Path: utzoo!utgpu!watmath!clyde!att!ncsc1!tsp From: tsp@ncsc1.ATT.COM (Tom Poindexter x8862) Newsgroups: comp.os.minix Subject: Re: (IBM V1.3) tty1 performance? Really Virtual Memory for pc minix Summary: Medium model propsal Keywords: virtual memory Message-ID: <260@ncsc1.ATT.COM> Date: 30 Nov 88 17:49:37 GMT References: <5687@louie.udel.EDU> <760@sdcc15.ucsd.edu> <6149@fluke.COM> Organization: AT&T NCSC, Englewood,CO Lines: 33 In article <6149@fluke.COM>, dcd@tc.fluke.COM (David Dyck) writes: > > John J Marco had a suggestion: > > > ......... (An even better idea would be for > > somebody to implement virtual memory for MINIX so that programs could > > be as large as memory. This would be feasable with AT computers. XTs > > would be quite slow, but at least they would run large programs. > > Back in Jan '88 MBECK@ai.ai.mit.edu (Mark E. Becker) posted > a virtual memory for MINIX. > I have also seen a hack for medium model programs ( code > 64kb, data < 64kb) made to fit in a small model architecture. It was Venix-86 2.0, Venturcom's V7 Unix for IBM PCs circa 1985 or so. The compiler spit out small model code (16 bit pointers), but the linker arranged the code by "pages" of 64kb or less. Each page could contain several functions, and function calls local to that page were handled normally. Function calls to off-page routines went throught some common handler, which looked up the page via a table, then performed a special interrupt to the kernel which would adjust the segment registers. I think that indirect function calls were handled properly (*ptr_to_function)(args). If anyone still has the documentation for this, perhaps you could post a better explanation. I seem to recall that the whole documenation for the medium model was about 1 page in one of the supplementary manuals. Any takers for implementing this? :-) -- Tom Poindexter ...!att!ncsc5!tsp or tsp@ncsc5.att.com