Merge git://git.kernel.org/pub/scm/linux/kernel/git/davem/sparc-2.6
[deliverable/linux.git] / arch / sparc / kernel / nmi.c
1 /* Pseudo NMI support on sparc64 systems.
2 *
3 * Copyright (C) 2009 David S. Miller <davem@davemloft.net>
4 *
5 * The NMI watchdog support and infrastructure is based almost
6 * entirely upon the x86 NMI support code.
7 */
8 #include <linux/kernel.h>
9 #include <linux/param.h>
10 #include <linux/init.h>
11 #include <linux/percpu.h>
12 #include <linux/nmi.h>
13 #include <linux/module.h>
14 #include <linux/kprobes.h>
15 #include <linux/kernel_stat.h>
16 #include <linux/slab.h>
17 #include <linux/kdebug.h>
18 #include <linux/delay.h>
19 #include <linux/smp.h>
20
21 #include <asm/ptrace.h>
22 #include <asm/local.h>
23 #include <asm/pcr.h>
24
25 /* We don't have a real NMI on sparc64, but we can fake one
26 * up using profiling counter overflow interrupts and interrupt
27 * levels.
28 *
29 * The profile overflow interrupts at level 15, so we use
30 * level 14 as our IRQ off level.
31 */
32
33 static int nmi_watchdog_active;
34 static int panic_on_timeout;
35
36 int nmi_usable;
37 EXPORT_SYMBOL_GPL(nmi_usable);
38
39 static unsigned int nmi_hz = HZ;
40
41 static DEFINE_PER_CPU(unsigned int, last_irq_sum);
42 static DEFINE_PER_CPU(local_t, alert_counter);
43 static DEFINE_PER_CPU(int, nmi_touch);
44
45 void touch_nmi_watchdog(void)
46 {
47 if (nmi_watchdog_active) {
48 int cpu;
49
50 for_each_present_cpu(cpu) {
51 if (per_cpu(nmi_touch, cpu) != 1)
52 per_cpu(nmi_touch, cpu) = 1;
53 }
54 }
55
56 touch_softlockup_watchdog();
57 }
58 EXPORT_SYMBOL(touch_nmi_watchdog);
59
60 static void die_nmi(const char *str, struct pt_regs *regs, int do_panic)
61 {
62 if (notify_die(DIE_NMIWATCHDOG, str, regs, 0,
63 pt_regs_trap_type(regs), SIGINT) == NOTIFY_STOP)
64 return;
65
66 console_verbose();
67 bust_spinlocks(1);
68
69 printk(KERN_EMERG "%s", str);
70 printk(" on CPU%d, ip %08lx, registers:\n",
71 smp_processor_id(), regs->tpc);
72 show_regs(regs);
73
74 bust_spinlocks(0);
75
76 if (do_panic || panic_on_oops)
77 panic("Non maskable interrupt");
78
79 local_irq_enable();
80 do_exit(SIGBUS);
81 }
82
83 notrace __kprobes void perfctr_irq(int irq, struct pt_regs *regs)
84 {
85 unsigned int sum, touched = 0;
86 int cpu = smp_processor_id();
87
88 clear_softint(1 << irq);
89 pcr_ops->write(PCR_PIC_PRIV);
90
91 local_cpu_data().__nmi_count++;
92
93 if (notify_die(DIE_NMI, "nmi", regs, 0,
94 pt_regs_trap_type(regs), SIGINT) == NOTIFY_STOP)
95 touched = 1;
96
97 sum = kstat_irqs_cpu(0, cpu);
98 if (__get_cpu_var(nmi_touch)) {
99 __get_cpu_var(nmi_touch) = 0;
100 touched = 1;
101 }
102 if (!touched && __get_cpu_var(last_irq_sum) == sum) {
103 local_inc(&__get_cpu_var(alert_counter));
104 if (local_read(&__get_cpu_var(alert_counter)) == 5 * nmi_hz)
