sys/vm/vm_zeroidle.c

   1 /*
   2  * (MPSAFE)
   3  *
   4  * Copyright (c) 1994 John Dyson
   5  * Copyright (c) 2001 Matt Dillon
   6  * Copyright (c) 2010 The DragonFly Project
   7  *
   8  * All Rights Reserved.
   9  *
  10  * This code is derived from software contributed to The DragonFly Project
  11  * by Venkatesh Srinivas <me@endeavour.zapto.org>
  12  *
  13  * Redistribution and use in source and binary forms, with or without
  14  * modification, are permitted provided that the following conditions
  15  * are met:
  16  * 1. Redistributions of source code must retain the above copyright
  17  *    notice, this list of conditions and the following disclaimer.
  18  * 2. Redistributions in binary form must reproduce the above copyright
  19  *    notice, this list of conditions and the following disclaimer in the
  20  *    documentation and/or other materials provided with the distribution.
  21  * 3. Neither the name of the University nor the names of its contributors
  22  *    may be used to endorse or promote products derived from this software
  23  *    without specific prior written permission.
  24  *
  25  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS
  26  * OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
  27  * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  28  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY
  29  * DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  30  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE
  31  * GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
  32  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
  33  * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
  34  * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
  35  * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  36  *
  37  *      from: @(#)vm_machdep.c  7.3 (Berkeley) 5/13/91
  38  *      Utah $Hdr: vm_machdep.c 1.16.1.1 89/06/23$
  39  * from FreeBSD: .../i386/vm_machdep.c,v 1.165 2001/07/04 23:27:04 dillon
  40  */
  41
  42 #include <sys/param.h>
  43 #include <sys/systm.h>
  44 #include <sys/kernel.h>
  45 #include <sys/proc.h>
  46 #include <sys/vmmeter.h>
  47 #include <sys/sched.h>
  48 #include <sys/sysctl.h>
  49 #include <sys/thread.h>
  50 #include <sys/kthread.h>
  51 #include <sys/unistd.h>
  52 #include <vm/vm.h>
  53 #include <vm/vm_page.h>
  54 #include <cpu/lwbuf.h>
  55
  56 #include <sys/thread2.h>
  57 #include <vm/vm_page2.h>
  58
  59 /*
  60  * Implement the pre-zeroed page mechanism.
  61  */
  62 #define ZIDLE_LO(v)     ((v) * 2 / 3)
  63 #define ZIDLE_HI(v)     ((v) * 4 / 5)
  64
  65 /* Number of bytes to zero between reschedule checks */
  66 #define IDLEZERO_RUN    (64)
  67
  68 /* Maximum number of pages per second to zero */
  69 #define NPAGES_RUN      (20000)
  70
  71 static int idlezero_enable = 1;
  72 TUNABLE_INT("vm.idlezero_enable", &idlezero_enable);
  73 SYSCTL_INT(_vm, OID_AUTO, idlezero_enable, CTLFLAG_RW, &idlezero_enable, 0,
  74            "Allow the kernel to use idle CPU cycles to zero pages");
  75 static int idlezero_rate = NPAGES_RUN;
  76 SYSCTL_INT(_vm, OID_AUTO, idlezero_rate, CTLFLAG_RW, &idlezero_rate, 0,
  77            "Maximum pages per second to zero");
  78 static int idlezero_nocache = -1;
  79 SYSCTL_INT(_vm, OID_AUTO, idlezero_nocache, CTLFLAG_RW, &idlezero_nocache, 0,
  80            "Maximum pages per second to zero");
  81
  82 static int idlezero_count = 0;
  83 SYSCTL_INT(_vm, OID_AUTO, idlezero_count, CTLFLAG_RD, &idlezero_count, 0,
  84            "The number of physical pages prezeroed at idle time");
  85
  86 enum zeroidle_state {
  87         STATE_IDLE,
  88         STATE_GET_PAGE,
  89         STATE_ZERO_PAGE,
  90         STATE_RELEASE_PAGE
  91 };
  92
  93 #define DEFAULT_SLEEP_TIME      (hz / 10)
  94 #define LONG_SLEEP_TIME         (hz * 10)
  95
  96 static int zero_state;
  97
  98 /*
  99  * Attempt to maintain approximately 1/2 of our free pages in a
 100  * PG_ZERO'd state. Add some hysteresis to (attempt to) avoid
 101  * generally zeroing a page when the system is near steady-state.
 102  * Otherwise we might get 'flutter' during disk I/O / IPC or
 103  * fast sleeps. We also do not want to be continuously zeroing
 104  * pages because doing so may flush our L1 and L2 caches too much.
 105  *
 106  * Returns non-zero if pages should be zerod.
 107  */
 108 static int
 109 vm_page_zero_check(void)
 110 {
 111         if (idlezero_enable == 0)
 112                 return (0);
 113         if (zero_state == 0) {
 114                 /*
 115                  * Wait for the count to fall to LO before starting
 116                  * to zero pages.
