[dpdk-stable] [PATCH 02/15] vhost: configure vDPA as soon as the device is ready

[Maxime Coquelin]

On 9/2/19 11:02 AM, Wang, Xiao W wrote:
> Hi,
> 
>> -----Original Message-----
>> From: Ye, Xiaolong
>> Sent: Monday, September 2, 2019 4:35 PM
>> To: Maxime Coquelin <maxime.coquelin at redhat.com>
>> Cc: Bie, Tiwei <tiwei.bie at intel.com>; Wang, Zhihong
>> <zhihong.wang at intel.com>; amorenoz at redhat.com; Wang, Xiao W
>> <xiao.w.wang at intel.com>; dev at dpdk.org; jfreimann at redhat.com;
>> stable at dpdk.org
>> Subject: Re: [PATCH 02/15] vhost: configure vDPA as soon as the device is
>> ready
>>
>> On 08/29, Maxime Coquelin wrote:
>>> There might not have any VHOST_USER_SET_VRING_CALL requests
>>> sent once virtio device is ready. When it happens, the vDPA
>>> device's dev_conf() callback may never be called.
>>>
>>> Fixes: 9f9014512822 ("vhost: configure vDPA device after set vring call
>> message")
>>> Cc: stable at dpdk.org
>>> Cc: xiaolong.ye at intel.com
>>>
>>> Signed-off-by: Maxime Coquelin <maxime.coquelin at redhat.com>
>>> ---
>>> lib/librte_vhost/vhost_user.c | 3 +--
>>> 1 file changed, 1 insertion(+), 2 deletions(-)
>>>
>>> diff --git a/lib/librte_vhost/vhost_user.c b/lib/librte_vhost/vhost_user.c
>>> index 0b72648a5..b1ea80c52 100644
>>> --- a/lib/librte_vhost/vhost_user.c
>>> +++ b/lib/librte_vhost/vhost_user.c
>>> @@ -2112,8 +2112,7 @@ vhost_user_msg_handler(int vid, int fd)
>>> 	did = dev->vdpa_dev_id;
>>> 	vdpa_dev = rte_vdpa_get_device(did);
>>> 	if (vdpa_dev && virtio_is_ready(dev) &&
>>> -			!(dev->flags & VIRTIO_DEV_VDPA_CONFIGURED) &&
>>> -			msg.request.master ==
>> VHOST_USER_SET_VRING_CALL) {
>>> +			!(dev->flags & VIRTIO_DEV_VDPA_CONFIGURED)) {
> 
> In the early beginning of vhost user messages, there seems to be a VHOST_USER_SET_VRING_CALL with invalid call fd,
> not sure if QEMU has any update on this point.
> If the virtio_is_ready() is based on that invalid call fd, then vdpa_dev_conf() cannot setup interrupt properly.
> I think that's why in our previous implementation, we wait for the real call fd and then call dev_conf().

I think that when we receive the first SET_VRING_CALL request from Qemu,
virtio_is_ready() should be false because the rings addresses won't be
received yet.

Did it fixed a real issue, or was it based on code/logs review?

Thanks for the explanations,
Maxime

> BRs,
> Xiao
>