FS#69160 - [linux] performance regression with 5.10.3

Attached to Project: Arch Linux
Opened by linearithmik (linearithmik) - Friday, 01 January 2021, 00:07 GMT
Last edited by Andreas Radke (AndyRTR) - Wednesday, 21 April 2021, 11:19 GMT
Task Type Bug Report
Category Kernel
Status Closed
Assigned To No-one
Architecture x86_64
Severity High
Priority Normal
Reported Version
Due in Version Undecided
Due Date Undecided
Percent Complete 100%
Votes 1
Private No

Details

Description: After upgrading from linux "5.9.14.arch1-1" to "5.10.3.arch1-1", the ufw "0.36-5" commands and service startup are about 10x slower.
Note that the file system used for root is ext4.

Additional info:
* The system is installed on a Samsung 970 EVO Plus SSD. If a spinning drive is used, the performance regression might not be observable.
* There is also a one-time graphical glitch introduced with the new kernel when launching alacritty "0.6.0-1". But, it might need its own task.

Steps to reproduce:
1. Install ufw "0.36-5" and enable its service ("ufw.service")
2. Reboot your system, run `systemd-analyze blame | grep ufw.service` and note the observed time
3. Downgrade the kernel to "5.9.14.arch1-1"
4. Repeat step "2." and compare the values

Observed values:
"5.9.14.arch1-1": around 150 ms
"5.10.3.arch1-1": around 1.6 s

Example of ufw commands:
```
sudo ufw --force reset
sudo ufw default deny incoming
sudo ufw default allow outgoing
sudo ufw enable
```

Those commands could be put into a file and timed with `time sh file.sh`
This task depends upon

Closed by  Andreas Radke (AndyRTR)
Wednesday, 21 April 2021, 11:19 GMT
Reason for closing:  Fixed
Comment by Robin Rosenberger (robinrosenberger) - Friday, 01 January 2021, 11:27 GMT
Same here with similar disk, Samsung SSD 840 EVO 120GB, but btrfs rootfs. The boot process is considerably slower with 5.10.3, approx. 2 minutes compared to 20 seconds using 5.9.14
Edit: Also present with 5.10.4 from testing
Edit2: Seems to affect the CPU, not the disk. The same Ardour-Session gives me 25% DSP load on 5.9.14, while it's >90% with 5.10.3/4
CPU is Intel(R) Core(TM) i5-4570
Comment by Robin Rosenberger (robinrosenberger) - Sunday, 03 January 2021, 11:33 GMT
Disabling cpupower-gui.service and cpupower-gui-helper.service seems to have solved it for me
Comment by linearithmik (linearithmik) - Wednesday, 31 March 2021, 21:58 GMT
It looks like `5.11.11.arch1-1` fixed the performance regression. The issue can be closed.

Loading...