Re: [PATCH 2/4] tcm_vhost: Introduce tcm_vhost_check_endpoint()

[Paolo Bonzini <pbonzini@redhat.com>]

Il 13/03/2013 04:02, Asias He ha scritto:
> On Tue, Mar 12, 2013 at 09:26:18AM +0100, Paolo Bonzini wrote:
>> Il 12/03/2013 03:42, Asias He ha scritto:
>>> This helper is useful to check if vs->vs_endpoint is setup by
>>> vhost_scsi_set_endpoint()
>>>
>>> Signed-off-by: Asias He <asias@redhat.com>
>>> Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
>>> ---
>>>  drivers/vhost/tcm_vhost.c | 12 ++++++++++++
>>>  1 file changed, 12 insertions(+)
>>>
>>> diff --git a/drivers/vhost/tcm_vhost.c b/drivers/vhost/tcm_vhost.c
>>> index b3e50d7..29612bc 100644
>>> --- a/drivers/vhost/tcm_vhost.c
>>> +++ b/drivers/vhost/tcm_vhost.c
>>> @@ -91,6 +91,18 @@ static int iov_num_pages(struct iovec *iov)
>>>  	       ((unsigned long)iov->iov_base & PAGE_MASK)) >> PAGE_SHIFT;
>>>  }
>>>  
>>> +static bool tcm_vhost_check_endpoint(struct vhost_scsi *vs)
>>> +{
>>> +	bool ret = false;
>>> +
>>> +	mutex_lock(&vs->dev.mutex);
>>> +	if (vs->vs_endpoint)
>>> +		ret = true;
>>> +	mutex_unlock(&vs->dev.mutex);
>>
>> The return value is invalid as soon as mutex_unlock is called, i.e.
>> before tcm_vhost_check_endpoint returns.  Instead, check vs->vs_endpoint
>> in the caller while the mutex is taken.
> 
> Do you mean 1) or 2)?
> 
>    1)
>    vhost_scsi_handle_vq()
>    {
>    
>       mutex_lock(&vs->dev.mutex);
>       check vs->vs_endpoint
>       mutex_unlock(&vs->dev.mutex);
>    
>       handle vq
>    }
>    
>    2)
>    vhost_scsi_handle_vq()
>    {
>    
>       lock vs->dev.mutex
>       check vs->vs_endpoint
>       handle vq
>       unlock vs->dev.mutex
>    }
> 
> 1) does not make any difference with the original
> one right?

Yes, it's just what you have with tcm_vhost_check_endpoint inlined.

> 2) would be too heavy. This might not be a problem in current 1 thread
> per vhost model. But if we want concurrent multiqueue, this will be
> killing us.

I mean (2).  You could use an rwlock to enable more concurrency.

Paolo