On Mon, Dec 09, 2024 at 02:32:37PM +0200, Roger Quadros wrote:
Hello Roger,
On 06/12/2024 12:07, Siddharth Vadapalli wrote:
[...]
Looking at the list of compatibles in dwc3-generic.c and focusing on the compatible "ti,keystone-dwc3", I see that this compatible is present in: dts/upstream/src/arm/ti/keystone/keystone.dtsi among other keystone device-tree files. The hierarchy of the nodes and compatibles for the USB Subsystem and the USB Controller in Keystone is similar to that of AM62*. The difference however is the following: A) For non AM62* devices:
note there are 2 cases here:
A1) that enable CONFIG_USB_DWC3_GENERIC
USB Subsystem [Wrapper] driver is dwc3-generic.c
Yes.
USB Controller [DWC3] driver is xhci-dwc3.cNo.
Note that none of the platforms that enable CONFIG_USB_DWC3_GENERIC enable CONFIG_USB_XHCI_DWC3. They should not because dwc3-generic.c takes care of registering the XHCI driver via xhci_register/deregister()
Thank you for pointing this out.
A2) that don't enable CONFIG_USB_DWC3_GENERIC
These are usually only interested in host mode, they enable CONFIG_USB_XHCI_DWC3 and may also enable CONFIG_USB_XHCI_DWC3_OF_SIMPLE
So USB Wrapper/glue driver is DWC3_OF_SIMPLE USB driver is xhci-dwc3.c
Some platforms have their own glue CONFIG_USB_XHCI_STI dwc3-sti-glue.c CONFIG_USB_XHCI_OCTEON dwc3-octeon-glue.c
B) For AM62* devices: USB Subsystem [Wrapper] driver is dwc3-am62.c which exposes its own ".glue_configure" callback similar to what is done in dwc3-generic.c for other devices.
The Subsystem driver is still dwc3-generic.c. Glue driver is dwc3-am62.c
USB Controller [DWC3] driver is xhci-dwc3.cNo. We don't require CONFIG_USB_XHCI_DWC3 as XHCI registration is done by dwc3-generic. USB Host driver is USB_XHCI_HCD.
Ok.
So maybe non AM62* devices have: dwc3-generic.c for Subsystem + xhci-dwc3.c for USB Controller while AM62* devices have: dwc3-am62.c for Subsystem + dwc3-generic.c (neither Subsystem nor Controller?) + xhci-dwc3.c for USB Controller We could probably end up with the two driver situation similar to non AM62* devices if we update dwc3-generic.c to handle the configuration performed by dwc3-am62.c. But that still means we have two drivers: dwc3-generic.c and xhci-dwc3.c Maybe viewing the output of "dm tree" on a non AM62* device with both CONFIG_USB_XHCI_DWC3 and CONFIG_USB_DWC3_GENERIC enabled will let us know whether that is a valid configuration or not.
Enabling both is an invalid configuration and no config does it. AM62 was doing it wrong and so we fix it with this patch.
Understood. Thank you for explaining. I agree that the fix in this patch is correct and your earlier suggestion of Kconfig guards: "Maybe we still need some Kconfig guards that prevent both CONFIG_USB_XHCI_DWC3 and USB_DWC3_GENERIC to be set together as 2 drivers will claim the controller if dr_mode is USB_DR_MODE_OTG or USB_DR_MODE_HOST." could be implemented in a follow up patch.
Regards, Siddharth.