Re: was: CURRENT [r308087] still crashing: Backtrace provided

From: O. Hartmann <ohartmann_at_walstatt.org>
Date: Sun, 13 Nov 2016 11:24:43 +0100
Am Sun, 13 Nov 2016 12:41:07 +0300
"Andrey V. Elsukov" <ae_at_FreeBSD.org> schrieb:

> On 13.11.2016 12:23, O. Hartmann wrote:
> >>>>> Great, thank you. I would first like to confirm that r307234 is indeed
> >>>>> causing the crash - since it appears to be easy to trigger, that should
> >>>>> be faster. If not, the core will help track down the real problem.    
> >>>>
> >>>> Although I was under the impression the in-kernel-config option
> >>>>
> >>>> makeoptions    DEBUG=-g
> >>>>
> >>>> would make debugging symbols available, I'm proved wrong.    
> >>
> >> Do you have option FLOWTABLE in your kernel config?  
> > 
> > Would you suggest to disable this feature in the kernel? Or does the feature, by
> > accident, influence the debugging ??  
> 
> Hi,
> 
> I never used FLOWTABLE, but as I know, when L2 caching was reintroduced
> the kernel with enabled FLOWTABLE has started crashing almost
> immediately. glebius_at_ added workaround in r300854 to prevent the panic.
> Then we discussed with him that the change in in_pcb.c should be
> reasonable. And he decided to revert the workaround. Now it seems
> without r300854 FLOWTABLE isn't usable.
> 

All right. At this very moment I compile a most recent world and kernel WITHOUT the
option FLOWTABLE. It takes a lot(!) of time to revert the whole system to the working
r307233 again (we figured out, that r307234 triggers the crashes so far).

The interesting thing to me is that the crash occured on all platforms/Intel architectures
I use and maintain immediately after r307233 and it lasted for at least r307884
(approx.), that was, when I reverted everything to r307157 (the point from which I
started non-crahsing). Now, I have several relatively modern systems (norebooks with
Haswell Mobile CPU, XEON Haswell Workstation and Skylake XEON Servers, AMD Jaguar based
PCEngines APU 2C4) with FLOWTABLE enabled and with most recent CURRENT - without a crash
in days under heavy load (network I/O) and poudriere builds - without issues. But
everything I have in my hands with IvyBridge (XEON, consumer grade CPUs) and lower (one
Intel Core2Duo 2-socket server) is crashing with CURRENT beyond r307233. Maybe I oversee
the different internal cash architectures and probably optimization problem.

Anyway, this just to repeat for the records ;-)

I'll give notice after my boxes have been compiled to most recent CURRENT with
FLOWTABLES.

Thanks and kinf regards,

Oliver 

-- 
O. Hartmann

Ich widerspreche der Nutzung oder Übermittlung meiner Daten für
Werbezwecke oder für die Markt- oder Meinungsforschung (§ 28 Abs. 4 BDSG).

Received on Sun Nov 13 2016 - 09:25:01 UTC

This archive was generated by hypermail 2.4.0 : Wed May 19 2021 - 11:41:08 UTC