[Spice-devel] [PATCH 19/24] server/red_channel: red_channel_pipe_get: enforce ack window
Alon Levy
alevy at redhat.com
Wed Feb 2 13:28:12 PST 2011
On Wed, Feb 02, 2011 at 07:53:28PM +0100, Marc-André Lureau wrote:
> nack, see comment in patch #22.
>
> On Wed, Jan 19, 2011 at 7:08 PM, Alon Levy <alevy at redhat.com> wrote:
> > We were not setting blocked flag if too many messages weren't
> > acked (twice the ack window size).
> > From red_worker.
> > ---
> > server/red_channel.c | 9 +++++----
> > 1 files changed, 5 insertions(+), 4 deletions(-)
> >
> > diff --git a/server/red_channel.c b/server/red_channel.c
> > index bfa6aaf..990b9d9 100644
> > --- a/server/red_channel.c
> > +++ b/server/red_channel.c
> > @@ -570,17 +570,18 @@ void red_channel_pipe_add_type(RedChannel *channel, int pipe_item_type)
> > red_channel_push(channel);
> > }
> >
> > -static PipeItem *red_channel_pipe_get(RedChannel *channel)
> > +static inline PipeItem *red_channel_pipe_get(RedChannel *channel)
>
> Why do you put inline here?
Assumed but didn't test that this improves performance. (btw,
we have 140 inlines in server/red_worker.[ch])
>
> > {
> > PipeItem *item;
> >
> > if (!channel || channel->send_data.blocked ||
> > - (channel->handle_acks &&
> > - (channel->ack_data.messages_window > channel->ack_data.client_window * 2)) ||
> > !(item = (PipeItem *)ring_get_tail(&channel->pipe))) {
> > return NULL;
> > }
> > -
> > + if (channel->handle_acks && (channel->ack_data.messages_window > channel->ack_data.client_window * 2)) {
> > + channel->send_data.blocked = TRUE;
> > + return NULL;
> > + }
> > --channel->pipe_size;
> > ring_remove(&item->link);
> > return item;
> > --
> > 1.7.3.4
> >
> > _______________________________________________
> > Spice-devel mailing list
> > Spice-devel at lists.freedesktop.org
> > http://lists.freedesktop.org/mailman/listinfo/spice-devel
> >
>
>
>
> --
> Marc-André Lureau
More information about the Spice-devel
mailing list