E3276: Failed to find primary AT port

[Bjørn Mork]

"Georg Lippitsch" <georg.lippitsch at gmx.at> writes:
> Am 11.06.2014, 20:09 Uhr, schrieb Bjørn Mork <bjorn at mork.no>:
>
>> No, I don't have any idea. Is there any sign of the modem being
>> plugged in the kernel log when this happens? Is the problem there if
>> you switch to another port?
>
> At the first *successful* connect, I get several "unknown notification
> 3 received: index 1 len 4".

Yes, that is known and harmless.  I believe the message is silenced to a
debug level in newer kernels.

We don't actually know what Huawei means by this notification - "3" is
not defined by the NCM or CDC specs. But then they aren't claiming these
devices are true NCM devices either, so they are of course free to add
their own extenstions.  I guess supporting this notification could be a
useful addition to the driver if anyone wants to debug it and figure out
what it means.  But we'll probably just end up in the same dilemma we
have with the CONNECTION_SPEED_CHANGE notification: There aren't really
any suitable userspace interfaces the driver can report it through.

> At the second time, no messages about huawei-cdc-ncm appear at all.
>
> With unpatched 3.13.0 kernel, I even got a null pointer dereference
> when connecting the modem for the second time, and a freeze when
> shutting down the machine afterwards.
> I also have to admit that I used older 3.13.0 kernel and only patched
> the modules cdc-wdm, usb-wwan, cdc-mbim, huawei-cdc-ncm and cdc-ncm
> Probably I have overlooked something.

I am hoping so :-)

> It's probably better to try recent kernel git, I will report when I
> have done this.

Thanks.  I appreciate that.  And if you can reproduce the crash with an
unmodified mainline or stable kernel, then it would be nice to have the
actual oops message.  A photo of the screen will do if you are unable to
use a serial console or netconsole.


Bjørn