diff options
-rw-r--r-- | arch/x86/kernel/Makefile | 2 | ||||
-rw-r--r-- | arch/x86/kernel/cpu/perfctr-watchdog.c | 4 | ||||
-rw-r--r-- | arch/x86/kernel/nmi.c (renamed from arch/x86/kernel/nmi_64.c) | 207 | ||||
-rw-r--r-- | arch/x86/kernel/nmi_32.c | 476 | ||||
-rw-r--r-- | arch/x86/kernel/traps_32.c | 8 | ||||
-rw-r--r-- | arch/x86/kernel/traps_64.c | 28 | ||||
-rw-r--r-- | include/asm-x86/nmi.h | 6 |
7 files changed, 166 insertions, 565 deletions
diff --git a/arch/x86/kernel/Makefile b/arch/x86/kernel/Makefile index d1d4ee895270..8ef4dbf32cf9 100644 --- a/arch/x86/kernel/Makefile +++ b/arch/x86/kernel/Makefile | |||
@@ -52,7 +52,7 @@ obj-$(CONFIG_X86_32_SMP) += smpcommon.o | |||
52 | obj-$(CONFIG_X86_64_SMP) += tsc_sync.o smpcommon.o | 52 | obj-$(CONFIG_X86_64_SMP) += tsc_sync.o smpcommon.o |
53 | obj-$(CONFIG_X86_TRAMPOLINE) += trampoline_$(BITS).o | 53 | obj-$(CONFIG_X86_TRAMPOLINE) += trampoline_$(BITS).o |
54 | obj-$(CONFIG_X86_MPPARSE) += mpparse.o | 54 | obj-$(CONFIG_X86_MPPARSE) += mpparse.o |
55 | obj-$(CONFIG_X86_LOCAL_APIC) += apic_$(BITS).o nmi_$(BITS).o | 55 | obj-$(CONFIG_X86_LOCAL_APIC) += apic_$(BITS).o nmi.o |
56 | obj-$(CONFIG_X86_IO_APIC) += io_apic_$(BITS).o | 56 | obj-$(CONFIG_X86_IO_APIC) += io_apic_$(BITS).o |
57 | obj-$(CONFIG_X86_REBOOTFIXUPS) += reboot_fixups_32.o | 57 | obj-$(CONFIG_X86_REBOOTFIXUPS) += reboot_fixups_32.o |
58 | obj-$(CONFIG_KEXEC) += machine_kexec_$(BITS).o | 58 | obj-$(CONFIG_KEXEC) += machine_kexec_$(BITS).o |
diff --git a/arch/x86/kernel/cpu/perfctr-watchdog.c b/arch/x86/kernel/cpu/perfctr-watchdog.c index f9ae93adffe5..ddda4b64f545 100644 --- a/arch/x86/kernel/cpu/perfctr-watchdog.c +++ b/arch/x86/kernel/cpu/perfctr-watchdog.c | |||
@@ -181,7 +181,9 @@ void disable_lapic_nmi_watchdog(void) | |||
181 | return; | 181 | return; |
182 | 182 | ||
183 | on_each_cpu(stop_apic_nmi_watchdog, NULL, 0, 1); | 183 | on_each_cpu(stop_apic_nmi_watchdog, NULL, 0, 1); |
184 | wd_ops->unreserve(); | 184 | |
185 | if (wd_ops) | ||
186 | wd_ops->unreserve(); | ||
185 | 187 | ||
186 | BUG_ON(atomic_read(&nmi_active) != 0); | 188 | BUG_ON(atomic_read(&nmi_active) != 0); |
187 | } | 189 | } |
diff --git a/arch/x86/kernel/nmi_64.c b/arch/x86/kernel/nmi.c index d62f3b66b529..9ebf71323c9a 100644 --- a/arch/x86/kernel/nmi_64.c +++ b/arch/x86/kernel/nmi.c | |||
@@ -6,10 +6,13 @@ | |||
6 | * Fixes: | 6 | * Fixes: |
7 | * Mikael Pettersson : AMD K7 support for local APIC NMI watchdog. | 7 | * Mikael Pettersson : AMD K7 support for local APIC NMI watchdog. |
8 | * Mikael Pettersson : Power Management for local APIC NMI watchdog. | 8 | * Mikael Pettersson : Power Management for local APIC NMI watchdog. |
9 | * Mikael Pettersson : Pentium 4 support for local APIC NMI watchdog. | ||
9 | * Pavel Machek and | 10 | * Pavel Machek and |
10 | * Mikael Pettersson : PM converted to driver model. Disable/enable API. | 11 | * Mikael Pettersson : PM converted to driver model. Disable/enable API. |
11 | */ | 12 | */ |
12 | 13 | ||
14 | #include <asm/apic.h> | ||
15 | |||
13 | #include <linux/nmi.h> | 16 | #include <linux/nmi.h> |
14 | #include <linux/mm.h> | 17 | #include <linux/mm.h> |
15 | #include <linux/delay.h> | 18 | #include <linux/delay.h> |
@@ -17,22 +20,26 @@ | |||
17 | #include <linux/module.h> | 20 | #include <linux/module.h> |
18 | #include <linux/sysdev.h> | 21 | #include <linux/sysdev.h> |
19 | #include <linux/sysctl.h> | 22 | #include <linux/sysctl.h> |
23 | #include <linux/percpu.h> | ||
20 | #include <linux/kprobes.h> | 24 | #include <linux/kprobes.h> |
21 | #include <linux/cpumask.h> | 25 | #include <linux/cpumask.h> |
26 | #include <linux/kernel_stat.h> | ||
22 | #include <linux/kdebug.h> | 27 | #include <linux/kdebug.h> |
28 | #include <linux/smp.h> | ||
23 | 29 | ||
24 | #include <asm/i8259.h> | 30 | #include <asm/i8259.h> |
25 | #include <asm/io_apic.h> | 31 | #include <asm/io_apic.h> |
26 | #include <asm/smp.h> | 32 | #include <asm/smp.h> |
27 | #include <asm/nmi.h> | 33 | #include <asm/nmi.h> |
28 | #include <asm/proto.h> | 34 | #include <asm/proto.h> |
35 | #include <asm/timer.h> | ||
36 | |||
29 | #include <asm/mce.h> | 37 | #include <asm/mce.h> |
30 | 38 | ||
31 | #include <mach_traps.h> | 39 | #include <mach_traps.h> |
32 | 40 | ||
33 | int unknown_nmi_panic; | 41 | int unknown_nmi_panic; |
34 | int nmi_watchdog_enabled; | 42 | int nmi_watchdog_enabled; |
35 | int panic_on_unrecovered_nmi; | ||
36 | 43 | ||
37 | static cpumask_t backtrace_mask = CPU_MASK_NONE; | 44 | static cpumask_t backtrace_mask = CPU_MASK_NONE; |
38 | 45 | ||
@@ -43,12 +50,47 @@ static cpumask_t backtrace_mask = CPU_MASK_NONE; | |||
43 | * 0: the lapic NMI watchdog is disabled, but can be enabled | 50 | * 0: the lapic NMI watchdog is disabled, but can be enabled |
44 | */ | 51 | */ |
45 | atomic_t nmi_active = ATOMIC_INIT(0); /* oprofile uses this */ | 52 | atomic_t nmi_active = ATOMIC_INIT(0); /* oprofile uses this */ |
46 | static int panic_on_timeout; | 53 | EXPORT_SYMBOL(nmi_active); |
47 | 54 | ||
48 | unsigned int nmi_watchdog = NMI_DEFAULT; | 55 | unsigned int nmi_watchdog = NMI_DEFAULT; |
49 | static unsigned int nmi_hz = HZ; | 56 | EXPORT_SYMBOL(nmi_watchdog); |
57 | |||
58 | static int panic_on_timeout; | ||
50 | 59 | ||
60 | static unsigned int nmi_hz = HZ; | ||
51 | static DEFINE_PER_CPU(short, wd_enabled); | 61 | static DEFINE_PER_CPU(short, wd_enabled); |
62 | static int endflag __initdata; | ||
63 | |||
64 | static inline unsigned int get_nmi_count(int cpu) | ||
65 | { | ||
66 | #ifdef CONFIG_X86_64 | ||
67 | return cpu_pda(cpu)->__nmi_count; | ||
68 | #else | ||
69 | return nmi_count(cpu); | ||
70 | #endif | ||
71 | } | ||
72 | |||
73 | static inline int mce_in_progress(void) | ||
74 | { | ||
75 | #if defined(CONFIG_X86_64) && defined(CONFIG_X86_MCE) | ||
76 | return atomic_read(&mce_entry) > 0; | ||
