[kaffe] frustrating news regarding the pthread instability

[Dalibor Topic]

Noa Resare wrote:
> ons 2004-10-20 klockan 18:47 +0200 skrev Guilhem Lavaux:
> 
>>Hi Noa,
>>
>>I've just commited it. Please test it extensively. I would like also a 
>>thread hog test to see if there are memory leaks with the new scheme.
>>
>>Cheers,
>>
>>Guilhem.
> 
> 
> After half an hour of testing here it looks much better than before.
> I'll run it overnight and see how it works.
> 
> On the other hand it seems like the pthread_cond_destroy() problem that
> we had before have resurfaced. When I'm not LD_PRELOADing a noop
> pthread_cond_destroy() I get hangs every two minutes or so. I'll look a
> this some now, and probably some more tomorrow.

Hi Noa, salut Guilhem,

you could try running kaffe under valgrind. export KAFFE_DEBUG=valgrind 
and running mauve compiled with jikes with the latest CVS gave me these 
issues last evening using valgrind's default memcheck tool:

CLASSPATH=$CLASSPATH:`/bin/pwd` JAVAC="jikes -bootclasspath 
/home/topic/Projects/current/jre/lib/rt.jar" 
JAVA="/home/topic/Projects/current/bin/kaffe" 
/home/topic/Projects/current/bin/kaffe gnu.testlet.SimpleTestHarness
==12293== Memcheck, a memory error detector for x86-linux.
==12293== Copyright (C) 2002-2004, and GNU GPL'd, by Julian Seward et al.
==12293== Using valgrind-2.2.0, a program supervision framework for 
x86-linux.
==12293== Copyright (C) 2000-2004, and GNU GPL'd, by Julian Seward et al.
==12293== For more details, rerun with: -v
==12293==
==12293== warning: Valgrind's pthread_attr_setschedparam does nothing
==12293==          (scheduling not changeable)
==12293==          your program may misbehave as a result
==12293== warning: Valgrind's pthread_attr_setschedparam does nothing
==12293==          (scheduling not changeable)
==12293==          your program may misbehave as a result
==12293== Thread 2:
==12293== Conditional jump or move depends on uninitialised value(s)
==12293==    at 0x1B97E709: jthread_on_current_stack (thread-impl.c:1446)
==12293==    by 0x1B940B75: locks_internal_waitCond (locks.c:318)
==12293==    by 0x1B9C986A: finaliserMan (gc-incremental.c:876)
==12293==    by 0x1B948ED8: startSpecialThread (thread.c:308)
==12293==
==12293== Thread 2:
==12293== Conditional jump or move depends on uninitialised value(s)
==12293==    at 0x1B940726: putHeavyLock (locks.c:154)
==12293==    by 0x1B940CC7: locks_internal_waitCond (locks.c:319)
==12293==    by 0x1B9C986A: finaliserMan (gc-incremental.c:876)
==12293==    by 0x1B948ED8: startSpecialThread (thread.c:308)
==12293==
==12293== Thread 2:
==12293== Conditional jump or move depends on uninitialised value(s)
==12293==    at 0x1B97E709: jthread_on_current_stack (thread-impl.c:1446)
==12293==    by 0x1B9408F3: slowUnlockMutex (locks.c:223)
==12293==    by 0x1B9CB263: gc_heap_malloc (gc-mem.c:440)
==12293==    by 0x1B9C9E1A: gcMalloc (gc-incremental.c:1044)
==12293==
==12293== Thread 2:
==12293== Conditional jump or move depends on uninitialised value(s)
