Does anybody know the reasoning behind having nmi_watchdog turned off by
default on i386 and on by default on x86_64. I've heard that i386 had
problems with false positives in the past, but that local apic watchdog
may make that concern obsolete.
Regards,
Mike Mason
Mike Mason writes:
> Does anybody know the reasoning behind having nmi_watchdog turned off by
> default on i386 and on by default on x86_64. I've heard that i386 had
> problems with false positives in the past, but that local apic watchdog
> may make that concern obsolete.
On i386 the problems are mainly hardware and BIOS. In particular,
lots of Dell laptops have capable hardware but broken BIOSen that
hang the machines if we try to enable anything sending performance
counter interrupts via the local APIC.
/Mikael
Mikael Pettersson <[email protected]> writes:
> Mike Mason writes:
> > Does anybody know the reasoning behind having nmi_watchdog turned off by
> > default on i386 and on by default on x86_64. I've heard that i386 had
> > problems with false positives in the past, but that local apic watchdog
> > may make that concern obsolete.
>
> On i386 the problems are mainly hardware and BIOS. In particular,
> lots of Dell laptops have capable hardware but broken BIOSen that
> hang the machines if we try to enable anything sending performance
> counter interrupts via the local APIC.
AFAIK that trouble was mostly when you forced the local APIC on against
the wishes of the BIOS. That was always a dumb idea and gladly
Linux doesn't try that by default anymore.
-Andi