Re: [lwip-users] Is this a memory fragmentation problem

[Keith Rubow]

I'll try to answer your questions. No, we haven't talked about this problem.
Yes, all lwip RAW API functions are being called from the same context 
(main loop). Nothing called from interrupt.
My driver was written by myself, not vendor supplied. It seems to work 
well. It is not interrupt driven, it is polled. Data throughput rate is 
not an issue for me.
Yes, one port is a modbus slave (up to 5 connections). The other port is 
a telnet (or raw TCP/IP) user interface. I use Teraterm to talk to my 
device on the user interface port.
I can both Rx and Tx fine for hours and hours on end. I have some modbus 
test software on my PC polling modbus data every 1 or 2 seconds. I can 
do tens of thousands of polls before my system gets into this strange 
failure mode where it can't send data because tcp_write() starts 
returning ERR_MEM.
I have a tcp_sent() callback, but it does nothing except for return 
ERR_OK. I always check tcp_sndbuf() before tcp_write(), and wait for 
enough free memory before I send anything. Does the tcp_sent() callback 
need to do anything else? Do I even need it if I always check 
tcp_sndbuf() before calling tcp_write()?
Yes, all my sent data is copied into the memory allocated from the pool 
defined by MEM_SIZE.
I'm not sure if my error handling in tcp_recv() callback is correct. If 
the callback is called with a non-zero err, I call pbuf_free(p) if p was 
not NULL, and then return err. Do I need to call tcp_recved() (or 
anything else) when handling an error in the tcp_recv() callback? Should 
I close the connection if there is ever any error in the receive callback?

I am going to insert some diagnostics the tell me if I am ever getting 
an err condition in my receive callbacks. Then my system will have to 
run (maybe the rest of the day) before it fails. I'll post more 
information when I have anything to report.

Keith Rubow


On 5/18/2018 6:35 AM, Sergio R. Caprile wrote:
> First things first.
> Maybe we've already talked about this, but I don't remember everyone in
> the list...
> Are you calling all lwIP RAW API functions from the same context ?
> Either main loop or interrupts but only one of them and always the same one.
> Did you run a known-good application long enough in order to make sure
> your driver and your port are OK ? Yes, vendor provided too, and those
> the most.
>
> Now, I assume you are a modbus slave (TCP server) since you said you
> were listening.
> Since you say you can't tx, I'll assume you are rx'ing (you see the poll
> but you can't answer)
> Do you have proper tcp_sent() callbacks ? I don't know if this could
> lead to a situation like yours, though.
> Are you allocing memory ? tx memory is taken as PBUF_RAM
>
> Yes, what you describe looks weird.
>
> _______________________________________________
> lwip-users mailing list
> address@hidden
> https://lists.nongnu.org/mailman/listinfo/lwip-users