Kernel 5.10.29: Problems with Huawei ME906s

[Aleksander Morgado]

Hey Torsten,

> > 2021-05-12 10:15:26.137 kernel: cdc_mbim 1-2:3.0: setting rx_max = 16384
> > 2021-05-12 10:15:26.149 kernel: cdc_mbim 1-2:3.0: NDP will be placed at end of frame for this device.
> > 2021-05-12 10:15:26.150 kernel: cdc_mbim 1-2:3.0: cdc-wdm0: USB WDM device
> > 2021-05-12 10:15:26.150 kernel: cdc_mbim 1-2:3.0 wwan0: register 'cdc_mbim' at usb-0000:00:14.0-2, CDC MBIM, 9e:90:a9:6f:2d:31
> > 2021-05-12 10:15:26.238 kernel: cdc_mbim 1-2:3.0 wwan0: unregister 'cdc_mbim' usb-0000:00:14.0-2, CDC MBIM
> > 2021-05-12 10:15:29.405 kernel: cdc_mbim 1-2:3.0: setting rx_max = 16384
> > 2021-05-12 10:15:29.411 kernel: cdc_mbim 1-2:3.0: NDP will be placed at end of frame for this device.
> > 2021-05-12 10:15:29.413 kernel: cdc_mbim 1-2:3.0: cdc-wdm0: USB WDM device
> > 2021-05-12 10:15:29.414 kernel: cdc_mbim 1-2:3.0 wwan0: register 'cdc_mbim' at usb-0000:00:14.0-2, CDC MBIM, 9e:90:a9:6f:2d:31
> > 2021-05-12 10:15:29.464 kernel: cdc_mbim 1-2:3.0 wwan0: unregister 'cdc_mbim' usb-0000:00:14.0-2, CDC MBIM
> > 2021-05-12 10:15:32.627 kernel: cdc_mbim 1-2:3.0: setting rx_max = 16384
> > 2021-05-12 10:15:32.640 kernel: cdc_mbim 1-2:3.0: NDP will be placed at end of frame for this device.
> > 2021-05-12 10:15:32.641 kernel: cdc_mbim 1-2:3.0: cdc-wdm0: USB WDM device
> >
> ...
> > Are there any known regressions for this device in newer kernels?
>
> I was able to identify the cause of the problem.
>
> Based on the discussion started in <56D715C5.2080605 at secunet.com>
> (Huawei me906s-158, March 2016) we had an udev rules in place to perform
> the following actions:
>
> - resetting the modem by writing 0 to bConfigurationValue
> - waiting 3 seconds
> - reconfiguring the modem by writing 3 to bConfigurationValue
>
> This rule was executed when udev action != "remove" (so also on change
> events).

Maybe your problem was really the "bind" events? This is not really a
very recent addition, it happened some years ago, but it also gave us
problems in MM, see:
https://lists.freedesktop.org/archives/modemmanager-devel/2018-May/006417.html

For every port "add" you'll also get a port "bind" event when the port
is bound to a driver. Maybe you were running your rules on both "add"
and "bind" and looking at what the rule was doing it would probably
trigger an unbind/rebind and start again? something like that?

>
> When I run a "udevadm monitor kernel" while the driver is reprobing I
> see a number of add, change, bind, unbind events. So it seems that the
> timing for this events must have changed between the kernel versions to
> cause problems in 5.10.29 (or some systemd-udevd behaviourial change
> between versions 243 and 246).
>
> I have modified the rule for this special device to trigger only when
> the udev action is set to "add" and the problem vanished.
>

Thanks a lot for replying with your discoveries! It's always nice to
have open threads resolved.

--
Aleksander
https://aleksander.es