Uploaded image for project: 'JDK'
  1. JDK
  2. JDK-8191362

[Graal] gc/g1/TestShrinkAuxiliaryData tests crash with "assert(check_klass_alignment(result)) failed: address not aligned"

    Details

    • Type: Bug
    • Status: Resolved
    • Priority: P3
    • Resolution: Fixed
    • Affects Version/s: 10
    • Fix Version/s: 10
    • Component/s: hotspot
    • Subcomponent:
    • Resolved In Build:
      b39

      Backports

        Description

        Test crash with one of the following asserts:

        # Internal Error (/workspace/open/src/hotspot/share/oops/klass.inline.hpp:66), pid=31106, tid=31126
        # assert(check_klass_alignment(result)) failed: address not aligned: 0x0000001000000005

        # Internal Error (/home/katya/work/JavaSE/Hotspot/ws/hs10.consolidated/open/src/hotspot/share/runtime/handles.cpp:37), pid=25578, tid=25631
        # assert(oopDesc::is_oop(obj)) failed: not an oop: 0x00000000cbde1068
        1. base.log
          5 kB
          Tobias Hartmann
        2. fix_nonFinal.log
          7 kB
          Tobias Hartmann
        3. fix.log
          5 kB
          Tobias Hartmann
        4. hs_err_8191362.log
          114 kB
          Ekaterina Pavlova

          Issue Links

            Activity

            Hide
            thartmann Tobias Hartmann added a comment -
            Thanks, Tom. Gilles mentioned that it would be better to preserve backwards compatibility of Graal with old (unpatched) JVMCI versions and therefore don't change the API.
            http://cr.openjdk.java.net/~thartmann/8191362/webrev.02/

            I've created GR-7679 to keep track of this and will add pull requests.
            Show
            thartmann Tobias Hartmann added a comment - Thanks, Tom. Gilles mentioned that it would be better to preserve backwards compatibility of Graal with old (unpatched) JVMCI versions and therefore don't change the API. http://cr.openjdk.java.net/~thartmann/8191362/webrev.02/ I've created GR-7679 to keep track of this and will add pull requests.
            Hide
            never Tom Rodriguez added a comment -
            But this does change the API while duplicating code into Graal. If we're attempting to preserve API I'd just assert that the passed in alignment value matches the value in HotSpotVMConfig. [~gdub]
            Show
            never Tom Rodriguez added a comment - But this does change the API while duplicating code into Graal. If we're attempting to preserve API I'd just assert that the passed in alignment value matches the value in HotSpotVMConfig. [~gdub]
            Hide
            thartmann Tobias Hartmann added a comment - - edited
            I think the point is that with webrev.02, Graal does still work with older versions of JVMCI because we don't rely on HotSpotMetaAccessProvider.computeArrayAllocationSize anymore.
            Show
            thartmann Tobias Hartmann added a comment - - edited I think the point is that with webrev.02, Graal does still work with older versions of JVMCI because we don't rely on HotSpotMetaAccessProvider.computeArrayAllocationSize anymore.
            Hide
            never Tom Rodriguez added a comment -
            In that case I'd advocate adding the proper API as a method like we did in 01 so that it's available in the right way even if graal doesn't use it for compatibility reasons.
            Show
            never Tom Rodriguez added a comment - In that case I'd advocate adding the proper API as a method like we did in 01 so that it's available in the right way even if graal doesn't use it for compatibility reasons.
            Hide
            hgupdate HG Updates added a comment -
            URL: http://hg.openjdk.java.net/jdk/jdk10/rev/5db30620a3db
            User: thartmann
            Date: 2018-01-10 08:08:42 +0000
            Show
            hgupdate HG Updates added a comment - URL: http://hg.openjdk.java.net/jdk/jdk10/rev/5db30620a3db User: thartmann Date: 2018-01-10 08:08:42 +0000

              People

              • Assignee:
                thartmann Tobias Hartmann
                Reporter:
                epavlova Ekaterina Pavlova
              • Votes:
                0 Vote for this issue
                Watchers:
                6 Start watching this issue

                Dates

                • Created:
                  Updated:
                  Resolved: