2005-04-17 06:20:36 +08:00
|
|
|
/*
|
|
|
|
* NMI watchdog support on APIC systems
|
|
|
|
*
|
|
|
|
* Started by Ingo Molnar <mingo@redhat.com>
|
|
|
|
*
|
|
|
|
* Fixes:
|
|
|
|
* Mikael Pettersson : AMD K7 support for local APIC NMI watchdog.
|
|
|
|
* Mikael Pettersson : Power Management for local APIC NMI watchdog.
|
|
|
|
* Mikael Pettersson : Pentium 4 support for local APIC NMI watchdog.
|
|
|
|
* Pavel Machek and
|
|
|
|
* Mikael Pettersson : PM converted to driver model. Disable/enable API.
|
|
|
|
*/
|
|
|
|
|
|
|
|
#include <linux/delay.h>
|
|
|
|
#include <linux/interrupt.h>
|
|
|
|
#include <linux/module.h>
|
|
|
|
#include <linux/nmi.h>
|
|
|
|
#include <linux/sysdev.h>
|
|
|
|
#include <linux/sysctl.h>
|
2006-06-26 19:57:01 +08:00
|
|
|
#include <linux/percpu.h>
|
2006-09-26 16:52:36 +08:00
|
|
|
#include <linux/kprobes.h>
|
2006-12-07 09:14:01 +08:00
|
|
|
#include <linux/cpumask.h>
|
2007-02-16 17:28:09 +08:00
|
|
|
#include <linux/kernel_stat.h>
|
2007-05-08 15:27:03 +08:00
|
|
|
#include <linux/kdebug.h>
|
2005-04-17 06:20:36 +08:00
|
|
|
|
|
|
|
#include <asm/smp.h>
|
|
|
|
#include <asm/nmi.h>
|
|
|
|
|
|
|
|
#include "mach_traps.h"
|
|
|
|
|
2006-09-30 07:47:55 +08:00
|
|
|
int unknown_nmi_panic;
|
|
|
|
int nmi_watchdog_enabled;
|
|
|
|
|
2007-04-16 16:30:27 +08:00
|
|
|
static cpumask_t backtrace_mask = CPU_MASK_NONE;
|
2007-05-03 01:27:20 +08:00
|
|
|
|
2005-04-17 06:20:36 +08:00
|
|
|
/* nmi_active:
|
2006-09-26 16:52:26 +08:00
|
|
|
* >0: the lapic NMI watchdog is active, but can be disabled
|
|
|
|
* <0: the lapic NMI watchdog has not been set up, and cannot
|
2005-04-17 06:20:36 +08:00
|
|
|
* be enabled
|
2006-09-26 16:52:26 +08:00
|
|
|
* 0: the lapic NMI watchdog is disabled, but can be enabled
|
2005-04-17 06:20:36 +08:00
|
|
|
*/
|
2006-09-26 16:52:26 +08:00
|
|
|
atomic_t nmi_active = ATOMIC_INIT(0); /* oprofile uses this */
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2006-09-26 16:52:26 +08:00
|
|
|
unsigned int nmi_watchdog = NMI_DEFAULT;
|
|
|
|
static unsigned int nmi_hz = HZ;
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2007-05-03 01:27:20 +08:00
|
|
|
static DEFINE_PER_CPU(short, wd_enabled);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2006-12-10 04:33:35 +08:00
|
|
|
static int endflag __initdata = 0;
|
|
|
|
|
2008-01-30 20:30:33 +08:00
|
|
|
#ifdef CONFIG_SMP
|
2005-10-31 06:59:40 +08:00
|
|
|
/* The performance counters used by NMI_LOCAL_APIC don't trigger when
|
|
|
|
* the CPU is idle. To make sure the NMI watchdog really ticks on all
|
|
|
|
* CPUs during the test make them busy.
