12 RFR(M) 8214583: AccessController.getContext may return wrong value after JDK-8212605

[dean.long at oracle.com]

On 12/16/18 7:39 PM, dean.long at oracle.com wrote:
> On 12/16/18 7:03 PM, David Holmes wrote:
>> On 17/12/2018 12:49 pm, dean.long at oracle.com wrote:
>>> On 12/16/18 4:06 PM, David Holmes wrote:
>>>> On 15/12/2018 10:59 am, dean.long at oracle.com wrote:
>>>>> https://bugs.openjdk.java.net/browse/JDK-8214583
>>>>> http://cr.openjdk.java.net/~dlong/8214583/webrev
>>>>>
>>>>> This change includes two new regression test that demonstrate the 
>>>>> problem, and a fix that allows the tests
>>>>> to pass.
>>>>>
>>>>> The problem happens when the JIT compiler's escape analysis 
>>>>> eliminates the allocation of the AccessControlContext object 
>>>>> passed to doPrivileged.  The compiler thinks this is safe because 
>>>>> it does not see that the object "escapes".
>>>>
>>>> Then surely the compiler's notion of "escapes" needs to be updated!
>>>>
>>>
>>> The compiler can inline the callee method and see that the value 
>>> doesn't escape.  This is a valid optimization in cases where the 
>>> callee method is known.
>>
>> But it's not a valid optimization in this case, so my comment stands.
>>
>> Is this stack walking something this is guaranteed by the spec to be 
>> always valid (and hence the JIT is violating the rules), or is the 
>> stack walking code making assumptions about whether it will find the 
>> context object in the stack?
>>
>
> The stack walking is in the VM and is an internal implementation 
> detail, not part of the AccessController API spec.  A different thread 
> running normal Java code would never be able to see a non-escaping 
> value.  The stack walking code does need to find the context object in 
> the stack.  Non-escaping objects won't show up in the stack.
>
>> If we have to hack around this with an annotation I'd rather see a 
>> specific annotation that addresses the problematic usecase than a 
>> generic "don't inline" one. E.g. @StackVisible or something like that.
>>
>
> That sounds like a good idea for 13, but would require changes to both 
> C2 and Graal, and it seems a little risky compared to using existing 
> mechanisms.
>

I forgot to address this in my last reply, but I'm not suggesting a 
@DontInline annotation.  That was Claes.  My fixes uses a native method.

dl

> dl
>
>> Cheers,
>> David
>>
>>>
>>> dl
>>>
>>>> David
>>>> -----
>>>>
>>>>   However, getContext needs to be able to find
>>>>> the object using a stack walk, so we need a way to tell the 
>>>>> compiler that it does indeed escape. To do this we pass the value 
>>>>> to a native method that does nothing.
>>>>>
>>>>> Microbenchmark results:
>>>>>
>>>>> jdk12-b18:
>>>>>
>>>>> Benchmark                Mode  Cnt    Score   Error  Units
>>>>> DoPrivileged.test        avgt   25  255.626 ± 6.446  ns/op
>>>>> DoPrivileged.testInline  avgt   25  250.968 ± 4.975  ns/op
>>>>>
>>>>>
>>>>> jdk12-b19:
>>>>>
>>>>> Benchmark                Mode  Cnt  Score    Error  Units
>>>>> DoPrivileged.test        avgt   25  5.689 ±  0.001  ns/op
>>>>> DoPrivileged.testInline  avgt   25  2.765 ±  0.001  ns/op
>>>>>
>>>>> this fix:
>>>>>
>>>>> Benchmark                Mode  Cnt  Score    Error  Units
>>>>> DoPrivileged.test        avgt   25  5.020 ±  0.001  ns/op
>>>>> DoPrivileged.testInline  avgt   25  2.774 ±  0.025  ns/op
>>>>>
>>>>>
>>>>> dl
>>>
>