IRQ issue on graphics card passthrough

IRQ issue on graphics card passthrough

[André Weidemann]

Hi,
as mentioned before I have successfully passed a graphics card from a 
Linux host to VM using qemu-kvm.

Shortly after starting the VM and before Windows7 initializes the 
graphics card, "info pci" looks like this:
   Bus  0, device   4, function 0:
     VGA controller: PCI device 1002:6719
       IRQ 10.
       BAR0: 32 bit prefetchable memory at 0xe0000000 [0xefffffff].
       BAR2: 32 bit memory at 0xfeba0000 [0xfebbffff].
       BAR4: I/O at 0xc000 [0xc0ff].
       BAR6: 32 bit memory at 0xffffffffffffffff [0x0001fffe].
       id "radeon6950"

After Windows has initialized the graphics card, "info pci" lists the 
following for the graphics card:
   Bus  0, device  4, function 0:
     VGA controller: PCI device 1002:6719
       IRQ 0.
       BAR0: 32 bit prefetchable memory at 0xe0000000 [0xefffffff].
       BAR2: 32 bit memory at 0xfeba0000 [0xfebbffff].
       BAR4: I/O at 0xc000 [0xc0ff].
       BAR6: 32 bit memory at 0xffffffffffffffff [0x0001fffe].
       id "radeon6950"

Notice that the IRQ changed from 10 to 0...

The graphics card's IRQ under Windows7 is displayed as:
"0xFFFFFFFE (-2)".

During another problem of assigning sound cards to a  VM, Jan Kiszka has 
pointed me to a patch that addresses an IRQ issue: 
http://thread.gmane.org/gmane.comp.emulators.qemu/102540
Unfortuantely it does not help here.

What can I do to examine this issue further?

Regards
  André

Re: IRQ issue on graphics card passthrough

[Jan Kiszka]

[-- Attachment #1: Type: text/plain, Size: 1712 bytes --]

On 2011-06-04 19:50, André Weidemann wrote:
> Hi,
> as mentioned before I have successfully passed a graphics card from a
> Linux host to VM using qemu-kvm.
> 
> Shortly after starting the VM and before Windows7 initializes the
> graphics card, "info pci" looks like this:
>   Bus  0, device   4, function 0:
>     VGA controller: PCI device 1002:6719
>       IRQ 10.
>       BAR0: 32 bit prefetchable memory at 0xe0000000 [0xefffffff].
>       BAR2: 32 bit memory at 0xfeba0000 [0xfebbffff].
>       BAR4: I/O at 0xc000 [0xc0ff].
>       BAR6: 32 bit memory at 0xffffffffffffffff [0x0001fffe].
>       id "radeon6950"
> 
> After Windows has initialized the graphics card, "info pci" lists the
> following for the graphics card:
>   Bus  0, device  4, function 0:
>     VGA controller: PCI device 1002:6719
>       IRQ 0.
>       BAR0: 32 bit prefetchable memory at 0xe0000000 [0xefffffff].
>       BAR2: 32 bit memory at 0xfeba0000 [0xfebbffff].
>       BAR4: I/O at 0xc000 [0xc0ff].
>       BAR6: 32 bit memory at 0xffffffffffffffff [0x0001fffe].
>       id "radeon6950"
> 
> Notice that the IRQ changed from 10 to 0...
> 
> The graphics card's IRQ under Windows7 is displayed as:
> "0xFFFFFFFE (-2)".

That's not necessarily pointing to a problem, that just means Windows is
using the device in MSI mode. Does it do the same when running natively?

> 
> During another problem of assigning sound cards to a  VM, Jan Kiszka has
> pointed me to a patch that addresses an IRQ issue:
> http://thread.gmane.org/gmane.comp.emulators.qemu/102540
> Unfortuantely it does not help here.

For sure, that patch addresses a legacy interrupt issue, nothing
MSI-related.

Jan


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 259 bytes --]

Re: IRQ issue on graphics card passthrough

[André Weidemann]

Hi Jan,

On 05.06.2011 10:30, Jan Kiszka wrote:
> On 2011-06-04 19:50, André Weidemann wrote:
>> Hi,
>> as mentioned before I have successfully passed a graphics card from a
>> Linux host to VM using qemu-kvm.
>>
>> Shortly after starting the VM and before Windows7 initializes the
>> graphics card, "info pci" looks like this:
>>    Bus  0, device   4, function 0:
>>      VGA controller: PCI device 1002:6719
>>        IRQ 10.
>>        BAR0: 32 bit prefetchable memory at 0xe0000000 [0xefffffff].
>>        BAR2: 32 bit memory at 0xfeba0000 [0xfebbffff].
>>        BAR4: I/O at 0xc000 [0xc0ff].
>>        BAR6: 32 bit memory at 0xffffffffffffffff [0x0001fffe].
>>        id "radeon6950"
>>
>> After Windows has initialized the graphics card, "info pci" lists the
>> following for the graphics card:
>>    Bus  0, device  4, function 0:
>>      VGA controller: PCI device 1002:6719
>>        IRQ 0.
>>        BAR0: 32 bit prefetchable memory at 0xe0000000 [0xefffffff].
>>        BAR2: 32 bit memory at 0xfeba0000 [0xfebbffff].
>>        BAR4: I/O at 0xc000 [0xc0ff].
>>        BAR6: 32 bit memory at 0xffffffffffffffff [0x0001fffe].
>>        id "radeon6950"
>>
>> Notice that the IRQ changed from 10 to 0...
>>
>> The graphics card's IRQ under Windows7 is displayed as:
>> "0xFFFFFFFE (-2)".
>
> That's not necessarily pointing to a problem, that just means Windows is
> using the device in MSI mode. Does it do the same when running natively?
>
>>
>> During another problem of assigning sound cards to a  VM, Jan Kiszka has
>> pointed me to a patch that addresses an IRQ issue:
>> http://thread.gmane.org/gmane.comp.emulators.qemu/102540
>> Unfortuantely it does not help here.
>
> For sure, that patch addresses a legacy interrupt issue, nothing
> MSI-related.

It took me some time to get a native Windows7 running on that machine, 
but you are right. The IRQ under Windows("0xFFFFFFF6 (-10)") is similar 
to the one inside the VM ("0xFFFFFFFE (-2)").

Thank you for your reply.
  André