> I also thought I’d message the vendor and ask them if they could share any specifications or docs regarding their protocol. To my surprise, Nanoleaf tech support responded to me within 4 hours, with a full description of the protocol that’s used both by the Desk Dock as well as their RGB strips.
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 almost had that experience with one of the popular PC liquid cooling hardware vendors around 10 years ago.
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)
Yeah that part of the article put a big smile on my face.
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 :)
It's not the IP, it's sadly how people react. Some folks will be appreciative of help, credit to them. Others will immediately get back how they tried it, it didn't work and now they need you to rewrite everything, or do their project for them, or redesign your product to match what they want it to be. And if you politely refuse, it quickly escalates to threats of trashing your business through every channel, and other things.
So, the safest thing to do is not give details at all, or "leak" them like another reply in this thread mentions.
I wish more vendors would put "Linux support" on the package. Or maybe "Unofficially supported in Linux" or "Linux community support" if they don't want to get their fingers burned.
Maybe this kind of thing should be enforced in the GPL (as many devices use Linux under the hood).
IME, Linux support is often mentioned on the package these days. But you still need to look it up anyway because it's often a vendor source dump for a particular kernel version. Such drivers are often not that good. Notable exception: my Brother printer-scanner-copier has been working fine with its vendor-provided user space driver for ~10 years now. I think it has not been updated for most of that time. It contains 32 bit binaries!
This has to put them in the top 0.01% of companies that make consumer electronics.
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!
It really makes me wonder why it's not more common. They literally don't have to write any firmware, just use one of the thousands already available provided that the license allows for it
I have a solar inveter from a company, aparticular German brand. I wanted to use home assistant with it so I needed rs232 data.. tried the support and they asked me to sign an NDA.
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.
They have asked to the legal team who basically don't know shit about what us do and who will always take the most conservative approach possible. So you'll get either: no answer, a "NO" answer or an "NDA" answer.
Yup. The default from lawyers will be "No" because why help? The correct way to use lawyers, either in-house counsel or a paid lawyer if you have a specific project is to tell them what you want to happen and then they figure out what the best legal course of action is to closest achieve that. Mundanely this might mean e.g "Fill out form X1234 and then send letters to this politician suggesting they support your X1234 application" but at the extreme it could mean closer to the situation in the UK where a ferry company illegally fired crews with zero notice, and their legal advice had clearly been "Worst case they put one of your executives in prison, probably not even that, but either way you replace union workers with people on slave wages and that means $$$".
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.
USB is designed to solve a complex, but necessary problem.
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.
Serial ports still exist today. Even in desktop computers (though usually in weird underpowered ones designed for industrial applications and such). Even the most GAMING of motherboards have headers ready for one or even more serial ports to mess with. All you need is a cable and a card to slot into a PCIe bracket!
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.
> 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.
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.
I sometimes think about this, starting from scratch with a computer hardware and software stack that disallowed all of the layers of abstraction that have built up over the decades.
Yeah many of the abstractions help with performance but maybe there's value giving up much of that performance in exchange for simplicity.
For toy tinkering, to replicate something like the old ability to just write a byte to an IO memory location and have it appear on the pins of a parallel port, you could design a simple USB peripheral that did similar. Give it a set of DIP switches to pick a number (4 for 0-f, more would probably be overkill but feel free) and have the driver present something like /proc/simpleio/device<num>/in and /proc/simpleio/device<num>/out/pin<num> for reading & writing¹. If you use the same set of pins for in and out, /proc/simpleio/device<num>/config to state which direction each is expected to work, and perhaps set voltage options (probably per device, per pin sounds overkill for this). Release the full design and let people make their own, sell pre-made boards and cases and/or let others do that (or just the boards and have people 3D-print their own cases). Devices could have as many pins as the maker likes.
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.
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.
USB parallel ports, data acquisition peripherals, or modern boards that run linux and have GPIO pins are already a thing and don't need to be created.
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.
Nice article here, really enjoys the explanation.
As a web developer, system programming or in this case developing a device driver feels like a different world to me. But, after reading this article, I got the impression that it is not that much different to developing a client to an API in web developer's world.
It’s a userspace USB HID driver in rust, which is honestly more interesting/applicable to me than a kernel driver, which is what I thought it meant from the title.
I really enjoyed the way this post was written, i.e. it includes the code, how it was run, the false paths, etc. You almost get to live through the author's journey and how he figured out just enough to get something working.
