On 31/12/11 00:22, Linus Torvalds wrote: > It's also possible that we should blame the firmware code, which is > expressly written to encourage these kinds of bugs. It may be that i > tshould be the firmware code that has a "get_firmware()" + > "put_firmware()" model, and it should cache the firmware explicitly if > the config supports suspend, so that a firmware read at resume time > would actually work. The whole "request_firmware()" interface really > is very prone to these kinds of bugs. How about these semantics for get_firmware / put_firmware: get_firmware -> checks for a cached copy of the firmware and returns that if it exists otherwise calls _get_firmware (i.e. _request_firmware). Either way we get a refcount to the firmware blob. This requires adding 3 members to struct firmware: the refcount, the name of the firmware (see @name in request_firmware) and a list head to cache with. We could also add and async version of this to match request_firmware_nowait. put_firmware -> Removes a refcount from the firmware and frees if there are no more users. put_firmware is called when we are sure that we no longer need to reload the firmware for that device - device close time for most drivers. USB is a more difficult case as devices can change ID but the firmware blob still needs to be associated with them. I don't know USB at all so I have no idea if we can track this. Some PCI devices may also suffer the same problem. Hope this is useful, Jack -- To unsubscribe from this list: send the line "unsubscribe linux-usb" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
