aboutsummaryrefslogtreecommitdiffstats
path: root/arch/mn10300/kernel/mn10300-watchdog.c
diff options
context:
space:
mode:
Diffstat (limited to 'arch/mn10300/kernel/mn10300-watchdog.c')
-rw-r--r--arch/mn10300/kernel/mn10300-watchdog.c100
1 files changed, 61 insertions, 39 deletions
diff --git a/arch/mn10300/kernel/mn10300-watchdog.c b/arch/mn10300/kernel/mn10300-watchdog.c
index f362d9d138f1..965dd61656c3 100644
--- a/arch/mn10300/kernel/mn10300-watchdog.c
+++ b/arch/mn10300/kernel/mn10300-watchdog.c
@@ -30,7 +30,7 @@
30static DEFINE_SPINLOCK(watchdog_print_lock); 30static DEFINE_SPINLOCK(watchdog_print_lock);
31static unsigned int watchdog; 31static unsigned int watchdog;
32static unsigned int watchdog_hz = 1; 32static unsigned int watchdog_hz = 1;
33unsigned int watchdog_alert_counter; 33unsigned int watchdog_alert_counter[NR_CPUS];
34 34
35EXPORT_SYMBOL(touch_nmi_watchdog); 35EXPORT_SYMBOL(touch_nmi_watchdog);
36 36
@@ -39,9 +39,6 @@ EXPORT_SYMBOL(touch_nmi_watchdog);
39 * is to check its timer makes IRQ counts. If they are not 39 * is to check its timer makes IRQ counts. If they are not
40 * changing then that CPU has some problem. 40 * changing then that CPU has some problem.
41 * 41 *
42 * as these watchdog NMI IRQs are generated on every CPU, we only
43 * have to check the current processor.
44 *
45 * since NMIs dont listen to _any_ locks, we have to be extremely 42 * since NMIs dont listen to _any_ locks, we have to be extremely
46 * careful not to rely on unsafe variables. The printk might lock 43 * careful not to rely on unsafe variables. The printk might lock
47 * up though, so we have to break up any console locks first ... 44 * up though, so we have to break up any console locks first ...
@@ -69,8 +66,8 @@ int __init check_watchdog(void)
69 66
70 printk(KERN_INFO "OK.\n"); 67 printk(KERN_INFO "OK.\n");
71 68
72 /* now that we know it works we can reduce NMI frequency to 69 /* now that we know it works we can reduce NMI frequency to something
73 * something more reasonable; makes a difference in some configs 70 * more reasonable; makes a difference in some configs
74 */ 71 */
75 watchdog_hz = 1; 72 watchdog_hz = 1;
76 73
@@ -121,15 +118,22 @@ void __init watchdog_go(void)
121 } 118 }
122} 119}
123 120
121#ifdef CONFIG_SMP
122static void watchdog_dump_register(void *dummy)
123{
124 printk(KERN_ERR "--- Register Dump (CPU%d) ---\n", CPUID);
125 show_registers(__frame);
126}
127#endif
128
124asmlinkage 129asmlinkage
125void watchdog_interrupt(struct pt_regs *regs, enum exception_code excep) 130void watchdog_interrupt(struct pt_regs *regs, enum exception_code excep)
126{ 131{
127
128 /* 132 /*
129 * Since current-> is always on the stack, and we always switch 133 * Since current-> is always on the stack, and we always switch
130 * the stack NMI-atomically, it's safe to use smp_processor_id(). 134 * the stack NMI-atomically, it's safe to use smp_processor_id().
131 */ 135 */
132 int sum, cpu = smp_processor_id(); 136 int sum, cpu;
133 int irq = NMIIRQ; 137 int irq = NMIIRQ;
134 u8 wdt, tmp; 138 u8 wdt, tmp;
135 139
@@ -138,43 +142,61 @@ void watchdog_interrupt(struct pt_regs *regs, enum exception_code excep)
138 tmp = WDCTR; 142 tmp = WDCTR;
139 NMICR = NMICR_WDIF; 143 NMICR = NMICR_WDIF;
140 144
141 nmi_count(cpu)++; 145 nmi_count(smp_processor_id())++;
142 kstat_incr_irqs_this_cpu(irq, irq_to_desc(irq)); 146 kstat_incr_irqs_this_cpu(irq, irq_to_desc(irq));
143 sum = irq_stat[cpu].__irq_count; 147
144 148 for_each_online_cpu(cpu) {
145 if (last_irq_sums[cpu] == sum) { 149
146 /* 150 sum = irq_stat[cpu].__irq_count;
147 * Ayiee, looks like this CPU is stuck ... 151
148 * wait a few IRQs (5 seconds) before doing the oops ... 152 if ((last_irq_sums[cpu] == sum)
149 */ 153#if defined(CONFIG_GDBSTUB) && defined(CONFIG_SMP)
150 watchdog_alert_counter++; 154 && !(CHK_GDBSTUB_BUSY()
151 if (watchdog_alert_counter == 5 * watchdog_hz) { 155 || atomic_read(&cpu_doing_single_step))
152 spin_lock(&watchdog_print_lock); 156#endif
157 ) {
153 /* 158 /*
154 * We are in trouble anyway, lets at least try 159 * Ayiee, looks like this CPU is stuck ...
155 * to get a message out. 160 * wait a few IRQs (5 seconds) before doing the oops ...
156 */ 161 */
157 bust_spinlocks(1); 162 watchdog_alert_counter[cpu]++;
158 printk(KERN_ERR 163 if (watchdog_alert_counter[cpu] == 5 * watchdog_hz) {
159 "NMI Watchdog detected LOCKUP on CPU%d," 164 spin_lock(&watchdog_print_lock);
160 " pc %08lx, registers:\n", 165 /*
161 cpu, regs->pc); 166 * We are in trouble anyway, lets at least try
162 show_registers(regs); 167 * to get a message out.
163 printk("console shuts up ...\n"); 168 */
164 console_silent(); 169 bust_spinlocks(1);
165 spin_unlock(&watchdog_print_lock); 170 printk(KERN_ERR
166 bust_spinlocks(0); 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);
167#ifdef CONFIG_GDBSTUB 188#ifdef CONFIG_GDBSTUB
168 if (gdbstub_busy) 189 if (CHK_GDBSTUB_BUSY_AND_ACTIVE())
169 gdbstub_exception(regs, excep); 190 gdbstub_exception(regs, excep);
170 else 191 else
171 gdbstub_intercept(regs, excep); 192 gdbstub_intercept(regs, excep);
172#endif 193#endif
173 do_exit(SIGSEGV); 194 do_exit(SIGSEGV);
195 }
196 } else {
197 last_irq_sums[cpu] = sum;
198 watchdog_alert_counter[cpu] = 0;
174 } 199 }
175 } else {
176 last_irq_sums[cpu] = sum;
177 watchdog_alert_counter = 0;
178 } 200 }
179 201
180 WDCTR = wdt | WDCTR_WDRST; 202 WDCTR = wdt | WDCTR_WDRST;