On Tue, Jun 26, 2018 at 11:19:37AM -0700, Paul E. McKenney wrote: > The initial reason for ____cacheline_internodealigned_in_smp was that > some of the fields can be accessed by random CPUs, while others are > used more locally, give or take our usual contention over the handling > of CPU numbers. ;-)
So that whole internode thing only matters for the insane VSMP case, where they take node to mean a cluster node, not a numa node. VSMP is a networked shared memory machine where, by necessity, the MESI protocol operates on page granularity. In general I tend to completely and utterly ignore that and let the VSMP people worry about things.
