[Qemu-devel] Re: [PATCH] vhost: force vhost off for non-MSI guests

[Alex Williamson <alex.williamson@redhat.com>]

On Thu, 2011-01-20 at 17:35 +0200, Michael S. Tsirkin wrote:
> When MSI is off, each interrupt needs to be bounced through the io
> thread when it's set/cleared, so vhost-net causes more context switches and
> higher CPU utilization than userspace virtio which handles networking in
> the same thread.
> 
> We'll need to fix this by adding level irq support in kvm irqfd,
> for now disable vhost-net in these configurations.

We need this if we want avoid bouncing vfio INtx through qemu too.

> Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
> ---
> 
> I need to report some error from virtio-pci
> that would be handled specially (disable but don't
> report an error) so I wanted one that's never likely to be used by a
> userspace ioctl. I selected ERANGE but it'd
> be easy to switch to something else. Comments?
> 
>  hw/vhost.c      |    4 +++-
>  hw/virtio-net.c |    6 ++++--
>  hw/virtio-pci.c |    3 +++
>  hw/virtio.h     |    2 ++
>  4 files changed, 12 insertions(+), 3 deletions(-)
> 
> diff --git a/hw/vhost.c b/hw/vhost.c
> index 1d09ed0..c79765a 100644
> --- a/hw/vhost.c
> +++ b/hw/vhost.c
> @@ -649,7 +649,9 @@ int vhost_dev_start(struct vhost_dev *hdev, VirtIODevice *vdev)
>  
>      r = vdev->binding->set_guest_notifiers(vdev->binding_opaque, true);
>      if (r < 0) {
> -        fprintf(stderr, "Error binding guest notifier: %d\n", -r);
> +	if (r != -EVIRTIO_DISABLED) {
> +		fprintf(stderr, "Error binding guest notifier: %d\n", -r);
> +	}

style - the above is tab indented.

>          goto fail_notifiers;
>      }
>  
> diff --git a/hw/virtio-net.c b/hw/virtio-net.c
> index ccb3e63..5de3fee 100644
> --- a/hw/virtio-net.c
> +++ b/hw/virtio-net.c
> @@ -121,8 +121,10 @@ static void virtio_net_vhost_status(VirtIONet *n, uint8_t status)
>      if (!n->vhost_started) {
>          int r = vhost_net_start(tap_get_vhost_net(n->nic->nc.peer), &n->vdev);
>          if (r < 0) {
> -            error_report("unable to start vhost net: %d: "
> -                         "falling back on userspace virtio", -r);
> +            if (r != -EVIRTIO_DISABLED) {
> +                error_report("unable to start vhost net: %d: "
> +                             "falling back on userspace virtio", -r);
> +            }
>          } else {
>              n->vhost_started = 1;
>          }
> diff --git a/hw/virtio-pci.c b/hw/virtio-pci.c
> index dd8887a..dbf4be0 100644
> --- a/hw/virtio-pci.c
> +++ b/hw/virtio-pci.c
> @@ -628,6 +628,9 @@ static int virtio_pci_set_guest_notifier(void *opaque, int n, bool assign)
>      EventNotifier *notifier = virtio_queue_get_guest_notifier(vq);
>  
>      if (assign) {
> +        if (!msix_enabled(&proxy->pci_dev)) {
> +            return -EVIRTIO_DISABLED;
> +        }
>          int r = event_notifier_init(notifier, 0);
>          if (r < 0) {
>              return r;
> diff --git a/hw/virtio.h b/hw/virtio.h
> index d8546d5..53bbdba 100644
> --- a/hw/virtio.h
> +++ b/hw/virtio.h
> @@ -98,6 +98,8 @@ typedef struct {
>      void (*vmstate_change)(void * opaque, bool running);
>  } VirtIOBindings;
>  
> +#define EVIRTIO_DISABLED ERANGE
> +
>  #define VIRTIO_PCI_QUEUE_MAX 64
>  
>  #define VIRTIO_NO_VECTOR 0xffff

I'm not a fan of having this special return value.  Why doesn't
virtio-pci only setup the set_guest_notifiers function pointer when msix
is enabled?  If not that, it could at least expose some
virtio_foo_enabled() type feature that vhost could check.  Thanks,

Alex