Re: [lwip-users] tcp_write() errors on snd_queuelen

[Tim Lambrix]

I posted this email below a while back to which I didn’t get a response.  I let it go as the code appeared to be working with the poll value of 1.  Now I need to figure out how to detect a broken connection somewhere between the lwIP device and my PC.  Is there a way to use tcp_poll for this or the keep-alive or some other

detection?  I can’t figure out how tcp_poll is suppose to work.  My application opens a connection and must remains open until the software closes it (weeks, months or years later).  I can detect the lwIP device-to-switch link broken and the same for the PC side; just not when the link is broken between two switches in the middle.  Any suggestions?

 

I have been looking at the timers some more and trying to figure out how to get

my buffer moved during the interrupt.  It looks like there are two timers: 1)

lwIPHostTimerHandler and 2) tcp_poll.  I think I have to use the host timer

handler to read information out of my buffers creatively(hopefully while not in

the middle of changing it).

 

I am trying to learn how the tcp_poll timer function works.  I have a function

and set:   

 

        tcp_poll(pcb, TCP_Poll, 1);

 

I found that increasing this value above a 1 can either A) never get called at

a half second interval and B) from wireshark, only sends data out at whatever

the value is.  For example, I put a 4 in there and WireShark displays a 2

second gap between send and receives.  However, with a 1 in there it sends and

receives data at a 200ms interval and if needed, due to more than one TCP

outgoing packet being full (2*TCP_MSS), it will send all packets at that time. (I can send the two Wireshark outputs showing the timing difference if it helps). 

I don't understand this timer value behavior as the TCP_Poll function I have

written, doesn't really do much (residual from StellarisWare). 

 

  if ((hs == NULL) && (pcb->state == ESTABLISHED))

  {

    tcp_abort(pcb);

    return ERR_ABRT;

  }

 

Is there an explanation of how the poll interval timer works this way?  I found

that the TCP_SND_QUEUELEN limit is reached very quickly if the value is NOT a 1

which should map to 500 ms based on the slow TCP timer. This makes some sense

as I can send up to sixty tcp_writes in 200 ms alone.  In two seconds, this is

over 200.  In other words, my code only functions without crashing (with the

disable Ethernet interrupt fix around tcp_write of course) with the poll

interval set to a 1.