9.2-RC4 amd64 panic: vm_page_unwire

[John Marshall]

On Mon, 30 Sep 2013, 09:24 +0200, Oliver Pinter wrote:
> On 9/29/13, John Marshall <john.marshall at riverwillow.com.au> wrote:
> > On Sat, 28 Sep 2013, 17:27 +0200, Oliver Pinter wrote:
> >> On 9/28/13, John Marshall <john.marshall at riverwillow.com.au> wrote:
> >> > On Fri, 27 Sep 2013, 11:12 +0300, Konstantin Belousov wrote:
> >> >> On Fri, Sep 27, 2013 at 10:07:28AM +1000, John Marshall wrote:
> >> >> > I have made the core.txt.[0-2] files available in the following
> >> >> > directory.  The directory is not browsable.
> >> >> >
> >> >> >   http://www.riverwillow.net.au/~john/92rc4/
> >> >>
> >> >> This might be fixed by r254087-r254090 on stable/9.

> >> try this from 9-STABLE:
> >> https://github.com/freebsd/freebsd/commit/98362329be8e2cdefda1988e5e9d32efbc94855f
> >
> > Thanks Oliver.  If I followed the links correctly, that looks like
> > r255507 on stable/9.  The second part of that patch didn't apply cleanly
> > to 9.2.  I investigated and it looks like that is because r254442 had
> > also been applied since 9.2.  I'm guessing I should also apply r254442
> > as well (first)?  I shall proceed with applying both patches, rebuild,
> > and see what happens.
> 
> This patch affected only those kernels, that has cpuctl. When no cpuctl loaded,
> has no effect.

OK.  So including it shouldn't have hurt anything.

So, with 9.2-RELEASE (releng/9.2 r255904) with patches...
 + r254087 /stable/9/sys/vm/vm_page.c
 + r254088 /stable/9/sys/vm/vm_object.c
 + r254088 /stable/9/sys/vm/vm_object.h
 + r254089 /stable/9/sys/vm/vm_map.c
 + r254089 /stable/9/sys/vm/vm_map.h
 + r254090 /stable/9/sys/vm/vm_fault.c
 + r254442 /stable/9/sys/dev/cpuctl/cpuctl.c (apparently not needed)
 + r255507 /stable/9/sys/dev/cpuctl/cpuctl.c

...I am still seeing vm_page_unwire panics.

I noticed that in all cases EXCEPT the sftp-server case, either ntpd
exiting or watchdogd exiting seem to be the trigger for this
vm_page_unwire panic, so I did some experiments.  All core.txt files
are available via the same address as earlier ones (referenced above).

core.txt.5
 - watchdogd AND ntpd running
 - panic during shutdown (ntpd)

core.txt.6
 - watchdogd AND ntpd running
 - panic on 'service ntpd stop'

core.txt.7
 - watchdogd running (ntpd NOT running)
 - panic during shutdown (watchdogd)

core.txt.8
 - ntpd running (watchdogd NOT running)
 - panic during shutdown (ntpd)

core.txt.9
 - watchdogd AND ntpd started during startup
 - I was able to stop ntpd without mishap within the first 2 minutes
 - panic during shutdown (watchdogd)

core.txt.10
 - watchdogd AND ntpd running
 - panic on 'service watchdogd stop'

With both watchdogd AND ntpd running, stopping either deamon will
trigger the panic.

With watchdogd XOR ntpd running, the daemon can be stopped and started
without mishap, but if it is running during shutdown, the system will
panic when trying to stop that daemon.

Another data point: both systems on which I have seen this panic have
the ipmi driver compiled in.  ipmi makes the BMC's watchdog timer
available to the system.

  ozsrv04# grep ^ipmi /var/run/dmesg.boot
  ipmi0: <IPMI System Interface> on isa0
  ipmi0: KCS mode found at io 0xca2 alignment 0x1 on isa
  ipmi0: IPMI device rev. 1, firmware rev. 2.43, version 2.0
  ipmi0: Number of channels 2
  ipmi0: Attached watchdog

-- 
John Marshall
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 196 bytes
Desc: not available
URL: <http://lists.freebsd.org/pipermail/freebsd-stable/attachments/20130930/93d63715/attachment.sig>