Messages in this thread | | | Date | Thu, 18 Nov 2010 16:16:38 -0500 | From | Don Zickus <> | Subject | Re: [V2 PATCH 0/6] x86, NMI: give NMI handler a face-lift |
| |
On Fri, Nov 19, 2010 at 12:01:59AM +0300, Cyrill Gorcunov wrote: > On Thu, Nov 18, 2010 at 03:52:03PM -0500, Don Zickus wrote: > ... > > > > > > ok, Don, but you mentioned there are false alarms on real P4 machine, right? > > > > Yeah, there are two problems. One is using kgdb tests on kvm guests. The > > other is using kgdb tests on a bare metal p4 machine. > > > > Cheers, > > Don > > > > ok, thanks, Jason just confirmed it too. I thing if we run with kgdb armed > it (kgdb) should obtain nmi handler first and perf only after. > > actually I'm not sure why p4 hangs here while core passes the test, most
because p4 reads the wrong high register which comes back zero. This will always set overflow=1, thus swallowing all NMI if something like the nmi watchdog has perf active.
Cheers, Don
| |