Re: GENERIC-NODEBUG panics, GENERIC does not

On Fri, Apr 15, 2016 at 05:11:54PM +0300, Vladimir Zakharov wrote:
> On Fri, Apr 15, 2016, Konstantin Belousov wrote:
> > On Fri, Apr 15, 2016 at 02:38:18PM +0300, Vladimir Zakharov wrote:
> > > On Fri, Apr 15, 2016, Konstantin Belousov wrote:
> > > > On Fri, Apr 15, 2016 at 01:26:44PM +0300, Vladimir Zakharov wrote:
> > > > > Hello
> > > > > 
> > > > > Setting 'kern.eventtimer.periodic = 1' in /etc/sysctl.conf leads to
> > > > > GENERIC-NODEBUG panic. GENERIC loads without problems.
> > > > > So do GENERIC-NODEBUG as of r297245.
> > > > > 
> > > > > https://imgur.com/0DtDU3m
> > > > > 
> > > > > $ uname -a
> > > > > 
> > > > > FreeBSD vzakharov 11.0-CURRENT FreeBSD 11.0-CURRENT #4 r298037: Fri Apr
> > > > > 15 12:27:40 MSK 2016 root_at_vzakharov:/home/obj/usr/src/sys/GENERIC-NODEBUG  amd64
> > > > > 
> > > > 
> > > > I cannot reproduce it locally.
> > > > Please provide me verbose dmesg from the successfull boot on your machine.
> > > 
> > > GENERIC with kern.eventtimer.periodic=1: http://pastebin.com/raw/TB0TKSje
> > > GENERIC-NODEBUG: http://pastebin.com/raw/05CchraS
> > I need _verbose_ dmesg.  I need the reports about LAPICs initializations,
> > with register dumps.
> 
> dmesg before patching (kern.eventtimer.periodic is not set to avoid
> panic):
> http://pastebin.com/raw/CnJ911gX
> 
> dmesg for patched kernel (kern.eventtimer.periodic = 1):
> http://pastebin.com/raw/9Azz9LxT
Thanks.

> 
> > 
> > > 
> > > > Also, for the panic, show me the output of the same data as on the
> > > > screenshot and the output of 'show registers'.
> > > 
> > > https://imgur.com/pO8Yc3w
> > Ok, this seems to be yet another case of missed count mode divisor
> > initialization. Please try the patch at the end of the messasge.
> 
> Patched kernel doesn't panic anymore. Thanks.
Committed as r298056.

> 
> > > > Also please load your kernel into kgdb and do
> > > > list *lapic_et_start+0x281
> > By 'your kernel' I mean 'your panicing kernel'.
> It _was_ panicing kernel. Than not sure what you meant. There is no any
> dump/core files because dumpdev is not configured yet at the moment of
> panic. What should I do to retrieve the required information?
> I've simply run: $ kgdb /boot/kernel/kernel

No, core is not needed, only kernel to translate faulted address into
the source line.  The translation that you show did not make sense.
The registers value was more useful.