I reproduced the failure using Ubuntu’s signed in-tree `atlantic` driver on kernel `7.0.0-31-generic`. The external Marvell 2.5.16.0 DKMS driver was not installed for this kernel and remained only in DKMS’s `built` state.
During boot, Linux enumerated the JHL7440 bridges and xHCI controller first and allocated their bridge windows. The AQC113 appeared approximately 165 ms later. Its downstream bridge could not obtain the required 5 MiB memory window, all AQC113 BAR assignments failed with `no space`, and the in-tree driver failed to probe with error `-5`. The Thunderbolt enclosure still appeared authorized at 40 Gb/s, but no Ethernet interface was created. Removing the complete JHL7440 PCI subtree and rescanning the empty parent bus caused Linux to redistribute the bridge windows and assign every AQC113 BAR. Ubuntu’s in-tree driver then bound successfully, and `enp8s0` established a 10,000 Mb/s link without rebooting. This rules out the external Marvell DKMS driver as the cause of the PCI resource-allocation failure. The evidence points to a PCI/Thunderbolt enumeration timing problem: the AQC113 appears after the initial bridge resource allocation, and the automatic hot-plug path does not resize the sibling bridge windows afterward. Testing continues with the in-tree driver to determine whether the separate spontaneous Thunderbolt Link Down also reproduces. ** Attachment added: "intree driver supplemental failure file" https://bugs.launchpad.net/ubuntu/+source/linux/+bug/2167202/+attachment/6000018/+files/intree-driver-failure.txt -- 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
