review (S) for 6970683: improvements to hs_err output

[Vladimir Kozlov]

Looks good.

One thing I notice is os_solaris_x86.cpp doesn't have' REG_' in reg names in 32-bit case but os_linux_x86.cpp has it. Why such difference?

Thanks,
Vladimir

Tom Rodriguez wrote:
> On Oct 12, 2010, at 11:04 AM, Vladimir Kozlov wrote:
> 
>> Tom Rodriguez wrote:
>>>> src/share/vm/runtime/os.cpp next values are the same why print twice?
>>>>
>>>> +       st->print_cr(INTPTR_FORMAT " is the thread: "
>>>> +                    INTPTR_FORMAT, addr, thread);
>>> I wanted to print something shorter since we dump all the threads later but I didn't closely enough about what would come out.  How about this:
>>> -      thread->print_on(st);                                                                                                                         +      if (verbose) {                                                                                                                                 +        thread->print_on(st);                                                                                                                       +      } else {                                                                                                                                       +        static char buf[256];                                                                                                                       +        thread->print_on_error(buf, sizeof(buf));                                                                                                   +        st->print_cr(INTPTR_FORMAT " is %s", addr, buf);                                 
                                                            +      }   This will print out in the same one line format that the rest of VMError uses.
>> It is too many lines (two :) ). I would just say it is thread, we can look it down in hs_err for more info.
>>
>> -      thread->print_on(st);
>> +      if (verbose) {
>> +        thread->print_on(st);
>> +      } else {
>> +       st->print_cr(INTPTR_FORMAT " is the thread", addr);
>> +      }
> 
>         st->print_cr(INTPTR_FORMAT " is a thread", addr);
> 
>>>> src/share/vm/code/codeCache.cpp closed ")" should be "]" (or it is indication that address in not included?):
>>>>
>>>> +   st->print_cr("Code Cache  [" INTPTR_FORMAT ", " INTPTR_FORMAT ", " INTPTR_FORMAT ")",
>>> It's the same format we use for the other sections and yes the final address isn't included.  It's the half open interval.
>> Then it is fine.
>>
>>>> also why you need to put ' ' around numbers for codecache values? Can you separate them using comma?
>>> It was a copy paste from the log sweeper printing which was xml.  I've removed them.  Why do you want commas?
>>> Code Cache  [0xfa800000, 0xfaa40000, 0xfd800000)
>>> total_blobs=398 nmethods=299 adapters=60 free_code_cache=49158848
>>> vs.
>>> Code Cache  [0xfa800000, 0xfaa40000, 0xfd800000)
>>> total_blobs=398, nmethods=299, adapters=60, free_code_cache=49158848
>> I am fine without commas (less file size).
> 
> Ok.
> 
>>>> src/os_cpu/windows_x86/vm/os_windows_x86.cpp missing #ifdef AMD64 in os::print_register_info() and end of method:
>>>> st->cr();
>>>> }
>>>>
>>>> Also why you left Register to memory dump the same? On solaris platforms you print only register name before call to print_location() (I prefer this).
>>> Oops.  I thought I'd finished the windows side.  It's done now, though it hasn't been through jprt yet.
>> I don't see that you updated print_register_info(), it is still original (several lines per reg).
> 
> I've updated the webrev.  There were also extra _cr's in the os_linux_x86 code that I removed.
> 
> tom
> 
>> Thanks,
>> Vladimir
>>
>>> tom
>>>> Vladimir
>>>>
>>>> Tom Rodriguez wrote:
>>>>> http://cr.openjdk.java.net/~never/6970683
>>>>> 6970683: improvements to hs_err output
>>>>> Reviewed-by:
>>>>> There are a few things missing from the hs_err dump that would be
>>>>> useful.  First we don't dump the sparc L and I registers.  Second some
>>>>> information about the size and contents of the code cache would be
>>>>> useful.  Third we should dump a larger region around the faulting
>>>>> instruction.  Additionally the new register to memory mapping output
>>>>> can crash which stops us from getting the stack and instructions at
>>>>> the faulting pc, so I moved it into it's own section.  block_start
>>>>> would assert in some cases so I augmented existing logic to just
>>>>> return null.  I also changed the formatting to remove all the extra
>>>>> whitespace and made some of the output more compact and eliminated
>>>>> most of the useless whitespace.
>