An unexpected perf feature

Posted May 21, 2013 23:39 UTC (Tue) by deater (subscriber, #11746)
Parent article: An unexpected perf feature

> Kernels configured without support for perf events (i.e.
> CONFIG_PERF_EVENTS not set) are unaffected by the bug as they lack
> the system call entirely.

As far as I know it's not possible to disable perf_event on x86 since about 2.6.37 or so, because it is automatically enabled to get debugger support. I'd be glad to be proven wrong though.

As far as trinity support, better perf_event_open() coverage that I contributed was merged today, so everyone can fuzz away.

to post comments

An unexpected perf feature

Posted May 21, 2013 23:43 UTC (Tue) by deater (subscriber, #11746) [Link]

> As far as trinity support, better perf_event_open() coverage that I
> contributed was merged today, so everyone can fuzz away.

Also, the new trinity support isn't as complete as it could be. Check out my perf_event_open() manpage linked above. The perf_event system call is fantastically complex with over 40 inter-dependent arguments that interact in complex ways.

I still sometimes wish that a thinner perf counter interface was merged. perfctr and perfmon2 both were much thinner wrappers over the perf counter MSRs. The "put everything in the kernel" strategy for perf_event makes it that hard to validate correctness.