In this paper, we introduce the concept of "Decentralized Distributed Massive MIMO" (multiple-input multipleoutput), which decentralizes the baseband processing to the network edge and distributes the Massive MIMO (also known as cell-free massive MIMO). In Decentralized Distributed Massive MIMO (DD-mMIMO), the signal processing is performed at edge processing units (EPUs) located closer to the access points (APs) rather than in a single remote central processing unit (CPU), significantly reducing the latency. This avoids cluster edge effects that may arise in cell-free massive MIMO. To avoid this, DD-mMIMO defines coordination regions which may overlap, which implies the APs in the overlap area may be coordinated by more than one EPU. To analyse the performance of the system, we derive an expression of the achievable uplink spectral efficiency (SE) within multiple-antenna APs for various data detection algorithms. The numerical results demonstrate that DD-mMIMO has the potential to fulfill the ultra-dense low-latency requirement in the next generation mobile network and beyond.