Re: [PATCH 2/8] vhost: add helper to check if a vq has been setup

[Jason Wang <jasowang@redhat.com>]

On 2020/9/24 上午3:12, Mike Christie wrote:
> On 9/21/20 9:02 PM, Jason Wang wrote:
>> On 2020/9/22 上午2:23, Mike Christie wrote:
>>> This adds a helper check if a vq has been setup. The next patches
>>> will use this when we move the vhost scsi cmd preallocation from per
>>> session to per vq. In the per vq case, we only want to allocate cmds
>>> for vqs that have actually been setup and not for all the possible
>>> vqs.
>>>
>>> Signed-off-by: Mike Christie <michael.christie@oracle.com>
>>> ---
>>>    drivers/vhost/vhost.c | 9 +++++++++
>>>    drivers/vhost/vhost.h | 1 +
>>>    2 files changed, 10 insertions(+)
>>>
>>> diff --git a/drivers/vhost/vhost.c b/drivers/vhost/vhost.c
>>> index b45519c..5dd9eb1 100644
>>> --- a/drivers/vhost/vhost.c
>>> +++ b/drivers/vhost/vhost.c
>>> @@ -305,6 +305,15 @@ static void vhost_vring_call_reset(struct vhost_vring_call *call_ctx)
>>>        spin_lock_init(&call_ctx->ctx_lock);
>>>    }
>>>    +bool vhost_vq_is_setup(struct vhost_virtqueue *vq)
>>> +{
>>> +    if (vq->avail && vq->desc && vq->used && vhost_vq_access_ok(vq))
>>> +        return true;
>>> +    else
>>> +        return false;
>>> +}
>>> +EXPORT_SYMBOL_GPL(vhost_vq_is_setup);
>>
>> This is probably ok but I wonder maybe we should have something like what vDPA did (VHOST_SET_VRING_ENABLE) to match virtio 1.0 device definition.
> It looks like I can make that work. Some questions:
>
> 1. Do you mean a generic VHOST_SET_VRING_ENABLE or a SCSI specific one VHOST_SCSI_SET_VRING_ENABLE?


It would be better if we can make it generic.


>
> 2. I can see the VHOST_VDPA_SET_VRING_ENABLE kernel code and the vhost_set_vring_enable qemu code, so I have an idea of how it should work for vhost scsi. However, I'm not sure the requirements for a generic VHOST_SET_VRING_ENABLE if that is what you meant. I could not find it in the spec either. Could you send me a pointer to the section?


In the spec, for PCI, it's the queue_enable for modern device.


>
> For example, for vhost-net we seem to enable a device in the VHOST_NET_SET_BACKEND ioctl, so I'm not sure what behavior should be or needs to be implemented for net and vsock.


Yes, but VHOST_NET_SET_BACKEND is for the whole device not a specific 
virtqueue.


Thanks


>