RFR(S): 8231457 Asserts on AIX because os::elapsed_counter() is not monotonic

[David Holmes]

Hi Arno,

On 27/09/2019 5:43 pm, Zeller, Arno wrote:
> Hi David,
> 
> the tests look good and the changeset is unchanged (webrev: http://cr.openjdk.java.net/~azeller/webrevs/8231457_2/).
> Feel free to push any time you like.

For future reference when you are having a change sponsored you should 
prepare the final committed changeset with the correct changeset comment 
i.e.

8231457: Asserts on AIX because os::elapsed_counter() is not monotonic
Summary: replace gettimeofday with javaTimeNanos
Reviewed-by: dholmes, stuefe

That way your sponsor can just do a direct hg import.

Change has been pushed.

Cheers,
David

> Thanks a lot for your support,
> Arno
> 
> 
> On 26/09/2019 15:53 David Holmes wrote:
>> Hi Arno,
>>
>> On 26/09/2019 11:11 pm, Zeller, Arno wrote:
>>> Hi David,
>>>
>>> thanks for reviewing and sponsoring. I just put the patch in our nightly test
>> and would prefer to wait till tomorrow for the results - but I expect no issues.
>>
>> No problem, just let me know when you're ready and point me to the final
>> changeset. It's after 11pm for me now anyway :)
>>
>> Cheers,
>> David
>>
>>> Best regards,
>>> Arno
>>>
>>>> -----Original Message-----
>>>> From: David Holmes <david.holmes at oracle.com>
>>>> Sent: Donnerstag, 26. September 2019 13:34
>>>> To: Zeller, Arno <arno.zeller at sap.com>; hotspot-runtime-
>>>> dev at openjdk.java.net
>>>> Subject: Re: RFR(S): 8231457 Asserts on AIX because
>> os::elapsed_counter() is
>>>> not monotonic
>>>>
>>>> Hi Arno,
>>>>
>>>> On 26/09/2019 8:37 pm, Zeller, Arno wrote:
>>>>> Hi,
>>>>>
>>>>> we have seen assertions on AIX 7.2 on Power9 that complain about
>>>> negative durations. The issue seems to be caused by the implementation
>> of
>>>> o::elapsed_counter() on AIX that uses gettimeofday which is not
>> monotonic.
>>>> To solve the issue I prepared a small patch that let os::elapsed_counter()
>> use
>>>> javaTimeNanos() which should be monotonic. I had to adjust also
>>>> os::elapsed_frequency() and os::elapsedTime() to be consistent.
>>>>>
>>>>> Bug: https://bugs.openjdk.java.net/browse/JDK-8231457
>>>>> Webrev: http://cr.openjdk.java.net/~azeller/webrevs/8231457_2/
>>>>>
>>>>> I will need a sponsor for this patch.
>>>>
>>>> Changes looks good. I will sponsor this for you. Will wait for a second
>>>> reviewer.
>>>>
>>>> Thanks,
>>>> David
>>>>
>>>>
>>>>> Best regards,
>>>>> Arno
>>>>>