Interrupt issue for nic cards

Interrupt issue for nic cards

[Wei Liu]

Hi, all.

I'm running the latest Xen unstable with Debian Squeeze as DomU in HVM mode.

I configured two nics:

vif = ['bridge=br0,model=virtio', 'type=ioemu,bridge=br0,model=rtl8139']

These two cards can be successfully configured in DomU. But they fail
to communicate with host bridge.

I can see packets coming out their taps, but no packets are going back
into DomU.

Stefano suggest I look into /proc/interrupts. It shows:

11      0    IO-APIC-fasteoi   eth0
48      0    PCI-MSI-edge   virtio0-config
49      0    PCI-MSI-edge   virtio0-input
50      0    PCI-MSI-edge   virtio0-output

It seems that there are no interrupts for those nic devices. I'm
pretty sure that QEMU has pushed the data back and set interrupts
(through msix_notify / qemu_set_irq).

Has anyone encountered this problem before?

-- 
Best regards
Wei Liu
Twitter: @iliuw
Site: http://liuw.name

Re: Interrupt issue for nic cards

[Konrad Rzeszutek Wilk]

On Fri, May 13, 2011 at 11:14:47AM +0800, Wei Liu wrote:
> Hi, all.
> 
> I'm running the latest Xen unstable with Debian Squeeze as DomU in HVM mode.
> 
> I configured two nics:
> 
> vif = ['bridge=br0,model=virtio', 'type=ioemu,bridge=br0,model=rtl8139']
> 
> These two cards can be successfully configured in DomU. But they fail
> to communicate with host bridge.
> 
> I can see packets coming out their taps, but no packets are going back
> into DomU.
> 
> Stefano suggest I look into /proc/interrupts. It shows:
> 
> 11      0    IO-APIC-fasteoi   eth0
> 48      0    PCI-MSI-edge   virtio0-config
> 49      0    PCI-MSI-edge   virtio0-input
> 50      0    PCI-MSI-edge   virtio0-output
> 
> It seems that there are no interrupts for those nic devices. I'm
> pretty sure that QEMU has pushed the data back and set interrupts
> (through msix_notify / qemu_set_irq).
> 
> Has anyone encountered this problem before?

I haven't. But it looks like the NIC isn't beeing interrupted - which
implies that QEMU hasn't sent the IRQ to the DomU to tell it: 'hey
you got data'. It might be worth looking in QEMU and see under which
circumstances it sends interrupts. And whether the mechanism for sending
interrupts is KVM specific instead of being more generic (or able to
use the Xen mechanism).

> 
> -- 
> Best regards
> Wei Liu
> Twitter: @iliuw
> Site: http://liuw.name
> 
> _______________________________________________
> Xen-devel mailing list
> Xen-devel@lists.xensource.com
> http://lists.xensource.com/xen-devel

Re: Interrupt issue for nic cards

[Stefano Stabellini]

On Fri, 13 May 2011, Konrad Rzeszutek Wilk wrote:
> On Fri, May 13, 2011 at 11:14:47AM +0800, Wei Liu wrote:
> > Hi, all.
> > 
> > I'm running the latest Xen unstable with Debian Squeeze as DomU in HVM mode.
> > 
> > I configured two nics:
> > 
> > vif = ['bridge=br0,model=virtio', 'type=ioemu,bridge=br0,model=rtl8139']
> > 
> > These two cards can be successfully configured in DomU. But they fail
> > to communicate with host bridge.
> > 
> > I can see packets coming out their taps, but no packets are going back
> > into DomU.
> > 
> > Stefano suggest I look into /proc/interrupts. It shows:
> > 
> > 11      0    IO-APIC-fasteoi   eth0
> > 48      0    PCI-MSI-edge   virtio0-config
> > 49      0    PCI-MSI-edge   virtio0-input
> > 50      0    PCI-MSI-edge   virtio0-output
> > 
> > It seems that there are no interrupts for those nic devices. I'm
> > pretty sure that QEMU has pushed the data back and set interrupts
> > (through msix_notify / qemu_set_irq).
> > 
> > Has anyone encountered this problem before?
> 
> I haven't. But it looks like the NIC isn't beeing interrupted - which
> implies that QEMU hasn't sent the IRQ to the DomU to tell it: 'hey
> you got data'. It might be worth looking in QEMU and see under which
> circumstances it sends interrupts. And whether the mechanism for sending
> interrupts is KVM specific instead of being more generic (or able to
> use the Xen mechanism).


Indeed. If I am not mistaked the function that should be called within
qemu to actually inject the interrupt is xen_piix3_set_irq and should be
called bt qemu_set_irq.

It is also worth making sure you are using the very last Xen from
xen-unstable, with the two patches I sent to the list in the last two
days (they fix interrupt injection issues to HVM domains):

http://marc.info/?l=xen-devel&m=130512868904804
http://marc.info/?l=xen-devel&m=130521059427303

Re: Interrupt issue for nic cards

[Wei Liu]

On Fri, May 13, 2011 at 9:49 PM, Stefano Stabellini
<stefano.stabellini@eu.citrix.com> wrote:
>
> Indeed. If I am not mistaked the function that should be called within
> qemu to actually inject the interrupt is xen_piix3_set_irq and should be
> called bt qemu_set_irq.
>
> It is also worth making sure you are using the very last Xen from
> xen-unstable, with the two patches I sent to the list in the last two
> days (they fix interrupt injection issues to HVM domains):
>
> http://marc.info/?l=xen-devel&m=130512868904804
> http://marc.info/?l=xen-devel&m=130521059427303
>

I applied your patches, but things are just the same.

Working on that now.

-- 
Best regards
Wei Liu
Twitter: @iliuw
Site: http://liuw.name