[lworld] RFR: 8369134: [lworld] We should preserve the markWord for all GCs

[Coleen Phillimore]

On Tue, 7 Oct 2025 13:02:10 GMT, Paul Hübner <phubner at openjdk.org> wrote:

> Hi all,
> 
> This PR follows up on changes made in [JDK-8368875](https://bugs.openjdk.org/browse/JDK-8368875). I updated the documentation to reflect the new information gained.
> 
> Furthermore, the investigation described in the issue yielded that we _must_ preserve the larval bit explicitly. I introduce a regression test. If we change the mark word preservation condition to `return (!is_unlocked() || !has_no_hash())` (i.e. what we had before), I have confirmed that we will crash a debug VM via failed assertion. Note that ZGC does not require mark word preservation. I've nonetheless decided to include it in case this behaviour is introduced later.
> 
> Testing: tiers 1-3.

The comment looks good.

test/hotspot/jtreg/runtime/valhalla/inlinetypes/LarvalMarkWordTest.java line 91:

> 89:   }
> 90: 
> 91:   private static void stressTheGC() {

There's a WhiteBox FullGC that you can call that might be a reliable way of calling System.gc().

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

PR Review: https://git.openjdk.org/valhalla/pull/1666#pullrequestreview-3351870930
PR Review Comment: https://git.openjdk.org/valhalla/pull/1666#discussion_r2441130180