Upon reading the title I thought this was going to be about how easy it is to write or modify a Linux driver when using a LLM even if you know nothing about the subject.
No LLM needed. Kernel driver code for simple things is usually copy-paste-modify. Find something that works with an HID interrupt based device, and modify that. If you want a /sys led, copy from that. It's only if you try to push to mainline you need to worry about understanding it, but they'd probably smell LLM trash from miles away.
I enjoyed this post, but I'm eager to hear what the next step would be for a real "production" userspace driver. Are these typically just daemons that are configured to run at start up? And then some configuration GUI communicates with it over a socket or something?
You could certainly do that, and it would make a ton of sense if there's both no standard software API for communicating with that type of device, and it's important that multiple pieces of software that communicate with the device are able to run at the same time (or you want to avoid repeating work when starting software multiple times). ADB (Android Debug Bridge) takes this approach.
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.
Typically production userspace drivers run as daemons (often systemd services) with udev rules to detect device connections, exposing control via D-Bus, sockets, or a custom API that GUI applications consume.
This is just so amazing, you've definitely made me want to try this out some day! Brings me back to engineering school days where we'd try to tweak just about anything with an interface!
I think they wrote the code more for the challenge of it than anything else. The final code doesn't seem to do more than the bare basics (which is probably enough for the author).
I don't think the style described in that doc is at all ugly. It's practical and simple. I say this as someone who spends a lot of time writing modern c++ and heavily using templates.
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.
I want to run FreeBSD on my laptop, but they don't have a [complete] driver for my wifi card. I've thought about diving into AI coding-assistant agents just to see if I could use one to finish throwing together a working driver... but figuring out the AI agents is frictiony enough that I'm leaving it be. (I'm not a VSCode user)
My understanding is that FreeBSD lacks the support for modern WiFi standards in any of its drivers, so it would be a significant effort of understanding the WiFi protocol in addition to understanding your specific device. That said, FreeBSD would be much easier to use on laptops if someone were to put in this effort
This is using rusb, a Rust wrapper around libusb. libusb is still written in plain old C: https://libusb.info/
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.
This is exactly the kind of simple tutorial I was looking for a few years ago. I forget what the project was that I wanted this for, though, haha. Thank you for the writeup!
Excellent post! I have a Nanoleaf 3d thing that I was hoping to use as an ambilight-like thing for the TV, but the experience was... meh. I know about the Hyperion project and hope to set it up one day, but meanwhile I think I could try to repurpose this code to control the strip from a Linux box.
Any programming language can be a solution, the programming language itself isn't the problem. Convincing device driver programmers around the world to come up with a universal API that includes all the possible features and works on every OS is.
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.
On a previous project I spent a lot of time trying to figure out a 'generic device driver'. That is actually a decently hard space to crack. As each device has randomly different ways of talking. Some are serial/local bus. Some are memory space. Some have a combination. Some systems broadcast data, some you have to query it, some you have to query then wait for the next broadcast. Some systems mildly follow the spec, some strictly follow it, some strictly follow their own flavor of the spec or an outdated one, some have their totally made up spec that they may or may not give you. Even from the same company you can have two models that on the outside if you use their stuff looks the same but under the hood is sorta different enough that your driver is garbage. I basically ended up here https://xkcd.com/927/
I agree with you on the Universal API. But It cannot cover any case, unfortunately.
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.
Neither microkernels nor rust will help if you can't get everybody to agree on an ABI, which is the hard part in the first place.
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.
> I also thought I’d message the vendor and ask them if they could share any specifications or docs regarding their protocol. To my surprise, Nanoleaf tech support responded to me within 4 hours, with a full description of the protocol that’s used both by the Desk Dock as well as their RGB strips.
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 almost had that experience with one of the popular PC liquid cooling hardware vendors around 10 years ago.
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)
Yeah that part of the article put a big smile on my face.
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 :)
> don't have a company policy to collaborate with academic research"
Strangely they all have a tacit policy to build their products at least partly on the results of academic research.
It's not the IP, it's sadly how people react. Some folks will be appreciative of help, credit to them. Others will immediately get back how they tried it, it didn't work and now they need you to rewrite everything, or do their project for them, or redesign your product to match what they want it to be. And if you politely refuse, it quickly escalates to threats of trashing your business through every channel, and other things.
So, the safest thing to do is not give details at all, or "leak" them like another reply in this thread mentions.
I wish more vendors would put "Linux support" on the package. Or maybe "Unofficially supported in Linux" or "Linux community support" if they don't want to get their fingers burned.
