RFR: 8158639: C2 compilation fails with SIGSEGV

[Jamsheed C m]

Hi Dean,

null_check_receiver_before_call() is made from callee *::generate fn,

and earlier, there should have been an implicit assumption that callsite 
call signature and callee defined method signature would be always 
same.. and so they would have used callee for these checks, which is no 
longer true!

as far as i know there are no other cases needed to be taken care.

Best Regards,

Jamsheed

On 8/23/2016 1:41 AM, dean.long at oracle.com wrote:
>
> It seems a little strange to continue passing a callee method to 
> null_check_receiver_before_call(), when it's only used for an assert.  
> It also makes me wonder if there are other places using arg_size() and 
> inc_sp() that need to be fixed.
>
> dl
>
>
> On 8/22/16 5:28 AM, Jamsheed C m wrote:
>>
>> Hi,
>>
>> bug id:https://bugs.openjdk.java.net/browse/JDK-8158639 
>> <https://bugs.openjdk.java.net/browse/JDK-8158639>
>> webrev:http://cr.openjdk.java.net/~jcm/8158639/webrev.00/ 
>> <http://cr.openjdk.java.net/%7Ejcm/8158639/webrev.00/>
>>
>> Request for review.
>>
>> Description
>> _linkTo* adapter elimination  in C2 cause wrong jvms  for null checks 
>> at callsite. which makes MemberName missing for trap and reexecute at 
>> these sites.
>>
>> and in reported bug, this makes wrong data getting loaded in profile 
>> data for trap and reexecute.and later getting crash at recompilation.
>>
>> Fix: fix the jvms based on _linkTo*(Declared method signature ) at 
>> these trap sites.
>>
>>
>> Best Regards,
>>
>> Jamsheed
>>
>>
>>
>

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.openjdk.java.net/pipermail/hotspot-compiler-dev/attachments/20160823/e17c9396/attachment.html>