[dpdk-dev] [PATCH] vhost: call rte_vhost_enable_guest_notification only on enabled queues

[dpdk-dev] [PATCH] vhost: call rte_vhost_enable_guest_notification only on enabled queues

[Rich Lane]

If the vhost PMD were configured with more queues than the guest, the old
code would segfault in rte_vhost_enable_guest_notification due to a NULL
virtqueue pointer.

Fixes: ee584e9710b9 ("vhost: add driver on top of the library")
Signed-off-by: Rich Lane <rich.lane@bigswitch.com>
---
 drivers/net/vhost/rte_eth_vhost.c | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)

diff --git a/drivers/net/vhost/rte_eth_vhost.c b/drivers/net/vhost/rte_eth_vhost.c
index b1eb082..310cbef 100644
--- a/drivers/net/vhost/rte_eth_vhost.c
+++ b/drivers/net/vhost/rte_eth_vhost.c
@@ -265,7 +265,6 @@ new_device(struct virtio_net *dev)
 		vq->device = dev;
 		vq->internal = internal;
 		vq->port = eth_dev->data->port_id;
-		rte_vhost_enable_guest_notification(dev, vq->virtqueue_id, 0);
 	}
 	for (i = 0; i < eth_dev->data->nb_tx_queues; i++) {
 		vq = eth_dev->data->tx_queues[i];
@@ -274,9 +273,11 @@ new_device(struct virtio_net *dev)
 		vq->device = dev;
 		vq->internal = internal;
 		vq->port = eth_dev->data->port_id;
-		rte_vhost_enable_guest_notification(dev, vq->virtqueue_id, 0);
 	}
 
+	for (i = 0; i < dev->virt_qp_nb * VIRTIO_QNUM; i++)
+		rte_vhost_enable_guest_notification(dev, i, 0);
+
 	dev->flags |= VIRTIO_DEV_RUNNING;
 	dev->priv = eth_dev;
 	eth_dev->data->dev_link.link_status = ETH_LINK_UP;
-- 
1.9.1

Re: [dpdk-dev] [PATCH] vhost: call rte_vhost_enable_guest_notification only on enabled queues

[Loftus, Ciara]

> 
> If the vhost PMD were configured with more queues than the guest, the old
> code would segfault in rte_vhost_enable_guest_notification due to a NULL
> virtqueue pointer.
> 
> Fixes: ee584e9710b9 ("vhost: add driver on top of the library")
> Signed-off-by: Rich Lane <rich.lane@bigswitch.com>
> ---
>  drivers/net/vhost/rte_eth_vhost.c | 5 +++--
>  1 file changed, 3 insertions(+), 2 deletions(-)
> 
> diff --git a/drivers/net/vhost/rte_eth_vhost.c
> b/drivers/net/vhost/rte_eth_vhost.c
> index b1eb082..310cbef 100644
> --- a/drivers/net/vhost/rte_eth_vhost.c
> +++ b/drivers/net/vhost/rte_eth_vhost.c
> @@ -265,7 +265,6 @@ new_device(struct virtio_net *dev)
>  		vq->device = dev;
>  		vq->internal = internal;
>  		vq->port = eth_dev->data->port_id;
> -		rte_vhost_enable_guest_notification(dev, vq-
> >virtqueue_id, 0);
>  	}
>  	for (i = 0; i < eth_dev->data->nb_tx_queues; i++) {
>  		vq = eth_dev->data->tx_queues[i];
> @@ -274,9 +273,11 @@ new_device(struct virtio_net *dev)
>  		vq->device = dev;
>  		vq->internal = internal;
>  		vq->port = eth_dev->data->port_id;
> -		rte_vhost_enable_guest_notification(dev, vq-
> >virtqueue_id, 0);
>  	}
> 
> +	for (i = 0; i < dev->virt_qp_nb * VIRTIO_QNUM; i++)
> +		rte_vhost_enable_guest_notification(dev, i, 0);
> +
>  	dev->flags |= VIRTIO_DEV_RUNNING;
>  	dev->priv = eth_dev;
>  	eth_dev->data->dev_link.link_status = ETH_LINK_UP;
> --
> 1.9.1

I see the same issue, and verified that this patch solves it. Thanks!

