[PATCH] s390x/pci: set kvm_msi_via_irqfd_allowed

[PATCH] s390x/pci: set kvm_msi_via_irqfd_allowed

[Matthew Rosato]

Allow irqfd to be used for virtio-pci on s390x if the kernel supports
it.

Signed-off-by: Matthew Rosato <mjrosato@linux.ibm.com>
---
 hw/s390x/s390-pci-bus.c | 1 +
 1 file changed, 1 insertion(+)

diff --git a/hw/s390x/s390-pci-bus.c b/hw/s390x/s390-pci-bus.c
index e8e41c8a9a..88852acf45 100644
--- a/hw/s390x/s390-pci-bus.c
+++ b/hw/s390x/s390-pci-bus.c
@@ -900,6 +900,7 @@ static void s390_pcihost_realize(DeviceState *dev, Error **errp)
     s390_pci_init_default_group();
     css_register_io_adapters(CSS_IO_ADAPTER_PCI, true, false,
                              S390_ADAPTER_SUPPRESSIBLE, errp);
+    kvm_msi_via_irqfd_allowed = kvm_irqfds_enabled();
 }
 
 static void s390_pcihost_unrealize(DeviceState *dev)
-- 
2.51.0

Re: [PATCH] s390x/pci: set kvm_msi_via_irqfd_allowed

[Matthew Rosato]

On 10/1/25 4:05 PM, Matthew Rosato wrote:
> Allow irqfd to be used for virtio-pci on s390x if the kernel supports
> it.
> 
> Signed-off-by: Matthew Rosato <mjrosato@linux.ibm.com>
> ---

FWIW we've been using this change in-house for a while now under test different scenarios - AFAICT failure to set this was just a past oversight since virtio-pci is not the default/recommendation for s390x.

>  hw/s390x/s390-pci-bus.c | 1 +
>  1 file changed, 1 insertion(+)
> 
> diff --git a/hw/s390x/s390-pci-bus.c b/hw/s390x/s390-pci-bus.c
> index e8e41c8a9a..88852acf45 100644
> --- a/hw/s390x/s390-pci-bus.c
> +++ b/hw/s390x/s390-pci-bus.c
> @@ -900,6 +900,7 @@ static void s390_pcihost_realize(DeviceState *dev, Error **errp)
>      s390_pci_init_default_group();
>      css_register_io_adapters(CSS_IO_ADAPTER_PCI, true, false,
>                               S390_ADAPTER_SUPPRESSIBLE, errp);
> +    kvm_msi_via_irqfd_allowed = kvm_irqfds_enabled();
>  }
>  
>  static void s390_pcihost_unrealize(DeviceState *dev)

Re: [PATCH] s390x/pci: set kvm_msi_via_irqfd_allowed

[David Hildenbrand]

On 01.10.25 22:05, Matthew Rosato wrote:
> Allow irqfd to be used for virtio-pci on s390x if the kernel supports
> it.
> 
> Signed-off-by: Matthew Rosato <mjrosato@linux.ibm.com>
> ---
>   hw/s390x/s390-pci-bus.c | 1 +
>   1 file changed, 1 insertion(+)
> 
> diff --git a/hw/s390x/s390-pci-bus.c b/hw/s390x/s390-pci-bus.c
> index e8e41c8a9a..88852acf45 100644
> --- a/hw/s390x/s390-pci-bus.c
> +++ b/hw/s390x/s390-pci-bus.c
> @@ -900,6 +900,7 @@ static void s390_pcihost_realize(DeviceState *dev, Error **errp)
>       s390_pci_init_default_group();
>       css_register_io_adapters(CSS_IO_ADAPTER_PCI, true, false,
>                                S390_ADAPTER_SUPPRESSIBLE, errp);
> +    kvm_msi_via_irqfd_allowed = kvm_irqfds_enabled();

Just wondering, what's the net result of this patch?

I assume better performance. Might be worth adding that "why" to the 
patch description.

-- 
Cheers

David / dhildenb

Re: [PATCH] s390x/pci: set kvm_msi_via_irqfd_allowed

[Matthew Rosato]

On 10/2/25 3:27 AM, David Hildenbrand wrote:
> On 01.10.25 22:05, Matthew Rosato wrote:
>> Allow irqfd to be used for virtio-pci on s390x if the kernel supports
>> it.
>>
>> Signed-off-by: Matthew Rosato <mjrosato@linux.ibm.com>
>> ---
>>   hw/s390x/s390-pci-bus.c | 1 +
>>   1 file changed, 1 insertion(+)
>>
>> diff --git a/hw/s390x/s390-pci-bus.c b/hw/s390x/s390-pci-bus.c
>> index e8e41c8a9a..88852acf45 100644
>> --- a/hw/s390x/s390-pci-bus.c
>> +++ b/hw/s390x/s390-pci-bus.c
>> @@ -900,6 +900,7 @@ static void s390_pcihost_realize(DeviceState *dev, Error **errp)
>>       s390_pci_init_default_group();
>>       css_register_io_adapters(CSS_IO_ADAPTER_PCI, true, false,
>>                                S390_ADAPTER_SUPPRESSIBLE, errp);
>> +    kvm_msi_via_irqfd_allowed = kvm_irqfds_enabled();
> 
> Just wondering, what's the net result of this patch?
> 

It allows the virtio-pci core code on s390x to ever decide using irqfd is OK (e.g. without this change we would always reach virtio_queue_set_guest_notifier_fd_handler(..., with_irqfd=false) because kvm_msi_via_irqfd_enabled() was guaranteed to be false.

> I assume better performance. Might be worth adding that "why" to the patch description.
> 

And yes, in a nutshell it means better performance.

How about something like...

Allow irqfd to be used for virtio-pci on s390x if the kernel supports
it.  This improves s390x virtio-pci performance when using kvm
acceleration by allowing kvm to deliver interrupts instead of QEMU.

Re: [PATCH] s390x/pci: set kvm_msi_via_irqfd_allowed

[Eric Farman]

On Wed, 2025-10-01 at 16:05 -0400, Matthew Rosato wrote:
> Allow irqfd to be used for virtio-pci on s390x if the kernel supports
> it.
> 
> Signed-off-by: Matthew Rosato <mjrosato@linux.ibm.com>
> ---
>  hw/s390x/s390-pci-bus.c | 1 +
>  1 file changed, 1 insertion(+)

As Matt alluded to in another reply, this is probably just historical oversight.

Reviewed-by: Eric Farman <farman@linux.ibm.com>

> 
> diff --git a/hw/s390x/s390-pci-bus.c b/hw/s390x/s390-pci-bus.c
> index e8e41c8a9a..88852acf45 100644
> --- a/hw/s390x/s390-pci-bus.c
> +++ b/hw/s390x/s390-pci-bus.c
> @@ -900,6 +900,7 @@ static void s390_pcihost_realize(DeviceState *dev, Error **errp)
>      s390_pci_init_default_group();
>      css_register_io_adapters(CSS_IO_ADAPTER_PCI, true, false,
>                               S390_ADAPTER_SUPPRESSIBLE, errp);
> +    kvm_msi_via_irqfd_allowed = kvm_irqfds_enabled();
>  }
>  
>  static void s390_pcihost_unrealize(DeviceState *dev)