On Mon, 30 Mar 2026 20:10:14 GMT, Chris Plummer <[email protected]> wrote:

> First it helpful to know that for hprof files, every java object is given an 
> ObjectID, and that id is simply the address of the java object in the java 
> heap. SA does the following to write a field that is reference type into the 
> hprof file:
> 
> 
> OopHandle handle = ((OopField)field).getValueAsOopHandle(oop);
> writeObjectID(getAddressValue(handle));
> 
> 
> The only validation it does is to make sure the address of the field is 
> valid. Note I'm referring to the address of the field itself, not the address 
> (oop) stored in the field. If the field is flattened, it may no longer have 
> an 8-byte alignment requirement. But getAddressValue(handle) does. So if the 
> field is not 8-byte aligned we get the UnalignedAddressException and you see 
> in the stack trace that is happening in all of the heap dumping test. If it 
> is 8-byte aligned, then the contents are just blindly loaded and treated as 
> an oop, which then gets written as an (invalid) ObjectID into the hprof file. 
> It then turns up as the hprof verification error that you see in 
> [JDK-8377387](https://bugs.openjdk.org/browse/JDK-8377387):
> 
> WARNING: Failed to resolve object id 0x10000 for field isRegularFile 
> (signature L)
> 
> The fix is to not write the (invalid) ObjectID if the field has been 
> flattened, and instead just write the ObjectID for null as if the field 
> actually contained a null. This gets us around the issues that cause test 
> failures, but eventually we need to properly handle the flattened field and 
> write its scalar contents to the hprof file. I'll file a separate CR for that.
> 
> Tested using --enable-preview with all tier1, tier2 svc, and tier5 svc.

This pull request has now been integrated.

Changeset: f31be3fb
Author:    Chris Plummer <[email protected]>
URL:       
https://git.openjdk.org/valhalla/commit/f31be3fba4feecfbc5d44fb1c9a1f0885b7f8e8a
Stats:     14 lines in 4 files changed: 5 ins; 8 del; 1 mod

8379925: [lworld] Several SA HeapDump tests fail with `Unrecognized heap dump 
sub-record type:  XX` with --enable-preview
8377387: [lworld] serviceability/sa/TestHeapDumpForInvokeDynamic.java test 
fails with --enable-preview

Reviewed-by: sspitsyn

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

PR: https://git.openjdk.org/valhalla/pull/2280

Reply via email to