Re: [PATCH 2/3] perf symbol: Implement a very simple ELF symbol parser

[Namhyung Kim]

On Mon, 10 Nov 2014 09:09:04 -0300, Arnaldo Carvalho de Melo wrote:
> Em Mon, Nov 10, 2014 at 03:36:48PM +0900, Namhyung Kim escreveu:
>> Hi Arnaldo,
>> 
>> On Fri, 7 Nov 2014 12:26:18 -0300, Arnaldo Carvalho de Melo wrote:
>> > Em Fri, Nov 07, 2014 at 02:20:05PM +0900, Namhyung Kim escreveu:
>> >> It'll be used to show (userspace) symbol names when libelf isn't (or
>> >> cannot be) linked.
>> >
>> > Does this deals with prelink, etc?
>> 
>> I believe so. :)
>> 
>> >  
>> >>   # Overhead  Command     Shared Object      Symbol
>> >>   # ........  ..........  .................  .........................
>> >>   #
>> >>       37.01%  mem-memcpy  libc-2.17.so       [.] __memcpy_ssse3_back
>> 
>>   namhyung@sejong:linux$ readelf -WS /lib64/libc-2.17.so | grep prelink
>>     [41] .gnu.prelink_undo PROGBITS        0000000000000000 200368 000c30 01      0   0  8
>> 
>>   namhyung@sejong:linux$ nm /lib64/libc-2.17.so | grep __memcpy_ssse3_back
>>   0000003153f46f40 t __memcpy_ssse3_back
>
> Right, in this case most of the samples seems to map to what is expected
> for that workload, and the binary was prelinked. Good.
>
> What about binaries that are not prelinked? IIRC there is code in the
> full blown ELF symbol-elf.c file to detect that and act accordingly,
> from a _very_ quick look I didn't saw it in this minimalistic ELF symtab
> reader, hence my question.

AFAIK we deal with prelink'ed binary and normal binary as same way - we
only cares about the file offsets.


  $ perf top --stdio
     PerfTop:     110 irqs/sec  kernel:70.0%  exact:  0.0% [4000Hz cycles],  (all, 12 CPUs)
  --------------------------------------------------------------------------------------------
     9.42%  perf           [.] map__process_kallsym_symbol       
     8.91%  [kernel]       [k] kallsyms_expand_symbol.constprop.1
     6.80%  [kernel]       [k] memcpy                            
     6.08%  [kernel]       [k] vsnprintf                         
     ...


Thanks,
Namhyung
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/