I've merged these (up to 5/6 so far).
Now I'm looking at all the uses of ->stopped.
* utrace_set_events
could test task_is_stopped_or_traced() || ->exit_state
* utrace_control:
could test task_is_traced() ||
->exit_state && !(->utrace_flags & _UTRACE_DEATH_EVENTS)
(or let utrace_do_stop test the latter)
* finish_utrace_stop
detects signal_wake_up(,1) (SIGKILL)
(And the rest are just the bookkeeping to keep it set/cleared.)
So I think it's now true that the only thing ->stopped is buying us is that
finish_utrace_stop test. Do you agree?
Perhaps we can just replace that with __fatal_signal_pending (or inline it,
that should be a one-insn test), or with a check on SIGNAL_GROUP_KILLED or
suchlike if we had that. (Some test for group-exit-or-exec.)
If we can indeed clean up each of those uses with a different test that is
sufficient for the case, then we can drop the utrace->stopped bit entirely.
That means we actually wouldn't need tracehook_finish_jctl for anything,
AFAICT.
Thanks,
Roland
