Round 3: RFR: 8013651 NMT: reserve/release sequence id's in incorrect order due to race

David Holmes david.holmes at oracle.com
Tue Jun 11 00:42:54 PDT 2013 

On 8/06/2013 12:55 AM, Zhengyu Gu wrote:
> On 6/6/2013 10:31 PM, David Holmes wrote:
>> I'm confused about the expectations of the constructor:
>>
>> MemTracker::Tracker::Tracker(MemoryOperation op, Thread* thr)
>>
>> Is thr, if not NULL, always the current thread? If so, then I don't
>> think it would be allocating from a SafepointSafe state; and if not
>> then it could change state immediately after you have checked it!
>>
> Yes, thr is the current thread, and we actually do classloading in
> _thread_in_native state, following is the stack.
>
>  >
> jvm.dll!MemTracker::Tracker::Tracker(MemTracker::Tracker::MemoryOperation op,
> Thread * thr)  Line 729    C++
>       jvm.dll!MemTracker::record_arena_size(unsigned char * addr,
> unsigned int size)  Line 344    C++
>       jvm.dll!Arena::set_size_in_bytes(unsigned int size)  Line 532 +
> 0xd bytes    C++
>       jvm.dll!Arena::grow(unsigned int x,
> AllocFailStrategy::AllocFailEnum alloc_failmode)  Line 570    C++
>       jvm.dll!Arena::Amalloc(unsigned int x,
> AllocFailStrategy::AllocFailEnum alloc_failmode)  Line 388 + 0x10
> bytes    C++
>       jvm.dll!ResourceArea::allocate_bytes(unsigned int size,
> AllocFailStrategy::AllocFailEnum alloc_failmode)  Line 71    C++
>       jvm.dll!resource_allocate_bytes(unsigned int size,
> AllocFailStrategy::AllocFailEnum alloc_failmode)  Line 39    C++
>       jvm.dll!ClassPathZipEntry::open_stream(const char * name)  Line
> 265 + 0xe bytes    C++
>       jvm.dll!LazyClassPathEntry::open_stream(const char * name) Line
> 323    C++
>       jvm.dll!ClassLoader::load_classfile(Symbol * h_name, Thread *
> __the_thread__)  Line 900 + 0x17 bytes    C++
>       jvm.dll!SystemDictionary::load_instance_class(Symbol * class_name,
> Handle class_loader, Thread * __the_thread__)  Line 1284 + 0x14 bytes
> C++
>       jvm.dll!SystemDictionary::resolve_instance_class_or_null(Symbol *
> name, Handle class_loader, Handle protection_domain, Thread *
> __the_thread__)  Line 769 + 0x18 bytes    C++
>       jvm.dll!SystemDictionary::resolve_or_null(Symbol * class_name,
> Handle class_loader, Handle protection_domain, Thread * __the_thread__)
> Line 227 + 0x15 bytes    C++
>       jvm.dll!SystemDictionary::resolve_or_fail(Symbol * class_name,
> Handle class_loader, Handle protection_domain, bool throw_error, Thread
> * __the_thread__)  Line 166 + 0x15 bytes    C++
>       jvm.dll!ConstantPool::klass_at_impl(constantPoolHandle this_oop,
> int which, Thread * __the_thread__)  Line 252 + 0x17 bytes    C++
>       jvm.dll!ConstantPool::klass_at(int which, Thread *
> __the_thread__)  Line 352 + 0x1b bytes    C++
>       jvm.dll!InterpreterRuntime::anewarray(JavaThread * thread,
> ConstantPool * pool, int index, int size)  Line 187 + 0x10 bytes C++
>       02ee6241()

So this is an IRT_ENTRY that puts us _thread_in_vm. Where does the 
transition to _thread_in_native occur in the subsequent frames?

