Subject: Re: FDDI: PC/BIOS and MTU
To: Chris G. Demetriou <cgd@netbsd.org>
From: Hal Murray <murray@pa.dec.com>
List: current-users
Date: 06/02/1999 02:21:23
> certainly not 'solely'.  I saw it in the context of slots behind
> PPBs.  For whatever reason, the designers thought non-bus-mastering
> slots were 'OK.'

That seems pretty strange to me.  (But there are lots of things in 
the PC world that I don't understand.) 

PCI slots are pretty expensive - after you include the connectors 
and a share of the host bridge.  By the time you get that far, it 
makes sense to toss in a few more bells and whistles and make the 
chip smart enough to do DMA.  Or am I missing something?

Are there any PCI chips that don't do DMA?  Any interesting ones?

-----

Most of the existing drivers turn on the master enable bit.  It might 
be interesting to see which ones don't do that (yet). 

As long as that code is in there, it seems cleaner to me to move 
it all to a central place.  (Maybe with some trick to disable it 
in case there is some driver/card that breaks.) 

A clean alternative would be to rip that code out of all the drivers 
and put a centralized trap in that printed a Fix-your-BIOS message 
when the bit wasn't set in a card known to require it. 

Having the fixup code in half the drivers seems like the only wrong 
way.