sys/i386/i386/mp_watchdog.c

   1 /*-
   2  * Copyright (c) 2004 Robert N. M. Watson
   3  * All rights reserved.
   4  *
   5  * Redistribution and use in source and binary forms, with or without
   6  * modification, are permitted provided that the following conditions
   7  * are met:
   8  * 1. Redistributions of source code must retain the above copyright
   9  *    notice, this list of conditions and the following disclaimer.
  10  * 2. Redistributions in binary form must reproduce the above copyright
  11  *    notice, this list of conditions and the following disclaimer in the
  12  *    documentation and/or other materials provided with the distribution.
  13  *
  14  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
  15  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  16  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  17  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
  18  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  19  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  20  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  21  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  22  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  23  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  24  * SUCH DAMAGE.
  25  *
  26  * $FreeBSD$
  27  */
  28
  29 #include "opt_mp_watchdog.h"
  30 #include "opt_sched.h"
  31
  32 #ifdef SCHED_ULE
  33 #error MP_WATCHDOG cannot currently be used with SCHED_ULE
  34 #endif
  35
  36 #include <sys/param.h>
  37 #include <sys/kdb.h>
  38 #include <sys/kernel.h>
  39 #include <sys/lock.h>
  40 #include <sys/mutex.h>
  41 #include <sys/pcpu.h>
  42 #include <sys/proc.h>
  43 #include <sys/sysctl.h>
  44 #include <sys/systm.h>
  45
  46 #include <machine/smp.h>
  47 #include <x86/apicreg.h>
  48 #include <x86/apicvar.h>
  49 #include <machine/mp_watchdog.h>
  50
  51 /*
  52  * mp_watchdog hijacks the idle thread on a specified CPU, prevents new work
  53  * from being scheduled there, and uses it as a "watchdog" to detect kernel
  54  * failure on other CPUs.  This is made reasonable by inclusion of logical
  55  * processors in Xeon hardware.  The watchdog is configured by setting the
  56  * debug.watchdog sysctl/tunable to the CPU of interest.  A callout will then
  57  * begin executing reseting a timer that is gradually lowered by the watching
  58  * thread.  If the timer reaches 0, the watchdog fires by ether dropping
  59  * directly to the debugger, or by sending an NMI IPI to the boot processor.
  60  * This is a somewhat less efficient substitute for dedicated watchdog
  61  * hardware, but can be quite an effective tool for debugging hangs.
  62  *
  63  * XXXRW: This should really use the watchdog(9)/watchdog(4) framework, but
  64  * doesn't yet.
  65  */
  66 static int      watchdog_cpu = -1;
  67 static int      watchdog_dontfire = 1;
  68 static int      watchdog_timer = -1;
  69 static int      watchdog_nmi = 1;
  70
  71 SYSCTL_INT(_debug, OID_AUTO, watchdog_nmi, CTLFLAG_RWTUN, &watchdog_nmi, 0,
  72     "IPI the boot processor with an NMI to enter the debugger");
  73
  74 static struct callout   watchdog_callout;
  75
  76 static void watchdog_change(int wdcpu);
  77
  78 /*
  79  * Number of seconds before the watchdog will fire if the callout fails to
  80  * reset the timer.
  81  */
  82 #define WATCHDOG_THRESHOLD      10
  83
  84 static void
  85 watchdog_init(void *arg)
  86 {
  87
  88         callout_init(&watchdog_callout, 1);
  89         if (watchdog_cpu != -1)
  90                 watchdog_change(watchdog_cpu);
  91 }
  92
  93 /*
  94  * This callout resets a timer until the watchdog kicks in.  It acquires some
  95  * critical locks to make sure things haven't gotten wedged with hose locks
  96  * held.
  97  */
  98 static void
  99 watchdog_function(void *arg)
 100 {
 101
 102         /*
 103          * Since the timer ran, we must not be wedged.  Acquire some critical
 104          * locks to make sure.  Then reset the timer.
