Exiting a spin-wait loop on a flag, with and without `pause`

A thread on an Intel x86 core with SMT enabled waits for another core to publish a flag. The loop compiles to the three instructions below with no pause. The flag's line stays in this core's L1D in Shared state while it spins, so every load hits. Eventually another core stores 1 to flag, invalidating this core's copy. What does the spinning core go through when that store arrives, and what would a pause inside the loop change?

; while (flag.load(std::memory_order_acquire) == 0) {}
.spin:
    mov   eax, dword ptr [flag]
    test  eax, eax
    jz    .spin

42 of 47 in Computer Architecture