RFR (XXS): 8179244: Assert failed in instanceMirrorKlass.inline.hpp
erik.helin at oracle.com
Wed Nov 22 15:49:02 UTC 2017
On 11/20/2017 12:56 PM, Thomas Schatzl wrote:
> Hi all,
> can I have reviews for this small comment update that clarifies that
> the assert in that location is too strong and should be commented out.
> The issue is that an assert fails during CMS precleaning if class
> loading occurred where the mirror ended up in the old generation (in
> some situations CMS allocates directly into old gen). This assert found
> that a klass that has had a NULL value suddenly changed its value to
> something else.
> This is normal behavior in the situation described above.
> Also checked that we do not miss iterating over a CLD for such oops -
> there is actually explicit support in class loading for this, where
> such CLDs (and their mirrors) are kept alive until remark and iterated
> over again there. (Look for "CMS support" in classLoaderData.?pp).
Looks good, Reviewed, I think the comment is on the right level.
Although the comment and the bug (and understanding how this all hangs
together) is far from trivial, I think this actual patch is trivial (it
just adds a comment). Therefore, in my opinion, this patch only requires
> local compilation (just a comment added)
More information about the hotspot-gc-dev