 117                  */
 118                 if (vm_page_zero_count <= ZIDLE_LO(vmstats.v_free_count))
 119                         zero_state = 1;
 120         } else {
 121                 /*
 122                  * Once we are zeroing pages wait for the count to
 123                  * increase to HI before we stop zeroing pages.
 124                  */
 125                 if (vm_page_zero_count >= ZIDLE_HI(vmstats.v_free_count))
 126                         zero_state = 0;
 127         }
 128         return (zero_state);
 129 }
 130
 131 /*
 132  * vm_pagezero should sleep for a longer time when idlezero is disabled or
 133  * when there is an excess of zeroed pages.
 134  */
 135 static int
 136 vm_page_zero_time(void)
 137 {
 138         if (idlezero_enable == 0)
 139                 return (LONG_SLEEP_TIME);
 140         if (vm_page_zero_count >= ZIDLE_HI(vmstats.v_free_count))
 141                 return (LONG_SLEEP_TIME);
 142         return (DEFAULT_SLEEP_TIME);
 143 }
 144
 145 /*
 146  * MPSAFE thread
 147  */
 148 static void
 149 vm_pagezero(void __unused *arg)
 150 {
 151         vm_page_t m = NULL;
 152         struct lwbuf *lwb = NULL;
 153         struct lwbuf lwb_cache;
 154         enum zeroidle_state state = STATE_IDLE;
 155         char *pg = NULL;
 156         int npages = 0;
 157         int sleep_time;
 158         int i = 0;
 159
 160         /*
 161          * Adjust thread parameters before entering our loop.  The thread
 162          * is started with the MP lock held and with normal kernel thread
 163          * priority.
 164          *
 165          * Also put us on the last cpu for now.
 166          *
 167          * For now leave the MP lock held, the VM routines cannot be called
 168          * with it released until tokenization is finished.
 169          */
 170         lwkt_setpri_self(TDPRI_IDLE_WORK);
 171         lwkt_setcpu_self(globaldata_find(ncpus - 1));
 172         sleep_time = DEFAULT_SLEEP_TIME;
 173
 174         /*
 175          * Loop forever
 176          */
 177         for (;;) {
 178                 switch(state) {
 179                 case STATE_IDLE:
 180                         /*
 181                          * Wait for work.
 182                          */
 183                         tsleep(&zero_state, 0, "pgzero", sleep_time);
 184                         if (vm_page_zero_check())
 185                                 npages = idlezero_rate / 10;
 186                         sleep_time = vm_page_zero_time();
 187                         if (npages)
 188                                 state = STATE_GET_PAGE; /* Fallthrough */
 189                         break;
 190                 case STATE_GET_PAGE:
 191                         /*
 192                          * Acquire page to zero
 193                          */
 194                         if (--npages == 0) {
 195                                 state = STATE_IDLE;
 196                         } else {
 197                                 m = vm_page_free_fromq_fast();
 198                                 if (m == NULL) {
 199                                         state = STATE_IDLE;
 200                                 } else {
 201                                         state = STATE_ZERO_PAGE;
 202                                         lwb = lwbuf_alloc(m, &lwb_cache);
 203                                         pg = (char *)lwbuf_kva(lwb);
 204                                         i = 0;
 205                                 }
 206                         }
 207                         break;
 208                 case STATE_ZERO_PAGE:
 209                         /*
 210                          * Zero-out the page
 211                          */
 212                         while (i < PAGE_SIZE) {
 213                                 if (idlezero_nocache == 1)
 214                                         bzeront(&pg[i], IDLEZERO_RUN);
 215                                 else
 216                                         bzero(&pg[i], IDLEZERO_RUN);
 217                                 i += IDLEZERO_RUN;
 218                                 lwkt_yield();
 219                         }
 220                         state = STATE_RELEASE_PAGE;
 221                         break;
 222                 case STATE_RELEASE_PAGE:
 223                         lwbuf_free(lwb);
 224                         vm_page_flag_set(m, PG_ZERO);
 225                         vm_page_free_toq(m);
 226                         state = STATE_GET_PAGE;
 227                         ++idlezero_count;
 228                         break;
 229                 }
 230                 lwkt_yield();
 231         }
 232 }
 233
 234 static void
 235 pagezero_start(void __unused *arg)
 236 {
 237         int error;
 238         struct thread *td;
 239
 240         if (idlezero_nocache < 0 && (cpu_mi_feature & CPU_MI_BZERONT))
 241                 idlezero_nocache = 1;
 242
 243         error = kthread_create(vm_pagezero, NULL, &td, "pagezero");
 244         if (error)
 245                 panic("pagezero_start: error %d", error);
 246 }
 247
 248 SYSINIT(pagezero, SI_SUB_KTHREAD_VM, SI_ORDER_ANY, pagezero_start, NULL);