Seems that DFPT method in VASP has many defects. It can't include vdw interaction. Now I found it also can't include dipole correction, which leads to a large negative frequency at $\Gamma$ point. I found this problem because two different phases of my structure has almost similar large negative frequency at $\Gamma$ point. Before this I've spent days tuning precision-related parameters.
So does anyone know the reason behind this?