vm_glue.c revision 193593
133965Sjdp/*- 260484Sobrien * Copyright (c) 1991, 1993 360484Sobrien * The Regents of the University of California. All rights reserved. 433965Sjdp * 533965Sjdp * This code is derived from software contributed to Berkeley by 633965Sjdp * The Mach Operating System project at Carnegie-Mellon University. 733965Sjdp * 833965Sjdp * Redistribution and use in source and binary forms, with or without 933965Sjdp * modification, are permitted provided that the following conditions 1033965Sjdp * are met: 1133965Sjdp * 1. Redistributions of source code must retain the above copyright 1233965Sjdp * notice, this list of conditions and the following disclaimer. 1333965Sjdp * 2. Redistributions in binary form must reproduce the above copyright 1433965Sjdp * notice, this list of conditions and the following disclaimer in the 1533965Sjdp * documentation and/or other materials provided with the distribution. 1633965Sjdp * 4. Neither the name of the University nor the names of its contributors 1733965Sjdp * may be used to endorse or promote products derived from this software 1833965Sjdp * without specific prior written permission. 1933965Sjdp * 2033965Sjdp * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 2133965Sjdp * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 2233965Sjdp * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 2333965Sjdp * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 2433965Sjdp * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 2533965Sjdp * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 2633965Sjdp * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 2733965Sjdp * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 2833965Sjdp * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 2933965Sjdp * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 3033965Sjdp * SUCH DAMAGE. 3133965Sjdp * 3233965Sjdp * from: @(#)vm_glue.c 8.6 (Berkeley) 1/5/94 3333965Sjdp * 3477298Sobrien * 3533965Sjdp * Copyright (c) 1987, 1990 Carnegie-Mellon University. 3633965Sjdp * All rights reserved. 3733965Sjdp * 3833965Sjdp * Permission to use, copy, modify and distribute this software and 3933965Sjdp * its documentation is hereby granted, provided that both the copyright 4033965Sjdp * notice and this permission notice appear in all copies of the 4133965Sjdp * software, derivative works or modified versions, and any portions 4233965Sjdp * thereof, and that both notices appear in supporting documentation. 4333965Sjdp * 4477298Sobrien * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS" 4533965Sjdp * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND 4633965Sjdp * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE. 4733965Sjdp * 4833965Sjdp * Carnegie Mellon requests users of this software to return to 4933965Sjdp * 5077298Sobrien * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU 5133965Sjdp * School of Computer Science 5233965Sjdp * Carnegie Mellon University 5333965Sjdp * Pittsburgh PA 15213-3890 5433965Sjdp * 5533965Sjdp * any improvements or extensions that they make and grant Carnegie the 5633965Sjdp * rights to redistribute these changes. 5733965Sjdp */ 5833965Sjdp 5933965Sjdp#include <sys/cdefs.h> 6033965Sjdp__FBSDID("$FreeBSD: head/sys/vm/vm_glue.c 193593 2009-06-06 20:13:14Z alc $"); 6133965Sjdp 6233965Sjdp#include "opt_vm.h" 6333965Sjdp#include "opt_kstack_pages.h" 6433965Sjdp#include "opt_kstack_max_pages.h" 6533965Sjdp 6633965Sjdp#include <sys/param.h> 6777298Sobrien#include <sys/systm.h> 6833965Sjdp#include <sys/limits.h> 6977298Sobrien#include <sys/lock.h> 7033965Sjdp#include <sys/mutex.h> 7133965Sjdp#include <sys/proc.h> 7233965Sjdp#include <sys/resourcevar.h> 7333965Sjdp#include <sys/sched.h> 7438889Sjdp#include <sys/sf_buf.h> 7538889Sjdp#include <sys/shm.h> 7638889Sjdp#include <sys/vmmeter.h> 7738889Sjdp#include <sys/sx.h> 7838889Sjdp#include <sys/sysctl.h> 7938889Sjdp 8038889Sjdp#include <sys/kernel.h> 8160484Sobrien#include <sys/ktr.h> 8277298Sobrien#include <sys/unistd.h> 8333965Sjdp 8433965Sjdp#include <vm/vm.h> 8577298Sobrien#include <vm/vm_param.h> 8633965Sjdp#include <vm/pmap.h> 8733965Sjdp#include <vm/vm_map.h> 8833965Sjdp#include <vm/vm_page.h> 8977298Sobrien#include <vm/vm_pageout.h> 9033965Sjdp#include <vm/vm_object.h> 9133965Sjdp#include <vm/vm_kern.h> 9233965Sjdp#include <vm/vm_extern.h> 9333965Sjdp#include <vm/vm_pager.h> 9433965Sjdp#include <vm/swap_pager.h> 9533965Sjdp 9633965Sjdpextern int maxslp; 9733965Sjdp 9833965Sjdp/* 9933965Sjdp * System initialization 10033965Sjdp * 10133965Sjdp * Note: proc0 from proc.h 10233965Sjdp */ 10333965Sjdpstatic void vm_init_limits(void *); 10433965SjdpSYSINIT(vm_limits, SI_SUB_VM_CONF, SI_ORDER_FIRST, vm_init_limits, &proc0); 10533965Sjdp 10633965Sjdp/* 10733965Sjdp * THIS MUST BE THE LAST INITIALIZATION ITEM!!! 10833965Sjdp * 10960484Sobrien * Note: run scheduling should be divorced from the vm system. 