[tlb]

It might be good enough to detect a large number of accesses to any single row and then initiate a complete refresh. This wouldn't be triggered often by normal software. Most exploits have to use cache flush instructions, and with modern several-way-associative caches it would be rare for normal code to trigger it accidentally. In that case, the DRAM maker just has to specify the maximum number of accesses to any row.