Re: callchain sampling bug in perf?

[Frederic Weisbecker]

On Sun, Aug 15, 2010 at 06:53:59PM -0400, Christoph Hellwig wrote:
> I'm trying to play with perf record -g, that is sampling the callchains.
> Doing this with simple workloads works fine, but when I try to do this
> with compilebench a lot of samples seem to get lost.
> 
> I'm doing the following:
> 
> perf probe --add xlog_sync
> perf record -g -e probe:xlog_sync -- ./compilebench
> 
> Trying to report it I do not get any callchains at all:
> 
> [root@virtlab106 compilebench-0.6]# perf report -g flat -n
> # Events: 9K cycles
> #
> # Overhead  Samples         Command      Shared Object     Symbol
> # ........ ..........  ............  .................  .........
> #
>     70.41%       6757  compilebench  [kernel.kallsyms]  [k] xlog_sync
>     22.61%       2170          sync  [kernel.kallsyms]  [k] xlog_sync
>      3.89%        373            sh  [kernel.kallsyms]  [k] xlog_sync
>      2.50%        240        python  [kernel.kallsyms]  [k] xlog_sync
>      0.33%         32         :3881  [kernel.kallsyms]  [k] xlog_sync
>      0.13%         12         :3971  [kernel.kallsyms]  [k] xlog_sync
>      0.11%         11         :3956  [kernel.kallsyms]  [k] xlog_sync
>      0.01%          1         :3972  [kernel.kallsyms]  [k] xlog_sync
> 
> Doing a perf report -g flat,0.0 -n shows lots of callgraph, but the
> percentag for them doesn't add up at all.


Hmm, against which tree are you running? I've just tested callchains
record on random kprobe based tracepoint and it seemed to work well
on report.

Could you send me your perf.data and also the resulting archive
after the following command: "perf archive" ?

Thanks.

--
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/