Re: [lwip-users] [Issue] TCP fragmentation over PPP - fragments are lost in LwIP if data is queued for sending

[Sylvain Rochet]

Hi Marco,

On Fri, Nov 25, 2016 at 01:58:05PM +0100, Marco Jakobs wrote:
> > > 
> > > Not sure where and why ... that's not modified and PPP runs in 
> > > it's own thread ...
> > 
> > How would PPP do that in vanilla lwIP 1.4.1? Only the input escaping 
> > can be done in a separate thread, the rest should run in 
> > tcpip_thread. If you changed that --> >threading issues (+ not 
> > vanilla 1.4.1 any more...)
> 
> No changes, it runs in the tcpip_thread!

I'm totally clueless about TCP details, especially in lwIP 1.4.1 I last 
used 4 years ago; just to be sure PPP is not involved into the issue 
here before you enter diging into my "here be dragons" zone, here is my 
check list:

You are using an RTOS with PPP_INPROC_MULTITHREADED (with or without 
PPP_INPROC_OWNTHREAD) for PPP input.

(  There are known race conditions in lwIP 1.4.1 about them but they
   can only occur when disconnecting and you'll have to be very unlucky
   to trigger them. PPP_INPROC_MULTITHREADED is IRQ safe but not thread
   safe as its name state so. )

You are not using the lwIP raw API outside of the lwIP thread.

SYS_LIGHTWEIGHT_PROT is true.

You are not running out of timers, check MEMP_NUM_SYS_TIMEOUT value.

Your timers are working well even if the stack is stuck for a few 
seconds, most TCP issues come from non working timers.

Sylvain

signature.asc
Description: Digital signature