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: releng/5.3/sys/i386/i386/mp_watchdog.c 133770 2004-08-15 20:32:40Z rwatson $
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 <machine/apicreg.h>
48 #include <machine/apicvar.h>
49 #include <machine/mp_watchdog.h>
50
51 /*
52 * mp_swatchdog 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_cpu sysctl 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_RW, &watchdog_nmi, 0,
72 "IPI the boot processor with an NMI to enter the debugger");
73
74 static struct callout watchdog_callout;
75
76 /*
77 * Number of seconds before the watchdog will fire if the callout fails to
78 * reset the timer.
79 */
80 #define WATCHDOG_THRESHOLD 10
81
82 static void
83 watchdog_init(void *arg)
84 {
85
86 callout_init(&watchdog_callout, CALLOUT_MPSAFE);
87 }
88
89 /*
90 * This callout resets a timer until the watchdog kicks in. It acquires some
91 * critical locks to make sure things haven't gotten wedged with hose locks
92 * held.
93 */
94 static void
95 watchdog_function(void *arg)
96 {
97
98 /*
99 * Since the timer ran, we must not be wedged. Acquire some critical
100 * locks to make sure. Then reset the timer.
101 */
102 mtx_lock(&Giant);
103 mtx_lock_spin(&sched_lock);
104 watchdog_timer = WATCHDOG_THRESHOLD;
105 mtx_unlock_spin(&sched_lock);
106 mtx_unlock(&Giant);
107 callout_reset(&watchdog_callout, 1 * hz, watchdog_function, NULL);
108 }
109 SYSINIT(watchdog_init, SI_SUB_DRIVERS, SI_ORDER_ANY, watchdog_init, NULL);
110
111 /*
112 * This sysctl sets which CPU is the watchdog CPU. Set to -1 or 0xffffffff
113 * to disable the watchdog.
114 */
115 static int
116 sysctl_watchdog(SYSCTL_HANDLER_ARGS)
117 {
118 int error, temp;
119
120 temp = watchdog_cpu;
121 error = sysctl_handle_int(oidp, &temp, 0, req);
122 if (error)
123 return (error);
124
125 if (req->newptr != NULL) {
126 if (temp == -1 || temp == 0xffffffff) {
127 /*
128 * Disable the watcdog.
129 */
130 watchdog_cpu = -1;
131 watchdog_dontfire = 1;
132 callout_stop(&watchdog_callout);
133 printf("watchdog stopped\n");
134 } else {
135 watchdog_timer = WATCHDOG_THRESHOLD;
136 watchdog_dontfire = 0;
137 watchdog_cpu = temp;
138 callout_reset(&watchdog_callout, 1 * hz,
139 watchdog_function, NULL);
140 }
141 }
142 return (0);
143 }
144 SYSCTL_PROC(_debug, OID_AUTO, watchdog, CTLTYPE_INT|CTLFLAG_RW, 0, 0,
145 sysctl_watchdog, "IU", "");
146
147 /*
148 * A badly behaved sysctl that leaks the sched lock when written to. Then
149 * spin holding it just to make matters worse. This can be used to test the
150 * effectiveness of the watchdog by generating a fairly hard and nast hang.
151 * Note that Giant is also held in the current world order when we get here.
152 */
153 static int
154 sysctl_leak_schedlock(SYSCTL_HANDLER_ARGS)
155 {
156 int error, temp;
157
158 temp = 0;
159 error = sysctl_handle_int(oidp, &temp, 0, req);
160 if (error)
161 return (error);
162
163 if (req->newptr != NULL) {
164 if (temp) {
165 printf("Leaking the sched lock...\n");
166 mtx_lock_spin(&sched_lock);
167 while (1);
168 }
169 }
170 return (0);
171 }
172 SYSCTL_PROC(_debug, OID_AUTO, leak_schedlock, CTLTYPE_INT|CTLFLAG_RW, 0, 0,
173 sysctl_leak_schedlock, "IU", "");
174
175 /*
176 * Drop into the debugger by sending an IPI NMI to the boot processor.
177 */
178 static void
179 watchdog_ipi_nmi(void)
180 {
181
182 /*
183 * Deliver NMI to the boot processor. Why not?
184 */
185 lapic_ipi_raw(APIC_DEST_DESTFLD | APIC_TRIGMOD_EDGE |
186 APIC_LEVEL_ASSERT | APIC_DESTMODE_PHY | APIC_DELMODE_NMI,
187 boot_cpu_id);
188 lapic_ipi_wait(-1);
189 }
190
191 /*
192 * ap_watchdog() is called by the SMP idle loop code. It works on the same
193 * premise that the disabling of logical processors does: that if the cpu is
194 * idle, then it can ignore the world from then on, as nothing will be
195 * scheduled on it. Leaving aside multi-runqueue schedulers (SCHED_ULE) and
196 * explicit process migration (sched_bind()), this is not an unreasonable
197 * assumption.
198 */
199 void
200 ap_watchdog(u_int cpuid)
201 {
202 char old_pcomm[MAXCOMLEN + 1];
203 struct proc *p;
204
205 if (watchdog_cpu != cpuid)
206 return;
207
208 printf("watchdog started on cpu %d\n", cpuid);
209 p = curproc;
210 bcopy(p->p_comm, old_pcomm, MAXCOMLEN + 1);
211 snprintf(p->p_comm, MAXCOMLEN + 1, "mp_watchdog cpu %d", cpuid);
212 while (1) {
213 DELAY(1000000); /* One second. */
214 if (watchdog_cpu != cpuid)
215 break;
216 atomic_subtract_int(&watchdog_timer, 1);
217 if (watchdog_timer < 4)
218 printf("Watchdog timer: %d\n", watchdog_timer);
219 if (watchdog_timer == 0 && watchdog_dontfire == 0) {
220 printf("Watchdog firing!\n");
221 watchdog_dontfire = 1;
222 if (watchdog_nmi)
223 watchdog_ipi_nmi();
224 else
225 kdb_enter("mp_watchdog");
226 }
227 }
228 bcopy(old_pcomm, p->p_comm, MAXCOMLEN + 1);
229 printf("watchdog stopped on cpu %d\n", cpuid);
230 }
Cache object: 29bf5ffa62c53cc466994a7addbe21ee
|