11060484Sobrien */ 11160484Sobrienstatic void scheduler(void *); 11277298SobrienSYSINIT(scheduler, SI_SUB_RUN_SCHEDULER, SI_ORDER_ANY, scheduler, NULL); 11377298Sobrien 11433965Sjdp#ifndef NO_SWAPPING 11533965Sjdpstatic int swapout(struct proc *); 11677298Sobrienstatic void swapclear(struct proc *); 11777298Sobrien#endif 11833965Sjdp 11938889Sjdp/* 12038889Sjdp * MPSAFE 12138889Sjdp * 12238889Sjdp * WARNING! This code calls vm_map_check_protection() which only checks 12338889Sjdp * the associated vm_map_entry range. It does not determine whether the 12460484Sobrien * contents of the memory is actually readable or writable. In most cases 12538889Sjdp * just checking the vm_map_entry is sufficient within the kernel's address 12638889Sjdp * space. 12738889Sjdp */ 12838889Sjdpint 12938889Sjdpkernacc(addr, len, rw) 13038889Sjdp void *addr; 13138889Sjdp int len, rw; 13238889Sjdp{ 13338889Sjdp boolean_t rv; 13477298Sobrien vm_offset_t saddr, eaddr; 13577298Sobrien vm_prot_t prot; 13677298Sobrien 13777298Sobrien KASSERT((rw & ~VM_PROT_ALL) == 0, 13877298Sobrien ("illegal ``rw'' argument to kernacc (%x)\n", rw)); 13933965Sjdp 14033965Sjdp if ((vm_offset_t)addr + len > kernel_map->max_offset || 14133965Sjdp (vm_offset_t)addr + len < (vm_offset_t)addr) 14260484Sobrien return (FALSE); 14377298Sobrien 14477298Sobrien prot = rw; 14533965Sjdp saddr = trunc_page((vm_offset_t)addr); 14633965Sjdp eaddr = round_page((vm_offset_t)addr + len); 14733965Sjdp vm_map_lock_read(kernel_map); 14860484Sobrien rv = vm_map_check_protection(kernel_map, saddr, eaddr, prot); 14933965Sjdp vm_map_unlock_read(kernel_map); 15033965Sjdp return (rv == TRUE); 15133965Sjdp} 15233965Sjdp 15333965Sjdp/* 15433965Sjdp * MPSAFE 15533965Sjdp * 15633965Sjdp * WARNING! This code calls vm_map_check_protection() which only checks 15733965Sjdp * the associated vm_map_entry range. It does not determine whether the 15833965Sjdp * contents of the memory is actually readable or writable. vmapbuf(), 15977298Sobrien * vm_fault_quick(), or copyin()/copout()/su*()/fu*() functions should be 16077298Sobrien * used in conjuction with this call. 16177298Sobrien */ 16277298Sobrienint 16377298Sobrienuseracc(addr, len, rw) 16477298Sobrien void *addr; 16577298Sobrien int len, rw; 16677298Sobrien{ 16777298Sobrien boolean_t rv; 16877298Sobrien vm_prot_t prot; 16977298Sobrien vm_map_t map; 17077298Sobrien 17177298Sobrien KASSERT((rw & ~VM_PROT_ALL) == 0, 17277298Sobrien ("illegal ``rw'' argument to useracc (%x)\n", rw)); 17377298Sobrien prot = rw; 17477298Sobrien map = &curproc->p_vmspace->vm_map; 17577298Sobrien if ((vm_offset_t)addr + len > vm_map_max(map) || 17633965Sjdp (vm_offset_t)addr + len < (vm_offset_t)addr) { 17777298Sobrien return (FALSE); 17877298Sobrien } 17977298Sobrien vm_map_lock_read(map); 18077298Sobrien rv = vm_map_check_protection(map, trunc_page((vm_offset_t)addr), 18133965Sjdp round_page((vm_offset_t)addr + len), prot); 18233965Sjdp vm_map_unlock_read(map); 18360484Sobrien return (rv == TRUE); 18460484Sobrien} 18560484Sobrien 18660484Sobrienint 18777298Sobrienvslock(void *addr, size_t len) 18877298Sobrien{ 18933965Sjdp vm_offset_t end, last, start; 19033965Sjdp vm_size_t npages; 19133965Sjdp int error; 19233965Sjdp 19333965Sjdp last = (vm_offset_t)addr + len; 19438889Sjdp start = trunc_page((vm_offset_t)addr); 19560484Sobrien end = round_page(last); 19660484Sobrien if (last < (vm_offset_t)addr || end < (vm_offset_t)addr) 19738889Sjdp return (EINVAL); 19860484Sobrien npages = atop(end - start); 19938889Sjdp if (npages > vm_page_max_wired) 20038889Sjdp return (ENOMEM); 20133965Sjdp PROC_LOCK(curproc); 20260484Sobrien if (ptoa(npages + 20360484Sobrien pmap_wired_count(vm_map_pmap(&curproc->p_vmspace->vm_map))) > 20460484Sobrien lim_cur(curproc, RLIMIT_MEMLOCK)) { 20533965Sjdp PROC_UNLOCK(curproc); 20638889Sjdp return (ENOMEM); 20760484Sobrien } 20838889Sjdp PROC_UNLOCK(curproc); 20938889Sjdp#if 0 21060484Sobrien /* 21138889Sjdp * XXX - not yet 21260484Sobrien * 21360484Sobrien * The limit for transient usage of wired pages should be 21477298Sobrien * larger than for "permanent" wired pages (mlock()). 21577298Sobrien * 21633965Sjdp * Also, the sysctl code, which is the only present user 21733965Sjdp * of vslock(), does a hard loop on EAGAIN. 