OpenJDK 8 Segfault on Startup

[David Holmes]

Hi,

On 27/11/2018 12:22 am, Attie Grande wrote:
> Hi All,
> 
> I apologise if this is the wrong list, if it is, please let me know.
> After getting a "waiting approval" / "members-only list" response, I
> had another look at the available lists, but still feel this is the
> most suitable.

hotspot-dev may be better.

> I've been trying to get OpenJDK 8 running on Linux on an ARM platform,
> using the Yocto Sumo recipes from meta-java [1].
> 
> Unfortunately, the result is a java executable that cannot startup
> without hitting a SIGSEGV. Even running "simple" things like `java` or
> `java -help` result in a segmentation fault.

FYI the JVM will trigger SEGV as part of normal startup. The signal 
handler will deal with it. If you run under a debugger you need to step 
over these initial SEGVs.

If you are crashing then it may be the signal handler has not been 
properly setup.

Of course it may be you are hitting the SEGV for a dozen other reasons 
related to the patches applied.

David

> Investigating with GDB shows that the segfault is in libjvm.so, but
> there are no symbols in this region of memory, and the stack has been
> destroyed (I suspect as part of the VM's operation?)
> 
> I've spent some time following the execution through this "fog of
> war", and library functions like malloc(), strlen(), etc... are called
> fairly regularly.
> 
> I'm confident that I've cleared the call to Arguments::parse() in
> Threads::create_vm() [2], but this is getting messy, and surely cannot
> be the correct way to debug JVM.
> 
> One of the last strings I see through strlen() before the segfault is
> "I2C/C2I adapters", which points directly at
> AdapterBlob::AdapterBlob() [3]... I don't know if that will give a
> hint of state / progress to anyone in the know.
> 
> I've attached an error report: hs_err_pid5803.log.
> 
> - How would I go about debugging OpenJDK to figure out what's going on here?
> - I have a series of patches applies (see below), are there any
> obvious missing or harmful patches?
> - Has anyone been down this road before?
> 
> Thanks in advance,
> Attie
> 
> $ quilt series
> patches/remove-shell-variables-from-autoheader.patch
> patches/filter-aclocal-copy-too.patch
> patches/dont-expect-fqpn-for-make.patch
> patches/openjdk8-no-genx11-in-headless.patch
> patches/openjdk8-no-unused-deps.patch
> patches/openjdk8-find-compiler-fix-env-respect.patch
> patches/openjdk8-prevent-debuginfo-in-favour-of-openembedded-package-split.patch
> patches/0001-jdk-comparison-between-pointer-and-integer.patch
> patches/0002-hotspot-fix-compilation-with-security-flags-enabled.patch
> patches/0003-jdk-Allow-using-a-system-installed-libjpeg.patch
> patches/0004-jdk-Allow-using-a-system-installed-libpng.patch
> patches/0005-hotspot-use-correct-include-for-poll.patch
> patches/0006-hotspot-don-t-rely-on-old-SysV-SIGCLD.patch
> patches/0007-jdk-use-correct-include-for-poll.patch
> patches/0008-jdk-use-correct-include-for-signal.patch
> patches/0009-jdk-disable-backtrace-musl-build-fix.patch
> patches/0010-build-fix-build-on-as-needed-toolchains-generic.patch
> patches/openjdk8-restrict-to-staging-dir.patch
> patches/openjdk8-fix-shark-build.patch
> patches/openjdk8-fix-shark-stdc++11.patch
> patches/openjdk8-fix-assembler-flag-handling-in-makefile.patch
> patches/openjdk8-fix-adlc-flags.patch
> patches/openjdk8-add-missing-linker-flags.patch
> patches/openjdk8-fix-libpng-neon-build.patch
> 
> [1]: http://git.yoctoproject.org/cgit/cgit.cgi/meta-java/tree/recipes-core/openjdk?h=sumo
> [2]: https://hg.openjdk.java.net/jdk8u/jdk8u/hotspot/file/d2e3920d2fd0/src/share/vm/runtime/thread.cpp#l3335
> [3]: https://hg.openjdk.java.net/jdk8u/jdk8u/hotspot/file/d2e3920d2fd0/src/share/vm/code/codeBlob.cpp#l270
>