Re: [patch] Re: Problem with exiting threads under NPTL

[Linus Torvalds]

On Sun, 14 Dec 2003, Ingo Molnar wrote:
>
> the code is a bit ugly, but it's necessary - a parent can decide _after_
> starting the child that it wants to detach it. (by setting SIGCHLD to
> SIG_IGN. The testcase doesnt do this.) So the only place where we can
> detect the detached-ness of a process is in do_notify_parent().

Hmm.. What if "leader->exit_signal" was -1 already _before_ we call
"do_notify_parent()"? In that case we'd never call "do_notify_parent()"
for the leader at all, and we would also not release it outselves, the way
you've done the test.

Or is that case impossible to trigger? Looks a bit like that. But if it
_is_ impossible to trigger (ie exit_signal cannot be -1 for a thread
leader), then why does the current code test for "&& leader->exit_signal
!= -1)" at all?

That code looks fragile as hell. I think you fixed a bug and it might be
the absolutely proper fix, but I'd be happier about it if it was more
obvious what the rules are and _why_ that is the only case that matters..

		Linus
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/