I sometimes use PSI to diagnose bottlenecks in my system. Now I notice it's not enabled, and oddly CONFIG_PSI does not appear in the kernel config at all, not even disabled! It's still in the 6.6 kernel according to https://cateee.net/lkddb/web-lkddb/PSI.html

Can anybody tell me where it has disappeared to in the land of Liquorix?
Back to top

PSI support was removed due to multiple reports of instability with it enabled: gitlab.com/alfredchen/linux-prjc/-/commit/ceb80e952a426e02d456f5f03b5171649c4e424f

I don't expect for it to make a return unless someone volunteers time to reintroduce it properly.
Back to top

Makes sense! Thanks for the quick response!
Back to top

Would it be possible to enable both




In Liquorix kernels then? Having PSI enabled is very useful for early OOM and setting the second option means it becomes a boot-time parameter, rather than always being enabled.[/code]
Back to top

Did you read the thread before posting?
Back to top

Yes, you said it been removed because enabling it causes stability issues in some cases.

That’s why I'm asking if it possible to add it back default disabled, so that users can test themselves whether its worth it to them.

And no need to be rude, btw…
Back to top

PSI cannot be turned on because the code in Project-C to support it was entirely removed. PSI also cannot be shimmed for the CPU scheduler, and even if someone figured out how, it would confuse all the software depending on PSI to provide it the right information to make decisions.

Maybe it was not obvious from my post, but this isn't something you "just" turn on. Someone needs to sit down and write / test code and verify that reintroduction of PSI support works (it already didn't, which is not a good starting point). That's asking a lot for a developer that isn't getting paid for this work.

Also, restoring the code that was removed and disabling PSI support by default will still create a support cost, as a number of users may unknowingly enable PSI through a kernel boot parameter or the like, only to find out their system is unstable. This is already something Liquorix works around for CPU governors and IO schedulers, since distributions like to set bad defaults that cripple Liquorix's out-of-box experience or stability. This has been discovered over the years whenever a new Phoronix benchmark comes out, only to find a bad result is caused by Ubuntu or whichever distribution forcing a setting that makes the kernel misbehave in drastic ways.

This is why I asked you if you read the thread, these points should be clear. If you had made a new thread however, I would be more understanding as the history on PSI support in Project-C is scattered between Gitlab/GitHub issues and forum posts.
Back to top

I see, I had been under the impression it had only been disabled using kernel configuration, rather than entirely removed. That’s too bad, but understandable.

When you asked about reading the “whole thread” I had only assumed this thread, not the linked things, sorry. :'-D
Back to top

PSI sounds like a good concept but the execution may seem to be a bit flawed so maybe it gets revisited sometime.
Back to top