77 | #endif | ||
78 | return 0; | ||
79 | } | ||
80 | |||
81 | /* | ||
82 | * Take the local apic timer and PIT/HPET into account. We don't | ||
83 | * know which one is active, when we have highres/dyntick on | ||
84 | */ | ||
85 | static inline unsigned int get_timer_irqs(int cpu) | ||
86 | { | ||
87 | #ifdef CONFIG_X86_64 | ||
88 | return read_pda(apic_timer_irqs) + read_pda(irq0_irqs); | ||
89 | #else | ||
90 | return per_cpu(irq_stat, cpu).apic_timer_irqs + | ||
91 | per_cpu(irq_stat, cpu).irq0_irqs; | ||
92 | #endif | ||
93 | } | ||
52 | 94 | ||
53 | /* Run after command line and cpu_init init, but before all other checks */ | 95 | /* Run after command line and cpu_init init, but before all other checks */ |
54 | void nmi_watchdog_default(void) | 96 | void nmi_watchdog_default(void) |
@@ -58,22 +100,23 @@ void nmi_watchdog_default(void) | |||
58 | nmi_watchdog = NMI_NONE; | 100 | nmi_watchdog = NMI_NONE; |
59 | } | 101 | } |
60 | 102 | ||
61 | static int endflag __initdata = 0; | ||
62 | |||
63 | #ifdef CONFIG_SMP | 103 | #ifdef CONFIG_SMP |
64 | /* The performance counters used by NMI_LOCAL_APIC don't trigger when | 104 | /* |
105 | * The performance counters used by NMI_LOCAL_APIC don't trigger when | ||
65 | * the CPU is idle. To make sure the NMI watchdog really ticks on all | 106 | * the CPU is idle. To make sure the NMI watchdog really ticks on all |
66 | * CPUs during the test make them busy. | 107 | * CPUs during the test make them busy. |
67 | */ | 108 | */ |
68 | static __init void nmi_cpu_busy(void *data) | 109 | static __init void nmi_cpu_busy(void *data) |
69 | { | 110 | { |
70 | local_irq_enable_in_hardirq(); | 111 | local_irq_enable_in_hardirq(); |
71 | /* Intentionally don't use cpu_relax here. This is | 112 | /* |
72 | to make sure that the performance counter really ticks, | 113 | * Intentionally don't use cpu_relax here. This is |
73 | even if there is a simulator or similar that catches the | 114 | * to make sure that the performance counter really ticks, |
74 | pause instruction. On a real HT machine this is fine because | 115 | * even if there is a simulator or similar that catches the |
75 | all other CPUs are busy with "useless" delay loops and don't | 116 | * pause instruction. On a real HT machine this is fine because |
76 | care if they get somewhat less cycles. */ | 117 | * all other CPUs are busy with "useless" delay loops and don't |
118 | * care if they get somewhat less cycles. | ||
119 | */ | ||
77 | while (endflag == 0) | 120 | while (endflag == 0) |
78 | mb(); | 121 | mb(); |
79 | } | 122 | } |
@@ -81,10 +124,10 @@ static __init void nmi_cpu_busy(void *data) | |||
81 | 124 | ||
82 | int __init check_nmi_watchdog(void) | 125 | int __init check_nmi_watchdog(void) |
83 | { | 126 | { |
84 | int *prev_nmi_count; | 127 | unsigned int *prev_nmi_count; |
85 | int cpu; | 128 | int cpu; |
86 | 129 | ||
87 | if ((nmi_watchdog == NMI_NONE) || (nmi_watchdog == NMI_DISABLED)) | 130 | if (nmi_watchdog == NMI_NONE || nmi_watchdog == NMI_DISABLED) |
88 | return 0; | 131 | return 0; |
89 | 132 | ||
90 | if (!atomic_read(&nmi_active)) | 133 | if (!atomic_read(&nmi_active)) |
@@ -101,20 +144,20 @@ int __init check_nmi_watchdog(void) | |||
101 | smp_call_function(nmi_cpu_busy, (void *)&endflag, 0, 0); | 144 | smp_call_function(nmi_cpu_busy, (void *)&endflag, 0, 0); |
102 | #endif | 145 | #endif |
103 | 146 | ||
104 | for (cpu = 0; cpu < nr_cpu_ids; cpu++) | 147 | for_each_possible_cpu(cpu) |
105 | prev_nmi_count[cpu] = cpu_pda(cpu)->__nmi_count; | 148 | prev_nmi_count[cpu] = get_nmi_count(cpu); |
106 | local_irq_enable(); | 149 | local_irq_enable(); |
107 | mdelay((20*1000)/nmi_hz); // wait 20 ticks | 150 | mdelay((20 * 1000) / nmi_hz); /* wait 20 ticks */ |
108 | 151 | ||
109 | for_each_online_cpu(cpu) { | 152 | for_each_online_cpu(cpu) { |
110 | if (!per_cpu(wd_enabled, cpu)) | 153 | if (!per_cpu(wd_enabled, cpu)) |
111 | continue; | 154 | continue; |
112 | if (cpu_pda(cpu)->__nmi_count - prev_nmi_count[cpu] <= 5) { | 155 | if (get_nmi_count(cpu) - prev_nmi_count[cpu] <= 5) { |
113 | printk(KERN_WARNING "WARNING: CPU#%d: NMI " | 156 | printk(KERN_WARNING "WARNING: CPU#%d: NMI " |
114 | "appears to be stuck (%d->%d)!\n", | 157 | "appears to be stuck (%d->%d)!\n", |
115 | cpu, | 158 | cpu, |
116 | prev_nmi_count[cpu], | 159 | prev_nmi_count[cpu], |
117 | cpu_pda(cpu)->__nmi_count); | 160 | get_nmi_count(cpu)); |
118 | per_cpu(wd_enabled, cpu) = 0; | 161 | per_cpu(wd_enabled, cpu) = 0; |
119 | atomic_dec(&nmi_active); | 162 | atomic_dec(&nmi_active); |
120 | } | 163 | } |
@@ -127,8 +170,10 @@ int __init check_nmi_watchdog(void) | |||
127 | } | 170 | } |
128 | printk("OK.\n"); | 171 | printk("OK.\n"); |
129 | 172 | ||
130 | /* now that we know it works we can reduce NMI frequency to | 173 | /* |
131 | something more reasonable; makes a difference in some configs */ | 174 | * now that we know it works we can reduce NMI frequency to |
175 | * something more reasonable; makes a difference in some configs | ||
176 | */ | ||
132 | if (nmi_watchdog == NMI_LOCAL_APIC) | 177 | if (nmi_watchdog == NMI_LOCAL_APIC) |
133 | nmi_hz = lapic_adjust_nmi_hz(1); | 178 | nmi_hz = lapic_adjust_nmi_hz(1); |
134 | 179 | ||
@@ -137,7 +182,6 @@ int __init check_nmi_watchdog(void) | |||
137 | error: | 182 | error: |
138 | if (nmi_watchdog == NMI_IO_APIC && !timer_through_8259) | 183 | if (nmi_watchdog == NMI_IO_APIC && !timer_through_8259) |
139 | disable_8259A_irq(0); | 184 | disable_8259A_irq(0); |
140 | |||
141 | return -1; | 185 | return -1; |
142 | } | 186 | } |
143 | 187 | ||
@@ -145,7 +189,7 @@ static int __init setup_nmi_watchdog(char *str) | |||
145 | { | 189 | { |
146 | int nmi; | 190 | int nmi; |
147 | 191 | ||
148 | if (!strncmp(str,"panic",5)) { | 192 | if (!strncmp(str, "panic", 5)) { |