21838889Sjdp */ 21977298Sobrien if (npages + cnt.v_wire_count > vm_page_max_wired) 22060484Sobrien return (EAGAIN); 22138889Sjdp#endif 22238889Sjdp error = vm_map_wire(&curproc->p_vmspace->vm_map, start, end, 22360484Sobrien VM_MAP_WIRE_SYSTEM | VM_MAP_WIRE_NOHOLES); 22460484Sobrien /* 22577298Sobrien * Return EFAULT on error to match copy{in,out}() behaviour 22677298Sobrien * rather than returning ENOMEM like mlock() would. 22777298Sobrien */ 22877298Sobrien return (error == KERN_SUCCESS ? 0 : EFAULT); 22977298Sobrien} 23033965Sjdp 23138889Sjdpvoid 23260484Sobrienvsunlock(void *addr, size_t len) 23333965Sjdp{ 23433965Sjdp 23560484Sobrien /* Rely on the parameter sanity checks performed by vslock(). */ 23660484Sobrien (void)vm_map_unwire(&curproc->p_vmspace->vm_map, 23760484Sobrien trunc_page((vm_offset_t)addr), round_page((vm_offset_t)addr + len), 23860484Sobrien VM_MAP_WIRE_SYSTEM | VM_MAP_WIRE_NOHOLES); 23960484Sobrien} 24077298Sobrien 24177298Sobrien/* 24277298Sobrien * Pin the page contained within the given object at the given offset. If the 24360484Sobrien * page is not resident, allocate and load it using the given object's pager. 24460484Sobrien * Return the pinned page if successful; otherwise, return NULL. 24560484Sobrien */ 24660484Sobrienstatic vm_page_t 24760484Sobrienvm_imgact_hold_page(vm_object_t object, vm_ooffset_t offset) 24860484Sobrien{ 24977298Sobrien vm_page_t m, ma[1]; 25077298Sobrien vm_pindex_t pindex; 25133965Sjdp int rv; 25233965Sjdp 25333965Sjdp VM_OBJECT_LOCK(object); 25433965Sjdp pindex = OFF_TO_IDX(offset); 25533965Sjdp m = vm_page_grab(object, pindex, VM_ALLOC_NORMAL | VM_ALLOC_RETRY); 25633965Sjdp if ((m->valid & VM_PAGE_BITS_ALL) != VM_PAGE_BITS_ALL) { 25733965Sjdp ma[0] = m; 25833965Sjdp rv = vm_pager_get_pages(object, ma, 1, 0); 25933965Sjdp m = vm_page_lookup(object, pindex); 26033965Sjdp if (m == NULL) 26133965Sjdp goto out; 26233965Sjdp if (rv != VM_PAGER_OK) { 26333965Sjdp vm_page_lock_queues(); 26477298Sobrien vm_page_free(m); 26533965Sjdp vm_page_unlock_queues(); 26633965Sjdp m = NULL; 26733965Sjdp goto out; 26833965Sjdp } 26933965Sjdp } 27033965Sjdp vm_page_lock_queues(); 27133965Sjdp vm_page_hold(m); 27233965Sjdp vm_page_unlock_queues(); 27333965Sjdp vm_page_wakeup(m); 27477298Sobrienout: 27577298Sobrien VM_OBJECT_UNLOCK(object); 27633965Sjdp return (m); 27733965Sjdp} 27833965Sjdp 27933965Sjdp/* 28033965Sjdp * Return a CPU private mapping to the page at the given offset within the 28133965Sjdp * given object. The page is pinned before it is mapped. 28233965Sjdp */ 28333965Sjdpstruct sf_buf * 28433965Sjdpvm_imgact_map_page(vm_object_t object, vm_ooffset_t offset) 28533965Sjdp{ 28633965Sjdp vm_page_t m; 28733965Sjdp 28838889Sjdp m = vm_imgact_hold_page(object, offset); 28933965Sjdp if (m == NULL) 29077298Sobrien return (NULL); 29133965Sjdp sched_pin(); 29260484Sobrien return (sf_buf_alloc(m, SFB_CPUPRIVATE)); 29333965Sjdp} 29433965Sjdp 29533965Sjdp/* 29660484Sobrien * Destroy the given CPU private mapping and unpin the page that it mapped. 29733965Sjdp */ 29833965Sjdpvoid 29933965Sjdpvm_imgact_unmap_page(struct sf_buf *sf) 30033965Sjdp{ 30177298Sobrien vm_page_t m; 30277298Sobrien 30333965Sjdp m = sf_buf_page(sf); 30433965Sjdp sf_buf_free(sf); 30533965Sjdp sched_unpin(); 30633965Sjdp vm_page_lock_queues(); 30733965Sjdp vm_page_unhold(m); 30833965Sjdp vm_page_unlock_queues(); 30933965Sjdp} 31060484Sobrien 31133965Sjdp#ifndef KSTACK_MAX_PAGES 31233965Sjdp#define KSTACK_MAX_PAGES 32 31338889Sjdp#endif 31477298Sobrien 31560484Sobrien/* 31633965Sjdp * Create the kernel stack (including pcb for i386) for a new thread. 31733965Sjdp * This routine directly affects the fork perf for a process and 31833965Sjdp * create performance for a thread. 31933965Sjdp */ 32033965Sjdpint 32138889Sjdpvm_thread_new(struct thread *td, int pages) 32260484Sobrien{ 32360484Sobrien vm_object_t ksobj; 32460484Sobrien vm_offset_t ks; 32577298Sobrien vm_page_t m, ma[KSTACK_MAX_PAGES]; 32633965Sjdp int i; 32777298Sobrien 32833965Sjdp /* Bounds check */ 32933965Sjdp if (pages <= 1) 33033965Sjdp pages = KSTACK_PAGES; 33133965Sjdp else if (pages > KSTACK_MAX_PAGES) 33233965Sjdp pages = KSTACK_MAX_PAGES; 33338889Sjdp /* 33433965Sjdp * Allocate an object for the kstack. 33577298Sobrien */ 33633965Sjdp ksobj = vm_object_allocate(OBJT_DEFAULT, pages); 33760484Sobrien 33833965Sjdp /* 33933965Sjdp * Get a kernel virtual address for this thread's kstack. 