On 16/12/15 05:35, David Holmes wrote: > It's kind of interesting we've never hit this before. It's very > frustrating that we hit it on a thread that couldn't care less about > the TLS variables that have claimed space on its stack. I guess they > need a notion of "inheritable thread-locals" too.
I think we're rather looking at abuse of TLS here. The obvious right way to use TLS is the way we do it in HotSpot: we have a base pointer to everything that is thread-local. So while this situation is a pain, the right answer to people who want to allocate a big block of TLS is "Don't do that, then." And yes, I know that we have no control over people who launch Java from their own applications. So it's a pain for us, and I am reaching out to glibc maintainers here at Red Hat. Andrew.
