On Wednesday, 26-Mar-2003 at 18:57 GMT, John Levon wrote:
> On Tue, Mar 25, 2003 at 01:42:10PM -0500, Ernie Petrides wrote:
> > In order to track this properly/efficiently, there would need to be a
> > per-cpu counter maintained by do_nmi() as follows:
> As I mentioned in private mail, it would likely need to be a per-counter
> counter too.
It's not clear why a per-profiling-counter count of "extra NMIs expected"
is necessary. The intent is to track the number of NMIs that a cpu is
likely to incur for which nmi_callback() will return 0, with the ultimate
goal being the avoidance of an inappropriate call to default_do_nmi().