We were following up on historical performance for some workloads. Notably, Dacapo/eclipse experienced about 3% regression after JDK-8314307 fix in JDK 21u, at least on AArch64. There is a relevant, but smaller regression in JDK 17u from the variant of the fix.

$ build/linux-aarch64-server-release/images/jdk/bin/java -Xmx4g -Xms4g -XX:+AlwaysPreTouch -jar dacapo-23.11-chopin.jar spring -n 10 -t 1 2>&1 | grep msec

JDK 21, at 10ccbebc8f60553325179d11267370e0028cbfa4
===== DaCapo 23.11-chopin spring PASSED in 24439 msec =====
===== DaCapo 23.11-chopin spring PASSED in 24394 msec =====
===== DaCapo 23.11-chopin spring PASSED in 24715 msec =====

JDK 21, at 10ccbebc8f60553325179d11267370e0028cbfa4~1
===== DaCapo 23.11-chopin spring PASSED in 23622 msec =====
===== DaCapo 23.11-chopin spring PASSED in 23651 msec =====
===== DaCapo 23.11-chopin spring PASSED in 23711 msec =====

JDK 17, at c5723c0948c902491f52a4ec9bb57c3f93552a0c
===== DaCapo 23.11-chopin spring PASSED in 24019 msec =====
===== DaCapo 23.11-chopin spring PASSED in 24100 msec =====
===== DaCapo 23.11-chopin spring PASSED in 24146 msec =====

JDK 17, at c5723c0948c902491f52a4ec9bb57c3f93552a0c~1
===== DaCapo 23.11-chopin spring PASSED in 23700 msec =====
===== DaCapo 23.11-chopin spring PASSED in 23760 msec =====
===== DaCapo 23.11-chopin spring PASSED in 23695 msec =====