[c-nsp] CRC errors on fastethernet interface

[Mike Andrews]

On 11/26/12 1:25 PM, Nick Hilliard wrote:
> On 26/11/2012 17:57, Mikael Abrahamsson wrote:
>> PA-FE and IO-FE does afaik, only support 100 meg (not 10 meg) and only half
>> and full duplex, not autoneg.
> i could never understand this.  the pa-fe and the io-fe both used DEC tulip
> chips, which were perfectly capable of 10/100, hd/fd.  obviously Cisco
> deliberately didn't support this in software, but it was never clear why.
>
Probably because some (most?) revisions of the Tulip had buggy/broken 
auto-negotiation logic.  They're probably the source of most the early 
hardcoding-speed-and-duplex bad practices.  The combo of Tulip to 2924XL 
was particularly awful in my experience and pretty much required 
hardcoding speed/duplex on servers and switches...

They were bad enough that I've seen PCI Tulip NICs in the wild (I think 
SMC made them) that had two RJ45 ports, one for 10 meg and one for 100 
meg.  I suspect the PA-FE-TX took that approach.

DEC made some nice things in their heyday, but the Tulip chip was 
reeeeeeallly not one of them...

I think the PA-2FE-TX has Intel NIC chips.