[TUHS] Re: Porting Unix v6 to i386

[Michael Davidson]

Johnny Billquist wrote:
> 
> On Wednesday 30 January 2002 10:18, P.A.Osborne wrote:
> > Having had a rummage and a chat with acolleague here at
> > UKC - it seems that V6 will be easier than V7,  partially because
> > of the Lions commentary - but mainly because 286 protected mode
> > gives a very similar handling on memory management as the PDP did.
> 
> What a silly argument. V6 and V7 both run on the PDP-11, so the memory
> management hardware used by them both are the same.
> 
> And while I'm not familiar with the 286 protected mode, I have heard
> people before claim that it is horrendous compared to the MMU on the
> PDP-11.
> 

Well, I tend to agree that there probably isn't really much difference
in the difficulty of porting either V6 or V7 to an Intel x86 processor.

The real deciding factor will (and should) be which version you *want*
to port and why you want to port it in the first place.

While the Intel architecture is very different from that of the PDP-11
it is possible to set up an IA-32 processor in such a way as to give
you an environment which is quite similar to the one in which V7
expected to run.

16 bit protected mode can effectively give you a 64k+64k split i&d
address space for both user and kernel mode and if you enable paging
then you can map the underlying physical memory in 4096 byte pages
wherever you want it to be. Once it's all set up correctly, most of
the Intel MMU architecture can be ignored and pretty much all the 
kernel has to do is to manipulate 64 entries in a page table (16+16
for user i+d and 16+16 for kernel i+d).

Rather than hack all of the code to build with a modern C compiler
I would look seriously at the possibility of getting either pcc or
the original Ritchie C compiler to generate x86 code. 

(This has, of course, been done before as anyone who ever used
Xenix version 2.x on Intel machines will know - right down to
32 bit longs with PDP-11 word ordering ...)

One possibility which actually isn't as bad as it sounds is to
treat the pdp-11 assembler output of the C compiler as an intermediate 
language and have an additional compiler pass that effectively
converted from pdp11 assembler to x86 assembler.

A good convertor would do enough basic block analysis to be
able to keep track of live registers and condition codes and
would essentially recompile from pdp11 to x86 assembler.
My guess is that with a little care it should be possible to
keep the code expansion resulting from such a translation
process to a minimum.

Even a fairly simple minded conversion probably wouldn't be
too bad.