Re: [PATCH 0/4] usb: dwc2: Add USB DRD support on CV1800 Series SoC

From: Chen-Yu Yeh

Date: Fri Nov 13 2026 - 02:45:12 EST


On Tue, Nov 10, 2026 at 03:06:21PM +0800, Inochi Amaoto wrote:
> The current DRD implementation of DRD will cause a interrupt flow on
> CV1800 Series SoC. It has two reasons:
> 1. The DRD does not sync the phy mode when change the controller mode,
> it is a violation of the DRD as it does not rely on a OTG phy.
> 2. The CV1800 is entered with a Session End Detected interrupt, but the
> DRD code has no change on it, so the controller will enter the
> interrupt storm.

Thanks, fixing this in dwc2 makes sense to me. I am dropping my PHY
series [1].

I tested patches 1-3 on a Milk-V Duo 256M (SG2002), on top of
f66469d9905b, with this on the board DTS:

&usb {
dr_mode = "otg";
usb-role-switch;
role-switch-default-mode = "peripheral";
status = "okay";
};

R=/sys/class/usb_role/4340000.usb-role-switch below.

Bare board on a PC, after booting:

# echo device > $R/role
# devmem 0x03000048 32
0x000003C0
# cat /sys/class/udc/*/state
configured

PC$ lsusb | grep 0525
Bus 003 Device 012: ID 0525:a4a2 Netchip Technology, Inc. Linux-USB Ethernet/RNDIS Gadget

# echo host > $R/role
# devmem 0x04340000 32 (GOTGCTL)
0x002C00FC
# devmem 0x04340004 32 (GOTGINT)
0x00000000

The interrupt count did not move over one second in host mode, so no
storm. Switching back to device gave "configured" again, and "none"
went back to device. I did one host/device round on each of two boots.

On the USB & Ethernet IO board, with a USB stick plugged in:

# echo host > $R/role
[ 136.556790] hub 1-1:1.0: 4 ports detected
[ 136.837392] usb 1-1.1: new high-speed USB device number 3 using dwc2
[ 139.174764] sda: sda1 sda2 sda3 sda4

GINTMSK 0xF300081E, HPRT0 0x00001005

Tested-by: Chen-Yu Yeh <chenyou910331@xxxxxxxxx> # Milk-V Duo 256M

> Note: This patches set does not resolve all the problems, as
> usb_add_hcd() will override the role set by dwc2, but I am not
> sure whether it should change skip_phy_initialization of hcd.
> So I kept it as it is and users still needs to set the role
> in the userspace as "role-switch-default-mode" in devicetree
> takes no effect.

I see the same at boot, and the "set mode" print in the PHY driver
shows the order:

[ 1.777738] phy phy-3000000.syscon:phy@48.0: set mode 6
[ 1.778979] phy phy-3000000.syscon:phy@48.0: set mode 5
[ 1.779005] phy phy-3000000.syscon:phy@48.0: set mode 1

That is PHY_MODE_USB_DEVICE from dwc2_drd_init(), then HOST_SS (not
supported by the PHY) and HOST from usb_add_hcd(). The PHY ends up in
host (0x03000048 = 0x142) while the controller is still in device mode
(GINTSTS = 0x04000020). On the bare board the UDC stays "not attached"
and the PC sees nothing. On the IO board the hub does not show up
either. So until a role is written, the port does not work in either
mode. This was the same on all three boots.

I am happy to test the next versions on the Duo 256M, in both device
and host mode, and send a Tested-by.

Once your series is merged, would you like me to send the DTS change
for the Duo 256M, or would you prefer to do it yourself?

[1] https://lore.kernel.org/all/avLC68zOj8WKuslb@inochi.infowork/

Regards,
Chen-Yu