Tested-by: Ciara Loftus <ciara.loftus@intel.com>

Thanks,
Ciara

Re: [dpdk-dev] [PATCH] vhost: call rte_vhost_enable_guest_notification only on enabled queues

[Tan, Jianfeng]

Hi,

On 4/7/2016 8:29 AM, Rich Lane wrote:
> If the vhost PMD were configured with more queues than the guest, the old
> code would segfault in rte_vhost_enable_guest_notification due to a NULL
> virtqueue pointer.
>
> Fixes: ee584e9710b9 ("vhost: add driver on top of the library")
> Signed-off-by: Rich Lane <rich.lane@bigswitch.com>
> ---
>   drivers/net/vhost/rte_eth_vhost.c | 5 +++--
>   1 file changed, 3 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/net/vhost/rte_eth_vhost.c b/drivers/net/vhost/rte_eth_vhost.c
> index b1eb082..310cbef 100644
> --- a/drivers/net/vhost/rte_eth_vhost.c
> +++ b/drivers/net/vhost/rte_eth_vhost.c
> @@ -265,7 +265,6 @@ new_device(struct virtio_net *dev)
>   		vq->device = dev;
>   		vq->internal = internal;
>   		vq->port = eth_dev->data->port_id;
> -		rte_vhost_enable_guest_notification(dev, vq->virtqueue_id, 0);
>   	}
>   	for (i = 0; i < eth_dev->data->nb_tx_queues; i++) {
>   		vq = eth_dev->data->tx_queues[i];
> @@ -274,9 +273,11 @@ new_device(struct virtio_net *dev)
>   		vq->device = dev;
>   		vq->internal = internal;
>   		vq->port = eth_dev->data->port_id;
> -		rte_vhost_enable_guest_notification(dev, vq->virtqueue_id, 0);
>   	}
>   
> +	for (i = 0; i < dev->virt_qp_nb * VIRTIO_QNUM; i++)
> +		rte_vhost_enable_guest_notification(dev, i, 0);
> +
>   	dev->flags |= VIRTIO_DEV_RUNNING;
>   	dev->priv = eth_dev;
>   	eth_dev->data->dev_link.link_status = ETH_LINK_UP;

Just one question, when qemu starts a vm, usually, only one queue is 
enabled, then only 1 tx and 1 rx are called 
rte_vhost_enable_guest_notification; but after system is up, we use 
"ethtool -K eth0 combined x" to enable multiqueues, there's no chance to 
call rte_vhost_enable_guest_notification for other queues, right?

Thanks,
Jianfeng

Re: [dpdk-dev] [PATCH] vhost: call rte_vhost_enable_guest_notification only on enabled queues

[Loftus, Ciara]

> On 4/7/2016 8:29 AM, Rich Lane wrote:
> > If the vhost PMD were configured with more queues than the guest, the
> old
> > code would segfault in rte_vhost_enable_guest_notification due to a NULL
> > virtqueue pointer.
> >
> > Fixes: ee584e9710b9 ("vhost: add driver on top of the library")
> > Signed-off-by: Rich Lane <rich.lane@bigswitch.com>
> > ---
> >   drivers/net/vhost/rte_eth_vhost.c | 5 +++--
> >   1 file changed, 3 insertions(+), 2 deletions(-)
> >
> > diff --git a/drivers/net/vhost/rte_eth_vhost.c
> b/drivers/net/vhost/rte_eth_vhost.c
> > index b1eb082..310cbef 100644
> > --- a/drivers/net/vhost/rte_eth_vhost.c
> > +++ b/drivers/net/vhost/rte_eth_vhost.c
> > @@ -265,7 +265,6 @@ new_device(struct virtio_net *dev)
> >   		vq->device = dev;
> >   		vq->internal = internal;
> >   		vq->port = eth_dev->data->port_id;
> > -		rte_vhost_enable_guest_notification(dev, vq-
> >virtqueue_id, 0);
> >   	}
> >   	for (i = 0; i < eth_dev->data->nb_tx_queues; i++) {
> >   		vq = eth_dev->data->tx_queues[i];
> > @@ -274,9 +273,11 @@ new_device(struct virtio_net *dev)
> >   		vq->device = dev;
> >   		vq->internal = internal;
> >   		vq->port = eth_dev->data->port_id;
> > -		rte_vhost_enable_guest_notification(dev, vq-
> >virtqueue_id, 0);
> >   	}
> >
> > +	for (i = 0; i < dev->virt_qp_nb * VIRTIO_QNUM; i++)
> > +		rte_vhost_enable_guest_notification(dev, i, 0);
> > +
> >   	dev->flags |= VIRTIO_DEV_RUNNING;
> >   	dev->priv = eth_dev;
> >   	eth_dev->data->dev_link.link_status = ETH_LINK_UP;
> 
> Just one question, when qemu starts a vm, usually, only one queue is
> enabled, then only 1 tx and 1 rx are called
> rte_vhost_enable_guest_notification; but after system is up, we use
> "ethtool -K eth0 combined x" to enable multiqueues, there's no chance to
> call rte_vhost_enable_guest_notification for other queues, right?

