Linux 2.6 emulation and Linux Java problem

[Peter Kostouros]

Roman Divacky wrote:
>> I used the normal kdump; unfortunately I could not install linux_kdump 
>> and did not persevere with it ("does not build with the default linux 
>> base, use the package instead" and I had a hiccup installing the 
>> package, too). I will look into dtrace over the weekend.
>>     
>  
> try the linux_kdump... the dtrace might have problems on its own as it
> is really new code...
>
>   
OK, I installed linux_kdump.

>>>> 1860 java RET  open 97/0x61
>>>>
>>>> 1860 java CALL freebsd6_mmap(0x61, 0x2b639970, PROT_EXEC, MAP_FILE, 
>>>> 0xa5a5a5a5, ..., 0xa5a5a5a5, 0, ..., 0, 0xc, 0xdead0002, ...
>>>>
>>>>
>>>> Note
>>>>
>>>> 1. These applications ran successfully with linux_base-fc4 and 
>>>> compat.linux.osrelease set to 2.4.2;
>>>> 2. The success of running java applications also depends on 
>>>> debug.witness.watch: I get more mileage from java applications when this 
>>>> sysctl is 0.
>>>>    
>>>>         
>>> Do you get witness warning/errors on the console? Please check and
>>> report them if there are any.
>>>       
>
> a LOR could lead to a deadlock not a crash.... if you still have the ktrace.out
> I believe you can put it on a web and we can run our own linux_kdump on it to
> see what happened. can you make that available?
>
>   
ktrace.out is about 500MB.

Anyways, one java process looks to be looping:

RET  linux_sys_futex -1 errno 110 Unknown error: 110

CALL linux_sys_futex(0x8092528,0x81,0x1,0xfffffffd,0x8092528,0x2affd250)

RET  linux_sys_futex 1

CALL linux_clock_gettime(0x1,0x2affd290)

RET  linux_clock_gettime 0

CALL gettimeofday(0x2affd2a8,0)

RET  gettimeofday 0

CALL linux_clock_gettime(0x1,0x2affd290)

RET  linux_clock_gettime 0

CALL linux_clock_gettime(0x1,0x2affd290)

RET  linux_clock_gettime 0

CALL gettimeofday(0x2affd240,0)

RET  gettimeofday 0

CALL linux_clock_gettime(0x1,0x2affd21c)


RET  linux_clock_gettime 0

CALL linux_sys_futex(0x80e0acc,0x80,0x1,0x2affd21c,0x1,0x2affd280)


and another over sched_yield (although I am uncertain as I have not 
examined the entire dump).

Let me know if you want anything from the trace.