==12293==    at 0x1B97E711: jthread_on_current_stack (thread-impl.c:1446)
==12293==    by 0x1B9408F3: slowUnlockMutex (locks.c:223)
==12293==    by 0x1B9CB263: gc_heap_malloc (gc-mem.c:440)
==12293==    by 0x1B9C9E1A: gcMalloc (gc-incremental.c:1044)
==12293==
==12293== Conditional jump or move depends on uninitialised value(s)
==12293==    at 0x1B97E709: jthread_on_current_stack (thread-impl.c:1446)
==12293==    by 0x1B94080D: slowLockMutex (locks.c:182)
==12293==    by 0x1B9C9E05: gcMalloc (thread-internal.h:116)
==12293==    by 0x1B949D1C: utf8ConstNew (utf8const.c:193)
==12293==
==12293== Conditional jump or move depends on uninitialised value(s)
==12293==    at 0x1B940816: slowLockMutex (locks.c:189)
==12293==    by 0x1B9C9E05: gcMalloc (thread-internal.h:116)
==12293==    by 0x1B949D1C: utf8ConstNew (utf8const.c:193)
==12293==    by 0x1B97ED70: Kaffe_FindClass (errors.h:124)
==12293==
==12293== Thread 3:
==12293== Conditional jump or move depends on uninitialised value(s)
==12293==    at 0x1B97E709: jthread_on_current_stack (thread-impl.c:1446)
==12293==    by 0x1B940B75: locks_internal_waitCond (locks.c:318)
==12293==    by 0x1B9C8E6A: gcMan (gc-incremental.c:512)
==12293==    by 0x1B948ED8: startSpecialThread (thread.c:308)
==12293==
==12293== Thread 3:
==12293== Conditional jump or move depends on uninitialised value(s)
==12293==    at 0x1B97E711: jthread_on_current_stack (thread-impl.c:1446)
==12293==    by 0x1B940B75: locks_internal_waitCond (locks.c:318)
==12293==    by 0x1B9C8E6A: gcMan (gc-incremental.c:512)
==12293==    by 0x1B948ED8: startSpecialThread (thread.c:308)
==12293==
==12293== Thread 3:
==12293== Conditional jump or move depends on uninitialised value(s)
==12293==    at 0x1B940726: putHeavyLock (locks.c:154)
==12293==    by 0x1B940CC7: locks_internal_waitCond (locks.c:319)
==12293==    by 0x1B9C8E6A: gcMan (gc-incremental.c:512)
==12293==    by 0x1B948ED8: startSpecialThread (thread.c:308)
==12293==
==12293== Thread 3:
==12293== Conditional jump or move depends on uninitialised value(s)
==12293==    at 0x1B97E709: jthread_on_current_stack (thread-impl.c:1446)
==12293==    by 0x1B94080D: slowLockMutex (locks.c:182)
==12293==    by 0x1B9C9E05: gcMalloc (thread-internal.h:116)
==12293==    by 0x1B93B6C5: jmalloc (gc.c:22)
==12293==
==12293== Thread 3:
==12293== Conditional jump or move depends on uninitialised value(s)
==12293==    at 0x1B940816: slowLockMutex (locks.c:189)
==12293==    by 0x1B9C9E05: gcMalloc (thread-internal.h:116)
==12293==    by 0x1B93B6C5: jmalloc (gc.c:22)
==12293==    by 0x1B947AAD: execute_java_constructor_v (errors.h:124)
--12293-- INTERNAL ERROR: Valgrind received a signal 11 (SIGSEGV) - exiting
--12293-- si_code=1 Fault EIP: 0xB0035465; Faulting address: 0x1C2EDC2B

valgrind: the `impossible' happened:
    Killed by fatal signal [1]

You can pass options to valgrind using the environment variable 
VALGRIND_OPTS. Running vagrind with VALGRIND_OPTS="--tool=helgrind" for 
the data race checker gives me these failures before vagrind blows up:

