On Tue, Aug 02, 2022 at 06:33:36PM +0200, Johan Hovold wrote: > On Mon, Aug 01, 2022 at 12:30:15PM +0530, Krishna Kurapati wrote: > > Dwc3 Qcom driver makes use of usb_hub_find_child API in its efforts > > to get speed of connected devices (HS/LS/FS) and enable interrupts > > accordingly. > > > usb_hub_find_child API is a part of usb core compiled > > either into the kernel or as a module (CONFIG_USB= Y or M). In some > > builds (make randconfig for i386) CONFIG_USB is not enabled and the > > usb core is not compiled resulting in linking errors. > > Please replace the above with something more succinct. Whether USB core > is built as a module or not is completely irrelevant. The problem is > that the qcom dwc3 driver can be built and used without host support. > > > Provide stubs for dwc3_qcom_read_usb2_speed function to use > > usb_hub_find_child API only if CONFIG_USB is enabled. Else return > > USB_SPEED_UNKNOWN. > > The fact that you need to do this is an indication that something is > wrong with the current implementation. The glue driver shouldn't be > touching the host driver internal state directly like this. > > As pointed out here: > > https://lore.kernel.org/all/20220802151404.1797-4-johan+linaro@xxxxxxxxxx/ > > dwc3_qcom_read_usb2_speed() is indeed broken and currently triggers a > NULL-pointer dereference when the controller is used in peripheral mode. > > But for now I guess something like this is needed even if we try to > avoid stubs in implementation files. This is how I think this should be fixed instead: https://lore.kernel.org/all/20220804151001.23612-4-johan+linaro@xxxxxxxxxx/ and which keeps the ifdeffery minimal. I included the patch in v2 of the series that addresses the other problems with this code. Johan
