Re: [Qemu-devel] [RFC] net: 'Remove vhostforce option in addition to vhost param'

[Pankaj Gupta <pagupta@redhat.com>]

Hi,

Could you please help/guide me here. As suggested by Jason I did other changes also.
But when I did testing still virtio-net.c functions like 'receive()' gets called when vhost is 'ON'.

I want to know is there anything I am missing here or is this expected behaviour?

I was also searching for "kvm eventfd support for injecting level-triggered interrupts", For non-MSIX
guests, can we remove vhost-force unless we have this feature?

Best regards,
Pankaj

----- Original Message -----
> From: "Jason Wang" <jasowang@redhat.com>
> To: "Pankaj Gupta" <pagupta@redhat.com>
> Cc: "Michael S. Tsirkin" <mst@redhat.com>, qemu-devel@nongnu.org, stefanha@redhat.com, aliguori@amazon.com
> Sent: Sunday, February 15, 2015 8:11:29 AM
> Subject: Re: [Qemu-devel] [RFC] net: 'Remove vhostforce option in	addition	to vhost param'
> 
> 
> 
> On Thu, Feb 12, 2015 at 11:41 PM, Pankaj Gupta <pagupta@redhat.com>
> wrote:
> >  
> >>  
> >>  > On Thu, Feb 12, 2015 at 11:50:05AM +0530, Pankaj Gupta wrote:
> >>  > >  vhostforce was added to enable use of vhost when
> >>  > >  guest don't have MSI-X support.
> >>  > >  Now, we have scenarios which  dont use interrupts
> >>  > >  like DPDK and still use vhost. Also, performance of
> >>  > >  guests without MSI-X support is getting less popular.
> >>  > > 
> >>  > >  Its ok to remove this extra option and enable vhost
> >>  > >  on the basis of vhost=ON/OFF.
> >>  > > 
> >>  > > Signed-off-by: Pankaj Gupta <pagupta@redhat.com>
> >>  > 
> >>  > The patch doesn't seem to do what it says.
> >>  > Did you try with a non MSIX guest and vhost=on, to check that
> >>  > it actually runs vhost and not userspace virtio?
> >>  
> >>  No, I have not. I just did basic tested a new guest without
> >> vhostforce.
> >>  I will test non-MSIX guest and share the result.
> > 
> > I tested this with RHEL 4 guest which don't have MSI-X. Though vhost
> > gets
> > created but still userspace virtio-net code executes.
> > 
> > So, vhostforce was added to disable vhost for non-MSI guest?
> 
> In fact to enable vhost.
> > 
> > I took the idea from KVM/Networking todo list.
> > 
> > Do we have some other dependency before we want to remove vhostforce?
> >   
> 
> You may want to take a look at the vhost_dev->force and
> vhost_dev_query().
> > 
> >>  
> >>  > 
> >>  > > ---
> >>  > >  net/tap.c        |  4 +---
> >>  > >  net/vhost-user.c | 16 ++--------------
> >>  > >  2 files changed, 3 insertions(+), 17 deletions(-)
> >>  > > 
> >>  > > diff --git a/net/tap.c b/net/tap.c
> >>  > > index 1fe0edf..bd2efa9 100644
> >>  > > --- a/net/tap.c
> >>  > > +++ b/net/tap.c
> >>  > > @@ -634,13 +634,11 @@ static int net_init_tap_one(const
> >> NetdevTapOptions
> >>  > > *tap, NetClientState *peer,
> >>  > >          }
> >>  > >      }
> >>  > >  
> >>  > > -    if (tap->has_vhost ? tap->vhost :
> >>  > > -        vhostfdname || (tap->has_vhostforce &&
> >> tap->vhostforce)) {
> >>  > > +    if (tap->has_vhost ? tap->vhost : vhostfdname) {
> >>  > >          VhostNetOptions options;
> >>  > >  