David
-----

>       jvm.dll!JavaCalls::call_helper(JavaValue * result, methodHandle *
> m, JavaCallArguments * args, Thread * __the_thread__)  Line 402 + 0x36
> bytes    C++
>       jvm.dll!os::os_exception_wrapper(void (JavaValue *, methodHandle
> *, JavaCallArguments *, Thread *)* f, JavaValue * value, methodHandle *
> method, JavaCallArguments * args, Thread * thread)  Line 113 + 0x13
> bytes    C++
>       jvm.dll!JavaCalls::call(JavaValue * result, methodHandle method,
> JavaCallArguments * args, Thread * __the_thread__)  Line 307 + 0x1a
> bytes    C++
> jvm.dll!InstanceKlass::call_class_initializer_impl(instanceKlassHandle
> this_oop, Thread * __the_thread__)  Line 1125 + 0x1f bytes    C++
>       jvm.dll!InstanceKlass::call_class_initializer(Thread *
> __the_thread__)  Line 1093 + 0xd bytes    C++
>       jvm.dll!InstanceKlass::initialize_impl(instanceKlassHandle
> this_oop, Thread * __the_thread__)  Line 843    C++
>       jvm.dll!InstanceKlass::initialize(Thread * __the_thread__) Line
> 502 + 0xd bytes    C++
>       jvm.dll!initialize_class(Symbol * class_name, Thread *
> __the_thread__)  Line 973 + 0x20 bytes    C++
>       jvm.dll!Threads::create_vm(JavaVMInitArgs * args, bool *
> canTryAgain)  Line 3479 + 0xf bytes    C++
>       jvm.dll!JNI_CreateJavaVM(JavaVM_ * * vm, void * * penv, void *
> args)  Line 5113 + 0xd bytes    C++
>       java.exe!013713c1()
>       [Frames below may be incorrect and/or missing, no symbols loaded
> for java.exe]
>       java.exe!01371e78()
>       java.exe!0137cf67()
>       java.exe!0137ab83()
>       java.exe!0137cfa5()
>       java.exe!0137ac0d()
>       kernel32.dll!768b3677()
>       ntdll.dll!76fe9f42()
>       ntdll.dll!76fe9f15()
>
>
>> In memTracker.hpp for the !INCLUDE_NMT case class Tracker doesn't have
>> an allocation-type as a supertype. I'm also unclear whether the
>> methods that have:
>>
>> return Tracker();
>>
>> are going to be returning a stack-allocated object, or whether this
>> will actually force use of a copy-constructor at the call site. We
>> really want this to be a no-op (or as close as possible) when NMT is
>> not compiled in. Perhaps a static singleton instance of Tracker could
>> be used instead?
>>
> Yes, static on can help solaris. Windows and Linux all optimize the copy
> constructor away.
>
> Thanks,
>
> -Zhengyu
>
>> Thanks,
>> David
>>
>> On 5/06/2013 12:26 AM, Zhengyu Gu wrote:
>>> Round 3:
>>>
>>> Based on Coleen and Stefan's comment, reverted most of NMT tracking
>>> calls to original to reduce code changes.
>>>
>>> http://cr.openjdk.java.net/~zgu/8013651/webrev.03/
>>> <http://cr.openjdk.java.net/%7Ezgu/8013651/webrev.03/>
>>>
>>>
>>> Tests:
>>>    JPRT
>>>    vm.quick.testlist on Linux 32
>>>
>>> Thanks,
>>>
>>> -Zhengyu
>>>
>>>
>>> On 5/22/2013 10:28 AM, Zhengyu Gu wrote:
>>>> Based on the discussion with Karen, Coleen and Harold, following
>>>> changes are made:
>>>>
>>>> 1) Renamed NMTTrackOp to NMTTracker, avoid the confusion with VM
>>>> operations.
>>>> 2) Used NMTTracker's dtor to discard the tracking operation if no
>>>> recording is done.
>>>>
>>>> Tests:
>>>>  - JPRT
>>>>  - vm.quick.testlist on Linux 32, Solaris sparcv9 and Windows 32.
>>>>
>>>> Thanks,
>>>>
>>>> -Zhengyu
>>>>
>>>>
>>>> On 5/14/2013 10:01 AM, Zhengyu Gu wrote:
>>>>> There can be race conditions between the memory operations and the
>>>>> book keeping records are written. For example, thread 1 releases a
>>>>> virtual memory block, before it can write the release record, thread
>>>>> 2 reserves the same virtual memory block and writes reservation
>>>>> first, as result, NMT indicates the block is "released".
>>>>>
>>>>> The solution is that, for those operations that can cause the race
>>>>> conditions, NMT should pre-reserve sequence number for it, if the
>>>>> operation succeeds, NMT uses pre-reserved sequence number to write
>>>>> the record.
>>>>>
>>>>> The tricky part is that, a sequence number is only good for the
>>>>> generation it is acquired, when there are reserved sequence number,
>>>>> NMT has to prevent itself from entering so called "sync-point" where
>>>>> the generation can be advanced.
>>>>>
>>>>>
>>>>> Bug: http://bugs.sun.com/view_bug.do?bug_id=8013651
>>>>> Webrev: http://cr.openjdk.java.net/~zgu/8013651/webrev/
>>>>> <http://cr.openjdk.java.net/%7Ezgu/8013651/webrev/>
>>>>>
>>>>>
>>>>> Tests:
>>>>>    1) JPRT
>>>>>    2) vm.quick.testlist on Linux 32, Linux x64 and Solaris Sparcv9
>>>>>    3) Kitchensink on Linux 32, Linux x64, Solaris Sparcv9 and Windows
>>>>> x64
>>>>>    4) NMT jtreg tests on Linux 32, Linux x64 and Solaris Sparcv9
>>>>>
>>>>>
>>>>> Thanks,
>>>>>
>>>>> -Zhengyu
>>>>
>>>
>

More information about the hotspot-runtime-dev mailing list