On Thu, Jul 23, 2020 at 03:06:51PM +0000, Yannick Schinko wrote: > Hello, > > so before I get into any detail I'd like to make clear that I am not at all > familiar with the Linux kernel itself. However I'm not an inexperienced > programmer. > Additionally the driver I'm describing here (and that I ultimately plan to > create) is not intended to merged into the kernel itself. It's made for a > specific piece of hardware that will only be produced in miniscule numbers. > In turn that means removing existing modules to replace them with my own is a > possibility.
We take kernel drivers into the tree for only one piece of hardware in the world, and one user. So don't think that we would not take such a thing, if it were actually needed (see below...) > > Now, why I am writing here in the first place: > Currently the company I work for is designing an inline UPS for a very > specific board (the APU2 series from PC engines). On that UPS there's a > microcontroller that communicates over USB with the help of a CH341 chip. > I have already successfully implemented an userland program that communicates > with the microcontroller. This works great! However there's two issues that > led me to the conclusion that I need to implement this as a kernel module: > - Startup and shutdown times/timings: I noticed early on that the program > starts too late on boot and stops too early on shutdown, no matter what I did > (excluding very very dirty and impractical solutions that should never be > used in any kind of production machine) I could not get it to start early > enough and survive the shut down sequence long enough. The only way I can > think of to circumvent this is by turning the program into a kernel module as > that will allow it to start before the OS and to shutdown after the OS, which > would be acceptable. Just fix your system to do this better. You can start your program _REALLY_ early at boot, and just never stop running it until you power down. That control is up to you, you do not have to be a kernel module to do that at all. In fact, you could run your code _before_ kernel modules are loaded, by putting it into the initramfs, if you really need it. This isn't a kernel issue, but a userspace configuration issue, that you have control over fixing. > - Providing sensor data: The microcontroller has voltage sensors and > temperature sensors that I would like to make available to userspace through > the normal hwmon interfaces. According to my research that is only possible > using a kernel module. Funnily enough that part is already implemented and > works provided I feed it mocked data. Why not use a userspace USB driver to talk to the device and do it that way? Why does it have to be a kernel driver? > In consequence I need to create a kernel module (or at least I think > so) that listens to data sent by the microcontroller over USB and then > sends back some other information in response. Userspace USB code works just fine for this too. > As a secondary functionality it would be amazing if I somehow could still > forward that device as a tty (as it currently is) so I can flash firmware > using avrdude. I was thinking that as long as nothing is using that tty in > userspace, the kernel module listens to the device and sends responses but > stops doing that as soon as the tty is in use and also continues listening > again as soon as it's no longer in use. While this would be amazing to have > it's not required for our use case. I don't understand what you mean by this. In short, I think you should be able to do everything you want/need from userspace today, no kernel programming needed. good luck! greg k-h