|
|
|
|
*/
|
|
|
|
static __init void nmi_cpu_busy(void *data)
|
|
|
|
{
|
2006-07-03 15:25:25 +08:00
|
|
|
local_irq_enable_in_hardirq();
|
2005-10-31 06:59:40 +08:00
|
|
|
/* Intentionally don't use cpu_relax here. This is
|
|
|
|
to make sure that the performance counter really ticks,
|
|
|
|
even if there is a simulator or similar that catches the
|
|
|
|
pause instruction. On a real HT machine this is fine because
|
|
|
|
all other CPUs are busy with "useless" delay loops and don't
|
|
|
|
care if they get somewhat less cycles. */
|
2006-12-10 04:33:35 +08:00
|
|
|
while (endflag == 0)
|
|
|
|
mb();
|
2007-11-10 05:39:38 +08:00
|
|
|
}
|
2008-01-30 20:30:33 +08:00
|
|
|
#endif
|
2005-10-31 06:59:40 +08:00
|
|
|
|
[PATCH] check nmi watchdog is broken
A bug against an xSeries system showed up recently noting that the
check_nmi_watchdog() test was failing.
I have been investigating it and discovered in both i386 and x86_64 the
recent change to the routine to use the cpu_callin_map has uncovered a
problem. Prior to that change, on an SMP box, the test was trivally
passing because all cpu's were found to not yet be online, but now with the
callin_map they are discovered, it goes on to test the counter and they
have not yet begun to increment, so it announces a CPU is stuck and bails
out.
On all the systems I have access to test, the announcement of failure is
also bougs... by the time you can login and check /proc/interrupts, the
NMI count is happily incrementing on all CPUs. Its just that the test is
being done too early.
I have tried moving the call to the test around a bit, and it was always
too early. I finally hit on this proposed solution, it delays the routine
via a late_initcall(), seems like the right solution to me.
Signed-off-by: Adrian Bunk <bunk@stusta.de>
Cc: Andi Kleen <ak@muc.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-05-01 23:58:48 +08:00
|
|
|
static int __init check_nmi_watchdog(void)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
2005-10-31 06:59:40 +08:00
|
|
|
unsigned int *prev_nmi_count;
|
2005-04-17 06:20:36 +08:00
|
|
|
int cpu;
|
|
|
|
|
2007-08-15 08:40:35 +08:00
|
|
|
if ((nmi_watchdog == NMI_NONE) || (nmi_watchdog == NMI_DISABLED))
|
2006-09-26 16:52:26 +08:00
|
|
|
return 0;
|
|
|
|
|
|
|
|
if (!atomic_read(&nmi_active))
|
[PATCH] check nmi watchdog is broken
A bug against an xSeries system showed up recently noting that the
check_nmi_watchdog() test was failing.
I have been investigating it and discovered in both i386 and x86_64 the
recent change to the routine to use the cpu_callin_map has uncovered a
problem. Prior to that change, on an SMP box, the test was trivally
passing because all cpu's were found to not yet be online, but now with the
callin_map they are discovered, it goes on to test the counter and they
have not yet begun to increment, so it announces a CPU is stuck and bails
out.
On all the systems I have access to test, the announcement of failure is
also bougs... by the time you can login and check /proc/interrupts, the
NMI count is happily incrementing on all CPUs. Its just that the test is
being done too early.
I have tried moving the call to the test around a bit, and it was always
too early. I finally hit on this proposed solution, it delays the routine
via a late_initcall(), seems like the right solution to me.
Signed-off-by: Adrian Bunk <bunk@stusta.de>
Cc: Andi Kleen <ak@muc.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-05-01 23:58:48 +08:00
|
|
|
return 0;
|
|
|
|
|
2005-10-31 06:59:40 +08:00
|
|
|
prev_nmi_count = kmalloc(NR_CPUS * sizeof(int), GFP_KERNEL);
|
|
|
|
if (!prev_nmi_count)
|
2008-01-22 17:23:01 +08:00
|
|
|
return -1;
|
2005-10-31 06:59:40 +08:00
|
|
|
|
[PATCH] check nmi watchdog is broken
A bug against an xSeries system showed up recently noting that the
check_nmi_watchdog() test was failing.
I have been investigating it and discovered in both i386 and x86_64 the
recent change to the routine to use the cpu_callin_map has uncovered a
problem. Prior to that change, on an SMP box, the test was trivally
passing because all cpu's were found to not yet be online, but now with the
callin_map they are discovered, it goes on to test the counter and they
have not yet begun to increment, so it announces a CPU is stuck and bails
out.
