vm_page.c revision 170292
143313Sbrian/*- 243313Sbrian * Copyright (c) 1991 Regents of the University of California. 343313Sbrian * All rights reserved. 443313Sbrian * 543313Sbrian * This code is derived from software contributed to Berkeley by 643313Sbrian * The Mach Operating System project at Carnegie-Mellon University. 743313Sbrian * 843313Sbrian * Redistribution and use in source and binary forms, with or without 943313Sbrian * modification, are permitted provided that the following conditions 1043313Sbrian * are met: 1143313Sbrian * 1. Redistributions of source code must retain the above copyright 1243313Sbrian * notice, this list of conditions and the following disclaimer. 1343313Sbrian * 2. Redistributions in binary form must reproduce the above copyright 1443313Sbrian * notice, this list of conditions and the following disclaimer in the 1543313Sbrian * documentation and/or other materials provided with the distribution. 1643313Sbrian * 4. Neither the name of the University nor the names of its contributors 1743313Sbrian * may be used to endorse or promote products derived from this software 1843313Sbrian * without specific prior written permission. 1943313Sbrian * 2043313Sbrian * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 2143313Sbrian * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 2243313Sbrian * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 2343313Sbrian * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 2443313Sbrian * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 2543313Sbrian * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 2650479Speter * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 2743313Sbrian * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 2843313Sbrian * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 2943313Sbrian * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 3043313Sbrian * SUCH DAMAGE. 3198132Sbrian * 3258032Sbrian * from: @(#)vm_page.c 7.4 (Berkeley) 5/7/91 3343313Sbrian */ 3443313Sbrian 3543313Sbrian/*- 3643313Sbrian * Copyright (c) 1987, 1990 Carnegie-Mellon University. 3743313Sbrian * All rights reserved. 3858032Sbrian * 3943313Sbrian * Authors: Avadis Tevanian, Jr., Michael Wayne Young 4058037Sbrian * 4158037Sbrian * Permission to use, copy, modify and distribute this software and 4296324Sbrian * its documentation is hereby granted, provided that both the copyright 4358037Sbrian * notice and this permission notice appear in all copies of the 4458037Sbrian * software, derivative works or modified versions, and any portions 4596324Sbrian * thereof, and that both notices appear in supporting documentation. 4658037Sbrian * 4758037Sbrian * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS" 4843313Sbrian * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND 4998132Sbrian * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE. 5098132Sbrian * 5198132Sbrian * Carnegie Mellon requests users of this software to return to 52102500Sbrian * 5343313Sbrian * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU 5443313Sbrian * School of Computer Science 5543313Sbrian * Carnegie Mellon University 5643693Sbrian * Pittsburgh PA 15213-3890 5743313Sbrian * 5850840Sbrian * any improvements or extensions that they make and grant Carnegie the 5950840Sbrian * rights to redistribute these changes. 6043313Sbrian */ 6146686Sbrian 6243313Sbrian/* 6343313Sbrian * GENERAL RULES ON VM_PAGE MANIPULATION 6443313Sbrian * 6543313Sbrian * - a pageq mutex is required when adding or removing a page from a 6643313Sbrian * page queue (vm_page_queue[]), regardless of other mutexes or the 6743313Sbrian * busy state of a page. 6843313Sbrian * 6943313Sbrian * - a hash chain mutex is required when associating or disassociating 7043313Sbrian * a page from the VM PAGE CACHE hash table (vm_page_buckets), 7143313Sbrian * regardless of other mutexes or the busy state of a page. 7243313Sbrian * 7343313Sbrian * - either a hash chain mutex OR a busied page is required in order 7481634Sbrian * to modify the page flags. A hash chain mutex must be obtained in 7581634Sbrian * order to busy a page. A page's flags cannot be modified by a 7643313Sbrian * hash chain mutex if the page is marked busy. 7781634Sbrian * 7843313Sbrian * - The object memq mutex is held when inserting or removing 7943313Sbrian * pages from an object (vm_page_insert() or vm_page_remove()). This 8043313Sbrian * is different from the object's main mutex. 8143313Sbrian * 8243313Sbrian * Generally speaking, you have to be aware of side effects when running 8343313Sbrian * vm_page ops. A vm_page_lookup() will return with the hash chain 8443313Sbrian * locked, whether it was able to lookup the page or not. vm_page_free(), 8543313Sbrian * vm_page_cache(), vm_page_activate(), and a number of other routines 8643693Sbrian * will release the hash chain mutex for you. Intermediate manipulation 8743693Sbrian * routines such as vm_page_flag_set() expect the hash chain to be held 8843693Sbrian * on entry and the hash chain will remain held on return. 8943693Sbrian * 9043693Sbrian * pageq scanning can only occur with the pageq in question locked. 9143693Sbrian * We have a known bottleneck with the active queue, but the cache 9243693Sbrian * and free queues are actually arrays already. 9381634Sbrian */ 9443313Sbrian 9596324Sbrian/* 9643313Sbrian * Resident memory management module. 9796324Sbrian */ 9896730Sbrian 9996324Sbrian#include <sys/cdefs.h> 10096324Sbrian__FBSDID("$FreeBSD: head/sys/vm/vm_page.c 170292 2007-06-04 21:45:18Z attilio $"); 10196324Sbrian 10296324Sbrian#include <sys/param.h> 10396324Sbrian#include <sys/systm.h> 10496730Sbrian#include <sys/lock.h> 10596730Sbrian#include <sys/kernel.h> 10696730Sbrian#include <sys/malloc.h> 10796730Sbrian#include <sys/mutex.h> 10896730Sbrian#include <sys/proc.h> 10996730Sbrian#include <sys/sysctl.h> 11096730Sbrian#include <sys/vmmeter.h> 11196730Sbrian#include <sys/vnode.h> 11298132Sbrian 11398132Sbrian#include <vm/vm.h> 11498132Sbrian#include <vm/vm_param.h> 11596324Sbrian#include <vm/vm_kern.h> 11696324Sbrian#include <vm/vm_object.h> 11798132Sbrian#include <vm/vm_page.h> 11898132Sbrian#include <vm/vm_pageout.h> 11998132Sbrian#include <vm/vm_pager.h> 12098132Sbrian#include <vm/vm_extern.h> 12198132Sbrian#include <vm/uma.h> 12298132Sbrian#include <vm/uma_int.h> 12398132Sbrian 12498132Sbrian#include <machine/md_var.h> 12598132Sbrian 12698132Sbrian/* 12798132Sbrian * Associated with page of user-allocatable memory is a 12898132Sbrian * page structure. 12998132Sbrian */ 13098132Sbrian 13198132Sbrianstruct mtx vm_page_queue_mtx; 13298132Sbrianstruct mtx vm_page_queue_free_mtx; 13398132Sbrian 13498132Sbrianvm_page_t vm_page_array = 0; 13598132Sbrianint vm_page_array_size = 0; 13698132Sbrianlong first_page = 0; 13798132Sbrianint vm_page_zero_count = 0; 13898132Sbrian 13998132Sbrianstatic int boot_pages = UMA_BOOT_PAGES; 14098132SbrianTUNABLE_INT("vm.boot_pages", &boot_pages); 14198132SbrianSYSCTL_INT(_vm, OID_AUTO, boot_pages, CTLFLAG_RD, &boot_pages, 0, 14298132Sbrian "number of pages allocated for bootstrapping the VM system"); 14398132Sbrian 14498132Sbrian/* 14598132Sbrian * vm_set_page_size: 14698132Sbrian * 14798132Sbrian * Sets the page size, perhaps based upon the memory 14898132Sbrian * size. Must be called before any use of page-size 14998132Sbrian * dependent functions. 15098132Sbrian */ 15198132Sbrianvoid 15298132Sbrianvm_set_page_size(void) 15398132Sbrian{ 15498132Sbrian if (cnt.v_page_size == 0) 15598132Sbrian cnt.v_page_size = PAGE_SIZE; 15698132Sbrian if (((cnt.v_page_size - 1) & cnt.v_page_size) != 0) 15798132Sbrian panic("vm_set_page_size: page size not a power of two"); 15898132Sbrian} 15998132Sbrian 16098132Sbrian/* 16198132Sbrian * vm_page_blacklist_lookup: 16298132Sbrian * 16398132Sbrian * See if a physical address in this page has been listed 16498132Sbrian * in the blacklist tunable. Entries in the tunable are 16598132Sbrian * separated by spaces or commas. If an invalid integer is 16698132Sbrian * encountered then the rest of the string is skipped. 