[Sablevm-developer] IA64 saga continues...

[Grzegorz B. P. <ga...@de...>]

Hi!

I am trying to get SableVM working on UP IA64 machine with inlined
threading engine. Here's what I have now.

I tried http://www-sor.inria.fr/~piumarta/pldi98/vm.c and it worked
well w/o any assembly. The ouput was almost the same as on my i386.
I ran it a few times, but nothing bad happened, so it didn't seem
too helpful.

I rebuilded sablevm and sablevm-native packages with these options:
--enable-signals-for-exceptions=no
--with-threading=inlined
--enable-debugging-features

Running it normally simply throwns "Segmentation fault", then I had:

gadek@itanic:~$ /usr/bin/sablevm -Y -v -p sablevm.verbose.methods=true
-p sablevm.verbose.instructions=true --classpath=./HelloWorld.jar:
HelloWorld
[verbose jni: JNI_CreateJavaVM]
[verbose gc: allocating initial heap (16777216 bytes)]
[verbose class: loading "java/lang/Object"]
[verbose class: loading "java/io/Serializable"]
[verbose class: loading "java/lang/Cloneable"]
[verbose class: creating "[B"]
[verbose class: loading "java/lang/Class"]
[verbose methods: entering method java/lang/Class.<init>([B)V]
[verbose instructions: executing @0x6000000000010f70 PREPARE_METHOD]
Segmentation fault

Tried GDB and my session looked like this:
gadek@itanic:~$ gdb /usr/bin/sablevm
(gdb) set args -Y -v -p sablevm.verbose.methods=true -p
sablevm.verbose.instructions=true --classpath=./HelloWorld.jar:
HelloWorld
(gdb) run
Starting program: /usr/bin/sablevm -Y -v -p sablevm.verbose.methods=true
-p sablevm.verbose.instructions=true --classpath=./HelloWorld.jar:
HelloWorld
[New Thread 1024 (LWP 1197)]
[verbose jni: JNI_CreateJavaVM]
[verbose gc: allocating initial heap (16777216 bytes)]
[verbose class: loading "java/lang/Object"]
[verbose class: loading "java/io/Serializable"]
[verbose class: loading "java/lang/Cloneable"]
[verbose class: creating "[B"]
[verbose class: loading "java/lang/Class"]
[verbose methods: entering method java/lang/Class.<init>([B)V]
[verbose instructions: executing @0x6000000000010f70 PREPARE_METHOD]

Program received signal SIGSEGV, Segmentation fault.
[Switching to Thread 1024 (LWP 1197)]
0x6000000000022a70 in ?? ()
(gdb) bt
#0  0x6000000000022a70 in ?? ()
#1  0x2000000000263070 in _svmf_interpreter (_env=0x600000000000d3e0)
    at instructions_preparation_inlined_threaded.c:17527
#2  0x20000000000a4760 in _svmf_new_class (env=0x3d73646f6874656d,
type=0x65757274) at util2.c:820

The 820th line is this:
  if (_svmm_invoke_nonvirtual_jlclass_init
      (env, class_instance, _svmf_cast_jobject (data)) != JNI_OK) {

The interesting thing is that it behaves exactly the same no matter
if in _svmf_iflush there IS this line:
__asm__ __volatile__ ("fc %0;; sync.i;; srlz.i;;" :: "r"(pword));
or it's commented out !

So either this flushing doesn't work OR we have another problem.
For example running with --debugging-features enabled gives tons
of warnings! I am attaching gzipped log. Maybe there's sth unusual
in it that causes the problems?

I'd like to hear any suggestions.

					Grzegorz B. Prokopski

PS: In the meantime I am gonna look for instructions that flush entire
data and code caches (if there are such instructions for IA64).
-- 
Grzegorz B. Prokopski <ga...@de...>
Debian http://www.debian.org/