RedHat/AXP

Thomas Bogendoerfer (tsbogend@bigbug.franken.de)
Mon, 1 Jan 1996 15:38:26 +0100 (MET)


> As far as I know, parity generation is done by the same circuitry,
> independent of whether a PCI memory or PCI I/O access is being
> performed. Thus, I don't think changing from I/O to memory space will
> make much difference w.r.t. bad parity problems.

but I saw only machine checks on IO accesses. When I used IO-mapped
accesses there were two addresses where the machine check happens.
With MEM-mapped accesses there was only one. So I hope the wrong
parity generation occurs only on IO-accesses (perhaps a timing problem
on the NCR chip).

> Have you tried 1.3.52 or some such? The latest ncr driver has the
> mb() instructions in the read/write macros, so it's more likely to
> work, though I haven't tried it myself.

OPTION_MEMORY_MAPPED still doesn't work in 1.3.52.

Thomas.