vm_zeroidle.c revision 134649 
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 134649 2004-09-02 18:59:15Z scottl $");
37
38#include <opt_sched.h>
39
40#include <sys/param.h>
41#include <sys/systm.h>
42#include <sys/kernel.h>
43#include <sys/proc.h>
44#include <sys/resourcevar.h>
45#include <sys/vmmeter.h>
46#include <sys/lock.h>
47#include <sys/mutex.h>
48#include <sys/sched.h>
49#include <sys/sysctl.h>
50#include <sys/kthread.h>
51#include <sys/unistd.h>
52
53#include <vm/vm.h>
54#include <vm/vm_page.h>
55
56SYSCTL_DECL(_vm_stats_misc);
57
58static int cnt_prezero;
59SYSCTL_INT(_vm_stats_misc, OID_AUTO, cnt_prezero, CTLFLAG_RD,
60    &cnt_prezero, 0, "");
61
62static int idlezero_enable_default = 1;
63TUNABLE_INT("vm.idlezero_enable", &idlezero_enable_default);
64/* Defer setting the enable flag until the kthread is running. */
65static int idlezero_enable = 0;
66SYSCTL_INT(_vm, OID_AUTO, idlezero_enable, CTLFLAG_RW, &idlezero_enable, 0, "");
67
68static int idlezero_maxrun = 16;
69SYSCTL_INT(_vm, OID_AUTO, idlezero_maxrun, CTLFLAG_RW, &idlezero_maxrun, 0, "");
70TUNABLE_INT("vm.idlezero_maxrun", &idlezero_maxrun);
71
72/*
73 * Implement the pre-zeroed page mechanism.
74 */
75
76#define ZIDLE_LO(v)	((v) * 2 / 3)
77#define ZIDLE_HI(v)	((v) * 4 / 5)
78
79static int zero_state;
80
81static int
82vm_page_zero_check(void)
83{
84
85	if (!idlezero_enable)
86		return (0);
87	/*
88	 * Attempt to maintain approximately 1/2 of our free pages in a
89	 * PG_ZERO'd state.   Add some hysteresis to (attempt to) avoid
90	 * generally zeroing a page when the system is near steady-state.
91	 * Otherwise we might get 'flutter' during disk I/O / IPC or
92	 * fast sleeps.  We also do not want to be continuously zeroing
93	 * pages because doing so may flush our L1 and L2 caches too much.
94	 */
95	if (zero_state && vm_page_zero_count >= ZIDLE_LO(cnt.v_free_count))
96		return (0);
97	if (vm_page_zero_count >= ZIDLE_HI(cnt.v_free_count))
98		return (0);
99	return (1);
100}
101
102static int
103vm_page_zero_idle(void)
104{
105	static int free_rover;
106	vm_page_t m;
107
108	mtx_lock_spin(&vm_page_queue_free_mtx);
109	zero_state = 0;
110	m = vm_pageq_find(PQ_FREE, free_rover, FALSE);
111	if (m != NULL && (m->flags & PG_ZERO) == 0) {
112		vm_pageq_remove_nowakeup(m);
113		mtx_unlock_spin(&vm_page_queue_free_mtx);
114		pmap_zero_page_idle(m);
115		mtx_lock_spin(&vm_page_queue_free_mtx);
116		m->flags |= PG_ZERO;
117		vm_pageq_enqueue(PQ_FREE + m->pc, m);
118		++vm_page_zero_count;
119		++cnt_prezero;
120		if (vm_page_zero_count >= ZIDLE_HI(cnt.v_free_count))
121			zero_state = 1;
122	}
123	free_rover = (free_rover + PQ_PRIME2) & PQ_L2_MASK;
124	mtx_unlock_spin(&vm_page_queue_free_mtx);
125	return (1);
126}
127
128/* Called by vm_page_free to hint that a new page is available. */
129void
130vm_page_zero_idle_wakeup(void)
131{
132
133	if (idlezero_enable && vm_page_zero_check())
134		wakeup(&zero_state);
135}
136
137static void
138vm_pagezero(void __unused *arg)
139{
140	struct proc *p;
141	struct rtprio rtp;
142	struct thread *td;
143	int pages, pri;
144
145	td = curthread;
146	p = td->td_proc;
147	rtp.prio = RTP_PRIO_MAX;
148	rtp.type = RTP_PRIO_IDLE;
149	pages = 0;
150	mtx_lock_spin(&sched_lock);
151	rtp_to_pri(&rtp, td->td_ksegrp);
152	pri = td->td_priority;
153	mtx_unlock_spin(&sched_lock);
154	idlezero_enable = idlezero_enable_default;
155
156	for (;;) {
157		if (vm_page_zero_check()) {
158			pages += vm_page_zero_idle();
159#ifndef PREEMPTION
160			if (pages > idlezero_maxrun || sched_runnable()) {
161				mtx_lock_spin(&sched_lock);
162				mi_switch(SW_VOL, NULL);
163				mtx_unlock_spin(&sched_lock);
164				pages = 0;
165			}
166#endif
167		} else {
168			tsleep(&zero_state, pri, "pgzero", hz * 300);
169			pages = 0;
170		}
171	}
172}
173
174static struct proc *pagezero_proc;
175
176static void
177pagezero_start(void __unused *arg)
178{
179	int error;
180
181	error = kthread_create(vm_pagezero, NULL, &pagezero_proc, RFSTOPPED, 0,
182	    "pagezero");
183	if (error)
184		panic("pagezero_start: error %d\n", error);
185	/*
186	 * We're an idle task, don't count us in the load.
187	 */
188	PROC_LOCK(pagezero_proc);
189	pagezero_proc->p_flag |= P_NOLOAD;
190	PROC_UNLOCK(pagezero_proc);
191	mtx_lock_spin(&sched_lock);
192	setrunqueue(FIRST_THREAD_IN_PROC(pagezero_proc), SRQ_BORING);
193	mtx_unlock_spin(&sched_lock);
194}
195SYSINIT(pagezero, SI_SUB_KTHREAD_VM, SI_ORDER_ANY, pagezero_start, NULL)
196