34033965Sjdp */ 34160484Sobrien ks = kmem_alloc_nofault(kernel_map, 34233965Sjdp (pages + KSTACK_GUARD_PAGES) * PAGE_SIZE); 34333965Sjdp if (ks == 0) { 34433965Sjdp printf("vm_thread_new: kstack allocation failed\n"); 34533965Sjdp vm_object_deallocate(ksobj); 34677298Sobrien return (0); 34777298Sobrien } 34833965Sjdp 34933965Sjdp if (KSTACK_GUARD_PAGES != 0) { 35033965Sjdp pmap_qremove(ks, KSTACK_GUARD_PAGES); 35133965Sjdp ks += KSTACK_GUARD_PAGES * PAGE_SIZE; 35233965Sjdp } 35333965Sjdp td->td_kstack_obj = ksobj; 35433965Sjdp td->td_kstack = ks; 35533965Sjdp /* 35633965Sjdp * Knowing the number of pages allocated is useful when you 35738889Sjdp * want to deallocate them. 35877298Sobrien */ 35960484Sobrien td->td_kstack_pages = pages; 36033965Sjdp /* 36133965Sjdp * For the length of the stack, link in a real page of ram for each 36233965Sjdp * page of stack. 36333965Sjdp */ 36433965Sjdp VM_OBJECT_LOCK(ksobj); 36538889Sjdp for (i = 0; i < pages; i++) { 36660484Sobrien /* 36760484Sobrien * Get a kernel stack page. 36860484Sobrien */ 36977298Sobrien m = vm_page_grab(ksobj, i, VM_ALLOC_NOBUSY | 37033965Sjdp VM_ALLOC_NORMAL | VM_ALLOC_RETRY | VM_ALLOC_WIRED); 37133965Sjdp ma[i] = m; 37233965Sjdp m->valid = VM_PAGE_BITS_ALL; 37333965Sjdp } 37433965Sjdp VM_OBJECT_UNLOCK(ksobj); 37533965Sjdp pmap_qenter(ks, ma, pages); 37633965Sjdp return (1); 37733965Sjdp} 37833965Sjdp 37933965Sjdp/* 38033965Sjdp * Dispose of a thread's kernel stack. 38133965Sjdp */ 38233965Sjdpvoid 38333965Sjdpvm_thread_dispose(struct thread *td) 38433965Sjdp{ 38533965Sjdp vm_object_t ksobj; 38633965Sjdp vm_offset_t ks; 38733965Sjdp vm_page_t m; 38833965Sjdp int i, pages; 38933965Sjdp 39033965Sjdp pages = td->td_kstack_pages; 39133965Sjdp ksobj = td->td_kstack_obj; 39233965Sjdp ks = td->td_kstack; 39333965Sjdp pmap_qremove(ks, pages); 39433965Sjdp VM_OBJECT_LOCK(ksobj); 39533965Sjdp for (i = 0; i < pages; i++) { 39633965Sjdp m = vm_page_lookup(ksobj, i); 39733965Sjdp if (m == NULL) 39833965Sjdp panic("vm_thread_dispose: kstack already missing?"); 39933965Sjdp vm_page_lock_queues(); 40033965Sjdp vm_page_unwire(m, 0); 40133965Sjdp vm_page_free(m); 40233965Sjdp vm_page_unlock_queues(); 40333965Sjdp } 40433965Sjdp VM_OBJECT_UNLOCK(ksobj); 40533965Sjdp vm_object_deallocate(ksobj); 40633965Sjdp kmem_free(kernel_map, ks - (KSTACK_GUARD_PAGES * PAGE_SIZE), 40733965Sjdp (pages + KSTACK_GUARD_PAGES) * PAGE_SIZE); 40833965Sjdp td->td_kstack = 0; 40933965Sjdp} 41033965Sjdp 41133965Sjdp/* 41233965Sjdp * Allow a thread's kernel stack to be paged out. 41377298Sobrien */ 41433965Sjdpvoid 41533965Sjdpvm_thread_swapout(struct thread *td) 41633965Sjdp{ 41733965Sjdp vm_object_t ksobj; 41833965Sjdp vm_page_t m; 41933965Sjdp int i, pages; 42033965Sjdp 42133965Sjdp cpu_thread_swapout(td); 42233965Sjdp pages = td->td_kstack_pages; 42333965Sjdp ksobj = td->td_kstack_obj; 42433965Sjdp pmap_qremove(td->td_kstack, pages); 42533965Sjdp VM_OBJECT_LOCK(ksobj); 42633965Sjdp for (i = 0; i < pages; i++) { 42733965Sjdp m = vm_page_lookup(ksobj, i); 42838889Sjdp if (m == NULL) 42938889Sjdp panic("vm_thread_swapout: kstack already missing?"); 43038889Sjdp vm_page_lock_queues(); 43138889Sjdp vm_page_dirty(m); 43238889Sjdp vm_page_unwire(m, 0); 43338889Sjdp vm_page_unlock_queues(); 43438889Sjdp } 43538889Sjdp VM_OBJECT_UNLOCK(ksobj); 43638889Sjdp} 43738889Sjdp 43838889Sjdp/* 43938889Sjdp * Bring the kernel stack for a specified thread back in. 44038889Sjdp */ 44138889Sjdpvoid 44238889Sjdpvm_thread_swapin(struct thread *td) 44338889Sjdp{ 44438889Sjdp vm_object_t ksobj; 44538889Sjdp vm_page_t m, ma[KSTACK_MAX_PAGES]; 44677298Sobrien int i, pages, rv; 44738889Sjdp 44838889Sjdp pages = td->td_kstack_pages; 44938889Sjdp ksobj = td->td_kstack_obj; 45038889Sjdp VM_OBJECT_LOCK(ksobj); 45138889Sjdp for (i = 0; i < pages; i++) { 45238889Sjdp m = vm_page_grab(ksobj, i, VM_ALLOC_NORMAL | VM_ALLOC_RETRY); 45338889Sjdp if (m->valid != VM_PAGE_BITS_ALL) { 45438889Sjdp rv = vm_pager_get_pages(ksobj, &m, 1, 0); 45538889Sjdp if (rv != VM_PAGER_OK) 45638889Sjdp panic("vm_thread_swapin: cannot get kstack for proc: %d", td->td_proc->p_pid); 45777298Sobrien m = vm_page_lookup(ksobj, i); 45838889Sjdp } 45938889Sjdp ma[i] = m; 46038889Sjdp vm_page_lock_queues(); 46138889Sjdp vm_page_wire(m); 46277298Sobrien vm_page_unlock_queues(); 46338889Sjdp vm_page_wakeup(m); 46438889Sjdp } 46538889Sjdp VM_OBJECT_UNLOCK(ksobj); 46638889Sjdp pmap_qenter(td->td_kstack, ma, pages); 46738889Sjdp cpu_thread_swapin(td); 46838889Sjdp} 46938889Sjdp 47038889Sjdp/* 47138889Sjdp * Set up a variable-sized alternate kstack. 47238889Sjdp */ 47338889Sjdpint 47438889Sjdpvm_thread_new_altkstack(struct thread *td, int pages) 47538889Sjdp{ 47638889Sjdp 47738889Sjdp td->td_altkstack = td->td_kstack; 47838889Sjdp td->td_altkstack_obj = td->td_kstack_obj; 47938889Sjdp td->td_altkstack_pages = td->td_kstack_pages; 48033965Sjdp 48133965Sjdp return (vm_thread_new(td, pages)); 48233965Sjdp} 48333965Sjdp 48433965Sjdp/* 48533965Sjdp * Restore the original kstack. 