Until we have the tracepoints, you can experiment with this test patch.
If my suspiction is correct, then this may resolve your issue:
diff --git a/drivers/usb/dwc3/gadget.c b/drivers/usb/dwc3/gadget.c
index 89fc690fdf34..29dbb889a0e2 100644
--- a/drivers/usb/dwc3/gadget.c
+++ b/drivers/usb/dwc3/gadget.c
@@ -2682,6 +2682,8 @@ static int dwc3_gadget_soft_disconnect(struct dwc3 *dwc)
*/
ret = dwc3_gadget_run_stop(dwc, false);
+ usb_gadget_set_state(dwc->gadget, USB_STATE_NOTATTACHED);
+
/*
* Stop the gadget after controller is halted, so that if needed, the
* events to update EP0 state can still occur while the run/stop
--
Thanks,
Thinh
