SIGABORT on linux-sparc

[Sébastien Bernard]

Andrew Haley a écrit :
> Sébastien Bernard wrote:
>> Andrew Haley a écrit :
>>> Sébastien Bernard wrote:
>>>  
>>>> Gary Benson a écrit :
>>>>    
>>>>> S�bastien Bernard wrote:
>>>>>  
>>>>>      
>>>>>> The sigbus happens in
>>>>>> icedtea6/openjdk-ecj/hotspot/src/cpu/zero/vm/cppInterpreter_zero.cpp
>>>>>> line 356.
>>>>>>
>>>>>> The line is :
>>>>>>
>>>>>>    case T_LONG:      SET_LOCALS_LONG(*(jlong *) result, 0);
>>>>>>      break;
>>>>>>
>>>>>> The SET_LOCALS_LONG is breaking the access.
>>>>>> The value of result according to gdb is  (long long int *) 0xf781c420
>>>>>>             
>>>>> That looks pretty aligned to me.
>>>>> What does gdb say the variable "locals" is at that point?
>>>>>         
>>>> #7  0xf7a96a9c in CppInterpreter::native_entry (method=0xf1645128,
>>>> __the_thread__=0x2ecb0)
>>>>    at
>>>> /home/seb/dev/icedtea6/openjdk-ecj/hotspot/src/cpu/zero/vm/cppInterpreter_zero.cpp:356
>>>>
>>>>
>>>> 356           SET_LOCALS_LONG(*(jlong *) result, 0);
>>>> (gdb) p (void*) locals
>>>> $1 = (void *) 0xf789aa58
>>>> (gdb)
>>>>     
>>> That's pretty aligned too.  Try
>>>
>>> x/i $pc
>>>
>>> and then tell us what the contents of the address register is.
>>>
>>> Andrew.
>>>   
>> Maybe an irc channel would be better suited for that kind of debugging.
> 
> We're on #openjdk at irc.oftc.net
> 
>> Anyway, here's the info
>> (gdb) up
>> #7  0xf7a96a9c in CppInterpreter::native_entry (method=0xf1645128,
>> __the_thread__=0x2ecb0)
>>    at
>> /home/seb/dev/icedtea6/openjdk-ecj/hotspot/src/cpu/zero/vm/cppInterpreter_zero.cpp:356
>>
>> 356           SET_LOCALS_LONG(*(jlong *) result, 0);
>> (gdb) x/i $pc
>> 0xf7a96a9c
>> <_ZN14CppInterpreter12native_entryEP13methodOopDescP6Thread+1456>:  
>> std  %g2, [ %g1 + -4 ]
> 
> Eww, nasty.  
> 
> p $g1 ?  
> 
> Andrew.

(gdb) x locals
0xf789aa58:     0xf789aaa4
(gdb) p/x $g1
$3 = 0xf789aa58
(gdb) x $g1
0xf789aa58:     Cannot access memory at address 0xf789aa58

	Seb