[16] RFR(T) 8248426: NMT: VirtualMemoryTracker::split_reserved_region() does not properly update summary counting

[Thomas Stüfe]

On Mon, Jun 29, 2020, 07:42 Thomas Stüfe <thomas.stuefe at gmail.com> wrote:

> Hi Zhengyu,
>
> thanks for finding this.
>
> VirtualMemoryTracker::remove_released_region() makes me nervous though
> because it does so much more than just removing the bookkeeping for the
> region.
> - removing committed regions - the space should be uncommitted
> - all that special CDS handling. Basically means I will never be able to
> split a reserved region tracked as mtClassShared.
>

(Scratch that, it still would work since this special handling only applies
to partwise release)

- splitting the reserved region
>
> Could we not simply add the call
> to VirtualMemorySummary::record_released_memory(size,
> reserved_rgn->flag()); to VirtualMemoryTracker::split_reserved_region()
> instead?
>
> Thanks, Thomas
>
>
>
>
>
> On Fri, Jun 26, 2020 at 10:54 PM Zhengyu Gu <zgu at redhat.com> wrote:
>
>> Hi,
>>
>> Please review this trivial patch that fixes summary counting in
>> VirtualMemoryTracker::split_reserved_region().
>>
>> The method uses internal method to remove a reserved region, which does
>> not update counting information. It should use high level tracking
>> method VirtualMemoryTracker::remove_released_region() instead.
>>
>> Without patch, NMT summary reports uncategorized memory, e.g.
>>
>> -                   Unknown (reserved=1060416KB, committed=0KB)
>>                              (mmap: reserved=1060416KB, committed=0KB)
>>
>>
>>
>>
>> Bug: https://bugs.openjdk.java.net/browse/JDK-8248426
>> Webrev: http://cr.openjdk.java.net/~zgu/JDK-8248426/webrev.00/
>>
>> Test:
>>    hotspot_nmt
>>    Submit test in progress
>>
>>
>> Thanks,
>>
>> -Zhengyu
>>
>>