7090802: segfault in GraphBuilder [Was: Re: JVM segfault]

Roland Westrelin roland.westrelin at oracle.com
Tue Feb 14 09:26:44 PST 2012

>> I've been able to make the JVM segfault fairly reliably when using TieredCompilation.  I reported this as bugid 7090802.
>> # Problematic frame:
>> # V  [libjvm.so+0x1aa058][thread -1705837712 also had an error]
>>  GraphBuilder::GraphBuilder(Compilation*, IRScope*)+0x378
>> I haven't been able to have a bullet proof way to make this fail.   However, starting Intellij so that it has to re-index the source code base seems to work reliably.    It fails on a single CPU virtualbox instance,  dual cpu, 8 cpu and 16cpu systems.   It's failing on Ubuntu, Suse 11.2 and 11.4 systems.
>> This isn't that big of a deal as I just turn off TieredCompilation.   I'm taking the time to point this problem out though as I think TieredCompilation is great and want to see it be bullet proof.    I have never had an issue using either the server or client compiler without TieredCompilation.    I have also tried tweaking Tier2CompileTheshold up to 50000 and that didn't seem to make a difference either.    With PrintCompilation enabled I can see that it's failing after about 14000 methods are compiled with the C1 compiler.
>> So, my apologies for posting this here.   How can I help resolve this issue?
> The bug is currently in "Incomplete" state and marked as "Need More Info".  I guess this is because of a missing test case to reproduce the problem.  And it's not filed against compiler1 category (which I think it should be).

It looks like a duplicate of 7126041 that I got reviewed yesterday and will push shortly.


More information about the hotspot-compiler-dev mailing list