[U-Boot-Users] Re: tftp timeout

[Andre Puschmann]

Jerry Van Baren wrote:
> Andre Puschmann wrote:
> 
>> hallo,
>> i know this newsgroup already talked about this subject and of course i
>> read this stuff but there are still some questions ..
>>
>> i am currently trying to download a kernel image via tftp from a
>> tftp-server on my hostpc which is connected through a ethernet-switch.
>> but i always get timeouts .. look at this:
>>
>> <log>
>> Load address: 0x20800000
>> Loading: *\0x08#######T ##################################T
>> ###################T ####T ##T #
>> \0x09 ###T #T #########T ########T #############T ##T #############T
>> ########T #############T
>> \0x09 #####T ###T ######T #######T #######T #############T ##T
>> ##############T ###########
>> \0x09 ###########
>> done
>> </log>
>>
>> i mean .. the transfer is ok .. but it takes ages :-)
>>
>>
>> it looks like this if i connect my board directly to my hostpc.
>>
>> <log>
>> Load address: 0x20800000
>> Loading: *\0x08T
>> #################################################################
>> \0x09 #################################################################
>> \0x09 #################################################################
>> \0x09 ###########
>> done
>> </log>
>>
>>
>> does anybody got the same issues?
>> i mean there is no huge traffic on the net and my local tftp client 
>> loads the 1MB kernel-image in seconds.
>> i am using u-boot-1.1.1 and i also tried u-boot-1.1.4!
>>
>> regards
>> andre
> 
> 
> My guess is that you have a full duplex/half duplex problem.  Verify 
> that your board/u-boot is setting your ethernet to the proper duplex 
> (full/half).  I'm guessing your board is half duplex but your switch is 
> full (typical of a switch ;-).
> 
> The switch sends traffic to your board while your board is 
> transmitting... that is a collision (late collision at that) to your 
> board but is OK to the switch.  This doesn't happen nearly as much with 
> your PC since you have a dedicated link without much asynchronous traffic.
> 
> This is a FAQ type problem where the software (u-boot/linux) needs to 
> poll the PHY chip for duplex and then (re)configure the MAC chip 
> (separate or built into the CPU) to match.  If the poll isn't happening 
> or has a bug, you have a problem.
> 
> gvb
> 

thanks for your tip ..
sorry i am not that introduced in how autonegotiation works but i guess 
the function dm9161_AutoNegotiate() in dm9161.c does the related stuff.
i mean the switch i am using is a no-name one and with an other 
netg**r-switch it works nice, just tested it ... so it seems to be a bug 
in that fuctions, right??

regards
andre