It's easy to express probe points that match hundreds or thousands of places in
the kernel. This can generate tons of C code, which takes tons of time to
compile. Investigate ways of reducing the penalty for big probe sessions.
Possibilities include elimination of duplicate probe bodies (thus reducing
number of emitted functions), simplifying emitted code (optimizing?), reducing
compiler optimization selection.
one chunk of code committed, investigation continuing
actually, new code seems so good that further improvements may not be urgent;
I think this is mostly as efficient as it can get. Tolerating a large number of
*different* probes (with their own probe bodies) is a different matter; but I
think we have it tolerating a single probe placed at a large number of locations.
*** This bug has been marked as a duplicate of 2060 ***