Date:   Fri, 14 Jan 2022 09:02:05 +0100
From:   Stefano Garzarella <sgarzare@redhat.com>
To:     Jason Wang <jasowang@redhat.com>
Cc:     virtualization <virtualization@lists.linux-foundation.org>,
        netdev <netdev@vger.kernel.org>,
        Stefan Hajnoczi <stefanha@redhat.com>,
        linux-kernel <linux-kernel@vger.kernel.org>,
        kvm <kvm@vger.kernel.org>, "Michael S. Tsirkin" <mst@redhat.com>
Subject: Re: [RFC PATCH] vhost: cache avail index in vhost_enable_notify()
Message-ID: <20220114080205.ls4txgj7qbqmc3q5@steredhat>
References: <20220113145642.205388-1-sgarzare@redhat.com>
 <CACGkMEsqY5RHL=9=iny6xRVs_=EdACUCfX-Rmpq+itpdoT_rrg@mail.gmail.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii; format=flowed
Content-Disposition: inline
In-Reply-To: <CACGkMEsqY5RHL=9=iny6xRVs_=EdACUCfX-Rmpq+itpdoT_rrg@mail.gmail.com>
Precedence: bulk

On Fri, Jan 14, 2022 at 02:18:01PM +0800, Jason Wang wrote:
>On Thu, Jan 13, 2022 at 10:57 PM Stefano Garzarella <sgarzare@redhat.com> wrote:
>>
>> In vhost_enable_notify() we enable the notifications and we read
>> the avail index to check if new buffers have become available in
>> the meantime. In this case, the device would go to re-read avail
>> index to access the descriptor.
>>
>> As we already do in other place, we can cache the value in `avail_idx`
>> and compare it with `last_avail_idx` to check if there are new
>> buffers available.
>>
>> Signed-off-by: Stefano Garzarella <sgarzare@redhat.com>
>
>Patch looks fine but I guess we won't get performance improvement
>since it doesn't save any userspace/VM memory access?

It should save the memory access when vhost_enable_notify() find 
something new in the VQ, so in this path:

     vhost_enable_notify() <- VM memory access for avail index
       == true

     vhost_disable_notify()
     ...

     vhost_get_vq_desc()   <- VM memory access for avail index
                              with the patch applied, this access is 
                              avoided since avail index is cached

In any case, I don't expect this to be a very common path, indeed we
usually use unlikely() for this path:

     if (unlikely(vhost_enable_notify(dev, vq))) {
         vhost_disable_notify(dev, vq);
         continue;
     }

So I don't expect a significant performance increase.

v1 coming with a better commit description.

Thanks,
Stefano