Fedora + Tuned-PPD Performance Regression Should Be Fixed With New Linux Kernel
([Fedora] 108 Minutes Ago
Framework Laptop 13 Pro)
- Reference: 0001654424
- News link: https://www.phoronix.com/news/Fedora-Tuned-PPD-Perf-Fix
- Source link:
Earlier this month when [1]benchmarking six Linux distributions on the new Framework Laptop 13 Pro powered by Intel Core Ultra Series 3 "Panther Lake", the performance was strangely slow on Fedora Workstation 44. When digging further into it, the performance loss was with [2]Fedora 44 using Tuned-PPD by default where as other Linux distributions default to Power Profiles Daemon. Ultimately, after Red Hat engineers got involved, was found to be a bug within the Intel P-State kernel driver exhibited by Tuned-PPD. The latest Linux kernel already has a fix in place and will be back-ported to Fedora 44.
Fedor Vorobev of Red Hat further investigated the issue following the Phoronix coverage of this problem on the Framework Laptop 13. While the performance regression was found when using tuned-ppd and not when using power-profiles-daemon, it was ultimately discovered that the Intel P-State driver was incorrectly handling the case where the governor is set to performance as was done by Tuned-PPD but not PPD itself when setting the "performance" mode.
Intel engineers happened to have already fixed the issue in a July patch with [3]intel_pstate: Adjust policy->cur in active mode to policy . That patch was merged as part of the power management material for the now open Linux 7.3 kernel cycle. It is also being back-ported to stable Linux kernel branches.
Fedora Rawhide thus is not affected by this issue any longer on affected laptops. Fedora will be back-porting the fix as well to the Fedora 44 release per [4]this merge request . The status is tracked via [5]this Red Hat bug report . Or this was the status as of yesterday when being alerted to this being "fixed". But now this morning when getting around to writing this article, there's new activity in the ticket as Vorobev is reporting that he is no longer able to reproduce in testing on a Lenovo ThinkPad T14 Gen5 so it sounds less certain. But hopefully this issue is indeed resolved at least for the Framework Laptop 13 Pro. I will be trying out the Fedora 44 updates on the Framework Laptop 13 Pro Core Ultra X9 388H laptop as time allows. One way or another, it sounds like this issue should be resolved soon.
[1] https://www.phoronix.com/review/framework-laptop-13-pro-linux
[2] https://www.phoronix.com/review/fedora-pantherlake-thermald-tuned
[3] https://lore.kernel.org/all/5144014.31r3eYUQgx@rafael.j.wysocki/
[4] https://gitlab.com/cki-project/kernel-ark/-/merge_requests/4711
[5] https://bugzilla.redhat.com/show_bug.cgi?id=2522083
Fedor Vorobev of Red Hat further investigated the issue following the Phoronix coverage of this problem on the Framework Laptop 13. While the performance regression was found when using tuned-ppd and not when using power-profiles-daemon, it was ultimately discovered that the Intel P-State driver was incorrectly handling the case where the governor is set to performance as was done by Tuned-PPD but not PPD itself when setting the "performance" mode.
Intel engineers happened to have already fixed the issue in a July patch with [3]intel_pstate: Adjust policy->cur in active mode to policy . That patch was merged as part of the power management material for the now open Linux 7.3 kernel cycle. It is also being back-ported to stable Linux kernel branches.
Fedora Rawhide thus is not affected by this issue any longer on affected laptops. Fedora will be back-porting the fix as well to the Fedora 44 release per [4]this merge request . The status is tracked via [5]this Red Hat bug report . Or this was the status as of yesterday when being alerted to this being "fixed". But now this morning when getting around to writing this article, there's new activity in the ticket as Vorobev is reporting that he is no longer able to reproduce in testing on a Lenovo ThinkPad T14 Gen5 so it sounds less certain. But hopefully this issue is indeed resolved at least for the Framework Laptop 13 Pro. I will be trying out the Fedora 44 updates on the Framework Laptop 13 Pro Core Ultra X9 388H laptop as time allows. One way or another, it sounds like this issue should be resolved soon.
[1] https://www.phoronix.com/review/framework-laptop-13-pro-linux
[2] https://www.phoronix.com/review/fedora-pantherlake-thermald-tuned
[3] https://lore.kernel.org/all/5144014.31r3eYUQgx@rafael.j.wysocki/
[4] https://gitlab.com/cki-project/kernel-ark/-/merge_requests/4711
[5] https://bugzilla.redhat.com/show_bug.cgi?id=2522083