On all the systems I have access to test, the announcement of failure is
also bougs... by the time you can login and check /proc/interrupts, the
NMI count is happily incrementing on all CPUs. Its just that the test is
being done too early.
I have tried moving the call to the test around a bit, and it was always
too early. I finally hit on this proposed solution, it delays the routine
via a late_initcall(), seems like the right solution to me.
Signed-off-by: Adrian Bunk <bunk@stusta.de>
Cc: Andi Kleen <ak@muc.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-05-01 23:58:48 +08:00
|
|
|
printk(KERN_INFO "Testing NMI watchdog ... ");
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2008-01-30 20:30:33 +08:00
|
|
|
#ifdef CONFIG_SMP
|
2005-10-31 06:59:40 +08:00
|
|
|
if (nmi_watchdog == NMI_LOCAL_APIC)
|
|
|
|
smp_call_function(nmi_cpu_busy, (void *)&endflag, 0, 0);
|
2008-01-30 20:30:33 +08:00
|
|
|
#endif
|
2005-10-31 06:59:40 +08:00
|
|
|
|
2006-03-28 17:56:39 +08:00
|
|
|
for_each_possible_cpu(cpu)
|
2008-01-30 20:30:33 +08:00
|
|
|
prev_nmi_count[cpu] = nmi_count(cpu);
|
2005-04-17 06:20:36 +08:00
|
|
|
local_irq_enable();
|
2007-04-02 18:14:12 +08:00
|
|
|
mdelay((20*1000)/nmi_hz); // wait 20 ticks
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2006-03-28 17:56:39 +08:00
|
|
|
for_each_possible_cpu(cpu) {
|
2005-04-17 06:20:36 +08:00
|
|
|
#ifdef CONFIG_SMP
|
|
|
|
/* Check cpu_callin_map here because that is set
|
|
|
|
after the timer is started. */
|
|
|
|
if (!cpu_isset(cpu, cpu_callin_map))
|
|
|
|
continue;
|
|
|
|
#endif
|
2007-05-03 01:27:20 +08:00
|
|
|
if (!per_cpu(wd_enabled, cpu))
|
2006-09-26 16:52:26 +08:00
|
|
|
continue;
|
2005-04-17 06:20:36 +08:00
|
|
|
if (nmi_count(cpu) - prev_nmi_count[cpu] <= 5) {
|
2007-12-05 00:19:07 +08:00
|
|
|
printk(KERN_WARNING "WARNING: CPU#%d: NMI "
|
|
|
|
"appears to be stuck (%d->%d)!\n",
|
2005-10-31 06:59:40 +08:00
|
|
|
cpu,
|
|
|
|
prev_nmi_count[cpu],
|
|
|
|
nmi_count(cpu));
|
2007-05-03 01:27:20 +08:00
|
|
|
per_cpu(wd_enabled, cpu) = 0;
|
2006-09-26 16:52:26 +08:00
|
|
|
atomic_dec(&nmi_active);
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
}
|
2007-09-06 22:59:54 +08:00
|
|
|
endflag = 1;
|
2006-09-26 16:52:26 +08:00
|
|
|
if (!atomic_read(&nmi_active)) {
|
|
|
|
kfree(prev_nmi_count);
|
|
|
|
atomic_set(&nmi_active, -1);
|
2008-01-22 17:23:01 +08:00
|
|
|
return -1;
|
2006-09-26 16:52:26 +08:00
|
|
|
}
|
2005-04-17 06:20:36 +08:00
|
|
|
printk("OK.\n");
|
|
|
|
|
|
|
|
/* now that we know it works we can reduce NMI frequency to
|
|
|
|
something more reasonable; makes a difference in some configs */
|
2007-05-03 01:27:20 +08:00
|
|
|
if (nmi_watchdog == NMI_LOCAL_APIC)
|
|
|
|
nmi_hz = lapic_adjust_nmi_hz(1);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2005-10-31 06:59:40 +08:00
|
|
|
kfree(prev_nmi_count);
|
2005-04-17 06:20:36 +08:00
|
|
|
return 0;
|
|
|
|
}
|
[PATCH] check nmi watchdog is broken
A bug against an xSeries system showed up recently noting that the
check_nmi_watchdog() test was failing.
