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