Xen devel - Oct 2012 - PCI USB Passthrough on Kernel 3.5 and 3.6 not working (HVM or PVM)

Hi Xenners,

I have tried to upgrade to kernel 3.5 and experienced the USB/PCI
passthrough problem.  This has not been "fixed" in 3.6 either.  I do
not
know if this is qemu, xen or kernel related.

I am using Xen 4.2 final and Xen 4.3-unstable to test.

DomU is Windows 7 64 bit or Linux Fedora 17.  In Windows, I get the
little yellow triangle with Code 10 (I know this means nothing...
really).  In linux, the PCI device is recognised, but a few seconds
later disabled due to a "fatal error", no more details than that in
dmesg.  

Dom0 retains full control over the USB controller (this is not the case
when passthrough works, xl switches between Dom0 and DomU depending on
who claims the device).  

In both Windows and Linux (lspci), the device is "seen" but cannot be
activated.  

In Dom0, the device is assigned to the guest (xl pci-list show that it
is assigned and has a valid guest PCI ID).

Xen logs are normal, devices get assigned as if they were working with a
kernel 3.3.4 Dom0 - no errors according to xen logs or xl dmesg.

VGA PCI passthrough works 100%.

I tried also with device model qemu-xen, but then passthrough doesn''t
work at all (no VGA either), even with upstream QEMU from git.

Please tell me if I''m missing something obvious or if I can provide
more
information for troubleshooting!

Kind regards,
Andi

> Hi Xenners,
> 
> I have tried to upgrade to kernel 3.5 and experienced the USB/PCI
> passthrough problem.  This has not been "fixed" in 3.6 either.  I
do not
> know if this is qemu, xen or kernel related.
> 
> I am using Xen 4.2 final and Xen 4.3-unstable to test.
> 
> DomU is Windows 7 64 bit or Linux Fedora 17.  In Windows, I get the
> little yellow triangle with Code 10 (I know this means nothing...
> really).  In linux, the PCI device is recognised, but a few seconds
> later disabled due to a "fatal error", no more details than that
in
> dmesg.
> 
> Dom0 retains full control over the USB controller (this is not the case
> when passthrough works, xl switches between Dom0 and DomU depending on
> who claims the device).
> 
> In both Windows and Linux (lspci), the device is "seen" but
cannot be
> activated.
> 
> In Dom0, the device is assigned to the guest (xl pci-list show that it
> is assigned and has a valid guest PCI ID).
> 
> Xen logs are normal, devices get assigned as if they were working with a
> kernel 3.3.4 Dom0 - no errors according to xen logs or xl dmesg.
> 
> VGA PCI passthrough works 100%.
> 
> I tried also with device model qemu-xen, but then passthrough
doesn''t
> work at all (no VGA either), even with upstream QEMU from git.
> 
> Please tell me if I''m missing something obvious or if I can
provide more
> information for troubleshooting!
> 
> Kind regards,
> Andi
Hi,
did you try the resource_allignment workaround?
What is your HW anyway?
-- 
Pavel Mateja

On Thu, Oct 11, 2012 at 03:14:22PM +0200, Pavel Mateja
wrote:
> > Hi Xenners,
> > 
> > I have tried to upgrade to kernel 3.5 and experienced the USB/PCI
> > passthrough problem.  This has not been "fixed" in 3.6
either.  I do not
> > know if this is qemu, xen or kernel related.
Does your lspci (in dom0) for the device you are passing in, show
''virtual'' for the BARs? If so, then it is a kernel issue.

I did another patch for this and it ought to be on the v3.6.x and
v3.5.x train:

commit c341ca45ce56143804ef5a8f4db753e554e640b4
Author: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Date:   Tue Sep 25 16:48:24 2012 -0400

    xen/pciback: Restore the PCI config space after an FLR.
> > 
> > I am using Xen 4.2 final and Xen 4.3-unstable to test.
> > 
> > DomU is Windows 7 64 bit or Linux Fedora 17.  In Windows, I get the
> > little yellow triangle with Code 10 (I know this means nothing...
> > really).  In linux, the PCI device is recognised, but a few seconds
> > later disabled due to a "fatal error", no more details than
that in
> > dmesg.
> > 
> > Dom0 retains full control over the USB controller (this is not the
case
> > when passthrough works, xl switches between Dom0 and DomU depending on
> > who claims the device).
> > 
> > In both Windows and Linux (lspci), the device is "seen" but
cannot be
> > activated.
> > 
> > In Dom0, the device is assigned to the guest (xl pci-list show that it
> > is assigned and has a valid guest PCI ID).
> > 
> > Xen logs are normal, devices get assigned as if they were working with
a
> > kernel 3.3.4 Dom0 - no errors according to xen logs or xl dmesg.
> > 
> > VGA PCI passthrough works 100%.
> > 
> > I tried also with device model qemu-xen, but then passthrough
doesn''t
> > work at all (no VGA either), even with upstream QEMU from git.
> > 
> > Please tell me if I''m missing something obvious or if I can
provide more
> > information for troubleshooting!
> > 
> > Kind regards,
> > Andi
> 
> Hi,
> did you try the resource_allignment workaround?
> What is your HW anyway?
> -- 
> Pavel Mateja
> 
> _______________________________________________
> Xen-devel mailing list
> Xen-devel@lists.xen.org
> http://lists.xen.org/xen-devel
>