48633965Sjdp */ 48733965Sjdpvoid 48833965Sjdpvm_thread_dispose_altkstack(struct thread *td) 48933965Sjdp{ 49033965Sjdp 49133965Sjdp vm_thread_dispose(td); 49233965Sjdp 49333965Sjdp td->td_kstack = td->td_altkstack; 49433965Sjdp td->td_kstack_obj = td->td_altkstack_obj; 49533965Sjdp td->td_kstack_pages = td->td_altkstack_pages; 49633965Sjdp td->td_altkstack = 0; 49733965Sjdp td->td_altkstack_obj = NULL; 49833965Sjdp td->td_altkstack_pages = 0; 49933965Sjdp} 50033965Sjdp 50133965Sjdp/* 50233965Sjdp * Implement fork's actions on an address space. 50377298Sobrien * Here we arrange for the address space to be copied or referenced, 50433965Sjdp * allocate a user struct (pcb and kernel stack), then call the 50577298Sobrien * machine-dependent layer to fill those in and make the new process 50633965Sjdp * ready to run. The new process is set up so that it returns directly 50733965Sjdp * to user mode to avoid stack copying and relocation problems. 50833965Sjdp */ 50933965Sjdpint 51033965Sjdpvm_forkproc(td, p2, td2, vm2, flags) 51133965Sjdp struct thread *td; 51233965Sjdp struct proc *p2; 51333965Sjdp struct thread *td2; 51433965Sjdp struct vmspace *vm2; 51533965Sjdp int flags; 51633965Sjdp{ 51733965Sjdp struct proc *p1 = td->td_proc; 51833965Sjdp int error; 51933965Sjdp 52033965Sjdp if ((flags & RFPROC) == 0) { 52133965Sjdp /* 52233965Sjdp * Divorce the memory, if it is shared, essentially 52333965Sjdp * this changes shared memory amongst threads, into 52433965Sjdp * COW locally. 52577298Sobrien */ 52677298Sobrien if ((flags & RFMEM) == 0) { 52777298Sobrien if (p1->p_vmspace->vm_refcnt > 1) { 52877298Sobrien error = vmspace_unshare(p1); 52977298Sobrien if (error) 53033965Sjdp return (error); 53133965Sjdp } 53233965Sjdp } 53333965Sjdp cpu_fork(td, p2, td2, flags); 53433965Sjdp return (0); 53533965Sjdp } 53633965Sjdp 53733965Sjdp if (flags & RFMEM) { 53833965Sjdp p2->p_vmspace = p1->p_vmspace; 53933965Sjdp atomic_add_int(&p1->p_vmspace->vm_refcnt, 1); 54033965Sjdp } 54133965Sjdp 54233965Sjdp while (vm_page_count_severe()) { 54333965Sjdp VM_WAIT; 54433965Sjdp } 54533965Sjdp 54633965Sjdp if ((flags & RFMEM) == 0) { 54733965Sjdp p2->p_vmspace = vm2; 54833965Sjdp if (p1->p_vmspace->vm_shm) 54933965Sjdp shmfork(p1, p2); 55033965Sjdp } 55133965Sjdp 55233965Sjdp /* 55333965Sjdp * cpu_fork will copy and update the pcb, set up the kernel stack, 55433965Sjdp * and make the child ready to run. 55533965Sjdp */ 55633965Sjdp cpu_fork(td, p2, td2, flags); 55733965Sjdp return (0); 55833965Sjdp} 55933965Sjdp 56033965Sjdp/* 56133965Sjdp * Called after process has been wait(2)'ed apon and is being reaped. 56233965Sjdp * The idea is to reclaim resources that we could not reclaim while 56333965Sjdp * the process was still executing. 56477298Sobrien */ 56533965Sjdpvoid 56633965Sjdpvm_waitproc(p) 56733965Sjdp struct proc *p; 56833965Sjdp{ 56933965Sjdp 57033965Sjdp vmspace_exitfree(p); /* and clean-out the vmspace */ 57133965Sjdp} 57233965Sjdp 57333965Sjdp/* 57433965Sjdp * Set default limits for VM system. 57533965Sjdp * Called for proc 0, and then inherited by all others. 57633965Sjdp * 57733965Sjdp * XXX should probably act directly on proc0. 57833965Sjdp */ 57933965Sjdpstatic void 58033965Sjdpvm_init_limits(udata) 58133965Sjdp void *udata; 58233965Sjdp{ 58333965Sjdp struct proc *p = udata; 58433965Sjdp struct plimit *limp; 58533965Sjdp int rss_limit; 58633965Sjdp 58733965Sjdp /* 58833965Sjdp * Set up the initial limits on process VM. Set the maximum resident 58933965Sjdp * set size to be half of (reasonably) available memory. Since this 59033965Sjdp * is a soft limit, it comes into effect only when the system is out 59133965Sjdp * of memory - half of main memory helps to favor smaller processes, 59233965Sjdp * and reduces thrashing of the object cache. 59333965Sjdp */ 59433965Sjdp limp = p->p_limit; 59533965Sjdp limp->pl_rlimit[RLIMIT_STACK].rlim_cur = dflssiz; 59633965Sjdp limp->pl_rlimit[RLIMIT_STACK].rlim_max = maxssiz; 59733965Sjdp limp->pl_rlimit[RLIMIT_DATA].rlim_cur = dfldsiz; 59833965Sjdp limp->pl_rlimit[RLIMIT_DATA].rlim_max = maxdsiz; 59933965Sjdp /* limit the limit to no less than 2MB */ 60033965Sjdp rss_limit = max(cnt.v_free_count, 512); 60133965Sjdp limp->pl_rlimit[RLIMIT_RSS].rlim_cur = ptoa(rss_limit); 60233965Sjdp limp->pl_rlimit[RLIMIT_RSS].rlim_max = RLIM_INFINITY; 60333965Sjdp} 60433965Sjdp 60533965Sjdpvoid 60633965Sjdpfaultin(p) 60733965Sjdp struct proc *p; 60833965Sjdp{ 60933965Sjdp#ifdef NO_SWAPPING 61077298Sobrien 61133965Sjdp PROC_LOCK_ASSERT(p, MA_OWNED); 61233965Sjdp if ((p->p_flag & P_INMEM) == 0) 61333965Sjdp panic("faultin: proc swapped out with NO_SWAPPING!"); 61433965Sjdp#else /* !NO_SWAPPING */ 61533965Sjdp struct thread *td; 61633965Sjdp 61733965Sjdp PROC_LOCK_ASSERT(p, MA_OWNED); 61833965Sjdp /* 61933965Sjdp * If another process is swapping in this process, 62033965Sjdp * just wait until it finishes. 