Going Magnum with BFS 357

As some of you may be aware, PCLinuxOS uses BFS by default. It also is a 32bit distro for maximum compatibility with desktop software. So, Texstar, who started PCLinuxOS was keen to try the newer BFS and built a 2.6.32.x kernel with BFS 350 on 32 bits. He was able to reproduce strange slowdowns and stalls and reported this back to me. After reviewing my code it was clear I had screwed up with 32 bit builds having moved to 64 bit niffy counters on BFS 350 and left a whole lot of ints and longs around that would overflow regularly. The ints would overflow even on 64 bit builds! Texstar has since confirmed for me that fixing the 32bit variables fixed the problem for him (thanks!).

I've committed those 32 bit fixes, added some more sanity checking on the crazy sched_clock interface using jiffy difference to determine upper bound and added some minor macro cleanups. I've bumped the version number up to 0.357 just because it sounds good. Testing on this has been done on 32 bit, uniprocessor, and an older kernel. Hopefully this means good things for android too!

Changelog follows:

I forgot about an awful lot of longs and ints that will overflow on 32 bit now
with u64 deadlines. Fix them.

Add some macro tidiness.

Make sched_clock sanity checking robust and standardised, using jiffy
difference as upper limit, and use nominal 1us when difference cannot be
trusted.

Go magnum.

I've uploaded a full patch for 2.6.35.7 and an incremental from 350 to 357 and will be uploading patches for older kernels shortly. Grab it now and do your worst!

BFS patches

Hopefully I can take a break from hacking for a bit and get back to my billion other pastimes while this version distils for a while... Then again, a new "stable" kernel will probably be out soon.

UPDATE: I've now uploaded patches for previous kernels as far back as 2.6.31.14.