From theory, since the return loss is the Log of the ratio Pin/Preflected, it has to be a positive number, as FvM said. However for historical reasons, sometimes (not so rarely) it is represented as a negative number, that is the Log of Preflected/Pin.
Your datasheet uses the last definition even if not strictly correct. Howeverer the RF3858 has an internal LNA that can be by-passed. When it's working as LNA its gain is about 21 dB with input and output return loss of about 8 dB. When it is in by-pass (you can use this feature to receive high level signal in order not to saturate the LNA) the gain is negative (that is a loss). Nothing is said about return loss under this last condition.
Of course you can choose (if you have to receive the signal in a wide dynamic range) to use an LNA combined with two external switches to a fixed attenuator, so in case of low signal you can pass it through the LNA and in case of large signal you can pass it through the attenuator. But in this case the component count will be higher as well as the needed PCB area.