149 | panic_on_timeout = 1; | 193 | panic_on_timeout = 1; |
150 | str = strchr(str, ','); | 194 | str = strchr(str, ','); |
151 | if (!str) | 195 | if (!str) |
@@ -155,15 +199,17 @@ static int __init setup_nmi_watchdog(char *str) | |||
155 | 199 | ||
156 | get_option(&str, &nmi); | 200 | get_option(&str, &nmi); |
157 | 201 | ||
158 | if ((nmi >= NMI_INVALID) || (nmi < NMI_NONE)) | 202 | if (nmi >= NMI_INVALID || nmi < NMI_NONE) |
159 | return 0; | 203 | return 0; |
160 | 204 | ||
161 | nmi_watchdog = nmi; | 205 | nmi_watchdog = nmi; |
162 | return 1; | 206 | return 1; |
163 | } | 207 | } |
164 | |||
165 | __setup("nmi_watchdog=", setup_nmi_watchdog); | 208 | __setup("nmi_watchdog=", setup_nmi_watchdog); |
166 | 209 | ||
210 | /* | ||
211 | * Suspend/resume support | ||
212 | */ | ||
167 | #ifdef CONFIG_PM | 213 | #ifdef CONFIG_PM |
168 | 214 | ||
169 | static int nmi_pm_active; /* nmi_active before suspend */ | 215 | static int nmi_pm_active; /* nmi_active before suspend */ |
@@ -202,7 +248,8 @@ static int __init init_lapic_nmi_sysfs(void) | |||
202 | { | 248 | { |
203 | int error; | 249 | int error; |
204 | 250 | ||
205 | /* should really be a BUG_ON but b/c this is an | 251 | /* |
252 | * should really be a BUG_ON but b/c this is an | ||
206 | * init call, it just doesn't work. -dcz | 253 | * init call, it just doesn't work. -dcz |
207 | */ | 254 | */ |
208 | if (nmi_watchdog != NMI_LOCAL_APIC) | 255 | if (nmi_watchdog != NMI_LOCAL_APIC) |
@@ -216,6 +263,7 @@ static int __init init_lapic_nmi_sysfs(void) | |||
216 | error = sysdev_register(&device_lapic_nmi); | 263 | error = sysdev_register(&device_lapic_nmi); |
217 | return error; | 264 | return error; |
218 | } | 265 | } |
266 | |||
219 | /* must come after the local APIC's device_initcall() */ | 267 | /* must come after the local APIC's device_initcall() */ |
220 | late_initcall(init_lapic_nmi_sysfs); | 268 | late_initcall(init_lapic_nmi_sysfs); |
221 | 269 | ||
@@ -223,7 +271,7 @@ late_initcall(init_lapic_nmi_sysfs); | |||
223 | 271 | ||
224 | static void __acpi_nmi_enable(void *__unused) | 272 | static void __acpi_nmi_enable(void *__unused) |
225 | { | 273 | { |
226 | apic_write(APIC_LVT0, APIC_DM_NMI); | 274 | apic_write_around(APIC_LVT0, APIC_DM_NMI); |
227 | } | 275 | } |
228 | 276 | ||
229 | /* | 277 | /* |
@@ -237,7 +285,7 @@ void acpi_nmi_enable(void) | |||
237 | 285 | ||
238 | static void __acpi_nmi_disable(void *__unused) | 286 | static void __acpi_nmi_disable(void *__unused) |
239 | { | 287 | { |
240 | apic_write(APIC_LVT0, APIC_DM_NMI | APIC_LVT_MASKED); | 288 | apic_write_around(APIC_LVT0, APIC_DM_NMI | APIC_LVT_MASKED); |
241 | } | 289 | } |
242 | 290 | ||
243 | /* | 291 | /* |
@@ -256,11 +304,12 @@ void setup_apic_nmi_watchdog(void *unused) | |||
256 | 304 | ||
257 | /* cheap hack to support suspend/resume */ | 305 | /* cheap hack to support suspend/resume */ |
258 | /* if cpu0 is not active neither should the other cpus */ | 306 | /* if cpu0 is not active neither should the other cpus */ |
259 | if ((smp_processor_id() != 0) && (atomic_read(&nmi_active) <= 0)) | 307 | if (smp_processor_id() != 0 && atomic_read(&nmi_active) <= 0) |
260 | return; | 308 | return; |
261 | 309 | ||
262 | switch (nmi_watchdog) { | 310 | switch (nmi_watchdog) { |
263 | case NMI_LOCAL_APIC: | 311 | case NMI_LOCAL_APIC: |
312 | /* enable it before to avoid race with handler */ | ||
264 | __get_cpu_var(wd_enabled) = 1; | 313 | __get_cpu_var(wd_enabled) = 1; |
265 | if (lapic_watchdog_init(nmi_hz) < 0) { | 314 | if (lapic_watchdog_init(nmi_hz) < 0) { |
266 | __get_cpu_var(wd_enabled) = 0; | 315 | __get_cpu_var(wd_enabled) = 0; |
@@ -276,9 +325,9 @@ void setup_apic_nmi_watchdog(void *unused) | |||
276 | void stop_apic_nmi_watchdog(void *unused) | 325 | void stop_apic_nmi_watchdog(void *unused) |
277 | { | 326 | { |
278 | /* only support LOCAL and IO APICs for now */ | 327 | /* only support LOCAL and IO APICs for now */ |
279 | if ((nmi_watchdog != NMI_LOCAL_APIC) && | 328 | if (nmi_watchdog != NMI_LOCAL_APIC && |
280 | (nmi_watchdog != NMI_IO_APIC)) | 329 | nmi_watchdog != NMI_IO_APIC) |
281 | return; | 330 | return; |
282 | if (__get_cpu_var(wd_enabled) == 0) | 331 | if (__get_cpu_var(wd_enabled) == 0) |
283 | return; | 332 | return; |
284 | if (nmi_watchdog == NMI_LOCAL_APIC) | 333 | if (nmi_watchdog == NMI_LOCAL_APIC) |
@@ -294,6 +343,11 @@ void stop_apic_nmi_watchdog(void *unused) | |||
294 | * | 343 | * |
295 | * as these watchdog NMI IRQs are generated on every CPU, we only | 344 | * as these watchdog NMI IRQs are generated on every CPU, we only |
296 | * have to check the current processor. | 345 | * have to check the current processor. |
346 | * | ||
347 | * since NMIs don't listen to _any_ locks, we have to be extremely | ||
348 | * careful not to rely on unsafe variables. The printk might lock | ||
349 | * up though, so we have to break up any console locks first ... | ||
350 | * [when there will be more tty-related locks, break them up here too!] | ||
297 | */ | 351 | */ |
298 | 352 | ||
299 | static DEFINE_PER_CPU(unsigned, last_irq_sum); | 353 | static DEFINE_PER_CPU(unsigned, last_irq_sum); |
@@ -302,11 +356,12 @@ static DEFINE_PER_CPU(int, nmi_touch); | |||
302 | 356 | ||
303 | void touch_nmi_watchdog(void) | 357 | void touch_nmi_watchdog(void) |
304 | { | 358 | { |
305 | if (nmi_watchdog > 0) { | 359 | if (nmi_watchdog == NMI_LOCAL_APIC || |
360 | nmi_watchdog == NMI_IO_APIC) { | ||
306 | unsigned cpu; | 361 | unsigned cpu; |
307 | 362 | ||
308 | /* | 363 | /* |
309 | * Tell other CPUs to reset their alert counters. We cannot | 364 | * Tell other CPUs to reset their alert counters. We cannot |
310 | * do it ourselves because the alert count increase is not | 365 | * do it ourselves because the alert count increase is not |
311 | * atomic. | 366 | * atomic. |
