PowerMac G5 fans going crazy

[Brandon Bergren]

On Wed, Feb 17, 2021, at 9:58 AM, Mark Millard via freebsd-ppc wrote:
> 
> No fixes for the mis-matched times across cpus issues have been
> committed to main or stable/13 . Sleeping threads can still sleep
> way too long.

Correct. And the reason that disabling smp fixes the issue is that with only one cpu running, there's no way for the timebase to get out of sync because there's only one timebase running in that case, so there's nothing for it to get out of sync *with*.

I've been working today on one for my G4, based on the recent Book-E changes so I can experiment with it. There will need to be several different bits written to handle turning the timebase on and off. (It's done in several different ways because it involves turning the timebase signal on and off via platform specific means so that the same value can be written to every cpu without worrying about execution timing differences, and then turning it back on after every cpu has the same value set.)

The difficulty here is that it's a bit messy because it's not always actually represented in the device tree, and a lot of the G5s use platform methods, which we don't have an interpreter for, so we'll have to implement it on a device by device basis in some cases.

If I can get my G4 working (should be relatively easy because it's a matter of twiddling a gpio pin), I'll go ahead and try and implement some of the other bits that are needed.

-- 
  Brandon Bergren
  bdragon at FreeBSD.org