RFR: 8245234: Still seeing missing mixed stack traces, even after JDK-8234624
Kevin Walls
kevinw at openjdk.org
Mon Oct 6 10:53:57 UTC 2025
On Sun, 5 Oct 2025 13:09:18 GMT, Yasumasa Suenaga <ysuenaga at openjdk.org> wrote:
> We haven't yet been able to unwind native frames in mixed mode jhsdb in some case even though we implement DWARF parser in [JDK-8234624](https://bugs.openjdk.org/browse/JDK-8234624). The cause is that `DwarfParser` would be reused.
>
> DWARF is encoded as a state machine, so we have to initialize when we try to find new frame via DWARF. So I made change to create new `DwarfParser` instance everytime in `LinuxAMD64CFrame::sender`.
I think this looks good, makes sense, the native DwarfParser has its own context. So the existing reuse of dwarf as nextDwarf at old line 150 might be unwise!
I see that getNextCFA, called down at new line 168, handles nextDwarf being null.
-------------
Marked as reviewed by kevinw (Reviewer).
PR Review: https://git.openjdk.org/jdk/pull/27636#pullrequestreview-3303845340
More information about the serviceability-dev
mailing list