[ECOS] TCP/IP and rom startup mode

[ECOS] TCP/IP and rom startup mode

[Fano Ramparany]

I'm using TCP/IP stack in a client application running on eCos (the
server runs
on a unix workstation). When running the client in "RAM startup" mode
(mode for debugging) everything works fine.
When running it in "ROM startup" mode I've noticed the following
problems:
1- it takes about one minute for the function
"init_all_network_interfaces" to complete.
  whereas in "RAM startup" mode although it takes some time for gdb to
download
  the client code on the board,  once done,
"init_all_network_interfaces" completes
  immediately.
2- I can establish a socket connection with the bootp server (the
argument
  of the tcpconnection function) only. If I my server runs on another
unix workstation,
  my client can't connect to it

Does this ring a bell to someone? or do you have any hint on how to
solve those
problems?

I'm working with eCos 23 march 2001 snapshot, on a EDB7212 board.

Thanks in advance for your help.

Fano

Re: [ECOS] TCP/IP and rom startup mode

[Jonathan Larmour]

Fano Ramparany wrote:
> 
> I'm using TCP/IP stack in a client application running on eCos (the
> server runs
> on a unix workstation). When running the client in "RAM startup" mode
> (mode for debugging) everything works fine.
> When running it in "ROM startup" mode I've noticed the following
> problems:
> 1- it takes about one minute for the function
> "init_all_network_interfaces" to complete.
>   whereas in "RAM startup" mode although it takes some time for gdb to
> download
>   the client code on the board,  once done,
> "init_all_network_interfaces" completes
>   immediately.

It sounds like maybe BOOTP/DHCP is timing out. Except I don't know why this
would be specific to ROM startup. Perhaps something else in this area of
your configuration changed?

> 2- I can establish a socket connection with the bootp server (the
> argument
>   of the tcpconnection function) only. If I my server runs on another
> unix workstation,
>   my client can't connect to it

Perhaps it needs routing?

Jifl
-- 
Red Hat, Rustat House, Clifton Road, Cambridge, UK. Tel: +44 (1223) 271062
Maybe this world is another planet's Hell -Aldous Huxley || Opinions==mine

Re: [ECOS] TCP/IP and rom startup mode

[Fano Ramparany]

I've investigated on this problem by adding some tracing in eCos net package.
I've found the following:
When operating in ROM startup mode, the long delay (approx. 2 mn) for
completing
init_all_network_interfaces is due to the execution of the DHCP protocol
(function do_dhcp in file dhcp_prot.c),
where there are many iterations in the DHCP state machine. More precisely,
controls
goes repetedly to state "DHCPSTATE_SELECTING" where it fails receiving the
DHCPOFFER message
from the DHCP server, and then listen again without retransmitting but with a
longer time out in case of repetitive failure.
(whereas, in the case of RAM startup mode, DHCPOFFER is received at the first
attempt)
However, I've found that if I wait long enough (1mn) between pushing the
ureset and pushing the wakeup button
(this is a specific procedure for running a ROM application on the EDB7212
board), I don't have
this 2mn delay.

So my conclusion is that this seems to be a HW problem.
I'm happy with this, but I though it might be helpful for other EDB7212 board
users,
to get this info in case they face the same problem.
If somebody has an explanation for why the delay between resetting and waking
up the board
impacts on the reception of the DHCPOFFER message, I'll be happy to hear about
it.

