Writing a basic Linux device driver when you know nothing about Linux drivers
crescentro.se
crescentro.se
How cool is that? Too many vendors still think that they have valuable intellectual property in such relative trivialities. And that handing out the specs freely helps their competitors more than themselves.
I can think of only a few companies that bother to publish any details... And most of them are focused on industrial customers where it isn't unreasonable to need certain protocol details for integration or even just compliance with certain regulatory systems.
Maybe things are changing?
I have noticed that some of the LED light controllers you see on AliExpress are leaning in to open firmware standards. 5 years ago, you bought the controller and had to flash your own firmware. Now, there's an option at checkout to select an open source firmware. Some even have a USB port built in for flashing!
I emailed them saying I'd be interested in developing drivers for their hardware for Linux as I was a happy customer and was immediately put in touch with one of the managers and their engineering team.
Made quite a bit of progress before the whole thing was shut down because one of their component vendors threatened them saying it'd be a breach of their contract with them.
Apparently that vendor sold a "datacenter" (non consumer) version of that hardware for which they charged a hefty license fee for the management software (which was Linux compatible).
Jokes on them, someone reverse engineered the whole thing with a USB analyzer years later and published it XD. (not me)
I did the same thing back in college, when I was in a lab. We wanted to do some research on Wi-Fi signals, and I happened to own a bunch of Wi-Fi adaptors produced by SomeSmallTech Co. Ltd., which featured relatively new Atheros chips and didn't have Linux drivers at the time.
So I sent an email to the company's public email address, asking for some datasheets, "for science". To my disappointment, presumably a PR person replied that they "don't have a company policy to collaborate with academic research". (But they did send a quick reply, kudos to that.)
Funnily enough, years later I ended up working for said company. Naturally, when I first logged into the company network, I searched for the datasheets I asked for. There were "classified" watermarks all over the PDFs :)
Strangely they all have a tacit policy to build their products at least partly on the results of academic research.
Okay, cool. I did with a fake name, address and everything and they sent a file..
Turns out the file is available online.
Facepalm pro Max.
So my question is, what kind of "IP" is in a data sheet that needs protection ? And this isnt even some secret product but a generic solar product sold by millions.
Rs-232 protocol ? Really ?
So, the safest thing to do is not give details at all, or "leak" them like another reply in this thread mentions.
Maybe this kind of thing should be enforced in the GPL (as many devices use Linux under the hood).
But it also makes me a little bit sad. The original parallel port and even ISA interface seemed so simple by comparison, with less layers of abstraction. Just run a wire, and write to a port.
I remember when I was a kid, I found a breakout board in an electronics store's random clearance parts bin, with an ISA header on an edge. On a whim I took it home and wire-wrapped a 7-segment LED onto it. Power and ground were easy. Each segment was hooked to a data line, through a simple buffer IC. I cheated and used only a minimal number of address lines to feed the enable port (guessing through a simple AND gate or something). I was amazed when I wrote to that address and it worked the first time!
I look at a protocol like USB, with hundreds of pages, and instead of that curious excitement and enablement I felt back then, I feel a bit overwhelmed.
Yeah many of the abstractions help with performance but maybe there's value giving up much of that performance in exchange for simplicity.
That would only cover the real basics: reading/writing the hi/low status of pins. Other things people might want is analogue voltage control/read (for many sensors) or pulse control (for controlling servos and such). Things like that could be mapped into the /proc/simpleio/device<num>/{in|out}/pin<num> files. Perhaps for setting/reading multiple pins at a time perhaps have something like /proc/simpleio/device<num>/{in|out}/allpins. You could expand the feature in many ways, though TBH beyond the simple hi/low thing people are probably better off getting an rPi or microcontroller and using all the available devices and plans there are out there already for their IO pins and using something newer.
--------
[1] Not sure what you'd do for Windows, I'd be inclined to release a Linux driver and let the Windows community worry about one for their platform.
What I'm talking about is a whole computer system + systems programming language + OS which is as simple to understand and program as computers were in the 80s. With modern speedups, without modern layers of abstraction.
I actually bought a real parallel port card years ago that plugs into a PCI port on my motherboard. Incidentally it also provides 4 serial ports.
Only tried it on Windows, but it works great. If I recall there was some kind of library I have to use in modern Windows versions due to a permissions or kernel isolation thing.
All those layers of abstraction is likely what allows us to hook up a single wire to our laptops and get multiple very fast ports from the docking station along with power and display output.
You get some, you lose some.
1. Hot-plug.
2. High speeds with long cables of dubious quality.
3. Multiplexing multiple devices on a single wire with hubs.
4. Reliable transmission on lower layer, so higher level protocols don't need to worry about it.
5. Multiple speeds with negotiation.
6. Newer USB standards support multiple power voltages with negotiation.
All that said, old USB protocols like USB 1.1 is not that hard. You don't need those hundreds of pages, only a subset of them. There are some tutorials in the Internets which will help you to understand everything, from wire signalling to application interface. Don't use USB reference as a learning source. These days ChatGPT probably will guide you over every layer. Just stick with old standards, they are simpler and plenty of devices use them.
