This paper investigates the problem of joint phase tracking and channel decoding in OFDM based Physical-layer Network Coding (PNC) systems. OFDM signaling can obviate the need for tight time synchronization among multiple simultaneous transmissions in the uplink of PNC systems. However, OFDM PNC systems are susceptible to phase drifts caused by residual carrier frequency offsets (CFOs). In the traditional OFDM system in which a receiver receives from only one transmitter, pilot tones are employed to aid phase tracking. In OFDM PNC systems, multiple transmitters transmit to a receiver, and these pilot tones must be shared among the multiple transmitters. This reduces the number of pilots that can be used by each transmitting node. Phase tracking in OFDM PNC is more challenging as a result. To overcome the degradation due to the reduced number of per-node pilots, this work supplements the pilots with the channel information contained in the data. In particular, we propose to solve the problems of phase tracking and channel decoding jointly. Our solution consists of the use of the expectation-maximization (EM) algorithm for phase tracking and the use of the belief propagation (BP) algorithm for channel decoding. The two problems are solved jointly through iterative processing between the EM and BP algorithms. Simulations and real experiments based on software-defined radio show that the proposed method can improve phase tracking as well as channel decoding performance.