Re: MC 7304 ipv4v6

[Bjørn Mork <bjorn mork no>]

Thomas Schäfer <tschaefer t-online de> writes:
> Am 25.11.2015 um 11:12 schrieb Bjørn Mork:
> > Bjørn Mork <bjorn mork no> writes:
> >
> > > But I believe I implemented it the last time this was discussed.  I'll
> > > see if I can dig up the old code for a demo.  Not going to rush this
> > > anyway.  I'd like everyone to be happy with the result.  And it's going
> > > to be immutable once it's in.
> >
> > git is a wonderful tool.  Nothing is ever lost :)
> >
> > I found my example code in an old v3.2 based branch, and turned it into
> > a demo on the bus to job.  Note that this requires a minor change to
> > usbnet.c as well, to prevent it from doing eth_type_trans() on the rx
> > skbs we pass on to it.  Therefore two, eh.. three, patches.
>
>
> You are so fast. :-)
>
>
> Some questions to your detailed solution and the invitation to test.
>
>
> I need a very recent kernel (I have seen 4.4 in patch 003)?

It will probably apply to older versions as well. There hasn't been that
many changes to the qmi_wwan driver.

Yes, I currently use v4.4-rcX.  But that is primarily because I have
found running a recent release candidate to be the best way to eliminate
bugs affecting me, not because it's necessary for this testing.  I don't
know if I'm just lucky, but there tend to be one or two bugs like that
in each release cycle.  The one making cdc_mbim crash with EM7305, or
any other Sierra MBIM device except for the MC74xx, is one current and
nasty example.  I believe it's good to avoid releasing with bugs like
that...  So I risk my file systems running release candidates.

> I need to follow your instructions (why do you use sometimes ifconfig
> instead of ip ?)

Arbitray choice made by fingers with no brain involvement.  Sort of like
DMA :)

> - with ModemManager/Networkmanager disabled - exclusively using qmi-cli
>
> so far I can follow.
>
> Your example has IPv4. Do I need additionally qmicli-commands for
> testing IPv4v6?

Probably.  I don't know how to do that.  Is there an IP family setting
for qmicli? Another alternative is modifying the 802.3 default in
ModemManager.  That would be this part of src/mm-port-qmi.c:

   if (ctx->set_data_format)
        flags |= (QMI_DEVICE_OPEN_FLAGS_NET_802_3 | QMI_DEVICE_OPEN_FLAGS_NET_NO_QOS_HEADER);

Changing it to something like

   if (ctx->set_data_format)
        flags |= (QMI_DEVICE_OPEN_FLAGS_NET_RAW_IP | QMI_DEVICE_OPEN_FLAGS_NET_NO_QOS_HEADER);

should do for a simple test. I don't think you can use NM to configure
the netdev, though.  So you'll have to do that manually after connecting.


> Because the test contains kernel-compilation, where I am totally out
> of practice, I cannot promise a date for test results.

No hurry.  The solution is pretty much guaranteed to solve your problem
if Reinhard's guess is correct.  And that is likely.  The real question
is not whether it will work around the firmware bug, but whether and
how we should implement the raw-ip support.



Bjørn