RFR: 8259021: SharedSecrets should avoid double racy reads from non-volatile fields [v2]
Hans Boehm
hboehm at google.com
Mon Jan 4 17:50:44 UTC 2021
On Mon, Jan 4, 2021 at 8:34 AM Peter Levart <plevart at openjdk.java.net>
wrote:
>
> On Mon, 4 Jan 2021 15:57:33 GMT, Richard Reingruber <rrich at openjdk.org>
wrote:
>
> >> The bug title and the PR title need to be the same.
> >> Editing either one is fine.
> >
> > But wouldn't it be legal for a compiler (java to bytecode or bytecode to
> > machinecode) to replace references of my_local_copy with references to
> > static_field?
> >
> > Foo my_local_copy = static_field;
> > if (my_copy == null) {
> > initialize();
> > my_local_copy = static_field;
> > }
> > return my_local_copy;
> >
> > Only if static_field was volatile this would be illegal, wouldn't it?
>
> @reinrich I don't think Java compilers may do that. If this was allowed,
such variables would not be called "local".
>
Indeed. Such transformations are allowed in C and C++ (since data races
result in undefined behavior, and thus the
compiler is allowed to assume there are no concurrent writes), but not in
Java.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://mail.openjdk.org/pipermail/security-dev/attachments/20210104/42dbb5b2/attachment.htm>
More information about the security-dev
mailing list