16798132Sbrian */ 16898132Sbrianstatic int 16998132Sbrianvm_page_blacklist_lookup(char *list, vm_paddr_t pa) 17098132Sbrian{ 17198132Sbrian vm_paddr_t bad; 17298132Sbrian char *cp, *pos; 17398132Sbrian 17498132Sbrian for (pos = list; *pos != '\0'; pos = cp) { 17598132Sbrian bad = strtoq(pos, &cp, 0); 17698132Sbrian if (*cp != '\0') { 17798132Sbrian if (*cp == ' ' || *cp == ',') { 17898132Sbrian cp++; 17998132Sbrian if (cp == pos) 18098132Sbrian continue; 18198132Sbrian } else 18298132Sbrian break; 18398132Sbrian } 18498132Sbrian if (pa == trunc_page(bad)) 18598132Sbrian return (1); 18698132Sbrian } 18798132Sbrian return (0); 18898132Sbrian} 18998132Sbrian 19098132Sbrian/* 19198132Sbrian * vm_page_startup: 19298132Sbrian * 19398132Sbrian * Initializes the resident memory module. 19498132Sbrian * 19598132Sbrian * Allocates memory for the page cells, and 19698132Sbrian * for the object/offset-to-page hash table headers. 19798132Sbrian * Each page cell is initialized and placed on the free list. 19898132Sbrian */ 19998132Sbrianvm_offset_t 20098132Sbrianvm_page_startup(vm_offset_t vaddr) 20198132Sbrian{ 20298132Sbrian vm_offset_t mapped; 20398132Sbrian vm_size_t npages; 20498132Sbrian vm_paddr_t page_range; 20598132Sbrian vm_paddr_t new_end; 20698132Sbrian int i; 20798132Sbrian vm_paddr_t pa; 20898132Sbrian int nblocks; 20998132Sbrian vm_paddr_t last_pa; 21098132Sbrian char *list; 21198132Sbrian 21298132Sbrian /* the biggest memory array is the second group of pages */ 21398132Sbrian vm_paddr_t end; 21498132Sbrian vm_paddr_t biggestsize; 21598132Sbrian vm_paddr_t low_water, high_water; 21698132Sbrian int biggestone; 21798132Sbrian 21898132Sbrian vm_paddr_t total; 219116622Sume 220116622Sume total = 0; 221116622Sume biggestsize = 0; 222116622Sume biggestone = 0; 223116622Sume nblocks = 0; 224116622Sume vaddr = round_page(vaddr); 225116622Sume 226116622Sume for (i = 0; phys_avail[i + 1]; i += 2) { 227116622Sume phys_avail[i] = round_page(phys_avail[i]); 228116622Sume phys_avail[i + 1] = trunc_page(phys_avail[i + 1]); 229116622Sume } 230116622Sume 231116622Sume low_water = phys_avail[0]; 232116622Sume high_water = phys_avail[1]; 233116622Sume 234116622Sume for (i = 0; phys_avail[i + 1]; i += 2) { 235116622Sume vm_paddr_t size = phys_avail[i + 1] - phys_avail[i]; 236116622Sume 237116622Sume if (size > biggestsize) { 23843693Sbrian biggestone = i; 23943693Sbrian biggestsize = size; 24043693Sbrian } 24143693Sbrian if (phys_avail[i] < low_water) 24243693Sbrian low_water = phys_avail[i]; 24343313Sbrian if (phys_avail[i + 1] > high_water) 24443313Sbrian high_water = phys_avail[i + 1]; 24543693Sbrian ++nblocks; 24696324Sbrian total += size; 24745910Sbrian } 24881634Sbrian 24981634Sbrian end = phys_avail[biggestone+1]; 25043693Sbrian 25171972Sbrian /* 25296324Sbrian * Initialize the locks. 25381634Sbrian */ 254116586Sume mtx_init(&vm_page_queue_mtx, "vm page queue mutex", NULL, MTX_DEF | 255116622Sume MTX_RECURSE); 256116586Sume mtx_init(&vm_page_queue_free_mtx, "vm page queue free mutex", NULL, 257116586Sume MTX_DEF); 25843313Sbrian 25943693Sbrian /* 26071972Sbrian * Initialize the queue headers for the free queue, the active queue 26143313Sbrian * and the inactive queue. 26243313Sbrian */ 26343313Sbrian vm_pageq_init(); 26471972Sbrian 26571972Sbrian /* 26671972Sbrian * Allocate memory for use when boot strapping the kernel memory 26771972Sbrian * allocator. 26871972Sbrian */ 26943313Sbrian new_end = end - (boot_pages * UMA_SLAB_SIZE); 27043313Sbrian new_end = trunc_page(new_end); 27143693Sbrian mapped = pmap_map(&vaddr, new_end, end, 27271972Sbrian VM_PROT_READ | VM_PROT_WRITE); 27396324Sbrian bzero((void *)mapped, end - new_end); 27496324Sbrian uma_startup((void *)mapped, boot_pages); 27596324Sbrian 27696324Sbrian#if defined(__amd64__) || defined(__i386__) 27796324Sbrian /* 27843693Sbrian * Allocate a bitmap to indicate that a random physical page 27943313Sbrian * needs to be included in a minidump. 28043313Sbrian * 28143693Sbrian * The amd64 port needs this to indicate which direct map pages 28271972Sbrian * need to be dumped, via calls to dump_add_page()/dump_drop_page(). 28371972Sbrian * 28443693Sbrian * However, i386 still needs this workspace internally within the 28543693Sbrian * minidump code. In theory, they are not needed on i386, but are 28643313Sbrian * included should the sf_buf code decide to use them. 28765178Sbrian */ 28871972Sbrian page_range = phys_avail[(nblocks - 1) * 2 + 1] / PAGE_SIZE; 28971972Sbrian vm_page_dump_size = round_page(roundup2(page_range, NBBY) / NBBY); 29071972Sbrian new_end -= vm_page_dump_size; 29171972Sbrian vm_page_dump = (void *)(uintptr_t)pmap_map(&vaddr, new_end, 29265178Sbrian new_end + vm_page_dump_size, VM_PROT_READ | VM_PROT_WRITE); 29365178Sbrian bzero((void *)vm_page_dump, vm_page_dump_size); 29465178Sbrian#endif 29565178Sbrian /* 29643313Sbrian * Compute the number of pages of memory that will be available for 29771972Sbrian * use (taking into account the overhead of a page structure per 29871972Sbrian * page). 29971972Sbrian */ 30043693Sbrian first_page = low_water / PAGE_SIZE; 30143693Sbrian#ifdef VM_PHYSSEG_SPARSE 30243313Sbrian page_range = 0; 30343313Sbrian for (i = 0; phys_avail[i + 1] != 0; i += 2) 30471972Sbrian page_range += atop(phys_avail[i + 1] - phys_avail[i]); 30543693Sbrian#elif defined(VM_PHYSSEG_DENSE) 30671972Sbrian page_range = high_water / PAGE_SIZE - first_page; 30771972Sbrian#else 30843693Sbrian#error "Either VM_PHYSSEG_DENSE or VM_PHYSSEG_SPARSE must be defined." 30943693Sbrian#endif 31043313Sbrian npages = (total - (page_range * sizeof(struct vm_page)) - 31143313Sbrian (end - new_end)) / PAGE_SIZE; 31296324Sbrian end = new_end; 31343313Sbrian 31443313Sbrian /* 31543313Sbrian * Reserve an unmapped guard page to trap access to vm_page_array[-1]. 31696324Sbrian */ 31796324Sbrian vaddr += PAGE_SIZE; 31843313Sbrian 31943313Sbrian /* 32096324Sbrian * Initialize the mem entry structures now, and put them in the free 32143313Sbrian * queue. 32243313Sbrian */ 32396153Sbrian new_end = trunc_page(end - page_range * sizeof(struct vm_page)); 32496153Sbrian mapped = pmap_map(&vaddr, new_end, end, 32596153Sbrian VM_PROT_READ | VM_PROT_WRITE); 32696153Sbrian vm_page_array = (vm_page_t) mapped; 32796324Sbrian#ifdef __amd64__ 32896153Sbrian /* 32996153Sbrian * pmap_map on amd64 comes out of the direct-map, not kvm like i386, 33096153Sbrian * so the pages must be tracked for a crashdump to include this data. 33196324Sbrian * This includes the vm_page_array and the early UMA bootstrap pages. 33296153Sbrian */ 33396153Sbrian for (pa = new_end; pa < phys_avail[biggestone + 1]; pa += PAGE_SIZE) 33496153Sbrian dump_add_page(pa); 33596153Sbrian#endif 33696324Sbrian phys_avail[biggestone + 1] = new_end; 33796153Sbrian 33896153Sbrian /* 33943313Sbrian * Clear all of the page structures 34043313Sbrian */ 34196324Sbrian bzero((caddr_t) vm_page_array, page_range * sizeof(struct vm_page)); 34243313Sbrian vm_page_array_size = page_range; 34343313Sbrian 34443313Sbrian /* 34543313Sbrian * This assertion tests the hypothesis that npages and total are 34696324Sbrian * redundant. XXX 34743313Sbrian */ 34843313Sbrian page_range = 0; 34943313Sbrian for (i = 0; phys_avail[i + 1] != 0; i += 2) 35043313Sbrian page_range += atop(phys_avail[i + 1] - phys_avail[i]); 35143313Sbrian KASSERT(page_range == npages, 35243313Sbrian ("vm_page_startup: inconsistent page counts")); 35343313Sbrian 35443313Sbrian /* 35543313Sbrian * Construct the free queue(s) in descending order (by physical 35643313Sbrian * address) so that the first 16MB of physical memory is allocated 35743313Sbrian * last rather than first. On large-memory machines, this avoids 35896324Sbrian * the exhaustion of low physical memory before isa_dma_init has run. 35943313Sbrian */ 36043313Sbrian cnt.v_page_count = 0; 36143313Sbrian cnt.v_free_count = 0; 36243313Sbrian list = getenv("vm.blacklist"); 36343313Sbrian for (i = 0; phys_avail[i + 1] != 0; i += 2) { 36443313Sbrian pa = phys_avail[i]; 36543313Sbrian last_pa = phys_avail[i + 1]; 36643313Sbrian while (pa < last_pa) { 36743313Sbrian if (list != NULL && 36843313Sbrian vm_page_blacklist_lookup(list, pa)) 36943313Sbrian printf("Skipping page with pa 0x%jx\n", 37043693Sbrian (uintmax_t)pa); 37196324Sbrian else 37243693Sbrian vm_pageq_add_new_page(pa); 37343693Sbrian pa += PAGE_SIZE; 37443313Sbrian } 37543313Sbrian } 37696324Sbrian freeenv(list); 37743693Sbrian return (vaddr); 37881634Sbrian} 379116586Sume 38081634Sbrianvoid 38143313Sbrianvm_page_flag_set(vm_page_t m, unsigned short bits) 38254914Sbrian{ 38354914Sbrian 38454914Sbrian mtx_assert(&vm_page_queue_mtx, MA_OWNED); 38554914Sbrian m->flags |= bits; 38643313Sbrian} 38743313Sbrian 38843313Sbrianvoid 38981634Sbrianvm_page_flag_clear(vm_page_t m, unsigned short bits) 39081634Sbrian{ 39143313Sbrian 39243313Sbrian mtx_assert(&vm_page_queue_mtx, MA_OWNED); 39343313Sbrian m->flags &= ~bits; 39481634Sbrian} 39581634Sbrian 39643313Sbrianvoid 39781634Sbrianvm_page_busy(vm_page_t m) 39881634Sbrian{ 39981634Sbrian 40043313Sbrian VM_OBJECT_LOCK_ASSERT(m->object, MA_OWNED); 40143313Sbrian KASSERT((m->oflags & VPO_BUSY) == 0, 40243313Sbrian ("vm_page_busy: page already busy!!!")); 40343313Sbrian m->oflags |= VPO_BUSY; 40443313Sbrian} 40543313Sbrian 40643313Sbrian/* 40781634Sbrian * vm_page_flash: 40843313Sbrian * 40981634Sbrian * wakeup anyone waiting for the page. 41043313Sbrian */ 41143313Sbrianvoid 41243313Sbrianvm_page_flash(vm_page_t m) 41381634Sbrian{ 41443313Sbrian 41543313Sbrian VM_OBJECT_LOCK_ASSERT(m->object, MA_OWNED); 41643313Sbrian if (m->oflags & VPO_WANTED) { 41796324Sbrian m->oflags &= ~VPO_WANTED; 41896324Sbrian wakeup(m); 41996324Sbrian } 42096324Sbrian} 42196324Sbrian 42296324Sbrian/* 42396324Sbrian * vm_page_wakeup: 42496324Sbrian * 42596324Sbrian * clear the VPO_BUSY flag and wakeup anyone waiting for the 42696324Sbrian * page. 42796324Sbrian * 42896324Sbrian */ 429116586Sumevoid 430116622Sumevm_page_wakeup(vm_page_t m) 431116622Sume{ 432116622Sume 433116622Sume VM_OBJECT_LOCK_ASSERT(m->object, MA_OWNED); 434116622Sume KASSERT(m->oflags & VPO_BUSY, ("vm_page_wakeup: page not busy!!!")); 435116622Sume m->oflags &= ~VPO_BUSY; 436116622Sume vm_page_flash(m); 437116586Sume} 438116586Sume 439116586Sumevoid 440116586Sumevm_page_io_start(vm_page_t m) 441116586Sume{ 442116586Sume 443116586Sume VM_OBJECT_LOCK_ASSERT(m->object, MA_OWNED); 444116586Sume m->busy++; 445116586Sume} 446116586Sume 447116586Sumevoid 448116586Sumevm_page_io_finish(vm_page_t m) 449116586Sume{ 450116586Sume 451116586Sume VM_OBJECT_LOCK_ASSERT(m->object, MA_OWNED); 452116586Sume m->busy--; 453116586Sume if (m->busy == 0) 454116586Sume vm_page_flash(m); 455116586Sume} 456116586Sume 457116586Sume/* 458116586Sume * Keep page from being freed by the page daemon 459116586Sume * much of the same effect as wiring, except much lower 460116586Sume * overhead and should be used only for *very* temporary 461116586Sume * holding ("wiring"). 462116586Sume */ 463116586Sumevoid 464116586Sumevm_page_hold(vm_page_t mem) 465116586Sume{ 466116586Sume 467116586Sume mtx_assert(&vm_page_queue_mtx, MA_OWNED); 468116586Sume mem->hold_count++; 469116586Sume} 470116586Sume 471116586Sumevoid 472116586Sumevm_page_unhold(vm_page_t mem) 473116586Sume{ 474116586Sume 475116586Sume mtx_assert(&vm_page_queue_mtx, MA_OWNED); 476116586Sume --mem->hold_count; 477116586Sume KASSERT(mem->hold_count >= 0, ("vm_page_unhold: hold count < 0!!!")); 478116586Sume if (mem->hold_count == 0 && VM_PAGE_INQUEUE2(mem, PQ_HOLD)) 479116586Sume vm_page_free_toq(mem); 480116586Sume} 481116586Sume 482116586Sume/* 483116586Sume * vm_page_free: 484116586Sume * 485116586Sume * Free a page. 486116586Sume */ 487116586Sumevoid 48896324Sbrianvm_page_free(vm_page_t m) 48996324Sbrian{ 49096324Sbrian 49196324Sbrian m->flags &= ~PG_ZERO; 49296324Sbrian vm_page_free_toq(m); 49396324Sbrian} 49496324Sbrian 49596324Sbrian/* 49696324Sbrian * vm_page_free_zero: 49796324Sbrian * 49896324Sbrian * Free a page to the zerod-pages queue 49996324Sbrian */ 50098132Sbrianvoid 50196324Sbrianvm_page_free_zero(vm_page_t m) 50296324Sbrian{ 50398149Sbrian 50498149Sbrian m->flags |= PG_ZERO; 50598149Sbrian vm_page_free_toq(m); 50698712Sbrian} 50798712Sbrian 50898712Sbrian/* 50998712Sbrian * vm_page_sleep: 51098712Sbrian * 51198712Sbrian * Sleep and release the page queues lock. 51298712Sbrian * 51398967Sbrian * The object containing the given page must be locked. 51498967Sbrian */ 51598967Sbrianvoid 51698712Sbrianvm_page_sleep(vm_page_t m, const char *msg) 51798712Sbrian{ 51898149Sbrian 51998149Sbrian VM_OBJECT_LOCK_ASSERT(m->object, MA_OWNED); 52098149Sbrian if (!mtx_owned(&vm_page_queue_mtx)) 52198149Sbrian vm_page_lock_queues(); 52298149Sbrian vm_page_flag_set(m, PG_REFERENCED); 52398149Sbrian vm_page_unlock_queues(); 52498149Sbrian 52596324Sbrian /* 52696324Sbrian * It's possible that while we sleep, the page will get 52796730Sbrian * unbusied and freed. If we are holding the object 52896730Sbrian * lock, we will assume we hold a reference to the object 52996730Sbrian * such that even if m->object changes, we can re-lock 53098149Sbrian * it. 53198149Sbrian */ 53298149Sbrian m->oflags |= VPO_WANTED; 53398712Sbrian msleep(m, VM_OBJECT_MTX(m->object), PVM, msg, 0); 53498712Sbrian} 53598712Sbrian 53698712Sbrian/* 53798712Sbrian * vm_page_dirty: 53898712Sbrian * 53998712Sbrian * make page all dirty 54098967Sbrian */ 54198967Sbrianvoid 54298967Sbrianvm_page_dirty(vm_page_t m) 54398712Sbrian{ 54498712Sbrian KASSERT(VM_PAGE_GETKNOWNQUEUE1(m) != PQ_CACHE, 54598149Sbrian ("vm_page_dirty: page in cache!")); 54698149Sbrian KASSERT(VM_PAGE_GETKNOWNQUEUE1(m) != PQ_FREE, 54798149Sbrian ("vm_page_dirty: page is free!")); 54898149Sbrian m->dirty = VM_PAGE_BITS_ALL; 54998149Sbrian} 55098149Sbrian 55198149Sbrian/* 55298149Sbrian * vm_page_splay: 55396730Sbrian * 55496730Sbrian * Implements Sleator and Tarjan's top-down splay algorithm. Returns 55598243Sbrian * the vm_page containing the given pindex. If, however, that 55698132Sbrian * pindex is not found in the vm_object, returns a vm_page that is 55798132Sbrian * adjacent to the pindex, coming before or after it. 55898132Sbrian */ 55998132Sbrianvm_page_t 56098132Sbrianvm_page_splay(vm_pindex_t pindex, vm_page_t root) 56198132Sbrian{ 56298132Sbrian struct vm_page dummy; 56398132Sbrian vm_page_t lefttreemax, righttreemin, y; 56498132Sbrian 56598132Sbrian if (root == NULL) 56698132Sbrian return (root); 56798132Sbrian lefttreemax = righttreemin = &dummy; 56898132Sbrian for (;; root = y) { 56998132Sbrian if (pindex < root->pindex) { 57098132Sbrian if ((y = root->left) == NULL) 57198132Sbrian break; 57298132Sbrian if (pindex < y->pindex) { 57398132Sbrian /* Rotate right. */ 57498132Sbrian root->left = y->right; 57598132Sbrian y->right = root; 57698132Sbrian root = y; 57798132Sbrian if ((y = root->left) == NULL) 57898132Sbrian break; 57998132Sbrian } 58096324Sbrian /* Link into the new root's right tree. */ 58196324Sbrian righttreemin->left = root; 58296324Sbrian righttreemin = root; 58396324Sbrian } else if (pindex > root->pindex) { 58496324Sbrian if ((y = root->right) == NULL) 58596324Sbrian break; 58696324Sbrian if (pindex > y->pindex) { 58796324Sbrian /* Rotate left. */ 58896324Sbrian root->right = y->left; 58996324Sbrian y->left = root; 59096324Sbrian root = y; 59196324Sbrian if ((y = root->right) == NULL) 59296324Sbrian break; 59396324Sbrian } 59496324Sbrian /* Link into the new root's left tree. */ 59596324Sbrian lefttreemax->right = root; 59696324Sbrian lefttreemax = root; 59743313Sbrian } else 59843313Sbrian break; 59943313Sbrian } 60096324Sbrian /* Assemble the new root. */ 60143693Sbrian lefttreemax->right = root->left; 60243693Sbrian righttreemin->left = root->right; 60343693Sbrian root->left = dummy.right; 60496324Sbrian root->right = dummy.left; 60596324Sbrian return (root); 60696324Sbrian} 60743693Sbrian 60843693Sbrian/* 60943313Sbrian * vm_page_insert: [ internal use only ] 61096324Sbrian * 61143693Sbrian * Inserts the given mem entry into the object and object list. 61243313Sbrian * 61343693Sbrian * The pagetables are not updated but will presumably fault the page 61458038Sbrian * in if necessary, or if a kernel page the caller will at some point 61543693Sbrian * enter the page into the kernel's pmap. We are not allowed to block 61643693Sbrian * here so we *can't* do this anyway. 