e1988Sjohnlev (ulong_t *)NULL, (size_t)NULL); 25087c478bd9Sstevel@tonic-gate break; 25097c478bd9Sstevel@tonic-gate 25107c478bd9Sstevel@tonic-gate /* 25117c478bd9Sstevel@tonic-gate * Store VM advise for mapped pages in segment layer. 25127c478bd9Sstevel@tonic-gate */ 25137c478bd9Sstevel@tonic-gate case MC_ADVISE: 25147c478bd9Sstevel@tonic-gate error = SEGOP_ADVISE(seg, raddr, ssize, (uint_t)arg); 25157c478bd9Sstevel@tonic-gate 25167c478bd9Sstevel@tonic-gate /* 25177c478bd9Sstevel@tonic-gate * Check for regular errors and special retry error 25187c478bd9Sstevel@tonic-gate */ 25197c478bd9Sstevel@tonic-gate if (error) { 25207c478bd9Sstevel@tonic-gate if (error == IE_RETRY) { 25217c478bd9Sstevel@tonic-gate /* 25227c478bd9Sstevel@tonic-gate * Need to acquire writers lock, so 25237c478bd9Sstevel@tonic-gate * have to drop readers lock and start 25247c478bd9Sstevel@tonic-gate * all over again 25257c478bd9Sstevel@tonic-gate */ 25267c478bd9Sstevel@tonic-gate AS_LOCK_EXIT(as, &as->a_lock); 25277c478bd9Sstevel@tonic-gate goto retry; 25287c478bd9Sstevel@tonic-gate } else if (error == IE_REATTACH) { 25297c478bd9Sstevel@tonic-gate /* 25307c478bd9Sstevel@tonic-gate * Find segment for current address 25317c478bd9Sstevel@tonic-gate * because current segment just got 25327c478bd9Sstevel@tonic-gate * split or concatenated 25337c478bd9Sstevel@tonic-gate */ 25347c478bd9Sstevel@tonic-gate seg = as_segat(as, raddr); 25357c478bd9Sstevel@tonic-gate if (seg == NULL) { 25367c478bd9Sstevel@tonic-gate AS_LOCK_EXIT(as, &as->a_lock); 25377c478bd9Sstevel@tonic-gate return (ENOMEM); 25387c478bd9Sstevel@tonic-gate } 25397c478bd9Sstevel@tonic-gate } else { 25407c478bd9Sstevel@tonic-gate /* 25417c478bd9Sstevel@tonic-gate * Regular error 25427c478bd9Sstevel@tonic-gate */ 25437c478bd9Sstevel@tonic-gate AS_LOCK_EXIT(as, &as->a_lock); 25447c478bd9Sstevel@tonic-gate return (error); 25457c478bd9Sstevel@tonic-gate } 25467c478bd9Sstevel@tonic-gate } 25477c478bd9Sstevel@tonic-gate break; 25487c478bd9Sstevel@tonic-gate 25497c478bd9Sstevel@tonic-gate /* 25507c478bd9Sstevel@tonic-gate * Can't happen. 25517c478bd9Sstevel@tonic-gate */ 25527c478bd9Sstevel@tonic-gate default: 25537c478bd9Sstevel@tonic-gate panic("as_ctl: bad operation %d", func); 25547c478bd9Sstevel@tonic-gate /*NOTREACHED*/ 25557c478bd9Sstevel@tonic-gate } 25567c478bd9Sstevel@tonic-gate 25577c478bd9Sstevel@tonic-gate rsize -= ssize; 25587c478bd9Sstevel@tonic-gate raddr += ssize; 25597c478bd9Sstevel@tonic-gate } 25607c478bd9Sstevel@tonic-gate 25617c478bd9Sstevel@tonic-gate if (func == MC_LOCK) 25627c478bd9Sstevel@tonic-gate kmem_free(mlock_map, mlock_size * sizeof (ulong_t)); 25637c478bd9Sstevel@tonic-gate AS_LOCK_EXIT(as, &as->a_lock); 25647c478bd9Sstevel@tonic-gate return (0); 25657c478bd9Sstevel@tonic-gate lockerr: 25667c478bd9Sstevel@tonic-gate 25677c478bd9Sstevel@tonic-gate /* 25687c478bd9Sstevel@tonic-gate * If the lower levels returned EDEADLK for a segment lockop, 25697c478bd9Sstevel@tonic-gate * it means that we should retry the operation. Let's wait 25707c478bd9Sstevel@tonic-gate * a bit also to let the deadlock causing condition clear. 25717c478bd9Sstevel@tonic-gate * This is part of a gross hack to work around a design flaw 25727c478bd9Sstevel@tonic-gate * in the ufs/sds logging code and should go away when the 25737c478bd9Sstevel@tonic-gate * logging code is re-designed to fix the problem. See bug 25747c478bd9Sstevel@tonic-gate * 4125102 for details of the problem. 25757c478bd9Sstevel@tonic-gate */ 25767c478bd9Sstevel@tonic-gate if (error == EDEADLK) { 25777c478bd9Sstevel@tonic-gate delay(deadlk_wait); 25787c478bd9Sstevel@tonic-gate error = 0; 25797c478bd9Sstevel@tonic-gate goto retry; 25807c478bd9Sstevel@tonic-gate } 25817c478bd9Sstevel@tonic-gate return (error); 25827c478bd9Sstevel@tonic-gate } 25837c478bd9Sstevel@tonic-gate 2584843e1988Sjohnlev int 2585843e1988Sjohnlev fc_decode(faultcode_t fault_err) 25867c478bd9Sstevel@tonic-gate { 25877c478bd9Sstevel@tonic-gate int error = 0; 25887c478bd9Sstevel@tonic-gate 25897c478bd9Sstevel@tonic-gate switch (FC_CODE(fault_err)) { 25907c478bd9Sstevel@tonic-gate case FC_OBJERR: 25917c478bd9Sstevel@tonic-gate error = FC_ERRNO(fault_err); 25927c478bd9Sstevel@tonic-gate break; 25937c478bd9Sstevel@tonic-gate case FC_PROT: 25947c478bd9Sstevel@tonic-gate error = EACCES; 25957c478bd9Sstevel@tonic-gate break; 25967c478bd9Sstevel@tonic-gate default: 25977c478bd9Sstevel@tonic-gate error = EFAULT; 25987c478bd9Sstevel@tonic-gate break; 25997c478bd9Sstevel@tonic-gate } 26007c478bd9Sstevel@tonic-gate return (error); 26017c478bd9Sstevel@tonic-gate } 26027c478bd9Sstevel@tonic-gate 2603a98e9dbfSaguzovsk /* 2604a98e9dbfSaguzovsk * Pagelock pages from a range that spans more than 1 segment. Obtain shadow 2605a98e9dbfSaguzovsk * lists from each segment and copy them to one contiguous shadow list (plist) 2606a98e9dbfSaguzovsk * as expected by the caller. Save pointers to per segment shadow lists at 2607a98e9dbfSaguzovsk * the tail of plist so that they can be used during as_pageunlock(). 2608a98e9dbfSaguzovsk */ 2609a98e9dbfSaguzovsk static int 2610a98e9dbfSaguzovsk as_pagelock_segs(struct as *as, struct seg *seg, struct page ***ppp, 2611a98e9dbfSaguzovsk caddr_t addr, size_t size, enum seg_rw rw) 2612a98e9dbfSaguzovsk { 2613a98e9dbfSaguzovsk caddr_t sv_addr = addr; 2614a98e9dbfSaguzovsk size_t sv_size = size; 2615a98e9dbfSaguzovsk struct seg *sv_seg = seg; 2616a98e9dbfSaguzovsk ulong_t segcnt = 1; 2617a98e9dbfSaguzovsk ulong_t cnt; 2618a98e9dbfSaguzovsk size_t ssize; 2619a98e9dbfSaguzovsk pgcnt_t npages = btop(size); 2620a98e9dbfSaguzovsk page_t **plist; 2621a98e9dbfSaguzovsk page_t **pl; 2622a98e9dbfSaguzovsk int error; 2623a98e9dbfSaguzovsk caddr_t eaddr; 2624a98e9dbfSaguzovsk faultcode_t fault_err = 0; 2625a98e9dbfSaguzovsk pgcnt_t pl_off; 2626a98e9dbfSaguzovsk extern struct seg_ops segspt_shmops; 2627a98e9dbfSaguzovsk 2628a98e9dbfSaguzovsk ASSERT(AS_LOCK_HELD(as, &as->a_lock)); 2629a98e9dbfSaguzovsk ASSERT(seg != NULL); 2630a98e9dbfSaguzovsk ASSERT(addr >= seg->s_base && addr < seg->s_base + seg->s_size); 2631a98e9dbfSaguzovsk ASSERT(addr + size > seg->s_base + seg->s_size); 2632a98e9dbfSaguzovsk ASSERT(IS_P2ALIGNED(size, PAGESIZE)); 2633a98e9dbfSaguzovsk ASSERT(IS_P2ALIGNED(addr, PAGESIZE)); 2634a98e9dbfSaguzovsk 2635a98e9dbfSaguzovsk /* 2636a98e9dbfSaguzovsk * Count the number of segments covered by the range we are about to 2637a98e9dbfSaguzovsk * lock. The segment count is used to size the shadow list we return 2638a98e9dbfSaguzovsk * back to the caller. 2639a98e9dbfSaguzovsk */ 2640a98e9dbfSaguzovsk for (; size != 0; size -= ssize, addr += ssize) { 2641a98e9dbfSaguzovsk if (addr >= seg->s_base + seg->s_size) { 2642a98e9dbfSaguzovsk 2643a98e9dbfSaguzovsk seg = AS_SEGNEXT(as, seg); 2644a98e9dbfSaguzovsk if (seg == NULL || addr != seg->s_base) { 2645a98e9dbfSaguzovsk AS_LOCK_EXIT(as, &as->a_lock); 2646a98e9dbfSaguzovsk return (EFAULT); 2647a98e9dbfSaguzovsk } 2648a98e9dbfSaguzovsk /* 2649a98e9dbfSaguzovsk * Do a quick check if subsequent segments 2650a98e9dbfSaguzovsk * will most likely support pagelock. 