Xen devel - Feb 2011 - PCI passthrough of multiple devices to HVM guest using stub domain not working

Only the first function of the first device that I pass-through to an
HVM guest using a stub domain successfully registers. If the first
device in the "pci=[....]" line in the configuration file has a single
function, the guest will boot and the first device will be
passed-through successfully, but only none of the devices listed after
the first will work. If the first device in the configuration file has
multiple functions, the guest will hang during start-up.

I''m using Xen 4.0.1 / pv_ops 2.6.32.25 Dom0. Works fine using qemu-dm
in Dom0 instead of stubdom-dm. I''ve tested this with multiple devices.

Here''s an excerpt from the QEMU log showing successful registration of
the first function of a device and failure of the second:

xs_read_watch() -> /local/domain/0/device-model/75/command dm-command
dm-command: hot insert pass-through pci dev
register_real_device: Assigning real physical device 08:00.0 ...
open(/sys/bus/pci/devices/0000:08:00.0/resource, 0)pt_libpci_fixup:
Error: Can''t open /sys/bus/pci/devices/0000:08:00.0/resource: I/O
error
register_real_device: Enable MSI translation via per device option
register_real_device: Disable power management
stat(/sys/bus/pci/devices/0000:08:00.0/physfn)
warning: pt_iomul not supported in stubdom 08:00.0
pt_register_regions: IO region registered (size=0x00020000 base_addr=0xf9600000)
pt_register_regions: IO region registered (size=0x00000020 base_addr=0x0000bf80)
pt_register_regions: IO region registered (size=0x00004000 base_addr=0xf9640000)
pt_msix_init: get MSI-X table bar base f9640000
open(/dev/mem) -> 12
pt_msix_init: table_off = 0, total_entries = 10
pt_msix_init: errno = 5
close(12)
pt_msix_init: mapping physical MSI-X table to 2807000
pt_msi_setup: Error: Mapping of MSI failed.
pt_enable_msi_translate: Error: MSI-INTx translation MSI setup failed, fallback
pci_intx: intx=1
register_real_device: Real physical device 08:00.0 registered successfuly!
IRQ type = INTx
xs_read_watch() -> /local/domain/0/device-model/75/command dm-command
xs_read(/local/domain/0/device-model/75/command): ENOENT
pcifront_watches: backend state changed:
/local/domain/0/backend/pci/76/0/state 7
pcifront_watches: writing device/pci/0/state 7
pcifront_watches: backend state changed:
/local/domain/0/backend/pci/76/0/state 5
pcifront_watches: done waiting
close pci: backend at /local/domain/0/backend/pci/76/0
xs_read_watch() -> /local/domain/0/device-model/75/command dm-command
dm-command: hot insert pass-through pci dev
register_real_device: Assigning real physical device 08:00.1 ...
register_real_device: Error: couldn''t locate device in libpci
structures
xs_read_watch() -> /local/domain/0/device-model/75/command dm-command
xs_read(/local/domain/0/device-model/75/command): ENOENT



Any suggestions/assistance greatly appreciated.

Jason

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

On Wed, 16 Feb 2011, Jason Sonnek wrote:
> Only the first function of the first device that I pass-through to an
> HVM guest using a stub domain successfully registers. If the first
> device in the "pci=[....]" line in the configuration file has a
single
> function, the guest will boot and the first device will be
> passed-through successfully, but only none of the devices listed after
> the first will work. If the first device in the configuration file has
> multiple functions, the guest will hang during start-up.
> 
> I''m using Xen 4.0.1 / pv_ops 2.6.32.25 Dom0. Works fine using
qemu-dm
> in Dom0 instead of stubdom-dm. I''ve tested this with multiple
devices.
I am not too surprised because PCI passthrough of multiple devices to
HVM guests with stubdoms is broken even on xen-unstable. IanC might be
able to give you more details.


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