Linux 5.15 Hit By Some Early Performance Regressions But Quickly Reverted
The newest performance regressions for Linux came from some new memcg accounting support additions merged last week. Intel's kernel test bot found some sizable performance regressions from the new code.
The memcg accounting for file lock caches was one of those regressions. The memcg accounting overhead turned out to be very great and Linus reverted the change while as "future work" are some ideas for lowering that overhead to be able to re-land the change. That patch was found to cause a ~33% performance hit in a scaling benchmark with Intel's test bot.
Similarly, memcg accounting for pollfd and select bits arrays caused another mighty performance hit from the small fundamental change. That change caused a ~15% performance regression in another scaling test.
Linus had to revert both of those patches on Tuesday due to the performance overhead. Meanwhile, he had another patch that also had to be reverted as it was found to be "complete broken".
At least these issues were caught quickly by Intel's kernel test robot.