@Changaco @davidrevoy It would have to check the address range *every* time before any allocation at minimum to be somewhat reliable, and that actually wouldn't cover the memory where the code to do those checks in the kernel is itself stored.

Alternatively triple allocation, lookup and so on with on-the-fly comparison.

But that still runs into the issue of corruption of the immediate check storage.

If CPUs exposed the cache explicitly and one trusted them to never get corrupted or flip (why such trust?) then one could have the core memory-checking runtime stored there.

Funny thing there is ECC memory somewhat addresses the case of memory corruption but the case of in-CPU corruption? That takes mainframe-grade CPUs to handle correctly, everything else likes to pretend CPUs are perfect and never do anything wrong (which has been empirically demonstrated to be very much false).