> The point that Ulrich was making was, I think, that Linux should do a
> read-modify-read cycle on the highest address *that was specified as
> legal* either by the BIOS or by command-line arguments. That will make
> problems show up earlier.
i would suggest to do this after mem_init(), since thats the last chance
to completely lock a memory area. (for any weird reason like a device mapped
into valid main memory or a broken motherboard).
-- mingo