ice: GNSS kthread worker causes hung task warnings when no GPS signal available

[ionutnechita]

Description

When using Intel E810 NICs with GNSS capability, the ice-gnss-* kthread workers cause hung task warnings when the GNSS receiver has no signal (e.g., no antenna connected or weak signal).

Environment

• Driver version: ice 2.5.4
• Kernel version: 6.18.21 / 6.18.22
• Hardware: Dell PowerEdge XR11
• NICs:
  • Intel E810-C for SFP (0000:51:00.x) - with GNSS antenna, working
  • Intel E810-C for QSFP (0000:8a:00.x) - no GNSS antenna connected

Steps to reproduce

1. Install E810 NIC with GNSS capability
2. Do not connect GNSS antenna (or have weak/no GPS signal)
3. Load ice driver
4. Wait ~120 seconds

Expected behavior

Driver should handle missing GNSS signal gracefully without blocking.

Actual behavior

Hung task warnings appear in dmesg:

[  248.064851] INFO: task ice-gnss-0000:5:603 blocked for more than 124 seconds.
[  248.064879]       Tainted: G           OE       6.18.21-xxx #8
[  248.064886] task:ice-gnss-0000:5 state:D stack:0     pid:603   tgid:603   ppid:2
[  248.064898] Call Trace:
[  248.064902]  <TASK>
[  248.064908]  __schedule+0x474/0x1210
[  248.064950]  schedule+0x27/0xf0
[  248.064959]  schedule_preempt_disabled+0x15/0x30
[  248.064968]  kthread+0xa4/0x220
[  248.064985]  ret_from_fork+0x273/0x2a0
[  248.065003]  ret_from_fork_asm+0x1a/0x30
[  248.065018]  </TASK>

GNSS status

# Working NIC (with antenna):
$ cat /dev/gnss0
<valid NMEA data with GPS fix, 9 satellites>

# NIC without antenna (causes hung task):
$ cat /dev/gnss1
$GNRMC,,V,,,,,,,,,,N,V*37
$GNGGA,,,,,,0,00,99.99,,,,,,*56

Root cause analysis

In ice_gnss_read() (src/ice_gnss.c), the polling loop iterates up to ICE_MAX_UBX_READ_TRIES (255) times with msleep(20) between attempts. When GNSS has no signal, this can block for ~5 seconds per cycle.

Additionally, during module unload, kthread_cancel_delayed_work_sync() waits for the work to complete, which can take a long time if stuck in the polling loop.

Suggested fix

Add a stopping flag to allow graceful termination of the read loop during shutdown:

1. Add bool stopping field to struct gnss_serial
2. Check stopping flag at the beginning of ice_gnss_read() and exit early
3. Check stopping flag inside the polling loop
4. Don't requeue the delayed work if stopping is true
5. Set stopping = true in ice_gnss_exit() before calling kthread_cancel_delayed_work_sync()

See attached patch for implementation.

Workaround

Connect GNSS antenna to all E810 NICs with GNSS capability, or disable hung task warnings:

sysctl -w kernel.hung_task_timeout_secs=0

0001-ice-gnss-fix-hung-task-when-GNSS-has-no-signal.patch

0001-ice-gnss-fix-hung-task-when-kthread-misses-initial-w.patch