Re: [perf] overflow/perf_count_sw_cpu_clock crashes recent kernels

[Lin Ming <ming.m.lin@intel.com>]

On Thu, 2011-07-28 at 09:54 +0800, Lin Ming wrote:
> On Thu, Jul 28, 2011 at 2:51 AM, Vince Weaver <vweaver1@eecs.utk.edu> wrote:
> > Hello
> >
> >> With 3.0.0 the PAPI "overflow_allcounters" test reliably locks up my
> >> Nehalem system.
> >
> > I finally managed to narrow this down to a small test, which is attached.
> >
> > Basically measuring overflow on the perf::perf_count_sw_cpu_clock
> > event will potentially *lock up* your system from user-space.
> >
> > This seems to be a long standing bug.  It will quickly lock solid
> > my Nehalem test box on 3.0, 2.6.39 and 2.6.38.
> >
> > On a Core2 2.6.32 box the crash testing program will wedge and become
> > unkillable, but it doesn't actually kill the machine.

Hi, Vince

I re-tested current -tip tree(commit 0931941) and it works OK now.

$ ./oflo_sw_cpu_clock_crash 
Matrix multiply sum: s=27665734022509.746094
Total overflows: 3460

Could you also have a try the current -tip tree?

Lin Ming