MT7622 USB Device

For Developers
1

Hi everyone,

I’ve been trying to enable USB Gadget (Device) mode on an MT7622-based board. According to the MT7622 datasheet, the USB 2.0 PHY supports "dual-mode" (host/device). However, the default mt7622.dtsi only defines the xhci node (Host-only mode).

Here is the default definition in mt7622.dtsi:

ssusb: usb@1a0c0000 {
    compatible = "mediatek,mt7622-xhci", "mediatek,mtk-xhci";
    reg = <0 0x1a0c0000 0 0x01000>,
          <0 0x1a0c4700 0 0x0100>;
    reg-names = "mac", "ippc";
    /* ... interrupts, clocks, phys ... */
    status = "disabled";
};

To switch to Gadget mode, I enabled the CONFIG_USB_MTU3 (MediaTek USB3 Dual Role controller) and CONFIG_USB_GADGET drivers via kernel_menuconfig, and applied the following DTS override to force the peripheral mode:

&ssusb {
    compatible = "mediatek,mt7622-mtu3", "mediatek,mtu3";
    
    reg = <0 0x1a0c0000 0 0x3000>,
          <0 0x1a0c4700 0 0x0100>;
    reg-names = "mac", "ippc";

    dr_mode = "peripheral";
    maximum-speed = "high-speed";

    phys = <&u2port0 PHY_TYPE_USB2>;
    vusb33-supply = <&reg_3p3v>;
    vbus-supply = <&reg_5v>;
    status = "okay";
};

The Issue:
The mtu3 driver probes successfully and registers the UDC (USB Device Controller), but it fails to initialize the Endpoints (EPs).

Here is the dmesg output:

[    1.076134] mtu3 1a0c0000.usb: failed to get phy-0
[    2.263130] mtu3 1a0c0000.usb: dr_mode: 2, drd: auto
[    2.278119] mtu3 1a0c0000.usb: irq 140
[    2.281891] mtu3 1a0c0000.usb: IP version 0x0(U2 IP)
[    2.291819] mtu3 1a0c0000.usb: fifosz/epnum: Tx=0x0/0, Rx=0x0/0
[    2.297756] mtu3 1a0c0000.usb: dma mask: 32 bits

Low-level MMIO Debugging (The Root Cause):
To determine if this is a driver initialization bug (e.g., missing clocks/power) or a physical hardware limitation, I performed a manual register dump. Register offsets and bitfields were identified using the mtu3_hw_regs.h header from the Linux 6.6 source tree (drivers/usb/mtu3/).

  1. Checking the Host (xHCI) base register:
    devmem 0x1a0c0000 32 -> 0x00960020 (xHCI controller is alive and powered).

  2. Checking Device Capability Registers (UDC/MTU3):

  • 0x1a0c0c10 (U3D_CAP_EPINFO) -> 0x00000000
  • 0x1a0c0c08 (U3D_CAP_EPNTXFFSZ) -> 0x00000000
  • 0x1a0c0c0c (U3D_CAP_EPNRXFFSZ) -> 0x00000000
  • 0x1a0c47a4 (U3D_SSUSB_HW_SUB_ID / IP_TRUNK_VERS) -> 0x00000000

Even after manually clearing the power-down bits in the IPPC registers (DEV_PDN in PW_CTRL2, and U2_CTRL_0P) and performing a SW reset via devmem, the Capability registers consistently return 0x00000000.

The fact that capability registers consistently return zero—even with the IPPC power-down bits cleared—suggests two possibilities. Either the digital UDC logic (MTU3 IP/SRAM) is physically absent/fused off in this specific SoC revision, or there is a critical, undocumented initialization sequence (such as a secure-bridge setting or a specific clock-gating dependency) required to expose the Device MAC.

Has anyone successfully initialized the MTU3 core on the MT7622 in peripheral mode? Is this a known hardware-level SKU limitation, or am I missing a "magic" sequence to wake up the UDC logic?

Thanks!

1 Like
2

Missing phy driver ?

3

If I understand correctly, the failed to get phy-0 at [ 1.076134] happens simply because the PHY driver hasn't finished initializing yet.

Then, the kernel retries at [ 2.263130], and by that time the PHY is already initialized, so the driver successfully picks it up and continues the boot process.

Is my assumption about this correct?

4

Do you have kmod-mt76-usb installed?

5

I didn't enable kmod-mt76-usb. As far as I understand, the mt76 stack in Linux is specifically designed for MediaTek Wi-Fi drivers. I understand it's needed when the SoC acts as a USB host for external wireless adapters. It shouldn't be related to the UDC mtu3 driver, which I'm trying to initialize. Or am I missing something?