> There's a few iffies I still have ... Mostly, I remember having a 
> converstaion with Chuck and Paul about the interaction bewteen the 
> write-back cache on the 4/200-series and DVMA, but I don't remember what 
> we decided the solution was (this came up in a conversation about the 
> 4/200-series and the Xylogics driver...).  In any case, I can't really 

It is probably sufficient to add

	if (vactype == VAC_WRITEBACK)
		cache_flush(...);

just before starting the DMA operation. I also recall that Chuck wanted
to optimize depending on whether you're reading or writing. I don't
know if this can be reliably done. You should be able to drop the
cache-flush after DMA completes though.

In any case, the current set of dvma_*() helper routines need to be
re-done to accommodate IO caches found in sun4/400 and sun4m machines.


-pk