JDK-4375107 : HS1.3.1(solx86) crashes on the stress test jck122007
  • Type: Bug
  • Component: hotspot
  • Sub-Component: runtime
  • Affected Version: 1.3.1
  • Priority: P3
  • Status: Closed
  • Resolution: Not an Issue
  • OS: solaris_8
  • CPU: x86
  • Submitted: 2000-09-29
  • Updated: 2001-06-19
  • Resolved: 2001-04-23
Related Reports
Relates :  
JDK-4285670 - Error happened during: scavenge, at thread.cpp, 1348
Relates :  
JDK-4284209 - 1.3: invalid page fault under tests jck12a005 & jck122007
Relates :  
JDK-4695808 - stress test crashes on Solaris/x86
Relates :  
JDK-4306103 - HS 1.3-rc1S crashes with exit code 128 against the stress-test jck122007
Description

Name: elC94547			Date: 09/29/2000



Java 2 SDK 1.3.1-b1 (solx86) (both client and server) crash while executing 
the following testbase_nsk test:
    nsk/stress/jck122/jck122007
which stresses-out some of JCK 1.2.2 tests by executing the involved JCK 
tests in parallel threads. 

This crash looks similar to HS 1.3 (win32) crashes described by the bugs:
    4284209 (P4/S4) 1.3: invalid page fault under tests jck12a005 & jck122007
    4285670 (P4/S5) Error happened during: scavenge, at thread.cpp, 1348
    4306103 (P4/S5) HS 1.3-rc1S crashes with exit code 128 against the stress-test jck122007

The test 'jck122017' is found in the testbase_nsk suite:
    /net/sqesvr/export/vsn/VM/testbase/testbase_nsk

======================================================================

Name: elC94547			Date: 09/29/2000



To reproduce this failure, please try the script 'doit.sh' 
found at the directory:
    /net/sqesvr/vsn/export/GammaBase/Bugs/4375107

I will send 4375107-stuff.zip for this GammaBase entry to 
<###@###.###> just now; and Igor will probably 
create that directory today.

======================================================================
Comments
EVALUATION This failure is a regression from the kestrel release on windows. jon.masamitsu@Eng 2000-11-16 The test within the test that is failing is registerTest("javasoft.sqe.tests.vm.astore_w.astore_w004.astore_w00401.astore_ w00401"); // JDK 1.3 (win32) #4256603 (P1/S3) method with large number of locals causes StackOverflowError The test fails on both sparc and i386 with a thread stack size of 256k and both pass with a thread stack size of 512k. The default thread stack size on sparc is 512k and on i386 is 256k. This test is causing the native thread to overflow its stack when the snaller stack size is used. jon.masamitsu@Eng 2001-04-23 The stack size requirements on sparc are generally larger than on i386 because of the register windows so the need for a larger default stack size on sparc is to be expected. There is no practical way to choose the default stack size on both architectures such that the any program will always run on the default stack. A sparc default stack that is twice the i386 default stack size is reasonable. The concensus is that the stack size on i386 should not be changed. This is not a bug. The test program should be run with a larger stack size if needed.
11-06-2004
WORK AROUND Increase thread stack size (-Xss option)
11-06-2004