FYI, I think my "usb: add a HCD_DMA flag instead of guestimating DMA capabilities"is the proper core fix for what your patch 1 works around, as the USBcore should not assume DMA capabilities based on the presence of a DMAmask.