Frequent Translation Lookaside Buffer (TLB) misses pose significant performance and energy overheads due to page walks required for fetching the translations. The address translation performance bottleneck is further exacerbated by the advent of big data and graph processing workloads due to their massive data footprints. Prefetching page table entries (PTEs) ahead of demand TLB accesses is an intuitively effective approach for alleviating the TLB performance bottleneck. However, each TLB prefetch request implies traversing the page table to fetch the corresponding PTE, triggering additional accesses to the memory hierarchy. Therefore, TLB prefetching is a promising, although costly, technique that may undermine performance when the prefetches are not accurate. This work exploits the locality in the last level of the page table to reduce the cost and enhance the performance benefits of TLB prefetching by prefetching adjacent PTEs “for free”. We design Dynamic Free TLB Prefetching (DFTP), a scheme that predicts via sampling the usefulness of these “free” PTEs and prefetches only the ones most likely to save TLB misses. DFTP can be combined with any TLB prefetcher to provide further performance enhancements by exploiting page table locality for both demand and prefetch page walks.