 105          */
 106         mtx_lock(&Giant);
 107         watchdog_timer = WATCHDOG_THRESHOLD;
 108         mtx_unlock(&Giant);
 109         callout_reset(&watchdog_callout, 1 * hz, watchdog_function, NULL);
 110 }
 111 SYSINIT(watchdog_init, SI_SUB_DRIVERS, SI_ORDER_ANY, watchdog_init, NULL);
 112
 113 static void
 114 watchdog_change(int wdcpu)
 115 {
 116
 117         if (wdcpu == -1 || wdcpu == 0xffffffff) {
 118                 /*
 119                  * Disable the watchdog.
 120                  */
 121                 watchdog_cpu = -1;
 122                 watchdog_dontfire = 1;
 123                 callout_stop(&watchdog_callout);
 124                 printf("watchdog stopped\n");
 125         } else {
 126                 watchdog_timer = WATCHDOG_THRESHOLD;
 127                 watchdog_dontfire = 0;
 128                 watchdog_cpu = wdcpu;
 129                 callout_reset(&watchdog_callout, 1 * hz, watchdog_function,
 130                     NULL);
 131         }
 132 }
 133
 134 /*
 135  * This sysctl sets which CPU is the watchdog CPU.  Set to -1 or 0xffffffff
 136  * to disable the watchdog.
 137  */
 138 static int
 139 sysctl_watchdog(SYSCTL_HANDLER_ARGS)
 140 {
 141         int error, temp;
 142
 143         temp = watchdog_cpu;
 144         error = sysctl_handle_int(oidp, &temp, 0, req);
 145         if (error)
 146                 return (error);
 147
 148         if (req->newptr != NULL)
 149                 watchdog_change(temp);
 150         return (0);
 151 }
 152 SYSCTL_PROC(_debug, OID_AUTO, watchdog, CTLTYPE_INT|CTLFLAG_RW, 0, 0,
 153     sysctl_watchdog, "I", "");
 154
 155 /*
 156  * Drop into the debugger by sending an IPI NMI to the boot processor.
 157  */
 158 static void
 159 watchdog_ipi_nmi(void)
 160 {
 161
 162         /*
 163          * Deliver NMI to the boot processor.  Why not?
 164          */
 165         lapic_ipi_raw(APIC_DEST_DESTFLD | APIC_TRIGMOD_EDGE |
 166             APIC_LEVEL_ASSERT | APIC_DESTMODE_PHY | APIC_DELMODE_NMI,
 167             boot_cpu_id);
 168         lapic_ipi_wait(-1);
 169 }
 170
 171 /*
 172  * ap_watchdog() is called by the SMP idle loop code.  It works on the same
 173  * premise that the disabling of logical processors does: that if the cpu is
 174  * idle, then it can ignore the world from then on, as nothing will be
 175  * scheduled on it.  Leaving aside multi-runqueue schedulers (SCHED_ULE) and
 176  * explicit process migration (sched_bind()), this is not an unreasonable
 177  * assumption.
 178  */
 179 void
 180 ap_watchdog(u_int cpuid)
 181 {
 182         char old_pcomm[MAXCOMLEN + 1];
 183         struct proc *p;
 184
 185         if (watchdog_cpu != cpuid)
 186                 return;
 187
 188         printf("watchdog started on cpu %d\n", cpuid);
 189         p = curproc;
 190         bcopy(p->p_comm, old_pcomm, MAXCOMLEN + 1);
 191         snprintf(p->p_comm, MAXCOMLEN + 1, "mp_watchdog cpu %d", cpuid);
 192         while (1) {
 193                 DELAY(1000000);                         /* One second. */
 194                 if (watchdog_cpu != cpuid)
 195                         break;
 196                 atomic_subtract_int(&watchdog_timer, 1);
 197                 if (watchdog_timer < 4)
 198                         printf("Watchdog timer: %d\n", watchdog_timer);
 199                 if (watchdog_timer == 0 && watchdog_dontfire == 0) {
 200                         printf("Watchdog firing!\n");
 201                         watchdog_dontfire = 1;
 202                         if (watchdog_nmi)
 203                                 watchdog_ipi_nmi();
 204                         else
 205                                 kdb_enter(KDB_WHY_WATCHDOG, "mp_watchdog");
 206                 }
 207         }
 208         bcopy(old_pcomm, p->p_comm, MAXCOMLEN + 1);
 209         printf("watchdog stopped on cpu %d\n", cpuid);
 210 }