As far as I know, virt_qp_nb will report the number of queues available, regardless of their state enabled/disabled. So for example if we have 4 queues, but only one enabled, virt_qp_nb should still = 4 and rte_vhost_enable_guest_notification() will be called for all of these queues.

Thanks,
Ciara

> 
> Thanks,
> Jianfeng

Re: [dpdk-dev] [PATCH] vhost: call rte_vhost_enable_guest_notification only on enabled queues

[Yuanhan Liu]

On Thu, Apr 07, 2016 at 03:29:32PM +0000, Loftus, Ciara wrote:
> > On 4/7/2016 8:29 AM, Rich Lane wrote:
> > > If the vhost PMD were configured with more queues than the guest, the
> > old
> > > code would segfault in rte_vhost_enable_guest_notification due to a NULL
> > > virtqueue pointer.
> > >
> > > Fixes: ee584e9710b9 ("vhost: add driver on top of the library")
> > > Signed-off-by: Rich Lane <rich.lane@bigswitch.com>
> > > ---
> > >   drivers/net/vhost/rte_eth_vhost.c | 5 +++--
> > >   1 file changed, 3 insertions(+), 2 deletions(-)
> > >
> > > diff --git a/drivers/net/vhost/rte_eth_vhost.c
> > b/drivers/net/vhost/rte_eth_vhost.c
> > > index b1eb082..310cbef 100644
> > > --- a/drivers/net/vhost/rte_eth_vhost.c
> > > +++ b/drivers/net/vhost/rte_eth_vhost.c
> > > @@ -265,7 +265,6 @@ new_device(struct virtio_net *dev)
> > >   		vq->device = dev;
> > >   		vq->internal = internal;
> > >   		vq->port = eth_dev->data->port_id;
> > > -		rte_vhost_enable_guest_notification(dev, vq-
> > >virtqueue_id, 0);
> > >   	}
> > >   	for (i = 0; i < eth_dev->data->nb_tx_queues; i++) {
> > >   		vq = eth_dev->data->tx_queues[i];
> > > @@ -274,9 +273,11 @@ new_device(struct virtio_net *dev)
> > >   		vq->device = dev;
> > >   		vq->internal = internal;
> > >   		vq->port = eth_dev->data->port_id;
> > > -		rte_vhost_enable_guest_notification(dev, vq-
> > >virtqueue_id, 0);
> > >   	}
> > >
> > > +	for (i = 0; i < dev->virt_qp_nb * VIRTIO_QNUM; i++)
> > > +		rte_vhost_enable_guest_notification(dev, i, 0);
> > > +
> > >   	dev->flags |= VIRTIO_DEV_RUNNING;
> > >   	dev->priv = eth_dev;
> > >   	eth_dev->data->dev_link.link_status = ETH_LINK_UP;
> > 
> > Just one question, when qemu starts a vm, usually, only one queue is
> > enabled, then only 1 tx and 1 rx are called
> > rte_vhost_enable_guest_notification; but after system is up, we use
> > "ethtool -K eth0 combined x" to enable multiqueues, there's no chance to
> > call rte_vhost_enable_guest_notification for other queues, right?
> 
> As far as I know, virt_qp_nb will report the number of queues available, regardless of their state enabled/disabled. So for example if we have 4 queues, but only one enabled, virt_qp_nb should still = 4 and rte_vhost_enable_guest_notification() will be called for all of these queues.

