Quoting Dmitry Dmitriev:

---

Today I noticed regression in java start-up time on my Linux-x64 system. Especially on fastdebug version. It seems that it happened after sync hs-rt repo with main.
I compare two versions with following JPRT id: 2015-12-14-130534.david.8145301 (push before sync with main) and 2015-12-14-225311.jprovino.hsrt-sync (sync with main).

Here a small results:
1) Bundle from 2015-12-14-130534.david.8145301 job

Running "time ./java -version" give following average time:
java version "9-internal"
Java(TM) SE Runtime Environment (fastdebug build 9-internal+0-2015-12-14-130534.david.8145301)
Java HotSpot(TM) 64-Bit Server VM (fastdebug build 9-internal+0-2015-12-14-130534.david.8145301, mixed mode)

real 0m0.248s
user 0m0.180s
sys 0m0.036s

Running "time ./java" give following average time:
...<some output>...
real 0m0.457s
user 0m0.488s
sys 0m0.104s

2) Bundle from 2015-12-14-225311.jprovino.hsrt-sync job

Running "time ./java -version" give following average time:
java version "9-internal"
Java(TM) SE Runtime Environment (fastdebug build 9-internal+0-2015-12-14-225311.jprovino.hsrt-sync)
Java HotSpot(TM) 64-Bit Server VM (fastdebug build 9-internal+0-2015-12-14-225311.jprovino.hsrt-sync, mixed mode)

real 0m0.734s
user 0m0.732s
sys 0m0.048s

Running "time ./java" give following average time:
real 0m1.572s
user 0m1.980s
sys 0m0.180s

Small update: I checked Windows-x64 system and not found regression in start-up time on it.

---

I reproduced and ruled out the JDK changes integrated, and with the added information about Windows not regressing JDK-8036003 stood out.

Building with --with-native-debug-symbols=none and backing out the changes in JDK-8036003 both get rid of the regression.