On Tue, 21 Apr 2026 18:40:06 GMT, Aleksey Shipilev <[email protected]> wrote:
>>> OK, I think I know what this is. The crash is in C2 AOT method at inlined >>> collection-set-check. Shenandoah expands that check in C2 IR and the >>> testing the native bitmap, which AFAICS got relocated. I guess we need to >>> `AOTRuntimeConstants` those to make them work. I am whipping up the patch >>> now... >> >> This should be enough to get AArch64 parts going: >> [shenandoah-aot-2.patch](https://github.com/user-attachments/files/26833509/shenandoah-aot-2.patch). >> There is still the same trouble in C1/x86 parts, I am going to implement >> those once I am closer to x86 machines next week. > >> There is still the same trouble in C1/x86 parts, I am going to implement >> those once I am closer to x86 machines next week. > > Here it is: > [shenandoah-aot-3.patch](https://github.com/user-attachments/files/26943718/shenandoah-aot-3.patch) > > Passes `CONF=linux-x86_64-server-fastdebug make images test TEST=runtime/cds/ > TEST_VM_OPTS=-XX:+UseShenandoahGC`, which is a very good sign. > > @vnkozlov, do you want me to collect these fixes and PR them to > `leyden/premain`, or do you have some other sync-up scheme in mind? @shipilev thank you for the patch. > @vnkozlov, do you want me to collect these fixes and PR them to > leyden/premain, or do you have some other sync-up scheme in mind? The best scenario would be if your changes can be applied to mainline. I tried and all 3 patches applied without issues. You can file JBS issue and push it there. AOT stubs caching is in mainline. Then I will merge them into this PR. Currently I collect all fixes here if I can't push them into mainline. And also push some changes here which later will be pushed into mainline. This PR changes diverged from leyden repo very significantly already. Currently I don't want spend time on Leyden repo merges. ------------- PR Comment: https://git.openjdk.org/jdk/pull/30778#issuecomment-4292173050
