RFR: 8297313: Refactor APIs for calculating address of CDS archive heap regions

[Calvin Cheung]

On Mon, 21 Nov 2022 06:22:49 GMT, Ioi Lam <iklam at openjdk.org> wrote:

> In anticipation of the following RFEs:
> 
> - [JDK-8296263](https://bugs.openjdk.org/browse/JDK-8296263): Uniform APIs for using archived heap regions
> - [JDK-8296344](https://bugs.openjdk.org/browse/JDK-8296344): Remove dependency on G1 for writing the CDS archive heap
> 
> We need to clean up the APIs for calculating the addresses of CDS archive heap regions.  The current APIs have confusing names and the implementation is convoluted.
> 
> The proposal is to change the above to three APIs with easy-to-understand semantics:
> 
> // The actual address of this region during dump time.
> address heap_region_dumptime_address(FileMapRegion* r)
> 
> // The address where this region can be mapped into the runtime heap without
> // patching any of the pointers that are embedded in this region.
> address heap_region_requested_address(FileMapRegion* r)
> 
> // The address where this shared heap region is actually mapped at runtime.
> address heap_region_mapped_address(FileMapRegion* r)
> 
> 
> The meaning of the `CDSFileMapRegion::_mapping_offset` field is changed slightly to simplify the implementation of the above 3 functions. Comments are improved explain what's going on.
> 
> With the above changes, `FileMapInfo::map_heap_regions_impl()` can also be simplified significantly.
> 
> ****
> Testing: tiers 1-4
> ****
> Misc changes:
> - added more assertions about exactly which combinations of collector and oop encoding mode are supported
>   - Specifically, uncompressed oops for non-G1 collectors are currently not supported;
>   - Change error message accordingly when the user selects an unsupported combination.

Looks good. One nit below.

src/hotspot/share/cds/archiveHeapLoader.hpp line 144:

> 142:   static bool      _mapped_heap_relocation_initialized;
> 143: 
> 144:   static void init_narrow_oop_decoding(address base, int shift) NOT_CDS_JAVA_HEAP_RETURN;

This is already inside the `#if INCLUDE_CDS_JAVA_HEAP` block. So the `NOT_CDS_JAVA_HEAP_RETURN` seems redundant.

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

Marked as reviewed by ccheung (Reviewer).

PR: https://git.openjdk.org/jdk/pull/11257