i386: Enable NMI watchdog by default does not work as intended

From: Keith Owens
Date: Thu Oct 26 2006 - 02:03:05 EST


This patch does not work as intended.

commit 1de84979dfc527c422abf63f27beabe43892989b
Author: Andi Kleen <ak@xxxxxxx>
Date: Tue Sep 26 10:52:27 2006 +0200

[PATCH] i386: Enable NMI watchdog by default

diff --git a/arch/i386/kernel/nmi.c b/arch/i386/kernel/nmi.c
index 8e4ed93..6e5085d 100644
--- a/arch/i386/kernel/nmi.c
+++ b/arch/i386/kernel/nmi.c
@@ -21,6 +21,7 @@ #include <linux/nmi.h>
#include <linux/sysdev.h>
#include <linux/sysctl.h>
#include <linux/percpu.h>
+#include <linux/dmi.h>

#include <asm/smp.h>
#include <asm/nmi.h>
@@ -204,6 +205,14 @@ static int __init check_nmi_watchdog(voi
unsigned int *prev_nmi_count;
int cpu;

+ /* Enable NMI watchdog for newer systems.
+ Actually it should be safe for most systems before 2004 too except
+ for some IBM systems that corrupt registers when NMI happens
+ during SMM. Unfortunately we don't have more exact information
+ on these and use this coarse check. */
+ if (nmi_watchdog == NMI_DEFAULT && dmi_get_year(DMI_BIOS_DATE) >= 2004)
+ nmi_watchdog = NMI_LOCAL_APIC;
+
if ((nmi_watchdog == NMI_NONE) || (nmi_watchdog == NMI_DEFAULT))
return 0;

It attempts to change NMI_DEFAULT to NMI_LOCAL_APIC for recent BIOS
versions, even when the user does _not_ specify nmi_watchdog= on the
command line. However a few lines further down is this test which
fails because nmi_active is 0.

if (!atomic_read(&nmi_active))
return 0;

The calling sequence is setup_local_APIC() -> setup_apic_nmi_watchdog()
-> atomic_inc(nmi_active), but only if nmi_watchdog contains
NMI_LOCAL_APIC or NMI_IO_APIC. Without a command line option,
setup_apic_nmi_watchdog() does nothing, nmi_active is still 0 and
check_nmi_watchdog() exits early. Only if you specify nmi_watchdog= on
the command line so nmi_active is non-zero before check_nmi_watchdog()
is called do you get an NMI handler. IOW, the automatic activation of
the NMI handler is too late.

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/