Why printk() is so complicated, and how to fix it (2019)

Short version: because it needs work whether it's a sunny day or the last embers of the burning world are cooling, and every condition in between. It needs to allow formatting of an arbitrary string, while not being able to allocate arbitrary memory when called, or take any locks, for that matter. Oh, and it needs to not slow the system to a crawl. And to be useful, its output has to end up somewhere a person can see it.

It would be a miracle if it wasn't incredibly complicated.

I liked the QNX solution - the kernel never prints anything. It never handles a string. It's just a microkernel; all it does is pass messages, allocate memory, and dispatch the CPU. Drivers are in user space, of course. There's a system management process, "proc", running in user space but with extra privileges, to manage things usually done in kernels. The kernel itself was very rarely modified.

More than that, there's no "console". QNX often ran on industrial hardware that didn't have a display or even a file system. Problems are reported by sending a message to a logging process. That might ship it over a network to somewhere else.

Some of this work has gone in now. printk now has a lockless ringbuffer, and that allows it to work from more contexts without needing special fallbacks.

Did anyone list all its requirements?