On Thu, 22 Jun 2006, Greg KH wrote: > > There are several possible ways to fix this. One is to add suspend and > > resume routines to the endpoint-device driver. Another is to change the > > code that checks for the children being suspended, to make it check only > > for child USB devices and not child endpoints. > > I think it needs to check for _USB_ devices, not just any old device > that could possibly be attached to the main USB device (as this one is.) > What's to stop any other struct device to bind here and cause the same > problem? In my upcoming patches for USB core suspend improvements, one of the changes affects this very piece of code. Instead of looping over all child devices in the driver-model sense, it loops over all interfaces in the active configuration, which is all we care about right here. > Ok, the problem is in verify_suspended(), we are not detecting what type > of device this is. > > Alan, what are you trying to check for here? What "bogus requests" were > you seeing from sysfs that you are trying to filter out? I didn't write that routine, Dave Brownell did. It has been there for ages. The "bogus requests" are attempts by the user to suspend a USB device (by writing to /sys/devices/.../power/state) without first suspending all its children and interfaces. (This can't happen when doing a global suspend because the PM core iterates through the entire device tree. It matters only for "runtime" or "selective" suspend.) The two easiest ways to fix the problem are: Change the code to look through the interfaces in the active configuration instead of using device_for_each_child; or Revert your "endpoints are devices" patch until my upcoming changes are in place. Alan Stern
