About time for an update...

Yeah, so I'm lazy and haven't updated this thing in quite a while now... The good news is I've finally arrived in Helsinki and got settled into the hotel etc. I'm going apartment hunting tomorrow, and I think I've found a nice looking place. :-)

On one of the flights I actually got asked by one of the flight attendants "Is that a pirated movie?" (In my defense, it could have been a really crappy rip of a DVD I just happen to own... ;-) I'm pretty sure he was joking about calling customs, though.  At least I didn't get stopped at that airport...

I did get stopped at London for an explosives test. I think that was just because I got a bit lost and ended up being last through the security checkpoint. Maybe they thought I ran off to plant bombs or something. :-P

Oh, yeah. It's a good idea to remember to install the wireless firmware before you go to another country and track through a snow storm to find a USB memory stick so you can download them from the business center. ;-P Actually the computers in the business center wouldn't load the USB stick, so I decided to get a taxi to a net café this time... After all that the hotel's DHCP server was down, so fail all around... At lest I'm online now. :)

I'm waiting for OProfile to download (at an amazing 5kB/s) so I can figure out why X is eating up 40-70% CPU?

I managed to get fglrx 8.11 downloaded and installed yesterday, so I might start playing around with tracing a userspace process MMIO access from within the kernel. Revenge works pretty well, but there are cases where it can miss the packets from the first few tests. I think this is because the RB pointers get updated too fast for Revenge to see them.

I had a conversation with Dave Airlie about this, and tried the idea of intercepting fglrx ioctls. This does work, but I'm not convinced as to it's reliability either. The ioctls contain more than just pointers to packers, and I think this extra information may be used by the DRM to insert extra packets into the RB (eg flush etc.) The ioctls may also change between fglrx versions.

The advantage to kernel tracing is that it should be able to reliably trace everything the driver does, although there are a few kinks to work out... The PPPT stuff used on newer hardware might be a problem. I'll see how it goes.

Oh, yeah, someday I'll post the rest of the China Trip...