Virtualization - Jan 2011 - [PATCH 1/2] virtio: console: Wake up outvq on host notifications

The outvq needs to be woken up on host notifications so that buffers
consumed by the host can be reclaimed, outvq freed, and application
writes may proceed again.

The need for this is now finally noticed when I have qemu patches ready
to use nonblocking IO and flow control.

CC: Hans de Goede <hdegoede at redhat.com>
CC: stable at kernel.org
Signed-off-by: Amit Shah <amit.shah at redhat.com>
---
 drivers/tty/hvc/virtio_console.c |   15 +++++++++++++--
 1 files changed, 13 insertions(+), 2 deletions(-)

diff --git a/drivers/tty/hvc/virtio_console.c b/drivers/tty/hvc/virtio_console.c
index c4b0255..1068095 100644
--- a/drivers/tty/hvc/virtio_console.c
+++ b/drivers/tty/hvc/virtio_console.c
@@ -1473,6 +1473,17 @@ static void control_work_handler(struct work_struct
*work)
 	spin_unlock(&portdev->cvq_lock);
 }
 
+static void out_intr(struct virtqueue *vq)
+{
+	struct port *port;
+
+	port = find_port_by_vq(vq->vdev->priv, vq);
+	if (!port)
+		return;
+
+	wake_up_interruptible(&port->waitqueue);
+}
+
 static void in_intr(struct virtqueue *vq)
 {
 	struct port *port;
@@ -1577,7 +1588,7 @@ static int init_vqs(struct ports_device *portdev)
 	 */
 	j = 0;
 	io_callbacks[j] = in_intr;
-	io_callbacks[j + 1] = NULL;
+	io_callbacks[j + 1] = out_intr;
 	io_names[j] = "input";
 	io_names[j + 1] = "output";
 	j += 2;
@@ -1591,7 +1602,7 @@ static int init_vqs(struct ports_device *portdev)
 		for (i = 1; i < nr_ports; i++) {
 			j += 2;
 			io_callbacks[j] = in_intr;
-			io_callbacks[j + 1] = NULL;
+			io_callbacks[j + 1] = out_intr;
 			io_names[j] = "input";
 			io_names[j + 1] = "output";
 		}
-- 
1.7.3.5

Signed-off-by: Amit Shah <amit.shah at redhat.com>
---
 drivers/tty/hvc/virtio_console.c |    3 ++-
 include/linux/virtio_console.h   |    3 ++-
 2 files changed, 4 insertions(+), 2 deletions(-)

diff --git a/drivers/tty/hvc/virtio_console.c b/drivers/tty/hvc/virtio_console.c
index 1068095..f03338f 100644
--- a/drivers/tty/hvc/virtio_console.c
+++ b/drivers/tty/hvc/virtio_console.c
@@ -1,6 +1,7 @@
 /*
  * Copyright (C) 2006, 2007, 2009 Rusty Russell, IBM Corporation
- * Copyright (C) 2009, 2010 Red Hat, Inc.
+ * Copyright (C) 2009, 2010, 2011 Red Hat, Inc.
+ * Copyright (C) 2009, 2010, 2011 Amit Shah <amit.shah at redhat.com>
  *
  * This program is free software; you can redistribute it and/or modify
  * it under the terms of the GNU General Public License as published by
diff --git a/include/linux/virtio_console.h b/include/linux/virtio_console.h
index a85064d..e4d3335 100644
--- a/include/linux/virtio_console.h
+++ b/include/linux/virtio_console.h
@@ -7,7 +7,8 @@
  * This header, excluding the #ifdef __KERNEL__ part, is BSD licensed so
  * anyone can use the definitions to implement compatible drivers/servers.
  *
- * Copyright (C) Red Hat, Inc., 2009, 2010
+ * Copyright (C) Red Hat, Inc., 2009, 2010, 2011
+ * Copyright (C) Amit Shah <amit.shah at redhat.com>, 2009, 2010, 2011
  */
 
 /* Feature bits */
-- 
1.7.3.5

Acked-by: Hans de Goede <hdegoede at redhat.com>

On 01/31/2011 08:36 AM, Amit Shah wrote:
> The outvq needs to be woken up on host notifications so that buffers
> consumed by the host can be reclaimed, outvq freed, and application
> writes may proceed again.
>
> The need for this is now finally noticed when I have qemu patches ready
> to use nonblocking IO and flow control.
>
> CC: Hans de Goede<hdegoede at redhat.com>
> CC: stable at kernel.org
> Signed-off-by: Amit Shah<amit.shah at redhat.com>
> ---
>   drivers/tty/hvc/virtio_console.c |   15 +++++++++++++--
>   1 files changed, 13 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/tty/hvc/virtio_console.c
b/drivers/tty/hvc/virtio_console.c
> index c4b0255..1068095 100644
> --- a/drivers/tty/hvc/virtio_console.c
> +++ b/drivers/tty/hvc/virtio_console.c
> @@ -1473,6 +1473,17 @@ static void control_work_handler(struct work_struct
*work)
>   	spin_unlock(&portdev->cvq_lock);
>   }
>
> +static void out_intr(struct virtqueue *vq)
> +{
> +	struct port *port;
> +
> +	port = find_port_by_vq(vq->vdev->priv, vq);
> +	if (!port)
> +		return;
> +
> +	wake_up_interruptible(&port->waitqueue);
> +}
> +
>   static void in_intr(struct virtqueue *vq)
>   {
>   	struct port *port;
> @@ -1577,7 +1588,7 @@ static int init_vqs(struct ports_device *portdev)
>   	 */
>   	j = 0;
>   	io_callbacks[j] = in_intr;
> -	io_callbacks[j + 1] = NULL;
> +	io_callbacks[j + 1] = out_intr;
>   	io_names[j] = "input";
>   	io_names[j + 1] = "output";
>   	j += 2;
> @@ -1591,7 +1602,7 @@ static int init_vqs(struct ports_device *portdev)
>   		for (i = 1; i<  nr_ports; i++) {
>   			j += 2;
>   			io_callbacks[j] = in_intr;
> -			io_callbacks[j + 1] = NULL;
> +			io_callbacks[j + 1] = out_intr;
>   			io_names[j] = "input";
>   			io_names[j + 1] = "output";
>   		}

On Mon, 31 Jan 2011 06:06:36 pm Amit Shah wrote:
> The outvq needs to be woken up on host notifications so that buffers
> consumed by the host can be reclaimed, outvq freed, and application
> writes may proceed again.
> 
> The need for this is now finally noticed when I have qemu patches ready
> to use nonblocking IO and flow control.
What about -stable?

Thanks,
Rusty.

On Mon, 31 Jan 2011 08:33:15 pm Rusty Russell wrote:
> On Mon, 31 Jan 2011 06:06:36 pm Amit Shah wrote:
> > The outvq needs to be woken up on host notifications so that buffers
> > consumed by the host can be reclaimed, outvq freed, and application
> > writes may proceed again.
> > 
> > The need for this is now finally noticed when I have qemu patches
ready
> > to use nonblocking IO and flow control.
> 
> What about -stable?
Sorry, I missed the CC...

Thanks,
Rusty.