> >>  > >          options.backend_type = VHOST_BACKEND_TYPE_KERNEL;
> >>  > >          options.net_backend = &s->nc;
> >>  > > -        options.force = tap->has_vhostforce && tap->vhostforce;
> >>  > >  
> >>  > >          if (tap->has_vhostfd || tap->has_vhostfds) {
> >>  > >              vhostfd = monitor_handle_fd_param(cur_mon,
> >> vhostfdname);
> >>  > > diff --git a/net/vhost-user.c b/net/vhost-user.c
> >>  > > index 24e050c..d2d7bf2 100644
> >>  > > --- a/net/vhost-user.c
> >>  > > +++ b/net/vhost-user.c
> >>  > > @@ -18,7 +18,6 @@
> >>  > >  typedef struct VhostUserState {
> >>  > >      NetClientState nc;
> >>  > >      CharDriverState *chr;
> >>  > > -    bool vhostforce;
> >>  > >      VHostNetState *vhost_net;
> >>  > >  } VhostUserState;
> >>  > >  
> >>  > > @@ -51,7 +50,6 @@ static int vhost_user_start(VhostUserState *s)
> >>  > >      options.backend_type = VHOST_BACKEND_TYPE_USER;
> >>  > >      options.net_backend = &s->nc;
> >>  > >      options.opaque = s->chr;
> >>  > > -    options.force = s->vhostforce;
> >>  > >  
> >>  > >      s->vhost_net = vhost_net_init(&options);
> >>  > >  
> >>  > > @@ -133,8 +131,7 @@ static void net_vhost_user_event(void
> >> *opaque, int
> >>  > > event)
> >>  > >  }
> >>  > >  
> >>  > >  static int net_vhost_user_init(NetClientState *peer, const
> >> char *device,
> >>  > > -                               const char *name,
> >> CharDriverState *chr,
> >>  > > -                               bool vhostforce)
> >>  > > +                               const char *name,
> >> CharDriverState *chr)
> >>  > >  {
> >>  > >      NetClientState *nc;
> >>  > >      VhostUserState *s;
> >>  > > @@ -149,7 +146,6 @@ static int
> >> net_vhost_user_init(NetClientState *peer,
> >>  > > const char *device,
> >>  > >      /* We don't provide a receive callback */
> >>  > >      s->nc.receive_disabled = 1;
> >>  > >      s->chr = chr;
> >>  > > -    s->vhostforce = vhostforce;
> >>  > >  
> >>  > >      qemu_chr_add_handlers(s->chr, NULL, NULL,
> >> net_vhost_user_event, s);
> >>  > >  
> >>  > > @@ -230,7 +226,6 @@ int net_init_vhost_user(const
> >> NetClientOptions *opts,
> >>  > > const char *name,
> >>  > >  {
> >>  > >      const NetdevVhostUserOptions *vhost_user_opts;
> >>  > >      CharDriverState *chr;
> >>  > > -    bool vhostforce;
> >>  > >  
> >>  > >      assert(opts->kind == NET_CLIENT_OPTIONS_KIND_VHOST_USER);
> >>  > >      vhost_user_opts = opts->vhost_user;
> >>  > > @@ -247,12 +242,5 @@ int net_init_vhost_user(const
> >> NetClientOptions
> >>  > > *opts,
> >>  > > const char *name,
> >>  > >          return -1;
> >>  > >      }
> >>  > >  
> >>  > > -    /* vhostforce for non-MSIX */
> >>  > > -    if (vhost_user_opts->has_vhostforce) {
> >>  > > -        vhostforce = vhost_user_opts->vhostforce;
> >>  > > -    } else {
> >>  > > -        vhostforce = false;
> >>  > > -    }
> >>  > > -
> >>  > > -    return net_vhost_user_init(peer, "vhost_user", name, chr,
> >>  > > vhostforce);
> >>  > > +    return net_vhost_user_init(peer, "vhost_user", name, chr);
> >>  > >  }
> >>  > > --
> >>  > > 1.8.3.1
> >>  > 
> >>  > 
> >>  
> > 
> 
>