312 | */ | 367 | */ |
@@ -316,6 +371,9 @@ void touch_nmi_watchdog(void) | |||
316 | } | 371 | } |
317 | } | 372 | } |
318 | 373 | ||
374 | /* | ||
375 | * Tickle the softlockup detector too: | ||
376 | */ | ||
319 | touch_softlockup_watchdog(); | 377 | touch_softlockup_watchdog(); |
320 | } | 378 | } |
321 | EXPORT_SYMBOL(touch_nmi_watchdog); | 379 | EXPORT_SYMBOL(touch_nmi_watchdog); |
@@ -323,7 +381,12 @@ EXPORT_SYMBOL(touch_nmi_watchdog); | |||
323 | notrace __kprobes int | 381 | notrace __kprobes int |
324 | nmi_watchdog_tick(struct pt_regs *regs, unsigned reason) | 382 | nmi_watchdog_tick(struct pt_regs *regs, unsigned reason) |
325 | { | 383 | { |
326 | int sum; | 384 | /* |
385 | * Since current_thread_info()-> is always on the stack, and we | ||
386 | * always switch the stack NMI-atomically, it's safe to use | ||
387 | * smp_processor_id(). | ||
388 | */ | ||
389 | unsigned int sum; | ||
327 | int touched = 0; | 390 | int touched = 0; |
328 | int cpu = smp_processor_id(); | 391 | int cpu = smp_processor_id(); |
329 | int rc = 0; | 392 | int rc = 0; |
@@ -335,7 +398,8 @@ nmi_watchdog_tick(struct pt_regs *regs, unsigned reason) | |||
335 | touched = 1; | 398 | touched = 1; |
336 | } | 399 | } |
337 | 400 | ||
338 | sum = read_pda(apic_timer_irqs) + read_pda(irq0_irqs); | 401 | sum = get_timer_irqs(cpu); |
402 | |||
339 | if (__get_cpu_var(nmi_touch)) { | 403 | if (__get_cpu_var(nmi_touch)) { |
340 | __get_cpu_var(nmi_touch) = 0; | 404 | __get_cpu_var(nmi_touch) = 0; |
341 | touched = 1; | 405 | touched = 1; |
@@ -345,28 +409,29 @@ nmi_watchdog_tick(struct pt_regs *regs, unsigned reason) | |||
345 | static DEFINE_SPINLOCK(lock); /* Serialise the printks */ | 409 | static DEFINE_SPINLOCK(lock); /* Serialise the printks */ |
346 | 410 | ||
347 | spin_lock(&lock); | 411 | spin_lock(&lock); |
348 | printk("NMI backtrace for cpu %d\n", cpu); | 412 | printk(KERN_WARNING "NMI backtrace for cpu %d\n", cpu); |
349 | dump_stack(); | 413 | dump_stack(); |
350 | spin_unlock(&lock); | 414 | spin_unlock(&lock); |
351 | cpu_clear(cpu, backtrace_mask); | 415 | cpu_clear(cpu, backtrace_mask); |
352 | } | 416 | } |
353 | 417 | ||
354 | #ifdef CONFIG_X86_MCE | 418 | /* Could check oops_in_progress here too, but it's safer not to */ |
355 | /* Could check oops_in_progress here too, but it's safer | 419 | if (mce_in_progress()) |
356 | not too */ | ||
357 | if (atomic_read(&mce_entry) > 0) | ||
358 | touched = 1; | 420 | touched = 1; |
359 | #endif | 421 | |
360 | /* if the apic timer isn't firing, this cpu isn't doing much */ | 422 | /* if the none of the timers isn't firing, this cpu isn't doing much */ |
361 | if (!touched && __get_cpu_var(last_irq_sum) == sum) { | 423 | if (!touched && __get_cpu_var(last_irq_sum) == sum) { |
362 | /* | 424 | /* |
363 | * Ayiee, looks like this CPU is stuck ... | 425 | * Ayiee, looks like this CPU is stuck ... |
364 | * wait a few IRQs (5 seconds) before doing the oops ... | 426 | * wait a few IRQs (5 seconds) before doing the oops ... |
365 | */ | 427 | */ |
366 | local_inc(&__get_cpu_var(alert_counter)); | 428 | local_inc(&__get_cpu_var(alert_counter)); |
367 | if (local_read(&__get_cpu_var(alert_counter)) == 5*nmi_hz) | 429 | if (local_read(&__get_cpu_var(alert_counter)) == 5 * nmi_hz) |
368 | die_nmi("NMI Watchdog detected LOCKUP on CPU %d\n", regs, | 430 | /* |
369 | panic_on_timeout); | 431 | * die_nmi will return ONLY if NOTIFY_STOP happens.. |
432 | */ | ||
433 | die_nmi("BUG: NMI Watchdog detected LOCKUP", | ||
434 | regs, panic_on_timeout); | ||
370 | } else { | 435 | } else { |
371 | __get_cpu_var(last_irq_sum) = sum; | 436 | __get_cpu_var(last_irq_sum) = sum; |
372 | local_set(&__get_cpu_var(alert_counter), 0); | 437 | local_set(&__get_cpu_var(alert_counter), 0); |
@@ -380,7 +445,8 @@ nmi_watchdog_tick(struct pt_regs *regs, unsigned reason) | |||
380 | rc |= lapic_wd_event(nmi_hz); | 445 | rc |= lapic_wd_event(nmi_hz); |
381 | break; | 446 | break; |
382 | case NMI_IO_APIC: | 447 | case NMI_IO_APIC: |
383 | /* don't know how to accurately check for this. | 448 | /* |
449 | * don't know how to accurately check for this. | ||
384 | * just assume it was a watchdog timer interrupt | 450 | * just assume it was a watchdog timer interrupt |
385 | * This matches the old behaviour. | 451 | * This matches the old behaviour. |
386 | */ | 452 | */ |
@@ -390,30 +456,6 @@ nmi_watchdog_tick(struct pt_regs *regs, unsigned reason) | |||
390 | return rc; | 456 | return rc; |
391 | } | 457 | } |
392 | 458 | ||
393 | static unsigned ignore_nmis; | ||
394 | |||
395 | asmlinkage notrace __kprobes void | ||
396 | do_nmi(struct pt_regs *regs, long error_code) | ||
397 | { | ||
398 | nmi_enter(); | ||
399 | add_pda(__nmi_count,1); | ||
400 | if (!ignore_nmis) | ||
401 | default_do_nmi(regs); | ||
402 | nmi_exit(); | ||
403 | } | ||
404 | |||
405 | void stop_nmi(void) | ||
406 | { | ||
407 | acpi_nmi_disable(); | ||
408 | ignore_nmis++; | ||
409 | } | ||
410 | |||
411 | void restart_nmi(void) | ||
412 | { | ||
413 | ignore_nmis--; | ||
414 | acpi_nmi_enable(); | ||
415 | } | ||
416 | |||
417 | #ifdef CONFIG_SYSCTL | 459 | #ifdef CONFIG_SYSCTL |
418 | 460 | ||
419 | static int unknown_nmi_panic_callback(struct pt_regs *regs, int cpu) | 461 | static int unknown_nmi_panic_callback(struct pt_regs *regs, int cpu) |
@@ -422,7 +464,7 @@ static int unknown_nmi_panic_callback(struct pt_regs *regs, int cpu) | |||
422 | char buf[64]; | 464 | char buf[64]; |
423 | 465 | ||
424 | sprintf(buf, "NMI received for unknown reason %02x\n", reason); | 466 | sprintf(buf, "NMI received for unknown reason %02x\n", reason); |
425 | die_nmi(buf, regs, 1); /* Always panic here */ | 467 | die_nmi(buf, regs, 1); /* Always panic here */ |
426 | return 0; | 468 | return 0; |
427 | } | 469 | } |
428 | 470 | ||
