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 <machine/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 */ 66static int watchdog_cpu = -1; 67static int watchdog_dontfire = 1; 68static int watchdog_timer = -1; 69static int watchdog_nmi = 1; 70 71TUNABLE_INT("debug.watchdog", &watchdog_cpu); 72SYSCTL_INT(_debug, OID_AUTO, watchdog_nmi, CTLFLAG_RW, &watchdog_nmi, 0, 73 "IPI the boot processor with an NMI to enter the debugger"); 74 75static struct callout watchdog_callout; 76 77static void watchdog_change(int wdcpu); 78 79/* 80 * Number of seconds before the watchdog will fire if the callout fails to 81 * reset the timer. 82 */ 83#define WATCHDOG_THRESHOLD 10 84 85static void 86watchdog_init(void *arg) 87{ 88 89 callout_init(&watchdog_callout, CALLOUT_MPSAFE); 90 if (watchdog_cpu != -1) 91 watchdog_change(watchdog_cpu); 92} 93 94/* 95 * This callout resets a timer until the watchdog kicks in. It acquires some 96 * critical locks to make sure things haven't gotten wedged with hose locks 97 * held. 98 */ 99static void 100watchdog_function(void *arg) 101{ 102 103 /* 104 * Since the timer ran, we must not be wedged. Acquire some critical 105 * locks to make sure. Then reset the timer. 106 */ 107 mtx_lock(&Giant); 108 watchdog_timer = WATCHDOG_THRESHOLD; 109 mtx_unlock(&Giant); 110 callout_reset(&watchdog_callout, 1 * hz, watchdog_function, NULL); 111} 112SYSINIT(watchdog_init, SI_SUB_DRIVERS, SI_ORDER_ANY, watchdog_init, NULL); 113 114static void 115watchdog_change(int wdcpu) 116{ 117 118 if (wdcpu == -1 || wdcpu == 0xffffffff) { 119 /* 120 * Disable the watchdog. 121 */ 122 watchdog_cpu = -1; 123 watchdog_dontfire = 1; 124 callout_stop(&watchdog_callout); 125 printf("watchdog stopped\n"); 126 } else { 127 watchdog_timer = WATCHDOG_THRESHOLD; 128 watchdog_dontfire = 0; 129 watchdog_cpu = wdcpu; 130 callout_reset(&watchdog_callout, 1 * hz, watchdog_function, 131 NULL); 132 } 133} 134 135/* 136 * This sysctl sets which CPU is the watchdog CPU. Set to -1 or 0xffffffff 137 * to disable the watchdog. 138 */ 139static int 140sysctl_watchdog(SYSCTL_HANDLER_ARGS) 141{ 142 int error, temp; 143 144 temp = watchdog_cpu; 145 error = sysctl_handle_int(oidp, &temp, 0, req); 146 if (error) 147 return (error); 148 149 if (req->newptr != NULL) 150 watchdog_change(temp); 151 return (0); 152} 153SYSCTL_PROC(_debug, OID_AUTO, watchdog, CTLTYPE_INT|CTLFLAG_RW, 0, 0, 154 sysctl_watchdog, "I", ""); 155 156/* 157 * Drop into the debugger by sending an IPI NMI to the boot processor. 158 */ 159static void 160watchdog_ipi_nmi(void) 161{ 162 163 /* 164 * Deliver NMI to the boot processor. Why not? 165 */ 166 lapic_ipi_raw(APIC_DEST_DESTFLD | APIC_TRIGMOD_EDGE | 167 APIC_LEVEL_ASSERT | APIC_DESTMODE_PHY | APIC_DELMODE_NMI, 168 boot_cpu_id); 169 lapic_ipi_wait(-1); 170} 171 172/* 173 * ap_watchdog() is called by the SMP idle loop code. It works on the same 174 * premise that the disabling of logical processors does: that if the cpu is 175 * idle, then it can ignore the world from then on, as nothing will be 176 * scheduled on it. Leaving aside multi-runqueue schedulers (SCHED_ULE) and 177 * explicit process migration (sched_bind()), this is not an unreasonable 178 * assumption. 179 */ 180void 181ap_watchdog(u_int cpuid) 182{ 183 char old_pcomm[MAXCOMLEN + 1]; 184 struct proc *p; 185 186 if (watchdog_cpu != cpuid) 187 return; 188 189 printf("watchdog started on cpu %d\n", cpuid); 190 p = curproc; 191 bcopy(p->p_comm, old_pcomm, MAXCOMLEN + 1); 192 snprintf(p->p_comm, MAXCOMLEN + 1, "mp_watchdog cpu %d", cpuid); 193 while (1) { 194 DELAY(1000000); /* One second. */ 195 if (watchdog_cpu != cpuid) 196 break; 197 atomic_subtract_int(&watchdog_timer, 1); 198 if (watchdog_timer < 4) 199 printf("Watchdog timer: %d\n", watchdog_timer); 200 if (watchdog_timer == 0 && watchdog_dontfire == 0) { 201 printf("Watchdog firing!\n"); 202 watchdog_dontfire = 1; 203 if (watchdog_nmi) 204 watchdog_ipi_nmi(); 205 else 206 kdb_enter(KDB_WHY_WATCHDOG, "mp_watchdog"); 207 } 208 } 209 bcopy(old_pcomm, p->p_comm, MAXCOMLEN + 1); 210 printf("watchdog stopped on cpu %d\n", cpuid); 211} 212