ylo@ngs.fi (Tatu Ylonen) (02/15/91)
> Or map 0xfffff000 page to 0x1ffff000. Thereafter, everything would be > reachable. Any weird problems with this? Yes. You also need to map it at 0xfffff000 - interrupt acknowledge looks for it in virtual address space (found out the hard way, but it is in fact documented in the data sheets). In my opinion, 512 megabytes of usable virtual address space is just barely sufficient. I would have mapped the kernel at 0xc0000000 or in that range, but that turned out to be very inpractical due to cpu brain damage. Now we are limited to about 256 megabytes for user mode virtual address space (it is enough for a machine like pc532, but it is not much). Are problems with sbitib+page faults documented (it sets msr to indicate a read cycle although it is a read-modify-cycle). (I don't have any bug lists for the cpu). Is it documented that movsu and movus use ptb1 even when you have specified that only ptb0 is in use? Smells like a cpu bug to me. BTW, Mach 3.0 now runs on pc532 (but it is not yet reliable). Tatu