RFR: JDK-8232069: Enable CDS even when UseCompressedClassPointers and/or UseCompressedOops are false
Calvin Cheung
calvin.cheung at oracle.com
Fri Feb 7 00:06:52 UTC 2020
Hi Yumin,
Changes look good overall. I have two comments:
1. Since you've added 2 fields in the archive header. I think the
CURRENT_CDS_ARCHIVE_VERSION should be updated in share/include/cds.h.
2. In the new test TestCombinedCompressedFlags.java:
Instead of:
171 System.out.println("Platform is not 64 bit, skipped");
it is better to throw a SkippedException like the following:
throw new SkippedException("Platform is not 64 bit, skipped");
You would need to 'import jtreg.SkippedException'.
The following import statements seem unnecessary:
37 import jdk.test.lib.Asserts;
41 import jdk.test.lib.cds.CDSTestUtils;
42 import jdk.test.lib.process.ProcessTools;
Do you need the following @module?
33 * java.management
thanks,
Calvin
On 2/6/20 2:07 PM, Yumin Qi wrote:
> Hi, Ioi and all
>
>
> Forget to mention tests: hs-tier1 - 4 passed.
>
>
> Thanks
>
> Yumin
>
> On 2/6/20 2:04 PM, Yumin Qi wrote:
>> Hi, Ioi
>>
>>
>> Thanks for the review. See answers embedded.
>>
>> New URL: http://cr.openjdk.java.net/~minqi/8232069/webrev-01/
>>
>> On 2/5/20 10:34 AM, Ioi Lam wrote:
>>> Hi Yumin,
>>>
>>>
>>> Looks good. Some nitpicks on the tests:
>>>
>>> CommandLineFlagComboNegative.java: the following line is not needed.
>>>
>>> import jdk.test.lib.cds.CDSTestUtils.Result;
>>>
>> Removed.
>>>
>>> TestCombinedCompressedFlags.java: DUMP_NORMAL_MSG doesn't seem to be
>>> used by the test.
>>>
>> Removed.
>>
>> Also changed to use TestCommon.dump/TestCommon.exec which better to
>> check dum/run results.
>>
>>> DifferentHeapSizes.java: after your change,
>>> CDSTestUtils.MSG_COMPRESSION_MUST_BE_USED is no longer used anywhere
>>> in the tests. Also, This message is no longer printed by the VM. I
>>> think the CDSTestUtils.MSG_COMPRESSION_MUST_BE_USED field should be
>>> removed.
>>>
>>>
>> Removed.
>>> IncompatibleOptions.java: COOPS_DUMP_WARNING can be removed.
>>>
>> Removed.
>>> The check_shared parameter for testExec() should be removed. It's
>>> currently used only when expectedToFail == false. However, in all
>>> the cases where expectedToFail == false, check_shared is always true.
>>>
>>>
>> Removed.
>>
>>> Also, in filemap.cpp, before checking "compressed_oops() !=
>>> UseCompressedOops ...", I think it will be good to add a log to
>>> print out the expected compressed mode. That way if the user gets an
>>> error, they can run with -Xlog:cds to find out exactly what the
>>> reason is. E.g.
>>>
>>> log_info(cds)("Archive was created with UseCompressedOops = %d,
>>> UseCompressedClassPointers = %d",
>>> compressed_oops(),
>>> compressed_class_pointers());
>>> if (compressed_oops() != UseCompressedOops ...
>>>
>> Added.
>>
>>
>> Thanks
>>
>> Yumin
>>
>>>
>>> On 2/4/20 6:01 PM, Yumin Qi wrote:
>>>> Hi,
>>>>
>>>> Please review fix:
>>>>
>>>> Bug: https://bugs.openjdk.java.net/browse/JDK-8232069
>>>>
>>>> Webrev:http://cr.openjdk.java.net/~minqi/8232069/webrev/
>>>>
>>>> Description: Currently CDS only works when using compressed
>>>> Oops turned on, -XX:+UseCompressedOops (default), the fix is
>>>> allowing CDS works with uncompressed oops and uncompressed class
>>>> pointers. The contents of InstanceKlass::_fields are depending on
>>>> UseCompressedOops and UseCompressedClassPointers on/off. There are
>>>> three suggestions as described in bug description:
>>>>
>>>> 1. Remember the compression mode at CDS dump time. At runtime,
>>>> disable CDS if the compression mode is changed.
>>>> 2. Same as 1, but save more than one CDS archives in the JDK image.
>>>> Choose the appropriate one at start-up
>>>> 3. Recompute re-compute InstanceKlass::_fields at runtime.
>>>>
>>>> The fix takes the first one, storing UseCompressedOops and
>>>> UseCompressedClassPointers in shared archive. CDS is only allowed
>>>> at runtime if the flags are consistent with the stored versions in
>>>> archive.
>>>>
>>>> Testing: hs-tier1-2
>>>>
>>>>
>>>> Thanks
>>>>
>>>> Yumin
>>>>
>>>
More information about the hotspot-runtime-dev
mailing list