[SRU][N][PATCH 0/2] TBT call trace while connecting TBT4 monitor on TBT5 port

[AceLan Kao]

From: "Chia-Lin Kao (AceLan)" <acelan.kao at canonical.com>

BugLink: https://bugs.launchpad.net/bugs/2137613

[Impact]

When connecting Thunderbolt devices (especially monitors like Dell
U2725QE), users see alarming kernel backtraces in dmesg during device
enumeration. While the devices eventually work after automatic
reconnection, the call traces cause user concern and can trigger
automated bug reporting tools.

Error log example:
```
[ 36.031530] thunderbolt 0000:c7:00.6: PCIe Down path activation failed
[ 36.031531] WARNING: drivers/thunderbolt/path.c:589 at 0x0, CPU#12: pool-/usr/libex/3145
[ 36.031605] CPU: 12 UID: 0 PID: 3145 Comm: pool-/usr/libex Tainted: G D W 6.18.0+ #8
[ 36.031610] RIP: 0010:tb_path_activate+0x126/0x530 [thunderbolt]
[ 36.031637] Call Trace:
[ 36.031638] <TASK>
...
```

The issue occurs when:
- Type-C connections have transient electrical issues
- During lane bonding transitions (single lane to dual lane)
- The Thunderbolt port's control channel is temporarily unavailable

The devices typically recover automatically within a few seconds and work
normally, but the kernel backtrace (tb_WARN) is generated unnecessarily
for these expected transient conditions.

Affected hardware:
- Dell U2725QE Thunderbolt monitor (USB4 device 8087:b26)
- Other Thunderbolt/USB4 devices experiencing similar transient connection issues
- AMD and Intel Thunderbolt controllers

[Fix]

Modify tb_path_activate() in drivers/thunderbolt/path.c to differentiate
between expected transient failures and actual errors:

- For -ENOTCONN errors: Use tb_warn() to log the error without generating a kernel backtrace
- For all other errors: Keep tb_WARN() to generate the full call trace for debugging

This approach aligns with the existing comment in drivers/thunderbolt/ctl.c
which states that TB_CFG_ERROR_PORT_NOT_CONNECTED "can happen during
surprise removal" and we should "not warn" about it.

The fix does not suppress the warning message itself - users and developers
can still see the path activation failure in dmesg. It only removes the
unnecessary kernel backtrace (stack dump, register dump, etc.) for this
specific expected transient condition.

Patch: https://lore.kernel.org/lkml/20260102031905.27416-1-acelan.kao@canonical.com/T/#u
("thunderbolt: Suppress call trace for transient -ENOTCONN errors during
path activation")

[Test Plan]

Hardware needed:
- Dell U2725QE Thunderbolt monitor or similar Thunderbolt device that
  exhibits transient connection issues
- System with Thunderbolt 3/4 or USB4 controller

Test steps:
```bash
# Clear dmesg
sudo dmesg -C

# Connect Dell U2725QE or similar Thunderbolt device
# Wait 10 seconds

# Check for call traces
dmesg | grep -A 30 "path activation failed"
```

Without the patch: A full kernel backtrace appears with WARNING, RIP, Call
Trace, register dump, etc.

With the patch: Only a simple warning message appears without the backtrace:
```
thunderbolt 0000:c7:00.6: PCIe Down path activation failed (port not connected)
```

[Where problems could occur]

The patch modifies error reporting in the Thunderbolt path activation code,
which could affect debugging and error handling:

1. **Thunderbolt subsystem**: If there are genuine bugs that manifest as
-ENOTCONN errors (not just transient issues), developers might miss
important debugging information because the full backtrace won't be
generated. This would make it harder to diagnose actual Thunderbolt
controller bugs or firmware issues.

Chia-Lin Kao (AceLan) (1):
  UBUNTU: SAUCE: thunderbolt: log path activation failures without WARN
    backtraces

Mika Westerberg (1):
  thunderbolt: Show path name in debug log when path is deactivated

 drivers/thunderbolt/path.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

--
2.51.0