[PATCH v9 35/37] [TESTING] Kconfig: enable NET_LWIP by default except for SANDBOX

[Simon Glass]

Hi Ilias,

On Fri, 30 Aug 2024 at 03:27, Ilias Apalodimas
<ilias.apalodimas at linaro.org> wrote:
>
> Hi Simon,
>
> On Fri, 30 Aug 2024 at 03:59, Simon Glass <sjg at chromium.org> wrote:
> >
> > Hi Jerome,
> >
> > On Thu, 29 Aug 2024 at 10:21, Jerome Forissier
> > <jerome.forissier at linaro.org> wrote:
> > >
> > >
> > >
> > > On 8/29/24 16:05, Simon Glass wrote:
> > > > Hi Jerome,
> > > >
> > > > On Fri, 23 Aug 2024 at 07:50, Jerome Forissier
> > > > <jerome.forissier at linaro.org> wrote:
> > > >>
> > > >> Enable NET_LWIP by default for testing purposes. SANDBOX doesn't support
> > > >> NET_LWIP so default to NET in this case.
> > > >
> > > > Sandbox needs to support NET_LWIP.
> > >
> > > I agree in principle, but AFAICT it is not a trivial task. Does it have to
> > > be done in this series or can it be dealt with later?
> >
> > Does it build OK with sandbox? We use it for almost all of our feature
> > development and testing, so I cannot imagine bringing this in if it
> > doesn't work. How did you develop this feature?
>
> sandbox is deeply rooted into the old network stack. It fakes ACK
> responses, handles the ping request/replies with the sandbox ethernet
> driver etc. I was looking on adding LWIP support for it, but it's not
> trivial and I don't want us to add another hacky 'glue layer'. Since
> LWIP is a big feature on its own -- and is working on real hardware
> and QEMU, we discussed adding it, so people can test and contribute
> fixes while we fix sandbox.

At the very least it needs to build on sandbox.

I am not sure what tests LWIP has. The U-Boot networking tests are
very simple on sandbox. The test provides a handler to generate a
packet that it wants, then does a networking operation. It should not
be hard to plumb that through.

I am nervous about this going in with an intention to fix it up later...

>
> >
> > >
> > > I would need to look into this more closely to give details, but if I
> > > remember correctly the sandbox tests expect a precise sequence of network
> > > packets. For example when doing a TFTP tests, it expects an ARP packet to
> > > go out first, but when lwIP is used the ARP might not go out because the IP
> > > may already be in the cache. So the sandbox is definitely not a black box
> > > in that respect which makes things a bit more difficult.
> >
> > The key thing is to have a way to reset the state. So long as you have
> > that, you should be able to put networking back to the initial state
> > before running a test.

Regards,
Simon