Notes from the LPC scheduler microconference

Notes from the LPC scheduler microconference

There's no reason, for instance, user-space doesn't have tables that know that versions prior to X have a specific list of events and a corresponding set of parameters, that versions X through Y have different parameters/values for foo() trace point, that versions Y and onwards differ again in whatever way they do, and so on.

This sort of approach worries me because it seems to assume a single linear set of kernel versions when in fact there's a huge forest of distro kernels which pick and choose from upstream.

So if possible you'd like the interface to be self-describing somehow.