RFR: 8370248: AOTMapLogger should check if pointer is in AOTMetaspace

[Vladimir Kozlov]

On Tue, 21 Oct 2025 15:37:06 GMT, Ioi Lam <iklam at openjdk.org> wrote:

> This fixes a recent bug in `AOTMapLogger` that's exposed in macOS 26 testing.
> 
> When running the JVM with a valid AOT cache (or AOT cache) with the flag `-Xlog:aot+map=debug`, we use `AOTMapLogger::RuntimeGatherArchivedMetaspaceObjs` to discover all metaspace objects that are contained in the AOT cache.
> 
> However, this code is executed after some of the AOT-cached classes are linked. As a result, `RuntimeGatherArchivedMetaspaceObjs::do_unique_ref()` could see instances of `AdapterHandlerEntry` which are dynamically allocated and their addresses are outside of the AOT cache.
> 
> In all platforms except macOS 26, the `AdapterHandlerEntry` pointers are all above the address range of the AOT cache, and they are filtered out by subsequent code in `AOTMapLogger`. However, on macOS 26, the `AdapterHandlerEntry` pointers are below the AOT cache, and they caused this assert to fail:
> 
> https://github.com/openjdk/jdk/blob/9a88d7f468cdd040bdf4e1ff9441dc9c66eab03e/src/hotspot/share/cds/aotMapLogger.cpp#L240-L242
> 
> The fix is simple: filter out all pointers that are outside of the address range of the AOT cache in `do_unique_ref()`.

Good.

-------------

Marked as reviewed by kvn (Reviewer).

PR Review: https://git.openjdk.org/jdk/pull/27918#pullrequestreview-3361548446