vm_zeroidle.c revision 134461 
1/*-
2 * Copyright (c) 1994 John Dyson
3 * Copyright (c) 2001 Matt Dillon
4 *
5 * All Rights Reserved.
6 * Redistribution and use in source and binary forms, with or without
7 * modification, are permitted provided that the following conditions
8 * are met:
9 * 1. Redistributions of source code must retain the above copyright
10 *    notice, this list of conditions and the following disclaimer.
11 * 2. Redistributions in binary form must reproduce the above copyright
12 *    notice, this list of conditions and the following disclaimer in the
13 *    documentation and/or other materials provided with the distribution.
14 * 4. Neither the name of the University nor the names of its contributors
15 *    may be used to endorse or promote products derived from this software
16 *    without specific prior written permission.
17 *
18 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS
19 * OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
20 * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
21 * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY
22 * DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
23 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE
24 * GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
25 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
26 * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
27 * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
28 * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
29 *
30 *	from: @(#)vm_machdep.c	7.3 (Berkeley) 5/13/91
31 *	Utah $Hdr: vm_machdep.c 1.16.1.1 89/06/23$
32 * from: FreeBSD: .../i386/vm_machdep.c,v 1.165 2001/07/04 23:27:04 dillon
33 */
34
35#include <sys/cdefs.h>
36__FBSDID("$FreeBSD: head/sys/vm/vm_zeroidle.c 134461 2004-08-29 01:02:33Z iedowse $");
37
38#include <sys/param.h>
39#include <sys/systm.h>
40#include <sys/kernel.h>
41#include <sys/proc.h>
42#include <sys/resourcevar.h>
43#include <sys/vmmeter.h>
44#include <sys/lock.h>
45#include <sys/mutex.h>
46#include <sys/sched.h>
47#include <sys/sysctl.h>
48#include <sys/kthread.h>
49#include <sys/unistd.h>
50
51#include <vm/vm.h>
52#include <vm/vm_page.h>
53
54SYSCTL_DECL(_vm_stats_misc);
55
56static int cnt_prezero;
57SYSCTL_INT(_vm_stats_misc, OID_AUTO, cnt_prezero, CTLFLAG_RD,
58    &cnt_prezero, 0, "");
59
60static int idlezero_enable_default = 1;
61TUNABLE_INT("vm.idlezero_enable", &idlezero_enable_default);
62/* Defer setting the enable flag until the kthread is running. */
63static int idlezero_enable = 0;
64SYSCTL_INT(_vm, OID_AUTO, idlezero_enable, CTLFLAG_RW, &idlezero_enable, 0, "");
65
66static int idlezero_maxrun = 16;
67SYSCTL_INT(_vm, OID_AUTO, idlezero_maxrun, CTLFLAG_RW, &idlezero_maxrun, 0, "");
68TUNABLE_INT("vm.idlezero_maxrun", &idlezero_maxrun);
69
70/*
71 * Implement the pre-zeroed page mechanism.
72 */
73
74#define ZIDLE_LO(v)	((v) * 2 / 3)
75#define ZIDLE_HI(v)	((v) * 4 / 5)
76
77static int zero_state;
78
79static int
80vm_page_zero_check(void)
81{
82
83	if (!idlezero_enable)
84		return (0);
85	/*
86	 * Attempt to maintain approximately 1/2 of our free pages in a
87	 * PG_ZERO'd state.   Add some hysteresis to (attempt to) avoid
88	 * generally zeroing a page when the system is near steady-state.
89	 * Otherwise we might get 'flutter' during disk I/O / IPC or
90	 * fast sleeps.  We also do not want to be continuously zeroing
91	 * pages because doing so may flush our L1 and L2 caches too much.
92	 */
93	if (zero_state && vm_page_zero_count >= ZIDLE_LO(cnt.v_free_count))
94		return (0);
95	if (vm_page_zero_count >= ZIDLE_HI(cnt.v_free_count))
96		return (0);
97	return (1);
98}
99
100static int
101vm_page_zero_idle(void)
102{
103	static int free_rover;
104	vm_page_t m;
105
106	mtx_lock_spin(&vm_page_queue_free_mtx);
107	zero_state = 0;
108	m = vm_pageq_find(PQ_FREE, free_rover, FALSE);
109	if (m != NULL && (m->flags & PG_ZERO) == 0) {
110		vm_pageq_remove_nowakeup(m);
111		mtx_unlock_spin(&vm_page_queue_free_mtx);
112		pmap_zero_page_idle(m);
113		mtx_lock_spin(&vm_page_queue_free_mtx);
114		m->flags |= PG_ZERO;
115		vm_pageq_enqueue(PQ_FREE + m->pc, m);
116		++vm_page_zero_count;
117		++cnt_prezero;
118		if (vm_page_zero_count >= ZIDLE_HI(cnt.v_free_count))
119			zero_state = 1;
120	}
121	free_rover = (free_rover + PQ_PRIME2) & PQ_L2_MASK;
122	mtx_unlock_spin(&vm_page_queue_free_mtx);
123	return (1);
124}
125
126/* Called by vm_page_free to hint that a new page is available. */
127void
128vm_page_zero_idle_wakeup(void)
129{
130
131	if (idlezero_enable && vm_page_zero_check())
132		wakeup(&zero_state);
133}
134
135static void
136vm_pagezero(void __unused *arg)
137{
138	struct proc *p;
139	struct rtprio rtp;
140	struct thread *td;
141	int pages, pri;
142
143	td = curthread;
144	p = td->td_proc;
145	rtp.prio = RTP_PRIO_MAX;
146	rtp.type = RTP_PRIO_IDLE;
147	pages = 0;
148	mtx_lock_spin(&sched_lock);
149	rtp_to_pri(&rtp, td->td_ksegrp);
150	pri = td->td_priority;
151	mtx_unlock_spin(&sched_lock);
152	idlezero_enable = idlezero_enable_default;
153
154	for (;;) {
155		if (vm_page_zero_check()) {
156			pages += vm_page_zero_idle();
157#ifndef PREEMPTION
158			if (pages > idlezero_maxrun || sched_runnable()) {
159				mtx_lock_spin(&sched_lock);
160				mi_switch(SW_VOL, NULL);
161				mtx_unlock_spin(&sched_lock);
162				pages = 0;
163			}
164#endif
165		} else {
166			tsleep(&zero_state, pri, "pgzero", hz * 300);
167			pages = 0;
168		}
169	}
170}
171
172static struct proc *pagezero_proc;
173
174static void
175pagezero_start(void __unused *arg)
176{
177	int error;
178
179	error = kthread_create(vm_pagezero, NULL, &pagezero_proc, RFSTOPPED, 0,
180	    "pagezero");
181	if (error)
182		panic("pagezero_start: error %d\n", error);
183	/*
184	 * We're an idle task, don't count us in the load.
185	 */
186	PROC_LOCK(pagezero_proc);
187	pagezero_proc->p_flag |= P_NOLOAD;
188	PROC_UNLOCK(pagezero_proc);
189	mtx_lock_spin(&sched_lock);
190	setrunqueue(FIRST_THREAD_IN_PROC(pagezero_proc));
191	mtx_unlock_spin(&sched_lock);
192}
193SYSINIT(pagezero, SI_SUB_KTHREAD_VM, SI_ORDER_ANY, pagezero_start, NULL)
194