RFR: 8240370: Provide Intel JCC Erratum opt-out
Erik Österlund
erik.osterlund at oracle.com
Tue Mar 3 17:27:30 UTC 2020
Hi Vladimir,
Thanks for the review.
/Erik
> On 3 Mar 2020, at 18:06, Vladimir Ivanov <vladimir.x.ivanov at oracle.com> wrote:
>
> Looks good.
>
> Best regards,
> Vladimir Ivanov
>
>> On 3 Mar 2020, at 19:35, Erik Österlund <erik.osterlund at oracle.com> wrote:
>>
>> Hi Vladimir,
>>
>> Sure, why not. I liked the second alternative:
>> http://cr.openjdk.java.net/~eosterlund/8240370/webrev.02/
>>
>> Thanks,
>> /Erik
>>
>>>> On 2020-03-03 16:29, Vladimir Ivanov wrote:
>>> Hi Erik,
>>>
>>>> New webrev: http://cr.openjdk.java.net/~eosterlund/8240370/webrev.01/
>>>
>>> IMO it makes sense to support both forcibly turning the flag on and off from the command-line.
>>>
>>> So, I prefer:
>>>
>>> bool VM_Version::compute_has_intel_jcc_erratum() {
>>> + if (!FLAG_IS_DEFAULT(IntelJccErratumMitigation)) {
>>> + return IntelJccErratumMitigation;
>>> + }
>>>
>>> Or even:
>>>
>>> if (FLAG_IS_DEFAULT(IntelJccErratumMitigation)) {
>>> _has_intel_jcc_erratum = compute_has_intel_jcc_erratum();
>>> } else {
>>> _has_intel_jcc_erratum = IntelJccErratumMitigation;
>>> }
>>>
>>> Best regards,
>>> Vladimir Ivanov
>>>
>>>>> /Claes
>>>>>
>>>>> On 2020-03-03 13:23, Erik Österlund wrote:
>>>>>> Hi,
>>>>>>
>>>>>> Today, the Intel JCC Erratum mitigation is automatically enabled when running on an affected machine.
>>>>>> While it is strongly recommended that customers run code in environments with the micro code updates
>>>>>> provided by Intel to mitigate the erratum, and hence are probably almost always better off with the
>>>>>> mitigation, there could be value in being able to opt out to analyze the performance characteristics
>>>>>> of an application, as a diagnostic option.
>>>>>>
>>>>>> This patch adds a diagnostic option that allows you to opt out with: -XX:-IntelJccErratum
>>>>>>
>>>>>> Webrev:
>>>>>> http://cr.openjdk.java.net/~eosterlund/8240370/webrev.00/
>>>>>>
>>>>>> Bug:
>>>>>> https://bugs.openjdk.java.net/browse/JDK-8240370
>>>>>>
>>>>>> Thanks,
>>>>>> /Erik
>>>>
>>
>
More information about the hotspot-compiler-dev
mailing list