2 * Copyright (c) 1991, 1993
3 * The Regents of the University of California. All rights reserved.
5 * This code is derived from software contributed to Berkeley by
6 * The Mach Operating System project at Carnegie-Mellon University.
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 * 3. All advertising materials mentioning features or use of this software
17 * must display the following acknowledgement:
18 * This product includes software developed by the University of
19 * California, Berkeley and its contributors.
20 * 4. Neither the name of the University nor the names of its contributors
21 * may be used to endorse or promote products derived from this software
22 * without specific prior written permission.
24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36 * from: @(#)vm_glue.c 8.6 (Berkeley) 1/5/94
39 * Copyright (c) 1987, 1990 Carnegie-Mellon University.
40 * All rights reserved.
42 * Permission to use, copy, modify and distribute this software and
43 * its documentation is hereby granted, provided that both the copyright
44 * notice and this permission notice appear in all copies of the
45 * software, derivative works or modified versions, and any portions
46 * thereof, and that both notices appear in supporting documentation.
48 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
49 * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
50 * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
52 * Carnegie Mellon requests users of this software to return to
54 * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU
55 * School of Computer Science
56 * Carnegie Mellon University
57 * Pittsburgh PA 15213-3890
59 * any improvements or extensions that they make and grant Carnegie the
60 * rights to redistribute these changes.
62 * $FreeBSD: src/sys/vm/vm_glue.c,v 1.94.2.4 2003/01/13 22:51:17 dillon Exp $
63 * $DragonFly: src/sys/vm/vm_glue.c,v 1.52 2007/02/25 23:17:13 corecode Exp $
68 #include <sys/param.h>
69 #include <sys/systm.h>
71 #include <sys/resourcevar.h>
74 #include <sys/vmmeter.h>
75 #include <sys/sysctl.h>
77 #include <sys/kernel.h>
78 #include <sys/unistd.h>
80 #include <machine/limits.h>
83 #include <vm/vm_param.h>
86 #include <vm/vm_map.h>
87 #include <vm/vm_page.h>
88 #include <vm/vm_pageout.h>
89 #include <vm/vm_kern.h>
90 #include <vm/vm_extern.h>
93 #include <vm/vm_page2.h>
94 #include <sys/thread2.h>
97 * System initialization
99 * Note: proc0 from proc.h
102 static void vm_init_limits (void *);
103 SYSINIT(vm_limits, SI_SUB_VM_CONF, SI_ORDER_FIRST, vm_init_limits, &proc0)
106 * THIS MUST BE THE LAST INITIALIZATION ITEM!!!
108 * Note: run scheduling should be divorced from the vm system.
110 static void scheduler (void *);
111 SYSINIT(scheduler, SI_SUB_RUN_SCHEDULER, SI_ORDER_FIRST, scheduler, NULL)
115 static int swap_debug = 0;
116 SYSCTL_INT(_vm, OID_AUTO, swap_debug,
117 CTLFLAG_RW, &swap_debug, 0, "");
121 static int scheduler_notify;
123 static void swapout (struct proc *);
126 kernacc(c_caddr_t addr, int len, int rw)
129 vm_offset_t saddr, eaddr;
132 KASSERT((rw & (~VM_PROT_ALL)) == 0,
133 ("illegal ``rw'' argument to kernacc (%x)\n", rw));
136 * The globaldata space is not part of the kernel_map proper,
137 * check access separately.
139 if (is_globaldata_space((vm_offset_t)addr, (vm_offset_t)(addr + len)))
143 * Nominal kernel memory access - check access via kernel_map.
145 if ((vm_offset_t)addr + len > kernel_map.max_offset ||
146 (vm_offset_t)addr + len < (vm_offset_t)addr) {
150 saddr = trunc_page((vm_offset_t)addr);
151 eaddr = round_page((vm_offset_t)addr + len);
152 vm_map_lock_read(&kernel_map);
153 rv = vm_map_check_protection(&kernel_map, saddr, eaddr, prot);
154 vm_map_unlock_read(&kernel_map);
159 useracc(c_caddr_t addr, int len, int rw)
164 vm_map_entry_t save_hint;
166 KASSERT((rw & (~VM_PROT_ALL)) == 0,
167 ("illegal ``rw'' argument to useracc (%x)\n", rw));
170 * XXX - check separately to disallow access to user area and user
171 * page tables - they are in the map.
173 * XXX - VM_MAX_USER_ADDRESS is an end address, not a max. It was once
174 * only used (as an end address) in trap.c. Use it as an end address
175 * here too. This bogusness has spread. I just fixed where it was
176 * used as a max in vm_mmap.c.
178 if ((vm_offset_t) addr + len > /* XXX */ VM_MAX_USER_ADDRESS
179 || (vm_offset_t) addr + len < (vm_offset_t) addr) {
182 map = &curproc->p_vmspace->vm_map;
183 vm_map_lock_read(map);
185 * We save the map hint, and restore it. Useracc appears to distort
186 * the map hint unnecessarily.
