user-space probes -- plan B from outer space

[Frank Ch. Eigler]

Hi -

> Code (and README, tests, bug report) attached.
> The slowdown is still about 14x (over LKML uprobes) [...]

Thanks, well done.  It looks a lot like plan B: no kernel-side changes.

> [...]  It slows down significantly (and often breaks -- see README)
> when probing multiple heavily-probe-burdened processes
> simultaneously. [...]

Unless one can get around it with clever signal manipulation, this may
be a good reason to recast libptp as a multiprocessing system.  (One
would have to experiment to see whether it could be "simply"
multithreaded or a full-blown multi-process widget is required.)  In
any case, it would avoid multiplexing multiple sources of ptrace
SIGCHLD on any particular probing process.

Another major stumbling block will be when attempting to instrument
multithreaded target programs, where single-stepping and opcode
replacement need to be synchronized.  This is a source of complexity
in real debuggers.

- FChE