2651a98e9dbfSaguzovsk */ 2652a98e9dbfSaguzovsk if (seg->s_ops == &segvn_ops) { 2653a98e9dbfSaguzovsk vnode_t *vp; 2654a98e9dbfSaguzovsk 2655a98e9dbfSaguzovsk if (SEGOP_GETVP(seg, addr, &vp) != 0 || 2656a98e9dbfSaguzovsk vp != NULL) { 2657a98e9dbfSaguzovsk AS_LOCK_EXIT(as, &as->a_lock); 2658a98e9dbfSaguzovsk goto slow; 2659a98e9dbfSaguzovsk } 2660a98e9dbfSaguzovsk } else if (seg->s_ops != &segspt_shmops) { 2661a98e9dbfSaguzovsk AS_LOCK_EXIT(as, &as->a_lock); 2662a98e9dbfSaguzovsk goto slow; 2663a98e9dbfSaguzovsk } 2664a98e9dbfSaguzovsk segcnt++; 2665a98e9dbfSaguzovsk } 2666a98e9dbfSaguzovsk if (addr + size > seg->s_base + seg->s_size) { 2667a98e9dbfSaguzovsk ssize = seg->s_base + seg->s_size - addr; 2668a98e9dbfSaguzovsk } else { 2669a98e9dbfSaguzovsk ssize = size; 2670a98e9dbfSaguzovsk } 2671a98e9dbfSaguzovsk } 2672a98e9dbfSaguzovsk ASSERT(segcnt > 1); 2673a98e9dbfSaguzovsk 2674a98e9dbfSaguzovsk plist = kmem_zalloc((npages + segcnt) * sizeof (page_t *), KM_SLEEP); 2675a98e9dbfSaguzovsk 2676a98e9dbfSaguzovsk addr = sv_addr; 2677a98e9dbfSaguzovsk size = sv_size; 2678a98e9dbfSaguzovsk seg = sv_seg; 2679a98e9dbfSaguzovsk 2680a98e9dbfSaguzovsk for (cnt = 0, pl_off = 0; size != 0; size -= ssize, addr += ssize) { 2681a98e9dbfSaguzovsk if (addr >= seg->s_base + seg->s_size) { 2682a98e9dbfSaguzovsk seg = AS_SEGNEXT(as, seg); 2683a98e9dbfSaguzovsk ASSERT(seg != NULL && addr == seg->s_base); 2684a98e9dbfSaguzovsk cnt++; 2685a98e9dbfSaguzovsk ASSERT(cnt < segcnt); 2686a98e9dbfSaguzovsk } 2687a98e9dbfSaguzovsk if (addr + size > seg->s_base + seg->s_size) { 2688a98e9dbfSaguzovsk ssize = seg->s_base + seg->s_size - addr; 2689a98e9dbfSaguzovsk } else { 2690a98e9dbfSaguzovsk ssize = size; 2691a98e9dbfSaguzovsk } 2692a98e9dbfSaguzovsk pl = &plist[npages + cnt]; 2693a98e9dbfSaguzovsk error = SEGOP_PAGELOCK(seg, addr, ssize, (page_t ***)pl, 2694a98e9dbfSaguzovsk L_PAGELOCK, rw); 2695a98e9dbfSaguzovsk if (error) { 2696a98e9dbfSaguzovsk break; 2697a98e9dbfSaguzovsk } 2698a98e9dbfSaguzovsk ASSERT(plist[npages + cnt] != NULL); 2699a98e9dbfSaguzovsk ASSERT(pl_off + btop(ssize) <= npages); 2700a98e9dbfSaguzovsk bcopy(plist[npages + cnt], &plist[pl_off], 2701a98e9dbfSaguzovsk btop(ssize) * sizeof (page_t *)); 2702a98e9dbfSaguzovsk pl_off += btop(ssize); 2703a98e9dbfSaguzovsk } 2704a98e9dbfSaguzovsk 2705a98e9dbfSaguzovsk if (size == 0) { 2706a98e9dbfSaguzovsk AS_LOCK_EXIT(as, &as->a_lock); 2707a98e9dbfSaguzovsk ASSERT(cnt == segcnt - 1); 2708a98e9dbfSaguzovsk *ppp = plist; 2709a98e9dbfSaguzovsk return (0); 2710a98e9dbfSaguzovsk } 2711a98e9dbfSaguzovsk 2712a98e9dbfSaguzovsk /* 2713a98e9dbfSaguzovsk * one of pagelock calls failed. The error type is in error variable. 2714a98e9dbfSaguzovsk * Unlock what we've locked so far and retry with F_SOFTLOCK if error 2715a98e9dbfSaguzovsk * type is either EFAULT or ENOTSUP. Otherwise just return the error 2716a98e9dbfSaguzovsk * back to the caller. 2717a98e9dbfSaguzovsk */ 2718a98e9dbfSaguzovsk 2719a98e9dbfSaguzovsk eaddr = addr; 2720a98e9dbfSaguzovsk seg = sv_seg; 2721a98e9dbfSaguzovsk 2722a98e9dbfSaguzovsk for (cnt = 0, addr = sv_addr; addr < eaddr; addr += ssize) { 2723a98e9dbfSaguzovsk if (addr >= seg->s_base + seg->s_size) { 2724a98e9dbfSaguzovsk seg = AS_SEGNEXT(as, seg); 2725a98e9dbfSaguzovsk ASSERT(seg != NULL && addr == seg->s_base); 2726a98e9dbfSaguzovsk cnt++; 2727a98e9dbfSaguzovsk ASSERT(cnt < segcnt); 2728a98e9dbfSaguzovsk } 2729a98e9dbfSaguzovsk if (eaddr > seg->s_base + seg->s_size) { 2730a98e9dbfSaguzovsk ssize = seg->s_base + seg->s_size - addr; 2731a98e9dbfSaguzovsk } else { 2732a98e9dbfSaguzovsk ssize = eaddr - addr; 2733a98e9dbfSaguzovsk } 2734a98e9dbfSaguzovsk pl = &plist[npages + cnt]; 2735a98e9dbfSaguzovsk ASSERT(*pl != NULL); 2736a98e9dbfSaguzovsk (void) SEGOP_PAGELOCK(seg, addr, ssize, (page_t ***)pl, 2737a98e9dbfSaguzovsk L_PAGEUNLOCK, rw); 2738a98e9dbfSaguzovsk } 2739a98e9dbfSaguzovsk 2740a98e9dbfSaguzovsk AS_LOCK_EXIT(as, &as->a_lock); 2741a98e9dbfSaguzovsk 2742a98e9dbfSaguzovsk kmem_free(plist, (npages + segcnt) * sizeof (page_t *)); 2743a98e9dbfSaguzovsk 2744a98e9dbfSaguzovsk if (error != ENOTSUP && error != EFAULT) { 2745a98e9dbfSaguzovsk return (error); 2746a98e9dbfSaguzovsk } 2747a98e9dbfSaguzovsk 2748a98e9dbfSaguzovsk slow: 2749a98e9dbfSaguzovsk /* 2750a98e9dbfSaguzovsk * If we are here because pagelock failed due to the need to cow fault 2751a98e9dbfSaguzovsk * in the pages we want to lock F_SOFTLOCK will do this job and in 2752a98e9dbfSaguzovsk * next as_pagelock() call for this address range pagelock will 2753a98e9dbfSaguzovsk * hopefully succeed. 2754a98e9dbfSaguzovsk */ 2755a98e9dbfSaguzovsk fault_err = as_fault(as->a_hat, as, sv_addr, sv_size, F_SOFTLOCK, rw); 2756a98e9dbfSaguzovsk if (fault_err != 0) { 2757a98e9dbfSaguzovsk return (fc_decode(fault_err)); 2758a98e9dbfSaguzovsk } 2759a98e9dbfSaguzovsk *ppp = NULL; 2760a98e9dbfSaguzovsk 2761a98e9dbfSaguzovsk return (0); 2762a98e9dbfSaguzovsk } 2763a98e9dbfSaguzovsk 27647c478bd9Sstevel@tonic-gate /* 27657c478bd9Sstevel@tonic-gate * lock pages in a given address space. Return shadow list. If 27667c478bd9Sstevel@tonic-gate * the list is NULL, the MMU mapping is also locked. 27677c478bd9Sstevel@tonic-gate */ 27687c478bd9Sstevel@tonic-gate int 27697c478bd9Sstevel@tonic-gate as_pagelock(struct as *as, struct page ***ppp, caddr_t addr, 27707c478bd9Sstevel@tonic-gate size_t size, enum seg_rw rw) 27717c478bd9Sstevel@tonic-gate { 27727c478bd9Sstevel@tonic-gate size_t rsize; 27737c478bd9Sstevel@tonic-gate caddr_t raddr; 27747c478bd9Sstevel@tonic-gate faultcode_t fault_err; 27757c478bd9Sstevel@tonic-gate struct seg *seg; 2776a98e9dbfSaguzovsk int err; 27777c478bd9Sstevel@tonic-gate 27787c478bd9Sstevel@tonic-gate TRACE_2(TR_FAC_PHYSIO, TR_PHYSIO_AS_LOCK_START, 27797c478bd9Sstevel@tonic-gate "as_pagelock_start: addr %p size %ld", addr, size); 27807c478bd9Sstevel@tonic-gate 27817c478bd9Sstevel@tonic-gate raddr = (caddr_t)((uintptr_t)addr & (uintptr_t)PAGEMASK); 27827c478bd9Sstevel@tonic-gate rsize = (((size_t)(addr + size) + PAGEOFFSET) & PAGEMASK) - 2783843e1988Sjohnlev (size_t)raddr; 2784a98e9dbfSaguzovsk 27857c478bd9Sstevel@tonic-gate /* 27867c478bd9Sstevel@tonic-gate * if the request crosses two segments let 27877c478bd9Sstevel@tonic-gate * as_fault handle it. 27887c478bd9Sstevel@tonic-gate */ 27897c478bd9Sstevel@tonic-gate AS_LOCK_ENTER(as, &as->a_lock, RW_READER); 2790a98e9dbfSaguzovsk 2791a98e9dbfSaguzovsk seg = as_segat(as, raddr); 2792a98e9dbfSaguzovsk if (seg == NULL) { 2793a98e9dbfSaguzovsk AS_LOCK_EXIT(as, &as->a_lock); 2794a98e9dbfSaguzovsk return (EFAULT); 2795a98e9dbfSaguzovsk } 2796a98e9dbfSaguzovsk ASSERT(raddr >= seg->s_base && raddr < seg->s_base + seg->s_size); 2797a98e9dbfSaguzovsk if (raddr + rsize > seg->s_base + seg->s_size) { 2798a98e9dbfSaguzovsk