I've found a post here: Small subgroup confinement attack on Diffie-Hellman
which says we can pick $k$ in this way:
And, as we know, $2$ will always be a prime-factor of $p−1$, therefore there will be a subgroup with two elements, that don’t generate anything besides themselves. Obviously the neutral element $1$ is in that subgroup, the other element is $p−1$; In this way, can we just pick $k = {(p-1) \over 2}$ as $w=2$; then, Eva can make sure that the so called shared key must be $1$ or $p-1$?
By the way, I really do some test and I find the result of shared key is really $1$....
But as far as I know, DH is really a secure algorithm, so I think I must have made a mistake; can anybody tell me where I'm mistaken?