RFR (S): 8181785: Remove the experimental ClearFPUAtPark JVM Flag

Erik Osterlund erik.osterlund at oracle.com
Fri Jun 9 17:02:55 UTC 2017


Thank you Dan.

/Erik

> On 9 Jun 2017, at 18:29, Daniel D. Daugherty <daniel.daugherty at oracle.com> wrote:
> 
> I agree. Primary reason is that it is experimental. Secondary reasons
> are "unsafe" and "unstable".
> 
> Dan
> 
> 
>> On 6/9/17 10:23 AM, Erik Osterlund wrote:
>> Hi Thomas, Dan and Robbin,
>> 
>> Thank you for reviewing. As Dan pointed out - this JVM flag has always been advertised as unsafe and unstable, hence nobody should be using it.
>> 
>> Do we agree we are okay with removing it without a CCC/CSR request?
>> 
>> Thanks,
>> /Erik
>> 
>>> On 9 Jun 2017, at 17:06, Daniel D. Daugherty <daniel.daugherty at oracle.com> wrote:
>>> 
>>> We have removed several experimental options without a CCC/CSR request.
>>> This one is identified as:
>>> 
>>> experimental(intx, ClearFPUAtPark, 0, "(Unsafe, Unstable)")
>>> 
>>> Since it is flagged as "Unsafe" and "Unstable" (like one or two I removed
>>> in JDK9), no one should be using it. It will not be noticed or missed.
>>> 
>>> Dan
>>> 
>>> 
>>> 
>>>>> On 6/9/17 6:36 AM, Robbin Ehn wrote:
>>>>> On 06/09/2017 02:27 PM, Thomas Schatzl wrote:
>>>>> Hi Erik,
>>>>> 
>>>>>> On Thu, 2017-06-08 at 17:04 +0200, Erik Österlund wrote:
>>>>>> Hi,
>>>>>> 
>>>>>> There is an experimental JVM flag called ClearFPUAtPark that is used
>>>>>> only by SPARC. It is turned off by default and is marked as Unsafe
>>>>>> and
>>>>>> Unstable.
>>>>>> The idea is to tell the Solaris kernel not to save floating point
>>>>>> registers when context switching.
>>>>>> I propose to remove it as it is an unstable option. It relies on an
>>>>>> implicit unstable OS contract with the kernel that may or may not be
>>>>>> honored.
>>>>>> 
>>>>>> Bug: https://bugs.openjdk.java.net/browse/JDK-8181785
>>>>>> 
>>>>>> Webrev: http://cr.openjdk.java.net/~eosterlund/8181785/webrev.00/
>>>>>> 
>>>>>> Testing: JPRT
>>>>>> 
>>>>>> Will need a sponsor.
>>>>>   looks good apart from copyright dates.
>>>>> 
>>>>> As even experimental VM options are part of the public VM interface, I
>>>>> think you need to do a CSR request.
>>>>> 
>>>>> Further I think we can't just remove these options but need to use the
>>>>> usual option deprecation/removal process.
>>>> Hi, here the runtime process:
>>>> http://mail.openjdk.java.net/pipermail/hotspot-runtime-dev/2015-June/015277.html
>>>> 
>>>>    release 1: Deprecate & Handle - warn and keep supporting
>>>>    release 2: Deprecate & Obsolete - warn and do nothing
>>>>    release 3: Dead - unrecognized
>>>> 
>>>> But in this case I suggest going for 2 directly and let CSR guys object if so..
>>>> 
>>>> /Robbin
>>>> 
>>>>> Thanks,
>>>>>   Thomas
>>>>> 
> 



More information about the hotspot-runtime-dev mailing list