With enough persistence and some fast enough MCU you should be able to bit-bang USB 1.1 LS (1.5 Mbps) and write some simple USB device. That will require to implement all layers of USB and I'm pretty sure it's not impossible task.
USB itself, at least the 1.1 protocol that's still used for some devices today, isn't all that complicated in itself: https://youtu.be/wdgULBpRoXk
Many if the complicated parts here, like device identifiers, are the things you'd need to manually configure if you were to use a serial port. Those hundreds of pages are similar to the hundreds of device types, vendors, and models that Windows would list when you clicked "add new hardware" back in Windows 98. When push comes to shove, you're just sending energy pulses down wires, there's just a chip in there that helps collect pulses so you don't need to manage the timing manually.
If there is an applicable standard software API (either multiplatform like a filesystem, or a special one exposed by the OS kernel [1] ), the driver probably belongs in the Linux kernel (or in the form of a Windows driver on that platform). My understanding is that GPU APIs are an exception on Linux, and are implemented in userspace by a piece of software called MESA. You could also use the daemon approach in this case if you don't want to bother with getting a driver added to the kernel.
For a more niche device where exclusive access is acceptable and every piece of software would need to add special support for this specific type of device anyway, it's a lot simpler to distribute the driver as a library that software authors can include in their program. If there are several devices that work similarly but communicate differently, you could have one library that either includes multiple drivers, or exposes a common interface that other libraries can implement.
A downside of any approach for USB devices on Linux that isn't a kernel driver is that one or more udev rules will need to be added (as the article described). This also applies when using a device that uses a supported USB protocol, but has different IDs than the ones listed in the kernel driver.
[1] More devices fall into this category than you might expect. For example, Linux has an API for communicating with CAN devices called SocketCAN, so if you're writing a driver for a CAN device that connects via USB and exposes the full CAN bus over USB (maybe something that goes in or connects to a car), you should write a kernel driver that converts data between SocketCAN and whatever USB protocol is being used (assuming one doesn't already exist, a lot of USB CAN devices use protocols that already have drivers in the kernel). SocketCAN only exposes the raw data extracted from the CAN frames, so if you want to expose an easy way to control a particular CAN device, that belongs in a userspace library that uses the SocketCAN API under the hood.
I understood that reference
With questionable grammar: [https://rajiv256.github.io//projects/ouros/](link)
Plus, to port this to OpenRGB, you'd need to rewrite the code into C++ (ugly, old C-inspired C++, at that: https://github.com/CalcProgrammer1/OpenRGB/blob/master/CONTR...) which would take most of the joy out of it for me at least.
I thoroughly enjoy rust, but I doubt not being able to use it should be grounds for avoiding contributing to a project. Unless you are going to write async heavy code, libusb is pretty easy to use in C.
If you don't want to learn a different programming language, you can take the exact same approach in any language you prefer and play along. You may need to turn the more object oriented calls into libusb_* calls, but if you're used to programming in C you probably won't have a problem getting that to work.
Are VFIO or eBPF sufficient; Does this code need to run in the kernel?
As a community we must find a way for tackling this issue.
Micro-Kernels are a solution where one can run different OSes but they will reuse the same device driver servers.
But it requires co-ordination and determination.
Rust can be a solution for sure.
In this case, the cross-platform libusb should make this code work on either Linux or Windows (if you install the signed Windows drivers). If other operating systems port libusb, they get this code for free.
Most "real" drivers still run in kernel mode, though, and not even Linux can keep their ABI stable (Windows has to, between releases, with the aid of compatibility wrappers that only work for a certain amount of releases).
It would probably be worth it more forbBespoke operating systems to implement either the Windows API (like ReactOS does) or the Linux API (pick an LTS version) to get existing drivers to work. Unless you pay them, most driver programmers aren't going to bother with anything than Windows, maybe Linux, possibly macOS.
NxM problem to N+M.
But I agree, that vendors do not make it easy to foster new OSes.
I set it up so the application layer could just say 'I want to change/get some datapoint'. But the underside of that the 'datapoint' object had to know quite a bit about what to feed into the system to bind up things correctly to do that data transfer. Last time I looked home assistant had something similar to what I was coming up with. I just could not get it to not have a leaky abstraction. The problem was one system might have a set of 16 bit address spaces. While the next one would be 8 bit, the next one would be 78 bit. Then properly decoding those things was also leaky. As many of these systems will do things like bit masking/shifting to pull data out partiular fields. Plus some systems are just simple memory read/writes and others are bind out an RS232 port and fill out this packet just right (dont forget the correct baud rates), oh and there are 6 other things packed into the return packet. The real painful ones were the vendors who got tricky with a DRM handshake before each call and a simple decode system.
But the micro-kernel can help here by translating the NxM problem to N+M by reusing device servers, which can have a universal API and can be written in any native or not programming language.
Now, you can virtualize your OS over the microkernel and re-use the servers. That was the initial dream.
That said, if you're willing to deal with other people's APIs, NetBSD's rump kernels are good for providing reusable drivers, and some projects have even had luck pulling drivers out of Linux to reuse, though obviously that's a little bit touchier.