1 /* MN10300 Watchdog timer
2 *
3 * Copyright (C) 2007 Red Hat, Inc. All Rights Reserved.
4 * Written by David Howells (dhowells@redhat.com)
5 * - Derived from arch/i386/kernel/nmi.c
6 *
7 * This program is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU General Public Licence
9 * as published by the Free Software Foundation; either version
10 * 2 of the Licence, or (at your option) any later version.
11 */
12 #include <linux/module.h>
13 #include <linux/sched.h>
14 #include <linux/kernel.h>
15 #include <linux/init.h>
16 #include <linux/delay.h>
17 #include <linux/interrupt.h>
18 #include <linux/kernel_stat.h>
19 #include <linux/nmi.h>
20 #include <asm/processor.h>
21 #include <asm/system.h>
22 #include <asm/atomic.h>
23 #include <asm/intctl-regs.h>
24 #include <asm/rtc-regs.h>
25 #include <asm/div64.h>
26 #include <asm/smp.h>
27 #include <asm/gdb-stub.h>
28 #include <proc/clock.h>
29
30 static DEFINE_SPINLOCK(watchdog_print_lock);
31 static unsigned int watchdog;
32 static unsigned int watchdog_hz = 1;
33 unsigned int watchdog_alert_counter[NR_CPUS];
34
35 EXPORT_SYMBOL(touch_nmi_watchdog);
36
37 /*
38 * the best way to detect whether a CPU has a 'hard lockup' problem
39 * is to check its timer makes IRQ counts. If they are not
40 * changing then that CPU has some problem.
41 *
42 * since NMIs dont listen to _any_ locks, we have to be extremely
43 * careful not to rely on unsafe variables. The printk might lock
44 * up though, so we have to break up any console locks first ...
45 * [when there will be more tty-related locks, break them up
46 * here too!]
47 */
48 static unsigned int last_irq_sums[NR_CPUS];
49
check_watchdog(void)50 int __init check_watchdog(void)
51 {
52 irq_cpustat_t tmp[1];
53
54 printk(KERN_INFO "Testing Watchdog... ");
55
56 memcpy(tmp, irq_stat, sizeof(tmp));
57 local_irq_enable();
58 mdelay((10 * 1000) / watchdog_hz); /* wait 10 ticks */
59 local_irq_disable();
60
61 if (nmi_count(0) - tmp[0].__nmi_count <= 5) {
62 printk(KERN_WARNING "CPU#%d: Watchdog appears to be stuck!\n",
63 0);
64 return -1;
65 }
66
67 printk(KERN_INFO "OK.\n");
68
69 /* now that we know it works we can reduce NMI frequency to something
70 * more reasonable; makes a difference in some configs
71 */
72 watchdog_hz = 1;
73
74 return 0;
75 }
76
setup_watchdog(char * str)77 static int __init setup_watchdog(char *str)
78 {
79 unsigned tmp;
80 int opt;
81 u8 ctr;
82
83 get_option(&str, &opt);
84 if (opt != 1)
85 return 0;
86
87 watchdog = opt;
88 if (watchdog) {
89 set_intr_stub(EXCEP_WDT, watchdog_handler);
90 ctr = WDCTR_WDCK_65536th;
91 WDCTR = WDCTR_WDRST | ctr;
92 WDCTR = ctr;
93 tmp = WDCTR;
94
95 tmp = __muldiv64u(1 << (16 + ctr * 2), 1000000, MN10300_WDCLK);
96 tmp = 1000000000 / tmp;
97 watchdog_hz = (tmp + 500) / 1000;
98 }
99
100 return 1;
101 }
102
103 __setup("watchdog=", setup_watchdog);
104
watchdog_go(void)105 void __init watchdog_go(void)
106 {
107 u8 wdt;
108
109 if (watchdog) {
110 printk(KERN_INFO "Watchdog: running at %uHz\n", watchdog_hz);
111 wdt = WDCTR & ~WDCTR_WDCNE;
112 WDCTR = wdt | WDCTR_WDRST;
113 wdt = WDCTR;
114 WDCTR = wdt | WDCTR_WDCNE;
115 wdt = WDCTR;
116
117 check_watchdog();
118 }
119 }
120
121 #ifdef CONFIG_SMP
watchdog_dump_register(void * dummy)122 static void watchdog_dump_register(void *dummy)
123 {
124 printk(KERN_ERR "--- Register Dump (CPU%d) ---\n", CPUID);
125 show_registers(current_frame());
126 }
127 #endif
128
129 asmlinkage
watchdog_interrupt(struct pt_regs * regs,enum exception_code excep)130 void watchdog_interrupt(struct pt_regs *regs, enum exception_code excep)
131 {
132 /*
133 * Since current-> is always on the stack, and we always switch
134 * the stack NMI-atomically, it's safe to use smp_processor_id().
135 */
136 int sum, cpu;
137 int irq = NMIIRQ;
138 u8 wdt, tmp;
139
140 wdt = WDCTR & ~WDCTR_WDCNE;
141 WDCTR = wdt;
142 tmp = WDCTR;
143 NMICR = NMICR_WDIF;
144
145 nmi_count(smp_processor_id())++;
146 kstat_incr_irqs_this_cpu(irq, irq_to_desc(irq));
147
148 for_each_online_cpu(cpu) {
149
150 sum = irq_stat[cpu].__irq_count;
151
152 if ((last_irq_sums[cpu] == sum)
153 #if defined(CONFIG_GDBSTUB) && defined(CONFIG_SMP)
154 && !(CHK_GDBSTUB_BUSY()
155 || atomic_read(&cpu_doing_single_step))
156 #endif
157 ) {
158 /*
159 * Ayiee, looks like this CPU is stuck ...
160 * wait a few IRQs (5 seconds) before doing the oops ...
161 */
162 watchdog_alert_counter[cpu]++;
163 if (watchdog_alert_counter[cpu] == 5 * watchdog_hz) {
164 spin_lock(&watchdog_print_lock);
165 /*
166 * We are in trouble anyway, lets at least try
167 * to get a message out.
168 */
169 bust_spinlocks(1);
170 printk(KERN_ERR
171 "NMI Watchdog detected LOCKUP on CPU%d,"
172 " pc %08lx, registers:\n",
173 cpu, regs->pc);
174 #ifdef CONFIG_SMP
175 printk(KERN_ERR
176 "--- Register Dump (CPU%d) ---\n",
177 CPUID);
178 #endif
179 show_registers(regs);
180 #ifdef CONFIG_SMP
181 smp_nmi_call_function(watchdog_dump_register,
182 NULL, 1);
183 #endif
184 printk(KERN_NOTICE "console shuts up ...\n");
185 console_silent();
186 spin_unlock(&watchdog_print_lock);
187 bust_spinlocks(0);
188 #ifdef CONFIG_GDBSTUB
189 if (CHK_GDBSTUB_BUSY_AND_ACTIVE())
190 gdbstub_exception(regs, excep);
191 else
192 gdbstub_intercept(regs, excep);
193 #endif
194 do_exit(SIGSEGV);
195 }
196 } else {
197 last_irq_sums[cpu] = sum;
198 watchdog_alert_counter[cpu] = 0;
199 }
200 }
201
202 WDCTR = wdt | WDCTR_WDRST;
203 tmp = WDCTR;
204 WDCTR = wdt | WDCTR_WDCNE;
205 tmp = WDCTR;
206 }
207