Andrew Clunis

So, I've had a new blag on the Internets set up for a while now, running Mephisto:

http://www.orospakr.ca/

The content is mostly nerdy, but hopefully some folks will find it interesting. :)

Test!

I'm trying to get all my Livejournal friends' locations plotted on a map - please add your location starting with this form. Username: (Then get your friends to!)

Which OS are You?


*sigh*

I suppose Elegance concerns me enough when it comes to software engineering... anyway.

I was seriously hoping for Debian GNU/Linux or some other Free system. Stupid test. :(

Which File Extension are You?


ARGH! NONONONONO!

*sniffle* annoying?

I mean, why couldn't I have been something cool, like ELF (well, granted, elf files usually have no file extension...).

heh, I have been looking at the GPL flash player a bit recently. Perhaps this is a sign from the gods that I should start hacking on it.,,

Wow, LitW was great fun!

So, I met tons of awesome people today, including Alan Cox! He's a pretty cool guy. Even complimented him on the -ac kernel morse code flaster. ;)

anyway, so I was asking him and a few other awesome kernel hackers about the stack corruption I've been having with UM Framebuffer, and one fellow asks, "hey, were you guy that posted about that to the fbdev list?"

Turns out it was Geert himself! AWESOME! So I spent a good while chatting about framebuffer stuff with him. Small world. At least when you get a huge portion of the community crammed into one city. ;)

Basically, the two best leads I have are either host userspace memory allocation not going through UML's wrappers properly, or threading conflicts. Loading SDL on from source with -g and just plain ol' reading the code should help, though. Jeff Dike will be at OLS, so I shall pick his brain, too.

Damn I'm looking forward to when OLS starts on Tuesday night. Woo!

PS. Has anyone else read It's Walky in less than 48 hours and still have time to go outside? :D

well, after a little bit of futzing around with my host kernel, I finally booted up with skas3.

And my modified SKAS uml kernel linked against SDL booted! No errors! No -lpthread crashing the whole works. It was awesome. Now that I have that squared away, I can actually get coding without worry. And since it linked OK with -lpthread, that means that I might be able to use the audio stuff in SDL, which may come in handy at some point...

Later!

Ugh, it is 15:00 already, and all I've really done today is fight with CUPS over the phone and stuff... *sigh*

I'm looking forward to getting home and working on uml framebuffer...

Well, well, well.

Turns out SDL wasn't being linked in at all. Apparently the `sdl-config --static-libs` option was being passed at the wrong place in the linker line for linux:. argh! No error or anything. It just didn't show up in the resultant ELF. No symbols, no -lpthread errors, no nothin'! Gah. I was able to fix this though, and was able to get the thing to actually acknowlege the existence of the `sdl-config --static-libs` argument.

My troubles didn't end there, though. Oh no. Of course, unliked dynamically linked stuff, static libs can't bring along their own dependencies. So, for instance, the libSDL that ships with Debian sid is compiled against aalib. Which is cool, but sdl-config --static-libs doesn't mention it. Yay for a crapload of "could not resolve symbol" errors. So, I had to pass `aalibconfig --static-libs` manually. Which means a dep that the end user's libSDL may or may not require is hardwired to the um build system. And the same thing for libgpm.a, which my Debian copy of libSDL requires, is not even included in the Debian gpm packages. Ugly, ugly, ugly.

So, of course, the magic bullet that solves all this crap is: link dynamically.

Unfortunately, tt mode in UML cannot be statically linked. Users will be obliged to use SKAS mode to use the SDL Framebuffer. Period. Either that or some crazy hacker finds out a way of making tt mode work with a dynamic kernel, but I doubt it.

SKAS mode is cleaner, faster, and more 'lifelike' to a real linux box (memory spaces are arranged much more like that of the host) than tt mode UML is. It is completely superior. Only drawback is the host kernel must have a patch which is not yet standard issue.

I set my UM source tree's .config to SKAS mode (and changed the sdl-config line to --libs rather than --static-libs) and it compiled cleanly! no errors! not so much as a warning. I wasn't really able to test my new image since my host doesn't have SKAS support in my host kernel though. The image seems to start OK though, up until the part where it tries to load init (and of course can't because it can't allocate the new address space). If my memory serves me correctly, it got past the part where -lpthread crashed my tt kernels back when I last tried this project. But we shall see when I load a SKAS kernel on my laptop tomorrow night...

Night all!

hm, I'm probably going to see Gothica tomorrow with a friend of mine. I guess I'll see if it is as freaky as the movie posters indicate it is... Although, what's the bet it is just americanised holywood crap? :P