[U-Boot] Fw: TFTP fails when using network switch

[Jerry Van Baren <gerald.vanbaren@ge.com>]

Hi Alex,

Loren A. Linden Levy wrote:
> Hi Everyone,
> 
> I am having an interesting problem with am m5282Lite Dev Kit. I had
> done all of the previous development work by connection the card
> directly to a spare ethernet card in my PC via a cross-over
> cable. This worked just fine and the tftp download was very fast. Now
> I have run a long Ethernet cable and can download the uClinux using
> u-boot, but the hub is 10Mbit. I get some "bad len" errors but
> eventually (seems to take a long time) the download completes and
> uClinux boots. I then tried to hook up a spare 100MBit switch i had
> running one Ethernet cable to the pC (tftp server) and one to the
> coldfire board (now using regular network cable) but the tftp just
> times out.

This sounds like a marginal cabling issue, 10bT (slower, half duplex, 
simple encoding) works, 100bT switched (faster, full duplex, 4b5 
encoding) doesn't.

How long is the cable in your new setup?  What type of cable (patch 
cord, Cat3/Cat4/Cat5, lamp cord)?  Are there any connectors/connections 
between the hub/switch and your Dev Kit?

> The funny thing is i seem to be able to ping the coldfire from the PC
> side but not the PC from the coldfire side. Does anyone have an idea
> of what is wrong? Thanks in advance!

Pinging defaults to small packets at a very slow rate - not very 
stressful.  Try flood pinging with large packets.

Were you able to ping both ways successfully with your original setup 
(short crossover cable)?  Does pinging work differently with the 10bT 
hub vs. 100bT switch?  Is your hardware autonegotiating the full/half 
duplex correctly?  Running one side half and the other side full results 
in lots of "collisions" and runt packets when the full side sends while 
the half side is sending (the half side detects a "collision" and aborts 
the packet).

> Alex

HTH,
gvb