[SOLVED] Current64 - Fri Apr 20 updates

burdi01 · 04-20-2018, 04:46 AM

Current64 - the Fir Apr 20 updates contain:

Code:

k/kernel-source-4.14.35-noarch-1.txz: Upgraded.
       RCU_EXPERT n -> y
       +RCU_FANOUT 32
       +RCU_FANOUT_LEAF 16
       +RCU_FAST_NO_HZ y
       +RCU_NOCB_CPU y
       Thanks to alienBOB.

What is the purpose of these changes? I could not find an LQ thread discussing things.
Please enlighten me.

Darth Vader · 04-20-2018, 05:22 AM

Quote:

Originally Posted by burdi01

What is the purpose of these changes? I could not find an LQ thread discussing things.
Please enlighten me.

Did say something for you the word "Ryzen" ? That's why.

Alien Bob · 04-20-2018, 06:55 AM

I applied those to my own custom kernel on my Ryzen powered build server and that put an end to the many kernel oopses (needs a "rcu_nocbs=0-<maxprox-1>" parameter on the kernel's boot commandline too).
Then I asked Pat to add it to the Slackware kernel config so that I can revert to stock kernel.

Darth Vader · 04-20-2018, 07:08 AM

Quote:

Originally Posted by Alien Bob

I applied those to my own custom kernel on my Ryzen powered build server and that put an end to the many kernel oopses (needs a "rcu_nocbs=0-<maxprox-1>" parameter on the kernel's boot commandline too).
Then I asked Pat to add it to the Slackware kernel config so that I can revert to stock kernel.

I bet that that changes aren't on your benefits only.

There are other Ryzen owners around. Not including me, as I am more than happy with my Bulldozers.

willysr · 04-20-2018, 12:13 PM

me included

aaazen · 04-20-2018, 01:41 PM

It looks like there is an ongoing discussion about Ryzen lockups and kernel parameters etc. in the kernel bug list:

https://bugzilla.kernel.org/show_bug.cgi?id=196683

bassmadrigal · 04-21-2018, 10:35 AM

Quote:

Originally Posted by burdi01

I could not find an LQ thread discussing things.

They've already covered the purpose behind the changes, but to cover where in LQ it was discussed...

Here's the thread where a user was running into similar lockups on Ryzen that I had experienced, so I pointed him to the changes I had made. I then mentioned that I had to make these changes in the thread about whether people use the stock kernels, stating this was the only change I had to make. It turns out Eric was running into the same lockups I had been, but I didn't know about it (otherwise I would've suggested it to him long ago -- I've been using it since at least Nov).

I had only thought that mfoley and I were the only ones affected by this issue on Slackware and neither of us were running -current.

burdi01 · 04-22-2018, 03:24 AM

@bassmadrigal:

Quote:

... so I pointed him to the changes I had made.

Thks. I searched the forum for "RCU_" and did not find that thread therefore. The ryzen angle first came later.

burdi01 · 04-22-2018, 04:58 AM

My kernels are of the distribution type - thus that they run on a variety of Intel (from Celeron upwards) and AMD (A and upwards) CPUs. At the moment they are at 4.16.3. Would these RCU settings applicable here too?

bassmadrigal · 04-23-2018, 06:54 AM

Quote:

Originally Posted by burdi01

My kernels are of the distribution type - thus that they run on a variety of Intel (from Celeron upwards) and AMD (A and upwards) CPUs. At the moment they are at 4.16.3. Would these RCU settings applicable here too?

As far as I know, this only takes effect if you pass the "rcu_nocbs=0-<maxprox-1>" option at boot. But I haven't seen anything from various bug reports stating this Ryzen issue has been fixed on newer kernel releases yet, so for us Ryzen owners, we'll probably need to continue doing this to prevent these lockups.

But I'm not familiar enough with what this setting does and why it prevents the Ryzen lockups to give you any definitive answers.

Petri Kaukasoina · 04-23-2018, 09:23 AM

The bugzilla thread (link in post #6) gives the impression that it's not a kernel problem or kernel config problem but something to be fixed with a BIOS update. Setting a new option "Power Supply Idle Control" to "Typical Current Idle" should fix it. The CPU will dissipate a few more watts when idle which makes it stable.

It seems that the kernel RCU settings just have a side effect which makes some systems stable, maybe because they make the CPU draw more power.

bassmadrigal · 04-26-2018, 11:09 PM

Quote:

Originally Posted by Petri Kaukasoina

The bugzilla thread (link in post #6) gives the impression that it's not a kernel problem or kernel config problem but something to be fixed with a BIOS update. Setting a new option "Power Supply Idle Control" to "Typical Current Idle" should fix it. The CPU will dissipate a few more watts when idle which makes it stable.

Thanks for the info! I never had the heart to fully read through the bug report since a lot of it went over my head. Once we get unpacked (in the middle of a move across the country right now), I'll have to see if there's an update for my motherboard that adds this option. If it's there, I can enable it and try to remove the kernel option on boot.