@@ -441,27 +483,37 @@ int proc_nmi_enabled(struct ctl_table *table, int write, struct file *file, | |||
441 | return 0; | 483 | return 0; |
442 | 484 | ||
443 | if (atomic_read(&nmi_active) < 0 || nmi_watchdog == NMI_DISABLED) { | 485 | if (atomic_read(&nmi_active) < 0 || nmi_watchdog == NMI_DISABLED) { |
444 | printk( KERN_WARNING "NMI watchdog is permanently disabled\n"); | 486 | printk(KERN_WARNING |
487 | "NMI watchdog is permanently disabled\n"); | ||
445 | return -EIO; | 488 | return -EIO; |
446 | } | 489 | } |
447 | 490 | ||
448 | /* if nmi_watchdog is not set yet, then set it */ | 491 | /* if nmi_watchdog is not set yet, then set it */ |
449 | nmi_watchdog_default(); | 492 | nmi_watchdog_default(); |
450 | 493 | ||
494 | #ifdef CONFIG_X86_32 | ||
495 | if (nmi_watchdog == NMI_NONE) { | ||
496 | if (lapic_watchdog_ok()) | ||
497 | nmi_watchdog = NMI_LOCAL_APIC; | ||
498 | else | ||
499 | nmi_watchdog = NMI_IO_APIC; | ||
500 | } | ||
501 | #endif | ||
502 | |||
451 | if (nmi_watchdog == NMI_LOCAL_APIC) { | 503 | if (nmi_watchdog == NMI_LOCAL_APIC) { |
452 | if (nmi_watchdog_enabled) | 504 | if (nmi_watchdog_enabled) |
453 | enable_lapic_nmi_watchdog(); | 505 | enable_lapic_nmi_watchdog(); |
454 | else | 506 | else |
455 | disable_lapic_nmi_watchdog(); | 507 | disable_lapic_nmi_watchdog(); |
456 | } else { | 508 | } else { |
457 | printk( KERN_WARNING | 509 | printk(KERN_WARNING |
458 | "NMI watchdog doesn't know what hardware to touch\n"); | 510 | "NMI watchdog doesn't know what hardware to touch\n"); |
459 | return -EIO; | 511 | return -EIO; |
460 | } | 512 | } |
461 | return 0; | 513 | return 0; |
462 | } | 514 | } |
463 | 515 | ||
464 | #endif | 516 | #endif /* CONFIG_SYSCTL */ |
465 | 517 | ||
466 | int do_nmi_callback(struct pt_regs *regs, int cpu) | 518 | int do_nmi_callback(struct pt_regs *regs, int cpu) |
467 | { | 519 | { |
@@ -484,6 +536,3 @@ void __trigger_all_cpu_backtrace(void) | |||
484 | mdelay(1); | 536 | mdelay(1); |
485 | } | 537 | } |
486 | } | 538 | } |
487 | |||
488 | EXPORT_SYMBOL(nmi_active); | ||
489 | EXPORT_SYMBOL(nmi_watchdog); | ||
diff --git a/arch/x86/kernel/nmi_32.c b/arch/x86/kernel/nmi_32.c deleted file mode 100644 index 6580dae46277..000000000000 --- a/arch/x86/kernel/nmi_32.c +++ /dev/null | |||
@@ -1,476 +0,0 @@ | |||
1 | /* | ||
2 | * NMI watchdog support on APIC systems | ||
3 | * | ||
4 | * Started by Ingo Molnar <mingo@redhat.com> | ||
5 | * | ||
6 | * Fixes: | ||
7 | * Mikael Pettersson : AMD K7 support for local APIC NMI watchdog. | ||
8 | * Mikael Pettersson : Power Management for local APIC NMI watchdog. | ||
9 | * Mikael Pettersson : Pentium 4 support for local APIC NMI watchdog. | ||
10 | * Pavel Machek and | ||
11 | * Mikael Pettersson : PM converted to driver model. Disable/enable API. | ||
12 | */ | ||
13 | |||
14 | #include <linux/delay.h> | ||
15 | #include <linux/interrupt.h> | ||
16 | #include <linux/module.h> | ||
17 | #include <linux/nmi.h> | ||
18 | #include <linux/sysdev.h> | ||
19 | #include <linux/sysctl.h> | ||
20 | #include <linux/percpu.h> | ||
21 | #include <linux/kprobes.h> | ||
22 | #include <linux/cpumask.h> | ||
23 | #include <linux/kernel_stat.h> | ||
24 | #include <linux/kdebug.h> | ||
25 | #include <linux/slab.h> | ||
26 | |||
27 | #include <asm/i8259.h> | ||
28 | #include <asm/io_apic.h> | ||
29 | #include <asm/smp.h> | ||
30 | #include <asm/nmi.h> | ||
31 | #include <asm/timer.h> | ||
32 | |||
33 | #include "mach_traps.h" | ||
34 | |||
35 | int unknown_nmi_panic; | ||
36 | int nmi_watchdog_enabled; | ||
37 | |||
38 | static cpumask_t backtrace_mask = CPU_MASK_NONE; | ||
39 | |||
40 | /* nmi_active: | ||
41 | * >0: the lapic NMI watchdog is active, but can be disabled | ||
42 | * <0: the lapic NMI watchdog has not been set up, and cannot | ||
43 | * be enabled | ||
44 | * 0: the lapic NMI watchdog is disabled, but can be enabled | ||
45 | */ | ||
46 | atomic_t nmi_active = ATOMIC_INIT(0); /* oprofile uses this */ | ||
47 | |||
48 | unsigned int nmi_watchdog = NMI_DEFAULT; | ||
49 | static unsigned int nmi_hz = HZ; | ||
50 | |||
51 | static DEFINE_PER_CPU(short, wd_enabled); | ||
52 | |||
53 | static int endflag __initdata = 0; | ||
54 | |||
55 | #ifdef CONFIG_SMP | ||
56 | /* The performance counters used by NMI_LOCAL_APIC don't trigger when | ||
57 | * the CPU is idle. To make sure the NMI watchdog really ticks on all | ||
58 | * CPUs during the test make them busy. | ||
59 | */ | ||
60 | static __init void nmi_cpu_busy(void *data) | ||
61 | { | ||
62 | local_irq_enable_in_hardirq(); | ||
63 | /* Intentionally don't use cpu_relax here. This is | ||
64 | to make sure that the performance counter really ticks, | ||
65 | even if there is a simulator or similar that catches the | ||
66 | pause instruction. On a real HT machine this is fine because | ||
67 | all other CPUs are busy with "useless" delay loops and don't | ||
68 | care if they get somewhat less cycles. */ | ||
69 | while (endflag == 0) | ||
70 | mb(); | ||
71 | } | ||
72 | #endif | ||
73 | |||
74 | int __init check_nmi_watchdog(void) | ||
75 | { | ||
76 | unsigned int *prev_nmi_count; | ||
77 | int cpu; | ||
78 | |||
79 | if ((nmi_watchdog == NMI_NONE) || (nmi_watchdog == NMI_DISABLED)) | ||
80 | return 0; | ||
81 | |||
82 | if (!atomic_read(&nmi_active)) | ||
83 | return 0; | ||
84 | |||
85 | prev_nmi_count = kmalloc(NR_CPUS * sizeof(int), GFP_KERNEL); | ||
86 | if (!prev_nmi_count) | ||
87 | goto error; | ||
88 | |||
89 | printk(KERN_INFO "Testing NMI watchdog ... "); | ||
90 | |||
91 | #ifdef CONFIG_SMP | ||
92 | if (nmi_watchdog == NMI_LOCAL_APIC) | ||
93 | smp_call_function(nmi_cpu_busy, (void *)&endflag, 0, 0); | ||
94 | #endif | ||
95 | |||
96 | for_each_possible_cpu(cpu) | ||
97 | prev_nmi_count[cpu] = nmi_count(cpu); | ||
98 | local_irq_enable(); | ||
