USB-C hub's built-in NVMe enclosure (RTL9210) is not powered on hot-plug; works on cold-plug and in macOS (MacBookPro16,2)

[Nazari]

Hardware: MacBook Pro 13" 2020, MacBookPro16,2 (i7-1068NG7, T2). Ubuntu 26.04.1 LTS, stock kernel 7.0.0-30-generic. Kernel cmdline: intel_iommu=on iommu=pt pm_async=off (no t2-specific parameters). ORICO USB-C hub with an integrated M.2 NVMe slot (Realtek RTL9210B-CG bridge), hub's external power attached.

Hot-plug (hub connected while Linux is running, either USB-C port):
The hub itself enumerates fine (Realtek 0bda:0432, SuperSpeed Plus Gen 2x1, self-powered), together with its microSD reader (05e3:0751, shows as an empty /dev/sda) and the ITE billboard (048d:5212). The NVMe bridge (0bda:9210) never appears — not even as a USB 2.0 fallback. uhubctl shows all four downstream ports powered but idle:

Current status for hub 2-2 [0bda:0432 Generic 4-Port USB 3.0 Hub, USB 3.20, 4 ports, ppps]
  Port 1: 02a0 power 5gbps Rx.Detect
  Port 2: 02a0 power 5gbps Rx.Detect
  Port 3: 02a0 power 5gbps Rx.Detect
  Port 4: 02a0 power 5gbps Rx.Detect

dmesg only shows the hub, card reader and billboard; there is no enumeration attempt or error for the bridge. Tried without success: uhubctl -l 2-2 -a cycle, -a off / -a on, echo on > /sys/bus/usb/devices/2-2/power/control (the hub was auto/suspended), unbind/bind of the hub, the other USB-C port. No typec class is exposed on this machine, so there is no PD information to inspect.

Cold-plug (hub connected before powering on the Mac): works immediately.

[    1.439198] usb 2-2.2: new SuperSpeed Plus Gen 2x1 USB device number 3 using xhci_hcd
[    1.464335] usb 2-2.2: New USB device found, idVendor=0bda, idProduct=9210, bcdDevice=f0.01
[    1.464347] usb 2-2.2: Product: RTL9210B-CG
[    1.464358] usb 2-2.2: Manufacturer: Realtek
  Port 2: 0203 power 5gbps U0 enable connect [0bda:9210 Realtek RTL9210B-CG 012345678944]

Driver uas, 10 Gbps, partitions mount normally.

Control tests: the same hub and cable hot-plugged into a desktop Linux machine (Ubuntu, conventional USB-C port) enumerate the RTL9210 right away. The same MacBook booted into macOS detects the SSD hot-plugged.

Interpretation: the hub seems to gate the SSD power rail on a USB-C (PD) negotiation. On T2 Macs that negotiation is done by the T2 firmware at boot, and Linux does not redo it on hot-plug, so devices behind the hub that need it stay unpowered. This looks like the same pattern as the Thunderbolt report in #207 ("connect the external SSD before turning on the Mac").

Workaround: connect the hub before booting (or reboot with it attached).

Reporting mainly for documentation — a note in the wiki's device state page (USB-C docks / PD) or a known-issue entry may save others the debugging.

[deqrocks]

That is an issue with the thunderbolt.c pm management where once entered D3, UAS functionality is lost. We are working on it. A workaround in the driver would be easy to upstream but will come at the cost of higher idle power consumption. There is a bigger picture around missing Apple specific ACPI methods in the thunderbolt driver.
Try running with kernel param acpi_osi=!Darwin and see if it fixes that. It will disable thunderbolt and use Linux acpi paths.

[Nazari]

Thanks — confirmed. With acpi_osi=!Darwin added to the kernel command line (same kernel 7.0.0-30-generic, hub connected after boot), the RTL9210 enumerates immediately on hot-plug:

[21:53:33] usb 2-1: New USB device found, idVendor=0bda, idProduct=0432   (hub)
[21:53:34] usb 2-1.2: new SuperSpeed Plus Gen 2x1 USB device number 3 using xhci_hcd
[21:53:34] usb 2-1.2: New USB device found, idVendor=0bda, idProduct=9210, bcdDevice=f0.01
[21:53:34] usb 2-1.2: Product: RTL9210B-CG
[21:53:34] scsi host1: uas
[21:53:35] sd 1:0:0:0: [sda] Attached SCSI disk

uhubctl shows port 2 in U0 enable connect and the exFAT partition auto-mounts. So it matches your description of the thunderbolt.c D3 issue rather than anything hub/PD related — feel free to retitle. For anyone landing here: cold-plugging (connecting before boot) also works without the parameter. I'll keep acpi_osi=!Darwin for now since I don't use Thunderbolt devices on this machine.