I have been investigating it and discovered in both i386 and x86_64 the
recent change to the routine to use the cpu_callin_map has uncovered a
problem. Prior to that change, on an SMP box, the test was trivally
passing because all cpu's were found to not yet be online, but now with the
callin_map they are discovered, it goes on to test the counter and they
have not yet begun to increment, so it announces a CPU is stuck and bails
out.
On all the systems I have access to test, the announcement of failure is
also bougs... by the time you can login and check /proc/interrupts, the
NMI count is happily incrementing on all CPUs. Its just that the test is
being done too early.
I have tried moving the call to the test around a bit, and it was always
too early. I finally hit on this proposed solution, it delays the routine
via a late_initcall(), seems like the right solution to me.
Signed-off-by: Adrian Bunk <bunk@stusta.de>
Cc: Andi Kleen <ak@muc.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-05-01 23:58:48 +08:00
|
|
|
/* This needs to happen later in boot so counters are working */
|
|
|
|
late_initcall(check_nmi_watchdog);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
|
|
|
static int __init setup_nmi_watchdog(char *str)
|
|
|
|
{
|
|
|
|
int nmi;
|
|
|
|
|
|
|
|
get_option(&str, &nmi);
|
|
|
|
|
2006-09-26 16:52:26 +08:00
|
|
|
if ((nmi >= NMI_INVALID) || (nmi < NMI_NONE))
|
2005-04-17 06:20:36 +08:00
|
|
|
return 0;
|
2007-01-23 12:40:34 +08:00
|
|
|
|
2006-09-26 16:52:26 +08:00
|
|
|
nmi_watchdog = nmi;
|
2005-04-17 06:20:36 +08:00
|
|
|
return 1;
|
|
|
|
}
|
|
|
|
|
|
|
|
__setup("nmi_watchdog=", setup_nmi_watchdog);
|
|
|
|
|
2007-02-13 20:26:24 +08:00
|
|
|
|
2007-05-03 01:27:20 +08:00
|
|
|
/* Suspend/resume support */
|
2007-02-13 20:26:24 +08:00
|
|
|
|
2005-04-17 06:20:36 +08:00
|
|
|
#ifdef CONFIG_PM
|
|
|
|
|
|
|
|
static int nmi_pm_active; /* nmi_active before suspend */
|
|
|
|
|
2005-04-17 06:25:24 +08:00
|
|
|
static int lapic_nmi_suspend(struct sys_device *dev, pm_message_t state)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
2006-09-26 16:52:27 +08:00
|
|
|
/* only CPU0 goes here, other CPUs should be offline */
|
2006-09-26 16:52:26 +08:00
|
|
|
nmi_pm_active = atomic_read(&nmi_active);
|
2006-09-26 16:52:27 +08:00
|
|
|
stop_apic_nmi_watchdog(NULL);
|
|
|
|
BUG_ON(atomic_read(&nmi_active) != 0);
|
2005-04-17 06:20:36 +08:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
static int lapic_nmi_resume(struct sys_device *dev)
|
|
|
|
{
|
2006-09-26 16:52:27 +08:00
|
|
|
/* only CPU0 goes here, other CPUs should be offline */
|
|
|
|
if (nmi_pm_active > 0) {
|
|
|
|
setup_apic_nmi_watchdog(NULL);
|
|
|
|
touch_nmi_watchdog();
|
|
|
|
}
|
2005-04-17 06:20:36 +08:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
static struct sysdev_class nmi_sysclass = {
|
2007-12-20 09:09:39 +08:00
|
|
|
.name = "lapic_nmi",
|
2005-04-17 06:20:36 +08:00
|
|
|
