Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
Date:   Mon, 9 Jul 2018 11:15:27 -0700
From:   Stephen Hemminger <stephen@networkplumber.org>
To:     Haiyang Zhang <haiyangz@linuxonhyperv.com>
Cc:     haiyangz@microsoft.com, davem@davemloft.net,
        netdev@vger.kernel.org, olaf@aepfle.de, sthemmin@microsoft.com,
        linux-kernel@vger.kernel.org, devel@linuxdriverproject.org,
        vkuznets@redhat.com
Subject: Re: [PATCH net] hv_netvsc: Fix napi reschedule while receive
 completion is busy
Message-ID: <20180709111527.70ab354c@xeon-e3>
In-Reply-To: <20180709164319.5058-1-haiyangz@linuxonhyperv.com>
References: <20180709164319.5058-1-haiyangz@linuxonhyperv.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=US-ASCII
Content-Transfer-Encoding: 7bit
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Mon,  9 Jul 2018 16:43:19 +0000
Haiyang Zhang <haiyangz@linuxonhyperv.com> wrote:

> From: Haiyang Zhang <haiyangz@microsoft.com>
> 
> If out ring is full temporarily and receive completion cannot go out,
> we may still need to reschedule napi if other conditions are met.
> Otherwise the napi poll might be stopped forever, and cause network
> disconnect.
> 
> Fixes: 7426b1a51803 ("netvsc: optimize receive completions")
> Signed-off-by: Haiyang Zhang <haiyangz@microsoft.com>
> ---
>  drivers/net/hyperv/netvsc.c | 8 ++++----
>  1 file changed, 4 insertions(+), 4 deletions(-)
> 
> diff --git a/drivers/net/hyperv/netvsc.c b/drivers/net/hyperv/netvsc.c
> index 8e9d0ee1572b..caaf5054f446 100644
> --- a/drivers/net/hyperv/netvsc.c
> +++ b/drivers/net/hyperv/netvsc.c
> @@ -1285,14 +1285,14 @@ int netvsc_poll(struct napi_struct *napi, int budget)
>  		nvchan->desc = hv_pkt_iter_next(channel, nvchan->desc);
>  	}
>  
> -	/* If send of pending receive completions suceeded
> -	 *   and did not exhaust NAPI budget this time
> +	send_recv_completions(ndev, net_device, nvchan);
> +
> +	/* If it did not exhaust NAPI budget this time
>  	 *   and not doing busy poll
>  	 * then re-enable host interrupts
>  	 *     and reschedule if ring is not empty.
>  	 */
> -	if (send_recv_completions(ndev, net_device, nvchan) == 0 &&
> -	    work_done < budget &&
> +	if (work_done < budget &&
>  	    napi_complete_done(napi, work_done) &&
>  	    hv_end_read(&channel->inbound) &&
>  	    napi_schedule_prep(napi)) {

This patch doesn't look right. I think the existing code works
as written.

If send_receive_completions is unable to send because ring is full
then vmbus_sendpacket will return -EBUSY which gets returns
from send_receive_completions.  Because the return is non-zero,
the driver will not call napi_complete_done.
Since napi_complete_done was not called, NAPI will reschedule
the napi poll routine.