Fano

Jonathan Larmour wrote:

> Fano Ramparany wrote:
> >
> > I'm using TCP/IP stack in a client application running on eCos (the
> > server runs
> > on a unix workstation). When running the client in "RAM startup" mode
> > (mode for debugging) everything works fine.
> > When running it in "ROM startup" mode I've noticed the following
> > problems:
> > 1- it takes about one minute for the function
> > "init_all_network_interfaces" to complete.
> >   whereas in "RAM startup" mode although it takes some time for gdb to
> > download
> >   the client code on the board,  once done,
> > "init_all_network_interfaces" completes
> >   immediately.
>
> It sounds like maybe BOOTP/DHCP is timing out. Except I don't know why this
> would be specific to ROM startup. Perhaps something else in this area of
> your configuration changed?
>
> > 2- I can establish a socket connection with the bootp server (the
> > argument
> >   of the tcpconnection function) only. If I my server runs on another
> > unix workstation,
> >   my client can't connect to it
>
> Perhaps it needs routing?
>
> Jifl
> --
> Red Hat, Rustat House, Clifton Road, Cambridge, UK. Tel: +44 (1223) 271062
> Maybe this world is another planet's Hell -Aldous Huxley || Opinions==mine

Re: [ECOS] TCP/IP and rom startup mode

[wangwei]

I met the same problem .
I used redboot instead of stubs,which startup in rom. when I use gdb-insight
to debug it via tcp/ip,
init_all_network_interfaces will get a long delay. if gdb connnect target
via serial ,
everything is ok.


----- Original Message -----
From: "Fano Ramparany" <Fano.Ramparany@rd.francetelecom.fr>
To: "Jonathan Larmour" <jlarmour@redhat.com>
Cc: "eCos Disuss" <ecos-discuss@sourceware.cygnus.com>
Sent: Tuesday, May 29, 2001 6:05 PM
Subject: Re: [ECOS] TCP/IP and rom startup mode


> I've investigated on this problem by adding some tracing in eCos net
package.
> I've found the following:
> When operating in ROM startup mode, the long delay (approx. 2 mn) for
> completing
> init_all_network_interfaces is due to the execution of the DHCP protocol
> (function do_dhcp in file dhcp_prot.c),
> where there are many iterations in the DHCP state machine. More precisely,
> controls
> goes repetedly to state "DHCPSTATE_SELECTING" where it fails receiving the
> DHCPOFFER message
> from the DHCP server, and then listen again without retransmitting but
with a
> longer time out in case of repetitive failure.
> (whereas, in the case of RAM startup mode, DHCPOFFER is received at the
first
> attempt)
> However, I've found that if I wait long enough (1mn) between pushing the
> ureset and pushing the wakeup button
> (this is a specific procedure for running a ROM application on the EDB7212
> board), I don't have
> this 2mn delay.
>
> So my conclusion is that this seems to be a HW problem.
> I'm happy with this, but I though it might be helpful for other EDB7212
board
> users,
> to get this info in case they face the same problem.
> If somebody has an explanation for why the delay between resetting and
waking
> up the board
> impacts on the reception of the DHCPOFFER message, I'll be happy to hear
about
> it.
>
> Fano
>
> Jonathan Larmour wrote:
>
> > Fano Ramparany wrote:
> > >
> > > I'm using TCP/IP stack in a client application running on eCos (the
> > > server runs
> > > on a unix workstation). When running the client in "RAM startup" mode
> > > (mode for debugging) everything works fine.
> > > When running it in "ROM startup" mode I've noticed the following
> > > problems:
> > > 1- it takes about one minute for the function
> > > "init_all_network_interfaces" to complete.
> > >   whereas in "RAM startup" mode although it takes some time for gdb to
> > > download
> > >   the client code on the board,  once done,
> > > "init_all_network_interfaces" completes
> > >   immediately.
> >
> > It sounds like maybe BOOTP/DHCP is timing out. Except I don't know why
this
> > would be specific to ROM startup. Perhaps something else in this area of
> > your configuration changed?
> >
> > > 2- I can establish a socket connection with the bootp server (the
> > > argument
> > >   of the tcpconnection function) only. If I my server runs on another
> > > unix workstation,
> > >   my client can't connect to it
> >
> > Perhaps it needs routing?
> >
> > Jifl
> > --
> > Red Hat, Rustat House, Clifton Road, Cambridge, UK. Tel: +44 (1223)
271062
> > Maybe this world is another planet's Hell -Aldous Huxley ||
Opinions==mine