Re: Futex problems with firebird3.0 and openjdk-8 on m68k
On 06/04/2017 09:59 AM, Finn Thain wrote:
>> http://users.physik.fu-berlin.de/~glaubitz/openjdk-8.tgz
>>
>
> Thanks. I used that tgz to generate the patch below, and then patched a
> fresh source tree.
Great!
> The jdk portion of the build did complete. But for many hours, there was a
> java process blocked in futex_wait_queue_me (according to the WCHAN column
> in top). It was this command:
>
> Compiling 9463 files for BUILD_JDK
> (/usr/lib/jvm/java-8-openjdk-m68k/bin/java -Xms64M -XX:ThreadStackSize=768 "-Xbootclasspath/p:/home/buildd/openjdk-8-8u131-b11/build/langtools/dist/bootstrap/lib/javac.jar" -cp /home/buildd/openjdk-8-8u131-b11/build/langtools/dist/bootstrap/lib/javac.jar com.sun.tools.javac.Main -bootclasspath /home/buildd/openjdk-8-8u131-b11/build/jdk/classes -source 8 -target 8 -encoding ascii -XDignore.symbol.file=true -Xlint:-unchecked,-deprecation,-overrides,auxiliaryclass,classfile,dep-ann,divzero,empty,try,varargs -Werror -implicit:none -sourcepath "/home/buildd/openjdk-8-8u131-b11/src/jdk/src/share/classes:/home/buildd/openjdk-8-8u131-b11/src/jdk/src/solaris/classes:/home/buildd/openjdk-8-8u131-b11/build/jdk/gensrc:/home/buildd/openjdk-8-8u131-b11/build/jdk/gensrc_no_srczip" -d /home/buildd/openjdk-8-8u131-b11/build/jdk/classes -h /home/buildd/openjdk-8-8u131-b11/build/jdk/gensrc_headers.BUILD_JDK.tmp @/home/buildd/openjdk-8-8u131-b11/build/jdk/classes/_the.BUILD_JDK_batch.tmp && /bin/mv /home/buildd/openjdk-8-8u131-b11/build/jdk/classes/_the.BUILD_JDK_batch.tmp /home/buildd/openjdk-8-8u131-b11/build/jdk/classes/_the.BUILD_JDK_batch)
>
> It eventually completed without error.
In this case, it's still executing the host JVM as you can see from the path
of the java binary.
> Following the jdk build, when "checking if we are using ecj as javac"
> during icedtea-sound/configure, the javac -version command also blocked in
> futex_wait_queue_me. This command was blocked for five hours until I
> finally killed it.
>
> I also ran openjdk-8-8u131-b11/build/images/j2sdk-image/bin/javac -help
> and that hung in futex_wait_queue_me too. It seems there's something weird
> going on here but there looks to be a consistent way to reproduce it.
And here it is actually the freshly compiled JDK as you can see from the
path above. So, it the bug must be either in the JDK itself or it's gcc
miscompiling the code.
More specifically: The problematic code is inside libjvm.so in:
/usr/lib/jvm/java-8-openjdk-m68k/jre/lib/m68k/server/libjvm.so
Replacing that libjvm.so with the one from Andreas' build fixes the
issue for me. In fact, that's how I built the current openjdk-8
package that's currently in the archives.
Andreas says that he doesn't have this issue and openjdk-8 works
fine for him, which is proven by the fact that his copy of libjvm.so
works. I checked and compared both his patch and the patch in Debian's
openjdk-8 package and they are identical as far as I can tell:
> https://people.debian.org/~glaubitz/java-1_8_0-openjdk-1.8.0.77-11.1.src.rpm
> http://bazaar.launchpad.net/~openjdk/openjdk/openjdk8/view/head:/debian/patches/m68k-support.diff
> We can no longer blame the kernel or the emulator. Next I'll try an older
> glibc.
Thanks.
> FYI, I've uploaded the log of the build up to the point that I killed it.
> http://www.telegraphics.com.au/~fthain/openjdk-8-8u131-b11-build-log
Yes, it's the exact same issue I also ran into.
On a side note, the issue with firebird3.0 seems to be a local issue. The
last build with a freshly created chroot now succeeded without any issues [1],
so please strike firebird3.0 from the original list.
Adrian
> [1] https://buildd.debian.org/status/fetch.php?pkg=firebird3.0&arch=m68k&ver=3.0.1.32609.ds4-14&stamp=1496366219&raw=0
--
.''`. John Paul Adrian Glaubitz
: :' : Debian Developer - glaubitz@debian.org
`. `' Freie Universitaet Berlin - glaubitz@physik.fu-berlin.de
`- GPG: 62FF 8A75 84E0 2956 9546 0006 7426 3B37 F5B5 F913
Reply to: