Second hardware combo hitting the same failure class System: - ThinkPad P1 Gen9 (21UFS0KC00), BIOS N4SET30W 1.09 - Lenovo ThinkPad Thunderbolt 5 Smart Dock 7500 - 40BA (not Dell TB4 — different dock vendor/gen, same failure class) - 3 monitors: Dell U2415 (1920x1200), LG HDR 4K (3840x2160), built-in eDP (3200x2000) - Kernel 7.0.0-34-generic (Linux Mint 22.2 / Ubuntu 24.04 base, Cinnamon 6.6.4) - Intel Panther Lake, xe driver — same platform as the original report
Unlike the original report (cold-boot-only), I have two additional 100%-reproducible triggers on this hardware, both ending in a genuine kernel-level hang (Xorg goes into uninterruptible D state, unrecoverable without a hard reset): 1. Opening Cinnamon's Display Settings applet with the dock's MST-tunneled monitors attached, changing any setting and applying it 2. The screensaver/DPMS wake-from-idle cycle. Both produce this signature in dmesg, with an escalating pre-freeze pattern before the hard hang: xe 0000:00:02.0: [drm] *ERROR* [CONNECTOR:...] Failed to get ACT after 3000 ms workqueue: intel_atomic_cleanup_work [xe] hogged CPU for >10000us 4 times, consider switching to WQ_UNBOUND workqueue: intel_atomic_cleanup_work [xe] hogged CPU for >10000us 5 times, ... workqueue: intel_atomic_cleanup_work [xe] hogged CPU for >10000us 11 times, ... workqueue: intel_atomic_cleanup_work [xe] hogged CPU for >10000us 19 times, ... (count keeps climbing — 4, 5, 7, 11, 19, 35, 67, 131... — until hard freeze) Additional isolation donet: - Not NVIDIA-related: this is a muxless hybrid laptop; NVIDIA never drives any display output regardless of driver state. Confirmed across nouveau, the -open driver, and the closed nvidia-driver-610 package (which actually refuses to load its kernel module on this GPU at all — NVRM: ... requires use of the NVIDIA open kernel modules). xe is the only driver touching any output in every configuration tested. Rules out NVIDIA as a contributing factor. - Cross-hardware control test: connected the identical dock model to a completely different laptop (ThinkPad P1 Gen4i, Tiger Lake, hardware-MUXed NVIDIA driving all outputs, i915 not even loaded). Zero ACT/link-training/atomic-cleanup-workqueue errors across multiple dock hotplug/replug cycles. This isolates the fault to the xe/Panther Lake/muxless side of the pairing, not the dock or monitors. - Also observed: after a wedge, only a full Thunderbolt tunnel teardown+rebuild that lands on a new enumeration path (e.g. thunderbolt port 1-1 → 1-3) clears the wedged state (a same-port reconnect reusing the existing tunnel does not recover it). This is consistent with your root-cause note that the tunnel isn't recreated once torn down early, in my case it's torn down by the driver getting into a bad MST/ACT state post-boot, not just at cold boot, but the underlying "xe doesn't cleanly re-establish the DP tunnel" mechanism looks like the same bug surfacing on a different trigger. Is this related? https://gitlab.freedesktop.org/drm/xe/kernel/-/work_items/9421 -- You received this bug notification because you are a member of Ubuntu Bugs, which is subscribed to Ubuntu. https://bugs.launchpad.net/bugs/2155195 Title: [Panther Lake/xe] DisplayPort-over-Thunderbolt external monitor not detected at cold boot (regression in linux 7.0.0-22) To manage notifications about this bug go to: https://bugs.launchpad.net/ubuntu/+source/linux/+bug/2155195/+subscriptions -- ubuntu-bugs mailing list [email protected] https://lists.ubuntu.com/mailman/listinfo/ubuntu-bugs
