[Bug 3533] [IcedTea8] HotSpot generates code with unaligned stack, crashes on SSE operations

[bugzilla-daemon at icedtea.classpath.org]

https://icedtea.classpath.org/bugzilla/show_bug.cgi?id=3533

--- Comment #27 from Maciej S. Szmigiero <mail at maciej.szmigiero.name> ---
(In reply to Andrew Haley from comment #26)
> (In reply to Maciej S. Szmigiero from comment #25)
> > As far as I can remember, significant additional amount of work is needed
> > in addition to the mentioned patch to fully check and fix the code generator
> > (that's why using a compiler flag instead as a workaround is so attractive).
> 
> Would that work, though?  If the call is to an external library, we can't
> guarantee that library has been specially compiled.
> I guess we know that all of the calls are to libc or to libjvm, so perhaps
> that's good enough.

As long as the Java VM is compiled with "-mincoming-stack-boundary=2" calling
C/C++ functions in the JVM from HotSpot-generated code should work fine (and it
seems this is indeed the case) because every JVM function will then (re)align
stack at its entry.
This means that any function further in the call chain (in the JVM or in an
external library) will see the stack being properly aligned.
This covers most of the cases of misaligned stack.

The only remaining problem could be the case that the generated code calls
directly a third party library with no JVM in between them (JNI seems to
operate like this).
I haven't hit such problem in practice but a static audit of the relevant
portion
of the code generator is probably necessary to be sure.

-- 
You are receiving this mail because:
You are on the CC list for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.openjdk.java.net/pipermail/distro-pkg-dev/attachments/20180523/db170ab0/attachment.html>