Uploaded image for project: 'JDK'
  1. JDK
  2. JDK-8145190

MinTLABSize can cause overflow problem with CMS GC

    XMLWordPrintable

    Details

    • Type: Bug
    • Status: Resolved
    • Priority: P2
    • Resolution: Fixed
    • Affects Version/s: 9
    • Fix Version/s: 9
    • Component/s: hotspot
    • Labels:
      None
    • Subcomponent:
      gc
    • Resolved In Build:
      b107

      Description

      Running 32-bit java on Linux system give crash with SIGFPE:
      java -XX:+UseConcMarkSweepGC -XX:MinTLABSize=4294967295 -version
      #
      # A fatal error has been detected by the Java Runtime Environment:
      #
      # SIGFPE (0x8) at pc=0xf6657ee8, pid=11974, tid=11975
      #
      # JRE version: (9.0) (build )
      # Java VM: Java HotSpot(TM) Server VM (9-internal+0-2015-12-07-211223.jprtadm.s, mixed mode, tiered, concurrent mark sweep gc, linux-x86)
      # Problematic frame:
      # V [libjvm.so+0x7b6ee8] CMSCollector::CMSCollector(ConcurrentMarkSweepGeneration*, CardTableRS*, ConcurrentMarkSweepPolicy*)+0xa88
      #
      # Core dump will be written. Default location: Core dumps may be processed with "/usr/share/apport/apport %p %s %c" (or dumping to /home/dmitry/work/bugs/8144578.CLOV_pass_options/bundle.i586/bin/core.11974)
      #
      # An error report file with more information is saved as:
      # /home/dmitry/work/bugs/8144578.CLOV_pass_options/bundle.i586/bin/hs_err_pid11974.log
      #
      ...

      I think that division by zero is happened in the following part of the code in CMSCollector::CMSCollector constructor(hotspot/src/share/vm/gc/cms/concurrentMarkSweepGeneration.cpp module):
      CMSCollector::CMSCollector(ConcurrentMarkSweepGeneration* cmsGen,
                                 CardTableRS* ct,
                                 ConcurrentMarkSweepPolicy* cp):
      ...
        // Support for parallelizing survivor space rescan
        if ((CMSParallelRemarkEnabled && CMSParallelSurvivorRemarkEnabled) || CMSParallelInitialMarkEnabled) {
          const size_t max_plab_samples =
            _young_gen->max_survivor_size() / (PLAB::min_size() * HeapWordSize);
      ...

      PLAB::min_size() returns 'align_object_size(MAX2(MinTLABSize / HeapWordSize, (uintx)oopDesc::header_size())) + AlignmentReserve'.

      Thus it seems that 'PLAB::min_size() * HeapWordSize' is overflow to 0.

      Range for MinTLABSize is (1,max_uintx). MinTLABSize is divided on HeapWordSize, but adding AlignmentReserve can cause overflow in CMSCollector::CMSCollector constructor.

      Thus, range or constraint should be corrected for MinTLABSize flag.

      I think that new test is not needed, because this flag will be tested for non-G1 GC mode after fixing JDK-8144578.

        Attachments

          Issue Links

            Activity

              People

              Assignee:
              sangheki Sangheon Kim
              Reporter:
              ddmitriev Dmitry Dmitriev
              Votes:
              0 Vote for this issue
              Watchers:
              4 Start watching this issue

                Dates

                Created:
                Updated:
                Resolved: