QMI disconnection errors detection in Netgear AC341U

[Dan Williams]

On Wed, 2017-09-06 at 09:44 +0200, Aleksander Morgado wrote:
> On Tue, Sep 5, 2017 at 11:10 PM, Dan Williams <dcbw at redhat.com>
> wrote:
> > On Tue, 2017-09-05 at 22:26 +0200, Aleksander Morgado wrote:
> > > Hey,
> > > 
> > > The Netgear AC341U is a QMI device that comes with autoconnect
> > > enabled (and disabling it doesn't look possible, firmware may
> > > even
> > > crash and power cycle itself). I suspect that due to autoconnect
> > > being enabled, the packet service indications aren't sent as we
> > > expect them, in particular to detect network initiated
> > > disconnections.
> > 
> > Instead, could we hook up eQMI_WDS_GET_MODEM_STATUS and the
> > eQMI_WDS_MODEM_IND indicator or something and watch
> > sWDSModemStatusReportIndication_Status to see if we're already
> > connected or when we get disconnected?  Maybe it sends that
> > indication
> > instead of packet service status for the NDIS autoconnect or
> > something.
> >  Worth a look perhaps.
> > 
> > It's just that we could get call failures from so many places and
> > checking them all seems icky...
> > 
> 
> Well, not that many really, the only place where we actively and
> periodically check something from the data connection is when loading
> stats. And the only other step after StartNetwork before saying we're
> connected is GetCurrentSettings.

OK true.

> But I agree that I should try to enable as many WDS event reports as
> possible and see if any of them tells us anything, will try that in
> the next days.

Thanks, I'm not sure I have time to check that out this week, but I'm
sure I have some modems that I could test.

> The "correct" solution here anyway would be to support the
> autoconnected setup somehow, e.g. creating a bearer automatically on
> boot when we detect autoconnect is enabled (even if not yet
> connected!). This is something we should look at more carefully, as
> we
> could then include in MM a lot more modems that we currently don't
> include (e.g. huawei hilink managed via the HTTP API).

Agreed, that was the subtext of my comment actually.  Being already
connected happens for at least 3 cases: (1) LTE default bearer, (2) QMI
NDIS autoconnect, and (3) hilink-type stuff.  Seems like we should
figure out a away to handle them all generically if we can.

Dan