[11] A question about the stub `generate_forward_exception`
Andy Law
question123446 at gmail.com
Thu Mar 12 11:13:04 UTC 2020
Hi,
(on X86_64) The `generate_forward_exception` and `generate_exception_blob`
code blobs can jump to `generate_deopt_blob`'s exception_offset position.
This time, all non-callee-saved registers are destroyed. Then depot blob
generates an OopMap which saves all registers in sight and when entering
into `Deoptimization::fetch_unroll_info` it will lookup deoptee's OopMap
and fill_in the local variable table. However, what if the deoptee's OopMap
contains non-callee-saved registers? I manually set a pending_exception in
`ThreadSafepointState::handle_polling_page_exception` and it had the same
crash stack as https://bugs.openjdk.java.net/browse/JDK-8028308, which
makes me ask this question.
Thanks,
Andy
More information about the hotspot-compiler-dev
mailing list