> 
> The problems with the bounce-buffer stuff that everyone is currently 
> talking about are:
> 
> 	- They are currently _very_ i386-specific.
>
On FreeBSD they ARE in the machine dependent portion of the tree.  The
existing 386 user base would have benefited from that "hack" or whatever
it might be called...

> 
> 	- They make it much harder to convert the drivers that need them
> 	  to a machine-independent DMA mapping system. [*]
> 
The FreeBSD (perhaps machine dependent, maybe not) scheme requires the
addition of only one subroutine call in each supported driver.  Also, there
is some support for DMAable malloc space also.  That subroutine call is
easily #ifdefed out (as it is in FreeBSD.)  Absolutely no driver restructuring
is/was necessary.

I think that the goal of a truly well thought out bounce or dma scheme
is laudable.  However, IMO that isn't an excuse to cause the users to
be disadvantaged until the "ideal" scheme is implemented.

John