Hi David, thank you for the feedback ! On Wed, Dec 5, 2018 at 9:05 PM David Lechner <david@xxxxxxxxxxxxxx> wrote: > > Does this actually need a new fieldbus subsystem or could it just be > implemented as a new network protocol? > > Then this generic interface to a fieldbus device could just be a socket. > This is a fundamental question, and I'm glad we are having this discussion. Caveat: I am by no means a Fieldbus expert. HMS Industrial Networks are. They have recently become aware of this mainlining effort; but they are not Linux focused. I really hope they will participate. Fieldbus is a collection of network protocols for industrial networks. The underlying network technology varies: e.g. Ethernet for Profinet, RS485 for Profibus, just to name a few. The actual layout of the raw packets exchanged over the network is proprietary; fieldbus cards will typically implement this internally, and hide the details. Some fieldbus protocols have hard real-time requirements: the CPU inside a fieldbus card will typically run an RTOS to satisfy these. Could Profinet be added as a new network protocol? Leaving aside proprietary and hard real-time concerns, the main issue is that cards implementing the standard won't provide access to the raw packets. As a result, the Fieldbus userspace API can only consist of whatever these cards are willing to expose. This varies widely among implementations and fieldbus protocols, but there are a few commonalities. For example, every fieldbus card will expose 'process memory' which you can think of as shared memory between a device and its controller (master). The fieldbus subsystem proposed in this patch exposes a minimal set of functionality that every fieldbus device is supposed to have. In many instances, this won't even be enough to get the card to work. For example, there is currently no way to provide initialization parameters. These vary from setting the card network id, to configuring a built-in HTTP server running inside the card (!!). Linus Walleij suggested that the KVP nature of Netlink sockets might be a good match for fieldbus initialization parameters. It sounds promising. Yet when it comes to accessing shared memory, I cannot see how Netlink could be of use? It would be ideal if actual fieldbus experts could help shape this interface, but I don't know of any that are willing to. One possible explanation is that the fieldbus corporate environment may be a strong believer in proprietary technologies. Sven