61743693Sbrian * 61843693Sbrian * The object and page must be locked. 61943693Sbrian * This routine may not block. 62043693Sbrian */ 62143313Sbrianvoid 62243693Sbrianvm_page_insert(vm_page_t m, vm_object_t object, vm_pindex_t pindex) 62343693Sbrian{ 62443693Sbrian vm_page_t root; 62543693Sbrian 62643693Sbrian VM_OBJECT_LOCK_ASSERT(object, MA_OWNED); 62743693Sbrian if (m->object != NULL) 62843693Sbrian panic("vm_page_insert: page already inserted"); 62943693Sbrian 63043693Sbrian /* 63143313Sbrian * Record the object/offset pair in this page 63243313Sbrian */ 63343693Sbrian m->object = object; 63443693Sbrian m->pindex = pindex; 63543693Sbrian 63643693Sbrian /* 63743693Sbrian * Now link into the object's ordered list of backed pages. 63843693Sbrian */ 63943693Sbrian root = object->root; 64043693Sbrian if (root == NULL) { 64143693Sbrian m->left = NULL; 64243693Sbrian m->right = NULL; 64343693Sbrian TAILQ_INSERT_TAIL(&object->memq, m, listq); 64443693Sbrian } else { 64543693Sbrian root = vm_page_splay(pindex, root); 64658028Sbrian if (pindex < root->pindex) { 64743693Sbrian m->left = root->left; 64843693Sbrian m->right = root; 64943693Sbrian root->left = NULL; 65043693Sbrian TAILQ_INSERT_BEFORE(root, m, listq); 65143693Sbrian } else if (pindex == root->pindex) 65258038Sbrian panic("vm_page_insert: offset already allocated"); 65343693Sbrian else { 65443693Sbrian m->right = root->right; 65558028Sbrian m->left = root; 65643693Sbrian root->right = NULL; 65743693Sbrian TAILQ_INSERT_AFTER(&object->memq, root, m, listq); 65843693Sbrian } 65943693Sbrian } 66043693Sbrian object->root = m; 66143693Sbrian object->generation++; 66243693Sbrian 66343693Sbrian /* 66443693Sbrian * show that the object has one more resident page. 66543693Sbrian */ 66643693Sbrian object->resident_page_count++; 66743693Sbrian /* 66843693Sbrian * Hold the vnode until the last page is released. 66943693Sbrian */ 67043693Sbrian if (object->resident_page_count == 1 && object->type == OBJT_VNODE) 67158038Sbrian vhold((struct vnode *)object->handle); 67243693Sbrian 67343693Sbrian /* 67458028Sbrian * Since we are inserting a new and possibly dirty page, 67543693Sbrian * update the object's OBJ_MIGHTBEDIRTY flag. 67643693Sbrian */ 67743693Sbrian if (m->flags & PG_WRITEABLE) 67843693Sbrian vm_object_set_writeable_dirty(object); 67943693Sbrian} 68043693Sbrian 68143693Sbrian/* 68258038Sbrian * vm_page_remove: 68343693Sbrian * NOTE: used by device pager as well -wfj 68443693Sbrian * 68558028Sbrian * Removes the given mem entry from the object/offset-page 68643693Sbrian * table and the object page list, but do not invalidate/terminate 68743693Sbrian * the backing store. 68843693Sbrian * 68943693Sbrian * The object and page must be locked. 69043693Sbrian * The underlying pmap entry (if any) is NOT removed here. 69143693Sbrian * This routine may not block. 69243693Sbrian */ 69343693Sbrianvoid 69443693Sbrianvm_page_remove(vm_page_t m) 69543313Sbrian{ 69643693Sbrian vm_object_t object; 69743693Sbrian vm_page_t root; 69843693Sbrian 69943693Sbrian if ((object = m->object) == NULL) 70043693Sbrian return; 70143693Sbrian VM_OBJECT_LOCK_ASSERT(object, MA_OWNED); 70243693Sbrian if (m->oflags & VPO_BUSY) { 70396324Sbrian m->oflags &= ~VPO_BUSY; 70496324Sbrian vm_page_flash(m); 70543693Sbrian } 70696324Sbrian mtx_assert(&vm_page_queue_mtx, MA_OWNED); 70796324Sbrian 70896324Sbrian /* 70996324Sbrian * Now remove from the object's list of backed pages. 71096324Sbrian */ 71196324Sbrian if (m != object->root) 71296324Sbrian vm_page_splay(m->pindex, object->root); 71396730Sbrian if (m->left == NULL) 71496324Sbrian root = m->right; 715116586Sume else { 716116622Sume root = vm_page_splay(m->pindex, m->left); 717116586Sume root->right = m->right; 718116586Sume } 71996324Sbrian object->root = root; 72098132Sbrian TAILQ_REMOVE(&object->memq, m, listq); 72198132Sbrian 72298132Sbrian /* 72398132Sbrian * And show that the object has one fewer resident page. 72498132Sbrian */ 72598132Sbrian object->resident_page_count--; 72696324Sbrian object->generation++; 72765178Sbrian /* 72843693Sbrian * The vnode may now be recycled. 72943693Sbrian */ 73043693Sbrian if (object->resident_page_count == 0 && object->type == OBJT_VNODE) 73143693Sbrian vdrop((struct vnode *)object->handle); 73243693Sbrian 73343693Sbrian m->object = NULL; 73443693Sbrian} 73543693Sbrian 73643693Sbrian/* 73765178Sbrian * vm_page_lookup: 73843693Sbrian * 73943693Sbrian * Returns the page associated with the object/offset 740116586Sume * pair specified; if none is found, NULL is returned. 741116586Sume * 742116586Sume * The object must be locked. 74396153Sbrian * This routine may not block. 74496153Sbrian * This is a critical path routine 74596730Sbrian */ 74696730Sbrianvm_page_t 74796324Sbrianvm_page_lookup(vm_object_t object, vm_pindex_t pindex) 74896324Sbrian{ 749116622Sume vm_page_t m; 750116622Sume 751116622Sume VM_OBJECT_LOCK_ASSERT(object, MA_OWNED); 752116622Sume if ((m = object->root) != NULL && m->pindex != pindex) { 75396324Sbrian m = vm_page_splay(pindex, m); 75496324Sbrian if ((object->root = m)->pindex != pindex) 75598132Sbrian m = NULL; 75698132Sbrian } 75798132Sbrian return (m); 75898132Sbrian} 75998132Sbrian 76098132Sbrian/* 76143693Sbrian * vm_page_rename: 76243693Sbrian * 76343693Sbrian * Move the given memory entry from its 76443693Sbrian * current object to the specified target object/offset. 76543693Sbrian * 76643693Sbrian * The object must be locked. 76796582Sbrian * This routine may not block. 76896582Sbrian * 76996582Sbrian * Note: swap associated with the page must be invalidated by the move. We 77096582Sbrian * have to do this for several reasons: (1) we aren't freeing the 77196582Sbrian * page, (2) we are dirtying the page, (3) the VM system is probably 77296582Sbrian * moving the page from object A to B, and will then later move 77396582Sbrian * the backing store from A to B and we can't have a conflict. 77496582Sbrian * 77596582Sbrian * Note: we *always* dirty the page. It is necessary both for the 77696582Sbrian * fact that we moved it, and because we may be invalidating 77796582Sbrian * swap. If the page is on the cache, we have to deactivate it 77896582Sbrian * or vm_page_dirty() will panic. Dirty pages are not allowed 77996582Sbrian * on the cache. 78096582Sbrian */ 78196582Sbrianvoid 78296582Sbrianvm_page_rename(vm_page_t m, vm_object_t new_object, vm_pindex_t new_pindex) 78396582Sbrian{ 78496582Sbrian 78596582Sbrian vm_page_remove(m); 78696582Sbrian vm_page_insert(m, new_object, new_pindex); 78796582Sbrian if (VM_PAGE_INQUEUE1(m, PQ_CACHE)) 78896582Sbrian vm_page_deactivate(m); 78996582Sbrian vm_page_dirty(m); 79096582Sbrian} 79196582Sbrian 79296582Sbrian/* 79396582Sbrian * vm_page_select_cache: 79496582Sbrian * 79596582Sbrian * Move a page of the given color from the cache queue to the free 79696582Sbrian * queue. As pages might be found, but are not applicable, they are 79796582Sbrian * deactivated. 79896582Sbrian * 79996582Sbrian * This routine may not block. 80096582Sbrian */ 80196582Sbrianvm_page_t 80296582Sbrianvm_page_select_cache(int color) 80396582Sbrian{ 80496582Sbrian vm_object_t object; 80596582Sbrian vm_page_t m; 80696582Sbrian boolean_t was_trylocked; 80796582Sbrian 80896582Sbrian mtx_assert(&vm_page_queue_mtx, MA_OWNED); 80996582Sbrian while ((m = vm_pageq_find(PQ_CACHE, color, FALSE)) != NULL) { 81096582Sbrian KASSERT(m->dirty == 0, ("Found dirty cache page %p", m)); 81196582Sbrian KASSERT(!pmap_page_is_mapped(m), 81243693Sbrian ("Found mapped cache page %p", m)); 81343693Sbrian KASSERT((m->flags & PG_UNMANAGED) == 0, 81443693Sbrian ("Found unmanaged cache page %p", m)); 81596730Sbrian KASSERT(m->wire_count == 0, ("Found wired cache page %p", m)); 81643693Sbrian if (m->hold_count == 0 && (object = m->object, 81796730Sbrian (was_trylocked = VM_OBJECT_TRYLOCK(object)) || 81898311Sbrian VM_OBJECT_LOCKED(object))) { 81943693Sbrian KASSERT((m->oflags & VPO_BUSY) == 0 && m->busy == 0, 82043693Sbrian ("Found busy cache page %p", m)); 82196582Sbrian vm_page_free(m); 82274049Sbrian if (was_trylocked) 82397738Sbrian VM_OBJECT_UNLOCK(object); 82450840Sbrian break; 82598132Sbrian } 82697738Sbrian vm_page_deactivate(m); 82796324Sbrian } 82896730Sbrian return (m); 82996730Sbrian} 83098311Sbrian 83196324Sbrian/* 83243693Sbrian * vm_page_alloc: 83343693Sbrian * 83496730Sbrian * Allocate and return a memory cell associated 83543693Sbrian * with this VM object/offset pair. 83643693Sbrian * 83743693Sbrian * page_req classes: 83843693Sbrian * VM_ALLOC_NORMAL normal process request 83943693Sbrian * VM_ALLOC_SYSTEM system *really* needs a page 84043693Sbrian * VM_ALLOC_INTERRUPT interrupt time request 84196730Sbrian * VM_ALLOC_ZERO zero page 84243693Sbrian * 84343693Sbrian * This routine may not block. 