Re: [BUG] thunderbolt: Anker Prime TB5 (1ea:83b5) reconnect-loops on Meteor Lake (8086:7ec3) unless clx=0

[Mika Westerberg <mika.westerberg@linux.intel.com>]

Hi,

On Thu, Sep 24, 2026 at 07:18:14AM +0000, Kurt Lieber wrote:
> Hi,
> 
> The Anker Prime Thunderbolt 5 dock reconnect-loops on an LG Gram with Meteor Lake unless the thunderbolt module is loaded with clx=0. With the default (clx=Y) the router is found, TMU setup fails with -ENOTCONN, the USB3 and DisplayPort tunnels are aborted, and the router is dropped. That repeats about once a second, including from a cold plug at boot.
> 
> Hardware:
>   Laptop: LG Gram 17Z90S-H.ADB9U1, Intel Core Ultra 7 155H
>   Host router: 0000:00:0d.3 Intel Core Ultra 200 Series Thunderbolt DMA1
>                [8086:7ec3] (rev 10), subsystem [1854:051d]
>   Dock: Anker Prime Docking Station (14-in-1, 8K, Thunderbolt 5)
>         kernel: vendor=0x1ea device=0x83b5
>         bolt uuid: 8ab48780-00ce-1da2-ffff-ffffffffffff
>         negotiated link: 40 Gb/s = 2 lanes * 20 Gb/s (TB4 fallback)
>   Retimer on the host port: vendor=0x8087 device=0xd9c
>   External display behind the dock: LG ULTRAFINE, 3840x2160
> 
> Software:
>   Linux anarchy 7.2.5-3-omarchy #1 SMP PREEMPT_DYNAMIC Mon, 14 Sep 2026
>   This is the Omarchy build of 7.2.5. modinfo reports clx as "allow low
>   power states on the high-speed lanes (default: true)". The published
>   linux-omarchy patch set does not touch drivers/thunderbolt/. bolt 0.9.11.
>   thunderbolt.host_reset was Y. pcie_aspm=off was not set.
> 
> What happens with clx=Y, dock already plugged in at boot:
> 
>   thunderbolt 0-1: new device found, vendor=0x1ea device=0x83b5
>   thunderbolt 0-1: Anker Prime Docking Station (14-in-1, 8K, Thunderbolt 5)
>   thunderbolt 0-0:1.1: new retimer found, vendor=0x8087 device=0xd9c
>   thunderbolt 0000:00:0d.3: 1: TMU: failed to enable mode uni-directional, LowRes: -107
>   thunderbolt 0000:00:0d.3: 1: failed to enable TMU
>   thunderbolt 0000:00:0d.3: 1: USB3 tunnel creation failed
>   thunderbolt 0000:00:0d.3: 1:12: DP tunnel activation failed, aborting
>   thunderbolt 0000:00:0d.3: 1:12: DP tunnel activation failed, aborting
>   thunderbolt 0-0:1.1: retimer disconnected
>   thunderbolt 0-1: device disconnected
> 
> The same sequence then repeats. While bolt policy was iommu, boltd also logged
> "auto-auth: authorization failed: kernel error: write error: Input/output error"
> and the kernel additionally logged "PCIe tunnel activation failed, aborting".
> Setting the stored bolt policy to manual stopped those authorization attempts.
> The TMU / USB3 / DP disconnect loop continued with the router left at
> authorized=0.
> 
> Workaround, same boot, no dock power cycle:
> 
>   modprobe -r thunderbolt   (typec consumers unloaded first)
>   modprobe thunderbolt clx=0
> 
> After that, /sys/module/thunderbolt/parameters/clx reads N, the router stays
> connected, DP-1 comes up at 3840x2160, and the dock USB3 hub, keyboard, and
> mouse enumerate. No further TMU or tunnel failures. Authorizing the router
> with boltctl after that succeeded on the first try. host_reset was left at Y.
> 
> This looks like the same class of failure as Launchpad 2165852 (AMD Rembrandt
> plus a ThinkPad TB4 dock, Goshen Ridge 0108:2031), which also loops on
> "failed to enable TMU" / "USB3 tunnel creation failed" unless thunderbolt.clx=0.
> The pair here is different: Meteor Lake host 8086:7ec3 and Anker router
> 1ea:83b5. QUIRK_NO_CLX already exists; this router does not hit it.
> 
> I can retest with thunderbolt.dyndbg=+p or try a patch if that is useful.

Can you test with the dynamic debug enabled and provide full dmesg? Or
alternatively run:

  # tbdump -r 1 -a 1 -vv -N2 LANE_ADP_CS_0

and provide the output. This needs to be done when the dock is connected
(with clx=0 is okay). This should show if the dock claims to support CL
states.

You can get tbdump from tbtools [1].

[1] https://github.com/intel/tbtools