On 9/10/2021 3:58 pm, Aleksey Shipilev wrote:
Hi,
This does not look a build issue per se, but rather a Hotspot bug.
The SIGILL in ThreadGroup.add looks to me like an issue with atomic
operations as that is the first synchronized method that gets executed
during VM initialization.
David
-----
On 10/8/21 7:46 PM, John Cummings wrote:
I am using the source from https://hg.openjdk.java.net/jdk/jdk11.
This is an extremely old JDK 11 tree. The most actual one is here:
https://github.com/openjdk/jdk11u-dev/
FWIW, ARM32 builds from that repo work for me on RPi 4:
https://builds.shipilev.net/openjdk-jdk11-dev/
I am at the point where I am unsure where to proceed or what to
change, and haven't been able to
find a solution anywhere thus far to build a client version of
OpenJDK-11. If there's any
additional information I can/should provide, please let me know.
Additional things to try:
*) Build fastdebug, trying to see if any VM assert is triggered; pass
--with-debug-level=fastdebug to configure
*) Build/test the JDK mainline (https://github.com/openjdk/jdk/),
trying to see if this is a fixed issue that is missing a 11u backport
*) Trim down --with-extra-cflags and --with-extra-cxxflags, looking
for some special flag that triggers the bug
There were plenty of ARM32 bugs fixed (backported) in 11u, but I also
see some that are only fixed in later JDKs, for example:
https://bugs.openjdk.java.net/browse/JDK-8222825
