I wonder if the following idea matches the scope of the a GSoC
project.
For LLVM there exists a (platform specific) feature called ASAN,
which
provides a good chunk of the functionality of valgrind. As it is
integrated into the compiler, it is quite fast. Since NetBSD is
missing
valgrind support and porting that is quite a lot work, ASAN provide
quite an improvement. I think porting it to userland can be done in a
month at most for a student, including dealing e.g. with libc
overrides for
assembler routines on x86. I also believe a basic working kernel
implementation in the rest of the time is feasible. That would fit
nicely between RUMP and normal kernel debugging as well. Thoughts?