RFR: 8256843: [PPC64] runtime/logging/RedefineClasses.java fails with assert: registers not saved on stack

[Richard Reingruber]

On Mon, 14 Dec 2020 13:43:28 GMT, Martin Doerr <mdoerr at openjdk.org> wrote:

> I was not aware of that __builtin_frame_address is available on all PPC64 platforms. Makes sense to use it and to get rid of the inline assembly.

Me neither. I found it in the implementation for x86_64 and learned that xlC supports it as well.
> 
> > Currently I don't understand why the result of __builtin_frame_address(0) has to be dereferenced?
> 
> I don't understand that, either. Seems like it was implemented by trial and error. I think trying to predict the C++ compiler's inlining is a terrible design, but it's only used by non-critical code like NMT stack traces, so we can focus on more critical issues.
> 
> We could use __builtin_frame_address(1) and remove the dereferencing. I think we should use that also in os::current_stack_pointer(). What do you think?

According to the documentation (https://gcc.gnu.org/onlinedocs/gcc/Return-Address.html) this is not really safe.

Calling this function with a nonzero argument can have unpredictable effects, including crashing the calling program

So I wouldn't do it.

We could still use it in os::current_stack_pointer() _and_ in os::current_frame().

> 
> Anyway, I like your proposal. Would you mind creating a PR for jdk16 as we need it there? I'm closing this one.

That's fine for me.

Thanks, Richard.

-------------

PR: https://git.openjdk.java.net/jdk/pull/1724