Comment 271 for bug 85488

It really seems like this should be fixed in libusb rather than the kernel. To fix it in the kernel apparently the approach is to blacklist specific scanners, which obviously leaves newer scanners still broken. Another possible kernel fix is to lengthen the timeout for sleep, but this obviously doesn't help because humans are involved and humans work on their own time. If libusb checked the sleep state before accessing the device and woke it if needed then that would solve the problem for all programs using libusb.

Or perhaps it should be fixed in the kernel. Why does the kernel not automatically wake the device when a program attempts to access it? Or rather, why does it only wake it when the device is accessed in a particular way. If the design was intentional then libusb should be fixed, if the implementation in the kernel is incomplete then the kernel should be fixed.

My vote, if it matters, is libusb.