[TUHS] PDP-7 Unix Paper

Erik E. Fair fair-tuhs at netbsd.org
Wed Feb 14 12:19:41 AEST 2018


That depends on the youngin' - the same technique was also used with the Dual Systems Version 7 Unix port to the mc68000 without an MMU in 1981: split the address space in half at 0x800000, top for kernel, bottom for running process, swap to context switch, and any reference to RAM below 0x800000 while in CPU user mode faulted, to protect the kernel from errant user code.

I can also tell you some S-100 (IEEE-696) device driver horror stories because CP/M-68k and its ilk were not multitasking like Unix and let us say that allowed the S-100 add-in board hardware designers to be ... lazy. The work-arounds in software were very ugly.

Main irritant: customers who claimed our hardware was broken because dereferencing a NULL pointer (0) resulted in a fault, rather than returning 0. "No, your C code is broken because address zero is not in the available address space of processes, and your code should not assume it is." (not a popular response from our customer service engineer). IIRC, most of the Unix code was well-behaved because it had grown up on computers with MMUs by the time we got it ... except for that issue.

Later, Dual produced a CPU board with an mc68451 MMU on it, and that helped. Sort of.

	Erik Fair


More information about the TUHS mailing list