[lwip-users] LWIP - netconn with cooperative scheduling

[Mark Phillips]

Hi, all!

I have ported the LWIP stack, and got it basically working.  However, I 
am running into some problems with multiple sockets, and I was wondering 
if my structure is wrong.

Unlike many OS'es, I am running under a cooperatively-scheduled OS.  A 
running task never gets interrupted except by a real interrupt, or when 
it voluntarily suspends.

I have 2 tasks running the basic stack; one calls ethernetif_input() 
repeatedly (and suspends after each call).  The other one calls 
tcpip_thread().

The only places where lwip code can suspend is either:
a) when reading a mailbox which is empty (suspends once and then 
returns) - this is, in general, how the tcpip_thread suspends.

b) When waiting on a semaphore (suspends repeatedly until sem available 
or timeout).

I have several application tasks which are using the sockets.  Some of 
these are listeners, some are initiators (they make the connection to a 
remote host).  They are using the netconn_new, netconn_connect, 
netconn_recv, netconn_write, etc. functions. These functions are always 
called only at the application level (i.e. never from an interrupt).

I assumed that this would be a safe environment; i.e. that none of the 
netconn functions would suspend (waiting for mbox or sem) when it was 
unsafe for a different task to call one of these functions, but I 
suspect I may be wrong here?

As I said above, this works perfectly when, for example, I just have one 
or two listening sockets, but when I try to mix listening and initiating 
sockets, the whole thing comes crashing down.

Any advice will be appreciated!