Messages in this thread | | | Date | Mon, 11 Nov 2013 22:13:45 +0100 | From | Ingo Molnar <> | Subject | Re: perf/tracepoint: another fuzzer generated lockup |
| |
* Peter Zijlstra <peterz@infradead.org> wrote:
> On Mon, Nov 11, 2013 at 01:44:19PM +0100, Ingo Molnar wrote: > > > > * Frederic Weisbecker <fweisbec@gmail.com> wrote: > > > > > > That said, I'm not sure what kernel you're running, but there were > > > > some issues with time-keeping hereabouts, but more importantly that > > > > second timing includes the printk() call of the first -- so that's > > > > always going to be fucked. > > > > > > It's a recent tip:master. So the delta debug printout is certainly > > > buggy, meanwhile these lockup only happen with Vince selftests, and they > > > trigger a lot of these NMI-too-long issues, or may be that's the other > > > way round :)... > > > > > > I'm trying to narrow down the issue, lets hope the lockup is not > > > actually due to printk itself. > > > > I'd _very_ strongly suggest to not include the printk() overhead in the > > execution time delta! What that function wants to report is pure NMI > > execution overhead, not problem reporting overhead. > > > > That way any large number reported there is always a bug somewhere, > > somehow. > > -ENOPATCH :-) > > You'll find that there's two levels of measuring NMI latency and the > outer will invariably include the reporting of the inner one; fixing > that is going to be hideously ugly. > > That said, I would very strongly suggest to tear that printk() from the > NMI path, its just waiting to wreck someone's machine :-)
So why not just write the value somewhere and printk once at the end of the NMI sequence, once everything is said and done?
Thanks,
Ingo
| |