[lworld] RFR: 8373787: [lworld] InstanceKlass::size mixes bytes and words

[Stefan Karlsson]

On Tue, 16 Dec 2025 12:26:42 GMT, Stefan Karlsson <stefank at openjdk.org> wrote:

> The following function in `InstanceKlass` is supposed to return size in words:
> 
>   static int size(int vtable_length, int itable_length,
>                   int nonstatic_oop_map_size,
>                   bool is_interface,
>                   bool is_inline_type) {
>     return align_metadata_size(header_size() +
>            vtable_length +
>            itable_length +
>            nonstatic_oop_map_size +
>            (is_interface ? (int)sizeof(Klass*)/wordSize : 0) +
>            (is_inline_type ? (int)sizeof(InlineKlassFixedBlock) : 0));
>   }
> 
> but `sizeof(InlineKlassFixedBlock)` returns a size in bytes. This adds 640 bytes instead of 80 bytes to all InlineKlasses.
> 
> This can be seen by running `jcmd <pid> VM.classes` to check the size of the InlineKlass of Integer before and after the fix for this:
> 
> Before:
> 
> 0x00000ffc0024d260 167 fully_initialized WS java.lang.Integer
> 
> After:
> 
> 0x000001800024d260 97 fully_initialized WS java.lang.Integer
> 
> 
> That is in words. So this bug added an extra `(167 - 97) * 8 == 560` bytes. 
> 
> I've so far only done some local smoke testing with TEST=hotspot_valhalla

Thanks for all the reviews! This passes tier1-3 and TEST=hotspot_valhalla and TEST=jdk_valhalla locally. I'm marking this for integration.

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

PR Comment: https://git.openjdk.org/valhalla/pull/1804#issuecomment-3661537255