Arch Linux

Ranguvar

Member

Registered: 2008-08-12

Posts: 2,549

Superhigh load bug - very strange

Hey all! It's been a long time since I frequented this forum!

I recently got back into Arch and it's a blast.
I was sad to see the System Administration board is down, which is where I'd have posted this, so apologies if it isn't appropriate here.

I have a very strange bug on my Thinkpad T470 laptop. i5-7300U, 16GiB of RAM, Samsung 850 EVO 500GiB with lzo BTRFS.

For the past month or two, whenever I've run intensive tasks - compiling, virtual machines - after some time, my system will completely freeze. Compiles completely freeze.
It will have been at 100% CPU load for some time, but suddenly, CPU usage will drop to 20% or lower, while system load will soar - past 8, past 10, I've seen 100+, 150+.
RAM usage is well within limits, I can't tell what is going on.

The screen may only update once every few _minutes_ in some cases.
In better cases, it will sluggishly respond for some time - X windows will literally disappear or glitch out, switching to a virtual console will take forever, and login on that VC even longer.

When this happens, only SysRq REISUB fixes the problem - sometimes the SIGTERM or SIGKILL fixes it (until I run intensive processes again).
Sometimes even this just leaves me with a black screen, and the full reboot is needed.

I've tried with Linux 4.12 and 4.13, -ck and regular, with stock configs. No crazy modules installed.
X seems not to be at fault - I repreduced this by only logging in on a console and starting a Chrome compile.

The issue happens on battery power with that tlp profile, and on AC with a max performance tlp profile, no battery saving at all.

At this point, I'm about to try an LTS kernel, disabling BTRFS compression, or a fresh install, but this install isn't even that old, and it has no real background services or bloat to speak of.

Any help diagnosing the massive load and low CPU usage / freezing would be much appreciated!
I found vmstat, but I don't think I see anything helpful - system CPU usage is even lower than the suddenly low user CPU usage, while load is insanely high.
I'm going to try for systemctl and dmesg logs, I think I remember them having nothing useful, but I'll confirm.

Last edited by Ranguvar (2017-10-18 22:34:55)

V1del

Forum Moderator

Registered: 2012-10-16

Posts: 21,728

Re: Superhigh load bug - very strange

Post your dmesg and systemctl logs, just because you don't see anything off, people might pick up on something. First thing that comes to mind reading this. Are your microcode updates in place and correctly applied? There is an issue with Kabylakes and HT which can make it go haywire similar to ways you describe, that has been fixed in a microcode update.

Ranguvar

Member

Registered: 2008-08-12

Posts: 2,549

Re: Superhigh load bug - very strange

Yeah, I have microcode.

I found the issue. Sometimes it takes posting about it to really think it through.

Appears to be a crash in btrfs. I'll update this solved after I figure out a solution and report it to the devs.