Many real-world networked systems can be modeled as scale-free networks. Due to the robust-yet-fragile nature of scale-free networks, it is vulnerable to the failure of hub nodes, which triggers cascading failures and finally causes the entire network to collapse. In this paper, we study the recovery of scale-free networks when cascading failures occur. We recover the network by repairing failed nodes, and each failed node requires a certain amount of resources to be repaired. A measure named resilience loss is used to quantify the recovery performance. We find that under the constraint of a fixed amount of total recovery resources, there exists an optimal resource allocation strategy to achieve the best network recovery performance, which has the lowest resilience loss. The results may be helpful to understand how real-world scale-free networks recover from cascading failures.