dom@wcwvax.UUCP (Dominic Sweetman) (11/13/85)
I am a long-standing hater of 8086's, from the time that I worked on one of the first European applications thereof: I could never understand why they didn't make it like a PDP-11 if they couldn't think of anything better. I have read the 386 documentation; although perpetrating some 8086 foolishness, it is basically cured. It is now a proper 32-bit machine with register/memory operations, looks quite fast and has an on-chip MMU. Major complaints are that there are only 6 general purpose registers and that the page tables don't have adequate protection information (can't implement kernel read-only pages, for instance). The segmentation scheme could be ignored; but Intel have realised after only seven years that stacks grow down, so stack segments should be extensible downward. This means that one could use the segmentation scheme to make good the protection deficiencies of the page tables. Incidentally, the National 32000 series MMU does support breakpoints on data references; they are a bit buggy on physical addresses but work OK on virtual addresses. The feature that got taken out was a pair of registers which traced non-sequential instruction fetches to do program flow analysis. I think they took it out because it didn't work, and they had other problems at the time. Dominic Sweetman Whitechapel Computer Works Ltd 75 Whitechapel Road London E1 1DU ENGLAND. uucp: ..seismo!mcvax!ukc!ucl-cs!wcwvax!dom phone:+44 1 377 8680