RFR: 8245106: ZGC: Fix incorrect setup when using -XX:+UseTransparentHugePages
Per Liden
per.liden at oracle.com
Mon May 18 10:24:03 UTC 2020
Thanks Stefan!
/Per
On 5/15/20 7:52 PM, Stefan Karlsson wrote:
> I talked to Per directly about some nits and additional comments. Other
> than that, this looks good to me.
>
> StefanK
>
> On 2020-05-15 14:04, Per Liden wrote:
>> When using -XX:+UseTransparentHugePages, ZGC mistakenly calls
>> madvise(..., MADV_HUGE); too late. The end result is that large pages
>> are never allocated on first page fault, only via THP defrag. This in
>> turn means that the majority of the heap will typically not be backed
>> by large pages, which can have a very noticeable performance impact.
>>
>> This patch depends on "8245098: Make SafeFetch32/N available earlier",
>> which is currently out for review on hotspot-dev.
>>
>> Bug: https://bugs.openjdk.java.net/browse/JDK-8245106
>> Webrev: http://cr.openjdk.java.net/~pliden/8245106/webrev.0
>>
>> /Per
>
More information about the hotspot-gc-dev
mailing list