Fri 02 Jun 2006 11:52:57 PM PDT
Steve Hastings has written a detailed report.
I like the unconference system, although with a room full of Linux and hardware people we didn't get the kind of clamor to put topics on the board that you might get in a room full of, say, bloggers. It's good that we had a selection of topics on the Wiki first.
I put a question up there to kick off the discussion, about the right names to give to files in sysfs, and FreedomHEC got a great talk from Greg K-H about how to organize the files that a driver creates in sysfs. One value per file, please. If you want extra information for debugging, use /sys/kernel/debug, where you can put whatever you want.
The good news is that sysfs actually looks much more complicated from the user side than from the kernel side. Yes, go browse /sys and you'll see a whole bunch of symlinks, but the C functions for creating the sysfs files from inside the kernel are pretty straightforward.
Randy Dunlap, in a talk on "Linux Kernel Social Engineering," quoted Oregon's "Things are different here" slogan. "We don't really want to see just ideas...You have an idea you need to back it up with some code. Sooner rather than later."
While Randy got some good laughs mentioning companies that consider "50 page Word documents" part of the development process, the really good point about the "social engineering" discussion was how many ways there are to get involved in kernel development if that's what's what you want to do. Check out The Linux Kernel Janitor Project for some cleanup projects that will give you a tour of the kernel and help you find an area of interest.
James Bottomley, SCSI subsystem maintainer, is also looking for developers. Any volunteers willing to reproduce and fix the bugs in the drivers for older, less commonly used SCSI host adapters?
Greg also gave a good example of why stable driver ABIs are nonsense: MSFT and Linux have both rewritten USB support 3 times. Linux hasn't kept the two old ones so can be lighter weight and more reliable.
Yes, I worked through Greg's driver tutorial with Steve and the rest of the audience. Tips to prepare if you're coming to the next one at OLS, LinuxWorld, or next FreedomHEC: make sure you can build the kernel that's on your machine, and make sure that CONFIG_MOD_UNLOAD is on. (You'll be insmod-ing and rmmod-ing the module a bunch of times as you read, build, and test different versions.)
I have a bunch of great notes on what to do and not to do in driver code. Definitely a worthwhile tutorial; I would have gotten a lot out of it if I had a half-done driver to ask questions about, too.
Jamey Sharp is making me think that our long nightmare of butthead 3D support in Free software is coming to a close. He's running r300, and says it's "in good shape". Better yet, he adds, "Intel is fully supporting X.org", and Greg predicts "The next-generation Intel stuff will be equal to Nvidia stuff." Yes, FreedomHEC sort of turned into a "3D on X and Linux BOF" for a little while, kicked off by Jamey's X update.
Is this the conference I would go to if I hadn't been involved in it? Hell yes. We even got some Open Source Marketing advice from James. "Windows vs. Linux The Great Battle" is very much a Marketing Myth. "The only people who believe this are people who are susceptible to believing marketing stuff...Companies trust suppliers who respect their choices."
Off to put together the Linux DDK mailing to the attendees. Must find CD labels. Peace, and don't create USB packets on the stack.