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

[address@hidden]

Mullanix, Todd wrote:
> For my port to SYS/BIOS, I have NO_SYS=0 and I'm making sure the netif->input is 
> being called from a thread and not in the interrupt. So I think I'm adhering to the 
> rules of the game in my case. Note: I'll make the change to use netif->input.
Yeah, the NO_SYS=0 case looks, OK, I think: timers are handled 
automatically by tcpip_thread and tcpip_input() is used for input packets.
> My understanding of the NO_SYS=1 case was the same as Tim's. So as long as 
> nothing is done in main() or another interrupt, it should be fine since there 
> are no threads.
Well, if you see it that way, the port is OK. But it seems like Tim 
calls tcp_write from main(). I know this is a point in lwIP which is not 
as easy to understand and at least Tim (as a user of Stellarisware) got 
it wrong resulting in some days of bug-hunting :-(
Also, having had a look at the code, I'm not sure at which point it 
would be safe to call tcp_write with NO_SYS=1: there seems to be no 
obvious way to call user code from the ethernet interrupt handler (which 
is obviously the "lwip execution context" in your port for NO_SYS=1).

Note that this is only a problem for asynchronous writes. Server-writes 
(like in the httpd) are either triggered by received packets or by lwIP 
timers, so their execution context is always correct.

Simon