RE: [lwip-users] Dying listener sockets - Email found in subject

[Kieran Mansley]

On Tue, 2010-02-09 at 15:43 +0000, Tamas Somogyi wrote:
> Hi,
> 
> First of all, thanks very much to Stéphane for the quick help.
> It seems indeed that upgrading lwIP to CVS head and specifying
> MEMP_NUM_NETCONN= MEMP_NUM_TCP_PCB+ MEMP_NUM_TCP_PCB_LISTEN solves the
> problem on both platforms for my test application.

Thanks for reporting your success.

> However I still have to fine-tune the constants in lwipopts.h to get
> it work in my real application on the target platform where there are
> strict limitations for resources (threads, memory, speed, etc.). E.g.
> it is not fully clear to me that MEMP_NUM_NETCONN should equal to
> MEMP_NUM_TCP_PCB+ MEMP_NUM_TCP_PCB_LISTEN+ MEMP_NUM_UDP_PCB+
> MEMP_NUM_RAW_PCB? E.g. lwip_sanity_check() complains if it is more and
> suggest to specify less (and not equal?) - so how does it affect the
> behaviour/performance?

Each netconn will require 1 PCB, so you can't do NETCONN > SUM(PCBS).
That is why there is the restriction in sanity check, but it should
probably include "equal" as a sane value.

> I try to narrow down the changes to find the root problem, but I
> cannot promise (quick) result as there are many differences
> (1.3.0<->CVS head, platforms, constants).

As I understand it the suggestion is that this bug has been fixed since
1.3.2 was released, so it should only be differences between 1.3.2<->CVS
head that are relevant.

Kieran