RFR(XS) 8010500: [parfait] Possible null pointer dereference at hotspot/src/share/vm/opto/loopnode.hpp

[Niclas Adlertz]

Hi David,

I'm sorry about that. You are right, I should have added some background 
information and reasoning. Luckily Vladimir Kozlov already added some 
information.

Background:
Sometimes the loop header can be optimized away in IGVN. Because of 
this, we need to check if loopnode() (which returns the loop header) 
returns NULL in get_pre_loop_end().
The complete pre-loop (header and end) is required for the storeword 
optimization.

As Vladimir said, get_pre_loop_end() is called in three places during 
the superword optimization. If get_pre_loop_end() returns NULL the first 
time, we abort the optimization. If not, we assume that the complete 
pre-loop exists throughout the whole optimization phase. That is why we 
add an assert at the two remaning calls to get_pre_loop_end().

Kind Regards,
Niclas Adlertz

On 02/17/2014 07:58 PM, David Chase wrote:
> I am not a Reviewer, but is this the best way to address a Parfait warning?
> I don't see any reasoning, either in the bug report comments or in the patch, about how
> the null pointer should not actually appear, and the assert merely changes one sort
> of crash into another for debug and fastdebug builds.
>
> David
>
> On 2014-02-17, at 3:49 PM, Niclas Adlertz <niclas.adlertz at oracle.com> wrote:
>
>> Hi all,
>>
>> Could I please have a review for this very small fix?
>>
>> http://cr.openjdk.java.net/~adlertz/JDK-8010500/webrev00/
>> https://bugs.openjdk.java.net/browse/JDK-8010500
>>
>> Kind Regards,
>> Niclas Adlertz
>