egisin@orchid.UUCP (09/14/87)
I saw a message in comp.unix.wizards claiming the ST version of Minix implements context switching by copying the active process in and out of a fixed region of memory. Is this true? It would take about 100 millisecs to copy a 100k process in any out. I got the impession from an earlier message that copying was only done for forked (but not exec'd) processes. Can someone clear this up?
ast@cs.vu.nl (Andy Tanenbaum) (09/16/87)
In article <10647@orchid.waterloo.edu> egisin@orchid.waterloo.edu (Eric Gisin) writes: >I saw a message in comp.unix.wizards claiming the ST version >of Minix implements context switching by copying the active process >in and out of a fixed region of memory. Is this true? >It would take about 100 millisecs to copy a 100k process in any out. > True enough, but the average MINIX process is about 5K, not 100K, so we are talking 5 msec, not 100 msec. Also, as soon as the child does an exec, the funny business is finished and the exec'd core image runs normally. After the exec, the parent can be put back where it belongs and both it and the child are not bothered with the relocation stuff any more. In practice, the loss in performance is negligible. This solution was considered better than generating position independent code, which is not very efficient. Andy Tanenbaum (ast@cs.vu.nl)
peter@sugar.UUCP (Peter da Silva) (09/22/87)
Andrew Tannenbaum writes... > True enough, but the average MINIX process is about 5K, not 100K, so we > are talking 5 msec, not 100 msec. Also, as soon as the child does an > exec, the funny business is finished and the exec'd core image runs > normally... What about using the glue chip to relocate logical zero, or do MINIX processes on the ST run in supervisor mode? -- -- Peter da Silva `-_-' ...!hoptoad!academ!uhnix1!sugar!peter -- 'U` Have you hugged your wolf today?
egisin@orchid.UUCP (09/24/87)
In article <790@sugar.UUCP>, peter@sugar.UUCP (Peter da Silva) writes: > What about using the glue chip to relocate logical zero, or do MINIX > processes on the ST run in supervisor mode? What are you talking about? Atari has said the glue chip doesn't do relocation.
jwt@atari.UUCP (Jim Tittsler) (09/24/87)
In article <790@sugar.UUCP>, peter@sugar.UUCP (Peter da Silva) writes: > What about using the glue chip to relocate logical zero, or do MINIX > processes on the ST run in supervisor mode? The Atari ST series have a linear, untranslated address space in both user and supervisor modes. Jim Tittsler, Atari Corporation
peter@sugar.UUCP (Peter da Silva) (09/27/87)
OK, already. The ATARI ST does not do address translation. All I have to say then is that the available documentation is even worse than I thought. Because that's how I interpreted teh Abacus manual. -- -- Peter da Silva `-_-' ...!hoptoad!academ!uhnix1!sugar!peter -- 'U` Have you hugged your wolf today? -- Disclaimer: These aren't mere opinions... these are *values*.
gert@nikhefh.UUCP (09/30/87)
In article <827@sugar.UUCP> peter@sugar.UUCP (Peter da Silva) writes: >OK, already. The ATARI ST does not do address translation. All I have to say >then is that the available documentation is even worse than I thought. Because >that's how I interpreted teh Abacus manual. >-- >-- Peter da Silva `-_-' ...!hoptoad!academ!uhnix1!sugar!peter >-- 'U` Have you hugged your wolf today? >-- Disclaimer: These aren't mere opinions... these are *values*. The Atari ST does not have address translation in a form that would be useful for solving context switch problems. It does however have some form of address translation. In the ST there are one or two memory banks. When two banks are present the ST can handle banks of different size, e.g, bank 0 may hold 512KB and bank 1 may hold 2M. The ''MMU'' then takes care of mapping the banks such that it appears as one contiguous block of memory. (-- This is not according to any book, but according to a disassembly of the ST roms I made ) Gert Poletiek NIKHEF-H, Dutch National Institute for Nuclear and High Energy Physics Kruislaan 409, P.O.Box 41882, 1009 DB Amsterdam, The Netherlands UUCP: {decvax,cernvax,unido,seismo}!mcvax!nikhefh!gert bitnet: nikhefh!gert@mcvax.bitnet, U00025@hasara5.bitnet