188 save_hint = map->hint;
189 rv = vm_map_check_protection(map,
190 trunc_page((vm_offset_t)addr), round_page((vm_offset_t)addr + len), prot);
191 map->hint = save_hint;
192 vm_map_unlock_read(map);
198 vslock(caddr_t addr, u_int len)
201 vm_map_wire(&curproc->p_vmspace->vm_map,
202 trunc_page((vm_offset_t)addr),
203 round_page((vm_offset_t)addr + len), 0);
208 vsunlock(caddr_t addr, u_int len)
211 vm_map_wire(&curproc->p_vmspace->vm_map,
212 trunc_page((vm_offset_t)addr),
213 round_page((vm_offset_t)addr + len),
219 * Implement fork's actions on an address space.
220 * Here we arrange for the address space to be copied or referenced,
221 * allocate a user struct (pcb and kernel stack), then call the
222 * machine-dependent layer to fill those in and make the new process
223 * ready to run. The new process is set up so that it returns directly
224 * to user mode to avoid stack copying and relocation problems.
227 vm_fork(struct lwp *lp1, struct proc *p2, int flags)
229 struct proc *p1 = lp1->lwp_proc;
232 if ((flags & RFPROC) == 0) {
234 * Divorce the memory, if it is shared, essentially
235 * this changes shared memory amongst threads, into
238 if ((flags & RFMEM) == 0) {
239 if (p1->p_vmspace->vm_refcnt > 1) {
243 cpu_fork(lp1, NULL, flags);
248 p2->p_vmspace = p1->p_vmspace;
249 p1->p_vmspace->vm_refcnt++;
252 while (vm_page_count_severe()) {
256 if ((flags & RFMEM) == 0) {
257 p2->p_vmspace = vmspace_fork(p1->p_vmspace);
259 pmap_pinit2(vmspace_pmap(p2->p_vmspace));
261 if (p1->p_vmspace->vm_shm)
265 td2 = lwkt_alloc_thread(NULL, LWKT_THREAD_STACK, -1, 0);
266 pmap_init_proc(p2, td2);
267 lwkt_setpri(td2, TDPRI_KERN_USER);
268 lwkt_set_comm(td2, "%s", p1->p_comm);
271 * cpu_fork will copy and update the pcb, set up the kernel stack,
272 * and make the child ready to run.
274 cpu_fork(lp1, td2->td_lwp, flags);
278 * Called after process has been wait(2)'ed apon and is being reaped.
279 * The idea is to reclaim resources that we could not reclaim while
280 * the process was still executing.
283 vm_waitproc(struct proc *p)
286 vmspace_exitfree(p); /* and clean-out the vmspace */
290 * Set default limits for VM system.
291 * Called for proc 0, and then inherited by all others.
293 * XXX should probably act directly on proc0.
296 vm_init_limits(void *udata)
298 struct proc *p = udata;
302 * Set up the initial limits on process VM. Set the maximum resident
303 * set size to be half of (reasonably) available memory. Since this
304 * is a soft limit, it comes into effect only when the system is out
305 * of memory - half of main memory helps to favor smaller processes,
306 * and reduces thrashing of the object cache.
308 p->p_rlimit[RLIMIT_STACK].rlim_cur = dflssiz;
309 p->p_rlimit[RLIMIT_STACK].rlim_max = maxssiz;
310 p->p_rlimit[RLIMIT_DATA].rlim_cur = dfldsiz;
311 p->p_rlimit[RLIMIT_DATA].rlim_max = maxdsiz;
312 /* limit the limit to no less than 2MB */
313 rss_limit = max(vmstats.v_free_count, 512);
314 p->p_rlimit[RLIMIT_RSS].rlim_cur = ptoa(rss_limit);
315 p->p_rlimit[RLIMIT_RSS].rlim_max = RLIM_INFINITY;
319 * Faultin the specified process. Note that the process can be in any
320 * state. Just clear P_SWAPPEDOUT and call wakeup in case the process is
324 faultin(struct proc *p)
326 if (p->p_flag & P_SWAPPEDOUT) {
328 * The process is waiting in the kernel to return to user
329 * mode but cannot until P_SWAPPEDOUT gets cleared.
332 p->p_flag &= ~(P_SWAPPEDOUT | P_SWAPWAIT);
335 kprintf("swapping in %d (%s)\n", p->p_pid, p->p_comm);
344 * Kernel initialization eventually falls through to this function,
345 * which is process 0.
347 * This swapin algorithm attempts to swap-in processes only if there
348 * is enough space for them. Of course, if a process waits for a long
349 * time, it will be swapped in anyway.
352 struct scheduler_info {
357 static int scheduler_callback(struct proc *p, void *data);
360 scheduler(void *dummy)
362 struct scheduler_info info;
365 KKASSERT(!IN_CRITICAL_SECT(curthread));
367 scheduler_notify = 0;
369 * Don't try to swap anything in if we are low on memory.