99 | mdelay((20*1000)/nmi_hz); // wait 20 ticks | ||
100 | |||
101 | for_each_possible_cpu(cpu) { | ||
102 | #ifdef CONFIG_SMP | ||
103 | /* Check cpu_callin_map here because that is set | ||
104 | after the timer is started. */ | ||
105 | if (!cpu_isset(cpu, cpu_callin_map)) | ||
106 | continue; | ||
107 | #endif | ||
108 | if (!per_cpu(wd_enabled, cpu)) | ||
109 | continue; | ||
110 | if (nmi_count(cpu) - prev_nmi_count[cpu] <= 5) { | ||
111 | printk(KERN_WARNING "WARNING: CPU#%d: NMI " | ||
112 | "appears to be stuck (%d->%d)!\n", | ||
113 | cpu, | ||
114 | prev_nmi_count[cpu], | ||
115 | nmi_count(cpu)); | ||
116 | per_cpu(wd_enabled, cpu) = 0; | ||
117 | atomic_dec(&nmi_active); | ||
118 | } | ||
119 | } | ||
120 | endflag = 1; | ||
121 | if (!atomic_read(&nmi_active)) { | ||
122 | kfree(prev_nmi_count); | ||
123 | atomic_set(&nmi_active, -1); | ||
124 | goto error; | ||
125 | } | ||
126 | printk("OK.\n"); | ||
127 | |||
128 | /* now that we know it works we can reduce NMI frequency to | ||
129 | something more reasonable; makes a difference in some configs */ | ||
130 | if (nmi_watchdog == NMI_LOCAL_APIC) | ||
131 | nmi_hz = lapic_adjust_nmi_hz(1); | ||
132 | |||
133 | kfree(prev_nmi_count); | ||
134 | return 0; | ||
135 | error: | ||
136 | if (nmi_watchdog == NMI_IO_APIC && !timer_through_8259) | ||
137 | disable_8259A_irq(0); | ||
138 | timer_ack = 0; | ||
139 | |||
140 | return -1; | ||
141 | } | ||
142 | |||
143 | static int __init setup_nmi_watchdog(char *str) | ||
144 | { | ||
145 | int nmi; | ||
146 | |||
147 | get_option(&str, &nmi); | ||
148 | |||
149 | if ((nmi >= NMI_INVALID) || (nmi < NMI_NONE)) | ||
150 | return 0; | ||
151 | |||
152 | nmi_watchdog = nmi; | ||
153 | return 1; | ||
154 | } | ||
155 | |||
156 | __setup("nmi_watchdog=", setup_nmi_watchdog); | ||
157 | |||
158 | |||
159 | /* Suspend/resume support */ | ||
160 | |||
161 | #ifdef CONFIG_PM | ||
162 | |||
163 | static int nmi_pm_active; /* nmi_active before suspend */ | ||
164 | |||
165 | static int lapic_nmi_suspend(struct sys_device *dev, pm_message_t state) | ||
166 | { | ||
167 | /* only CPU0 goes here, other CPUs should be offline */ | ||
168 | nmi_pm_active = atomic_read(&nmi_active); | ||
169 | stop_apic_nmi_watchdog(NULL); | ||
170 | BUG_ON(atomic_read(&nmi_active) != 0); | ||
171 | return 0; | ||
172 | } | ||
173 | |||
174 | static int lapic_nmi_resume(struct sys_device *dev) | ||
175 | { | ||
176 | /* only CPU0 goes here, other CPUs should be offline */ | ||
177 | if (nmi_pm_active > 0) { | ||
178 | setup_apic_nmi_watchdog(NULL); | ||
179 | touch_nmi_watchdog(); | ||
180 | } | ||
181 | return 0; | ||
182 | } | ||
183 | |||
184 | |||
185 | static struct sysdev_class nmi_sysclass = { | ||
186 | .name = "lapic_nmi", | ||
187 | .resume = lapic_nmi_resume, | ||
188 | .suspend = lapic_nmi_suspend, | ||
189 | }; | ||
190 | |||
191 | static struct sys_device device_lapic_nmi = { | ||
192 | .id = 0, | ||
193 | .cls = &nmi_sysclass, | ||
194 | }; | ||
195 | |||
196 | static int __init init_lapic_nmi_sysfs(void) | ||
197 | { | ||
198 | int error; | ||
199 | |||
200 | /* should really be a BUG_ON but b/c this is an | ||
201 | * init call, it just doesn't work. -dcz | ||
202 | */ | ||
203 | if (nmi_watchdog != NMI_LOCAL_APIC) | ||
204 | return 0; | ||
205 | |||
206 | if (atomic_read(&nmi_active) < 0) | ||
207 | return 0; | ||
208 | |||
209 | error = sysdev_class_register(&nmi_sysclass); | ||
210 | if (!error) | ||
211 | error = sysdev_register(&device_lapic_nmi); | ||
212 | return error; | ||
213 | } | ||
214 | /* must come after the local APIC's device_initcall() */ | ||
215 | late_initcall(init_lapic_nmi_sysfs); | ||
216 | |||
217 | #endif /* CONFIG_PM */ | ||
218 | |||
219 | static void __acpi_nmi_enable(void *__unused) | ||
220 | { | ||
221 | apic_write_around(APIC_LVT0, APIC_DM_NMI); | ||
222 | } | ||
223 | |||
224 | /* | ||
225 | * Enable timer based NMIs on all CPUs: | ||
226 | */ | ||
227 | void acpi_nmi_enable(void) | ||
228 | { | ||
229 | if (atomic_read(&nmi_active) && nmi_watchdog == NMI_IO_APIC) | ||
230 | on_each_cpu(__acpi_nmi_enable, NULL, 0, 1); | ||
231 | } | ||
232 | |||
233 | static void __acpi_nmi_disable(void *__unused) | ||
234 | { | ||
235 | apic_write(APIC_LVT0, APIC_DM_NMI | APIC_LVT_MASKED); | ||
236 | } | ||
237 | |||
238 | /* | ||
239 | * Disable timer based NMIs on all CPUs: | ||
240 | */ | ||
241 | void acpi_nmi_disable(void) | ||
242 | { | ||
243 | if (atomic_read(&nmi_active) && nmi_watchdog == NMI_IO_APIC) | ||
244 | on_each_cpu(__acpi_nmi_disable, NULL, 0, 1); | ||
245 | } | ||
246 | |||
247 | void setup_apic_nmi_watchdog(void *unused) | ||
248 | { | ||
249 | if (__get_cpu_var(wd_enabled)) | ||
250 | return; | ||
251 | |||
252 | /* cheap hack to support suspend/resume */ | ||
253 | /* if cpu0 is not active neither should the other cpus */ | ||
254 | if ((smp_processor_id() != 0) && (atomic_read(&nmi_active) <= 0)) | ||
255 | return; | ||
256 | |||
257 | switch (nmi_watchdog) { | ||
258 | case NMI_LOCAL_APIC: | ||
259 | __get_cpu_var(wd_enabled) = 1; /* enable it before to avoid race with handler */ | ||
260 | if (lapic_watchdog_init(nmi_hz) < 0) { | ||
261 | __get_cpu_var(wd_enabled) = 0; | ||
262 | return; | ||
263 | } | ||
264 | /* FALL THROUGH */ | ||
265 | case NMI_IO_APIC: | ||
266 | __get_cpu_var(wd_enabled) = 1; | ||
267 | atomic_inc(&nmi_active); | ||
268 | } | ||
269 | } | ||
270 | |||
271 | void stop_apic_nmi_watchdog(void *unused) | ||
272 | { | ||
273 | /* only support LOCAL and IO APICs for now */ | ||
274 | if ((nmi_watchdog != NMI_LOCAL_APIC) && | ||
275 | (nmi_watchdog != NMI_IO_APIC)) | ||
276 | return; | ||
277 | if (__get_cpu_var(wd_enabled) == 0) | ||
278 | return; | ||
279 | if (nmi_watchdog == NMI_LOCAL_APIC) | ||
280 | lapic_watchdog_stop(); | ||
281 | __get_cpu_var(wd_enabled) = 0; | ||
