Re: [PATCH] LWIP over SLIP

["David Ho" <davidkwho at gmail dot com>]

Hi,

Sorry for the delay, I have been trying to get httpd running over LWIP
on eCos, now that I have resolved all the issues, I can attend this
your questions.

> Please could you explain what goes wrong with loopif and how your
> change fixes this.
>
> The current code is:
>
> #if LWIP_HAVE_LOOPIF
>         IP4_ADDR(&gw, 127,0,0,1);
>         IP4_ADDR(&ipaddr, 127,0,0,1);
>         IP4_ADDR(&netmask, 255,0,0,0);
>
>         netif_add(&ecos_loopif, &ipaddr, &netmask, &gw, NULL, loopif_init,
>             tcpip_input);
> #endif
>
> Reading netif_add, it will call loopif_init, so it is not obvious to
> me why you need to call it before.

I actually hadn't test this.  I just noticed that when I enabled debug
the message never printed out the loopif interface name properly, I
thought this one liner would fix it.

> > #if LWIP_SLIP
> > -       lwip_set_addr(&mynetif);
> >        slipif_init(&mynetif);
> > -       netif_set_default(&mynetif);
> > +       lwip_set_addr(&mynetif);
> > #elif PPP_SUPPORT
> >        pppInit();
> > #if PAP_SUPPORT || CHAP_SUPPORT
>
> Here you no longer call netif_set_default(). Did you test this change
> with routing to packets via a default gateway?

If you simply look further down the file, where lwip_set_addr is
defined, the function calls netif_set_default.  Essentially
netif_set_default was called twice in the original code.

Note that I have to move the call to lwip_set_addr after slipif_init
resets the netif->flags.  Inside lwip_set_addr, it attempts to bring
the interface up by setting a bit in netif->flags, which slipif_init
clears.

> > diff --git a/packages/net/lwip_tcpip/current/src/ecos/sio.c
> > b/packages/net/lwip_tcpip/current/src/ecos/sio.c
> > index 60004c4..87a4eb7 100644
> > --- a/packages/net/lwip_tcpip/current/src/ecos/sio.c
> > +++ b/packages/net/lwip_tcpip/current/src/ecos/sio.c
> > @@ -82,7 +82,7 @@ void *
> > sio_open(int devnum)
> > {
> >        int res;
> > -       cyg_uint32 nb = 0, len = 4;
> > +       cyg_uint32 nb = 1, len = 4;
> >
> > #if LWIP_SLIP
> >        #define SIODEV SLIP_DEV
> > @@ -95,9 +95,11 @@ #endif
> >
> >        res = cyg_io_set_config(ser, CYG_IO_SET_CONFIG_READ_BLOCKING,
> > &nb, &len);
> >        res = cyg_io_set_config(ser, CYG_IO_SET_CONFIG_WRITE_BLOCKING,
> > &nb, &len);
> > +#if 0
> >        len = 4;
> >        nb = 0xFFFFFFFF;
> >        res = cyg_io_set_config(ser,
> > CYG_IO_SET_CONFIG_SERIAL_FLOW_CONTROL_METHOD, &nb, &len);
> > +#endif
> >        if (res != ENOERR)
> >                        diag_printf("set_config flow_control returned
> > an error\n");
> >        return &ser;
>
> Again, please could you explain this change. It is not obvious what is
> wrong.

If I don't set up the serial port driver to blocking read, I would be
continously receiving garbage that is returned from a non-blocking
read, which is not what I want.  So I enabled blocking on both reads
and writes and that made things behave a lot better.

There is no FLOW_CONTROL_METHOD option implemented in the at91 serial
driver, trying to do that on the driver will stop the LWIP
initialization process.

Hope this helps,
David