Yes.

	--yliu

Re: [dpdk-dev] [PATCH] vhost: call rte_vhost_enable_guest_notification only on enabled queues

[Yuanhan Liu]

On Wed, Apr 06, 2016 at 05:29:06PM -0700, Rich Lane wrote:
> If the vhost PMD were configured with more queues than the guest, the old
> code would segfault in rte_vhost_enable_guest_notification due to a NULL
> virtqueue pointer.
> 
> Fixes: ee584e9710b9 ("vhost: add driver on top of the library")
> Signed-off-by: Rich Lane <rich.lane@bigswitch.com>

Acked-by: Yuanhan Liu <yuanhan.liu@linux.intel.com>

Thanks.

	--yliu

Re: [dpdk-dev] [PATCH] vhost: call rte_vhost_enable_guest_notification only on enabled queues

[Tan, Jianfeng]

On 4/7/2016 11:29 PM, Loftus, Ciara wrote:
>> On 4/7/2016 8:29 AM, Rich Lane wrote:
>>> If the vhost PMD were configured with more queues than the guest, the
>> old
>>> code would segfault in rte_vhost_enable_guest_notification due to a NULL
>>> virtqueue pointer.
>>>
>>> Fixes: ee584e9710b9 ("vhost: add driver on top of the library")
>>> Signed-off-by: Rich Lane <rich.lane@bigswitch.com>
>>> ---
>>>    drivers/net/vhost/rte_eth_vhost.c | 5 +++--
>>>    1 file changed, 3 insertions(+), 2 deletions(-)
>>>
>>> diff --git a/drivers/net/vhost/rte_eth_vhost.c
>> b/drivers/net/vhost/rte_eth_vhost.c
>>> index b1eb082..310cbef 100644
>>> --- a/drivers/net/vhost/rte_eth_vhost.c
>>> +++ b/drivers/net/vhost/rte_eth_vhost.c
>>> @@ -265,7 +265,6 @@ new_device(struct virtio_net *dev)
>>>    		vq->device = dev;
>>>    		vq->internal = internal;
>>>    		vq->port = eth_dev->data->port_id;
>>> -		rte_vhost_enable_guest_notification(dev, vq-
>>> virtqueue_id, 0);
>>>    	}
>>>    	for (i = 0; i < eth_dev->data->nb_tx_queues; i++) {
>>>    		vq = eth_dev->data->tx_queues[i];
>>> @@ -274,9 +273,11 @@ new_device(struct virtio_net *dev)
>>>    		vq->device = dev;
>>>    		vq->internal = internal;
>>>    		vq->port = eth_dev->data->port_id;
>>> -		rte_vhost_enable_guest_notification(dev, vq-
>>> virtqueue_id, 0);
>>>    	}
>>>
>>> +	for (i = 0; i < dev->virt_qp_nb * VIRTIO_QNUM; i++)
>>> +		rte_vhost_enable_guest_notification(dev, i, 0);
>>> +
>>>    	dev->flags |= VIRTIO_DEV_RUNNING;
>>>    	dev->priv = eth_dev;
>>>    	eth_dev->data->dev_link.link_status = ETH_LINK_UP;
>> Just one question, when qemu starts a vm, usually, only one queue is
>> enabled, then only 1 tx and 1 rx are called
>> rte_vhost_enable_guest_notification; but after system is up, we use
>> "ethtool -K eth0 combined x" to enable multiqueues, there's no chance to
>> call rte_vhost_enable_guest_notification for other queues, right?
> As far as I know, virt_qp_nb will report the number of queues available, regardless of their state enabled/disabled. So for example if we have 4 queues, but only one enabled, virt_qp_nb should still = 4 and rte_vhost_enable_guest_notification() will be called for all of these queues.