282 | atomic_dec(&nmi_active); | ||
283 | } | ||
284 | |||
285 | /* | ||
286 | * the best way to detect whether a CPU has a 'hard lockup' problem | ||
287 | * is to check it's local APIC timer IRQ counts. If they are not | ||
288 | * changing then that CPU has some problem. | ||
289 | * | ||
290 | * as these watchdog NMI IRQs are generated on every CPU, we only | ||
291 | * have to check the current processor. | ||
292 | * | ||
293 | * since NMIs don't listen to _any_ locks, we have to be extremely | ||
294 | * careful not to rely on unsafe variables. The printk might lock | ||
295 | * up though, so we have to break up any console locks first ... | ||
296 | * [when there will be more tty-related locks, break them up | ||
297 | * here too!] | ||
298 | */ | ||
299 | |||
300 | static unsigned int | ||
301 | last_irq_sums [NR_CPUS], | ||
302 | alert_counter [NR_CPUS]; | ||
303 | |||
304 | void touch_nmi_watchdog(void) | ||
305 | { | ||
306 | if (nmi_watchdog > 0) { | ||
307 | unsigned cpu; | ||
308 | |||
309 | /* | ||
310 | * Just reset the alert counters, (other CPUs might be | ||
311 | * spinning on locks we hold): | ||
312 | */ | ||
313 | for_each_present_cpu(cpu) { | ||
314 | if (alert_counter[cpu]) | ||
315 | alert_counter[cpu] = 0; | ||
316 | } | ||
317 | } | ||
318 | |||
319 | /* | ||
320 | * Tickle the softlockup detector too: | ||
321 | */ | ||
322 | touch_softlockup_watchdog(); | ||
323 | } | ||
324 | EXPORT_SYMBOL(touch_nmi_watchdog); | ||
325 | |||
326 | extern void die_nmi(struct pt_regs *, const char *msg); | ||
327 | |||
328 | notrace __kprobes int | ||
329 | nmi_watchdog_tick(struct pt_regs *regs, unsigned reason) | ||
330 | { | ||
331 | |||
332 | /* | ||
333 | * Since current_thread_info()-> is always on the stack, and we | ||
334 | * always switch the stack NMI-atomically, it's safe to use | ||
335 | * smp_processor_id(). | ||
336 | */ | ||
337 | unsigned int sum; | ||
338 | int touched = 0; | ||
339 | int cpu = smp_processor_id(); | ||
340 | int rc = 0; | ||
341 | |||
342 | /* check for other users first */ | ||
343 | if (notify_die(DIE_NMI, "nmi", regs, reason, 2, SIGINT) | ||
344 | == NOTIFY_STOP) { | ||
345 | rc = 1; | ||
346 | touched = 1; | ||
347 | } | ||
348 | |||
349 | if (cpu_isset(cpu, backtrace_mask)) { | ||
350 | static DEFINE_SPINLOCK(lock); /* Serialise the printks */ | ||
351 | |||
352 | spin_lock(&lock); | ||
353 | printk("NMI backtrace for cpu %d\n", cpu); | ||
354 | dump_stack(); | ||
355 | spin_unlock(&lock); | ||
356 | cpu_clear(cpu, backtrace_mask); | ||
357 | } | ||
358 | |||
359 | /* | ||
360 | * Take the local apic timer and PIT/HPET into account. We don't | ||
361 | * know which one is active, when we have highres/dyntick on | ||
362 | */ | ||
363 | sum = per_cpu(irq_stat, cpu).apic_timer_irqs + | ||
364 | per_cpu(irq_stat, cpu).irq0_irqs; | ||
365 | |||
366 | /* if the none of the timers isn't firing, this cpu isn't doing much */ | ||
367 | if (!touched && last_irq_sums[cpu] == sum) { | ||
368 | /* | ||
369 | * Ayiee, looks like this CPU is stuck ... | ||
370 | * wait a few IRQs (5 seconds) before doing the oops ... | ||
371 | */ | ||
372 | alert_counter[cpu]++; | ||
373 | if (alert_counter[cpu] == 5*nmi_hz) | ||
374 | /* | ||
375 | * die_nmi will return ONLY if NOTIFY_STOP happens.. | ||
376 | */ | ||
377 | die_nmi(regs, "BUG: NMI Watchdog detected LOCKUP"); | ||
378 | } else { | ||
379 | last_irq_sums[cpu] = sum; | ||
380 | alert_counter[cpu] = 0; | ||
381 | } | ||
382 | /* see if the nmi watchdog went off */ | ||
383 | if (!__get_cpu_var(wd_enabled)) | ||
384 | return rc; | ||
385 | switch (nmi_watchdog) { | ||
386 | case NMI_LOCAL_APIC: | ||
387 | rc |= lapic_wd_event(nmi_hz); | ||
388 | break; | ||
389 | case NMI_IO_APIC: | ||
390 | /* don't know how to accurately check for this. | ||
391 | * just assume it was a watchdog timer interrupt | ||
392 | * This matches the old behaviour. | ||
393 | */ | ||
394 | rc = 1; | ||
395 | break; | ||
396 | } | ||
397 | return rc; | ||
398 | } | ||
399 | |||
400 | #ifdef CONFIG_SYSCTL | ||
401 | |||
402 | static int unknown_nmi_panic_callback(struct pt_regs *regs, int cpu) | ||
403 | { | ||
404 | unsigned char reason = get_nmi_reason(); | ||
405 | char buf[64]; | ||
406 | |||
407 | sprintf(buf, "NMI received for unknown reason %02x\n", reason); | ||
408 | die_nmi(regs, buf); | ||
409 | return 0; | ||
410 | } | ||
411 | |||
412 | /* | ||
413 | * proc handler for /proc/sys/kernel/nmi | ||
414 | */ | ||
415 | int proc_nmi_enabled(struct ctl_table *table, int write, struct file *file, | ||
416 | void __user *buffer, size_t *length, loff_t *ppos) | ||
417 | { | ||
418 | int old_state; | ||
419 | |||
420 | nmi_watchdog_enabled = (atomic_read(&nmi_active) > 0) ? 1 : 0; | ||
421 | old_state = nmi_watchdog_enabled; | ||
422 | proc_dointvec(table, write, file, buffer, length, ppos); | ||
423 | if (!!old_state == !!nmi_watchdog_enabled) | ||
424 | return 0; | ||
425 | |||
426 | if (atomic_read(&nmi_active) < 0 || nmi_watchdog == NMI_DISABLED) { | ||
427 | printk( KERN_WARNING "NMI watchdog is permanently disabled\n"); | ||
428 | return -EIO; | ||
429 | } | ||
430 | |||
431 | if (nmi_watchdog == NMI_DEFAULT) { | ||
432 | if (lapic_watchdog_ok()) | ||
433 | nmi_watchdog = NMI_LOCAL_APIC; | ||
434 | else | ||
435 | nmi_watchdog = NMI_IO_APIC; | ||
436 | } | ||
437 | |||
438 | if (nmi_watchdog == NMI_LOCAL_APIC) { | ||
439 | if (nmi_watchdog_enabled) | ||
440 | enable_lapic_nmi_watchdog(); | ||
441 | else | ||
442 | disable_lapic_nmi_watchdog(); | ||
443 | } else { | ||
444 | printk( KERN_WARNING | ||
445 | "NMI watchdog doesn't know what hardware to touch\n"); | ||
446 | return -EIO; | ||
447 | } | ||
448 | return 0; | ||
449 | } | ||
450 | |||
451 | #endif | ||
452 | |||
453 | int do_nmi_callback(struct pt_regs *regs, int cpu) | ||
