Originally Posted by
MemberBerry
I have no experience with that, so I can't comment on it, but at least in the x86 world you needed additional hardware to do that, for example DMA controllers. With PCI that changed, each PCI device could take control of the bus when it needed it, unless another device was using it. Of course it's not really that simple, there is still the chipset managing that and preventing conflicts.
There are a lot of variations on this theme, the most recent being I/O devices that can write directly to the CPU cache memory to improve performance even further. Anyway got too carried away, sorry for going off topic.
Most engineers that designed embedded control systems in the 80's and afterwards were weaned on the PDP11 or other DEC machines that used memory-mapped I/O and included it in their designs. Motorola used it on the 6800 and Intel were the ones that promoted separate I/O instructions. Nowadays most control systems use memory-mapped I/O. So as soon as machines were developed that had memory addressing, I/O could not be access in user mode. Maybe memory-mapped I/O was the majority of designs for technical reasons, but I think the fact that you had to buy a $10,000 workstation to develop for the 8008.8080, and Motorola gave away kits of parts for engineers to build and write code on a cross-assembler on their department PDP11 went a long way to giving memory-mapped addressing the edge that it now enjoys.