[dpdk-stable] [PATCH v2 3/3] virtio-user: set status on virtio-user reconnect

[Adrian Moreno]

On 10/22/20 9:37 AM, Xia, Chenbo wrote:
> Hi Maxime,
> 
>> -----Original Message-----
>> From: Maxime Coquelin <maxime.coquelin at redhat.com>
>> Sent: Thursday, October 22, 2020 3:14 PM
>> To: Wang, Yinan <yinan.wang at intel.com>; Adrian Moreno
>> <amorenoz at redhat.com>; dev at dpdk.org; Xia, Chenbo <chenbo.xia at intel.com>
>> Cc: Fu, Patrick <patrick.fu at intel.com>; stable at dpdk.org; Wang, Zhihong
>> <zhihong.wang at intel.com>; Xu, Qian Q <qian.q.xu at intel.com>
>> Subject: Re: [PATCH v2 3/3] virtio-user: set status on virtio-user
>> reconnect
>>
>> Hi Yinan,
>>
>> On 10/22/20 6:01 AM, Wang, Yinan wrote:
>>> Hi Maxime/ Adrian，
>>>
>>> Thanks for the patch.  we can launch vhost-user with client mode with
>> this fix patch.
>>> But still fail to get throughput with basic vhost/virtio-user server
>> mode loopback test. This is another problem which introduced by
>> 57912824615fd7787a48a7b18e40661466.
>>> Bugzilla: https://bugs.dpdk.org/show_bug.cgi?id=541
>>
>> Thanks for reporting the issue, I will look at it this morning.
>>
>> BTW, this virtio-user server mode is broken by design, we should really
>> fix it.
>>
>> For example, it assumes features to be supported by the backend before
>> the negotiation took place:
>> https://git.dpdk.org/dpdk/tree/drivers/net/virtio/virtio_user/virtio_user_
>> dev.c#n513
>>
>> As part of my rework, I suggest we implement the same behaviour as QEMU
>> does, which will be reliable and consistent with QEMU. It means that if
>> server mode is enabled in device command-line, the driver waits until
>> the socket is ready.
>>
>> Chenbo, Adrian, what do you think?
> 
> Yes! I totally agree and have the same opinion for a long time. As I remember, this
> feature assumption has caused problems before and I notice the new STATUS feature is
> also affected. It's good to make this consistent with QEMU 😊.
> 
+1

In fact, I was looking into how to get rid of some error messages originated
during the initialization of the virtio-pci layer when the socket is still not
ready.


> Cheers,
> Chenbo
> 
>>
>> Thanks,
>> Maxime
>>
>>> BR,
>>> Yinan
>>>> -----Original Message-----
>>>> From: Maxime Coquelin <maxime.coquelin at redhat.com>
>>>> Sent: 2020年10月21日 0:43
>>>> To: Adrian Moreno <amorenoz at redhat.com>; dev at dpdk.org
>>>> Cc: Wang, Yinan <yinan.wang at intel.com>; Fu, Patrick
>> <patrick.fu at intel.com>;
>>>> stable at dpdk.org; Xia, Chenbo <chenbo.xia at intel.com>; Wang, Zhihong
>>>> <zhihong.wang at intel.com>
>>>> Subject: Re: [PATCH v2 3/3] virtio-user: set status on virtio-user
>> reconnect
>>>>
>>>>
>>>>
>>>> On 10/20/20 5:20 PM, Adrian Moreno wrote:
>>>>> Newer vhost-user backends will rely on SET_STATUS to start the device
>>>>> so this required to support them.
>>>>>
>>>>> Fixes: 57912824615f ("net/virtio-user: support vhost status setting")
>>>>> Cc: maxime.coquelin at redhat.com
>>>>> Cc: stable at dpdk.org
>>>>>
>>>>> Signed-off-by: Adrian Moreno <amorenoz at redhat.com>
>>>>> ---
>>>>>  drivers/net/virtio/virtio_user_ethdev.c | 14 ++++++++++++--
>>>>>  1 file changed, 12 insertions(+), 2 deletions(-)
>>>>>
>>>>> diff --git a/drivers/net/virtio/virtio_user_ethdev.c
>>>> b/drivers/net/virtio/virtio_user_ethdev.c
>>>>> index e870fb2ff..d8bea4537 100644
>>>>> --- a/drivers/net/virtio/virtio_user_ethdev.c
>>>>> +++ b/drivers/net/virtio/virtio_user_ethdev.c
>>>>> @@ -78,6 +78,13 @@ virtio_user_server_reconnect(struct virtio_user_dev
>>>> *dev)
>>>>>  		return -1;
>>>>>
>>>>>  	dev->vhostfd = connectfd;
>>>>> +
>>>>> +	vtpci_reset(hw);
>>>>> +
>>>>> +	vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_ACK);
>>>>> +
>>>>> +	vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_DRIVER);
>>>>> +
>>>>>  	if (dev->ops->send_request(dev, VHOST_USER_GET_FEATURES,
>>>>>  				   &dev->device_features) < 0) {
>>>>>  		PMD_INIT_LOG(ERR, "get_features failed: %s",
>>>>> @@ -111,6 +118,8 @@ virtio_user_server_reconnect(struct
>> virtio_user_dev
>>>> *dev)
>>>>>
>>>>>  	dev->features &= dev->device_features;
>>>>>
>>>>> +	vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_FEATURES_OK);
>>>>> +
>>>>>  	/* For packed ring, resetting queues is required in reconnection. */
>>>>>  	if (vtpci_packed_queue(hw) &&
>>>>>  	   (vtpci_get_status(hw) & VIRTIO_CONFIG_STATUS_DRIVER_OK)) {
>>>>> @@ -119,8 +128,9 @@ virtio_user_server_reconnect(struct
>> virtio_user_dev
>>>> *dev)
>>>>>  		virtio_user_reset_queues_packed(eth_dev);
>>>>>  	}
>>>>>
>>>>> -	ret = virtio_user_start_device(dev);
>>>>> -	if (ret < 0)
>>>>> +	/*  Start the device */
>>>>> +	vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_DRIVER_OK);
>>>>> +	if (!dev->started)
>>>>>  		return -1;
>>>>>
>>>>>  	if (dev->queue_pairs > 1) {
>>>>>
>>>>
>>>> Reviewed-by: Maxime Coquelin <maxime.coquelin at redhat.com>
>>>>
>>>> Thanks,
>>>> Maxime
>>>
> 

-- 
Adrián Moreno