62133965Sjdp */ 62233965Sjdp if (p->p_flag & P_SWAPPINGIN) { 62333965Sjdp while (p->p_flag & P_SWAPPINGIN) 62433965Sjdp msleep(&p->p_flag, &p->p_mtx, PVM, "faultin", 0); 62577298Sobrien return; 62633965Sjdp } 62733965Sjdp if ((p->p_flag & P_INMEM) == 0) { 62833965Sjdp /* 62977298Sobrien * Don't let another thread swap process p out while we are 63033965Sjdp * busy swapping it in. 63133965Sjdp */ 63233965Sjdp ++p->p_lock; 63333965Sjdp p->p_flag |= P_SWAPPINGIN; 63433965Sjdp PROC_UNLOCK(p); 63533965Sjdp 63633965Sjdp /* 63733965Sjdp * We hold no lock here because the list of threads 63833965Sjdp * can not change while all threads in the process are 63933965Sjdp * swapped out. 64033965Sjdp */ 64133965Sjdp FOREACH_THREAD_IN_PROC(p, td) 64233965Sjdp vm_thread_swapin(td); 64333965Sjdp PROC_LOCK(p); 64433965Sjdp swapclear(p); 64533965Sjdp p->p_swtick = ticks; 64633965Sjdp 64733965Sjdp wakeup(&p->p_flag); 64833965Sjdp 64933965Sjdp /* Allow other threads to swap p out now. */ 65033965Sjdp --p->p_lock; 65133965Sjdp } 65233965Sjdp#endif /* NO_SWAPPING */ 65333965Sjdp} 65433965Sjdp 65533965Sjdp/* 65633965Sjdp * This swapin algorithm attempts to swap-in processes only if there 65733965Sjdp * is enough space for them. Of course, if a process waits for a long 65833965Sjdp * time, it will be swapped in anyway. 65933965Sjdp * 66033965Sjdp * Giant is held on entry. 66133965Sjdp */ 66233965Sjdp/* ARGSUSED*/ 66333965Sjdpstatic void 66433965Sjdpscheduler(dummy) 66533965Sjdp void *dummy; 66633965Sjdp{ 66733965Sjdp struct proc *p; 66833965Sjdp struct thread *td; 66933965Sjdp struct proc *pp; 67033965Sjdp int slptime; 67177298Sobrien int swtime; 67233965Sjdp int ppri; 67333965Sjdp int pri; 67433965Sjdp 67533965Sjdp mtx_assert(&Giant, MA_OWNED | MA_NOTRECURSED); 67633965Sjdp mtx_unlock(&Giant); 67733965Sjdp 67833965Sjdploop: 67933965Sjdp if (vm_page_count_min()) { 68033965Sjdp VM_WAIT; 68133965Sjdp goto loop; 68233965Sjdp } 68333965Sjdp 68477298Sobrien pp = NULL; 68533965Sjdp ppri = INT_MIN; 68633965Sjdp sx_slock(&allproc_lock); 68733965Sjdp FOREACH_PROC_IN_SYSTEM(p) { 68833965Sjdp PROC_LOCK(p); 68933965Sjdp if (p->p_flag & (P_SWAPPINGOUT | P_SWAPPINGIN | P_INMEM)) { 69033965Sjdp PROC_UNLOCK(p); 69133965Sjdp continue; 69233965Sjdp } 69333965Sjdp swtime = (ticks - p->p_swtick) / hz; 69433965Sjdp FOREACH_THREAD_IN_PROC(p, td) { 69533965Sjdp /* 69633965Sjdp * An otherwise runnable thread of a process 69733965Sjdp * swapped out has only the TDI_SWAPPED bit set. 69833965Sjdp * 69933965Sjdp */ 70033965Sjdp thread_lock(td); 70133965Sjdp if (td->td_inhibitors == TDI_SWAPPED) { 70233965Sjdp slptime = (ticks - td->td_slptick) / hz; 70333965Sjdp pri = swtime + slptime; 70433965Sjdp if ((td->td_flags & TDF_SWAPINREQ) == 0) 70533965Sjdp pri -= p->p_nice * 8; 70633965Sjdp /* 70733965Sjdp * if this thread is higher priority 70833965Sjdp * and there is enough space, then select 70933965Sjdp * this process instead of the previous 71033965Sjdp * selection. 71133965Sjdp */ 71277298Sobrien if (pri > ppri) { 71333965Sjdp pp = p; 71433965Sjdp ppri = pri; 71533965Sjdp } 71633965Sjdp } 71733965Sjdp thread_unlock(td); 71833965Sjdp } 71933965Sjdp PROC_UNLOCK(p); 72033965Sjdp } 72138889Sjdp sx_sunlock(&allproc_lock); 72233965Sjdp 72338889Sjdp /* 72477298Sobrien * Nothing to do, back to sleep. 72538889Sjdp */ 72638889Sjdp if ((p = pp) == NULL) { 72733965Sjdp tsleep(&proc0, PVM, "sched", maxslp * hz / 2); 72833965Sjdp goto loop; 72977298Sobrien } 73038889Sjdp PROC_LOCK(p); 73138889Sjdp 73277298Sobrien /* 73338889Sjdp * Another process may be bringing or may have already 73477298Sobrien * brought this process in while we traverse all threads. 73538889Sjdp * Or, this process may even be being swapped out again. 73638889Sjdp */ 73738889Sjdp if (p->p_flag & (P_INMEM | P_SWAPPINGOUT | P_SWAPPINGIN)) { 73838889Sjdp PROC_UNLOCK(p); 73938889Sjdp goto loop; 74038889Sjdp } 74138889Sjdp 74238889Sjdp /* 74338889Sjdp * We would like to bring someone in. (only if there is space). 