On 29/11/17(Wed) 12:13, Jason Tubnor wrote:
> On 28 November 2017 at 20:33, Martin Pieuchot <[email protected]> wrote:
> 
> >
> > Could you do:
> >
> > # sysctl kern.splassert=2
> >
> > And report the tracebacks?
> >
> 
> Hi Martin,
> 
> Please find attached a compressed /var/log/messages file.  This only
> contains data gathered when conducting the iperf tests.  Please advise if
> there are any further tests or data you need.

Thanks, the problem is that vxlan(4)+HFSC is currently broken as it
doesn't grab the NET_LOCK():

  splassert: ip_output: want 2 have 0
  Starting stack trace...
  ip_output(ffff8000005024b8,2,0,ffffff0036bd1f00,ffff800000502000,cae9f2e3efe2f
0bc) at ip_output+0x79
  vxlan_output(ffff800000502230,ffff800000502000) at vxlan_output+0x2f1
  vxlanstart(4) at vxlanstart+0x2b
  if_qstart_compat(ffff8000005022a8) at if_qstart_compat+0x33
  ifq_serialize(ffffffff81b17700,ffff800000502230) at ifq_serialize+0xe7
  hfsc_deferred(ffffffff81b17700) at hfsc_deferred+0x42
  softclock(ffffffff81aac260) at softclock+0xd6
  softintr_dispatch(ffff80000001a8a0) at softintr_dispatch+0xcb
  Xsoftclock() at Xsoftclock+0x1f
  --- interrupt ---

I guess the correct solution would be to use a task in vxlanstart()...

Reply via email to