Xen devel - Sep 2012 - 4.2 kernel 3.5 USB PCI Passthrough

Hi Xenners,

I have been using 4.2 for a very long time now, and recently (since 
upgrading to rc2), I''ve had some strange issues cropping up:

None of my disk images boot with xl create;  I have moved the important 
ones to LVM disks, since phy: boots fine.  I have hacked some of the xen 
code to make this work (but forgot which hack got it working, it has to 
do with the caching modes in the file open calls - I will double check 
and get back to the list).  My qemu disks are always reported as 
0mb-sized volumes in the BIOS screen when this bug manifests itself.  
Neither raw nor qcow2 images work.

When I upgraded to kernel 3.5, my USB PCI passthrough stopped working.  
The secondary VGA passthrough still works fine, just the USB host 
controller that is passed through gets claimed by xen, but the guest 
cannot load drivers for the controller.  I.e. the guest (Windoze 7) sees 
it in device manager, but the device fails to activate.

Last night I upgraded to the latest 3.5.4 in the fc17 repo, and the USB 
issue was still there.  The weird thing though, is that it seems that my 
raw images magically started booting.  I still need to investigate this, 
in more detail to make sure I''m not mixing up versions and get 
conclusive results.

Apologies, this post is not too detailed and contains too many
"maybes",
but I though there may be a simple answer that someone else had stumbled 
across the same issues recently.

I will compile rc3 later on and run decent comparisons on both kernels, 
as well as raw and LVM machines.

Kind regards,
Andi

On Tue, Sep 25, 2012 at 09:09:57AM +0200, Andi Reinbrech
wrote:
> Hi Xenners,
> 
> I have been using 4.2 for a very long time now, and recently (since
> upgrading to rc2), I''ve had some strange issues cropping up:
> 
> None of my disk images boot with xl create;  I have moved the
> important ones to LVM disks, since phy: boots fine.  I have hacked
> some of the xen code to make this work (but forgot which hack got it
> working, it has to do with the caching modes in the file open calls
> - I will double check and get back to the list).  My qemu disks are
> always reported as 0mb-sized volumes in the BIOS screen when this
> bug manifests itself.  Neither raw nor qcow2 images work.
>
I think there was fixes/changes to Xen qemu-dm caching modes
during the late 4.2-rc''s. 
 
> When I upgraded to kernel 3.5, my USB PCI passthrough stopped
> working.  The secondary VGA passthrough still works fine, just the
> USB host controller that is passed through gets claimed by xen, but
> the guest cannot load drivers for the controller.  I.e. the guest
> (Windoze 7) sees it in device manager, but the device fails to
> activate.
> 
Any errors in dom0 dmesg or in xen dmesg? 
> Last night I upgraded to the latest 3.5.4 in the fc17 repo, and the
> USB issue was still there.  The weird thing though, is that it seems
> that my raw images magically started booting.  I still need to
> investigate this, in more detail to make sure I''m not mixing up
> versions and get conclusive results.
> 
Did you upgrade Xen to 4.2.0 final at the same time? 
> Apologies, this post is not too detailed and contains too many
> "maybes", but I though there may be a simple answer that someone
> else had stumbled across the same issues recently.
> 
> I will compile rc3 later on and run decent comparisons on both
> kernels, as well as raw and LVM machines.
> 
You probably should post a different thread for each problem..

-- Pasi