84443693Sbrian * 84565178Sbrian * Additional special handling is required when called from an 84665178Sbrian * interrupt (VM_ALLOC_INTERRUPT). We are not allowed to mess with 84796730Sbrian * the page cache in this case. 84843693Sbrian */ 84943693Sbrianvm_page_t 85043693Sbrianvm_page_alloc(vm_object_t object, vm_pindex_t pindex, int req) 85143693Sbrian{ 85243693Sbrian vm_page_t m = NULL; 85396730Sbrian int color, flags, page_req; 85443693Sbrian 85543693Sbrian page_req = req & VM_ALLOC_CLASS_MASK; 85643693Sbrian KASSERT(curthread->td_intr_nesting_level == 0 || 85743693Sbrian page_req == VM_ALLOC_INTERRUPT, 85843693Sbrian ("vm_page_alloc(NORMAL|SYSTEM) in interrupt context")); 85996730Sbrian 86043693Sbrian if ((req & VM_ALLOC_NOOBJ) == 0) { 86143693Sbrian KASSERT(object != NULL, 86299418Sbrian ("vm_page_alloc: NULL object.")); 86343693Sbrian VM_OBJECT_LOCK_ASSERT(object, MA_OWNED); 86443693Sbrian color = (pindex + object->pg_color) & PQ_COLORMASK; 86543693Sbrian } else 86643693Sbrian color = pindex & PQ_COLORMASK; 86796730Sbrian 86843693Sbrian /* 86943693Sbrian * The pager is allowed to eat deeper into the free page list. 87096324Sbrian */ 87196324Sbrian if ((curproc == pageproc) && (page_req != VM_ALLOC_INTERRUPT)) { 87296324Sbrian page_req = VM_ALLOC_SYSTEM; 87396324Sbrian }; 87496324Sbrian 87543693Sbrianloop: 87643693Sbrian mtx_lock(&vm_page_queue_free_mtx); 87796730Sbrian if (cnt.v_free_count > cnt.v_free_reserved || 87843693Sbrian (page_req == VM_ALLOC_SYSTEM && 87996324Sbrian cnt.v_cache_count == 0 && 88096324Sbrian cnt.v_free_count > cnt.v_interrupt_free_min) || 88196324Sbrian (page_req == VM_ALLOC_INTERRUPT && cnt.v_free_count > 0)) { 88296324Sbrian /* 88396324Sbrian * Allocate from the free queue if the number of free pages 88496324Sbrian * exceeds the minimum for the request class. 88596730Sbrian */ 88696324Sbrian m = vm_pageq_find(PQ_FREE, color, (req & VM_ALLOC_ZERO) != 0); 88796324Sbrian } else if (page_req != VM_ALLOC_INTERRUPT) { 88896324Sbrian mtx_unlock(&vm_page_queue_free_mtx); 88996730Sbrian /* 89096324Sbrian * Allocatable from cache (non-interrupt only). On success, 89196324Sbrian * we must free the page and try again, thus ensuring that 89296324Sbrian * cnt.v_*_free_min counters are replenished. 89396324Sbrian */ 89496324Sbrian vm_page_lock_queues(); 89596324Sbrian if ((m = vm_page_select_cache(color)) == NULL) { 89696730Sbrian KASSERT(cnt.v_cache_count == 0, 89796324Sbrian ("vm_page_alloc: cache queue is missing %d pages", 89896730Sbrian cnt.v_cache_count)); 89996324Sbrian vm_page_unlock_queues(); 90096730Sbrian atomic_add_int(&vm_pageout_deficit, 1); 90196324Sbrian pagedaemon_wakeup(); 90296730Sbrian 90396730Sbrian if (page_req != VM_ALLOC_SYSTEM) 90496730Sbrian return (NULL); 90596730Sbrian 90696730Sbrian mtx_lock(&vm_page_queue_free_mtx); 90796324Sbrian if (cnt.v_free_count <= cnt.v_interrupt_free_min) { 90896730Sbrian mtx_unlock(&vm_page_queue_free_mtx); 90996730Sbrian return (NULL); 91096324Sbrian } 91196324Sbrian m = vm_pageq_find(PQ_FREE, color, (req & VM_ALLOC_ZERO) != 0); 91296324Sbrian } else { 91398311Sbrian vm_page_unlock_queues(); 91496730Sbrian goto loop; 91596730Sbrian } 91696730Sbrian } else { 91796730Sbrian /* 91896730Sbrian * Not allocatable from cache from interrupt, give up. 91998311Sbrian */ 92096730Sbrian mtx_unlock(&vm_page_queue_free_mtx); 92196730Sbrian atomic_add_int(&vm_pageout_deficit, 1); 92296730Sbrian pagedaemon_wakeup(); 92398311Sbrian return (NULL); 92498311Sbrian } 92596730Sbrian 92698311Sbrian /* 92798311Sbrian * At this point we had better have found a good page. 92896730Sbrian */ 92996730Sbrian 93096730Sbrian KASSERT( 93196730Sbrian m != NULL, 93296324Sbrian ("vm_page_alloc(): missing page on free queue") 93396324Sbrian ); 93498243Sbrian 93596324Sbrian /* 93696324Sbrian * Remove from free queue 93796730Sbrian */ 93896324Sbrian vm_pageq_remove_nowakeup(m); 93943693Sbrian 94043693Sbrian /* 94150840Sbrian * Initialize structure. Only the PG_ZERO flag is inherited. 94250840Sbrian */ 94350840Sbrian flags = 0; 94497738Sbrian if (m->flags & PG_ZERO) { 94550840Sbrian vm_page_zero_count--; 94650840Sbrian if (req & VM_ALLOC_ZERO) 94750840Sbrian flags = PG_ZERO; 94850840Sbrian } 94950840Sbrian if (object != NULL && object->type == OBJT_PHYS) 95050840Sbrian flags |= PG_UNMANAGED; 95196730Sbrian m->flags = flags; 95250840Sbrian if (req & (VM_ALLOC_NOBUSY | VM_ALLOC_NOOBJ)) 95350840Sbrian m->oflags = 0; 95497738Sbrian else 95550840Sbrian m->oflags = VPO_BUSY; 95650840Sbrian if (req & VM_ALLOC_WIRED) { 95750840Sbrian atomic_add_int(&cnt.v_wire_count, 1); 95850840Sbrian m->wire_count = 1; 95996730Sbrian } else 96050840Sbrian m->wire_count = 0; 96150840Sbrian m->hold_count = 0; 96250840Sbrian m->act_count = 0; 96396582Sbrian m->busy = 0; 96450840Sbrian m->valid = 0; 96571972Sbrian KASSERT(m->dirty == 0, ("vm_page_alloc: free/cache page %p was dirty", m)); 96643693Sbrian mtx_unlock(&vm_page_queue_free_mtx); 96743693Sbrian 96843693Sbrian if ((req & VM_ALLOC_NOOBJ) == 0) 96943693Sbrian vm_page_insert(m, object, pindex); 97043693Sbrian else 97143693Sbrian m->pindex = pindex; 97243693Sbrian 97371972Sbrian /* 97443693Sbrian * Don't wakeup too often - wakeup the pageout daemon when 97543693Sbrian * we would be nearly out of memory. 97643693Sbrian */ 97796730Sbrian if (vm_paging_needed()) 97896730Sbrian pagedaemon_wakeup(); 97943693Sbrian 98043693Sbrian return (m); 981116588Sume} 982116588Sume 983116588Sume/* 984116588Sume * vm_wait: (also see VM_WAIT macro) 985116588Sume * 986116588Sume * Block until free pages are available for allocation 987116588Sume * - Called in various places before memory allocations. 988116588Sume */ 989116588Sumevoid 990116588Sumevm_wait(void) 991116588Sume{ 992116588Sume 993116588Sume mtx_lock(&vm_page_queue_free_mtx); 994116588Sume if (curproc == pageproc) { 995116588Sume vm_pageout_pages_needed = 1; 996116588Sume msleep(&vm_pageout_pages_needed, &vm_page_queue_free_mtx, 997116588Sume PDROP | PSWP, "VMWait", 0); 998116588Sume } else { 999116588Sume if (!vm_pages_needed) { 1000116588Sume vm_pages_needed = 1; 100143693Sbrian wakeup(&vm_pages_needed); 100265178Sbrian } 100365178Sbrian msleep(&cnt.v_free_count, &vm_page_queue_free_mtx, PDROP | PVM, 100465178Sbrian "vmwait", 0); 100598243Sbrian } 1006116588Sume} 100765178Sbrian 100865178Sbrian/* 100996582Sbrian * vm_waitpfault: (also see VM_WAITPFAULT macro) 101074049Sbrian * 101197738Sbrian * Block until free pages are available for allocation 101265178Sbrian * - Called only in vm_fault so that processes page faulting 101398132Sbrian * can be easily tracked. 101497738Sbrian * - Sleeps at a lower priority than vm_wait() so that vm_wait()ing 101565178Sbrian * processes will be able to grab memory first. Do not change 101665178Sbrian * this balance without careful testing first. 101765178Sbrian */ 101865178Sbrianvoid 101965178Sbrianvm_waitpfault(void) 102065178Sbrian{ 102165178Sbrian 102265178Sbrian mtx_lock(&vm_page_queue_free_mtx); 102365178Sbrian if (!vm_pages_needed) { 102465178Sbrian vm_pages_needed = 1; 102565178Sbrian wakeup(&vm_pages_needed); 102698132Sbrian } 102765178Sbrian msleep(&cnt.v_free_count, &vm_page_queue_free_mtx, PDROP | PUSER, 102867133Sbrian "pfault", 0); 102965178Sbrian} 103065178Sbrian 103165178Sbrian/* 103265178Sbrian * vm_page_activate: 103365178Sbrian * 103465178Sbrian * Put the specified page on the active list (if appropriate). 