Instruction fetch and a load in the same cycle on a unified memory

A classic 5-stage in-order pipeline (IF, ID, EX, MEM, WB) uses a single unified memory with exactly one access port shared by both instruction fetch and data-memory access (no separate instruction/data caches). In some cycle, instruction I is in its MEM stage performing a load, while instruction I+3 is simultaneously trying to use IF to fetch the next instruction. What happens?

1 of 47 in Computer Architecture