Invocation API and -Xss Issues?

Michaelt-Oracle

Are there any known issues with using -Xss to set thread stack size when embedding the JVM in a native process via JNI_CreateJavaVM? We are seeing SEGV raised by non-Java threads when -Xss is used. The crashes stop when -Xss is removed.

jschellSomeoneStoleMyAlias

One alternative could be that there is a bug in the JNI code.  The use of the option ends up allowing the bug to impact the application space in a way that the OS detects it - and thus the SEGV.

Michaelt-Oracle

While a user level bug is certainly a possibility what's strange about this case is there are no user level threads making JNI calls at the time of the crash. We've also determined the crash occurs on OEL but not on Solaris Sparc, other platforms to be tested soon. The question here is simply if there are any known issues. My queries of bugdb did not yield any.

jschellSomeoneStoleMyAlias

> there are no user level threads making JNI calls at the time of the crash.

Irrelevant.

SEGV occurs because the OS detected that the application did something bad.  That can occur any time once the bug occurs.  It might be immediately or hundreds of lines later.  For example if the code corrupts the native heap then the fault might not occur until the native heap manager attempts to compress the heap which can occur much later.

> We've also determined the crash occurs on OEL but not on Solaris Sparc

Different execution paths which means the potential for different behavior.  So in one case it does something the OS sees in the other it doesn't.