RK3588 USB3 DRD support

RK3588 has two USB3 DRD controllers (called TYPEC0 and TYPEC1 in Rock 5B schematics). One of them is specific to RK3588, the other one also exists in RK3588S. Unlike other RK3588 IP it does not follow the RK356x design, but the RK3399 design. The USB controller is based on Synopsys DesignWare Core 3 and handled by the dwc3 kernel driver. The only downstream change is adding quirk-skip-phy-init DT property. To use the controller two PHYs are involved - one for USB2 and USB3 respectively. The USB2 PHY has already been upstreamed. The USB3 PHY is new (drivers/phy/rockchip/phy-rockchip-usbdp.c) and also handles DisplayPort. Testing the Displayport functionality is not part of this issue.

Both, EVB1 and Rock 5B use usbdrd3_0 for their USB-C port and usbdrd3_1 for a USB3 Type A port in host-only mode. In case of Rock 5B it's the lower USB3 port (close to the PCB). The upper USB3 Type A port from Rock 5B is handled by a different IP with yet another PHY and already supported upstream.

Sub-Tasks:

• Get the USB3 DRD based ports working
  • Port and phy-rockchip-usbdp driver
  • Investigate quirk-skip-phy-init (it's probably not upstreamable), introducing commmit
  • Add DT binding for the PHY and new fallback compatible for the DWC3 (Note: T41616 needs the same)
  • Add USB3 DRD nodes (incl. related PHYs) to rk3588s.dtsi and rk3588.dtsi
  • Enable USB3 DRD ports in EVB1 DTS
  • Enable USB3 DRD ports in Rock 5B DTS
  • Enable USB3 DRD port in Rock 5A DTS
  • Add new phy driver to arm64 defconfig
• Cleanup PHY driver
• Send patches upstream
  • PHY driver + bindings
  • DT patches for RK3588, Rock 5A, Rock 5B, EVB1
• Integrate into our rk3588 integration branch
• Test functionality with the CI Debian image
• Wait for patches to land upstream

added T41615 label

assigned to @sre

Current status:

1. I'm mostly happy with the usbdp code from the rk3588-test branch. It works perfectly fine for the USB-A ports found on Rock 5A, Rock 5B and EVB1 and the generic bits have been cleaned up to use latest common kernel APIs.

2. When doing usb start in U-Boot (with a USB3 enabled U-Boot, upstream does not yet support it and vendor U-Boot does not support it either. Our U-Boot branch and our Rock 5B images do support it), the USBDP PHY fails to probe:

[    1.183568] rockchip-usbdp-phy fed90000.phy: cmn ana lcpll lock timeout
[    1.184771] rockchip-usbdp-phy fed90000.phy: failed to init combophy
[    1.185931] phy phy-fed90000.phy.8: phy init failed --> -110
[    1.186957] dwc3 fc400000.usb: error -ETIMEDOUT: failed to initialize core
[    1.188224] dwc3: probe of fc400000.usb failed with error -110

3. When enabling full USB-C support, the Rock 5B resets. Apparently this happens because the board is supplied via USB-C PD and fusb302 driver does something, which results in a power-supply reset. This still needs to be investigated. Current workaround: Mark the fusb302 DT node as status = "disabled";

4. On EVB1 the board is not powered via USB-C, so enabling is easier. Here I noticed some issues with port data role. Apparently the port often switches into DEVICE mode, even when it should go into HOST mode. It does work with a powered USB-C hub, but otherwise uses wrong mode.

For proper USB-C support, vendor DT currently is using svid property. Hans de Goede added core support for that some time ago, but did not add any DT bindings since it was only used with in-kernel device properties. But Dmitry Baryshkov recently worked on a series adding support:

https://lore.kernel.org/all/20231204020303.2287338-1-dmitry.baryshkov@linaro.org/

There is no response, but the binding patch is in linux-next and should land in 6.8 via the USB tree.

mentioned in commit 2acc59dd

I received a few cleanup patches from Heiko Stübner for the usbphy. They are now added to the rk3588-test branch.

I received a new driver from Rockchip, with cleanups from their side. This has now also been merged into the rk3588-test branch. Additionally I did a bunch of further cleanups and more tests. I now noticed, that the EVB1 USB3 Type-A port handled by this PHY negotiates 480M when the device is plugged since boot, but 5000M when the device is hot-plugged. The reason for that is not yet clear.

I also noticed, that my USB-C hub with HDMI is only detected as 480M, even though it's capable of SuperSpeed. OTOH plugging a USB3 capable device directly gets it to 5000M.

Considering 480M is better than nothing, I think this should not block further upstreaming.

marked the checklist item Cleanup PHY driver as completed

marked the checklist item Investigate quirk-skip-phy-init (it's probably not upstreamable), introducing commmit as completed

marked the checklist item DT patches for RK3588, Rock 5A, Rock 5B, EVB1 as completed

marked the checklist item PHY driver + bindings as completed

marked the checklist item Send patches upstream as completed

PATCHv1: https://lore.kernel.org/linux-rockchip/20240209181831.104687-1-sebastian.reichel@collabora.com/

PATCHv2: https://lore.kernel.org/linux-rockchip/20240213163609.44930-1-sebastian.reichel@collabora.com/

PATCHv3: https://lore.kernel.org/all/20240216170514.75200-1-sebastian.reichel@collabora.com/

This needs a PATCHv4 once v6.9-rc1 has been tagged instead of a simple resend. Only change is rk_udphy_phy_xlate now needing a const struct of_phandle_args *args argument.

PATCHv4: https://lore.kernel.org/all/20240325141653.84910-1-sebastian.reichel@collabora.com/

PATCHv5: https://lore.kernel.org/all/20240408225109.128953-1-sebastian.reichel@collabora.com/

The DT binding and the PHY driver just got merged by Vinod. After my discussion with Heiko at Embedded World I'm pretty sure the other patches will also be landing in linux-next soon.

marked the checklist item Wait for patches to land upstream as completed