On 08/10/2025 15.19, Matthew Rosato wrote:
On 10/8/25 8:51 AM, Thomas Huth wrote:
On 07/10/2025 22.56, Matthew Rosato wrote:
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.
Reviewed-by: Eric Farman <[email protected]>
Signed-off-by: Matthew Rosato <[email protected]>
---
v2:
- add review tag
- tweak commit message (David)
---
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();
}
Hi Matthew,
this unfortunately fails to compile without KVM (e.g. on x86 hosts):
../../devel/qemu/hw/s390x/s390-pci-bus.c: In function ‘s390_pcihost_realize’:
../../devel/qemu/hw/s390x/s390-pci-bus.c:903:5: error:
‘kvm_msi_via_irqfd_allowed’ undeclared (first use in this function); did you
mean ‘kvm_msi_via_irqfd_enabled’?
903 | kvm_msi_via_irqfd_allowed = kvm_irqfds_enabled();
| ^~~~~~~~~~~~~~~~~~~~~~~~~
| kvm_msi_via_irqfd_enabled
../../devel/qemu/hw/s390x/s390-pci-bus.c:903:5: note: each undeclared
identifier is reported only once for each function it appears in
Argh, thanks for pointing this out. Of course this value only exists when
CONFIG_KVM...
What do you think about using a routine in hw/s390x/s390-pci-kvm.c to set the
value (and is a NOP stub when !CONFIG_KVM)?
Yes, since we're doing this for some other KVM-related functions there
already, that sounds like the right way to go here.
Thomas
