The current policy of always excluding stacks that aren't
owned by the current thread only works well on MPU systems,
where it is very fast to have a reserved MPU region corresponding
to the stack buffer of the current thread.
This unfortunately is not efficient on MMU-based systems, it
would require either a different set of page tables per thread
or an expensive page table update on context switch.
This PR relaxes the documented policy to leave it up to the
implementation whether threads that all belong to the same
memory domain can see each other's stacks.
There's lots of precedent for this, for example on Linux threads
in the same process can all access each other's stacks.
Signed-off-by: Andrew Boie <andrew.p.boie@intel.com>
4abc38b342