RFR: 8221732: Introduce CollectedHeap::hash_oop()

Thomas Schatzl thomas.schatzl at oracle.com
Mon Apr 1 10:36:04 UTC 2019


Hi,

On Mon, 2019-04-01 at 09:39 +0200, Per Liden wrote:
> The JVMTITagMap hashes oops by casts them to 32-bit unsigned
> integers. 
> For ZGC, which has colored oop, this happens to work because all 
> metadata bits are above 32-bit. In theory, ZGC could store its
> metadata bits in the lower 32-bit part of the address, in which case
> this hash scheme would break. Today we have an ugly assert in 
> JvmtiTagHashmap::hash() to make sure this doesn't happen. However,
> the JVMITagMap should not make assumptions on what an oop looks like,
> and should instead delegate this hashing to CollectedHeap.
> 
> Bug: https://bugs.openjdk.java.net/browse/JDK-8221732
> Webrev: http://cr.openjdk.java.net/~pliden/8221732/webrev.0

  looks good.

Thomas





More information about the hotspot-gc-dev mailing list