Re: [Xen-devel] Re: Xen 3.4 multi-function pass-through tree, isn't working...

[Tom Rotenberg <tom.rotenberg@xxxxxxxxx>]

Simon,

My results:

1. Trying to isolate the device that causes this:  - well it looks
like whenever i try to PT more than one device it happens, here is the
log:
t500-testing:~# xm create /etc/xen/hvm pci='00:1a.0,1,2,7'
Using config file "/etc/xen/xp2_ati".
Error: int() argument must be a string or a number, not 'NoneType'
t500-testing:~# xm create /etc/xen/hvm_ati pci='00:1a.0,1,2'
Using config file "/etc/xen/xp2_ati".
Error: int() argument must be a string or a number, not 'NoneType'
t500-testing:~# xm create /etc/xen/hvm_ati pci='00:1a.0,1'
Using config file "/etc/xen/xp2_ati".
Error: int() argument must be a string or a number, not 'NoneType'
t500-testing:~# xm create /etc/xen/hvm_ati pci='00:1a.0'
Using config file "/etc/xen/xp2_ati".
Started domain hvm (id=4)
t500-testing:~# xm create /etc/xen/hvm_ati pci='00:1a.0' pci='00:1d.0'
Using config file "/etc/xen/xp2_ati".
Error: int() argument must be a string or a number, not 'NoneType'
t500-testing:~# xm create /etc/xen/hvm_ati pci='00:1a.0'
Using config file "/etc/xen/xp2_ati".
Started domain hvm (id=6)
t500-testing:~# xm create /etc/xen/hvm_ati pci='00:1a.0,2'
Using config file "/etc/xen/xp2_ati".
Error: int() argument must be a string or a number, not 'NoneType'
t500-testing:~# xm create /etc/xen/xp2_ati pci='00:1a.0,1'
Using config file "/etc/xen/xp2_ati".
Error: int() argument must be a string or a number, not 'NoneType'

2. The result of the bisection of the tree, to find the problematic changest.
Well, it looks like, changeset: 19694, caused this problem. The
changeset details:
changeset:   19694:b19ed7753827
user:        Simon Horman <horms@xxxxxxxxxxxx>
date:        Mon Jul 13 18:48:05 2009 +1000
summary:     xend: hot-plug PCI devices at boot-time

Tom

On Wed, Jul 22, 2009 at 9:11 AM, Simon Horman<horms@xxxxxxxxxxxx> wrote:
> On Tue, Jul 21, 2009 at 08:15:00PM +1000, Simon Horman wrote:
>> On Tue, Jul 21, 2009 at 10:22:19AM +0300, Tom Rotenberg wrote:
>> > Simon,
>> >
>> > First of all, my workaround seems to work ok, so i think you should
>> > consider including it anyway (if there was an exception during state
>> > retrieval, then the state should be unknown, instead of throw an
>> > exception). Do u think you will add this patch to the multi-function
>> > code?
>>
>> I am certainly considering adding it if we can't determine why this
>> happens. But it would be better to know why :-)
>>
>> > I'm reproducing this, on a Lenovo T500 machine, which is different
>> > than your machine, since it also contains the 00:1a.2 device.
>> > Also, i don't specify the devices to be pass-through in the boot line,
>> > but rather, bind them to pciback after the machine boots.
>>
>> Thanks, I'll see if binding them after boot sheds any light on the problem.
>
> Hi Tom,
>
> unfortuantely binding devices to pci-back after booting dom0 didn't
> help me to reproduce the problem.
>
> I'm wondering if you could run a few tests to try and isolate the problem
>
> 1) Try passing through the PCI devices one by one to try and
>   see which devices cause the problem that you are seeing and;
>
> 2) Do a bisection on the tree - say back to v3.4.1-rc6, to see
>   which changeset is causing the problem.
>
>

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-devel