103565178Sbrian * Ensure that act_count is at least ACT_INIT but do not otherwise 103665178Sbrian * mess with it. 103765178Sbrian * 103865178Sbrian * The page queues must be locked. 103965178Sbrian * This routine may not block. 104065178Sbrian */ 104165178Sbrianvoid 104265178Sbrianvm_page_activate(vm_page_t m) 104365178Sbrian{ 104465178Sbrian 104565178Sbrian mtx_assert(&vm_page_queue_mtx, MA_OWNED); 104665178Sbrian if (VM_PAGE_GETKNOWNQUEUE2(m) != PQ_ACTIVE) { 104765178Sbrian if (VM_PAGE_INQUEUE1(m, PQ_CACHE)) 104865178Sbrian PCPU_INC(cnt.v_reactivated); 104999418Sbrian vm_pageq_remove(m); 105065178Sbrian if (m->wire_count == 0 && (m->flags & PG_UNMANAGED) == 0) { 105165178Sbrian if (m->act_count < ACT_INIT) 105265178Sbrian m->act_count = ACT_INIT; 105398243Sbrian vm_pageq_enqueue(PQ_ACTIVE, m); 105465178Sbrian } 105597904Sbrian } else { 105697904Sbrian if (m->act_count < ACT_INIT) 105799418Sbrian m->act_count = ACT_INIT; 105865178Sbrian } 105965178Sbrian} 106065178Sbrian 106165178Sbrian/* 106265178Sbrian * vm_page_free_wakeup: 106365178Sbrian * 106465178Sbrian * Helper routine for vm_page_free_toq() and vm_page_cache(). This 106565178Sbrian * routine is called when a page has been added to the cache or free 106665178Sbrian * queues. 1067116588Sume * 106865178Sbrian * The page queues must be locked. 106965178Sbrian * This routine may not block. 107065178Sbrian */ 107165178Sbrianstatic inline void 1072116588Sumevm_page_free_wakeup(void) 1073116588Sume{ 1074116588Sume 1075116588Sume mtx_assert(&vm_page_queue_free_mtx, MA_OWNED); 1076116588Sume /* 1077116588Sume * if pageout daemon needs pages, then tell it that there are 1078116588Sume * some free. 1079116588Sume */ 1080116588Sume if (vm_pageout_pages_needed && 1081116588Sume cnt.v_cache_count + cnt.v_free_count >= cnt.v_pageout_free_min) { 1082116588Sume wakeup(&vm_pageout_pages_needed); 1083116588Sume vm_pageout_pages_needed = 0; 1084116588Sume } 1085116588Sume /* 1086116588Sume * wakeup processes that are waiting on memory if we hit a 1087116588Sume * high water mark. And wakeup scheduler process if we have 1088116588Sume * lots of memory. this process will swapin processes. 1089116622Sume */ 1090116622Sume if (vm_pages_needed && !vm_page_count_min()) { 1091116622Sume vm_pages_needed = 0; 1092116622Sume wakeup(&cnt.v_free_count); 1093116622Sume } 1094116622Sume} 1095116622Sume 1096116622Sume/* 1097116622Sume * vm_page_free_toq: 1098116622Sume * 1099116622Sume * Returns the given page to the PQ_FREE list, 1100116622Sume * disassociating it with any VM object. 1101116588Sume * 1102116588Sume * Object and page must be locked prior to entry. 1103116588Sume * This routine may not block. 1104116588Sume */ 1105116588Sume 1106116588Sumevoid 110765178Sbrianvm_page_free_toq(vm_page_t m) 110865178Sbrian{ 110965178Sbrian struct vpgqueues *pq; 111065178Sbrian 111197738Sbrian if (VM_PAGE_GETQUEUE(m) != PQ_NONE) 111265178Sbrian mtx_assert(&vm_page_queue_mtx, MA_OWNED); 111365178Sbrian KASSERT(!pmap_page_is_mapped(m), 111465178Sbrian ("vm_page_free_toq: freeing mapped page %p", m)); 111565178Sbrian PCPU_INC(cnt.v_tfree); 111665178Sbrian 111765178Sbrian if (m->busy || VM_PAGE_INQUEUE1(m, PQ_FREE)) { 111865178Sbrian printf( 111965178Sbrian "vm_page_free: pindex(%lu), busy(%d), VPO_BUSY(%d), hold(%d)\n", 112065178Sbrian (u_long)m->pindex, m->busy, (m->oflags & VPO_BUSY) ? 1 : 0, 112197738Sbrian m->hold_count); 112265178Sbrian if (VM_PAGE_INQUEUE1(m, PQ_FREE)) 112365178Sbrian panic("vm_page_free: freeing free page"); 112465178Sbrian else 112565178Sbrian panic("vm_page_free: freeing busy page"); 112665178Sbrian } 112765178Sbrian 112865178Sbrian /* 112965178Sbrian * unqueue, then remove page. Note that we cannot destroy 113096582Sbrian * the page here because we do not want to call the pager's 113165178Sbrian * callback routine until after we've put the page on the 113265178Sbrian * appropriate free queue. 113398243Sbrian */ 113465178Sbrian vm_pageq_remove_nowakeup(m); 113565178Sbrian vm_page_remove(m); 113698243Sbrian 113765178Sbrian /* 113865178Sbrian * If fictitious remove object association and 113965178Sbrian * return, otherwise delay object association removal. 114065178Sbrian */ 114165178Sbrian if ((m->flags & PG_FICTITIOUS) != 0) { 114265178Sbrian return; 114365178Sbrian } 114465178Sbrian 114565178Sbrian m->valid = 0; 114665178Sbrian vm_page_undirty(m); 114765178Sbrian 114865178Sbrian if (m->wire_count != 0) { 114965178Sbrian if (m->wire_count > 1) { 115065178Sbrian panic("vm_page_free: invalid wire count (%d), pindex: 0x%lx", 115165178Sbrian m->wire_count, (long)m->pindex); 115265178Sbrian } 115365178Sbrian panic("vm_page_free: freeing wired page"); 115465178Sbrian } 115565178Sbrian if (m->hold_count != 0) { 115665178Sbrian m->flags &= ~PG_ZERO; 115771972Sbrian vm_pageq_enqueue(PQ_HOLD, m); 115865178Sbrian return; 115965178Sbrian } 116065178Sbrian VM_PAGE_SETQUEUE1(m, PQ_FREE); 116165178Sbrian mtx_lock(&vm_page_queue_free_mtx); 116265178Sbrian pq = &vm_page_queues[VM_PAGE_GETQUEUE(m)]; 116365178Sbrian pq->lcnt++; 116471972Sbrian ++(*pq->cnt); 116565178Sbrian 116665178Sbrian /* 116765178Sbrian * Put zero'd pages on the end ( where we look for zero'd pages 116865178Sbrian * first ) and non-zerod pages at the head. 116965178Sbrian */ 117065178Sbrian if (m->flags & PG_ZERO) { 117143693Sbrian TAILQ_INSERT_TAIL(&pq->pl, m, pageq); 117243693Sbrian ++vm_page_zero_count; 117343693Sbrian } else { 117443313Sbrian TAILQ_INSERT_HEAD(&pq->pl, m, pageq); 117543313Sbrian vm_page_zero_idle_wakeup(); 117671657Sbrian } 117771657Sbrian vm_page_free_wakeup(); 117843313Sbrian mtx_unlock(&vm_page_queue_free_mtx); 117971657Sbrian} 118071657Sbrian 118171657Sbrian/* 118271657Sbrian * vm_page_wire: 118396324Sbrian * 118498132Sbrian * Mark this page as wired down by yet 118598132Sbrian * another map, removing it from paging queues 118698132Sbrian * as necessary. 118798132Sbrian * 118898132Sbrian * The page queues must be locked. 118998132Sbrian * This routine may not block. 119098132Sbrian */ 119198132Sbrianvoid 119296730Sbrianvm_page_wire(vm_page_t m) 119396730Sbrian{ 119496324Sbrian 119543313Sbrian /* 119671657Sbrian * Only bump the wire statistics if the page is not already wired, 1197116586Sume * and only unqueue the page if it is on some queue (if it is unmanaged 1198116586Sume * it is already off the queues). 1199116586Sume */ 1200116586Sume mtx_assert(&vm_page_queue_mtx, MA_OWNED); 120143313Sbrian if (m->flags & PG_FICTITIOUS) 120243313Sbrian return; 120343313Sbrian if (m->wire_count == 0) { 1204 if ((m->flags & PG_UNMANAGED) == 0) 1205 vm_pageq_remove(m); 1206 atomic_add_int(&cnt.v_wire_count, 1); 1207 } 1208 m->wire_count++; 1209 KASSERT(m->wire_count != 0, ("vm_page_wire: wire_count overflow m=%p", m)); 1210} 1211 1212/* 1213 * vm_page_unwire: 1214 * 1215 * Release one wiring of this page, potentially 1216 * enabling it to be paged again. 1217 * 1218 * Many pages placed on the inactive queue should actually go 1219 * into the cache, but it is difficult to figure out which. What 1220 * we do instead, if the inactive target is well met, is to put 1221 * clean pages at the head of the inactive queue instead of the tail. 1222 * This will cause them to be moved to the cache more quickly and 1223 * if not actively re-referenced, freed more quickly. If we just 1224 * stick these pages at the end of the inactive queue, heavy filesystem 1225 * meta-data accesses can cause an unnecessary paging load on memory bound 1226 * processes. This optimization causes one-time-use metadata to be 1227 * reused more quickly. 1228 * 1229 * BUT, if we are in a low-memory situation we have no choice but to 1230 * put clean pages on the cache queue. 1231 * 1232 * A number of routines use vm_page_unwire() to guarantee that the page 1233 * will go into either the inactive or active queues, and will NEVER 1234 * be placed in the cache - for example, just after dirtying a page. 1235 * dirty pages in the cache are not allowed. 