Re: can kprobes be modular?

[Roland McGrath <roland at redhat dot com>]

> Well, there's at least one call to kprobes from deep in the kernel:
> finish_task_switch() calls kprobe_flush_task().

Ah, that's the kind of gotcha I was looking for.  This appears to be just
a destructor hook for task-associated data when a task_struct is cleaned
up.  I don't see a strong reason why it needs to be done in
finish_task_switch rather than later like in __put_task_struct (which is
after a zombie is reaped, but that isn't usually forever).  With
CONFIG_PROFILING=y, task_handoff_register is already exported and can be
used for that.

> There may be other gotchas.  As far as I know, nobody has investigated
> the idea for quite a while.

For now I'm just seeking all the concrete gotchas people can think of.  I
haven't seen a concrete issue so far that didn't seem trivially addressed
in a cleaner and more modular fashion than #ifdef CONFIG_KPROBES.


Thanks,
Roland