.resume = lapic_nmi_resume,
|
|
|
|
.suspend = lapic_nmi_suspend,
|
|
|
|
};
|
|
|
|
|
|
|
|
static struct sys_device device_lapic_nmi = {
|
|
|
|
.id = 0,
|
|
|
|
.cls = &nmi_sysclass,
|
|
|
|
};
|
|
|
|
|
|
|
|
static int __init init_lapic_nmi_sysfs(void)
|
|
|
|
{
|
|
|
|
int error;
|
|
|
|
|
2006-09-26 16:52:26 +08:00
|
|
|
/* should really be a BUG_ON but b/c this is an
|
|
|
|
* init call, it just doesn't work. -dcz
|
|
|
|
*/
|
|
|
|
if (nmi_watchdog != NMI_LOCAL_APIC)
|
|
|
|
return 0;
|
|
|
|
|
2007-05-03 01:27:20 +08:00
|
|
|
if (atomic_read(&nmi_active) < 0)
|
2005-04-17 06:20:36 +08:00
|
|
|
return 0;
|
|
|
|
|
|
|
|
error = sysdev_class_register(&nmi_sysclass);
|
|
|
|
if (!error)
|
|
|
|
error = sysdev_register(&device_lapic_nmi);
|
|
|
|
return error;
|
|
|
|
}
|
|
|
|
/* must come after the local APIC's device_initcall() */
|
|
|
|
late_initcall(init_lapic_nmi_sysfs);
|
|
|
|
|
|
|
|
#endif /* CONFIG_PM */
|
|
|
|
|
2007-05-03 01:27:20 +08:00
|
|
|
static void __acpi_nmi_enable(void *__unused)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
2007-05-03 01:27:20 +08:00
|
|
|
apic_write_around(APIC_LVT0, APIC_DM_NMI);
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
|
2007-05-03 01:27:20 +08:00
|
|
|
/*
|
|
|
|
* Enable timer based NMIs on all CPUs:
|
|
|
|
*/
|
|
|
|
void acpi_nmi_enable(void)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
2007-05-03 01:27:20 +08:00
|
|
|
if (atomic_read(&nmi_active) && nmi_watchdog == NMI_IO_APIC)
|
|
|
|
on_each_cpu(__acpi_nmi_enable, NULL, 0, 1);
|
2006-09-26 16:52:26 +08:00
|
|
|
}
|
|
|
|
|
2007-05-03 01:27:20 +08:00
|
|
|
static void __acpi_nmi_disable(void *__unused)
|
2006-09-26 16:52:27 +08:00
|
|
|
{
|
2007-05-03 01:27:20 +08:00
|
|
|
apic_write(APIC_LVT0, APIC_DM_NMI | APIC_LVT_MASKED);
|
2006-09-26 16:52:27 +08:00
|
|
|
}
|
|
|
|
|
2007-05-03 01:27:20 +08:00
|
|
|
/*
|
|
|
|
* Disable timer based NMIs on all CPUs:
|
|
|
|
*/
|
|
|
|
void acpi_nmi_disable(void)
|
2006-09-26 16:52:27 +08:00
|
|
|
{
|
2007-05-03 01:27:20 +08:00
|
|
|
if (atomic_read(&nmi_active) && nmi_watchdog == NMI_IO_APIC)
|
|
|
|
on_each_cpu(__acpi_nmi_disable, NULL, 0, 1);
|
2006-09-26 16:52:27 +08:00
|
|
|
}
|
|
|
|
|
2008-01-30 20:30:33 +08:00
|
|
|
void setup_apic_nmi_watchdog(void *unused)
|
2006-09-26 16:52:26 +08:00
|
|
|
{
|
2007-05-03 01:27:20 +08:00
|
|
|
if (__get_cpu_var(wd_enabled))
|
2008-01-30 20:30:33 +08:00
|
|
|
return;
|
2006-09-26 16:52:27 +08:00
|
|
|
|
|
|
|
/* cheap hack to support suspend/resume */
|
|
|
|
/* if cpu0 is not active neither should the other cpus */
|
|
|
|
if ((smp_processor_id() != 0) && (atomic_read(&nmi_active) <= 0))
|
|
|
|
return;
|
|
|
|
|
2007-05-03 01:27:20 +08:00
|
|
|
switch (nmi_watchdog) {
|
|
|
|
case NMI_LOCAL_APIC:
|
|
|
|
__get_cpu_var(wd_enabled) = 1; /* enable it before to avoid race with handler */
|
|
|
|
if (lapic_watchdog_init(nmi_hz) < 0) {
|
|
|
|
__get_cpu_var(wd_enabled) = 0;
|
