2 * Copyright (c) 2004 Robert N. M. Watson
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
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.
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
29 #include "opt_mp_watchdog.h"
30 #include "opt_sched.h"
33 #error MP_WATCHDOG cannot currently be used with SCHED_ULE
36 #include <sys/param.h>
38 #include <sys/kernel.h>
40 #include <sys/mutex.h>
43 #include <sys/sysctl.h>
44 #include <sys/systm.h>
46 #include <machine/smp.h>
47 #include <x86/apicreg.h>
48 #include <machine/apicvar.h>
49 #include <machine/mp_watchdog.h>
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.
63 * XXXRW: This should really use the watchdog(9)/watchdog(4) framework, but
66 static int watchdog_cpu = -1;
67 static int watchdog_dontfire = 1;
68 static int watchdog_timer = -1;
69 static int watchdog_nmi = 1;
71 TUNABLE_INT("debug.watchdog", &watchdog_cpu);
72 SYSCTL_INT(_debug, OID_AUTO, watchdog_nmi, CTLFLAG_RW, &watchdog_nmi, 0,
73 "IPI the boot processor with an NMI to enter the debugger");
75 static struct callout watchdog_callout;
77 static void watchdog_change(int wdcpu);
80 * Number of seconds before the watchdog will fire if the callout fails to
83 #define WATCHDOG_THRESHOLD 10
86 watchdog_init(void *arg)
89 callout_init(&watchdog_callout, CALLOUT_MPSAFE);
90 if (watchdog_cpu != -1)
91 watchdog_change(watchdog_cpu);
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
100 watchdog_function(void *arg)
104 * Since the timer ran, we must not be wedged. Acquire some critical
105 * locks to make sure. Then reset the timer.
108 watchdog_timer = WATCHDOG_THRESHOLD;
110 callout_reset(&watchdog_callout, 1 * hz, watchdog_function, NULL);
112 SYSINIT(watchdog_init, SI_SUB_DRIVERS, SI_ORDER_ANY, watchdog_init, NULL);
115 watchdog_change(int wdcpu)
118 if (wdcpu == -1 || wdcpu == 0xffffffff) {
120 * Disable the watchdog.
123 watchdog_dontfire = 1;
124 callout_stop(&watchdog_callout);
125 printf("watchdog stopped\n");
127 watchdog_timer = WATCHDOG_THRESHOLD;
128 watchdog_dontfire = 0;
129 watchdog_cpu = wdcpu;
130 callout_reset(&watchdog_callout, 1 * hz, watchdog_function,
136 * This sysctl sets which CPU is the watchdog CPU. Set to -1 or 0xffffffff
137 * to disable the watchdog.
140 sysctl_watchdog(SYSCTL_HANDLER_ARGS)
145 error = sysctl_handle_int(oidp, &temp, 0, req);
149 if (req->newptr != NULL)
150 watchdog_change(temp);
153 SYSCTL_PROC(_debug, OID_AUTO, watchdog, CTLTYPE_INT|CTLFLAG_RW, 0, 0,
154 sysctl_watchdog, "I", "");
157 * Drop into the debugger by sending an IPI NMI to the boot processor.
160 watchdog_ipi_nmi(void)
164 * Deliver NMI to the boot processor. Why not?
166 lapic_ipi_raw(APIC_DEST_DESTFLD | APIC_TRIGMOD_EDGE |
167 APIC_LEVEL_ASSERT | APIC_DESTMODE_PHY | APIC_DELMODE_NMI,
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
181 ap_watchdog(u_int cpuid)
183 char old_pcomm[MAXCOMLEN + 1];
186 if (watchdog_cpu != cpuid)
189 printf("watchdog started on cpu %d\n", cpuid);
191 bcopy(p->p_comm, old_pcomm, MAXCOMLEN + 1);
192 snprintf(p->p_comm, MAXCOMLEN + 1, "mp_watchdog cpu %d", cpuid);
194 DELAY(1000000); /* One second. */
195 if (watchdog_cpu != cpuid)
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;
206 kdb_enter(KDB_WHY_WATCHDOG, "mp_watchdog");
209 bcopy(old_pcomm, p->p_comm, MAXCOMLEN + 1);
210 printf("watchdog stopped on cpu %d\n", cpuid);