74438889Sjdp * [What checks the space? ] 74538889Sjdp */ 74638889Sjdp faultin(p); 74738889Sjdp PROC_UNLOCK(p); 74838889Sjdp goto loop; 74938889Sjdp} 75038889Sjdp 75138889Sjdpvoid 75238889Sjdpkick_proc0(void) 75338889Sjdp{ 75438889Sjdp 75538889Sjdp wakeup(&proc0); 75638889Sjdp} 75777298Sobrien 75838889Sjdp#ifndef NO_SWAPPING 75938889Sjdp 76038889Sjdp/* 76138889Sjdp * Swap_idle_threshold1 is the guaranteed swapped in time for a process 76238889Sjdp */ 76338889Sjdpstatic int swap_idle_threshold1 = 2; 76438889SjdpSYSCTL_INT(_vm, OID_AUTO, swap_idle_threshold1, CTLFLAG_RW, 76538889Sjdp &swap_idle_threshold1, 0, "Guaranteed swapped in time for a process"); 76638889Sjdp 76738889Sjdp/* 76838889Sjdp * Swap_idle_threshold2 is the time that a process can be idle before 76977298Sobrien * it will be swapped out, if idle swapping is enabled. 77038889Sjdp */ 77177298Sobrienstatic int swap_idle_threshold2 = 10; 77277298SobrienSYSCTL_INT(_vm, OID_AUTO, swap_idle_threshold2, CTLFLAG_RW, 77338889Sjdp &swap_idle_threshold2, 0, "Time before a process will be swapped out"); 77433965Sjdp 77533965Sjdp/* 77677298Sobrien * Swapout is driven by the pageout daemon. Very simple, we find eligible 77733965Sjdp * procs and swap out their stacks. We try to always "swap" at least one 77877298Sobrien * process in case we need the room for a swapin. 77933965Sjdp * If any procs have been sleeping/stopped for at least maxslp seconds, 78077298Sobrien * they are swapped. Else, we swap the longest-sleeping or stopped process, 78133965Sjdp * if any, otherwise the longest-resident process. 78277298Sobrien */ 78377298Sobrienvoid 78433965Sjdpswapout_procs(action) 78533965Sjdpint action; 78633965Sjdp{ 78777298Sobrien struct proc *p; 78833965Sjdp struct thread *td; 78933965Sjdp int didswap = 0; 79077298Sobrien 79133965Sjdpretry: 79233965Sjdp sx_slock(&allproc_lock); 79377298Sobrien FOREACH_PROC_IN_SYSTEM(p) { 79477298Sobrien struct vmspace *vm; 79533965Sjdp int minslptime = 100000; 79633965Sjdp int slptime; 79733965Sjdp 79833965Sjdp /* 79938889Sjdp * Watch out for a process in 80033965Sjdp * creation. It may have no 80177298Sobrien * address space or lock yet. 80233965Sjdp */ 80333965Sjdp if (p->p_state == PRS_NEW) 80433965Sjdp continue; 80538889Sjdp /* 80677298Sobrien * An aio daemon switches its 80738889Sjdp * address space while running. 80877298Sobrien * Perform a quick check whether 80933965Sjdp * a process has P_SYSTEM. 81038889Sjdp */ 81177298Sobrien if ((p->p_flag & P_SYSTEM) != 0) 81277298Sobrien continue; 81377298Sobrien /* 81477298Sobrien * Do not swapout a process that 81577298Sobrien * is waiting for VM data 81677298Sobrien * structures as there is a possible 81777298Sobrien * deadlock. Test this first as 81877298Sobrien * this may block. 81977298Sobrien * 82077298Sobrien * Lock the map until swapout 82138889Sjdp * finishes, or a thread of this 82277298Sobrien * process may attempt to alter 82377298Sobrien * the map. 82477298Sobrien */ 82538889Sjdp vm = vmspace_acquire_ref(p); 82633965Sjdp if (vm == NULL) 82738889Sjdp continue; 82838889Sjdp if (!vm_map_trylock(&vm->vm_map)) 82938889Sjdp goto nextproc1; 83033965Sjdp 83138889Sjdp PROC_LOCK(p); 83238889Sjdp if (p->p_lock != 0 || 83338889Sjdp (p->p_flag & (P_STOPPED_SINGLE|P_TRACED|P_SYSTEM|P_WEXIT) 83433965Sjdp ) != 0) { 83533965Sjdp goto nextproc; 83638889Sjdp } 83738889Sjdp /* 83838889Sjdp * only aiod changes vmspace, however it will be 83933965Sjdp * skipped because of the if statement above checking 84038889Sjdp * for P_SYSTEM 84138889Sjdp */ 84238889Sjdp if ((p->p_flag & (P_INMEM|P_SWAPPINGOUT|P_SWAPPINGIN)) != P_INMEM) 84360484Sobrien goto nextproc; 84460484Sobrien 84560484Sobrien switch (p->p_state) { 84660484Sobrien default: 84733965Sjdp /* Don't swap out processes in any sort 84838889Sjdp * of 'special' state. */ 84960484Sobrien break; 85033965Sjdp 85133965Sjdp case PRS_NORMAL: 85233965Sjdp /* 85333965Sjdp * do not swapout a realtime process 85433965Sjdp * Check all the thread groups.. 85533965Sjdp */ 85638889Sjdp FOREACH_THREAD_IN_PROC(p, td) { 85738889Sjdp thread_lock(td); 85838889Sjdp if (PRI_IS_REALTIME(td->td_pri_class)) { 85933965Sjdp thread_unlock(td); 86033965Sjdp goto nextproc; 86133965Sjdp } 86233965Sjdp slptime = (ticks - td->td_slptick) / hz; 86338889Sjdp /* 86433965Sjdp * Guarantee swap_idle_threshold1 86533965Sjdp * time in memory. 86633965Sjdp */ 86733965Sjdp if (slptime < swap_idle_threshold1) { 86833965Sjdp thread_unlock(td); 86933965Sjdp goto nextproc; 87060484Sobrien } 87160484Sobrien 87260484Sobrien /* 87360484Sobrien * Do not swapout a process if it is 87460484Sobrien * waiting on a critical event of some 87560484Sobrien * kind or there is a thread whose 87660484Sobrien * pageable memory may be accessed. 87760484Sobrien * 87860484Sobrien * This could be refined to support 87960484Sobrien * swapping out a thread. 