> I do not know what to do with ptrace_detach()->wake_up_process(). It and ptrace_resume()->wake_up_process() are both clearly wrong in general. I think all we really need to worry about is making sure that the ptrace-tests suite has tests that express all the assumptions and requirements of existing applications (i.e. old gdb versions). If those are inherently contradictory then we can work out the right compromises and get the suite to test "as compatible as we want it" manifest ptrace behavior. Then as long as we don't regress there, our kernel changes are OK.
I think the mayhem of poor bookkeeping might make it too much hassle to try to fix the existing upstream code to use the proper conditional wake-ups. It's not worth putting time into it if that's so. Thanks, Roland