[9] RFR(S): 8040085: dtrace/jsdt tests crash on solaris. found an unadvertised bad scavengable oop in the code cache

[Christian Thalinger]

On Apr 28, 2014, at 7:29 PM, Albert <albert.noll at oracle.com> wrote:

> Hi Chris,
> 
> thanks for looking at this. Unfortunately, the change is already pushed.
> Shall I file a new bug for adding the code you proposed? Alternatively,
> we could also do the change when fixing: 
> 
> https://bugs.openjdk.java.net/browse/JDK-8022968
> 
> I would update the bug accordingly.
> 
> What do you think?

Yeah, push it as part of this one.

> 
> Best,
> Albert
> 
> On 04/28/2014 09:46 PM, Christian Thalinger wrote:
>> Not sure if this was pushed already but you should do something like AdapterHandlerLibrary::create_native_wrapper to print compilation and handle a full code cache:
>> 
>>   // Install the generated code.
>>   if (nm != NULL) {
>>     if (PrintCompilation) {
>>       ttyLocker ttyl;
>>       CompileTask::print_compilation(tty, nm, method->is_static() ? "(static)" : "");
>>     }
>>     nm->post_compiled_method_load_event();
>>   } else {
>>     // CodeCache is full, disable compilation
>>     CompileBroker::handle_full_code_cache();
>>   }
>> 
>> Perhaps without the load event.
>> 
>> On Apr 24, 2014, at 10:40 PM, Albert <albert.noll at oracle.com> wrote:
>> 
>>> Hi Roland,
>>> 
>>> thanks for your quick response. Here is the updated webrev that keeps the print().
>>> 
>>> http://cr.openjdk.java.net/~anoll/8040085/webrev.02/
>>> 
>>> Best,
>>> Albert
>>> 
>>> On 04/25/2014 10:27 AM, Roland Westrelin wrote:
>>>>> Here is a new webrev, which includes the clean-up.
>>>>> http://cr.openjdk.java.net/~anoll/8040085/webrev.01/
>>>> I think we still want the:
>>>> 
>>>> tty->print("---   n  “);
>>>> 
>>>> part to match the PrintCompilation output for other methods (it doesn’t really match anymore, actually).
>>>> 
>>>> Roland.
>>>> 
>>>>> Best,
>>>>> Albert
>>>>> 
>>>>> On 04/25/2014 09:59 AM, Roland Westrelin wrote:
>>>>>>> 7017732 was pushed 3 years ago. What does this show up now? Also, CodeCache::add_scavenge_root_nmethod() is required                     if the nmethod embeds oops but I don’t see where in SharedRuntime::generate_dtrace_nmethod() we have constant oops.
>>>>>> I took a look and found that the oop comes from:
>>>>>> 
>>>>>> // Add the class loader of Method* for the nmethod itself
>>>>>> append_oop_references(&oops, mh->method_holder());
>>>>>> 
>>>>>> in finalize_oop_references() that is called from nmethod::new_dtrace_nmethod()
>>>>>> 
>>>>>> So Albert’s fix look good to me.
>>>>>> 
>>>>>> I also noticed this in AdapterHandlerLibrary::create_dtrace_nmethod():
>>>>>>   if (PrintCompilation) {
>>>>>>     ttyLocker ttyl;
>>>>>>     tty->print("---   n%s  ");
>>>>>> 
>>>>>> %s but no argument passed. Can you clean that up as well, Albert?
>>>>>> 
>>>>>> Roland.
>>> 
>> 
> 

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.openjdk.java.net/pipermail/hotspot-compiler-dev/attachments/20140429/8164d189/attachment.html>