tst-backtrace failures on AArch64

[Roland McGrath]

On some other machines, either __foo_nocancel is wholly separate, so that
foo never jumps into it (e.g. arm); or __foo_nocancel appears after foo
(and inside the range of foo's st_size), so that a PC inside __foo_nocancel
is seen as being inside foo (e.g. x86_64).  But on some others, it looks
more like aarch64 (e.g. alpha, s390) and I don't know what makes backtrace
happy for them.