[U-Boot] [PATCH 0/4] Network defrag

[Ben Warren]

Robin Getz wrote:
> On Wed 12 Aug 2009 16:05, Ben Warren pondered:
>   
>> Hi Robin,
>>
>> Robin Getz wrote:
>>     
>>> On Mon 10 Aug 2009 15:57, Ben Warren pondered:
>>>   
>>>       
>>>> Robin Getz wrote:
>>>>     
>>>>         
>>>>> Thanks to Alessandro for putting it together.
>>>>>
>>>>> Feel free to add my Signed-off (once the docs have been updated
>>>>> explaining what this is all for).
>>>>>
>>>>>   
>>>>>       
>>>>>           
>>>> I'll do that.  Thanks for all your help.
>>>>     
>>>>         
>>> Some info for the docs, when I was troubleshooting a Ubuntu 9.04 install.
>>> ======
>>>
>>> It appears that some tftp servers (the older BSD version,
>>> Debian's "tftpd") doesn't support RFC 2348 (blksize), and always use
>>> a block size of 512 :( You  need to make sure that you install the 
>>> the Peter Anvin version, which does support  RFC 2348, and blksize up
>>> to 65,464 bytes (Debian's "tftpd-hpa"):
>>> http://www.kernel.org/pub/software/network/tftp/ 
>>>   
>>>       
>> Maybe it would make sense to have a message printed if the user 
>> specifies a higher blocksize but the TFTP server doesn't respond to the 
>> 'blksize' request.  Thoughts?
>>     
>
> It doesn't happen today...
>
>   
Obviously.  I'm asking if you think it would be helpful.
> We request 1468 (today), and if don't get an respose to the blksize request, 
> it falls back to 512 (with the BSD tftpd) - no message to the user - it just 
> takes longer, and people complain about a crappy network driver...
>
> To me - this is independent of the defragmentation patch. If we request 1468 
> (MTU) or 1469 (which will be fragmented) - the logic in tftp.c is the same...
>
>   
Agreed.  The code to request 'blksize' has been in place for a while, 
but may be more relevant now that we can have blocks MUCH bigger than 
the default.
> There is already a debug("Blocksize ack... in the code - so I'm assuming that 
> if someone wanted to figure it out - they just need to turn on DEBUG in 
> tftp.c - rather than printing it out all the time...
>
>   
Sure, if you don't mind re-compiling.  I think it might be an 
opt-outable message via puts_quiet()
> -Robin
>   
regards,
Ben