more perf breakage from sample identifier

more perf breakage from sample identifier

[David Ahern]

Adrian:

'perf kvm stat live' command fails on Linus' latest tree:
$ perf kvm stat live
Failed to parse sample

git bisect points to:

75562573bab35b129cfd342fc2bcf89da84a6644 is the first bad commit
commit 75562573bab35b129cfd342fc2bcf89da84a6644
Author: Adrian Hunter <adrian.hunter@intel.com>
Date:   Tue Aug 27 11:23:09 2013 +0300

     perf tools: Add support for PERF_SAMPLE_IDENTIFIER

     Enable parsing of samples with sample format bit 
PERF_SAMPLE_IDENTIFIER.
     In addition, if the kernel supports it, prefer it to selecting
     PERF_SAMPLE_ID thereby allowing non-matching sample types.

     Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>


perf_evlist__event2evsel() is failing. From what I can see the problem 
is that evlist->heads is not getting populated. This command does not 
read/write files but processes events as they come in. I have another 
similar command that is also breaking for the same reason.

David

Re: more perf breakage from sample identifier

[David Ahern]

On 9/5/13 3:35 PM, David Ahern wrote:
> Adrian:
>
> 'perf kvm stat live' command fails on Linus' latest tree:
> $ perf kvm stat live
> Failed to parse sample
>
> git bisect points to:
>
> 75562573bab35b129cfd342fc2bcf89da84a6644 is the first bad commit
> commit 75562573bab35b129cfd342fc2bcf89da84a6644
> Author: Adrian Hunter <adrian.hunter@intel.com>
> Date:   Tue Aug 27 11:23:09 2013 +0300
>
>      perf tools: Add support for PERF_SAMPLE_IDENTIFIER
>
>      Enable parsing of samples with sample format bit
> PERF_SAMPLE_IDENTIFIER.
>      In addition, if the kernel supports it, prefer it to selecting
>      PERF_SAMPLE_ID thereby allowing non-matching sample types.
>
>      Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
>
>
> perf_evlist__event2evsel() is failing. From what I can see the problem
> is that evlist->heads is not getting populated. This command does not
> read/write files but processes events as they come in. I have another
> similar command that is also breaking for the same reason.

That's not the problem. The hash is getting populated. The problem is a 
bad assumption on what fields are in the sample -- like PERF_SAMPLE_IP. 
The kvm command does not care about the IP so it is excluded:

attr->sample_type &= ~PERF_SAMPLE_IP;

If I remove that line the command works fine.

Since this a legitimate option to not get the IP in the sample, the 
evsel lookup code needs to get fixed in 3.12.

David