Request for Review (S) 6852078: HSX 14/16 in jdk 5.0: api/javax_management api/org_omg jck tests crashes or make tnameserv crash

[Changpeng Fang]

On 10/29/09 17:18, Vladimir Kozlov wrote:
> I don't think having assert there is correct
> since there is nothing wrong with ideal graph.
>
1924   //unsafe reference could not be aligned appropriately without runtime checking
1925   if (base == NULL || base->bottom_type() == Type::TOP) {
1926     assert(!valid(), "unsafe access");
1927     return;
1928   }


The assert is nothing related to ideal graph. It just makes sure that 
the SWPointer returned from
here is not useable. You may see many places that use SWPointer::valid() 
checking before  doing
actual thing (alignment, for example).

Thanks,

Changpeng


> Vladimir
>
> Changpeng Fang wrote:
>> Was sent  to the gc mailing list. Resend to the correct compiler 
>> list. Thanks,
>>
>>> http://cr.openjdk.java.net/~cfang/6852078/webrev.00/
>>>
>>> Note that I am requesting for change to the current jdk source (For 
>>> the SubCR)
>>>
>>> 6852078: HSX 14/16 in jdk 5.0: api/javax_management api/org_omg jck 
>>> tests crashes or make tnameserv crash
>>>
>>> Problem: Current implementation of superword optimization does not 
>>> check for unsafe memory accesses. Since
>>> unsafe memory accesses could not be aligned appropriately without 
>>> runtime checking, a un-aligned superword
>>> read/write could cause runtime failure.
>>>
>>> Solution: Simply disable superword for unsafe memory access whose 
>>> base is top.
>>>
>>> Tests: Test case in 6852078
>>>
>>> Thanks,
>>>
>>> Changpeng
>>>
>>>
>>