Revisiting the DOS memory models - by Julio Merino
At the beginning of the year, I wrote a bunch of articles on the various tricks DOS played to overcome the tight memory limits of x86’s real mode. There was one question that came up and remained unanswered: what were the various “models” that the compilers of the day offered? Take a look at Borland Turbo C++’s code generation menu:
Tiny, small, medium, compact, large, huge… What did these options mean? What were their effects? And, more importantly… is any of that legacy relevant today in the world of 64-bit machines and gigabytes of RAM? To answer those questions, we must start with a brief review of the 8086 architecture and the binary formats supported by DOS.
In the 8086 architecture, which is the architecture that DOS targeted, memory references are composed of two parts: a 2-byte segment “identifier” and a 2-byte offset within the segment. These pairs are often expressed as segment:offset.
Segments are contiguous 64KB chunks of memory and are identified by their base address. To be able to address the full 1MB of memory that the 8086 supports, segments are offset from each other by 16 bytes. As you can deduce from this, segments overlap, which means that a specific physical memory position can be referenced by many segment/offset pairs.
