RFR (XS) 8220198: Lots of com/sun/crypto/provider/Cipher tests fail on x86_32 due to missing SHA512 stubs

[Vladimir Kozlov]

Thank you, Aleksey

vm_version_x86.cpp changes are good. And I agree with assert in GraphKit.

But I would also suggest to add (stubAddr == NULL) check in 
LibraryCallKit::inline_sha_implCompress(). We do such check in other intrinsics [1].

And please file a new bug to fix other intrinsics which missing the check - I see such cases in 
library_call.cpp.

Thanks,
Vladimir

[1] http://hg.openjdk.java.net/jdk/jdk/file/c12b897021ea/src/hotspot/share/opto/library_call.cpp#l5986

On 3/26/19 6:19 AM, Aleksey Shipilev wrote:
> Bug:
>    https://bugs.openjdk.java.net/browse/JDK-8220198
> 
> Fix:
>    http://cr.openjdk.java.net/~shade/8220198/webrev.01/
> 
> Actually, we might consider enabling these stubs for 32-bit builds, but it would require more work.
> I want to get this easy patch in to make it cleanly backportable to 12u and 11u, where x86_32 is
> broken too. Not very sure about the assert in GraphKit: it turns the cryptic crash into proper
> failure, and I _think_ there are no false negatives from it, because calling to NULL stub would
> crash eventually anyway.
> 
> Testing: Linux {x86_64, x86_32} tier1, Cipher tests are fixed on 32-bit, jdk-submit (running)
> 
> Thanks,
> -Aleksey
> 
>