371 if (vm_page_count_min()) {
377 * Look for a good candidate to wake up
381 allproc_scan(scheduler_callback, &info);
384 * Nothing to do, back to sleep for at least 1/10 of a second. If
385 * we are woken up, immediately process the next request. If
386 * multiple requests have built up the first is processed
387 * immediately and the rest are staggered.
389 if ((p = info.pp) == NULL) {
390 tsleep(&proc0, 0, "nowork", hz / 10);
391 if (scheduler_notify == 0)
392 tsleep(&scheduler_notify, 0, "nowork", 0);
397 * Fault the selected process in, then wait for a short period of
400 * XXX we need a heuristic to get a measure of system stress and
401 * then adjust our stagger wakeup delay accordingly.
406 tsleep(&proc0, 0, "swapin", hz / 10);
411 scheduler_callback(struct proc *p, void *data)
413 struct scheduler_info *info = data;
418 if (p->p_flag & P_SWAPWAIT) {
420 FOREACH_LWP_IN_PROC(lp, p) {
421 /* XXX lwp might need a different metric */
422 pri += lp->lwp_slptime;
424 pri += p->p_swtime - p->p_nice * 8;
427 * The more pages paged out while we were swapped,
428 * the more work we have to do to get up and running
429 * again and the lower our wakeup priority.
431 * Each second of sleep time is worth ~1MB
433 pgs = vmspace_resident_count(p->p_vmspace);
434 if (pgs < p->p_vmspace->vm_swrss) {
435 pri -= (p->p_vmspace->vm_swrss - pgs) /
436 (1024 * 1024 / PAGE_SIZE);
440 * If this process is higher priority and there is
441 * enough space, then select this process instead of
442 * the previous selection.
444 if (pri > info->ppri) {
458 if (scheduler_notify == 0) {
459 scheduler_notify = 1;
460 wakeup(&scheduler_notify);
466 #define swappable(p) \
467 (((p)->p_lock == 0) && \
468 ((p)->p_flag & (P_TRACED|P_SYSTEM|P_SWAPPEDOUT|P_WEXIT)) == 0)
472 * Swap_idle_threshold1 is the guaranteed swapped in time for a process
474 static int swap_idle_threshold1 = 15;
475 SYSCTL_INT(_vm, OID_AUTO, swap_idle_threshold1,
476 CTLFLAG_RW, &swap_idle_threshold1, 0, "");
479 * Swap_idle_threshold2 is the time that a process can be idle before
480 * it will be swapped out, if idle swapping is enabled. Default is
483 static int swap_idle_threshold2 = 60;
484 SYSCTL_INT(_vm, OID_AUTO, swap_idle_threshold2,
485 CTLFLAG_RW, &swap_idle_threshold2, 0, "");
488 * Swapout is driven by the pageout daemon. Very simple, we find eligible
489 * procs and mark them as being swapped out. This will cause the kernel
490 * to prefer to pageout those proc's pages first and the procs in question
491 * will not return to user mode until the swapper tells them they can.
493 * If any procs have been sleeping/stopped for at least maxslp seconds,
494 * they are swapped. Else, we swap the longest-sleeping or stopped process,
495 * if any, otherwise the longest-resident process.
498 static int swapout_procs_callback(struct proc *p, void *data);
501 swapout_procs(int action)
503 allproc_scan(swapout_procs_callback, &action);
507 swapout_procs_callback(struct proc *p, void *data)
511 int action = *(int *)data;
520 * We only consider active processes.
522 if (p->p_stat != SACTIVE && p->p_stat != SSTOP)
525 FOREACH_LWP_IN_PROC(lp, p) {
527 * do not swap out a realtime process
529 if (RTP_PRIO_IS_REALTIME(lp->lwp_rtprio.type))
533 * Guarentee swap_idle_threshold time in memory
535 if (lp->lwp_slptime < swap_idle_threshold1)
539 * If the system is under memory stress, or if we
540 * are swapping idle processes >= swap_idle_threshold2,
541 * then swap the process out.
543 if (((action & VM_SWAP_NORMAL) == 0) &&
544 (((action & VM_SWAP_IDLE) == 0) ||
545 (lp->lwp_slptime < swap_idle_threshold2))) {
549 if (minslp == -1 || lp->lwp_slptime < minslp)
550 minslp = lp->lwp_slptime;
556 * If the process has been asleep for awhile, swap
559 if ((action & VM_SWAP_NORMAL) ||
560 ((action & VM_SWAP_IDLE) &&
561 (minslp > swap_idle_threshold2))) {
566 * cleanup our reference
574 swapout(struct proc *p)
578 kprintf("swapping out %d (%s)\n", p->p_pid, p->p_comm);
582 * remember the process resident count
584 p->p_vmspace->vm_swrss = vmspace_resident_count(p->p_vmspace);
585 p->p_flag |= P_SWAPPEDOUT;
589 #endif /* !NO_SWAPPING */