Linearizability is an important correctness criterion that guarantees the safety of concurrent data structures. Due to the nondeterminism of concurrent executions, reproduction and localization of a linearizability fault still remain challenging. The existing work mainly focuses on model checking the thread schedule space of a concurrent program on a fine-grained (state) level, and hence suffers from the severe problem of state space explosion. This paper presents a tool called CGVT to build a small test case that is sufficient enough for reproducing a linearizability fault. Given a possibly long history that has been detected non-linearizable, CGVT first locates the operations causing a linearizability violation, and then synthesizes a short test case for further investigation. Moreover, we present several optimization techniques to improve the effectiveness and efficiency of CGVT. We have applied CGVT to 10 concurrent objects, while the linearizability of some of the concurrent objects is unknown yet. The experiments show that CGVT is powerful and efficient enough to build the test cases adaptable for a fine-grained analysis.