Re: [lwip-users] the sequentiality of the lwip stack

[address@hidden]

Fabian Koch wrote:
> I'm still trying to figure out exactly what parts of the LwIP stack 
> stop us from doing that and I would love to get some pointers (from 
> you, Simon?) where there would be work to do to make that work.

There are multiple issues:
- When an application thread makes a socket call, it issues a function 
call in the tcpip-thread and waits for it to complete. This uses a 
semaphore, which only exists once where multiple semaphores would be 
needed when multiple threads could call for one socket at once.
- A different issue is that the socket/netconn contains a state which 
could get confused when reading, writing and closing is done concurrently.
- Next, when closing a socket, we would have to make sure all threads 
blocking on this socket (reading, writing) would get unblocked with an 
error code and that closing is not executed simultaneously more than 
once (when it needs some time).

I'm not saying it can't be done, but from the lines above, it's clear 
that it would increase the footprint, which would have to be prevented 
when not needed.


Simon