return (as_pagelock_segs(as, seg, ppp, raddr, rsize, rw)); 2799a98e9dbfSaguzovsk } 2800a98e9dbfSaguzovsk if (raddr + rsize <= raddr) { 28017c478bd9Sstevel@tonic-gate AS_LOCK_EXIT(as, &as->a_lock); 2802a98e9dbfSaguzovsk return (EFAULT); 28037c478bd9Sstevel@tonic-gate } 28047c478bd9Sstevel@tonic-gate 28057c478bd9Sstevel@tonic-gate TRACE_2(TR_FAC_PHYSIO, TR_PHYSIO_SEG_LOCK_START, 28067c478bd9Sstevel@tonic-gate "seg_lock_1_start: raddr %p rsize %ld", raddr, rsize); 28077c478bd9Sstevel@tonic-gate 28087c478bd9Sstevel@tonic-gate /* 28097c478bd9Sstevel@tonic-gate * try to lock pages and pass back shadow list 28107c478bd9Sstevel@tonic-gate */ 2811a98e9dbfSaguzovsk err = SEGOP_PAGELOCK(seg, raddr, rsize, ppp, L_PAGELOCK, rw); 28127c478bd9Sstevel@tonic-gate 28137c478bd9Sstevel@tonic-gate TRACE_0(TR_FAC_PHYSIO, TR_PHYSIO_SEG_LOCK_END, "seg_lock_1_end"); 28147c478bd9Sstevel@tonic-gate 2815a98e9dbfSaguzovsk AS_LOCK_EXIT(as, &as->a_lock); 28167c478bd9Sstevel@tonic-gate 2817a98e9dbfSaguzovsk if (err == 0 || (err != ENOTSUP && err != EFAULT)) { 2818a98e9dbfSaguzovsk return (err); 28197c478bd9Sstevel@tonic-gate } 28207c478bd9Sstevel@tonic-gate 28217c478bd9Sstevel@tonic-gate /* 2822a98e9dbfSaguzovsk * Use F_SOFTLOCK to lock the pages because pagelock failed either due 2823a98e9dbfSaguzovsk * to no pagelock support for this segment or pages need to be cow 2824a98e9dbfSaguzovsk * faulted in. If fault is needed F_SOFTLOCK will do this job for 2825a98e9dbfSaguzovsk * this as_pagelock() call and in the next as_pagelock() call for the 2826a98e9dbfSaguzovsk * same address range pagelock call will hopefull succeed. 28277c478bd9Sstevel@tonic-gate */ 28287c478bd9Sstevel@tonic-gate fault_err = as_fault(as->a_hat, as, addr, size, F_SOFTLOCK, rw); 28297c478bd9Sstevel@tonic-gate if (fault_err != 0) { 2830843e1988Sjohnlev return (fc_decode(fault_err)); 28317c478bd9Sstevel@tonic-gate } 28327c478bd9Sstevel@tonic-gate *ppp = NULL; 28337c478bd9Sstevel@tonic-gate 28347c478bd9Sstevel@tonic-gate TRACE_0(TR_FAC_PHYSIO, TR_PHYSIO_AS_LOCK_END, "as_pagelock_end"); 28357c478bd9Sstevel@tonic-gate return (0); 28367c478bd9Sstevel@tonic-gate } 28377c478bd9Sstevel@tonic-gate 2838a98e9dbfSaguzovsk /* 2839a98e9dbfSaguzovsk * unlock pages locked by as_pagelock_segs(). Retrieve per segment shadow 2840a98e9dbfSaguzovsk * lists from the end of plist and call pageunlock interface for each segment. 2841a98e9dbfSaguzovsk * Drop as lock and free plist. 2842a98e9dbfSaguzovsk */ 2843a98e9dbfSaguzovsk static void 2844a98e9dbfSaguzovsk as_pageunlock_segs(struct as *as, struct seg *seg, caddr_t addr, size_t size, 2845a98e9dbfSaguzovsk struct page **plist, enum seg_rw rw) 2846a98e9dbfSaguzovsk { 2847a98e9dbfSaguzovsk ulong_t cnt; 2848a98e9dbfSaguzovsk caddr_t eaddr = addr + size; 2849a98e9dbfSaguzovsk pgcnt_t npages = btop(size); 2850a98e9dbfSaguzovsk size_t ssize; 2851a98e9dbfSaguzovsk page_t **pl; 2852a98e9dbfSaguzovsk 2853a98e9dbfSaguzovsk ASSERT(AS_LOCK_HELD(as, &as->a_lock)); 2854a98e9dbfSaguzovsk ASSERT(seg != NULL); 2855a98e9dbfSaguzovsk ASSERT(addr >= seg->s_base && addr < seg->s_base + seg->s_size); 2856a98e9dbfSaguzovsk ASSERT(addr + size > seg->s_base + seg->s_size); 2857a98e9dbfSaguzovsk ASSERT(IS_P2ALIGNED(size, PAGESIZE)); 2858a98e9dbfSaguzovsk ASSERT(IS_P2ALIGNED(addr, PAGESIZE)); 2859a98e9dbfSaguzovsk ASSERT(plist != NULL); 2860a98e9dbfSaguzovsk 2861a98e9dbfSaguzovsk for (cnt = 0; addr < eaddr; addr += ssize) { 2862a98e9dbfSaguzovsk if (addr >= seg->s_base + seg->s_size) { 2863a98e9dbfSaguzovsk seg = AS_SEGNEXT(as, seg); 2864a98e9dbfSaguzovsk ASSERT(seg != NULL && addr == seg->s_base); 2865a98e9dbfSaguzovsk cnt++; 2866a98e9dbfSaguzovsk } 2867a98e9dbfSaguzovsk if (eaddr > seg->s_base + seg->s_size) { 2868a98e9dbfSaguzovsk ssize = seg->s_base + seg->s_size - addr; 2869a98e9dbfSaguzovsk } else { 2870a98e9dbfSaguzovsk ssize = eaddr - addr; 2871a98e9dbfSaguzovsk } 2872a98e9dbfSaguzovsk pl = &plist[npages + cnt]; 2873a98e9dbfSaguzovsk ASSERT(*pl != NULL); 2874a98e9dbfSaguzovsk (void) SEGOP_PAGELOCK(seg, addr, ssize, (page_t ***)pl, 2875a98e9dbfSaguzovsk L_PAGEUNLOCK, rw); 2876a98e9dbfSaguzovsk } 2877a98e9dbfSaguzovsk ASSERT(cnt > 0); 2878a98e9dbfSaguzovsk AS_LOCK_EXIT(as, &as->a_lock); 2879a98e9dbfSaguzovsk 2880a98e9dbfSaguzovsk cnt++; 2881a98e9dbfSaguzovsk kmem_free(plist, (npages + cnt) * sizeof (page_t *)); 2882a98e9dbfSaguzovsk } 2883a98e9dbfSaguzovsk 28847c478bd9Sstevel@tonic-gate /* 28857c478bd9Sstevel@tonic-gate * unlock pages in a given address range 28867c478bd9Sstevel@tonic-gate */ 28877c478bd9Sstevel@tonic-gate void 28887c478bd9Sstevel@tonic-gate as_pageunlock(struct as *as, struct page **pp, caddr_t addr, size_t size, 28897c478bd9Sstevel@tonic-gate enum seg_rw rw) 28907c478bd9Sstevel@tonic-gate { 28917c478bd9Sstevel@tonic-gate struct seg *seg; 28927c478bd9Sstevel@tonic-gate size_t rsize; 28937c478bd9Sstevel@tonic-gate caddr_t raddr; 28947c478bd9Sstevel@tonic-gate 28957c478bd9Sstevel@tonic-gate TRACE_2(TR_FAC_PHYSIO, TR_PHYSIO_AS_UNLOCK_START, 28967c478bd9Sstevel@tonic-gate "as_pageunlock_start: addr %p size %ld", addr, size); 28977c478bd9Sstevel@tonic-gate 28987c478bd9Sstevel@tonic-gate /* 28997c478bd9Sstevel@tonic-gate * if the shadow list is NULL, as_pagelock was 29007c478bd9Sstevel@tonic-gate * falling back to as_fault 29017c478bd9Sstevel@tonic-gate */ 29027c478bd9Sstevel@tonic-gate if (pp == NULL) { 29037c478bd9Sstevel@tonic-gate (void) as_fault(as->a_hat, as, addr, size, F_SOFTUNLOCK, rw); 29047c478bd9Sstevel@tonic-gate return; 29057c478bd9Sstevel@tonic-gate } 2906a98e9dbfSaguzovsk 29077c478bd9Sstevel@tonic-gate raddr = (caddr_t)((uintptr_t)addr & (uintptr_t)PAGEMASK); 29087c478bd9Sstevel@tonic-gate rsize = (((size_t)(addr + size) + PAGEOFFSET) & PAGEMASK) - 2909843e1988Sjohnlev (size_t)raddr; 2910a98e9dbfSaguzovsk 29117c478bd9Sstevel@tonic-gate AS_LOCK_ENTER(as, &as->a_lock, RW_READER); 2912a98e9dbfSaguzovsk seg = as_segat(as, raddr); 2913a98e9dbfSaguzovsk ASSERT(seg != NULL); 2914a98e9dbfSaguzovsk 29157c478bd9Sstevel@tonic-gate TRACE_2(TR_FAC_PHYSIO, TR_PHYSIO_SEG_UNLOCK_START, 29167c478bd9Sstevel@tonic-gate "seg_unlock_start: raddr %p rsize %ld", raddr, rsize); 2917a98e9dbfSaguzovsk 2918a98e9dbfSaguzovsk ASSERT(raddr >= seg->s_base && raddr < seg->s_base + seg->s_size); 2919a98e9dbfSaguzovsk if (raddr + rsize <= seg->s_base + seg->s_size) { 2920a98e9dbfSaguzovsk SEGOP_PAGELOCK(seg, raddr, rsize, &pp, L_PAGEUNLOCK, rw); 2921a98e9dbfSaguzovsk } else { 2922a98e9dbfSaguzovsk as_pageunlock_segs(as, seg, raddr, rsize, pp, rw); 2923a98e9dbfSaguzovsk return; 2924a98e9dbfSaguzovsk } 29257c478bd9Sstevel@tonic-gate AS_LOCK_EXIT(as, &as->a_lock); 29267c478bd9Sstevel@tonic-gate TRACE_0(TR_FAC_PHYSIO, TR_PHYSIO_AS_UNLOCK_END, "as_pageunlock_end"); 29277c478bd9Sstevel@tonic-gate } 29287c478bd9Sstevel@tonic-gate 29297c478bd9Sstevel@tonic-gate int 29307c478bd9Sstevel@tonic-gate as_setpagesize(struct as *as, caddr_t addr, size_t size, uint_t szc, 