==32743== warning: Valgrind's pthread_attr_setschedparam does nothing
==32743==          (scheduling not changeable)
==32743==          your program may misbehave as a result
==32743== Possible data race writing variable at 0x8052B6C
==32743==    at 0x1D4EA8CB: linkNativeAndJavaThread (thread.c:94)
==32743==    by 0x1D4EB067: createDaemon (thread.c:376)
==32743==    by 0x1D56C694: gcEnable (gc-incremental.c:1304)
==32743==    by 0x1D4CD48B: initialiseKaffe (baseClasses.c:235)
==32743==  Address 0x8052B6C is not stack'd, malloc'd or (recently) free'd
==32743==  Previous state: exclusively owned by thread 2
==32743== warning: Valgrind's pthread_attr_setschedparam does nothing
==32743==          (scheduling not changeable)
==32743==          your program may misbehave as a result
==32743==
==32743== Thread 2:
==32743== Possible data race writing variable at 0x1D57158C (gc_lock)
==32743==    at 0x1D4E2EDC: locks_internal_lockMutex (locks.c:432)
==32743==    by 0x1D56BE05: gcMalloc (thread-internal.h:116)
==32743==    by 0x1D4DD6C5: jmalloc (gc.c:22)
==32743==    by 0x1D4E9AAD: execute_java_constructor_v (errors.h:124)
==32743==  Address 0x1D57158C is in BSS section of 
/home/topic/Projects/current/jre/lib/i386/libkaffegc-1.1.x-cvs.so
==32743==  Previous state: shared RO, no locks
==32743==
==32743== Thread 2:
==32743== Possible data race writing variable at 0x1D571A14 (gc_heap_lock)
==32743==    at 0x1D4E2EDC: locks_internal_lockMutex (locks.c:432)
==32743==    by 0x1D56D0DC: gc_heap_malloc (thread-internal.h:116)
==32743==    by 0x1D56BE1A: gcMalloc (gc-incremental.c:1044)
==32743==    by 0x1D4DD6C5: jmalloc (gc.c:22)
==32743==  Address 0x1D571A14 is in BSS section of 
/home/topic/Projects/current/jre/lib/i386/libkaffegc-1.1.x-cvs.so
==32743==  Previous state: shared RO, no locks
==32743==
==32743== Possible data race writing variable at 0x8052C5C
==32743==    at 0x1D4EA8CB: linkNativeAndJavaThread (thread.c:94)
==32743==    by 0x1D4EB067: createDaemon (thread.c:376)
==32743==    by 0x1D56C6CC: gcEnable (gc-incremental.c:1307)
==32743==    by 0x1D4CD48B: initialiseKaffe (baseClasses.c:235)
==32743==  Address 0x8052C5C is not stack'd, malloc'd or (recently) free'd
==32743==  Previous state: exclusively owned by thread 3
==32743==
==32743== Thread 3:
==32743== Possible data race writing variable at 0x1DF91CA8
==32743==    at 0x1D4E281E: slowLockMutex (locks.c:198)
==32743==    by 0x1D56BE05: gcMalloc (thread-internal.h:116)
==32743==    by 0x1D4DD6C5: jmalloc (gc.c:22)
==32743==    by 0x1D4E9AAD: execute_java_constructor_v (errors.h:124)
==32743==  Address 0x1DF91CA8 is not stack'd, malloc'd or (recently) free'd
==32743==  Previous state: shared RO, no locks
==32743==
==32743== Thread 3:
==32743== Possible data race writing variable at 0x1DF91CAC
==32743==    at 0x1D4E281E: slowLockMutex (locks.c:198)
==32743==    by 0x1D56BE05: gcMalloc (thread-internal.h:116)
==32743==    by 0x1D4DD6C5: jmalloc (gc.c:22)
==32743==    by 0x1D4E9AAD: execute_java_constructor_v (errors.h:124)
==32743==  Address 0x1DF91CAC is not stack'd, malloc'd or (recently) free'd
==32743==  Previous state: shared RO, no locks
--32743-- INTERNAL ERROR: Valgrind received a signal 11 (SIGSEGV) - exiting
--32743-- si_code=1 Fault EIP: 0xB0035465; Faulting address: 0x1DE8FC2B

Using the addrcheck tool simply blows up valgrind without failures ;)

Do these warnings look useful?

cheers,
dalibor topic

[1] Yeah, Kaffe destroys valgrind after a while. It does the same to 
some kernels, gcc, gdb .. it's simply a good test for any platform's 
(toolchain's) stability ;)