Re: Huawei E220 can't connect on Ubuntu 11.10

[Dan Williams <dcbw redhat com>]

On Mon, 2012-01-09 at 11:27 +0100, Aleksander Morgado wrote:
> Hey Dan,
> 
> >>
> >> On Ubuntu 11.10 there is problem with connection via modem Huawei E220.
> >>
> >> On Ubuntu 11.10 the NetworkManager is in version 0.9.1.90-0ubuntu5.1
> >> On Ubunto 11.04 there is version 0.8.4~git.20110319t175609.d14809b-0ubuntu3
> >>
> >> The issue for Ubuntu was submitted:
> >> https://bugs.launchpad.net/ubuntu/+source/network-manager/+bug/868034
> >>
> >> I suspect that this problem could be related to NetworkManager.
> >> One of the most interesting comment is available at:
> >> https://bugs.launchpad.net/ubuntu/+source/network-manager/+bug/868034/comments/86
> >>
> >> Could you please take a look at this bug?
> > 
> > Looks like https://bugzilla.gnome.org/show_bug.cgi?id=662113 to me.
> > 
> > What would be helpful would be to test this with the latest
> > NetworkManager, from git head. You'll need libnl3 3.2.3 for that,
> > which is not yet in Ubuntu (or to patch the configure.ac file to avoid
> > blocking the build for that). There's preliminary packages for libnl3
> > 3.2.3 in the daily builds PPA for NetworkManager for Ubuntu:
> > http://launchpad.net/~network-manager/+archive/trunk. You may want to
> > test this, which also includes a very recent snapshot from git (from
> > the 10th of December). The problem is, these packages are largely
> > untested, so no garantees that anything is fixed, or that it doesn't
> > break your system. This is further complicated by the fact that
> > reverting from libnl3 3.2.3 might not be all that easy, but it's
> > definitely possible.
> > 
> > I'm updating the Ubuntu bug report now to add a procedure for testing.
> > 
> 
> This issue seems to have been tracked up to commit dc89c0. With that
> patch on, the modems get rebooted when they start to get commands in the
> secondary port. From what I saw in the logs, we do enable unsolicited
> messages in the secondary port during modem enable with AT^CURC=1, and
> then we also send commands to it (AT+CREG/CGREG=2, to enable unsolicited
> registration messages), which is the time when the hang & reboot happens.

I did try and reproduce this with my E220 (a Telia branded one with
latest E220 firmware) and could not.  But the change in question I think
was for a modem that turned out not to require it, so we could revert
that.  We do want unsolicited notifications on the secondary port
though, because otherwise we don't  get anything useful when the primary
port is being used for PPP.  One possibility here is that the Windows
drivers tag usbif 0 as the "modem" port (ie, to be used for PPP) and the
secondary port as the "pcui" port.  I think it's actually the "pcui"
port that the Windows stuff uses normally, and only turns on usbif 0
when it needs PPP (since usbif 0 has the interrupt endpoint but we don't
need that for plain AT commands).

I'd done a patch a long time ago to better separate the primary port
from the port used for PPP, since right now we assume the primary port
should be used for both command&status and PPP, but that may not be the
case all the time.  Perhaps it's time to dust that patch off so people
could at least test this theory out on E220s that see the problem.

Also, what firmware version are the problematic modems running?  Mine is
on 11.117.07.00.07.  My E226, which is basically an E220 with Americas
3G bands (850/1900), hasn't had issues so far.

Dan

> I would suggest to revert that commit until a better solution is found
> and ideally make a new stable release.
> 
> Also, did we try to completely avoid sending any command to the
> secondary port once unsolicited messages were enabled on it? I can try
> to hack such behaviour and then ask bug reporters to test it.
>