88060484Sobrien */ 88160484Sobrien if (!thread_safetoswapout(td)) { 88260484Sobrien thread_unlock(td); 88360484Sobrien goto nextproc; 88460484Sobrien } 88560484Sobrien /* 88660484Sobrien * If the system is under memory stress, 88760484Sobrien * or if we are swapping 88860484Sobrien * idle processes >= swap_idle_threshold2, 88960484Sobrien * then swap the process out. 89077298Sobrien */ 89133965Sjdp if (((action & VM_SWAP_NORMAL) == 0) && 89233965Sjdp (((action & VM_SWAP_IDLE) == 0) || 89333965Sjdp (slptime < swap_idle_threshold2))) { 89477298Sobrien thread_unlock(td); 89533965Sjdp goto nextproc; 89633965Sjdp } 89733965Sjdp 89833965Sjdp if (minslptime > slptime) 89933965Sjdp minslptime = slptime; 90033965Sjdp thread_unlock(td); 90133965Sjdp } 90233965Sjdp 90333965Sjdp /* 90433965Sjdp * If the pageout daemon didn't free enough pages, 90533965Sjdp * or if this process is idle and the system is 90633965Sjdp * configured to swap proactively, swap it out. 90733965Sjdp */ 90833965Sjdp if ((action & VM_SWAP_NORMAL) || 90933965Sjdp ((action & VM_SWAP_IDLE) && 91033965Sjdp (minslptime > swap_idle_threshold2))) { 91133965Sjdp if (swapout(p) == 0) 91233965Sjdp didswap++; 91333965Sjdp PROC_UNLOCK(p); 91433965Sjdp vm_map_unlock(&vm->vm_map); 91533965Sjdp vmspace_free(vm); 91633965Sjdp sx_sunlock(&allproc_lock); 91733965Sjdp goto retry; 91833965Sjdp } 91933965Sjdp } 92033965Sjdpnextproc: 92133965Sjdp PROC_UNLOCK(p); 92233965Sjdp vm_map_unlock(&vm->vm_map); 92333965Sjdpnextproc1: 92433965Sjdp vmspace_free(vm); 92533965Sjdp continue; 92633965Sjdp } 92733965Sjdp sx_sunlock(&allproc_lock); 92833965Sjdp /* 92933965Sjdp * If we swapped something out, and another process needed memory, 93033965Sjdp * then wakeup the sched process. 93133965Sjdp */ 93233965Sjdp if (didswap) 93333965Sjdp wakeup(&proc0); 93433965Sjdp} 93533965Sjdp 93633965Sjdpstatic void 93733965Sjdpswapclear(p) 93877298Sobrien struct proc *p; 93933965Sjdp{ 94033965Sjdp struct thread *td; 94133965Sjdp 94233965Sjdp PROC_LOCK_ASSERT(p, MA_OWNED); 94333965Sjdp 94433965Sjdp FOREACH_THREAD_IN_PROC(p, td) { 94533965Sjdp thread_lock(td); 94633965Sjdp td->td_flags |= TDF_INMEM; 94733965Sjdp td->td_flags &= ~TDF_SWAPINREQ; 94833965Sjdp TD_CLR_SWAPPED(td); 94933965Sjdp if (TD_CAN_RUN(td)) 95033965Sjdp if (setrunnable(td)) { 95133965Sjdp#ifdef INVARIANTS 95233965Sjdp /* 95333965Sjdp * XXX: We just cleared TDI_SWAPPED 95433965Sjdp * above and set TDF_INMEM, so this 95533965Sjdp * should never happen. 95633965Sjdp */ 95733965Sjdp panic("not waking up swapper"); 95833965Sjdp#endif 95933965Sjdp } 96033965Sjdp thread_unlock(td); 96133965Sjdp } 96233965Sjdp p->p_flag &= ~(P_SWAPPINGIN|P_SWAPPINGOUT); 96333965Sjdp p->p_flag |= P_INMEM; 96433965Sjdp} 96533965Sjdp 96633965Sjdpstatic int 96733965Sjdpswapout(p) 96833965Sjdp struct proc *p; 96977298Sobrien{ 97033965Sjdp struct thread *td; 97133965Sjdp 97233965Sjdp PROC_LOCK_ASSERT(p, MA_OWNED); 97333965Sjdp#if defined(SWAP_DEBUG) 97433965Sjdp printf("swapping out %d\n", p->p_pid); 97533965Sjdp#endif 97633965Sjdp 97733965Sjdp /* 97833965Sjdp * The states of this process and its threads may have changed 97977298Sobrien * by now. Assuming that there is only one pageout daemon thread, 98033965Sjdp * this process should still be in memory. 98177298Sobrien */ 98277298Sobrien KASSERT((p->p_flag & (P_INMEM|P_SWAPPINGOUT|P_SWAPPINGIN)) == P_INMEM, 98377298Sobrien ("swapout: lost a swapout race?")); 98433965Sjdp 98560484Sobrien /* 98660484Sobrien * remember the process resident count 98760484Sobrien */ 98860484Sobrien p->p_vmspace->vm_swrss = vmspace_resident_count(p->p_vmspace); 98960484Sobrien /* 99060484Sobrien * Check and mark all threads before we proceed. 99160484Sobrien */ 99260484Sobrien p->p_flag &= ~P_INMEM; 99360484Sobrien p->p_flag |= P_SWAPPINGOUT; 99460484Sobrien FOREACH_THREAD_IN_PROC(p, td) { 99560484Sobrien thread_lock(td); 99660484Sobrien if (!thread_safetoswapout(td)) { 99760484Sobrien thread_unlock(td); 99860484Sobrien swapclear(p); 99960484Sobrien return (EBUSY); 100060484Sobrien } 100177298Sobrien td->td_flags &= ~TDF_INMEM; 100260484Sobrien TD_SET_SWAPPED(td); 100377298Sobrien thread_unlock(td); 100477298Sobrien } 100560484Sobrien td = FIRST_THREAD_IN_PROC(p); 100660484Sobrien ++td->td_ru.ru_nswap; 100760484Sobrien PROC_UNLOCK(p); 100860484Sobrien 100960484Sobrien /* 101060484Sobrien * This list is stable because all threads are now prevented from 101160484Sobrien * running. The list is only modified in the context of a running 101260484Sobrien * thread in this process. 101360484Sobrien */ 101460484Sobrien FOREACH_THREAD_IN_PROC(p, td) 101560484Sobrien vm_thread_swapout(td); 101660484Sobrien 101777298Sobrien PROC_LOCK(p); 101860484Sobrien p->p_flag &= ~P_SWAPPINGOUT; 101960484Sobrien p->p_swtick = ticks; 102060484Sobrien return (0); 102160484Sobrien} 102260484Sobrien#endif /* !NO_SWAPPING */ 102360484Sobrien