SIGBUS in Access<1572864UL>::store_at<int> on Solaris/SPARC

[David Holmes]

Fix has been pushed.

David

On 23/02/2018 9:49 PM, Roman Kennke wrote:
> I think it will.
> 
> Roman
> 
> On Fri, Feb 23, 2018 at 12:45 PM,  <jesper.wilhelmsson at oracle.com> wrote:
>> Hi Roman,
>>
>> Will this fix be pushed today?  If not I would prefer to back out JDK-8197999 to get clean results in the Friday nightly.
>>
>> Thanks,
>> /Jesper
>>
>>
>>> On 22 Feb 2018, at 22:17, Roman Kennke <rkennke at redhat.com> wrote:
>>>
>>> I took the bug.
>>> I tried with casting similar to what you suggested, but that failed my
>>> test. Maybe I had the parenthesis differently? In any case, I made it
>>> so that it matches what is in oop.inline.hpp boolean accessor.
>>>
>>> I'll move the test to a new file and post an RFR separately.
>>>
>>> Thanks, Roman
>>>
>>>
>>> On Thu, Feb 22, 2018 at 10:04 PM, Stefan Karlsson
>>> <stefan.karlsson at oracle.com> wrote:
>>>> On 2018-02-22 21:41, Roman Kennke wrote:
>>>>>
>>>>> Ok thank you.
>>>>> I tried your patch and can confirm that it works/passes. :-) It also
>>>>> gives me some ideas how gtest works.
>>>>>
>>>>> I modified the test so that it fails without the fix, and passes with the
>>>>> fix:
>>>>>
>>>>> http://cr.openjdk.java.net/~rkennke/8198564/webrev.00/
>>>>>
>>>>> If you think that's good, then I can post a formal RFR and take over the
>>>>> bug.
>>>>
>>>>
>>>> Yes, this seems good. A similar patch using (jboolean)(((jint)contents) & 1)
>>>> passes tests that used to fail on sparc.
>>>>
>>>> You might want to consider moving the test to a test_typeArrayOop.cpp file.
>>>>
>>>> Thanks,
>>>> StefanK
>>>>
>>>>
>>>>>
>>>>> Roman
>>>>>
>>>>> On Thu, Feb 22, 2018 at 9:00 PM, Stefan Karlsson
>>>>> <stefan.karlsson at oracle.com> wrote:
>>>>>>
>>>>>> On 2018-02-22 20:47, Roman Kennke wrote:
>>>>>>>
>>>>>>> On Thu, Feb 22, 2018 at 8:22 PM, Stefan Karlsson
>>>>>>> <stefan.karlsson at oracle.com> wrote:
>>>>>>>>
>>>>>>>> On 2018-02-22 20:14, Roman Kennke wrote:
>>>>>>>>>
>>>>>>>>> Right. This looks like possible and likely cause of the problem. And
>>>>>>>>> it worked before because of implicit conversion back to jboolean:
>>>>>>>>>
>>>>>>>>> -  void bool_at_put(int which, jboolean contents)  {
>>>>>>>>> *bool_at_addr(which) = (((jint)contents) & 1); }
>>>>>>>>>
>>>>>>>>>
>>>>>>>>> Can you test it? Because, I can't ;-)
>>>>>>>>
>>>>>>>>
>>>>>>>> Yes. I'm kicking of some testing on sparc. Could you write a gtest for
>>>>>>>> this?
>>>>>>>
>>>>>>> I can try. I never wrote a gtest before ;-) Is there an existing one
>>>>>>> that I could use as template, and/or pointers how to start?
>>>>>>
>>>>>>
>>>>>> You can look at the existing tests in test/hotspot/gtest. I suggest you
>>>>>> read
>>>>>> the official googletest doc to get started. There might be some other
>>>>>> document about our adaption of googletest, but I don't know where it is.
>>>>>>
>>>>>> Maybe something like this would work:
>>>>>>
>>>>>> diff --git a/test/hotspot/gtest/oops/test_arrayOop.cpp
>>>>>> b/test/hotspot/gtest/oops/test_arrayOop.cpp
>>>>>> --- a/test/hotspot/gtest/oops/test_arrayOop.cpp
>>>>>> +++ b/test/hotspot/gtest/oops/test_arrayOop.cpp
>>>>>> @@ -22,6 +22,7 @@
>>>>>>    */
>>>>>>
>>>>>>   #include "precompiled.hpp"
>>>>>> +#include "memory/universe.hpp"
>>>>>>   #include "oops/arrayOop.hpp"
>>>>>>   #include "oops/oop.inline.hpp"
>>>>>>   #include "unittest.hpp"
>>>>>> @@ -86,4 +87,37 @@
>>>>>>   TEST_VM(arrayOopDesc, narrowOop) {
>>>>>>     ASSERT_PRED1(check_max_length_overflow, T_NARROWOOP);
>>>>>>   }
>>>>>> +
>>>>>> +TEST_VM(arrayOopDesc, bool_at_put) {
>>>>>> +  char mem[100];
>>>>>> +  memset(mem, 0, ARRAY_SIZE(mem));
>>>>>> +
>>>>>> +  char* addr = align_up(mem, 16);
>>>>>> +
>>>>>> +  typeArrayOop o = (typeArrayOop) addr;
>>>>>> +  o->set_klass(Universe::boolArrayKlassObj());
>>>>>> +  o->set_length(10);
>>>>>> +
>>>>>> +
>>>>>> +  ASSERT_EQ((jboolean)0, o->bool_at(0));
>>>>>> +  ASSERT_EQ((jboolean)0, o->bool_at(1));
>>>>>> +  ASSERT_EQ((jboolean)0, o->bool_at(2));
>>>>>> +  ASSERT_EQ((jboolean)0, o->bool_at(3));
>>>>>> +  ASSERT_EQ((jboolean)0, o->bool_at(4));
>>>>>> +  ASSERT_EQ((jboolean)0, o->bool_at(5));
>>>>>> +  ASSERT_EQ((jboolean)0, o->bool_at(6));
>>>>>> +  ASSERT_EQ((jboolean)0, o->bool_at(7));
>>>>>> +
>>>>>> +  o->bool_at_put(0, 1);
>>>>>> +
>>>>>> +  ASSERT_EQ((jboolean)1, o->bool_at(0));
>>>>>> +  ASSERT_EQ((jboolean)0, o->bool_at(1));
>>>>>> +  ASSERT_EQ((jboolean)0, o->bool_at(2));
>>>>>> +  ASSERT_EQ((jboolean)0, o->bool_at(3));
>>>>>> +  ASSERT_EQ((jboolean)0, o->bool_at(4));
>>>>>> +  ASSERT_EQ((jboolean)0, o->bool_at(5));
>>>>>> +  ASSERT_EQ((jboolean)0, o->bool_at(6));
>>>>>> +  ASSERT_EQ((jboolean)0, o->bool_at(7));
>>>>>> +}
>>>>>> +
>>>>>>   // T_VOID and T_ADDRESS are not supported by max_array_length()
>>>>>>
>>>>>> And then run with:
>>>>>> ../build/fastdebug/hotspot/variant-server/libjvm/gtest/gtestLauncher -jdk
>>>>>> ../build/fastdebug/jdk --gtest_filter="arrayOopDesc*"
>>>>>>
>>>>>> StefanK
>>>>>>
>>>>>>> Roman
>>>>>>
>>>>>>
>>>>>>
>>>>
>>