29317c478bd9Sstevel@tonic-gate boolean_t wait) 29327c478bd9Sstevel@tonic-gate { 29337c478bd9Sstevel@tonic-gate struct seg *seg; 29347c478bd9Sstevel@tonic-gate size_t ssize; 29357c478bd9Sstevel@tonic-gate caddr_t raddr; /* rounded down addr */ 29367c478bd9Sstevel@tonic-gate size_t rsize; /* rounded up size */ 29377c478bd9Sstevel@tonic-gate int error = 0; 29387c478bd9Sstevel@tonic-gate size_t pgsz = page_get_pagesize(szc); 29397c478bd9Sstevel@tonic-gate 29407c478bd9Sstevel@tonic-gate setpgsz_top: 29417c478bd9Sstevel@tonic-gate if (!IS_P2ALIGNED(addr, pgsz) || !IS_P2ALIGNED(size, pgsz)) { 29427c478bd9Sstevel@tonic-gate return (EINVAL); 29437c478bd9Sstevel@tonic-gate } 29447c478bd9Sstevel@tonic-gate 29457c478bd9Sstevel@tonic-gate raddr = addr; 29467c478bd9Sstevel@tonic-gate rsize = size; 29477c478bd9Sstevel@tonic-gate 29487c478bd9Sstevel@tonic-gate if (raddr + rsize < raddr) /* check for wraparound */ 29497c478bd9Sstevel@tonic-gate return (ENOMEM); 29507c478bd9Sstevel@tonic-gate 29517c478bd9Sstevel@tonic-gate AS_LOCK_ENTER(as, &as->a_lock, RW_WRITER); 29527c478bd9Sstevel@tonic-gate as_clearwatchprot(as, raddr, rsize); 29537c478bd9Sstevel@tonic-gate seg = as_segat(as, raddr); 29547c478bd9Sstevel@tonic-gate if (seg == NULL) { 29557c478bd9Sstevel@tonic-gate as_setwatch(as); 29567c478bd9Sstevel@tonic-gate AS_LOCK_EXIT(as, &as->a_lock); 29577c478bd9Sstevel@tonic-gate return (ENOMEM); 29587c478bd9Sstevel@tonic-gate } 29597c478bd9Sstevel@tonic-gate 29607c478bd9Sstevel@tonic-gate for (; rsize != 0; rsize -= ssize, raddr += ssize) { 29617c478bd9Sstevel@tonic-gate if (raddr >= seg->s_base + seg->s_size) { 29627c478bd9Sstevel@tonic-gate seg = AS_SEGNEXT(as, seg); 29637c478bd9Sstevel@tonic-gate if (seg == NULL || raddr != seg->s_base) { 29647c478bd9Sstevel@tonic-gate error = ENOMEM; 29657c478bd9Sstevel@tonic-gate break; 29667c478bd9Sstevel@tonic-gate } 29677c478bd9Sstevel@tonic-gate } 29687c478bd9Sstevel@tonic-gate if ((raddr + rsize) > (seg->s_base + seg->s_size)) { 29697c478bd9Sstevel@tonic-gate ssize = seg->s_base + seg->s_size - raddr; 29707c478bd9Sstevel@tonic-gate } else { 29717c478bd9Sstevel@tonic-gate ssize = rsize; 29727c478bd9Sstevel@tonic-gate } 29737c478bd9Sstevel@tonic-gate 2974a98e9dbfSaguzovsk retry: 29757c478bd9Sstevel@tonic-gate error = SEGOP_SETPAGESIZE(seg, raddr, ssize, szc); 29767c478bd9Sstevel@tonic-gate 29777c478bd9Sstevel@tonic-gate if (error == IE_NOMEM) { 29787c478bd9Sstevel@tonic-gate error = EAGAIN; 29797c478bd9Sstevel@tonic-gate break; 29807c478bd9Sstevel@tonic-gate } 29817c478bd9Sstevel@tonic-gate 29827c478bd9Sstevel@tonic-gate if (error == IE_RETRY) { 29837c478bd9Sstevel@tonic-gate AS_LOCK_EXIT(as, &as->a_lock); 29847c478bd9Sstevel@tonic-gate goto setpgsz_top; 29857c478bd9Sstevel@tonic-gate } 29867c478bd9Sstevel@tonic-gate 29877c478bd9Sstevel@tonic-gate if (error == ENOTSUP) { 29887c478bd9Sstevel@tonic-gate error = EINVAL; 29897c478bd9Sstevel@tonic-gate break; 29907c478bd9Sstevel@tonic-gate } 29917c478bd9Sstevel@tonic-gate 29927c478bd9Sstevel@tonic-gate if (wait && (error == EAGAIN)) { 29937c478bd9Sstevel@tonic-gate /* 29947c478bd9Sstevel@tonic-gate * Memory is currently locked. It must be unlocked 29957c478bd9Sstevel@tonic-gate * before this operation can succeed through a retry. 29967c478bd9Sstevel@tonic-gate * The possible reasons for locked memory and 29977c478bd9Sstevel@tonic-gate * corresponding strategies for unlocking are: 29987c478bd9Sstevel@tonic-gate * (1) Normal I/O 29997c478bd9Sstevel@tonic-gate * wait for a signal that the I/O operation 30007c478bd9Sstevel@tonic-gate * has completed and the memory is unlocked. 30017c478bd9Sstevel@tonic-gate * (2) Asynchronous I/O 30027c478bd9Sstevel@tonic-gate * The aio subsystem does not unlock pages when 30037c478bd9Sstevel@tonic-gate * the I/O is completed. Those pages are unlocked 30047c478bd9Sstevel@tonic-gate * when the application calls aiowait/aioerror. 30057c478bd9Sstevel@tonic-gate * So, to prevent blocking forever, cv_broadcast() 30067c478bd9Sstevel@tonic-gate * is done to wake up aio_cleanup_thread. 30077c478bd9Sstevel@tonic-gate * Subsequently, segvn_reclaim will be called, and 30087c478bd9Sstevel@tonic-gate * that will do AS_CLRUNMAPWAIT() and wake us up. 30097c478bd9Sstevel@tonic-gate * (3) Long term page locking: 30107c478bd9Sstevel@tonic-gate * This is not relevant for as_setpagesize() 30117c478bd9Sstevel@tonic-gate * because we cannot change the page size for 30127c478bd9Sstevel@tonic-gate * driver memory. The attempt to do so will 30137c478bd9Sstevel@tonic-gate * fail with a different error than EAGAIN so 30147c478bd9Sstevel@tonic-gate * there's no need to trigger as callbacks like 30157c478bd9Sstevel@tonic-gate * as_unmap, as_setprot or as_free would do. 30167c478bd9Sstevel@tonic-gate */ 30177c478bd9Sstevel@tonic-gate mutex_enter(&as->a_contents); 3018a98e9dbfSaguzovsk if (!AS_ISNOUNMAPWAIT(as)) { 3019a98e9dbfSaguzovsk if (AS_ISUNMAPWAIT(as) == 0) { 3020a98e9dbfSaguzovsk cv_broadcast(&as->a_cv); 3021a98e9dbfSaguzovsk } 3022a98e9dbfSaguzovsk AS_SETUNMAPWAIT(as); 3023a98e9dbfSaguzovsk AS_LOCK_EXIT(as, &as->a_lock); 3024a98e9dbfSaguzovsk while (AS_ISUNMAPWAIT(as)) { 3025a98e9dbfSaguzovsk cv_wait(&as->a_cv, &as->a_contents); 3026a98e9dbfSaguzovsk } 3027a98e9dbfSaguzovsk } else { 3028a98e9dbfSaguzovsk /* 3029a98e9dbfSaguzovsk * We may have raced with 3030a98e9dbfSaguzovsk * segvn_reclaim()/segspt_reclaim(). In this 3031a98e9dbfSaguzovsk * case clean nounmapwait flag and retry since 3032a98e9dbfSaguzovsk * softlockcnt in this segment may be already 3033a98e9dbfSaguzovsk * 0. We don't drop as writer lock so our 3034a98e9dbfSaguzovsk * number of retries without sleeping should 3035a98e9dbfSaguzovsk * be very small. See segvn_reclaim() for 3036a98e9dbfSaguzovsk * more comments. 3037a98e9dbfSaguzovsk */ 3038a98e9dbfSaguzovsk AS_CLRNOUNMAPWAIT(as); 3039a98e9dbfSaguzovsk mutex_exit(&as->a_contents); 3040a98e9dbfSaguzovsk goto retry; 30417c478bd9Sstevel@tonic-gate } 30427c478bd9Sstevel@tonic-gate mutex_exit(&as->a_contents); 30437c478bd9Sstevel@tonic-gate goto setpgsz_top; 30447c478bd9Sstevel@tonic-gate } else if (error != 0) { 30457c478bd9Sstevel@tonic-gate break; 30467c478bd9Sstevel@tonic-gate } 30477c478bd9Sstevel@tonic-gate } 30487c478bd9Sstevel@tonic-gate as_setwatch(as); 30497c478bd9Sstevel@tonic-gate AS_LOCK_EXIT(as, &as->a_lock); 30507c478bd9Sstevel@tonic-gate return (error); 30517c478bd9Sstevel@tonic-gate } 30527c478bd9Sstevel@tonic-gate 3053ec25b48fSsusans /* 3054ec25b48fSsusans * as_iset3_default_lpsize() just calls SEGOP_SETPAGESIZE() on all segments 3055ec25b48fSsusans * in its chunk where s_szc is less than the szc we want to set. 3056ec25b48fSsusans */ 3057ec25b48fSsusans static int 3058ec25b48fSsusans as_iset3_default_lpsize(struct as *as, caddr_t raddr, size_t rsize, uint_t szc, 3059ec25b48fSsusans int *retry) 3060ec25b48fSsusans { 3061ec25b48fSsusans struct seg *seg; 3062ec25b48fSsusans size_t ssize; 3063ec25b48fSsusans int error; 3064ec25b48fSsusans 3065a98e9dbfSaguzovsk ASSERT(AS_WRITE_HELD(as, &as->a_lock)); 3066a98e9dbfSaguzovsk 3067ec25b48fSsusans seg = as_segat(as, raddr); 3068ec25b48fSsusans if (seg == NULL) { 3069ec25b48fSsusans panic("as_iset3_default_lpsize: no seg"); 3070ec25b48fSsusans } 3071ec25b48fSsusans 3072ec25b48fSsusans for (; rsize != 0; rsize -= ssize, raddr += ssize) { 3073ec25b48fSsusans if (raddr >= seg->s_base + seg->s_size) { 3074ec25b48fSsusans seg = AS_SEGNEXT(as, seg); 3075ec25b48fSsusans if (seg == NULL || raddr != seg->s_base) { 3076ec25b48fSsusans panic("as_iset3_default_lpsize: as changed"); 3077ec25b48fSsusans } 3078ec25b48fSsusans } 3079ec25b48fSsusans if ((raddr + rsize) > (seg->s_base + seg->s_size)) { 3080ec25b48fSsusans ssize = seg->s_base + seg->s_size - raddr; 3081ec25b48fSsusans } else { 3082ec25b48fSsusans ssize = rsize; 3083ec25b48fSsusans } 3084ec25b48fSsusans 3085ec25b48fSsusans if (szc > seg->s_szc) { 3086ec25b48fSsusans error = SEGOP_SETPAGESIZE(seg, raddr, ssize, szc); 3087ec25b48fSsusans /* Only retry on EINVAL segments that have no vnode. */ 3088ec25b48fSsusans if (error == EINVAL) { 3089ec25b48fSsusans vnode_t *vp = NULL; 3090ec25b48fSsusans if ((SEGOP_GETTYPE(seg, raddr) & MAP_SHARED) && 3091ec25b48fSsusans (SEGOP_GETVP(seg, raddr, &vp) != 0 || 3092ec25b48fSsusans vp == NULL)) { 3093ec25b48fSsusans *retry = 1; 3094ec25b48fSsusans } else { 3095ec25b48fSsusans *retry = 0; 3096ec25b48fSsusans } 3097ec25b48fSsusans } 3098ec25b48fSsusans if (error) { 3099ec25b48fSsusans return (error); 3100ec25b48fSsusans } 3101ec25b48fSsusans } 3102ec25b48fSsusans } 3103ec25b48fSsusans return (0); 3104ec25b48fSsusans } 3105ec25b48fSsusans 3106ec25b48fSsusans /* 3107ec25b48fSsusans * as_iset2_default_lpsize() calls as_iset3_default_lpsize() to set the 3108ec25b48fSsusans * pagesize on each segment in its range, but if any fails with EINVAL, 3109ec25b48fSsusans * then it reduces the pagesizes to the next size in the bitmap and 3110ec25b48fSsusans * retries as_iset3_default_lpsize(). The reason why the code retries 3111ec25b48fSsusans * smaller allowed sizes on EINVAL is because (a) the anon offset may not 3112ec25b48fSsusans * match the bigger sizes, and (b) it's hard to get this offset (to begin 3113ec25b48fSsusans * with) to pass to map_pgszcvec(). 3114ec25b48fSsusans */ 3115ec25b48fSsusans static int 3116ec25b48fSsusans as_iset2_default_lpsize(struct as *as, caddr_t addr, size_t size, uint_t szc, 3117ec25b48fSsusans uint_t szcvec) 3118ec25b48fSsusans { 3119ec25b48fSsusans int error; 3120ec25b48fSsusans int retry; 3121ec25b48fSsusans 3122a98e9dbfSaguzovsk ASSERT(AS_WRITE_HELD(as, &as->a_lock)); 3123a98e9dbfSaguzovsk 3124ec25b48fSsusans for (;;) { 3125ec25b48fSsusans error = as_iset3_default_lpsize(as, addr, size, szc, &retry); 3126ec25b48fSsusans if (error == EINVAL && retry) { 3127ec25b48fSsusans szcvec &= ~(1 << szc); 3128ec25b48fSsusans if (szcvec <= 1) { 3129ec25b48fSsusans return (EINVAL); 3130ec25b48fSsusans } 3131ec25b48fSsusans szc = highbit(szcvec) - 1; 3132ec25b48fSsusans } else { 3133ec25b48fSsusans return (error); 3134ec25b48fSsusans } 3135ec25b48fSsusans } 3136ec25b48fSsusans } 3137ec25b48fSsusans 3138ec25b48fSsusans /* 3139ec25b48fSsusans * as_iset1_default_lpsize() breaks its chunk into areas where existing 3140ec25b48fSsusans * segments have a smaller szc than we want to set. For each such area, 3141ec25b48fSsusans * it calls as_iset2_default_lpsize() 3142ec25b48fSsusans */ 3143ec25b48fSsusans static int 3144ec25b48fSsusans as_iset1_default_lpsize(struct as *as, caddr_t raddr, size_t rsize, uint_t szc, 3145ec25b48fSsusans uint_t szcvec) 3146ec25b48fSsusans { 3147ec25b48fSsusans struct seg *seg; 3148ec25b48fSsusans size_t ssize; 3149ec25b48fSsusans caddr_t setaddr = raddr; 3150ec25b48fSsusans size_t setsize = 0; 3151ec25b48fSsusans int set; 3152ec25b48fSsusans int error; 3153ec25b48fSsusans 3154ec25b48fSsusans ASSERT(AS_WRITE_HELD(as, &as->a_lock)); 3155ec25b48fSsusans 3156ec25b48fSsusans seg = as_segat(as, raddr); 3157ec25b48fSsusans if (seg == NULL) { 3158ec25b48fSsusans panic("as_iset1_default_lpsize: no seg"); 3159ec25b48fSsusans } 3160ec25b48fSsusans if (seg->s_szc < szc) { 3161ec25b48fSsusans set = 1; 3162ec25b48fSsusans } else { 3163ec25b48fSsusans set = 0; 3164ec25b48fSsusans } 3165ec25b48fSsusans 3166ec25b48fSsusans for (; rsize != 0; rsize -= ssize, raddr += ssize, setsize += ssize) { 3167ec25b48fSsusans if (raddr >= seg->s_base + seg->s_size) { 3168ec25b48fSsusans seg = AS_SEGNEXT(as, seg); 3169ec25b48fSsusans if (seg == NULL || raddr != seg->s_base) { 3170ec25b48fSsusans panic("as_iset1_default_lpsize: as changed"); 3171ec25b48fSsusans } 3172ec25b48fSsusans if (seg->s_szc >= szc && set) { 3173ec25b48fSsusans ASSERT(setsize != 0); 3174ec25b48fSsusans error = as_iset2_default_lpsize(as, 3175ec25b48fSsusans setaddr, setsize, szc, szcvec); 3176ec25b48fSsusans if (error) { 3177ec25b48fSsusans return (error); 3178ec25b48fSsusans } 3179ec25b48fSsusans set = 0; 3180ec25b48fSsusans } else if (seg->s_szc < szc && !set) { 3181ec25b48fSsusans setaddr = raddr; 3182ec25b48fSsusans setsize = 0; 3183ec25b48fSsusans set = 1; 3184ec25b48fSsusans } 3185ec25b48fSsusans } 3186ec25b48fSsusans if ((raddr + rsize) > (seg->s_base + seg->s_size)) { 3187ec25b48fSsusans ssize = seg->s_base + seg->s_size - raddr; 3188ec25b48fSsusans } else { 3189ec25b48fSsusans ssize = rsize; 3190ec25b48fSsusans } 3191ec25b48fSsusans } 3192ec25b48fSsusans error = 0; 3193ec25b48fSsusans if (set) { 3194ec25b48fSsusans ASSERT(setsize != 0); 3195ec25b48fSsusans error = as_iset2_default_lpsize(as, setaddr, setsize, 3196ec25b48fSsusans szc, szcvec); 3197ec25b48fSsusans } 3198ec25b48fSsusans return (error); 3199ec25b48fSsusans } 3200ec25b48fSsusans 3201ec25b48fSsusans /* 3202ec25b48fSsusans * as_iset_default_lpsize() breaks its chunk according to the size code bitmap 3203ec25b48fSsusans * returned by map_pgszcvec() (similar to as_map_segvn_segs()), and passes each 3204ec25b48fSsusans * chunk to as_iset1_default_lpsize(). 