Re: [lwip-users] Broadcast packets breaking my network traffic

[Kieran Mansley]

On Thu, 2007-07-19 at 18:53 +0300, Caglar Akyuz wrote:
> Kieran Mansley wrote:
> > On Thu, 2007-07-19 at 18:11 +0300, Caglar Akyuz wrote:
> >> Kieran Mansley wrote:
> >>> On Thu, 2007-07-19 at 15:35 +0100, Kieran Mansley wrote:
> >>>> Can you get a small ethereal capture to illustrate this?  The fact that
> >>>> the hub makes a difference is interesting.  I wonder if it is a problem
> >>>> with the boards seeing packets from one of the other boards.  Does it
> >>>> happen in response to all broadcast ARP packets, or only some?  Does it
> >>>> happen in response to other broadcast packets?  
> >>> Is there any traffic (other than broadcast packets and TCP ACKs) going
> >>> back from the PC to the devices?  If not, it could be just that the
> >>> board receiving data is causing them problems (rather than it being
> >>> broadcast packets in particular).  You could try sending a UDP stream at
> >>> the board (which should just drop the packets if there is nothing
> >>> listening to receive them) and see what that does.
> >>>
> >> No there isn't no other traffic other than packets and ACKS. Do you
> >> think pinging may cause the same effect because I'm pinging the devices
> >> while packets are flying. Some times pinging was ok, sometimes it
> >> crushed the boards.
> > 
> > If pinging kills the board I suggest there is something seriously wrong.
> > Not sure where though I would guess it might be something to do with
> > protection of the core stack.  Perhaps you have an rx packet and a tx
> > packet being procesed at the same time and this causes corruption.  How
> > do you synchronise processing of RX and TX in your port?
> 
> I rely on sys_arch_protect and sys_arch_unprotect. They're
> disabling/enabling interrupts at the cpu level. I also tried adding some
> other protection. For instance, I disabled passing broadcast packets
> from my mac driver to lwIP while I'm transmitting any packets.(disabled
> before calling tcp_ouput, enabled after tcp_sent callback)

I don't think that is sufficient: you don't just need to protect the
tcp_output to tcp_send call, but also any other calls into lwIP.  This
includes preventing having your sending thread, any of the TCP timers,
or any receive packet processing happening concurrently.  Even
processing a receive packet while you're in a call to tcp_write() for
example would be a problem as they're both modifying the same state.

> One thing I suspect is that there is something wrong with physical
> layer, or at least there are more than one problems and one of them is
> related to phy. For that reason, I disabled full duplex mode and I'm
> using half-duplex mode at the both sides, I think full-duplex mode
> doesn't help me much while using a hub.(I'm not %100 sure...)

I think the protection of the stack is more likely than a phy problem.

Kieran