Re: [lwip-users] [Issue] TCP fragmentation over PPP - fragments are lost in LwIP if data is queued for sending

[Marco Jakobs]

Hi Sylvain, Hi Simon,

OK ... I've got an update. It's *not* that the 376 bytes fragments (2 and 4 in 
my overview) gets lost! I've written a test where I immediately send the 2x1400 
bytes to the TCP connection so also the ETH is facing the queued data.

As LwIP is smart *g*, it'll fragment different here. Instead of:

1) 1024 bytes
2) 376 bytes
3) 1024 bytes
4) 376 bytes

it'll fragment it like this:

1) 1024 bytes
2) 1024 bytes
3) 752 bytes

Writing a known pattern in the payload I confirm that the first two packets of 
1024 bytes are matching! So the 3rd fragment of 752 is just not sent over the 
PPP.

Using the PPP_DEBUG I can see that pppifOutput is just called 2 times with a 
LEN of 1024 bytes each. For the last 752 bytes fragment, pppifOutput is not 
called.

We try to dig deeper here, but as we don't have a deep knowledge of the 
internal structure of LwIP any help or idea how to debug is appreciated ;-)

Marco


>Using an RTOS doesn't change anything, your uart tx function must wait if your 
>fifo/queue/whatever is full.
>But… PPPoS in lwIP 1.4.1 is not thread safe and hard to do right. On which 
>context are you calling the PPP input function ?

>Sylvain