3205ec25b48fSsusans */ 3206ec25b48fSsusans static int 3207ec25b48fSsusans as_iset_default_lpsize(struct as *as, caddr_t addr, size_t size, int flags, 3208ec25b48fSsusans int type) 3209ec25b48fSsusans { 3210ec25b48fSsusans int rtype = (type & MAP_SHARED) ? MAPPGSZC_SHM : MAPPGSZC_PRIVM; 3211ec25b48fSsusans uint_t szcvec = map_pgszcvec(addr, size, (uintptr_t)addr, 3212843e1988Sjohnlev flags, rtype, 1); 3213ec25b48fSsusans uint_t szc; 3214ec25b48fSsusans uint_t nszc; 3215ec25b48fSsusans int error; 3216ec25b48fSsusans caddr_t a; 3217ec25b48fSsusans caddr_t eaddr; 3218ec25b48fSsusans size_t segsize; 3219ec25b48fSsusans size_t pgsz; 3220ec25b48fSsusans uint_t save_szcvec; 3221ec25b48fSsusans 3222ec25b48fSsusans ASSERT(AS_WRITE_HELD(as, &as->a_lock)); 3223ec25b48fSsusans ASSERT(IS_P2ALIGNED(addr, PAGESIZE)); 3224ec25b48fSsusans ASSERT(IS_P2ALIGNED(size, PAGESIZE)); 3225ec25b48fSsusans 3226ec25b48fSsusans szcvec &= ~1; 3227ec25b48fSsusans if (szcvec <= 1) { /* skip if base page size */ 3228ec25b48fSsusans return (0); 3229ec25b48fSsusans } 3230ec25b48fSsusans 3231ec25b48fSsusans /* Get the pagesize of the first larger page size. */ 3232ec25b48fSsusans szc = lowbit(szcvec) - 1; 3233ec25b48fSsusans pgsz = page_get_pagesize(szc); 3234ec25b48fSsusans eaddr = addr + size; 3235ec25b48fSsusans addr = (caddr_t)P2ROUNDUP((uintptr_t)addr, pgsz); 3236ec25b48fSsusans eaddr = (caddr_t)P2ALIGN((uintptr_t)eaddr, pgsz); 3237ec25b48fSsusans 3238ec25b48fSsusans save_szcvec = szcvec; 3239ec25b48fSsusans szcvec >>= (szc + 1); 3240ec25b48fSsusans nszc = szc; 3241ec25b48fSsusans while (szcvec) { 3242ec25b48fSsusans if ((szcvec & 0x1) == 0) { 3243ec25b48fSsusans nszc++; 3244ec25b48fSsusans szcvec >>= 1; 3245ec25b48fSsusans continue; 3246ec25b48fSsusans } 3247ec25b48fSsusans nszc++; 3248ec25b48fSsusans pgsz = page_get_pagesize(nszc); 3249ec25b48fSsusans a = (caddr_t)P2ROUNDUP((uintptr_t)addr, pgsz); 3250ec25b48fSsusans if (a != addr) { 3251ec25b48fSsusans ASSERT(szc > 0); 3252ec25b48fSsusans ASSERT(a < eaddr); 3253ec25b48fSsusans segsize = a - addr; 3254ec25b48fSsusans error = as_iset1_default_lpsize(as, addr, segsize, szc, 3255ec25b48fSsusans save_szcvec); 3256ec25b48fSsusans if (error) { 3257ec25b48fSsusans return (error); 3258ec25b48fSsusans } 3259ec25b48fSsusans addr = a; 3260ec25b48fSsusans } 3261ec25b48fSsusans szc = nszc; 3262ec25b48fSsusans szcvec >>= 1; 3263ec25b48fSsusans } 3264ec25b48fSsusans 3265ec25b48fSsusans ASSERT(addr < eaddr); 3266ec25b48fSsusans szcvec = save_szcvec; 3267ec25b48fSsusans while (szcvec) { 3268ec25b48fSsusans a = (caddr_t)P2ALIGN((uintptr_t)eaddr, pgsz); 3269ec25b48fSsusans ASSERT(a >= addr); 3270ec25b48fSsusans if (a != addr) { 3271ec25b48fSsusans ASSERT(szc > 0); 3272ec25b48fSsusans segsize = a - addr; 3273ec25b48fSsusans error = as_iset1_default_lpsize(as, addr, segsize, szc, 3274ec25b48fSsusans save_szcvec); 3275ec25b48fSsusans if (error) { 3276ec25b48fSsusans return (error); 3277ec25b48fSsusans } 3278ec25b48fSsusans addr = a; 3279ec25b48fSsusans } 3280ec25b48fSsusans szcvec &= ~(1 << szc); 3281ec25b48fSsusans if (szcvec) { 3282ec25b48fSsusans szc = highbit(szcvec) - 1; 3283ec25b48fSsusans pgsz = page_get_pagesize(szc); 3284ec25b48fSsusans } 3285ec25b48fSsusans } 3286ec25b48fSsusans ASSERT(addr == eaddr); 3287ec25b48fSsusans 3288ec25b48fSsusans return (0); 3289ec25b48fSsusans } 3290ec25b48fSsusans 3291ec25b48fSsusans /* 3292ec25b48fSsusans * Set the default large page size for the range. Called via memcntl with 3293ec25b48fSsusans * page size set to 0. as_set_default_lpsize breaks the range down into 3294ec25b48fSsusans * chunks with the same type/flags, ignores-non segvn segments, and passes 3295ec25b48fSsusans * each chunk to as_iset_default_lpsize(). 3296ec25b48fSsusans */ 3297ec25b48fSsusans int 3298ec25b48fSsusans as_set_default_lpsize(struct as *as, caddr_t addr, size_t size) 3299ec25b48fSsusans { 3300ec25b48fSsusans struct seg *seg; 3301ec25b48fSsusans caddr_t raddr; 3302ec25b48fSsusans size_t rsize; 3303ec25b48fSsusans size_t ssize; 3304ec25b48fSsusans int rtype, rflags; 3305ec25b48fSsusans int stype, sflags; 3306ec25b48fSsusans int error; 3307ec25b48fSsusans caddr_t setaddr; 3308ec25b48fSsusans size_t setsize; 3309ec25b48fSsusans int segvn; 3310ec25b48fSsusans 3311ec25b48fSsusans if (size == 0) 3312ec25b48fSsusans return (0); 3313ec25b48fSsusans 3314ec25b48fSsusans AS_LOCK_ENTER(as, &as->a_lock, RW_WRITER); 3315ec25b48fSsusans again: 3316ec25b48fSsusans error = 0; 3317ec25b48fSsusans 3318ec25b48fSsusans raddr = (caddr_t)((uintptr_t)addr & (uintptr_t)PAGEMASK); 3319ec25b48fSsusans rsize = (((size_t)(addr + size) + PAGEOFFSET) & PAGEMASK) - 3320ec25b48fSsusans (size_t)raddr; 3321ec25b48fSsusans 3322ec25b48fSsusans if (raddr + rsize < raddr) { /* check for wraparound */ 3323ec25b48fSsusans AS_LOCK_EXIT(as, &as->a_lock); 3324ec25b48fSsusans return (ENOMEM); 3325ec25b48fSsusans } 3326ec25b48fSsusans as_clearwatchprot(as, raddr, rsize); 3327ec25b48fSsusans seg = as_segat(as, raddr); 3328ec25b48fSsusans if (seg == NULL) { 3329ec25b48fSsusans as_setwatch(as); 3330ec25b48fSsusans AS_LOCK_EXIT(as, &as->a_lock); 3331ec25b48fSsusans return (ENOMEM); 3332ec25b48fSsusans } 3333ec25b48fSsusans if (seg->s_ops == &segvn_ops) { 3334ec25b48fSsusans rtype = SEGOP_GETTYPE(seg, addr); 3335ec25b48fSsusans rflags = rtype & (MAP_TEXT | MAP_INITDATA); 3336ec25b48fSsusans rtype = rtype & (MAP_SHARED | MAP_PRIVATE); 3337ec25b48fSsusans segvn = 1; 3338ec25b48fSsusans } else { 3339ec25b48fSsusans segvn = 0; 3340ec25b48fSsusans } 3341ec25b48fSsusans setaddr = raddr; 3342ec25b48fSsusans setsize = 0; 3343ec25b48fSsusans 3344ec25b48fSsusans for (; rsize != 0; rsize -= ssize, raddr += ssize, setsize += ssize) { 3345ec25b48fSsusans if (raddr >= (seg->s_base + seg->s_size)) { 3346ec25b48fSsusans seg = AS_SEGNEXT(as, seg); 3347ec25b48fSsusans if (seg == NULL || raddr != seg->s_base) { 3348ec25b48fSsusans error = ENOMEM; 3349ec25b48fSsusans break; 3350ec25b48fSsusans } 3351ec25b48fSsusans if (seg->s_ops == &segvn_ops) { 3352ec25b48fSsusans stype = SEGOP_GETTYPE(seg, raddr); 3353ec25b48fSsusans sflags = stype & (MAP_TEXT | MAP_INITDATA); 3354ec25b48fSsusans stype &= (MAP_SHARED | MAP_PRIVATE); 3355ec25b48fSsusans if (segvn && (rflags != sflags || 3356ec25b48fSsusans rtype != stype)) { 3357ec25b48fSsusans /* 3358ec25b48fSsusans * The next segment is also segvn but 3359ec25b48fSsusans * has different flags and/or type. 3360ec25b48fSsusans */ 3361ec25b48fSsusans ASSERT(setsize != 0); 3362ec25b48fSsusans error = as_iset_default_lpsize(as, 3363ec25b48fSsusans setaddr, setsize, rflags, rtype); 3364ec25b48fSsusans if (error) { 3365ec25b48fSsusans break; 3366ec25b48fSsusans } 3367ec25b48fSsusans rflags = sflags; 3368ec25b48fSsusans rtype = stype; 3369ec25b48fSsusans setaddr = raddr; 3370ec25b48fSsusans setsize = 0; 3371ec25b48fSsusans } else if (!segvn) { 3372ec25b48fSsusans rflags = sflags; 3373ec25b48fSsusans rtype = stype; 3374ec25b48fSsusans setaddr = raddr; 3375ec25b48fSsusans setsize = 0; 3376ec25b48fSsusans segvn = 1; 3377ec25b48fSsusans } 3378ec25b48fSsusans } else if (segvn) { 3379ec25b48fSsusans /* The next segment is not segvn. */ 3380ec25b48fSsusans ASSERT(setsize != 0); 3381ec25b48fSsusans error = as_iset_default_lpsize(as, 3382ec25b48fSsusans setaddr, setsize, rflags, rtype); 3383ec25b48fSsusans if (error) { 3384ec25b48fSsusans break; 3385ec25b48fSsusans } 3386ec25b48fSsusans segvn = 0; 3387ec25b48fSsusans } 3388ec25b48fSsusans } 3389ec25b48fSsusans if ((raddr + rsize) > (seg->s_base + seg->s_size)) { 3390ec25b48fSsusans ssize = seg->s_base + seg->s_size - raddr; 3391ec25b48fSsusans } else { 3392ec25b48fSsusans ssize = rsize; 3393ec25b48fSsusans } 3394ec25b48fSsusans } 3395ec25b48fSsusans if (error == 0 && segvn) { 3396ec25b48fSsusans /* The last chunk when rsize == 0. */ 3397ec25b48fSsusans ASSERT(setsize != 0); 3398ec25b48fSsusans error = as_iset_default_lpsize(as, setaddr, setsize, 3399ec25b48fSsusans rflags, rtype); 3400ec25b48fSsusans } 3401ec25b48fSsusans 3402ec25b48fSsusans if (error == IE_RETRY) { 3403ec25b48fSsusans goto again; 3404ec25b48fSsusans } else if (error == IE_NOMEM) { 3405ec25b48fSsusans error = EAGAIN; 3406ec25b48fSsusans } else if (error == ENOTSUP) { 3407ec25b48fSsusans error = EINVAL; 3408ec25b48fSsusans } else if (error == EAGAIN) { 3409ec25b48fSsusans mutex_enter(&as->a_contents); 3410a98e9dbfSaguzovsk if (!AS_ISNOUNMAPWAIT(as)) { 3411a98e9dbfSaguzovsk if (AS_ISUNMAPWAIT(as) == 0) { 3412a98e9dbfSaguzovsk cv_broadcast(&as->a_cv); 3413a98e9dbfSaguzovsk } 3414a98e9dbfSaguzovsk AS_SETUNMAPWAIT(as); 3415a98e9dbfSaguzovsk AS_LOCK_EXIT(as, &as->a_lock); 3416a98e9dbfSaguzovsk while (AS_ISUNMAPWAIT(as)) { 3417a98e9dbfSaguzovsk cv_wait(&as->a_cv, &as->a_contents); 3418a98e9dbfSaguzovsk } 3419a98e9dbfSaguzovsk mutex_exit(&as->a_contents); 3420a98e9dbfSaguzovsk AS_LOCK_ENTER(as, &as->a_lock, RW_WRITER); 3421a98e9dbfSaguzovsk } else { 3422a98e9dbfSaguzovsk /* 3423a98e9dbfSaguzovsk * We may have raced with 3424a98e9dbfSaguzovsk * segvn_reclaim()/segspt_reclaim(). In this case 3425a98e9dbfSaguzovsk * clean nounmapwait flag and retry since softlockcnt 3426a98e9dbfSaguzovsk * in this segment may be already 0. We don't drop as 3427a98e9dbfSaguzovsk * writer lock so our number of retries without 3428a98e9dbfSaguzovsk * sleeping should be very small. See segvn_reclaim() 3429a98e9dbfSaguzovsk * for more comments. 3430a98e9dbfSaguzovsk */ 3431a98e9dbfSaguzovsk AS_CLRNOUNMAPWAIT(as); 3432a98e9dbfSaguzovsk mutex_exit(&as->a_contents); 3433ec25b48fSsusans } 3434ec25b48fSsusans goto again; 3435ec25b48fSsusans } 3436ec25b48fSsusans 3437ec25b48fSsusans as_setwatch(as); 3438ec25b48fSsusans AS_LOCK_EXIT(as, &as->a_lock); 3439ec25b48fSsusans return (error); 3440ec25b48fSsusans } 3441ec25b48fSsusans 34427c478bd9Sstevel@tonic-gate /* 34437c478bd9Sstevel@tonic-gate * Setup all of the uninitialized watched pages that we can. 34447c478bd9Sstevel@tonic-gate */ 34457c478bd9Sstevel@tonic-gate void 34467c478bd9Sstevel@tonic-gate as_setwatch(struct as *as) 34477c478bd9Sstevel@tonic-gate { 34487c478bd9Sstevel@tonic-gate struct watched_page *pwp; 34497c478bd9Sstevel@tonic-gate struct seg *seg; 34507c478bd9Sstevel@tonic-gate caddr_t vaddr; 34517c478bd9Sstevel@tonic-gate uint_t prot; 34527c478bd9Sstevel@tonic-gate int err, retrycnt; 34537c478bd9Sstevel@tonic-gate 34547c478bd9Sstevel@tonic-gate if (avl_numnodes(&as->a_wpage) == 0) 34557c478bd9Sstevel@tonic-gate return; 34567c478bd9Sstevel@tonic-gate 34577c478bd9Sstevel@tonic-gate ASSERT(AS_WRITE_HELD(as, &as->a_lock)); 34587c478bd9Sstevel@tonic-gate 34597c478bd9Sstevel@tonic-gate for (pwp = avl_first(&as->a_wpage); pwp != NULL; 34607c478bd9Sstevel@tonic-gate pwp = AVL_NEXT(&as->a_wpage, pwp)) { 34617c478bd9Sstevel@tonic-gate retrycnt = 0; 34627c478bd9Sstevel@tonic-gate retry: 34637c478bd9Sstevel@tonic-gate vaddr = pwp->wp_vaddr; 34647c478bd9Sstevel@tonic-gate if (pwp->wp_oprot != 0 || /* already set up */ 34657c478bd9Sstevel@tonic-gate (seg = as_segat(as, vaddr)) == NULL || 34667c478bd9Sstevel@tonic-gate SEGOP_GETPROT(seg, vaddr, 0, &prot) != 0) 34677c478bd9Sstevel@tonic-gate continue; 34687c478bd9Sstevel@tonic-gate 34697c478bd9Sstevel@tonic-gate pwp->wp_oprot = prot; 34707c478bd9Sstevel@tonic-gate if (pwp->wp_read) 34717c478bd9Sstevel@tonic-gate prot &= ~(PROT_READ|PROT_WRITE|PROT_EXEC); 34727c478bd9Sstevel@tonic-gate if (pwp->wp_write) 34737c478bd9Sstevel@tonic-gate prot &= ~PROT_WRITE; 34747c478bd9Sstevel@tonic-gate if (pwp->wp_exec) 34757c478bd9Sstevel@tonic-gate prot &= ~(PROT_READ|PROT_WRITE|PROT_EXEC); 34767c478bd9Sstevel@tonic-gate if (!(pwp->wp_flags & WP_NOWATCH) && prot != pwp->wp_oprot) { 34777c478bd9Sstevel@tonic-gate err = SEGOP_SETPROT(seg, vaddr, PAGESIZE, prot); 34787c478bd9Sstevel@tonic-gate if (err == IE_RETRY) { 34797c478bd9Sstevel@tonic-gate pwp->wp_oprot = 0; 34807c478bd9Sstevel@tonic-gate ASSERT(retrycnt == 0); 34817c478bd9Sstevel@tonic-gate retrycnt++; 34827c478bd9Sstevel@tonic-gate goto retry; 34837c478bd9Sstevel@tonic-gate } 34847c478bd9Sstevel@tonic-gate } 34857c478bd9Sstevel@tonic-gate pwp->wp_prot = prot; 34867c478bd9Sstevel@tonic-gate } 34877c478bd9Sstevel@tonic-gate } 34887c478bd9Sstevel@tonic-gate 34897c478bd9Sstevel@tonic-gate /* 34907c478bd9Sstevel@tonic-gate * Clear all of the watched pages in the address space. 34917c478bd9Sstevel@tonic-gate */ 34927c478bd9Sstevel@tonic-gate void 34937c478bd9Sstevel@tonic-gate as_clearwatch(struct as *as) 34947c478bd9Sstevel@tonic-gate { 34957c478bd9Sstevel@tonic-gate struct watched_page *pwp; 34967c478bd9Sstevel@tonic-gate struct seg *seg; 34977c478bd9Sstevel@tonic-gate caddr_t vaddr; 34987c478bd9Sstevel@tonic-gate uint_t prot; 34997c478bd9Sstevel@tonic-gate int err, retrycnt; 35007c478bd9Sstevel@tonic-gate 35017c478bd9Sstevel@tonic-gate if (avl_numnodes(&as->a_wpage) == 0) 35027c478bd9Sstevel@tonic-gate return; 35037c478bd9Sstevel@tonic-gate 35047c478bd9Sstevel@tonic-gate ASSERT(AS_WRITE_HELD(as, &as->a_lock)); 35057c478bd9Sstevel@tonic-gate 35067c478bd9Sstevel@tonic-gate for (pwp = avl_first(&as->a_wpage); pwp != NULL; 35077c478bd9Sstevel@tonic-gate pwp = AVL_NEXT(&as->a_wpage, pwp)) { 35087c478bd9Sstevel@tonic-gate retrycnt = 0; 35097c478bd9Sstevel@tonic-gate retry: 35107c478bd9Sstevel@tonic-gate vaddr = pwp->wp_vaddr; 35117c478bd9Sstevel@tonic-gate if (pwp->wp_oprot == 0 || /* not set up */ 35127c478bd9Sstevel@tonic-gate (seg = as_segat(as, vaddr)) == NULL) 35137c478bd9Sstevel@tonic-gate continue; 35147c478bd9Sstevel@tonic-gate 35157c478bd9Sstevel@tonic-gate if ((prot = pwp->wp_oprot) != pwp->wp_prot) { 35167c478bd9Sstevel@tonic-gate err = SEGOP_SETPROT(seg, vaddr, PAGESIZE, prot); 35177c478bd9Sstevel@tonic-gate if (err == IE_RETRY) { 35187c478bd9Sstevel@tonic-gate ASSERT(retrycnt == 0); 35197c478bd9Sstevel@tonic-gate retrycnt++; 35207c478bd9Sstevel@tonic-gate goto retry; 35217c478bd9Sstevel@tonic-gate } 35227c478bd9Sstevel@tonic-gate } 35237c478bd9Sstevel@tonic-gate pwp->wp_oprot = 0; 35247c478bd9Sstevel@tonic-gate pwp->wp_prot = 0; 35257c478bd9Sstevel@tonic-gate } 35267c478bd9Sstevel@tonic-gate } 35277c478bd9Sstevel@tonic-gate 35287c478bd9Sstevel@tonic-gate /* 35297c478bd9Sstevel@tonic-gate * Force a new setup for all the watched pages in the range. 