[Ptools-perfapi] kernel 3.0, papi_overflow and exec

Vince Weaver vweaver1 at eecs.utk.edu
Tue Aug 16 16:27:19 EDT 2011


Thanks for the bug report.

I've noticed that the overflow-related PAPI regression tests fail on 3.0 
as well.  I was working with the kernel devs pre-3.0 on some overflow 
handling stuff, but the changes they made just made things worse.  Then 
they stopped working on it for a while and I forgot to do more testing 
before the release.

> Here's how to reproduce the problem.  This is with kernel 3.0.0,
> perf_events and PAPI 4.1.3 on an AMD x86-64.  My PAPI overflow tests
> are available via anon svn checkout from:
> 
>    svn co http://svn.rice.edu/r/krentel/pub/papi-tests
> 
> The 'exec' test demonstrates the problem.  It starts PAPI_overflow()
> and calls exec().  The child should be able to run on its own, but
> instead it dies on "Real-time signal 2".

I'll try this and also see if I can get a small reproducible test to send 
the kernel devs.  Hopefully this doesn't require a bisect as that takes 
forever :(

Vince



More information about the Ptools-perfapi mailing list