The Latest On The Linux 5.9 Kernel Regression Stemming From Page Lock Fairness
On Saturday Linus Torvalds sent out a further explanation of the happenings to the various mailing lists and relevant developers.
The problem ultimately comes down to fairness. Linus explained, "It's the fairness. Fairness is good, but fairness is usually bad for performance even if it does get rid of the worst-case issues. In this case, it's _really_ bad for performance, because that page lock has always been unfair, and we have a lot of patterns that have basically come to (unintentionally) depend on that unfairness."
As of yesterday evening, Linus doesn't have any great solution but polling other developers in weighing between reverting the offending commit, adding some busy-spinning, adding a reader-writer page lock, or working more to de-emphasize the page lock. More details for those interested in that aforelinked mailing list.
At the moment he is pursuing another attempt at a new hybrid model for the page locking.
Once any new patch(es) are available, multiple test systems are ready to go in benchmarking the affected workloads and other test cases to spot any other performance changes. Thanks to Phoronix Premium members and other supporters that make all of the daily Linux benchmarking possible.
To be continued.
If you enjoyed this article consider joining Phoronix Premium to view this site ad-free, multi-page articles on a single page, and other benefits. PayPal or Stripe tips are also graciously accepted. Thanks for your support.