RFR: 8362657: Make tables used in AOT assembly phase GC-safe
Ashutosh Mehra
asmehra at openjdk.org
Mon Jul 21 19:13:51 UTC 2025
On Fri, 18 Jul 2025 20:46:17 GMT, Ioi Lam <iklam at openjdk.org> wrote:
> This fixes crashes when using `-Xlog:aot+map=trace,aot+map+oops=trace:file=aot.oops.txt:none:filesize=0` during the assembly phase.
>
> I basically changed a few raw `oop` pointers to `OopHandle`. I also simplified `ArchiveHeapWriter::is_marked_as_native_pointer()`.
>
> Note that for `HeapShared::archived_object_cache()`, I calculate the hashcode using the raw oop. This way, we can avoid calling `oopDesc::identity_hash()` for every archived oops. As a result, when we use this table after exiting the CDS safepoint, we must rehash it. See changes in `ArchiveBuilder::CDSMapLogger::log()`.
>
> - It might be alright to pre-calculate the identity hash of the archived oops in the assembly phase, but I think we should do that only if we find a reason to do so. We shouldn't do it just because it's convenient for implementing an internal table,
Marked as reviewed by asmehra (Committer).
@iklam there are arm compile failures likely due to some missing NOT_CDS_ macros. Otherwise this looks good to me.
-------------
PR Review: https://git.openjdk.org/leyden/pull/86#pullrequestreview-3039364698
PR Comment: https://git.openjdk.org/leyden/pull/86#issuecomment-3098042847
More information about the leyden-dev
mailing list