454 | { | ||
455 | #ifdef CONFIG_SYSCTL | ||
456 | if (unknown_nmi_panic) | ||
457 | return unknown_nmi_panic_callback(regs, cpu); | ||
458 | #endif | ||
459 | return 0; | ||
460 | } | ||
461 | |||
462 | void __trigger_all_cpu_backtrace(void) | ||
463 | { | ||
464 | int i; | ||
465 | |||
466 | backtrace_mask = cpu_online_map; | ||
467 | /* Wait for up to 10 seconds for all CPUs to do the backtrace */ | ||
468 | for (i = 0; i < 10 * 1000; i++) { | ||
469 | if (cpus_empty(backtrace_mask)) | ||
470 | break; | ||
471 | mdelay(1); | ||
472 | } | ||
473 | } | ||
474 | |||
475 | EXPORT_SYMBOL(nmi_active); | ||
476 | EXPORT_SYMBOL(nmi_watchdog); | ||
diff --git a/arch/x86/kernel/traps_32.c b/arch/x86/kernel/traps_32.c index 08d752de4eee..cf37d20b1ba7 100644 --- a/arch/x86/kernel/traps_32.c +++ b/arch/x86/kernel/traps_32.c | |||
@@ -756,9 +756,9 @@ unknown_nmi_error(unsigned char reason, struct pt_regs *regs) | |||
756 | 756 | ||
757 | static DEFINE_SPINLOCK(nmi_print_lock); | 757 | static DEFINE_SPINLOCK(nmi_print_lock); |
758 | 758 | ||
759 | void notrace __kprobes die_nmi(struct pt_regs *regs, const char *msg) | 759 | void notrace __kprobes die_nmi(char *str, struct pt_regs *regs, int do_panic) |
760 | { | 760 | { |
761 | if (notify_die(DIE_NMIWATCHDOG, msg, regs, 0, 2, SIGINT) == NOTIFY_STOP) | 761 | if (notify_die(DIE_NMIWATCHDOG, str, regs, 0, 2, SIGINT) == NOTIFY_STOP) |
762 | return; | 762 | return; |
763 | 763 | ||
764 | spin_lock(&nmi_print_lock); | 764 | spin_lock(&nmi_print_lock); |
@@ -767,10 +767,12 @@ void notrace __kprobes die_nmi(struct pt_regs *regs, const char *msg) | |||
767 | * to get a message out: | 767 | * to get a message out: |
768 | */ | 768 | */ |
769 | bust_spinlocks(1); | 769 | bust_spinlocks(1); |
770 | printk(KERN_EMERG "%s", msg); | 770 | printk(KERN_EMERG "%s", str); |
771 | printk(" on CPU%d, ip %08lx, registers:\n", | 771 | printk(" on CPU%d, ip %08lx, registers:\n", |
772 | smp_processor_id(), regs->ip); | 772 | smp_processor_id(), regs->ip); |
773 | show_registers(regs); | 773 | show_registers(regs); |
774 | if (do_panic) | ||
775 | panic("Non maskable interrupt"); | ||
774 | console_silent(); | 776 | console_silent(); |
775 | spin_unlock(&nmi_print_lock); | 777 | spin_unlock(&nmi_print_lock); |
776 | bust_spinlocks(0); | 778 | bust_spinlocks(0); |
diff --git a/arch/x86/kernel/traps_64.c b/arch/x86/kernel/traps_64.c index ec6d3b2130c4..80ba6d37bfe0 100644 --- a/arch/x86/kernel/traps_64.c +++ b/arch/x86/kernel/traps_64.c | |||
@@ -75,7 +75,9 @@ asmlinkage void alignment_check(void); | |||
75 | asmlinkage void machine_check(void); | 75 | asmlinkage void machine_check(void); |
76 | asmlinkage void spurious_interrupt_bug(void); | 76 | asmlinkage void spurious_interrupt_bug(void); |
77 | 77 | ||
78 | int panic_on_unrecovered_nmi; | ||
78 | static unsigned int code_bytes = 64; | 79 | static unsigned int code_bytes = 64; |
80 | static unsigned ignore_nmis; | ||
79 | 81 | ||
80 | static inline void conditional_sti(struct pt_regs *regs) | 82 | static inline void conditional_sti(struct pt_regs *regs) |
81 | { | 83 | { |
@@ -613,7 +615,9 @@ die_nmi(char *str, struct pt_regs *regs, int do_panic) | |||
613 | * We are in trouble anyway, lets at least try | 615 | * We are in trouble anyway, lets at least try |
614 | * to get a message out. | 616 | * to get a message out. |
615 | */ | 617 | */ |
616 | printk(str, smp_processor_id()); | 618 | printk(KERN_EMERG "%s", str); |
619 | printk(" on CPU%d, ip %08lx, registers:\n", | ||
620 | smp_processor_id(), regs->ip); | ||
617 | show_registers(regs); | 621 | show_registers(regs); |
618 | if (kexec_should_crash(current)) | 622 | if (kexec_should_crash(current)) |
619 | crash_kexec(regs); | 623 | crash_kexec(regs); |
@@ -862,6 +866,28 @@ asmlinkage notrace __kprobes void default_do_nmi(struct pt_regs *regs) | |||
862 | io_check_error(reason, regs); | 866 | io_check_error(reason, regs); |
863 | } | 867 | } |
864 | 868 | ||
869 | asmlinkage notrace __kprobes void | ||
870 | do_nmi(struct pt_regs *regs, long error_code) | ||
871 | { | ||
872 | nmi_enter(); | ||
873 | add_pda(__nmi_count, 1); | ||
874 | if (!ignore_nmis) | ||
875 | default_do_nmi(regs); | ||
876 | nmi_exit(); | ||
877 | } | ||
878 | |||
879 | void stop_nmi(void) | ||
880 | { | ||
881 | acpi_nmi_disable(); | ||
882 | ignore_nmis++; | ||
883 | } | ||
884 | |||
885 | void restart_nmi(void) | ||
886 | { | ||
887 | ignore_nmis--; | ||
888 | acpi_nmi_enable(); | ||
889 | } | ||
890 | |||
865 | /* runs on IST stack. */ | 891 | /* runs on IST stack. */ |
866 | asmlinkage void __kprobes do_int3(struct pt_regs * regs, long error_code) | 892 | asmlinkage void __kprobes do_int3(struct pt_regs * regs, long error_code) |
867 | { | 893 | { |
diff --git a/include/asm-x86/nmi.h b/include/asm-x86/nmi.h index 05449ef830a7..f0e435dd38fb 100644 --- a/include/asm-x86/nmi.h +++ b/include/asm-x86/nmi.h | |||
@@ -17,12 +17,10 @@ int do_nmi_callback(struct pt_regs *regs, int cpu); | |||
17 | 17 | ||
18 | #ifdef CONFIG_X86_64 | 18 | #ifdef CONFIG_X86_64 |
19 | extern void default_do_nmi(struct pt_regs *); | 19 | extern void default_do_nmi(struct pt_regs *); |
20 | extern void die_nmi(char *str, struct pt_regs *regs, int do_panic); | ||
21 | extern void nmi_watchdog_default(void); | ||
22 | #else | ||
23 | #define nmi_watchdog_default() do {} while (0) | ||
24 | #endif | 20 | #endif |
25 | 21 | ||
22 | extern void die_nmi(char *str, struct pt_regs *regs, int do_panic); | ||
23 | extern void nmi_watchdog_default(void); | ||
26 | extern int check_nmi_watchdog(void); | 24 | extern int check_nmi_watchdog(void); |
27 | extern int nmi_watchdog_enabled; | 25 | extern int nmi_watchdog_enabled; |
28 | extern int avail_to_resrv_perfctr_nmi_bit(unsigned int); | 26 | extern int avail_to_resrv_perfctr_nmi_bit(unsigned int); |