2009

Thu, Jun 11

<21:11 EEST> I think I've resolved a keyboard problem that's been bugging me for a couple of weeks. It is a rather convoluted issue between hardware, kernel and userspace drivers, and the workaround is far from obvious.

The problem is an erratic behaviour of Sigmatrix's keyboard in X, especially under heavy load such as parallel emerge. Every few keypresses are dropped, and some even changed in ordering. When typing fast, the text clearly lags behind. Typing slowly, punctuating each key s e p a r a t e l y, gives a flawless result, but it is incredibly frustrating for a fast typist. Previously, I have attempted to fix similar but lesser bugs with a kernel parameter such as i8042.nopnp. However, this time the effect is restricted to X, and I have switched to virtual consoles for most typing. More interestingly, the problem is less apparent with a USB keyboard.

The solution, at least for now, is to change the Xorg keyboard driver from the old kbd to the more general evdev. This involves some configuration in the form of the /dev/input/ device path. Evdev should also work for mice, but since I didn't get it working immediately, I keep using the old mouse driver.

Apparently, evdev is a part of the new hotplug system that, for example, enables launching X without a xorg.conf. However, in my case I am left without working input devices, so clearly there is something wrong. (This lack of autodetection is probably why I had to configure the keyboard device explicitly.)

Interestingly enough, playing with userspace drivers has helped resolve, what is clearly a hardware issue. Sigmatrix is the only machine where I have encountered this.

Fri, Jun 5

<18:00 EEST> In an unexpected twist of goodness, a new Linux/PPC client for Distributed.net was released this week. I had been waiting for this for about a year, ever since I got my first PPC machine. The former client was from 2004 and it was unable to run any current project besides RC5-72.

Another problem with the old client was the lack of new, faster code, which affected the older projects as well. This was particularly notable on Altivec machines. It turned out possible to run the Linux/PS3 client on G4/G5, since the Cell is a PPC at heart. It is actually 64-bit capable, but Distributed.net had chosen to make this a 32-bit binary, fortunately for us with a G4.

Nevertheless, using the Cell client has its share of problems. More importantly though, I (along with many others, I believe) was left wondering why they had released the new, OGR-NG capable client for so many other PPC platforms, but not for any Free OS. The OS-independent part had been there for a long time, and there was even the Linux/PS3 client, with the Cell instruction set being a superset of PPC. I was actually expecting a more general client for PPC, with or without the SPUs, since the number and type of crunchers are variable anyway.

Then again, the main problem with running the PS3 client on a G4 was that, the results were counted as PS3. So I actually ran the old client for some time, trying to give a message that the Linux/PPC platform is alive and well. I guess it may have worked after all, and there is probably no government conspiracy against OGR-NG on a Free OS. Though the changelog does point to bug #4187 which is closed from the public... :-j