105 die_nmi("BUG: NMI Watchdog detected LOCKUP",
106 regs, panic_on_timeout);
107 } else {
108 __get_cpu_var(last_irq_sum) = sum;
109 local_set(&__get_cpu_var(alert_counter), 0);
110 }
111 if (nmi_usable) {
112 write_pic(picl_value(nmi_hz));
113 pcr_ops->write(pcr_enable);
114 }
115 }
116
117 static inline unsigned int get_nmi_count(int cpu)
118 {
119 return cpu_data(cpu).__nmi_count;
120 }
121
122 static int endflag __initdata;
123
124 static __init void nmi_cpu_busy(void *data)
125 {
126 local_irq_enable_in_hardirq();
127 while (endflag == 0)
128 mb();
129 }
130
131 static void report_broken_nmi(int cpu, int *prev_nmi_count)
132 {
133 printk(KERN_CONT "\n");
134
135 printk(KERN_WARNING
136 "WARNING: CPU#%d: NMI appears to be stuck (%d->%d)!\n",
137 cpu, prev_nmi_count[cpu], get_nmi_count(cpu));
138
139 printk(KERN_WARNING
140 "Please report this to bugzilla.kernel.org,\n");
141 printk(KERN_WARNING
142 "and attach the output of the 'dmesg' command.\n");
143
144 nmi_usable = 0;
145 }
146
147 static void stop_watchdog(void *unused)
148 {
149 pcr_ops->write(PCR_PIC_PRIV);
150 }
151
152 static int __init check_nmi_watchdog(void)
153 {
154 unsigned int *prev_nmi_count;
155 int cpu, err;
156
157 prev_nmi_count = kmalloc(nr_cpu_ids * sizeof(unsigned int), GFP_KERNEL);
158 if (!prev_nmi_count) {
159 err = -ENOMEM;
160 goto error;
161 }
162
163 printk(KERN_INFO "Testing NMI watchdog ... ");
164
165 smp_call_function(nmi_cpu_busy, (void *)&endflag, 0);
166
167 for_each_possible_cpu(cpu)
168 prev_nmi_count[cpu] = get_nmi_count(cpu);
169 local_irq_enable();
170 mdelay((20 * 1000) / nmi_hz); /* wait 20 ticks */
171
172 for_each_online_cpu(cpu) {
173 if (get_nmi_count(cpu) - prev_nmi_count[cpu] <= 5)
174 report_broken_nmi(cpu, prev_nmi_count);
175 }
176 endflag = 1;
177 if (!nmi_usable) {
178 kfree(prev_nmi_count);
179 err = -ENODEV;
180 goto error;
181 }
182 printk("OK.\n");
183
184 nmi_hz = 1;
185
186 kfree(prev_nmi_count);
187 return 0;
188 error:
189 on_each_cpu(stop_watchdog, NULL, 1);
190 return err;
191 }
192
193 static void start_watchdog(void *unused)
194 {
195 pcr_ops->write(PCR_PIC_PRIV);
196 write_pic(picl_value(nmi_hz));
197
198 pcr_ops->write(pcr_enable);
199 }
200
201 void nmi_adjust_hz(unsigned int new_hz)
202 {
203 nmi_hz = new_hz;
204 on_each_cpu(start_watchdog, NULL, 1);
205 }
206 EXPORT_SYMBOL_GPL(nmi_adjust_hz);
207
208 int __init nmi_init(void)
209 {
210 nmi_usable = 1;
211
212 on_each_cpu(start_watchdog, NULL, 1);
213
214 return check_nmi_watchdog();
215 }
216
217 static int __init setup_nmi_watchdog(char *str)
218 {
219 if (!strncmp(str, "panic", 5))
220 panic_on_timeout = 1;
221
222 return 0;
223 }
224 __setup("nmi_watchdog=", setup_nmi_watchdog);
This page took 0.03519 seconds and 6 git commands to generate.