RFR(S): 8041481: JVM crashes with collect_args_for_profiling

[Roland Westrelin]

Thanks for the review Vladimir & Chris.
I’ll fix the copyright year.

Roland.

On Apr 24, 2014, at 7:52 PM, Christian Thalinger <christian.thalinger at oracle.com> wrote:

> The copyright year of the test is wrong:
>   2  * Copyright (c) 2013, Oracle and/or its affiliates. All rights reserved.
> Otherwise this looks good.
> 
> On Apr 24, 2014, at 6:08 AM, Roland Westrelin <roland.westrelin at oracle.com> wrote:
> 
>> The failure happens at a MethodHandle.linkTo* to a C1 intrinsic. The MethodHandle.linkTo* is inlined, and the argument profiling code tries to profile the member name argument that is already popped. The fix is to make sure we stop iterating over arguments once we’ve reached the end of the argument list and not only rely on the number of arguments that we expect to encounter at the call.
>> 
>> I also hit an assert failure in LIRGenerator::profile_type() which I think is due to class loading occurring concurrently with ci object initializations. I removed the asserts that made assumptions about the state of the ci objects and turned them into actual checks.
>> 
>> This will need to be backported to 8u20.
>> 
>> http://cr.openjdk.java.net/~roland/8041481/webrev.00/
>> 
>> Roland.
>