Re: [PATCH hyperv-fixes] hv_netvsc: Fix unwanted wakeup after tx_disable

[Stephen Hemminger <stephen@networkplumber.org>]

On Thu, 28 Mar 2019 19:00:18 +0000
Haiyang Zhang <haiyangz@microsoft.com> wrote:

> > -----Original Message-----
> > From: Stephen Hemminger <stephen@networkplumber.org>
> > Sent: Thursday, March 28, 2019 2:38 PM
> > To: Haiyang Zhang <haiyangz@linuxonhyperv.com>
> > Cc: sashal@kernel.org; linux-hyperv@vger.kernel.org; Haiyang Zhang
> > <haiyangz@microsoft.com>; KY Srinivasan <kys@microsoft.com>; Stephen
> > Hemminger <sthemmin@microsoft.com>; olaf@aepfle.de; vkuznets
> > <vkuznets@redhat.com>; davem@davemloft.net; netdev@vger.kernel.org;
> > linux-kernel@vger.kernel.org
> > Subject: Re: [PATCH hyperv-fixes] hv_netvsc: Fix unwanted wakeup after
> > tx_disable
> > 
> > On Thu, 28 Mar 2019 17:48:45 +0000
> > Haiyang Zhang <haiyangz@linuxonhyperv.com> wrote:
> >   
> > > +static inline void netvsc_tx_enable(struct netvsc_device *nvscdev,
> > > +				    struct net_device *ndev)
> > > +{
> > > +	nvscdev->tx_disable = false;
> > > +	mb(); /* ensure queue wake up mechanism is on */
> > > +
> > > +	netif_tx_wake_all_queues(ndev);
> > > +}  
> > 
> > You don't need a full mb(). virt_wmb() should be sufficient.  
> 
> I will make this change. 
> 
> > Could I suggest an alternative approach.
> > You don't need to introduce a local tx_disable flag, the only place where a
> > wakeup could cause problems is after a send_completion was processed
> > during detach state.
> > 
> > Instead, just avoid wakeup in that place.  
> 
> In netvsc_detach(), after netif_tx_disable(), we call  netvsc_wait_until_empty(nvdev);
> TX patch should not be waken up again while waiting for in/out ring to becomes empty.
> 
> In my tests before this patch, there are wakeup happens before netif_device_detach(), 
> so netif_device_present(ndev) is still true at that time.
> 
> In other places, like netvsc_close(), link_change(), we also don't want wakeup after tx_disable.
> 
> Thanks.
> - Haiyang
> 
> > 
> > --- a/drivers/net/hyperv/netvsc.c
> > +++ b/drivers/net/hyperv/netvsc.c
> > @@ -720,6 +720,7 @@ static void netvsc_send_tx_complete(struct
> > net_device *ndev,
> >                 struct netdev_queue *txq = netdev_get_tx_queue(ndev, q_idx);
> > 
> >                 if (netif_tx_queue_stopped(txq) &&
> > +                   netif_device_present(ndev) &&
> >                     (hv_get_avail_to_write_percent(&channel->outbound) >
> >                      RING_AVAIL_PERCENT_HIWATER || queue_sends < 1)) {
> >                         netif_tx_wake_queue(txq);  


Then what about doing netif_detach earlier in netvsc_detach.

The state management is already (too) complex in netvsc and adding another
boolean flag just makes it harder to understand.