rwessel (rwessel.delete@this.yahoo.com) on October 1, 2021 6:33 am wrote:

> Not really. Stack accesses continue to be relative to SS, (normal) data access to
> DS. But the problem is pointers to both get passed all over the place. Consider:
>
> int global;
> void f()
> {
> int local;
> int *allocated = malloc(sizeof int);
> g(&global, &local, allocated);
> }
>
> If you put the stack (assuming conventional stack allocation on x86) in a different segment
> (or not one aliased with DS), you now have to pass far pointers to g(). The performance overhead
> of that is why (general) large model programs were so disfavored in the 16-bit era. Many programs
> ended up mixed model so that they could use more memory, without that overhead.
>


I try again. Far pointers are segment selectors plus offset so of course using far pointers are avoidable for performance reasons. And if you put stack and data into different segments both addressing start from zero so unified addressing can't be used.

But I386 can also invert segment addressing and I believe that it's also it's default behavior for stack. So in special case of one stack and one data segment stack addressing starts from top of memory(growing down) and is using inverted offset( so max offset is still at top of memory and is also segment max size, 64KB or 4GB) and data starts addressing from bottom of memory(growing up) and as offset starts from zero addressing only overlap where those segments overlaps in linear address space.

Because of that i386 segmentation makes possible to use different segments for data and stack which still can be used through one linear address space - it pretty much can be totally transparent from programming view - and nobody still didn't use it.....


I have to give credits to Intel's MMU-design, that design is still some forty year later pretty awesome.