Re: [Xen-users] xl devd won't start in DomU: failed to get own domid

[Rüdiger Otte <ruediger.otte@xxxxxxx>]

On Fri, 6 May 2016 20:41:06 +0200
Roger Pau Monné wrote:

> On Sat, Apr 30, 2016 at 11:09:27PM +0200, Rüdiger Otte wrote:
> > On Mon, 25 Apr 2016 11:11:47 +0200
> > Roger Pau Monné wrote:
> > 
> > > On Sun, Apr 24, 2016 at 01:41:34PM +0200, Rüdiger Otte wrote:
> > > > Hi,
> > > > 
> > > > I'm successfully running my debian system with Xen 4.4 for
> > > > more than two years now. When trying to upgrade to the recent
> > > > 4.6 release I ran into a problem with the network driver
> > > > domain.
> > > > 
> > > > As Xen 4.6 requires devd to run in a driver domain, I tried to
> > > > start it there but it fails:
> > > > 
> > > > root@svc-fw:~# xl devd -F
> > > > libxl: error: libxl.c:3864:libxl_device_events_handler:
> > > > unable to get domain id
> > > > 
> > > > Strange thing is that 'xenstore-read domid' outputs the domid
> > > > without any problems.
> > > 
> > > It is indeed quite weird, xl does exactly the same as
> > > 'xenstore-read domid' AFAICT.
> > > 
> > > Can you paste the configuration file of your driver domain?
> > > Also, do you have driver_domain=1 in your configuration file?
> > > (although AFAICT the error for not having driver_domain=1
> > > should be different than the one you are getting).
> > 
> > I hadn't driver_domain=1 in the configuration, but adding it
> > didn't changed anything:
> > 
> > root@svc-fw:~# xl devd -F
> > libxl: error: libxl.c:1972:libxl__get_domid: failed to get own
> > domid (domid)
> > libxl: error: libxl.c:4553:libxl_device_events_handler: unable to
> > get domain id
> > 
> > 
> > xl.cfg:
> > 
> > name = 'svc-fw'
> > uuid = 'fc2c7c06-c301-4f6d-9862-d7357eacccda'
> > disk = [ '/xen/disk-image/svc-fw-debian.img,raw,hda,w' , 
> >        ]
> > bootloader = 'pygrub'
> > extra = 'iommu=soft'
> > cpus = "0-1"
> > vcpus = 1
> > cpu_weight = 64
> > memory = 256
> > device_model_version = 'qemu-xen'
> > device_model_override = '/usr/bin/qemu-system-x86_64'
> > on_poweroff = 'destroy'
> > on_reboot = 'restart'
> > on_crash = 'destroy'
> > driver_domain = 1
> > pci = [ '02:00.0' ]
> > 
> > 
> > The domain is running with Debian Wheezy with Xen packages and
> > dependencies from Jessie / Stretch.
> 
> TBH, I don't see anything wrong with your config.
> 
> Could you try to run `xl -vvv devd -F` in order to see if there are
> more messages that might help clarify what's going on?
> 
> Roger. 

Hi,

thanks for your answers. I already tried that but it didn't give more
than the regular debug output. But I also did some more investigation
and finally found the culprit. It was me, having modified Xen
initscripts in the driver domain :)

Background: package configuration of xen-utils is relying on a successful
start of /etc/init.d/xen. The script itself does a check if it is
running in Dom0 and dropping an error otherwise. So I simply
commented out that check and package configuration completed.

What also happened: after the modification the initscript ignored the
fact of running in DomU and always started a local instance of
xenstored. For some reason xenstore-utils don't have a problem with
that while xl devd has.

After having disabled the local xenstored, xl devd started and other
domains vif's can connect to the driver domain again. Unfortunately
this is isn't true for Dom0:

root@dox64:~# xl network-attach 0 type=vif bridge=br-dm0 backend=svc-fw 
rate=500Mb/s
libxl: debug: libxl.c:4226:libxl_device_nic_add: ao 0x25f6950: create: 
how=(nil) callback=(nil) poller=0x25f69e0
libxl: error: libxl.c:1998:device_addrm_aocomplete: unable to add device
libxl: debug: libxl_event.c:1874:libxl__ao_complete: ao 0x25f6950: complete, 
rc=-16
libxl: debug: libxl.c:4226:libxl_device_nic_add: ao 0x25f6950: inprogress: 
poller=0x25f69e0, flags=ic
libxl: debug: libxl_event.c:1843:libxl__ao__destroy: ao 0x25f6950: destroy
libxl_device_nic_add failed.
xc: debug: hypercall buffer: total allocations:20 total releases:20
xc: debug: hypercall buffer: current allocations:0 maximum allocations:2
xc: debug: hypercall buffer: cache current size:2
xc: debug: hypercall buffer: cache hits:12 misses:2 toobig:6


Any ideas?

Ruediger

_______________________________________________
Xen-users mailing list
Xen-users@xxxxxxxxxxxxx
http://lists.xen.org/xen-users