Update: USB4 disconnects and failed automatic NIC recovery,
September 17–18, 2026. All clock times are UTC−07:00.

Configuration:
- GEEKOM A7 Max: BIOS 1.29, EC 1.03.
- WAVLINK WL-NWU342G: JHL7440 NVM 69.84,
  AQC113 firmware 1.5.48.
- Kernel 7.0.0-31-generic, in-tree Atlantic driver.
- pci=nocrs removed. Remaining relevant parameters:
  pcie_aspm.policy=performance iommu=pt pcie_port_pm=off

At September 17 18:07:50, the kernel logged:
- thunderbolt 1-0:2.1: retimer disconnected
- pcieport 0000:00:04.1: Link Down / Card not present
- thunderbolt 1-2: device disconnected

The enclosure returned approximately 0.92 seconds later.
Thunderbolt authorization succeeded, but Ethernet did not recover.

The allocation sequence shows:
[193849.829835] 66:00.0: bridge window [mem 0xac000000-0xc3ffffff]: assigned
[193849.830282] 67:02.0: bridge window [mem 0xac000000-0xac0fffff]: assigned
[193849.830444] 67:04.0: bridge window [mem 0xac100000-0xc3efffff]: assigned
[193849.834855] 69:00.0: xHCI Host Controller
[193849.998084] 68:00.0: [1d6a:04c0] PCIe Endpoint
[193850.001240] 67:01.0: bridge window [mem size 0x00500000]: can't assign; no 
space
[193850.006123] 68:00.0: probe with driver atlantic failed with error -5

The parent bridge received 384 MiB of non-prefetchable MMIO,
but almost all was assigned to sibling bridge 67:04.0.
The NIC appeared approximately 163 ms after xHCI initialization,
after its empty branch's memory window had been disabled.
Its required 5 MiB window and BARs then failed allocation.

At September 18 05:38:57, another complete connection loss
occurred while the NIC was already unbound. Automatic rediscovery
and successful authorization were again followed by the same
bridge/BAR allocation failure and Atlantic probe error -5.

The captured late-event windows do not show the earlier
unsupported-request AER/TLP pattern preceding these disconnects.
The filtered system-journal windows show no system sleep/suspend
transition; device runtime power management remains unexcluded.

An RJ45 cable swap occurred September 17 at an unknown time.
The USB-C connection was not changed.

Recovery:
On September 18 at approximately 11:17, the guarded script removed
the complete enclosure subtree at 0000:66:00.0 and rescanned the
parent bus. Without rebooting or changing cables, this restored:
- AQC113 at 0000:68:00.0, bound to atlantic.
- enp8s0 UP/LOWER_UP.
- Speed: 10000Mb/s; Link detected: yes.

The initiating USB4 disconnect remains unexplained. The subsequent
failure to recover Ethernet is documented as a PCIe resource-
allocation failure during late NIC discovery, reproducing without
pci=nocrs. Please investigate bridge-window reservation/rebalancing
during hotplug rediscovery.

GEEKOM and WAVLINK support have also been updated.

** Attachment added: "kernel-before-recovery.log"
   
https://bugs.launchpad.net/ubuntu/+source/linux/+bug/2167202/+attachment/6001100/+files/kernel-before-recovery.log

-- 
You received this bug notification because you are a member of Ubuntu
Bugs, which is subscribed to Ubuntu.
https://bugs.launchpad.net/bugs/2167202

Title:
  thunderbolt/pci JHL7440/AQC113 fails to recover after tunnel
  disconnect; BAR allocation fails

To manage notifications about this bug go to:
https://bugs.launchpad.net/ubuntu/+source/linux/+bug/2167202/+subscriptions


-- 
ubuntu-bugs mailing list
[email protected]
https://lists.ubuntu.com/mailman/listinfo/ubuntu-bugs

Reply via email to