Maybe this kind of thing should be enforced in the GPL (as many devices use Linux under the hood).
IME, Linux support is often mentioned on the package these days. But you still need to look it up anyway because it's often a vendor source dump for a particular kernel version. Such drivers are often not that good. Notable exception: my Brother printer-scanner-copier has been working fine with its vendor-provided user space driver for ~10 years now. I think it has not been updated for most of that time. It contains 32 bit binaries!
I had the same reaction. Nano leaf is extremely cool for that.
This has to put them in the top 0.01% of companies that make consumer electronics.
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!
It really makes me wonder why it's not more common. They literally don't have to write any firmware, just use one of the thousands already available provided that the license allows for it
Or worse, they just reply “Linux is not supported” …
I have a solar inveter from a company, aparticular German brand. I wanted to use home assistant with it so I needed rs232 data.. tried the support and they asked me to sign an NDA.
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 ?
They have asked to the legal team who basically don't know shit about what us do and who will always take the most conservative approach possible. So you'll get either: no answer, a "NO" answer or an "NDA" answer.
Yup. The default from lawyers will be "No" because why help? The correct way to use lawyers, either in-house counsel or a paid lawyer if you have a specific project is to tell them what you want to happen and then they figure out what the best legal course of action is to closest achieve that. Mundanely this might mean e.g "Fill out form X1234 and then send letters to this politician suggesting they support your X1234 application" but at the extreme it could mean closer to the situation in the UK where a ferry company illegally fired crews with zero notice, and their legal advice had clearly been "Worst case they put one of your executives in prison, probably not even that, but either way you replace union workers with people on slave wages and that means $$$".
Was it Alpha-ESS? they make it so stupidly hard to get your information outside of their smartphone app.
Phocos
Germans germaning!
Even though there are much fewer lawyers in Germany and they are less involved in running companies (like in any country, compared to the US).
This is awesome!
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.
USB is designed to solve a complex, but necessary problem.
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.
Serial ports still exist today. Even in desktop computers (though usually in weird underpowered ones designed for industrial applications and such). Even the most GAMING of motherboards have headers ready for one or even more serial ports to mess with. All you need is a cable and a card to slot into a PCIe bracket!
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.
> 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.
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.
I sometimes think about this, starting from scratch with a computer hardware and software stack that disallowed all of the layers of abstraction that have built up over the decades.
Yeah many of the abstractions help with performance but maybe there's value giving up much of that performance in exchange for simplicity.
For toy tinkering, to replicate something like the old ability to just write a byte to an IO memory location and have it appear on the pins of a parallel port, you could design a simple USB peripheral that did similar. Give it a set of DIP switches to pick a number (4 for 0-f, more would probably be overkill but feel free) and have the driver present something like /proc/simpleio/device<num>/in and /proc/simpleio/device<num>/out/pin<num> for reading & writing¹. If you use the same set of pins for in and out, /proc/simpleio/device<num>/config to state which direction each is expected to work, and perhaps set voltage options (probably per device, per pin sounds overkill for this). Release the full design and let people make their own, sell pre-made boards and cases and/or let others do that (or just the boards and have people 3D-print their own cases). Devices could have as many pins as the maker likes.
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.
Thanks for the info!
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.
USB parallel ports, data acquisition peripherals, or modern boards that run linux and have GPIO pins are already a thing and don't need to be created.
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.
Maybe something like Oberon 2?
It would be neat to design a whole new contemporary stack from scratch.
It would also be an enormous undertaking, requiring man-centuries of work and has the risk of falling into the 'second systems effect' trap...
Nice article here, really enjoys the explanation. As a web developer, system programming or in this case developing a device driver feels like a different world to me. But, after reading this article, I got the impression that it is not that much different to developing a client to an API in web developer's world.
It’s a userspace USB HID driver in rust, which is honestly more interesting/applicable to me than a kernel driver, which is what I thought it meant from the title.
I really enjoyed the way this post was written, i.e. it includes the code, how it was run, the false paths, etc. You almost get to live through the author's journey and how he figured out just enough to get something working.
The title was misleading because the author wrote a userspace Rust driver and not a Linux kernel driver in C.
Upon reading the title I thought this was going to be about how easy it is to write or modify a Linux driver when using a LLM even if you know nothing about the subject.
No LLM needed. Kernel driver code for simple things is usually copy-paste-modify. Find something that works with an HID interrupt based device, and modify that. If you want a /sys led, copy from that. It's only if you try to push to mainline you need to worry about understanding it, but they'd probably smell LLM trash from miles away.