2006-09-26 16:52:26 +08:00
|
|
|
return;
|
|
|
|
}
|
2007-05-03 01:27:20 +08:00
|
|
|
/* FALL THROUGH */
|
|
|
|
case NMI_IO_APIC:
|
|
|
|
__get_cpu_var(wd_enabled) = 1;
|
|
|
|
atomic_inc(&nmi_active);
|
2006-09-26 16:52:26 +08:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2006-09-26 16:52:27 +08:00
|
|
|
void stop_apic_nmi_watchdog(void *unused)
|
2006-09-26 16:52:26 +08:00
|
|
|
{
|
|
|
|
/* only support LOCAL and IO APICs for now */
|
|
|
|
if ((nmi_watchdog != NMI_LOCAL_APIC) &&
|
|
|
|
(nmi_watchdog != NMI_IO_APIC))
|
|
|
|
return;
|
2007-05-03 01:27:20 +08:00
|
|
|
if (__get_cpu_var(wd_enabled) == 0)
|
2006-09-26 16:52:27 +08:00
|
|
|
return;
|
2007-05-03 01:27:20 +08:00
|
|
|
if (nmi_watchdog == NMI_LOCAL_APIC)
|
|
|
|
lapic_watchdog_stop();
|
|
|
|
__get_cpu_var(wd_enabled) = 0;
|
2006-09-26 16:52:26 +08:00
|
|
|
atomic_dec(&nmi_active);
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* the best way to detect whether a CPU has a 'hard lockup' problem
|
|
|
|
* is to check it's local APIC timer IRQ counts. If they are not
|
|
|
|
* changing then that CPU has some problem.
|
|
|
|
*
|
|
|
|
* as these watchdog NMI IRQs are generated on every CPU, we only
|
|
|
|
* have to check the current processor.
|
|
|
|
*
|
|
|
|
* since NMIs don't listen to _any_ locks, we have to be extremely
|
|
|
|
* careful not to rely on unsafe variables. The printk might lock
|
|
|
|
* up though, so we have to break up any console locks first ...
|
|
|
|
* [when there will be more tty-related locks, break them up
|
|
|
|
* here too!]
|
|
|
|
*/
|
|
|
|
|
|
|
|
static unsigned int
|
|
|
|
last_irq_sums [NR_CPUS],
|
|
|
|
alert_counter [NR_CPUS];
|
|
|
|
|
2007-07-17 19:03:57 +08:00
|
|
|
void touch_nmi_watchdog(void)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
2006-12-07 09:14:09 +08:00
|
|
|
if (nmi_watchdog > 0) {
|
|
|
|
unsigned cpu;
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2006-12-07 09:14:09 +08:00
|
|
|
/*
|
|
|
|
* Just reset the alert counters, (other CPUs might be
|
|
|
|
* spinning on locks we hold):
|
|
|
|
*/
|
2007-07-17 19:03:57 +08:00
|
|
|
for_each_present_cpu(cpu) {
|
|
|
|
if (alert_counter[cpu])
|
|
|
|
alert_counter[cpu] = 0;
|
|
|
|
}
|
2006-12-07 09:14:09 +08:00
|
|
|
}
|
2005-09-07 06:16:27 +08:00
|
|
|
|
|
|
|
/*
|
|
|
|
* Tickle the softlockup detector too:
|
|
|
|
*/
|
|
|
|
touch_softlockup_watchdog();
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
2006-07-30 18:03:29 +08:00
|
|
|
EXPORT_SYMBOL(touch_nmi_watchdog);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
|
|
|
extern void die_nmi(struct pt_regs *, const char *msg);
|
|
|
|
|
2006-09-26 16:52:36 +08:00
|
|
|
__kprobes int nmi_watchdog_tick(struct pt_regs * regs, unsigned reason)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Since current_thread_info()-> is always on the stack, and we
|
|
|
|
* always switch the stack NMI-atomically, it's safe to use
|
|
|
|
* smp_processor_id().
