BUG? domU reboot fails when using "pci =" option

BUG? domU reboot fails when using "pci =" option

[list user]

Hi all,

I believe I've found a bug.

Running xen-3.0.2-2 with linux-2.6.16 built from sources.

Domain-0 has one hidden PCI slot:  pciback.hide=(00:0d.0)

It is exposed in a domU with pci = ['00:0d.0'].

If that domU is shutdown with xm destroy, or with a shutdown -h now, all
works as expected.  But if instead it is rebooted, either by xm reboot
or shutdown -r now, it shuts down, attempts to restart and fails
silently.  This is evidenced by the fact that an xm list has used up one
increment on its domU counter.  At that point a manual start works.

Now, if the pci device is NOT enabled in the domU config file, reboot
works correctly.

Checking xend.log provides a clue:

   dev_config referenced before assignment in
                         /usr/lib/xen/xend/server/pciif.py

Not really a python coder but it looks like the invocation in question
is on line 97, which indicates that whole section of code may be borked,
since it now thinks it's xen-2.0.

Hope that's helpful to someone.

Mike Wright

Re: BUG? domU reboot fails when using "pci =" option

[list user]

list user wrote:
> Hi all,
> 
> I believe I've found a bug.
> 
> Running xen-3.0.2-2 with linux-2.6.16 built from sources.
> 
> Domain-0 has one hidden PCI slot:  pciback.hide=(00:0d.0)
> 
> It is exposed in a domU with pci = ['00:0d.0'].
> 
> If that domU is shutdown with xm destroy, or with a shutdown -h now, all
> works as expected.  But if instead it is rebooted, either by xm reboot
> or shutdown -r now, it shuts down, attempts to restart and fails
> silently.  This is evidenced by the fact that an xm list has used up one
> increment on its domU counter.  At that point a manual start works.
> 
> Now, if the pci device is NOT enabled in the domU config file, reboot
> works correctly.
> 
> Checking xend.log provides a clue:
> 
>   dev_config referenced before assignment in
>                         /usr/lib/xen/xend/server/pciif.py
> 
> Not really a python coder but it looks like the invocation in question
> is on line 97, which indicates that whole section of code may be borked,
> since it now thinks it's xen-2.0.
> 

Additional info.

If xend and xendomains are set to run automatically at boot time via 
chkconfig the same error occurs, but ONLY when the pci device has been 
exported to a domU.  An "xm list" starts at domid 2, showing that one 
domU has failed at startup.

Again, the error is familiar, but this time displays on the domain-0 
console:

   local variable dev_config referenced before assignment

:m)

> Hope that's helpful to someone.
> 
> Mike Wright
> 
> 
> 
> _______________________________________________
> Xen-users mailing list
> Xen-users@lists.xensource.com
> http://lists.xensource.com/xen-users
> 
> 

Re: BUG? domU reboot fails when using "pci =" option

[Ryan]

On Thu, 2006-05-04 at 10:53 -0700, list user wrote:
> list user wrote:
> 
> Additional info.
> 
> If xend and xendomains are set to run automatically at boot time via 
> chkconfig the same error occurs, but ONLY when the pci device has been 
> exported to a domU.  An "xm list" starts at domid 2, showing that one 
> domU has failed at startup.
> 
> Again, the error is familiar, but this time displays on the domain-0 
> console:
> 
>    local variable dev_config referenced before assignment
> 
> :m)
> 
> > Hope that's helpful to someone.
> > 
> > Mike Wright
> > 

I'm not quite sure how this python error could be causing your reboot
problem, but you did find a typo in that script. I'll submit a patch for
it to xen-unstable. If you'd like to fix it yourself, just change
dev_config to config.

Did you upgrade your box from xen 2.0.x to xen 3.0.2? That code path is
unlikely to be followed unless you either used an SXP configuration file
(which I didn't think you had done because "pci = ['00:0d.0']" wouldn't
work in an SXP file) or your Xen 3.0.2 installation didn't completely
overwrite all of the Xen 2.0.x files.

Ryan

Re: [Xen-devel] Re: BUG? domU reboot fails when using "pci =" option

[list user]

Ryan wrote:
> On Thu, 2006-05-04 at 10:53 -0700, list user wrote:
> 
>>list user wrote:
>>
>>Additional info.
>>
>>If xend and xendomains are set to run automatically at boot time via 
>>chkconfig the same error occurs, but ONLY when the pci device has been 
>>exported to a domU.  An "xm list" starts at domid 2, showing that one 
>>domU has failed at startup.
>>
>>Again, the error is familiar, but this time displays on the domain-0 
>>console:
>>
>>   local variable dev_config referenced before assignment
>>
>>:m)
>>
>>
>>>Hope that's helpful to someone.
>>>
>>>Mike Wright
>>>
> 
> 
> I'm not quite sure how this python error could be causing your reboot
> problem, but you did find a typo in that script. I'll submit a patch for
> it to xen-unstable. If you'd like to fix it yourself, just change
> dev_config to config.

Thanks Ryan,

I tried the fix but it just gave a different symptom.  (same error, 
different variable)

So... rpm --erase xen JIC there were any ghosts around.  Manually 
removed all traces of xen from the system.  Then did a ./install.sh from 
  the xen-3.0.2-2 directory, which had a clean "make world".

The results are the same.

I've found a more generic error about a local variable being used before 
assignment, but I think it's referring to dev_config.

Again, this only happens on auto-boot or domU reboot, and only when 
there is a "pci =" line in the domU's definition file.

Any ideas how I can help debug this?

:m)
> 
> Did you upgrade your box from xen 2.0.x to xen 3.0.2? That code path is
> unlikely to be followed unless you either used an SXP configuration file
> (which I didn't think you had done because "pci = ['00:0d.0']" wouldn't
> work in an SXP file) or your Xen 3.0.2 installation didn't completely
> overwrite all of the Xen 2.0.x files.
> 
> Ryan
> 
> 
> _______________________________________________
> Xen-devel mailing list
> Xen-devel@lists.xensource.com
> http://lists.xensource.com/xen-devel
> 
>