FS#79177 - Kernel 6.4.4 has regression with high iowait on one core

Attached to Project: Arch Linux
Opened by Mike Cloaked (mcloaked) - Sunday, 23 July 2023, 13:51 GMT
Last edited by Toolybird (Toolybird) - Monday, 24 July 2023, 01:15 GMT
Task Type Bug Report
Category Packages: Core
Status Closed
Assigned To No-one
Architecture All
Severity High
Priority Normal
Reported Version
Due in Version Undecided
Due Date Undecided
Percent Complete 100%
Votes 3
Private No


Kernel 6.4.4 has a regression causing high iowait percentage on one core.

Additional info:
* package version(s) 6.4.4 and 6.4.5
* config and/or log files etc.
* link to upstream bug report, if any

See below

Steps to reproduce:

update to 6.4.4 kernel - if running mysql/mariadb, then one core has 95% iowait percentage. This has been bisected and the problematic commit identified. Kernel 6.4.4 with the problematic commit reverted no longer has this issue. Either downgrade back to 6.4.3 or build your own kernel with the regression commit reverted.

See https://bugzilla.kernel.org/show_bug.cgi?id=217700 and https://bugzilla.kernel.org/show_bug.cgi?id=217699 and there is a discussion also on LKML about this.

Kernel 6.4.5 is now released with the problematic commit still not reverted, so perhaps the arch kernel 6.4.5 when building should have the commit concerned also reverted, until the issue is fixed upstream.
This task depends upon

Closed by  Toolybird (Toolybird)
Monday, 24 July 2023, 01:15 GMT
Reason for closing:  Not a bug
Additional comments about closing:  See comments
Comment by Oleksandr Natalenko (post-factum) - Sunday, 23 July 2023, 13:58 GMT Comment by Mike Cloaked (mcloaked) - Sunday, 23 July 2023, 14:08 GMT
Kernel 6.4.5 built with commit bd4f737b145d85c7183ec879ce46b57ce64362e1 reverted works without any iowait problems.
Comment by Mike Cloaked (mcloaked) - Sunday, 23 July 2023, 19:56 GMT
The LKML thread now suggests that the 100% display on cpu monitors is false, and needs to be fixed. The kernel is doing the correct thing, and giving improved performance, even if the iowait is high (correctly) in the mariadbd case. Therefore the commit that was suggested should be reverted should remain, and userland code which displays high utilisation needs to be fixed instead where it reports high when in fact the core is idle, but iowait is high.
Comment by Toolybird (Toolybird) - Monday, 24 July 2023, 01:14 GMT
Thanks for the report. But as per comments above, it turns out this isn't a kernel bug.

Please report upstream bugs against any affected userland tools that need fixing.