fix misalignment of dtv in static-linked programs with odd-sized TLS

both static and dynamic linked versions of the __copy_tls function
have a hidden assumption that the alignment of the beginning or end of
the memory passed is suitable for storing an array of pointers for the
dtv. pthread_create satisfies this requirement except when
libc.tls_size is misaligned, which cannot happen with dynamic linking
due to way update_tls_size computes the total size, but could happen
with static linking and odd-sized TLS.
1 file changed