1236 * 1237 * The page queues must be locked. 1238 * This routine may not block. 1239 */ 1240void 1241vm_page_unwire(vm_page_t m, int activate) 1242{ 1243 1244 mtx_assert(&vm_page_queue_mtx, MA_OWNED); 1245 if (m->flags & PG_FICTITIOUS) 1246 return; 1247 if (m->wire_count > 0) { 1248 m->wire_count--; 1249 if (m->wire_count == 0) { 1250 atomic_subtract_int(&cnt.v_wire_count, 1); 1251 if (m->flags & PG_UNMANAGED) { 1252 ; 1253 } else if (activate) 1254 vm_pageq_enqueue(PQ_ACTIVE, m); 1255 else { 1256 vm_page_flag_clear(m, PG_WINATCFLS); 1257 vm_pageq_enqueue(PQ_INACTIVE, m); 1258 } 1259 } 1260 } else { 1261 panic("vm_page_unwire: invalid wire count: %d", m->wire_count); 1262 } 1263} 1264 1265 1266/* 1267 * Move the specified page to the inactive queue. If the page has 1268 * any associated swap, the swap is deallocated. 1269 * 1270 * Normally athead is 0 resulting in LRU operation. athead is set 1271 * to 1 if we want this page to be 'as if it were placed in the cache', 1272 * except without unmapping it from the process address space. 1273 * 1274 * This routine may not block. 1275 */ 1276static inline void 1277_vm_page_deactivate(vm_page_t m, int athead) 1278{ 1279 1280 mtx_assert(&vm_page_queue_mtx, MA_OWNED); 1281 1282 /* 1283 * Ignore if already inactive. 1284 */ 1285 if (VM_PAGE_INQUEUE2(m, PQ_INACTIVE)) 1286 return; 1287 if (m->wire_count == 0 && (m->flags & PG_UNMANAGED) == 0) { 1288 if (VM_PAGE_INQUEUE1(m, PQ_CACHE)) 1289 PCPU_INC(cnt.v_reactivated); 1290 vm_page_flag_clear(m, PG_WINATCFLS); 1291 vm_pageq_remove(m); 1292 if (athead) 1293 TAILQ_INSERT_HEAD(&vm_page_queues[PQ_INACTIVE].pl, m, pageq); 1294 else 1295 TAILQ_INSERT_TAIL(&vm_page_queues[PQ_INACTIVE].pl, m, pageq); 1296 VM_PAGE_SETQUEUE2(m, PQ_INACTIVE); 1297 vm_page_queues[PQ_INACTIVE].lcnt++; 1298 1299 /* 1300 * Just not use an atomic here since vm_page_queues_lock 1301 * alredy protects this field. 1302 */ 1303 cnt.v_inactive_count++; 1304 } 1305} 1306 1307void 1308vm_page_deactivate(vm_page_t m) 1309{ 1310 _vm_page_deactivate(m, 0); 1311} 1312 1313/* 1314 * vm_page_try_to_cache: 1315 * 1316 * Returns 0 on failure, 1 on success 1317 */ 1318int 1319vm_page_try_to_cache(vm_page_t m) 1320{ 1321 1322 mtx_assert(&vm_page_queue_mtx, MA_OWNED); 1323 VM_OBJECT_LOCK_ASSERT(m->object, MA_OWNED); 1324 if (m->dirty || m->hold_count || m->busy || m->wire_count || 1325 (m->oflags & VPO_BUSY) || (m->flags & PG_UNMANAGED)) { 1326 return (0); 1327 } 1328 pmap_remove_all(m); 1329 if (m->dirty) 1330 return (0); 1331 vm_page_cache(m); 1332 return (1); 1333} 1334 1335/* 1336 * vm_page_try_to_free() 1337 * 1338 * Attempt to free the page. If we cannot free it, we do nothing. 1339 * 1 is returned on success, 0 on failure. 1340 */ 1341int 1342vm_page_try_to_free(vm_page_t m) 1343{ 1344 1345 mtx_assert(&vm_page_queue_mtx, MA_OWNED); 1346 if (m->object != NULL) 1347 VM_OBJECT_LOCK_ASSERT(m->object, MA_OWNED); 1348 if (m->dirty || m->hold_count || m->busy || m->wire_count || 1349 (m->oflags & VPO_BUSY) || (m->flags & PG_UNMANAGED)) { 1350 return (0); 1351 } 1352 pmap_remove_all(m); 1353 if (m->dirty) 1354 return (0); 1355 vm_page_free(m); 1356 return (1); 1357} 1358 1359/* 1360 * vm_page_cache 1361 * 1362 * Put the specified page onto the page cache queue (if appropriate). 1363 * 1364 * This routine may not block. 1365 */ 1366void 1367vm_page_cache(vm_page_t m) 1368{ 1369 1370 mtx_assert(&vm_page_queue_mtx, MA_OWNED); 1371 VM_OBJECT_LOCK_ASSERT(m->object, MA_OWNED); 1372 if ((m->flags & PG_UNMANAGED) || (m->oflags & VPO_BUSY) || m->busy || 1373 m->hold_count || m->wire_count) { 1374 printf("vm_page_cache: attempting to cache busy page\n"); 1375 return; 1376 } 1377 if (VM_PAGE_INQUEUE1(m, PQ_CACHE)) 1378 return; 1379 1380 /* 1381 * Remove all pmaps and indicate that the page is not 1382 * writeable or mapped. 1383 */ 1384 pmap_remove_all(m); 1385 if (m->dirty != 0) { 1386 panic("vm_page_cache: caching a dirty page, pindex: %ld", 1387 (long)m->pindex); 1388 } 1389 vm_pageq_remove_nowakeup(m); 1390 vm_pageq_enqueue(PQ_CACHE + m->pc, m); 1391 mtx_lock(&vm_page_queue_free_mtx); 1392 vm_page_free_wakeup(); 1393 mtx_unlock(&vm_page_queue_free_mtx); 1394} 1395 1396/* 1397 * vm_page_dontneed 1398 * 1399 * Cache, deactivate, or do nothing as appropriate. This routine 1400 * is typically used by madvise() MADV_DONTNEED. 1401 * 1402 * Generally speaking we want to move the page into the cache so 1403 * it gets reused quickly. However, this can result in a silly syndrome 1404 * due to the page recycling too quickly. Small objects will not be 1405 * fully cached. On the otherhand, if we move the page to the inactive 1406 * queue we wind up with a problem whereby very large objects 1407 * unnecessarily blow away our inactive and cache queues. 1408 * 1409 * The solution is to move the pages based on a fixed weighting. We 1410 * either leave them alone, deactivate them, or move them to the cache, 1411 * where moving them to the cache has the highest weighting. 1412 * By forcing some pages into other queues we eventually force the 1413 * system to balance the queues, potentially recovering other unrelated 1414 * space from active. The idea is to not force this to happen too 1415 * often. 1416 */ 1417void 1418vm_page_dontneed(vm_page_t m) 1419{ 1420 static int dnweight; 1421 int dnw; 1422 int head; 1423 1424 mtx_assert(&vm_page_queue_mtx, MA_OWNED); 1425 dnw = ++dnweight; 1426 1427 /* 1428 * occassionally leave the page alone 1429 */ 1430 if ((dnw & 0x01F0) == 0 || 1431 VM_PAGE_INQUEUE2(m, PQ_INACTIVE) || 1432 VM_PAGE_INQUEUE1(m, PQ_CACHE) 1433 ) { 1434 if (m->act_count >= ACT_INIT) 1435 --m->act_count; 1436 return; 1437 } 1438 1439 if (m->dirty == 0 && pmap_is_modified(m)) 1440 vm_page_dirty(m); 1441 1442 if (m->dirty || (dnw & 0x0070) == 0) { 1443 /* 1444 * Deactivate the page 3 times out of 32. 1445 */ 1446 head = 0; 1447 } else { 1448 /* 1449 * Cache the page 28 times out of every 32. Note that 1450 * the page is deactivated instead of cached, but placed 1451 * at the head of the queue instead of the tail. 1452 */ 1453 head = 1; 1454 } 1455 _vm_page_deactivate(m, head); 1456} 1457 1458/* 1459 * Grab a page, waiting until we are waken up due to the page 1460 * changing state. We keep on waiting, if the page continues 1461 * to be in the object. If the page doesn't exist, first allocate it 1462 * and then conditionally zero it. 1463 * 1464 * This routine may block. 1465 */ 1466vm_page_t 1467vm_page_grab(vm_object_t object, vm_pindex_t pindex, int allocflags) 1468{ 1469 vm_page_t m; 1470 1471 VM_OBJECT_LOCK_ASSERT(object, MA_OWNED); 1472retrylookup: 1473 if ((m = vm_page_lookup(object, pindex)) != NULL) { 1474 if (vm_page_sleep_if_busy(m, TRUE, "pgrbwt")) { 1475 if ((allocflags & VM_ALLOC_RETRY) == 0) 1476 return (NULL); 1477 goto retrylookup; 1478 } else { 1479 if ((allocflags & VM_ALLOC_WIRED) != 0) { 1480 vm_page_lock_queues(); 1481 vm_page_wire(m); 1482 vm_page_unlock_queues(); 1483 } 1484 if ((allocflags & VM_ALLOC_NOBUSY) == 0) 1485 vm_page_busy(m); 1486 return (m); 1487 } 1488 } 1489 m = vm_page_alloc(object, pindex, allocflags & ~VM_ALLOC_RETRY); 1490 if (m == NULL) { 1491 VM_OBJECT_UNLOCK(object); 1492 VM_WAIT; 1493 VM_OBJECT_LOCK(object); 1494 if ((allocflags & VM_ALLOC_RETRY) == 0) 1495 return (NULL); 1496 goto retrylookup; 1497 } 1498 if (allocflags & VM_ALLOC_ZERO && (m->flags & PG_ZERO) == 0) 1499 pmap_zero_page(m); 1500 return (m); 1501} 1502 1503/* 1504 * Mapping function for valid bits or for dirty bits in 1505 * a page. May not block. 1506 * 1507 * Inputs are required to range within a page. 1508 */ 1509inline int 1510vm_page_bits(int base, int size) 1511{ 1512 int first_bit; 1513 int last_bit; 1514 1515 KASSERT( 1516 base + size <= PAGE_SIZE, 1517 ("vm_page_bits: illegal base/size %d/%d", base, size) 1518 ); 1519 1520 if (size == 0) /* handle degenerate case */ 1521 return (0); 1522 1523 first_bit = base >> DEV_BSHIFT; 1524 last_bit = (base + size - 1) >> DEV_BSHIFT; 1525 1526 return ((2 << last_bit) - (1 << first_bit)); 1527} 1528 1529/* 1530 * vm_page_set_validclean: 1531 * 1532 * Sets portions of a page valid and clean. The arguments are expected 1533 * to be DEV_BSIZE aligned but if they aren't the bitmap is inclusive 1534 * of any partial chunks touched by the range. The invalid portion of 1535 * such chunks will be zero'd. 1536 * 1537 * This routine may not block. 