RFR: Disable ShHeapUncommit on too-large large-pages
Roman Kennke
rkennke at redhat.com
Tue Sep 11 08:22:09 UTC 2018
Am 11.09.2018 um 09:33 schrieb Aleksey Shipilev:
> On 09/11/2018 09:29 AM, Roman Kennke wrote:
>>> Yes, like that. Does it still pass tier3 with uber-large pages?
>>
>> No. Of course this is a mistake:
>>
>> diff -r 03c6138dd68c
>> src/hotspot/share/gc/shenandoah/shenandoahHeapRegion.cpp
>> --- a/src/hotspot/share/gc/shenandoah/shenandoahHeapRegion.cpp Tue Sep
>> 11 03:15:46 2018 -0400
>> +++ b/src/hotspot/share/gc/shenandoah/shenandoahHeapRegion.cpp Tue Sep
>> 11 03:29:01 2018 -0400
>> @@ -618,7 +618,7 @@
>> // Otherwise, uncommitting one region may falsely uncommit the adjacent
>> // regions too.
>> // Also see shenandoahArguments.cpp, where it handles UseLargePages.
>> - if (ShenandoahUncommit) {
>> + if (UseLargePages && ShenandoahUncommit) {
>> region_size = MAX2(region_size, os::large_page_size());
>> }
>>
>> Ok?
>
> Right. That makes more sense.
>
> -Aleksey
>
I also need the following fixes to heap alignment to make it work for real:
Incremental:
http://cr.openjdk.java.net/~rkennke/disable-uncommit/webrev.02.diff/
Full:
http://cr.openjdk.java.net/~rkennke/disable-uncommit/webrev.02/
This passes tier3_gc_shenandoah except for the remaining TAMS failure.
Ok?
Roman
More information about the shenandoah-dev
mailing list