Oh, yes. I failed to really get multiqueue configured right, which leads 
to the wrong conclusion. Sorry.

Thanks,
Jianfeng

>
> Thanks,
> Ciara
>
>> Thanks,
>> Jianfeng

Re: [dpdk-dev] [PATCH] vhost: call rte_vhost_enable_guest_notification only on enabled queues

[Thomas Monjalon]

> > If the vhost PMD were configured with more queues than the guest, the old
> > code would segfault in rte_vhost_enable_guest_notification due to a NULL
> > virtqueue pointer.
> > 
> > Fixes: ee584e9710b9 ("vhost: add driver on top of the library")
> > Signed-off-by: Rich Lane <rich.lane@bigswitch.com>
> 
> Acked-by: Yuanhan Liu <yuanhan.liu@linux.intel.com>

Applied, thanks

Re: [dpdk-dev] [PATCH] vhost: call rte_vhost_enable_guest_notification only on enabled queues

[Tetsuya Mukawa]

On 2016/04/08 2:20, Thomas Monjalon wrote:
>>> If the vhost PMD were configured with more queues than the guest, the old
>>> code would segfault in rte_vhost_enable_guest_notification due to a NULL
>>> virtqueue pointer.
>>>
>>> Fixes: ee584e9710b9 ("vhost: add driver on top of the library")
>>> Signed-off-by: Rich Lane <rich.lane@bigswitch.com>
>> Acked-by: Yuanhan Liu <yuanhan.liu@linux.intel.com>
> Applied, thanks

Hi Rich and Yuanhan,

I am sorry for late reply. I was out of office yesterday.
I've tested it also today, and seems to be good.

Thanks,
Tetsuya

Re: [dpdk-dev] [PATCH] vhost: call rte_vhost_enable_guest_notification only on enabled queues

[Yuanhan Liu]

On Fri, Apr 08, 2016 at 10:45:33AM +0900, Tetsuya Mukawa wrote:
> On 2016/04/08 2:20, Thomas Monjalon wrote:
> >>> If the vhost PMD were configured with more queues than the guest, the old
> >>> code would segfault in rte_vhost_enable_guest_notification due to a NULL
> >>> virtqueue pointer.
> >>>
> >>> Fixes: ee584e9710b9 ("vhost: add driver on top of the library")
> >>> Signed-off-by: Rich Lane <rich.lane@bigswitch.com>
> >> Acked-by: Yuanhan Liu <yuanhan.liu@linux.intel.com>
> > Applied, thanks
> 
> Hi Rich and Yuanhan,
> 
> I am sorry for late reply. I was out of office yesterday.

Tetsuya, no worry! Man takes leaves, and isn't the reason we have 2 or
more maintainers on some blocks? :)

	--yliu

> I've tested it also today, and seems to be good.
> 
> Thanks,
> Tetsuya

Re: [dpdk-dev] [PATCH] vhost: call rte_vhost_enable_guest_notification only on enabled queues

[Tetsuya Mukawa]

On 2016/04/08 15:14, Yuanhan Liu wrote:
> On Fri, Apr 08, 2016 at 10:45:33AM +0900, Tetsuya Mukawa wrote:
>> On 2016/04/08 2:20, Thomas Monjalon wrote:
>>>>> If the vhost PMD were configured with more queues than the guest, the old
>>>>> code would segfault in rte_vhost_enable_guest_notification due to a NULL
>>>>> virtqueue pointer.
>>>>>
>>>>> Fixes: ee584e9710b9 ("vhost: add driver on top of the library")
>>>>> Signed-off-by: Rich Lane <rich.lane@bigswitch.com>
>>>> Acked-by: Yuanhan Liu <yuanhan.liu@linux.intel.com>
>>> Applied, thanks
>> Hi Rich and Yuanhan,
>>
>> I am sorry for late reply. I was out of office yesterday.
> Tetsuya, no worry! Man takes leaves, and isn't the reason we have 2 or
> more maintainers on some blocks? :)
>
> 	--yliu

Yeah, thanks!

Tetsuya