RFR: 8280087: G1: Handle out-of-mark stack situations during reference processing more gracefully [v5]

[Albert Mingkun Yang]

On Thu, 7 Dec 2023 20:03:36 GMT, Ivan Walulya <iwalulya at openjdk.org> wrote:

>> Hi all,
>> 
>> Please review this change to modify how we grow the global marking stack used by Concurrent Marking. This patch allows for growing the marking stack without the need to copy over elements. Consequently, we can grow the marking stack even during the reference processing phase without the need to restart the marking cycle.
>> 
>> This mainly addresses the issue where object marking work created during reference processing may overflow the global marking stack. Currently G1 just bails out with a fatal error  as expanding the marking stack would require a restart which is not valid during the reference processing phase.
>> 
>> We have decided to maintain the restart concurrent marking when the global mark stack overflows during the marking phase of the concurrent cycle as this offers better memory utilisation.
>> 
>> Testing: Tier 1-5.
>
> Ivan Walulya has updated the pull request incrementally with one additional commit since the last revision:
> 
>   Albert Review

src/hotspot/share/gc/g1/g1ConcurrentMark.hpp line 162:

> 160:     //        |    |------->|     |     |     |     |     |     |     |    |
> 161:     //        +----+        +-----+-----+-----+-----+-----+-----+-----+----+
> 162:     //

Thank you for the diagram. It's much easier to understand now. However, it's unclear why it's important that "... doubles the capacity of  the growable array". What happens if all buckets have the same capacity?

Maybe I am misunderstanding the text/diagram, but I thought the last sentence should be sth like "each new bucket doubles the capacity of this chunk-allocator".

src/hotspot/share/gc/g1/g1ConcurrentMark.hpp line 235:

> 233:   // Allocate a new chunk from the reserved memory, using the high water mark. Returns
> 234:   // null if out of memory.
> 235:   TaskQueueEntryChunk* allocate_new_chunk();

Seems unused now.

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

PR Review Comment: https://git.openjdk.org/jdk/pull/16979#discussion_r1419664750
PR Review Comment: https://git.openjdk.org/jdk/pull/16979#discussion_r1419659722