Love it! I wrote an RTL8139 Ethernet driver in Rust as my undergrad thesis!
With questionable grammar: [https://rajiv256.github.io//projects/ouros/](link)
I enjoyed this post, but I'm eager to hear what the next step would be for a real "production" userspace driver. Are these typically just daemons that are configured to run at start up? And then some configuration GUI communicates with it over a socket or something?
You could certainly do that, and it would make a ton of sense if there's both no standard software API for communicating with that type of device, and it's important that multiple pieces of software that communicate with the device are able to run at the same time (or you want to avoid repeating work when starting software multiple times). ADB (Android Debug Bridge) takes this approach.
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.
Typically production userspace drivers run as daemons (often systemd services) with udev rules to detect device connections, exposing control via D-Bus, sockets, or a custom API that GUI applications consume.
Given the title, I'm surprised (in a good way) that it's not about AI coding.
I was kinda hopeful... I mean, letting an LLM write a driver, what could possibly go wrong :)
> Let’s run it again to make sure it was not a fluke!
I understood that reference
This is just so amazing, you've definitely made me want to try this out some day! Brings me back to engineering school days where we'd try to tweak just about anything with an interface!
The author should probably have implemented support in OpenRGB instead to better benefit others, but this is cool nonetheless.
I think they wrote the code more for the challenge of it than anything else. The final code doesn't seem to do more than the bare basics (which is probably enough for the author).
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 don't think the style described in that doc is at all ugly. It's practical and simple. I say this as someone who spends a lot of time writing modern c++ and heavily using templates.
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.
I want to run FreeBSD on my laptop, but they don't have a [complete] driver for my wifi card. I've thought about diving into AI coding-assistant agents just to see if I could use one to finish throwing together a working driver... but figuring out the AI agents is frictiony enough that I'm leaving it be. (I'm not a VSCode user)
My understanding is that FreeBSD lacks the support for modern WiFi standards in any of its drivers, so it would be a significant effort of understanding the WiFi protocol in addition to understanding your specific device. That said, FreeBSD would be much easier to use on laptops if someone were to put in this effort
Claude code, being a CLI interface, might be more your style? Expensive though
AI can also be used to help reverse engineer blobs.
I wish this was done in C so I didn't have to learn Rust. But maybe it is time to learn Rust.
Rust isn't that difficult really. Easier in many ways than C if you're building normal software and not, like, a library of fancy data structures.
This is using rusb, a Rust wrapper around libusb. libusb is still written in plain old C: https://libusb.info/
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.
This is exactly the kind of simple tutorial I was looking for a few years ago. I forget what the project was that I wanted this for, though, haha. Thank you for the writeup!
Thanks for this. Good read and also kinda inspiring.
Excellent post! I have a Nanoleaf 3d thing that I was hoping to use as an ambilight-like thing for the TV, but the experience was... meh. I know about the Hyperion project and hope to set it up one day, but meanwhile I think I could try to repurpose this code to control the strip from a Linux box.
Does this provide protection against opening then device from two processes at the same time?
Does O_EXCL work with linux USB devices?
Recommendations for a tutorial on how to write a formally verified kernel module in C? Rust?
Are VFIO or eBPF sufficient; Does this code need to run in the kernel?
[dead]
There are many innovative OSes that are killed by the lack of Device Drivers.
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.
Any programming language can be a solution, the programming language itself isn't the problem. Convincing device driver programmers around the world to come up with a universal API that includes all the possible features and works on every OS is.
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.
On a previous project I spent a lot of time trying to figure out a 'generic device driver'. That is actually a decently hard space to crack. As each device has randomly different ways of talking. Some are serial/local bus. Some are memory space. Some have a combination. Some systems broadcast data, some you have to query it, some you have to query then wait for the next broadcast. Some systems mildly follow the spec, some strictly follow it, some strictly follow their own flavor of the spec or an outdated one, some have their totally made up spec that they may or may not give you. Even from the same company you can have two models that on the outside if you use their stuff looks the same but under the hood is sorta different enough that your driver is garbage. I basically ended up here https://xkcd.com/927/
The XKCD is wrong in the microkernel concept since the micro-kernel translates the
NxM problem to N+M.
But I agree, that vendors do not make it easy to foster new OSes.
I agree with you on the Universal API. But It cannot cover any case, unfortunately.
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.
Neither microkernels nor rust will help if you can't get everybody to agree on an ABI, which is the hard part in the first place.
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.