Traditional mmio in balloon makes qemu do balloon inflation in the same
thread as vcpu thread.In a CPU overcommitment scenario, host may run
more than one vcpu threads on one host thread, which makes
madvise_dontneed_free() wait for a long time due to the function
cond_resched() at host side.

If using SEV/ES and the kernel provided by AMD, the overhead will
become even much larger.

With ioeventfd, the thread for host to do balloon inflation will
be seperated from the vcpu thread, leading to better performance
for the whole process of balloon inflation.

Signed-off-by: Zheyun Shen <szy0127@sjtu.edu.cn>
---
 hw/virtio/virtio-balloon-pci.c | 7 +++++++
 1 file changed, 7 insertions(+)

diff --git a/hw/virtio/virtio-balloon-pci.c b/hw/virtio/virtio-balloon-pci.c
index ce2645b..7195322 100644
--- a/hw/virtio/virtio-balloon-pci.c
+++ b/hw/virtio/virtio-balloon-pci.c
@@ -35,6 +35,12 @@ struct VirtIOBalloonPCI {
     VirtIOBalloon vdev;
 };
 
+static Property virtio_balloon_properties[] = {
+    DEFINE_PROP_BIT("ioeventfd", VirtIOPCIProxy, flags,
+            VIRTIO_PCI_FLAG_USE_IOEVENTFD_BIT, false),
+    DEFINE_PROP_END_OF_LIST(),
+};
+
 static void virtio_balloon_pci_realize(VirtIOPCIProxy *vpci_dev, Error **errp)
 {
     VirtIOBalloonPCI *dev = VIRTIO_BALLOON_PCI(vpci_dev);
@@ -51,6 +57,7 @@ static void virtio_balloon_pci_class_init(ObjectClass *klass, void *data)
     PCIDeviceClass *pcidev_k = PCI_DEVICE_CLASS(klass);
     k->realize = virtio_balloon_pci_realize;
     set_bit(DEVICE_CATEGORY_MISC, dc->categories);
+    device_class_set_props(dc, virtio_balloon_properties);
     pcidev_k->vendor_id = PCI_VENDOR_ID_REDHAT_QUMRANET;
     pcidev_k->device_id = PCI_DEVICE_ID_VIRTIO_BALLOON;
     pcidev_k->revision = VIRTIO_PCI_ABI_VERSION;
-- 
2.34.1

Please CC the relevant maintainers, I only stumbled over that by luck.

See scripts/get_maintainer.pl


On 12.12.23 08:50, Zheyun Shen wrote:
> Traditional mmio in balloon makes qemu do balloon inflation in the same
> thread as vcpu thread.In a CPU overcommitment scenario, host may run
> more than one vcpu threads on one host thread, which makes
> madvise_dontneed_free() wait for a long time due to the function
> cond_resched() at host side.
> 
> If using SEV/ES and the kernel provided by AMD, the overhead will
> become even much larger.
> 
> With ioeventfd, the thread for host to do balloon inflation will
> be seperated from the vcpu thread, leading to better performance
> for the whole process of balloon inflation.

I didn't really look into ieventfd so far, but from what I understand, 
an iothread will take care of processing these events, instead of the 
vcpu thread.

Do you have some performance numbers?

I'll note that whatever thread calls madvise(), we're holding the BQL in 
that code and will block any other threads that need it.


> 
> Signed-off-by: Zheyun Shen <szy0127@sjtu.edu.cn>
> ---
>   hw/virtio/virtio-balloon-pci.c | 7 +++++++
>   1 file changed, 7 insertions(+)
> 
> diff --git a/hw/virtio/virtio-balloon-pci.c b/hw/virtio/virtio-balloon-pci.c
> index ce2645b..7195322 100644
> --- a/hw/virtio/virtio-balloon-pci.c
> +++ b/hw/virtio/virtio-balloon-pci.c
> @@ -35,6 +35,12 @@ struct VirtIOBalloonPCI {
>       VirtIOBalloon vdev;
>   };
>   
> +static Property virtio_balloon_properties[] = {
> +    DEFINE_PROP_BIT("ioeventfd", VirtIOPCIProxy, flags,
> +            VIRTIO_PCI_FLAG_USE_IOEVENTFD_BIT, false),
> +    DEFINE_PROP_END_OF_LIST(),
> +};
> +
>   static void virtio_balloon_pci_realize(VirtIOPCIProxy *vpci_dev, Error **errp)
>   {
>       VirtIOBalloonPCI *dev = VIRTIO_BALLOON_PCI(vpci_dev);
> @@ -51,6 +57,7 @@ static void virtio_balloon_pci_class_init(ObjectClass *klass, void *data)
>       PCIDeviceClass *pcidev_k = PCI_DEVICE_CLASS(klass);
>       k->realize = virtio_balloon_pci_realize;
>       set_bit(DEVICE_CATEGORY_MISC, dc->categories);
> +    device_class_set_props(dc, virtio_balloon_properties);
>       pcidev_k->vendor_id = PCI_VENDOR_ID_REDHAT_QUMRANET;
>       pcidev_k->device_id = PCI_DEVICE_ID_VIRTIO_BALLOON;
>       pcidev_k->revision = VIRTIO_PCI_ABI_VERSION;

-- 
Cheers,

David / dhildenb