EM7565 failing if host reboots whithout power cycling the modem

[Aleksander Morgado]

Hey!

On Sun, Mar 19, 2023 at 11:40 AM Bjørn Mork <bjorn at mork.no> wrote:
>
> This is not an issue caused by ModemManager or any of the host drivers.
> Nut I don't know where else to ask. And this documents the issue for my
> next Google search :-)
>

Hahaha, been there.

> I'm using a Sierra Wireless EM7565 connected to the USB3 port of a
> Linksys WRT1900ACv1 "Mamba" (Marvell Armada 370/XP SoC), which has an
> Etron EJ168 PCIe connected xhci controller.  The system runs a recent
> OpenWrt build (v5.15 kernel) with ModemManager.
>
> This works fine after the initial power-on or after physically
> reconnecting the modem each time the host is rebooted.  But it fails
> consistently if I reboot the host without disconnecting the modem.
>
> The xhci driver will then log these messages while MM is probing the
> modem:
>
> [   78.014486] xhci_hcd 0000:01:00.0: WARN Successful completion on short TX
> [   78.021419] xhci_hcd 0000:01:00.0: WARN Successful completion on short TX
> [   78.028341] xhci_hcd 0000:01:00.0: WARN Successful completion on short TX
> [   78.035270] xhci_hcd 0000:01:00.0: WARN Successful completion on short TX
> [   78.043035] xhci_hcd 0000:01:00.0: WARN Successful completion on short TX
> [   78.049973] xhci_hcd 0000:01:00.0: WARN Successful completion on short TX
> [   78.056886] xhci_hcd 0000:01:00.0: WARN Successful completion on short TX
> [   78.063808] xhci_hcd 0000:01:00.0: WARN Successful completion on short TX
> [   78.070723] xhci_hcd 0000:01:00.0: WARN Successful completion on short TX
> [   78.077633] xhci_hcd 0000:01:00.0: WARN Successful completion on short TX
> [   78.084552] xhci_hcd 0000:01:00.0: WARN Successful completion on short TX
> [   78.092256] xhci_hcd 0000:01:00.0: WARN Successful completion on short TX
> [   78.099154] xhci_hcd 0000:01:00.0: WARN Successful completion on short TX
> [   78.106082] xhci_hcd 0000:01:00.0: WARN Successful completion on short TX
> [   81.007691] xhci_hcd 0000:01:00.0: WARN Successful completion on short TX
> [   81.014629] xhci_hcd 0000:01:00.0: WARN Successful completion on short TX
> [   81.021554] xhci_hcd 0000:01:00.0: WARN Successful completion on short TX
> [   81.028474] xhci_hcd 0000:01:00.0: WARN Successful completion on short TX
> [   81.035391] xhci_hcd 0000:01:00.0: WARN Successful completion on short TX
> [   81.042306] xhci_hcd 0000:01:00.0: WARN Successful completion on short TX
> [   81.049212] xhci_hcd 0000:01:00.0: WARN Successful completion on short TX
> [   81.056669] xhci_hcd 0000:01:00.0: WARN Successful completion on short TX
> [   81.063590] xhci_hcd 0000:01:00.0: WARN Successful completion on short TX
> [   81.070504] xhci_hcd 0000:01:00.0: WARN Successful completion on short TX
> [   81.077411] xhci_hcd 0000:01:00.0: WARN Successful completion on short TX
>
>
> and probing eventually fails.  Likewise with any attempt to use mbimcli
> etc.  The /dev/cdc-wdm0 device appears dead.  So I assume those warnings
> are real errors, masked by the XHCI_TRUST_TX_LENGTH quirk applied to
> this xhci device a decade ago.

Are we sure this is not an issue in the xhci driver itself? Maybe the
specific combination of modem and xhci controller triggers some corner
case that is not fully implemented correctly in xhci?
Last time I found something similar it took me 2 weeks of reading the
USB standard to come up with the fix
(45ba2154d12fc43b70312198ec47085f10be801a).

Did you report this at the LKML already? Maybe they have some hints on
how to best handle this.

-- 
Aleksander