On Thu, 19 Sep 2024 14:52:07 GMT, Martin Doerr <mdo...@openjdk.org> wrote:
> The SA model of the UpcallStub doesn't work reliably on PPC64. We can use the > Backchain which works for all Power Architecture 64-Bit ELF ABI compliant > frames. (The backtrace should stop when hitting a non-compliant frame which > can occur with some compiler options. Also see comment in JBS issue.) > Note that this enables walking the native frames which is possible on PPC64, > but not on some other platforms due to missing backlinks. > > "test/hotspot/jtreg/serviceability" tests have passed with this PR. > I'm not sure whether this PR is correct - if the app has several call stacks > which do not compliant the ABI, this PR would not work. See [#20789 > (comment)](https://github.com/openjdk/jdk/pull/20789#issuecomment-2324964613) > for details. > > It is better to use JavaFrameAnchor like other platforms if we can. We don't support frames which don't follow the ABI. I think stopping the backtrace immediately is an appropriate reaction. Other debuggers like GDB would do the same. Nevertheless, I couldn't spot any bug in your new code. So, I've debugged the issue and found [JDK-8340657](https://bugs.openjdk.org/browse/JDK-8340657) which causes the problem. ------------- PR Comment: https://git.openjdk.org/jdk/pull/21092#issuecomment-2368595882
