=== halvors1 is now known as halvors
=== mIk3_09 is now known as mIk3_08
[15:14] <Wimpress> I've run into an issue with the schedutil governor on Threadripper using Hirsute.
[15:14] <Wimpress> Schedutil is default for Zen CPUs now.
[15:15] <Wimpress> I run 4x NVME Gen4 AIC in RAID 0. 
[15:16] <Wimpress> The system will hard lock when high I/O
[15:18] <Wimpress> When I backup the Gen4 AIC RAID 0 array to a NVME Gen 3 RAID 0 array, the throughput reach 1 GigaBYTE per second.
[15:19] <Wimpress> If schedutil is the elected governor, the system will hard lock. I can reproduce this 2 out of 3 backups.
[15:19] <Wimpress> I have since made OnDemand the default governor and stability is restored. 
[15:20] <Wimpress> Anyone seen similar reports?
[15:24] <kwilczynski> Wimpress: Hi. I am not sure how you report issues to Ubuntu's Kernel team, but this might also warrant a Kernel Bugzilla report, if you can be asked. :)
[15:31] <Wimpress> kwilczynski: Yes, I think so. I thought I'd go hunting for someone else to confirm what I seeing first :)
[15:31] <kwilczynski> :)
[15:31] <Wimpress> I have a couple of people I can collab with on the testing.
[15:31] <kwilczynski> I wish I could help you.
[15:32] <kwilczynski> But I will be able to afford Threadripper sometime before retirement :)
[15:32] <Wimpress> Who have similar 3970X systems and AICs
[15:32] <kwilczynski> Ask Linus :D
[15:32] <Wimpress> Well, indeed. I see his workstation is a good candidate for testing this :D
[15:33] <kwilczynski> If you open Bugzilla, then posting on LKML might... might... not get you lunched if you include the Bugzilla link.
[15:33] <kwilczynski> lynched*
[15:33] <Wimpress> Thanks for the advice.
[15:33] <Wimpress> Noted.
[15:33] <kwilczynski> In all seriousness. It's an expensive kit, so not a lot of folks would have it.
[15:34] <kwilczynski> Which is why I recommended Bugzilla, as often kernel hackers lurk there and they might have similar machine.
[15:34] <kwilczynski> Plus then it's easier to post to LKML asking.
[15:34] <Wimpress> I know some enthusiasts with appropriate kit and one of them is tweaker who will enjoy digging into this.
[15:34] <kwilczynski> Nice!
[15:34] <kwilczynski> Yeah, if they can help, and have similar kit, then it would be ace.
[15:35] <Wimpress> I have tried to rule out common issues, so disable PBO and XMP. Tried two BIOS revisions.
[15:36] <kwilczynski> One thing to keep in mind.
[15:36] <kwilczynski> Did you try latest stable?
[15:36] <Wimpress> Stable what? Kernel?
[15:36] <kwilczynski> If you have 5.4 or 4.5 or something... the first thing people will do is to ask you to run latest.
[15:36] <kwilczynski> Yeah.
[15:36] <kwilczynski> The is a chance... some chance, that this was an issue that was already addressed
[15:36] <Wimpress> Yes. I was running 20.04 until 2 days ago.
[15:36] <kwilczynski> 20.04 is which one?
[15:37] <kwilczynski> 5.10?
[15:37] <Wimpress> So, was running a 5.4 kernel.
[15:37] <kwilczynski> Ahh, well.
[15:37] <kwilczynski> That's hardly recent.
[15:37] <Wimpress> Which doesn't have schedutil
[15:37] <kwilczynski> If you know how, try building 5.12
[15:37] <Wimpress> I can test that.
[15:38] <kwilczynski> If that also has issues, then also if you know how (easy to learn how) you could try to bisect changes across the kernel releases to see which one might cause it. But this would be a lot of work.
[15:41] <Wimpress> Well, my bisect would be pretty coarse right now, schedutil bad, ondemand good ;)
[15:41] <Wimpress> But I can see if schedutil was reliable at some point in it's introduction.
[15:41] <kwilczynski> :)
[15:42] <kwilczynski> That does help, indeed. :)
[15:42] <kwilczynski> Wimpress: If you spent time doing this, I'd be grateful to you a lot, for helping others, etc. :)
[15:43] <Wimpress> I will test it. I also have some other AMD/Ryzen hardware and Gen4 NVME. So, I'll also see if this is reproducible on not exotic hardware.
[15:46] <kwilczynski> Ace!
[15:46] <kwilczynski> If you need someone running 7+ years old hardware, do reach out :P