[lwip-users] Re: curious large packets and transmit stuck

[Andre Puschmann]

Kieran Mansley wrote:
> On Mon, 2007-01-29 at 20:06 +0100, Andre Puschmann wrote:
>> hello guys,
>> i did a port of lwip to our custom mpc5200 board.
>> transmission runs quite fast so far, but i still have stability problems.
>> i am not sure if this is a lwip related problem or if it comes from
>> lower levels, e.g. fec driver, interrupt handling or even dma problems.
>>
>> with my lwip app i constantly transmit 1000Byte long packets. this
>> works quite well up to a certain point. sometimes i even transmit 3GB of
>> data until it starts to stuck, but sometimes it hangs after 10MB.
> 
> At what level are you transmitting these packets?  Is it an application
> doing 1000 byte writes to the network, or do you have something low
> level that is output the packets directly?
> 

currently i am using the netconn-*-api .. which means the 1000byte limit
is just on application side! my ethernet drivers send method takes a
whole pbuf. so no limits on that side.

>> in my ethernet transmit function i did some sanity checks, doing so
>> showed up that lwip sometimes tries to transmit packets larger than
>> 1300Byte which shouldn't be the case.
> 
> If an application does 1000 byte writes, lwIP (or any other TCP stack)
> can decide to split them or combine them as it wishes to form packets up
> to the MTU of the network.  At the application level there is no concept
> of packets: it is just one single stream of bytes.  The splitting into
> packets is done by the stack, and it can choose where it does these
> splits, and so the lengths of the packets.  

i am aware of this fact. the curious thing IMO is the correlation
between the occurrences of this "larger packets" and the stuck of the
whole stack.

> 
> Hope that helps, but it doesn't sound like the large packets are the
> cause of your problems.

i don't think so either. but i am not sure. maybe the cause of the large
packets will bring up the cause of the entire problem.

is it possible to somehow reset the whole stack after this happens. is
it enough to call the init functions again?

> 
> Kieran

best regards,

andre