35307c478bd9Sstevel@tonic-gate */ 35317c478bd9Sstevel@tonic-gate static void 35327c478bd9Sstevel@tonic-gate as_setwatchprot(struct as *as, caddr_t addr, size_t size, uint_t prot) 35337c478bd9Sstevel@tonic-gate { 35347c478bd9Sstevel@tonic-gate struct watched_page *pwp; 35357c478bd9Sstevel@tonic-gate struct watched_page tpw; 35367c478bd9Sstevel@tonic-gate caddr_t eaddr = addr + size; 35377c478bd9Sstevel@tonic-gate caddr_t vaddr; 35387c478bd9Sstevel@tonic-gate struct seg *seg; 35397c478bd9Sstevel@tonic-gate int err, retrycnt; 35407c478bd9Sstevel@tonic-gate uint_t wprot; 35417c478bd9Sstevel@tonic-gate avl_index_t where; 35427c478bd9Sstevel@tonic-gate 35437c478bd9Sstevel@tonic-gate if (avl_numnodes(&as->a_wpage) == 0) 35447c478bd9Sstevel@tonic-gate return; 35457c478bd9Sstevel@tonic-gate 35467c478bd9Sstevel@tonic-gate ASSERT(AS_WRITE_HELD(as, &as->a_lock)); 35477c478bd9Sstevel@tonic-gate 35487c478bd9Sstevel@tonic-gate tpw.wp_vaddr = (caddr_t)((uintptr_t)addr & (uintptr_t)PAGEMASK); 35497c478bd9Sstevel@tonic-gate if ((pwp = avl_find(&as->a_wpage, &tpw, &where)) == NULL) 35507c478bd9Sstevel@tonic-gate pwp = avl_nearest(&as->a_wpage, where, AVL_AFTER); 35517c478bd9Sstevel@tonic-gate 35527c478bd9Sstevel@tonic-gate while (pwp != NULL && pwp->wp_vaddr < eaddr) { 35537c478bd9Sstevel@tonic-gate retrycnt = 0; 35547c478bd9Sstevel@tonic-gate vaddr = pwp->wp_vaddr; 35557c478bd9Sstevel@tonic-gate 35567c478bd9Sstevel@tonic-gate wprot = prot; 35577c478bd9Sstevel@tonic-gate if (pwp->wp_read) 35587c478bd9Sstevel@tonic-gate wprot &= ~(PROT_READ|PROT_WRITE|PROT_EXEC); 35597c478bd9Sstevel@tonic-gate if (pwp->wp_write) 35607c478bd9Sstevel@tonic-gate wprot &= ~PROT_WRITE; 35617c478bd9Sstevel@tonic-gate if (pwp->wp_exec) 35627c478bd9Sstevel@tonic-gate wprot &= ~(PROT_READ|PROT_WRITE|PROT_EXEC); 35637c478bd9Sstevel@tonic-gate if (!(pwp->wp_flags & WP_NOWATCH) && wprot != pwp->wp_oprot) { 35647c478bd9Sstevel@tonic-gate retry: 35657c478bd9Sstevel@tonic-gate seg = as_segat(as, vaddr); 35667c478bd9Sstevel@tonic-gate if (seg == NULL) { 35677c478bd9Sstevel@tonic-gate panic("as_setwatchprot: no seg"); 35687c478bd9Sstevel@tonic-gate /*NOTREACHED*/ 35697c478bd9Sstevel@tonic-gate } 35707c478bd9Sstevel@tonic-gate err = SEGOP_SETPROT(seg, vaddr, PAGESIZE, wprot); 35717c478bd9Sstevel@tonic-gate if (err == IE_RETRY) { 35727c478bd9Sstevel@tonic-gate ASSERT(retrycnt == 0); 35737c478bd9Sstevel@tonic-gate retrycnt++; 35747c478bd9Sstevel@tonic-gate goto retry; 35757c478bd9Sstevel@tonic-gate } 35767c478bd9Sstevel@tonic-gate } 35777c478bd9Sstevel@tonic-gate pwp->wp_oprot = prot; 35787c478bd9Sstevel@tonic-gate pwp->wp_prot = wprot; 35797c478bd9Sstevel@tonic-gate 35807c478bd9Sstevel@tonic-gate pwp = AVL_NEXT(&as->a_wpage, pwp); 35817c478bd9Sstevel@tonic-gate } 35827c478bd9Sstevel@tonic-gate } 35837c478bd9Sstevel@tonic-gate 35847c478bd9Sstevel@tonic-gate /* 35857c478bd9Sstevel@tonic-gate * Clear all of the watched pages in the range. 35867c478bd9Sstevel@tonic-gate */ 35877c478bd9Sstevel@tonic-gate static void 35887c478bd9Sstevel@tonic-gate as_clearwatchprot(struct as *as, caddr_t addr, size_t size) 35897c478bd9Sstevel@tonic-gate { 35907c478bd9Sstevel@tonic-gate caddr_t eaddr = addr + size; 35917c478bd9Sstevel@tonic-gate struct watched_page *pwp; 35927c478bd9Sstevel@tonic-gate struct watched_page tpw; 35937c478bd9Sstevel@tonic-gate uint_t prot; 35947c478bd9Sstevel@tonic-gate struct seg *seg; 35957c478bd9Sstevel@tonic-gate int err, retrycnt; 35967c478bd9Sstevel@tonic-gate avl_index_t where; 35977c478bd9Sstevel@tonic-gate 35987c478bd9Sstevel@tonic-gate if (avl_numnodes(&as->a_wpage) == 0) 35997c478bd9Sstevel@tonic-gate return; 36007c478bd9Sstevel@tonic-gate 36017c478bd9Sstevel@tonic-gate tpw.wp_vaddr = (caddr_t)((uintptr_t)addr & (uintptr_t)PAGEMASK); 36027c478bd9Sstevel@tonic-gate if ((pwp = avl_find(&as->a_wpage, &tpw, &where)) == NULL) 36037c478bd9Sstevel@tonic-gate pwp = avl_nearest(&as->a_wpage, where, AVL_AFTER); 36047c478bd9Sstevel@tonic-gate 36057c478bd9Sstevel@tonic-gate ASSERT(AS_WRITE_HELD(as, &as->a_lock)); 36067c478bd9Sstevel@tonic-gate 36077c478bd9Sstevel@tonic-gate while (pwp != NULL && pwp->wp_vaddr < eaddr) { 36087c478bd9Sstevel@tonic-gate 36097c478bd9Sstevel@tonic-gate if ((prot = pwp->wp_oprot) != 0) { 36107c478bd9Sstevel@tonic-gate retrycnt = 0; 36117c478bd9Sstevel@tonic-gate 36127c478bd9Sstevel@tonic-gate if (prot != pwp->wp_prot) { 36137c478bd9Sstevel@tonic-gate retry: 36147c478bd9Sstevel@tonic-gate seg = as_segat(as, pwp->wp_vaddr); 36157c478bd9Sstevel@tonic-gate if (seg == NULL) 36167c478bd9Sstevel@tonic-gate continue; 36177c478bd9Sstevel@tonic-gate err = SEGOP_SETPROT(seg, pwp->wp_vaddr, 36187c478bd9Sstevel@tonic-gate PAGESIZE, prot); 36197c478bd9Sstevel@tonic-gate if (err == IE_RETRY) { 36207c478bd9Sstevel@tonic-gate ASSERT(retrycnt == 0); 36217c478bd9Sstevel@tonic-gate retrycnt++; 36227c478bd9Sstevel@tonic-gate goto retry; 36237c478bd9Sstevel@tonic-gate 36247c478bd9Sstevel@tonic-gate } 36257c478bd9Sstevel@tonic-gate } 36267c478bd9Sstevel@tonic-gate pwp->wp_oprot = 0; 36277c478bd9Sstevel@tonic-gate pwp->wp_prot = 0; 36287c478bd9Sstevel@tonic-gate } 36297c478bd9Sstevel@tonic-gate 36307c478bd9Sstevel@tonic-gate pwp = AVL_NEXT(&as->a_wpage, pwp); 36317c478bd9Sstevel@tonic-gate } 36327c478bd9Sstevel@tonic-gate } 36337c478bd9Sstevel@tonic-gate 36347c478bd9Sstevel@tonic-gate void 36357c478bd9Sstevel@tonic-gate as_signal_proc(struct as *as, k_siginfo_t *siginfo) 36367c478bd9Sstevel@tonic-gate { 36377c478bd9Sstevel@tonic-gate struct proc *p; 36387c478bd9Sstevel@tonic-gate 36397c478bd9Sstevel@tonic-gate mutex_enter(&pidlock); 36407c478bd9Sstevel@tonic-gate for (p = practive; p; p = p->p_next) { 36417c478bd9Sstevel@tonic-gate if (p->p_as == as) { 36427c478bd9Sstevel@tonic-gate mutex_enter(&p->p_lock); 36437c478bd9Sstevel@tonic-gate if (p->p_as == as) 36447c478bd9Sstevel@tonic-gate sigaddq(p, NULL, siginfo, KM_NOSLEEP); 36457c478bd9Sstevel@tonic-gate mutex_exit(&p->p_lock); 36467c478bd9Sstevel@tonic-gate } 36477c478bd9Sstevel@tonic-gate } 36487c478bd9Sstevel@tonic-gate mutex_exit(&pidlock); 36497c478bd9Sstevel@tonic-gate } 36507c478bd9Sstevel@tonic-gate 36517c478bd9Sstevel@tonic-gate /* 36527c478bd9Sstevel@tonic-gate * return memory object ID 36537c478bd9Sstevel@tonic-gate */ 36547c478bd9Sstevel@tonic-gate int 36557c478bd9Sstevel@tonic-gate as_getmemid(struct as *as, caddr_t addr, memid_t *memidp) 36567c478bd9Sstevel@tonic-gate { 36577c478bd9Sstevel@tonic-gate struct seg *seg; 36587c478bd9Sstevel@tonic-gate int sts; 36597c478bd9Sstevel@tonic-gate 36607c478bd9Sstevel@tonic-gate AS_LOCK_ENTER(as, &as->a_lock, RW_READER); 36617c478bd9Sstevel@tonic-gate seg = as_segat(as, addr); 36627c478bd9Sstevel@tonic-gate if (seg == NULL) { 36637c478bd9Sstevel@tonic-gate AS_LOCK_EXIT(as, &as->a_lock); 36647c478bd9Sstevel@tonic-gate return (EFAULT); 36657c478bd9Sstevel@tonic-gate } 36667c478bd9Sstevel@tonic-gate /* 36677c478bd9Sstevel@tonic-gate * catch old drivers which may not support getmemid 36687c478bd9Sstevel@tonic-gate */ 36697c478bd9Sstevel@tonic-gate if (seg->s_ops->getmemid == NULL) { 36707c478bd9Sstevel@tonic-gate AS_LOCK_EXIT(as, &as->a_lock); 36717c478bd9Sstevel@tonic-gate return (ENODEV); 36727c478bd9Sstevel@tonic-gate } 36737c478bd9Sstevel@tonic-gate 36747c478bd9Sstevel@tonic-gate sts = SEGOP_GETMEMID(seg, addr, memidp); 36757c478bd9Sstevel@tonic-gate 36767c478bd9Sstevel@tonic-gate AS_LOCK_EXIT(as, &as->a_lock); 36777c478bd9Sstevel@tonic-gate return (sts); 36787c478bd9Sstevel@tonic-gate } 3679