> On every sun4m, including the Hypersparc, this never causes a
> fault trap.  On every sun4m *except* the Hypersparc, this also
> never saves away a fault type.  On the Hypersparc, it does.
> 

Well, that's good to know, though it's probably not the reason why
NetBSD fails on the Hypersparc at this moment, since I'm not currently
using the probe ASI to see whether a translation exists except in
one place (on a `SFSR_FT_TRANSERR' fault, which does not appear to
occur). But I'll fix this case anyway.

-pk

PS I did notice something else though that might induce lossage:
   The cache-bit in IOMMU PTEs is always on even though bus_dma_load()
   doesn't (yet) care to allocate cache-aligned DVMA addresses for
   the kernel buffer it's mapping. Perhaps this causes cache lossage
   even when cache flushes are done as if the machine hadn't a physically
   tagged cache (Erik's last test kernel ran in that mode).