Wednesday 13 May 2009

Signals and such

After a couple of late nights I've hacked up a few more bits and pieces for my 'os'.

Signals were pretty easy to get going. These are a bit like posix signals, but also not at all the same. A low-level asynchronous bit-level synchronisation mechanism basically. You can allocate a signal 'bit', wait on it (task goes to sleep) or send it to another process (thus waking it up, and perhaps pre-emptying yourself if it has a higher priority). It took a bit longer than I'd hopped because I aso 'cleaned up' the code - started creating a better directory structure and moving things out of the single c file I had. I rewrote the interrupt handler and system call assembly too - they're pretty trivial so it wasn't much.

I also did a lot of reading about `OS4', the latest AmigaOS. Some interesting ideas there, particularly the extension of the basic object oriented nature of the system further. Hmm, food for thought.

And investigated possible ways to get video output in the future. Sigh. PC video hardware sucks. It's not something I want to get bogged down in, but I might have to if it comes to that. I guess I can get some drivers from some generic library/X/or linux even, if they can be extracted or wrapped somehow anyway.

Being annoyed at the PC hardware I was looking at getting a beagleboard to play with - ARM looks like a lot more fun. But just as I was about to hit the buy button I had second thoughts. The display hardware is completely proprietary and undocumented and I can't really see the point if I can't get that to work. A real pity, as otherwise it looks like nice hardware that would be interesting to work with. Maybe somewhere down the line.

Then the next night I started looking at how to implement message ports and message passing. On AmigaOS these are really just arbitrated queues associated with a (possibly shared) signal bit. Very simple, very fast. Question is how to add memory protection in a way which maintains the simplicity. And since I hadn't done any paging stuff yet I got side-tracked learning how that works.

For message passing w/ memory protection in OS4 they've implemented MEMF_PUBLIC, so it actually means something, but I presume it also means all public memory is public to all processes, which is probably not what I want. That allows all processes to access the shared resources like message ports using the same virtual memory address. I thought instead perhaps I could just allocate the memory in a fixed 'public/shared' virtual address space, and migrate the memory around to the current owner using the paging system. But after reading various comments from kernel developers about the performance, I'm not sure it's a good idea. Then again, maybe it's just the cost of using an MMU and can't be avoided. And then I got so side-tracked with how the VM system should work (do i leave it in kernel space, or implement it as a service, etc) I feared I was getting too bogged down and had to put it aside for later.

And even as far as message ports go I probably need to move to a 'handle' model, rather than passing around pointers. So that meant an allocator that could initialise all the user fields, and OS4 has a generic tag (varagsish) based allocator for those so I had to look into tag utilities and such before even having a chance to go further.

And today i've got a little cold. Hrmph.

No comments: