On Thu, Jan 26, 2023 at 12:49:32PM +0100, Troels Liebe Bentsen wrote: > Hi, > > We have a hardware projekt where something is off with power ON > timing. It sometimes gets started in a broken state where the device > is seen by the USB system but does not respond to descriptor reads. Ah, a broken USB device, not much the kernel can do about that :( > When this happens this causes lsusb and libusb based tools to hang > until the descriptor read in the USB subsystem timeout out after 30 > seconds or so. It looks like the tools are trying to read > /sys/bus/usb/devices/.../descriptors and it blocks until the timeout > happens. > > We should fix our hardware and have done so in the next revision but > why should one device be able to block the descriptors file that most > user land USB code seem to use. If the device does not respond, what is the kernel or userspace supposed to do? > Would there be any reasoning against just serving the descriptors file > as it looked before inserting the broken USB device instead of > blocking the read? So a different device's descriptor file is being blocked by a broken device? Are you sure? They should all be independent. > And if we wanted to create a pull request for a change like that would > it be accepted or would it be considered breaking the API? It depends on what the kernel change looks like. Have you tried anything that worked for you that you wish to propose? thanks, greg k-h
