[review] New 'fibocom' plugin

[Aleksander Morgado]

>> >> > On Chrome OS, we simply use ID_MM_PORT_IGNORE to ignore all AT
>> >> > ports
>> >> > on the L850-GL module (e.g.
>> >> > https://chromium.googlesource.com/chromiumos/overlays/chromiumos-ov
>> >> > erlay/+/master/net-misc/modemmanager-next/files/77-mm-fibocom-port-
>> >> > types.rules).
>> >> >
>> >> > The modem is functional with just MBIM and supported via the
>> >> > generic
>> >> > plugin, which is the approach we took on Chrome OS. IMHO, there is
>> >> > little added benefit with a new modem plugin. More importantly, AT
>> >> > probing takes quite some time, and I observed it often took 7
>> >> > seconds.
>> >> > We've been trying hard for some time to move everything to purely
>> >> > MBIM
>> >> > or QMI :)
>> >> >
>> >>
>> >> I'm going to extend this plugin in the following weeks with more
>> >> features that will require the use of the AT port, e.g. GPS location
>> >> and such. I agree that probing is a bit slow right now, but that is
>> >> just because the DIAG port goes first through a series of AT probing
>> >> steps which are not needed. For devices with fixed layouts like this
>> >> one, adding udev tags to specify port types isn't bad IMO, and we
>> >> should probably do that for the DIAG port (i.e. flag it as "not AT"
>> >> for example).
>> >
>> > Agreed; ideally we can begin to restrict which ports we bother probing
>> > for DIAG, since with QMI and MBIM DIAG is much less useful these days.
>> >
>
> IIRC, I initially blacklisted two out of the three AT ports, but
> initial AT probing still took seconds to complete (and seconds matter
> a lot on Chrome OS as we're pretty sensitive to the overall time from
> boot to connect), which drove me into blacklisting all AT ports. For
> functionality that isn't implemented by standard MBIM command set,
> we've pushing for MBIM extensions -- actually, you'll expect a few
> related patches from me soon :-)
>
> Regarding your patches, I think they won't interfere with our AT
> blacklist on this modem. In the worst case, we can simply skip
> installing the fibocom plugin, which should allow us to gracefully
> fall back to the generic plugin and use only MBIM.

Yes, the plugin won't interfere in any way as long as you keep it
MBIM-only by blacklisting all TTYs.

I may understand what you say about the slow probing, though. Looks
like that may be similar to what I've seen with the also Intel-based
TOBY-L4, where the ttys are exposed early in the kernel, but then they
require some time to be ready for AT. That would cause slow probing of
TTYs indeed. I guess users may need to decide between having more
features given by an AT fallback or gaining some extra seconds during
boot. For the generic MM case we should worry about the former I'd
say.

-- 
Aleksander
https://aleksander.es