Spent a bit of time working on AICA / SPU related things on the DC today. Started out writing a module for the g2 dmac to do spu dma directly from the aica channel, though this still needs much debugging. So far we don't seem to be able to keep consistent data in registers (ie, write-out a p2seg addr and get back the same address in an entirely different segment). other things, always read 0, which in itself isn't a problem, but the lack of the completion interrupt firing certainly creates some issues. the joys of undocumented hardware.. back to the wince dump.
On a similar note, we can use a channel on the sh dmac itself for writing out the buffer, which is what is happening for testing now. unfortunately since we only have 4 channels on the 7750, this isn't an option. at least now I can look at optimizing some of the completion / signalling code in the subsystem so we don't use quite as many cycles (polling for residue sucks). however, the good news is, even when we're constantly polling for residue, cpu usage is still down from the old manual copy / wait on fifo method. Once the remaining performance things are ironed out, there should be no problems dealing with any high-bitrate samples thrown at it. This will be even more fun once zx80user gets his alsa driver finished and supporting all of the aica channels, instead of just the two channels supported by the oss driver.
Merged / cleaned / rewrote random parts of SnapGear's SH-DSP patch from the uClinux tree a few days ago, which turned out to be quite fun. As a result, rewrote most of the cpu init code, which should now be much easier to work with (especially for adding probe hooks / setting cpu flags). So far this is proving to be quite clean.
Also got most of the 8139too hacks cleaned up and merged in both 2.4 and 2.6, which handily knocks off another thing from the HEAD TODO list, and nukes yet more common code cruft from CVS.
More random tree maintenance. Cleaned up random bits of the SMP code, and made the SMP kernel compile again. Though it will likely be awhile before I'll get around to testing this. Being the sole user of the SMP code however, makes this perfectly reasonable.
Quite a bit of this stuff also needs documenting, which I can safely say is one of my least favourite activities (which Documentation/sh seems to reflect). Perhaps its time to take another look at the DocBook stuff, as it would be quite nice to organize most of this into a general sort of sh architecture guide, instead of just random text files. This also incidentally happens to be another point on the TODO list. Now I just need the motivation to write documentation instead of hacking on code. The general tediousness of debugging SPU DMA might spur this on quicker than anticipated.