Random input/output errors with /dev/cdc-wdm* in kernel 4.9.7

[Kasper Holtze]

Hello,

After upgrading Linux from 4.4.8 to 4.9.7, I'm seeing some strange 
issues connecting to my modems. The modem is Sierra MC7304, in QMI mode.

As it happened after upgrading the kernel, I'm not sure if it is 
directly related to libqmi.. Could be the qmi_wwan driver?

I see the issues with both qmicli and my custom software.

libqmi is version 1.16.2.

Usually it works fine when the system is first booted, or the modems are 
reset. After a while, I'm getting i/o errors reading from /dev/cdc-wdm*:

Sometimes it just hangs, most of the times I get this message:

Qmi-WARNING **: Error reading from istream: Error reading from file 
descriptor: Input/output error

I have enabled debugging in debugfs, and get this in the syslog:

[ 5664.134111] wdm_open: qmi_wwan 1-1.3:1.8: draining queued data
[ 5664.134778] wdm_in_callback: qmi_wwan 1-1.3:1.8: nonzero urb status 
received: -EPIPE
[ 5664.136365] wdm_int_callback: qmi_wwan 1-1.3:1.8: 
NOTIFY_RESPONSE_AVAILABLE received: index 8 len 0
[ 5664.136390] wdm_int_callback: qmi_wwan 1-1.3:1.8: submit response URB 0
[ 5664.136903] wdm_in_callback: qmi_wwan 1-1.3:1.8: nonzero urb status 
received: -EPIPE
[ 5664.137551] wdm_release: qmi_wwan 1-1.3:1.8: wdm_release: cleanup

and this:

# grep wdm /sys/kernel/debug/dynamic_debug/control
drivers/usb/class/cdc-wdm.c:1093 [cdc_wdm]wdm_suspend =pf 
"wdm%d_suspend\012"
drivers/usb/class/cdc-wdm.c:697 [cdc_wdm]wdm_open =pf "draining queued 
data\012"
drivers/usb/class/cdc-wdm.c:460 [cdc_wdm]wdm_write =pf "Tx URB has been 
submitted index=%d\012"
drivers/usb/class/cdc-wdm.c:1083 [cdc_wdm]wdm_disconnect =pf "%d open 
files - postponing cleanup\012"
drivers/usb/class/cdc-wdm.c:726 [cdc_wdm]wdm_release =pf "wdm_release: 
cleanup\012"
drivers/usb/class/cdc-wdm.c:734 [cdc_wdm]wdm_release =pf "cdc_wdm %s: 
device gone - cleaning up\012"
drivers/usb/class/cdc-wdm.c:288 [cdc_wdm]wdm_int_callback =pf 
"NOTIFY_RESPONSE_AVAILABLE received: index %d len %d\012"
drivers/usb/class/cdc-wdm.c:295 [cdc_wdm]wdm_int_callback =pf 
"NOTIFY_NETWORK_CONNECTION %s network\012"
drivers/usb/class/cdc-wdm.c:299 [cdc_wdm]wdm_int_callback =pf 
"SPEED_CHANGE received (len %u)\012"
drivers/usb/class/cdc-wdm.c:317 [cdc_wdm]wdm_int_callback =pf "submit 
response URB %d\012"
drivers/usb/class/cdc-wdm.c:1143 [cdc_wdm]wdm_resume =pf "wdm%d_resume\012"
drivers/usb/class/cdc-wdm.c:576 [cdc_wdm]wdm_read =pf "zero length - 
clearing WDM_READ\012"
drivers/usb/class/cdc-wdm.c:174 [cdc_wdm]wdm_in_callback =pf "nonzero 
urb status received: -ENOENT\012"
drivers/usb/class/cdc-wdm.c:178 [cdc_wdm]wdm_in_callback =pf "nonzero 
urb status received: -ECONNRESET\012"
drivers/usb/class/cdc-wdm.c:182 [cdc_wdm]wdm_in_callback =pf "nonzero 
urb status received: -ESHUTDOWN\012"
drivers/usb/class/cdc-wdm.c:186 [cdc_wdm]wdm_in_callback =pf "nonzero 
urb status received: -EPIPE\012"
drivers/usb/class/cdc-wdm.c:227 [cdc_wdm]wdm_in_callback =pf "got %d 
bytes without notification\012"

If I try accessing /dev/cdc-wdm* directly, I get the same input/output 
error (I know it doesn't make sense to cat the file, but just to show it..):

root at imx6qsabresd:~# cat /dev/cdc-wdm0
cat: read error: Input/output error

I have two modems and see the same issue. I can't really find the 
pattern, it seems that calling multiple commands quickly can trigger it. 
Sometimes I can wait a few minutes, then I can send a command again, 
other times I have to reset the modem.

I tried opening the device with QMI_DEVICE_OPEN_FLAGS_SYNC, and get this 
(not sure if it's related):

(process:17861): Qmi-CRITICAL **: __qmi_message_ctl_sync_response_parse: 
assertion 'qmi_message_get_message_id (message) == QMI_MESSAGE_CTL_SYNC' 
failed

(process:17861): GLib-GIO-CRITICAL **: g_simple_async_result_take_error: 
assertion 'error != NULL' failed

(process:17861): Qmi-CRITICAL **: qmi_message_ctl_sync_output_ref: 
assertion 'self != NULL' failed

(process:17861): GLib-GIO-CRITICAL **: g_simple_async_result_take_error: 
assertion 'error != NULL' failed

Any clues?

Thanks!

Kasper