Re: [RFC PATCH] thunderbolt: Tunnel only the first DP OUT of Apple Studio Displays

From: Mika Westerberg

Date: Fri Nov 13 2026 - 07:17:53 EST


Hi,

On Thu, Nov 12, 2026 at 05:38:46PM -0500, Duff O'Melia wrote:
> Hi Mika,
>
> On Mon, Nov 09, 2026 at 11:30:39AM +0100, Mika Westerberg wrote:
> > Can you provide full dmesg with thunderbolt.dyndbg=+p and drm.debug=0x10e
> > in the kernel command line when you have everything connected and the issue
> > reproduces?
>
> Sure, it's attached, xz-compressed. It's from a reboot with everything
> connected, on Omarchy's 7.2.15 kernel with its stock Thunderbolt driver
> (my patch not applied). I added log_buf_len=64M so the start of the boot
> isn't lost. I removed MAC addresses, USB serial numbers and some
> firewall log lines, and changed nothing else.
>
> In short:
>
> - 0-3 is the 2026 model (0x8025). Both of its DP OUT adapters got
> tunneled: 0:5 <-> 3:11 and 0:10 <-> 3:12.
> - 1-1 is a 2022 model (0x801f) on the second host router. It got
> 0:6 <-> 1:10.
> - 0-1 is the other 2022 model. It came up last, found no free DP IN
> adapter ("no suitable DP IN adapter available, not tunneling") and
> stays dark.
> - In xe, DP-1 (DPTUN 0:1:5) is the 2026 model's first tile, DP-3
> (DPTUN 0:1:10) its second tile, and DP-2 (DPTUN 1:1:6) the 2022
> model. DP-1 and DP-2 run at 5120x2880.
>
> One correction to my patch description: on this boot DP-3 isn't
> disabled. Pipe D is active on it at 2560x2880 with no framebuffer on its
> primary plane, and the compositor doesn't list it as a monitor. So the
> second tile's stream does come up, but it shows nothing.

Thanks! Confirmed:

[ 1.099566] thunderbolt 0000:00:0d.2: 0:5: DP IN resource available
[ 1.101235] thunderbolt 0000:00:0d.2: 0:10: DP IN resource available

Both DP INs of domain0

[ 2.417970] thunderbolt 0000:00:0d.2: 0:5 <-> 3:11 (DP): activating

First DP tunnel

[ 2.428951] thunderbolt 0000:00:0d.3: acking hot unplug event on 0:10
[ 2.429566] thunderbolt 0000:00:0d.2: 0: allocated DP resource for port 10
[ 2.433756] thunderbolt 0000:00:0d.2: 0:10 <-> 3:12 (DP): activating

Secodn DP tunnel. However, the DP resource in 0:10 is shared with domain1
0:10 as well so that leaves only one DP IN for domain1 (0:6).

[ 1.169534] thunderbolt 0000:00:0d.3: 0:6: DP IN resource available
[ 1.170368] thunderbolt 0000:00:0d.3: 0:10: DP IN resource available
[ 2.699961] thunderbolt 0000:00:0d.3: 0:6 <-> 1:10 (DP): activating

And this is that tunnel. Then you run out of DP INs.

It would be good if Imre could take a look from DP perspective if that
second tile in the 2026 model can be used.

I do have an internal patch series where userspace can affect what DP
IN/OUT pairs should be selected via a BPF hook but that's still work in
progress. That would help here.