Advice on finding out the cause of test failure of VarHandleTestAccessBoolean on windows x64

[Paul Sandoz]

> On 12 Aug 2016, at 11:09, dean.long at oracle.com wrote:
> 
> Tobias, does this sound like 8038348?
> 

I applied the patch [1] referenced in the issue but the bug was still reproducible.

I have observed the failure to occur on access to instance fields, static fields or arrays, via Unsafe access.

Thanks,
Paul.

[1] http://cr.openjdk.java.net/~thartmann/8038348/webrev.00/

> dl
> 
> 
> On 8/12/16 8:36 AM, Paul Sandoz wrote:
>> Hi,
>> 
>> Please can someone help/advice how to find the cause of a test failure i am observing.
>> 
>> The patch for VarHandles bitwise operations [1] is failing for boolean, the test VarHandleTestAccessBoolean, but only on windows x64. I have observed it can reliably fail in different assertions when testing bitwise ops and AFAICT so far only for C2 (and definitely not in the interpreter).
>> 
>> Here is an example:
>> 
>>   // get and bitwise and
>>   {
>>       vh.set(true);
>> 
>>       boolean o = (boolean) vh.getAndBitwiseAnd(false);
>>       assertEquals(o, true, "getAndBitwiseAnd boolean”);  // <— PASSES for the previous value
>>       boolean x = (boolean) vh.get();
>>       assertEquals(x, (boolean)(true & false), "getAndBitwiseAnd boolean value”);  // <— FAILS for the updated value, expected false got true
>>   }
>> 
>> 
>> It appears as if the atomic read modify write of “vh.getAndBitwiseAnd" failed and when getting the value (“vh.get”) it reads the previous value.
>> 
>> I don’t know what could be different about the windows x64 platform that such a bug could arise just on that platform.
>> 
>> 
>> vh.set/set will defer to Unsafe.put/getBoolean, where as the vh.getAndBitwiseAnd will call:
>> 
>>   @ForceInline
>>   public final boolean getAndBitwiseAndBoolean(Object o, long offset, boolean mask) {
>>       return byte2bool(getAndBitwiseAndByte(o, offset, bool2byte(mask)));
>>   }
>> 
>>   @ForceInline
>>   public final byte getAndBitwiseAndByte(Object o, long offset, byte mask) {
>>       byte current;
>>       do {
>>           current = getByteVolatile(o, offset);
>>       } while (!weakCompareAndSwapByteVolatile(o, offset,
>>                                                 current, (byte) (current & mask)));
>>       return current;
>>   }
>> 
>> which defers to reading/writing the boolean as the VM byte representation. So there is a mismatch going on between T_BOOLEAN and T_BYTE but i gather that should not be an issue.
>> 
>> If i disable the Unsafe _getByte and _getByteVolatile intrinsics the test passes.
>> 
>> (I have yet to work out how to compile and use hdis on windows x64).
>> 
>> Thanks,
>> Paul.
>> 
>> [1] http://cr.openjdk.java.net/~psandoz/jdk9/JDK-8161444-vhs-bitwise-atomics/webrev/
>> 
> 

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 841 bytes
Desc: Message signed with OpenPGP using GPGMail
URL: <http://mail.openjdk.java.net/pipermail/hotspot-compiler-dev/attachments/20160812/0efb3f9e/signature.asc>