Re: [lwip-users] TCP tmr endless loop

[Erik Ekman]

On 10/24/07, Erik Ekman <address@hidden> wrote:
> On 10/24/07, Jonathan Larmour <address@hidden> wrote:
> > Kieran Mansley wrote:
> > > On Wed, 2007-10-24 at 13:11 +0200, Erik Ekman wrote:
> > >
> > >
> > >>The problem is that sometimes it gets stuck in the tcp_slowtmr and
> > >>tcp_fasttmr, just like described earlier in the list. The
> > >>tcp_active_pcb has a pointer next which points to itself.
> > >
> > >
> > > That sort of list corruption very much points to two threads being
> > > active in the core of lwIP at the same time.
> >
> > Specifically your lwip_tick() and signal_send() functions are probably
> > being called overlapping each other, and each is making its own calls into
> > lwIP.
> >
> > Jifl
>
> I see. But my OS does not allow multiple tasks running at the same
> time. The "scheduler" is simply a loop through all registered tasks,
> and it calls the function for each task that has a message waiting.
> The next task will then be called when the first has returned. The
> receive packet interrupt allocates a message in constant time and
> sends the message to lwip_tick. The timer also just puts messages in
> the task queues, so it will not interrupt the execution.
>
> The system only uses cooperative multitasking and some interrupts, and
> if I need to add an RTOS for some reason, the existing OS can be run
> as a thread in it.
>
> I will investigate some more, thanks for the hints.
>
> /Erik
>

FYI:
The problem disappeared when I removed tcp_close() from the error
handler. I also reset callbacks to NULL before exiting.

/Erik