|
|
|
|
*/
|
2006-03-28 17:56:52 +08:00
|
|
|
unsigned int sum;
|
2006-09-26 16:52:26 +08:00
|
|
|
int touched = 0;
|
2006-03-28 17:56:52 +08:00
|
|
|
int cpu = smp_processor_id();
|
2008-01-30 20:30:33 +08:00
|
|
|
int rc = 0;
|
2006-09-26 16:52:26 +08:00
|
|
|
|
|
|
|
/* check for other users first */
|
|
|
|
if (notify_die(DIE_NMI, "nmi", regs, reason, 2, SIGINT)
|
|
|
|
== NOTIFY_STOP) {
|
2006-09-26 16:52:26 +08:00
|
|
|
rc = 1;
|
2006-09-26 16:52:26 +08:00
|
|
|
touched = 1;
|
|
|
|
}
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2006-12-07 09:14:01 +08:00
|
|
|
if (cpu_isset(cpu, backtrace_mask)) {
|
|
|
|
static DEFINE_SPINLOCK(lock); /* Serialise the printks */
|
|
|
|
|
|
|
|
spin_lock(&lock);
|
|
|
|
printk("NMI backtrace for cpu %d\n", cpu);
|
|
|
|
dump_stack();
|
|
|
|
spin_unlock(&lock);
|
|
|
|
cpu_clear(cpu, backtrace_mask);
|
|
|
|
}
|
|
|
|
|
2007-02-16 17:28:09 +08:00
|
|
|
/*
|
|
|
|
* Take the local apic timer and PIT/HPET into account. We don't
|
|
|
|
* know which one is active, when we have highres/dyntick on
|
|
|
|
*/
|
2007-10-13 05:04:06 +08:00
|
|
|
sum = per_cpu(irq_stat, cpu).apic_timer_irqs +
|
|
|
|
per_cpu(irq_stat, cpu).irq0_irqs;
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2007-02-16 17:28:09 +08:00
|
|
|
/* if the none of the timers isn't firing, this cpu isn't doing much */
|
2006-09-26 16:52:26 +08:00
|
|
|
if (!touched && last_irq_sums[cpu] == sum) {
|
2005-04-17 06:20:36 +08:00
|
|
|
/*
|
|
|
|
* Ayiee, looks like this CPU is stuck ...
|
|
|
|
* wait a few IRQs (5 seconds) before doing the oops ...
|
|
|
|
*/
|
|
|
|
alert_counter[cpu]++;
|
|
|
|
if (alert_counter[cpu] == 5*nmi_hz)
|
2005-09-04 06:56:48 +08:00
|
|
|
/*
|
|
|
|
* die_nmi will return ONLY if NOTIFY_STOP happens..
|
|
|
|
*/
|
2006-03-23 19:00:54 +08:00
|
|
|
die_nmi(regs, "BUG: NMI Watchdog detected LOCKUP");
|
2006-03-08 13:55:29 +08:00
|
|
|
} else {
|
2005-04-17 06:20:36 +08:00
|
|
|
last_irq_sums[cpu] = sum;
|
|
|
|
alert_counter[cpu] = 0;
|
|
|
|
}
|
2006-09-26 16:52:26 +08:00
|
|
|
/* see if the nmi watchdog went off */
|
2007-05-03 01:27:20 +08:00
|
|
|
if (!__get_cpu_var(wd_enabled))
|
|
|
|
return rc;
|
|
|
|
switch (nmi_watchdog) {
|
|
|
|
case NMI_LOCAL_APIC:
|
|
|
|
rc |= lapic_wd_event(nmi_hz);
|
|
|
|
break;
|
|
|
|
case NMI_IO_APIC:
|
|
|
|
/* don't know how to accurately check for this.
|
|
|
|
* just assume it was a watchdog timer interrupt
|
|
|
|
* This matches the old behaviour.
