On Mon, 01 Sep 1997 16:07:22 -0700 
 Jonathan Stone <jonathan@DSG.Stanford.EDU> wrote:

 > True.  But for the 3c905, that' sa limitation of our driver, not the
 > hardware. We now have documentation on the 3Com 3c905 DMA, which (as
 > far as I can tell) does not have the four-byte-alignment restriction
 > which forces the current DEC cards to use an extra copy.
 > 
 > By the time the `fixed' DEC chips (21143?) come out, a 3c905 driver
 > might just be faster than the 2114x available now.
 > 
 > And if these machines are going to be on 10Mbit Ethernet (was 21x4x a
 > typo for 2114x?), it won't make a signficiant difference either way.

The copy isn't strictly required on arch's that can do unaligned access,
e.g. the i386, although aligned access is always certainly more optimal.

I'm pretty sure that the Tulip driver could be made to detect the buggy
chips and only do the copy on those revisions, otherwise doing a 2-byte
fudge, much like I implemented in the "fxp" driver when porting it from
FreeBSD.

Jason R. Thorpe                                       thorpej@nas.nasa.gov
NASA Ames Research Center                            Home: +1 408 866 1912
NAS: M/S 258-6                                       Work: +1 415 604 0935
Moffett Field, CA 94035                             Pager: +1 415 428 6939