1538 * 1539 * (base + size) must be less then or equal to PAGE_SIZE. 1540 */ 1541void 1542vm_page_set_validclean(vm_page_t m, int base, int size) 1543{ 1544 int pagebits; 1545 int frag; 1546 int endoff; 1547 1548 mtx_assert(&vm_page_queue_mtx, MA_OWNED); 1549 VM_OBJECT_LOCK_ASSERT(m->object, MA_OWNED); 1550 if (size == 0) /* handle degenerate case */ 1551 return; 1552 1553 /* 1554 * If the base is not DEV_BSIZE aligned and the valid 1555 * bit is clear, we have to zero out a portion of the 1556 * first block. 1557 */ 1558 if ((frag = base & ~(DEV_BSIZE - 1)) != base && 1559 (m->valid & (1 << (base >> DEV_BSHIFT))) == 0) 1560 pmap_zero_page_area(m, frag, base - frag); 1561 1562 /* 1563 * If the ending offset is not DEV_BSIZE aligned and the 1564 * valid bit is clear, we have to zero out a portion of 1565 * the last block. 1566 */ 1567 endoff = base + size; 1568 if ((frag = endoff & ~(DEV_BSIZE - 1)) != endoff && 1569 (m->valid & (1 << (endoff >> DEV_BSHIFT))) == 0) 1570 pmap_zero_page_area(m, endoff, 1571 DEV_BSIZE - (endoff & (DEV_BSIZE - 1))); 1572 1573 /* 1574 * Set valid, clear dirty bits. If validating the entire 1575 * page we can safely clear the pmap modify bit. We also 1576 * use this opportunity to clear the VPO_NOSYNC flag. If a process 1577 * takes a write fault on a MAP_NOSYNC memory area the flag will 1578 * be set again. 1579 * 1580 * We set valid bits inclusive of any overlap, but we can only 1581 * clear dirty bits for DEV_BSIZE chunks that are fully within 1582 * the range. 1583 */ 1584 pagebits = vm_page_bits(base, size); 1585 m->valid |= pagebits; 1586#if 0 /* NOT YET */ 1587 if ((frag = base & (DEV_BSIZE - 1)) != 0) { 1588 frag = DEV_BSIZE - frag; 1589 base += frag; 1590 size -= frag; 1591 if (size < 0) 1592 size = 0; 1593 } 1594 pagebits = vm_page_bits(base, size & (DEV_BSIZE - 1)); 1595#endif 1596 m->dirty &= ~pagebits; 1597 if (base == 0 && size == PAGE_SIZE) { 1598 pmap_clear_modify(m); 1599 m->oflags &= ~VPO_NOSYNC; 1600 } 1601} 1602 1603void 1604vm_page_clear_dirty(vm_page_t m, int base, int size) 1605{ 1606 1607 mtx_assert(&vm_page_queue_mtx, MA_OWNED); 1608 m->dirty &= ~vm_page_bits(base, size); 1609} 1610 1611/* 1612 * vm_page_set_invalid: 1613 * 1614 * Invalidates DEV_BSIZE'd chunks within a page. Both the 1615 * valid and dirty bits for the effected areas are cleared. 1616 * 1617 * May not block. 1618 */ 1619void 1620vm_page_set_invalid(vm_page_t m, int base, int size) 1621{ 1622 int bits; 1623 1624 VM_OBJECT_LOCK_ASSERT(m->object, MA_OWNED); 1625 bits = vm_page_bits(base, size); 1626 mtx_assert(&vm_page_queue_mtx, MA_OWNED); 1627 if (m->valid == VM_PAGE_BITS_ALL && bits != 0) 1628 pmap_remove_all(m); 1629 m->valid &= ~bits; 1630 m->dirty &= ~bits; 1631 m->object->generation++; 1632} 1633 1634/* 1635 * vm_page_zero_invalid() 1636 * 1637 * The kernel assumes that the invalid portions of a page contain 1638 * garbage, but such pages can be mapped into memory by user code. 1639 * When this occurs, we must zero out the non-valid portions of the 1640 * page so user code sees what it expects. 1641 * 1642 * Pages are most often semi-valid when the end of a file is mapped 1643 * into memory and the file's size is not page aligned. 1644 */ 1645void 1646vm_page_zero_invalid(vm_page_t m, boolean_t setvalid) 1647{ 1648 int b; 1649 int i; 1650 1651 VM_OBJECT_LOCK_ASSERT(m->object, MA_OWNED); 1652 /* 1653 * Scan the valid bits looking for invalid sections that 1654 * must be zerod. Invalid sub-DEV_BSIZE'd areas ( where the 1655 * valid bit may be set ) have already been zerod by 1656 * vm_page_set_validclean(). 1657 */ 1658 for (b = i = 0; i <= PAGE_SIZE / DEV_BSIZE; ++i) { 1659 if (i == (PAGE_SIZE / DEV_BSIZE) || 1660 (m->valid & (1 << i)) 1661 ) { 1662 if (i > b) { 1663 pmap_zero_page_area(m, 1664 b << DEV_BSHIFT, (i - b) << DEV_BSHIFT); 1665 } 1666 b = i + 1; 1667 } 1668 } 1669 1670 /* 1671 * setvalid is TRUE when we can safely set the zero'd areas 1672 * as being valid. We can do this if there are no cache consistancy 1673 * issues. e.g. it is ok to do with UFS, but not ok to do with NFS. 1674 */ 1675 if (setvalid) 1676 m->valid = VM_PAGE_BITS_ALL; 1677} 1678 1679/* 1680 * vm_page_is_valid: 1681 * 1682 * Is (partial) page valid? Note that the case where size == 0 1683 * will return FALSE in the degenerate case where the page is 1684 * entirely invalid, and TRUE otherwise. 1685 * 1686 * May not block. 1687 */ 1688int 1689vm_page_is_valid(vm_page_t m, int base, int size) 1690{ 1691 int bits = vm_page_bits(base, size); 1692 1693 VM_OBJECT_LOCK_ASSERT(m->object, MA_OWNED); 1694 if (m->valid && ((m->valid & bits) == bits)) 1695 return 1; 1696 else 1697 return 0; 1698} 1699 1700/* 1701 * update dirty bits from pmap/mmu. May not block. 1702 */ 1703void 1704vm_page_test_dirty(vm_page_t m) 1705{ 1706 if ((m->dirty != VM_PAGE_BITS_ALL) && pmap_is_modified(m)) { 1707 vm_page_dirty(m); 1708 } 1709} 1710 1711int so_zerocp_fullpage = 0; 1712 1713void 1714vm_page_cowfault(vm_page_t m) 1715{ 1716 vm_page_t mnew; 1717 vm_object_t object; 1718 vm_pindex_t pindex; 1719 1720 object = m->object; 1721 pindex = m->pindex; 1722 1723 retry_alloc: 1724 pmap_remove_all(m); 1725 vm_page_remove(m); 1726 mnew = vm_page_alloc(object, pindex, VM_ALLOC_NORMAL | VM_ALLOC_NOBUSY); 1727 if (mnew == NULL) { 1728 vm_page_insert(m, object, pindex); 1729 vm_page_unlock_queues(); 1730 VM_OBJECT_UNLOCK(object); 1731 VM_WAIT; 1732 VM_OBJECT_LOCK(object); 1733 vm_page_lock_queues(); 1734 goto retry_alloc; 1735 } 1736 1737 if (m->cow == 0) { 1738 /* 1739 * check to see if we raced with an xmit complete when 1740 * waiting to allocate a page. If so, put things back 1741 * the way they were 1742 */ 1743 vm_page_free(mnew); 1744 vm_page_insert(m, object, pindex); 1745 } else { /* clear COW & copy page */ 1746 if (!so_zerocp_fullpage) 1747 pmap_copy_page(m, mnew); 1748 mnew->valid = VM_PAGE_BITS_ALL; 1749 vm_page_dirty(mnew); 1750 mnew->wire_count = m->wire_count - m->cow; 1751 m->wire_count = m->cow; 1752 } 1753} 1754 1755void 1756vm_page_cowclear(vm_page_t m) 1757{ 1758 1759 mtx_assert(&vm_page_queue_mtx, MA_OWNED); 1760 if (m->cow) { 1761 m->cow--; 1762 /* 1763 * let vm_fault add back write permission lazily 1764 */ 1765 } 1766 /* 1767 * sf_buf_free() will free the page, so we needn't do it here 1768 */ 1769} 1770 1771void 1772vm_page_cowsetup(vm_page_t m) 1773{ 1774 1775 mtx_assert(&vm_page_queue_mtx, MA_OWNED); 1776 m->cow++; 1777 pmap_remove_write(m); 1778} 1779 1780#include "opt_ddb.h" 1781#ifdef DDB 1782#include <sys/kernel.h> 1783 1784#include <ddb/ddb.h> 1785 1786DB_SHOW_COMMAND(page, vm_page_print_page_info) 1787{ 1788 db_printf("cnt.v_free_count: %d\n", cnt.v_free_count); 1789 db_printf("cnt.v_cache_count: %d\n", cnt.v_cache_count); 1790 db_printf("cnt.v_inactive_count: %d\n", cnt.v_inactive_count); 1791 db_printf("cnt.v_active_count: %d\n", cnt.v_active_count); 1792 db_printf("cnt.v_wire_count: %d\n", cnt.v_wire_count); 1793 db_printf("cnt.v_free_reserved: %d\n", cnt.v_free_reserved); 1794 db_printf("cnt.v_free_min: %d\n", cnt.v_free_min); 1795 db_printf("cnt.v_free_target: %d\n", cnt.v_free_target); 1796 db_printf("cnt.v_cache_min: %d\n", cnt.v_cache_min); 1797 db_printf("cnt.v_inactive_target: %d\n", cnt.v_inactive_target); 1798} 1799 1800DB_SHOW_COMMAND(pageq, vm_page_print_pageq_info) 1801{ 1802 int i; 1803 db_printf("PQ_FREE:"); 1804 for (i = 0; i < PQ_NUMCOLORS; i++) { 1805 db_printf(" %d", vm_page_queues[PQ_FREE + i].lcnt); 1806 } 1807 db_printf("\n"); 1808 1809 db_printf("PQ_CACHE:"); 1810 for (i = 0; i < PQ_NUMCOLORS; i++) { 1811 db_printf(" %d", vm_page_queues[PQ_CACHE + i].lcnt); 1812 } 1813 db_printf("\n"); 1814 1815 db_printf("PQ_ACTIVE: %d, PQ_INACTIVE: %d\n", 1816 vm_page_queues[PQ_ACTIVE].lcnt, 1817 vm_page_queues[PQ_INACTIVE].lcnt); 1818} 1819#endif /* DDB */ 1820