[14] RFR(S): 8225653: Provide more information when hitting SIGILL from HaltNode

[Vladimir Ivanov]

Overall, looks good.

What do you think about reusing MacroAssembler::stop()? Otherwise, you 
have to add MacroAssembler::halt for x86-32 as well.

Best regards,
Vladimir Ivanov

On 04/09/2019 10:43, Christian Hagedorn wrote:
> Hi
> 
> Please review the following patch:
> https://bugs.openjdk.java.net/browse/JDK-8225653
> http://cr.openjdk.java.net/~chagedorn/8225653/webrev.00/
> 
> This enhancement provides more information about executing a HaltNode on 
> x86. Up to now, an illegal instruction was emitted for a HaltNode which 
> resulted in a crash with SIGILL at runtime if executed. This made it 
> hard to determine if the crash was actually caused by a HaltNode or not 
> without doing a closer analysis.
> 
> This patch adds a string to each HaltNode stating the purpose of it. 
> Code generation for a HaltNode adds a runtime call which executes 
> fatal() with the newly provided HaltNode string. The VM then crashes 
> with this additional information which should make a further analysis 
> easier. I filed an RFE [1] to also provide this information for 
> architectures other than x86.
> 
> The assert at [2] is not necessary anymore since there is a new 
> DEFINE_CLASS_ID entry for HaltNode with ID 15. ClassMask_Halt now equals 
> USHORT_MAX, making the assert always true. It might be a good idea to 
> file an RFE to change the current encoding as the next entry with ID 16 
> will exceed the 16 bit limit.
> 
> Thank you!
> 
> Best regards,
> Christian
> 
> 
> [1] https://bugs.openjdk.java.net/browse/JDK-8230552
> [2] 
> http://hg.openjdk.java.net/jdk/jdk/file/d8f22418ca99/src/hotspot/share/opto/node.hpp#l750 
>