|
|
|
|
*/
|
|
|
|
rc = 1;
|
|
|
|
break;
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
2006-09-26 16:52:26 +08:00
|
|
|
return rc;
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
#ifdef CONFIG_SYSCTL
|
|
|
|
|
|
|
|
static int unknown_nmi_panic_callback(struct pt_regs *regs, int cpu)
|
|
|
|
{
|
|
|
|
unsigned char reason = get_nmi_reason();
|
|
|
|
char buf[64];
|
|
|
|
|
2006-09-26 16:52:27 +08:00
|
|
|
sprintf(buf, "NMI received for unknown reason %02x\n", reason);
|
|
|
|
die_nmi(regs, buf);
|
2005-04-17 06:20:36 +08:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2006-09-26 16:52:27 +08:00
|
|
|
/*
|
2006-09-26 16:52:27 +08:00
|
|
|
* proc handler for /proc/sys/kernel/nmi
|
2006-09-26 16:52:27 +08:00
|
|
|
*/
|
|
|
|
int proc_nmi_enabled(struct ctl_table *table, int write, struct file *file,
|
|
|
|
void __user *buffer, size_t *length, loff_t *ppos)
|
|
|
|
{
|
|
|
|
int old_state;
|
|
|
|
|
|
|
|
nmi_watchdog_enabled = (atomic_read(&nmi_active) > 0) ? 1 : 0;
|
|
|
|
old_state = nmi_watchdog_enabled;
|
|
|
|
proc_dointvec(table, write, file, buffer, length, ppos);
|
|
|
|
if (!!old_state == !!nmi_watchdog_enabled)
|
|
|
|
return 0;
|
|
|
|
|
2007-08-15 08:40:35 +08:00
|
|
|
if (atomic_read(&nmi_active) < 0 || nmi_watchdog == NMI_DISABLED) {
|
2006-09-26 16:52:27 +08:00
|
|
|
printk( KERN_WARNING "NMI watchdog is permanently disabled\n");
|
|
|
|
return -EIO;
|
2006-09-26 16:52:27 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
if (nmi_watchdog == NMI_DEFAULT) {
|
2007-05-03 01:27:20 +08:00
|
|
|
if (lapic_watchdog_ok())
|
2006-09-26 16:52:27 +08:00
|
|
|
nmi_watchdog = NMI_LOCAL_APIC;
|
|
|
|
else
|
|
|
|
nmi_watchdog = NMI_IO_APIC;
|
|
|
|
}
|
|
|
|
|
2006-09-26 16:52:27 +08:00
|
|
|
if (nmi_watchdog == NMI_LOCAL_APIC) {
|
2006-09-26 16:52:27 +08:00
|
|
|
if (nmi_watchdog_enabled)
|
|
|
|
enable_lapic_nmi_watchdog();
|
|
|
|
else
|
|
|
|
disable_lapic_nmi_watchdog();
|
|
|
|
} else {
|
|
|
|
printk( KERN_WARNING
|
|
|
|
"NMI watchdog doesn't know what hardware to touch\n");
|
|
|
|
return -EIO;
|
|
|
|
}
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2005-04-17 06:20:36 +08:00
|
|
|
#endif
|
|
|
|
|
2008-02-03 15:40:30 +08:00
|
|
|
int do_nmi_callback(struct pt_regs *regs, int cpu)
|
|
|
|
{
|
|
|
|
#ifdef CONFIG_SYSCTL
|
|
|
|
if (unknown_nmi_panic)
|
|
|
|
return unknown_nmi_panic_callback(regs, cpu);
|
|
|
|
#endif
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2006-12-07 09:14:01 +08:00
|
|
|
void __trigger_all_cpu_backtrace(void)
|
|
|
|
{
|
|
|
|
int i;
|
|
|
|
|
|
|
|
backtrace_mask = cpu_online_map;
|
|
|
|
/* Wait for up to 10 seconds for all CPUs to do the backtrace */
|
|
|
|
for (i = 0; i < 10 * 1000; i++) {
|
|
|
|
if (cpus_empty(backtrace_mask))
|
|
|
|
break;
|
|
|
|
mdelay(1);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2005-04-17 06:20:36 +08:00
|
|
|
EXPORT_SYMBOL(nmi_active);
|
|
|
|
EXPORT_SYMBOL(nmi_watchdog);
|